Intelligent Decision Support System for precision medicine
Abstract
Este trabalho de doutoramento corresponde a uma exploração abrangente da integração de Sistemas de Suporte à Decisão Inteligente (SSDI) no contexto da medicina de precisão. O capítulo introdutório estabelece o cenário oferecendo uma visão detalhada da investigação, delineando motivações, objetivos e a importância do trabalho. Motivada pela necessidade imperativa de ir além da abordagem tradicional baseada em sintomas na tomada de decisão clínica, a pesquisa procura abordar questões fundamentais relacionadas à adaptação de estratégias de tratamento das características individuais do paciente através da abordagem baseada em dados. Uma análise crítica aos protocolos de tomada de decisão clínica revela limitações significativas na adequação dos tratamentos às circunstâncias individuais dos pacientes. O paradigma da medicina de precisão destaca a necessidade de uma introspeção baseada nos dados por forma a aprimorar os processos decisórios. Apesar dos avanços nos Sistemas de Suporte à Decisão Clínica (SSDC), desafios como a qualidade dos dados e interoperabilidade persistem, exigindo uma abordagem transformadora. Com base em dois projetos distintos, o estudo desenvolve-se em domínios interdisciplinares, preenchendo lacunas numa perspetiva baseada em dados, para aperfeiçoar os SSDI no âmbito da Medicina de Precisão (IDSS4PM). Ao explorar conjuntos de dados oriundos da psicoterapia e da medicina intensiva, a investigação centra-se no processamento de dados, capacidade analítica e na modelação preditiva, estabelecendo as bases para aprimorar os protocolos de tomada de decisão clínica. O enquadramento teórico foi inspirado no modelo de tomada de decisão de Simon para orientação do desenho do IDSS4PM. Nos capítulos subsequentes encontra-se a definição da metodologia de pesquisa e um conjunto de capítulos onde são apresentados os resultados experimentais que contribuem para o aperfeiçoamento dos SSDI. O estudo visa fazer uma contribuição significativa para a medicina de precisão, tirando partido de perceções baseadas em dados. Nosso objetivo é promover uma abordagem precisa e eficaz à saúde e orientar mudanças transformadoras nas práticas de tomada de decisão clínica. Na conclusão, é feito um conjunto recomendações para futuros trabalhos de pesquisa, reconhecendo as limitações do trabalho atual.
Full text
Nasim Sadat Mosavi April 2024 UMinho | 2024 Intelligent Decision Support System for Precision Medicine Universidade do Minho Escola de Engenharia Nasim Sadat Mosavi Intelligent Decision Support System for Precision Medicine
April 2024 Doctoral Program in Information Systems and Technology Thesis performed under the supervision of Professor Doutor Manuel Filipe Santos Nasim Sadat Mosavi Intelligent Decision Support System for Precision Medicine Universidade do Minho Escola de Engenharia
ii COPYRIGHT Este é um trabalho académico que pode ser utilizado por terceiros desde que respeitadas as regras e boas práticas internacionalmente aceites, no que concerne aos direitos de autor e direitos conexos. Assim, o presente trabalho pode ser utilizado nos termos previstos na licença abaixo indicada. Caso o utilizador necessite de permissão para poder fazer um uso do trabalho em condições não previstas no licenciamento indicado, deverá contactar o autor, através do RepositóriUM da Universidade do Minho. O estudante deverá escolher uma das seguintes licenças (os textos das licenças são transcrições ipsis verbis do Despacho RT-31/2019 – Anexo 3)/The student must choose one of the following licenses. Atribuição CC BY https://creativecommons.org/licenses/by/4.0/
iii ACKNOWLEDGMENTS I would like to begin by expressing my deepest gratitude to Professor Manuel Filipe Santos, my esteemed PhD supervisor, for his unwavering guidance, expertise, and mentorship throughout my doctoral journey. His insightful feedback, constructive criticism, and steadfast support have played a pivotal role in shaping my research and academic development. I am profoundly grateful for the opportunity to learn from him, and his mentorship has been instrumental in my academic growth. To my beloved parents, whose constant support has been the cornerstone of my journey, I owe an immeasurable debt of gratitude. Their endless encouragement, sage advice, and strong belief in my abilities have been a constant source of strength and inspiration. Their sacrifices and dedication to my success have shaped me into the person I am today, and for that, I am eternally thankful. I am also deeply appreciative of my son Kourosh and my daughter Elena for their deep understanding and steadfast support. Their presence in my life has brought immeasurable joy and motivation, and their belief in my capabilities has propelled me forward during challenging times. Their unwavering faith in me has been a guiding light, driving my determination and resilience. Special recognition is extended to my husband, Roozbeh, whose unconditional love, support, and understanding have been my rock throughout this journey. His persistent encouragement and belief in my potential have provided me with the strength and confidence to overcome obstacles and achieve success. His steadfast support has created a nurturing environment that allowed me to focus wholeheartedly on my research and academic pursuits. Lastly, I extend my heartfelt gratitude to all my family members for their encouragement, and belief in my potential. Their continuous support has been the wind beneath my wings, propelling me toward my academic goals. I am truly fortunate to have such incredible motivations on my side, and I am deeply grateful for their presence in my life.
iv DECLARATION OF INTEGRITY I hereby declare having conducted this academic work with integrity. I confirm that I have not used plagiarism or any form of undue use of information or falsification of results along the process leading to its elaboration. I further declare that I have fully acknowledged the Code of Ethical Conduct of the University of Minho.
v RESUMO Sistema de Suporte à Decisão Inteligente para Medicina de Precisão Sistema de Suporte à Decisão Inteligente para Medicina de Precisão Este trabalho de doutoramento corresponde a uma exploração abrangente da integração de Sistemas de Suporte à Decisão Inteligente (SSDI) no contexto da medicina de precisão. O capítulo introdutório estabelece o cenário oferecendo uma visão detalhada da investigação, delineando motivações, objetivos e a importância do trabalho. Motivada pela necessidade imperativa de ir além da abordagem tradicional baseada em sintomas na tomada de decisão clínica, a pesquisa procura abordar questões fundamentais relacionadas à adaptação de estratégias de tratamento das características individuais do paciente através da abordagem baseada em dados. Uma análise crítica aos protocolos de tomada de decisão clínica revela limitações significativas na adequação dos tratamentos às circunstâncias individuais dos pacientes. O paradigma da medicina de precisão destaca a necessidade de uma introspeção baseada nos dados por forma a aprimorar os processos decisórios. Apesar dos avanços nos Sistemas de Suporte à Decisão Clínica (SSDC), desafios como a qualidade dos dados e interoperabilidade persistem, exigindo uma abordagem transformadora. Com base em dois projetos distintos, o estudo desenvolve-se em domínios interdisciplinares, preenchendo lacunas numa perspetiva baseada em dados, para aperfeiçoar os SSDI no âmbito da Medicina de Precisão (IDSS4PM). Ao explorar conjuntos de dados oriundos da psicoterapia e da medicina intensiva, a investigação centra-se no processamento de dados, capacidade analítica e na modelação preditiva, estabelecendo as bases para aprimorar os protocolos de tomada de decisão clínica. O enquadramento teórico foi inspirado no modelo de tomada de decisão de Simon para orientação do desenho do IDSS4PM. Nos capítulos subsequentes encontra-se a definição da metodologia de pesquisa e um conjunto de capítulos onde são apresentados os resultados experimentais que contribuem para o aperfeiçoamento dos SSDI. O estudo visa fazer uma contribuição significativa para a medicina de precisão, tirando partido de perceções baseadas em dados. Nosso objetivo é promover uma abordagem precisa e eficaz à saúde e orientar mudanças transformadoras nas práticas de tomada de decisão clínica. Na conclusão, é feito um conjunto recomendações para futuros trabalhos de pesquisa, reconhecendo as limitações do trabalho atual. Palavras-chave: Inteligência Artificial, Medicina de Precisão, Sistemas de Suporte à Decisão Inteligente Baseados em Dados, Tomada de Decisão Clínica,.
vi ABSTRACT Intelligent Decision Support System for Precision Medicine This doctoral thesis embarks on a comprehensive exploration into the integration of Intelligent Decision Support Systems (IDSS) within the evolving landscape of Precision Medicine (PM). The introductory chapter sets the stage by offering a detailed overview of the research, outlining motivations, objectives, and the significance of the study. Motivated by the imperative need to move beyond the traditional symptom-driven approach in clinical decision-making, the research seeks to address fundamental questions surrounding the tailoring of treatment strategies to individual patient characteristics via a datadriven approach. A critical analysis of existing clinical decision-making protocols reveals significant limitations, particularly in the realm of tailoring treatments to individual patient's unique circumstances. The emergence of Precision Medicine as a paradigm shift underscores the necessity for data-driven insights to inform decision-making processes effectively. Despite the advancements in Clinical Decision Support Systems (CDSS), challenges such as data quality, interoperability, and fragmented data generation persist, necessitating a transformative approach. Through two distinct projects, the study navigates interdisciplinary domains, bridging gaps in data-driven perspectives to advance the framework of IDSS for Precision Medicine (IDSS4PM). By exploring psychotherapy and Intensive Care Unit (ICU) datasets, the research delves into data processing, analytical insights, and predictive modeling, laying the groundwork for enhancing clinical decision-making protocols. The theoretical framework draws inspiration from Simon's model of decision-making, guiding the proposal of IDSS4PM. Chapters unfold to elucidate the research methodology, practical phases, and contributions toward refining the IDSS framework. The study aimed to make a meaningful contribution to PM by leveraging data-driven insights. Our goal is to foster a precise and effective approach to healthcare and guide transformative shifts in clinical decisionmaking practices. In conclusion, the thesis offers exhaustive recommendations for future research endeavors, acknowledging limitations. Keywords: Artificial Intelligence, Clinical Decision Making, Data Driven, Intelligent Decision Support Systems, Precision Medicine.
vii INDEX Copyright............................................................................................................................................. ii Acknowledgments ............................................................................................................................... iii Declaration of integrity ........................................................................................................................ iv Resumo............................................................................................................................................... v Abstract.............................................................................................................................................. vi List of abbreviations/Acronyms .......................................................................................................... xii List of Figures ................................................................................................................................... xiv List of Equations ............................................................................................................................... xvii List of tables .................................................................................................................................... xviii 1. Introduction ................................................................................................................................ 1 1.3 Motivation ............................................................................................................................ 1 2.3 Research Question And Objectives ....................................................................................... 5 3.3 Scope of The Study .............................................................................................................. 6 4.3 Significance And Contributions ............................................................................................. 7 1.5 Theoretical Framework......................................................................................................... 8 1.6 Structure of Thesis ............................................................................................................... 9 2. State of the art ......................................................................................................................... 11 2.1 Development of Decision Support Systems ......................................................................... 11 2.1.1 Types of Decision Support Systems ............................................................................ 17 2.1.2 Intelligent Decision Support Systems .......................................................................... 19 2.2 Precision Medicine as a New Approach in Clinical Decision Making..................................... 20 2.2.1 Precision Medicine: an Evolving Paradigm in Clinical Decision-Making ......................... 20 2.2.2 Catalyst of Change Is Revolutionizing The Clinical Decision-Making ............................. 23 2.2.3 IDSS4PM; Gaps and Limitations ................................................................................ 30
xiv LIST OF FIGURES Figure 1: Literature-Map 1960s to 2000s 13 Figure 2: Literature-Map 2000s to 2019s 14 Figure 3: Maturity model of DSS 16 Figure 4: Types of DSS 19 Figure 5: Precision Medicine 23 Figure 6: Various Drivers of evolving 24 Figure 7: Sources of Big Data 25 Figure 8: Healthcare Analytics 28 Figure 9:IDSS4PM; gaps and limitations 31 Figure 10: Design science research methodology 38 Figure 11: CRISP-DM 1.0 41 Figure 12: Mapping methodologies 43 Figure 13: Literature review strategy 44 Figure 14: Data exploration (Psychotherapy) 47 Figure 15: Relationship between session and WAI score (TA) for clients; termination=1 49 Figure 16: Relationship between session and WAI score (TA) for clients; termination=0 50 Figure 17: Relationship between session and WAI score (TA) for therapists; termination=1 51 Figure 18: Relationship between session and WAI score (TA) for therapists; termination=0 52 Figure 19: Average of WAI score (TA) by sex – client 53 Figure 20: Average WAI score (TA) by sex therapist 54 Figure 21: Average client’s WAI score (TA) by client ID 54 Figure 22: Average therapist’s WAI score (TA) by therapist ID 55 Figure 23: Therapy sessions attended by clients 56 Figure 24: Therapy sessions attended by therapists 56 Figure 25: Relationship between HR and WAI score (TA) for clients; termination=1 57 Figure 26: Relationship between HR and WAI score (TA) for clients; termination=0 57 Figure 27: Relationship between HR and WAI score (TA) for therapists; termination=1 58 Figure 28: Relationship between HR and WAI score (TA) for therapists; termination=0 59 Figure 29: Relationship between EDA and WAI score (TA) for clients; termination=1 59 Figure 30: Relationship between EDA and WAI score (TA) for clients; termination=0 60
xv Figure 31: Relationship between EDA and WAI score (TA) for therapists; termination=1 61 Figure 32: Relationship between EDA and WAI score (TA) for therapists; termination=0 61 Figure 33:Contributions IDSS4PM; Analytical Insights 62 Figure 34: Pearson Correlation Coefficients Matrix 64 Figure 35: Contributions to IDSS4PM; Predictive Modelling 71 Figure 36: Key characteristics of IDSS4PM ‘s framework 73 Figure 37: Data exploration 75 Figure 38: Steps to perform analytical insight 76 Figure 39: Structure of CEid 78 Figure 40: Access the Patient’s analytical insight 79 Figure 41: Length of stay in ICU 79 Figure 42: Analysing Clinical Data for Patients with process number:”1000025” 80 Figure 43: Temporal analysis of diverse clinical transactions 80 Figure 44: Temporal analysis of Diagnostic transactions 81 Figure 45: Temporal analysis of intervention transactions 82 Figure 46: Temporal analysis of oxygen saturation 83 Figure 47: Temporal analysis of Heart Rate 83 Figure 48: Temporal analysis of laboratory exams 84 Figure 49: Temporal analysis of sepsis 84 Figure 50: Temporal analysis of medications 85 Figure 51: Contributions to IDSS4PM; Analytical Insights 90 Figure 52: Towards temporal clustering analysis 91 Figure 53: Pipeline to unify datasets 92 Figure 54: Elbow method for the optimal number of clusters 93 Figure 55: Contributions to IDSS4PM; Clustering 97 Figure 56: Cluster-based data mapping and cluster assignment 97 Figure 57: Behaviour of data in each cluster 100 Figure 58: Behaviour of Glasgow coma in each cluster 105 Figure 59: Behaviour of systolic arterial blood pressure in each cluster 105 Figure 60: Behaviour of arterial (mean) blood pressure in each cluster 106 Figure 61: Behaviour of pulse rate of arterial blood pressure in each cluster 106 Figure 62: : Behaviour of diastolic arterial blood pressure in each cluste 106
xvi Figure 63: Behaviour of oxygen saturation level in each cluster 107 Figure 64: Behaviour of heart rate in each cluster 107 Figure 65: Behaviour of body temperature in each cluster 108 Figure 66: Behavior of sepsis in each cluster 108 Figure 67: Analysing the total days of clinical transactions in each cluste 109 Figure 68: Temporal distribution of each cluster 110 Figure 69: Behaviour of laboratory exams & results in each cluster 111 Figure 70: Distribution of interventions (codes) in each cluster 112 Figure 71: Distribution of diagnostic (codes) in each cluster 112 Figure 72: Distribution of medications (codes) in each cluster 113 Figure 73: Distribution of procedure-local (codes) in each cluster 114 Figure 74: Distribution of procedure-zona (codes) in each cluster 114 Figure 75: Contributions to IDSS4PM; Analytical Insight 116 Figure 76: Significant features of RF 120 Figure 77: Contributions to IDSS4PM; Predictive analysis 121 Figure 78: Theoretical reference; Project I 124 Figure 79: Contributions to IDSS4PM; Project I 126 Figure 80: Theoretical reference; Project II 133 Figure 81: :Contribution to IDSS4PM; Project II 135
xvii LIST OF EQUATIONS Equation 1.Accuracy 117 Equation 2.Precision 118 Equation 3.Recall 118 Equation 4.F1Score 118 Equation 5.Cohen's Kappa 119
xviii LIST OF TABLES Table 1: Features of Big Data in healthcare (Wang, 2017) 26 Table 2: Techniques to deal with limitations associated with data processing 33 Table 3: Pearson Correlation Between TA (WAI Scores) And Predictors 65 Table 4: Pearson Correlation Among” WAI_Client” Predictors 66 Table 5: Pearson Correlation Among” WAI_Therapist” Predictors 66 Table 6: List of selected predictors 67 Table 7: Ranking algorithms 68 Table 8: Significance of predictor 69 Table 9: Key Transformations 77 Table 10: Characteristics of each cluster 94 Table 11: Cluster references 95 Table 12: Analysing the behavior of data in each cluster 103 Table 13: Model Performance 119
1 1. INTRODUCTION The introduction chapter serves as the gateway to this PhD thesis, offering a comprehensive overview of our research. Here, we provide the backdrop and context of the study, elucidate our motivations, and outline our objectives. Additionally, we define the scope of the study, establish the theoretical foundation, and emphasize the significance of our contributions. Finally, we outline the structure of the document, guiding readers through the subsequent chapters. Our goal with this introduction is to provide readers with a clear understanding of the journey ahead and the importance of our research within its domain. 1.3 Motivation The concept of tailoring treatment to individual patients is not novel. Dating back to 1946, when Austin Bradford Hill successfully demonstrated the efficacy of streptomycin in treating tuberculosis, significant methodological advancements have shaped the landscape of clinical trials (Kosorok & Laber, 2019b). However, despite these advancements, the integration of data, physician expertise, and experience, coupled with clinical judgment, often led to a medical decision-making process reliant on trial and error. This symptom-driven model, founded on empirical evidence, tends to generalize solutions and treat all patients with similar symptoms uniformly, (Banappagoudar et al., 2023; S. Liu et al., 2010a). For decades, this evidence-based medicine has been the primary approach for all patients regardless of the individual patient’s variabilities and disease conditions; this approach views patients with a single disease and multiple conditions from one lens only and has resulted in over-treatments, costly, inefficient, and complex medication protocols (Beckmann & Lew, 2016), (Pencina & Peterson, 2016). In addition, delays in disease diagnosis and uncertainties regarding treatment response significantly impact disease progression and treatment outcomes (Kosorok & Laber, 2019a), (Naithani et al., 2021), (Gopal et al., 2019). Moreover, Recent literature underscores the prevalence of medication errors and delayed treatments. Despite these figures, the analysis of fatalities resulting from incorrect medication remains inadequate (Alqenae et al., 2020). Estimates indicate that approximately 44 million individuals worldwide are currently living with dementia. As the global population continues to age, projections suggest that this number will more than triple by the year 2050. With this demographic shift, the economic burden of dementia is also expected to escalate significantly. In the United States alone, experts anticipate that the
2 annual cost associated with dementia may surpass an astonishing $600 billion by 2050 (Alzheimer’s and Dementia, 2017). In addition, the surge in consumerism, coupled with heightened demand for enhanced services, has propelled the healthcare industry to adapt (Shrank, 2017). The escalating number of individuals grappling with intricate, multiple health conditions further underscores the need for a healthcare system that goes beyond a one-size-fits-all approach (Kizer, 2001). In response, there has been a shift towards ensuring that people receive personalized care tailored to their unique circumstances, moving away from a rigid adherence to evidence-based practices alone. This evolution reflects a commitment to addressing the diverse and nuanced needs of individuals in the pursuit of optimal healthcare outcomes (König et al., 2017). Consequently, fundamental questions persist: Why do certain drugs work effectively for some patients but not others? What factors contribute to medication side effects in specific individuals? Why do some individuals develop cancer while others do not? Addressing these questions necessitates a paradigm shift toward tailoring treatment strategies to each patient's unique characteristics and circumstances(Mosavi & Santos, 2020),(Gopal et al., 2019), (Bonkhoff & Grefkes, 2022), (Ahmed et al., 2020). The rise of Precision Medicine (PM) gained significant momentum following its endorsement by Barack Obama in 2015 and has been further bolstered by extensive scientific research from institutions such as the National Library of Medicine and the National Health Service (NHS). This movement has not only showcased remarkable technological advancements but has also brought to the forefront the inherent challenges in traditional clinical decision-making protocols(Borovska et al., 2018) (Bonkhoff & Grefkes, 2022) , (Haque et al., 2020), (Mosavi et al., 2023). Despite the development of Clinical Decision Support Systems (CDSS), which have been in use since the 1980s and their advancements, CDSSs face several limitations and challenges. Issues such as data quality, data processing and management, transportability/interoperability, and integration with other hospitals or systems pose significant hurdles. Moreover, the disrupted/fragmented workflow associated with CDSSs can make them inefficient, hindering the dissemination and scaling of otherwise high-quality systems (Sutton et al., 2020), (Hee Lee & Yoon, 2021), (Bekbolatova et al., 2024),(Mosavi et al., 2023). Additionally, they often fall short in addressing requirements related to genetic/genomic and biological data, such as interpreting genetic test results and understanding their implications for family members and future generations in medical decision-making (Sadat Mosavi & Filipe Santos, 2021), (Bonkhoff & Grefkes, 2022).
3 In the early twenty-first century, the limitations of traditional clinical decision-making models became evident, exacerbated by the vast amount of healthcare big data available, which often overwhelmed human cognition in making timely decisions(Mesko, 2017a). This period also witnessed transformative changes in the healthcare landscape, driven by technological advancements such as Cognitive Computing (CC), Artificial Intelligence (AI), and Big Data Analytics(Gopal et al., 2019), (Sriram & Subrahmanian, 2020). Recognizing the inherent complexity of patient heterogeneity in treatment assessment, particularly highlighted in the late twentieth century, alongside the rise of data analytics around 2010 and subsequent advancements in AI and CC from 2015 to 2020, has emphasized the necessity for a data-driven approach. This paradigm shift prioritizes precise and timely decision-making tailored to individual variables (Watson, 2017b, 2017a). This shift aims to optimize the utilization of patient clinical data and harness the power of AI and CC to tailor precise treatment pathways (K. B. Johnson et al., 2021), (Jabbar et al., 2018). Through deep research and extensive literature studies, PM has evolved from a theoretical concept to a practical approach, to maximize treatment efficacy and patient outcomes(Tarassoli, 2019) ,(Leone et al., 2021). In 2011, the concept of PM was introduced by the National Research Council (NRC), emphasizing the tailoring of medical interventions to individual patient characteristics. This approach involves categorizing patients based on their susceptibilities to diseases and responses to treatments. PM plays a pivotal role in guiding healthcare decisions, thereby enhancing the quality of care while minimizing unnecessary tests and therapies. As highlighted by Geoffrey S. Ginsburg and Kathryn A. Phillips in 2018, this method optimizes interventions, directing resources where they yield the greatest benefits and mitigating costs and potential side effects for patients (Maglaveras et al., 2016). According to the U.S. National Library of Medicine, PM represents an emerging paradigm that considers individual differences, including genes, environment, and lifestyle, to prevent and treat specific diseases for everyone (Francis S. Collins, 2015; Mesko, 2017; Haque et al., 2020). Additionally, the U.S. National Cancer Institute defines PM as "a form of medicine that uses information about a person’s genes, proteins, and environment to prevent, diagnose, and treat disease" (Haque et al., 2020). This progressive methodology seeks to replace the antiquated “one size fits all” model with a patient-centric “patient like me” approach. Its core mission is to address issues like inefficient treatments and medical errors, ultimately reducing the burden of overtreatment and hospitalization while saving more lives(Glen H. Murata, Albuquerque, 2014). Moreover, this approach represents a pivotal shift in medical science, where the core objectives include predicting the likelihood of developing a disease, achieving precise diagnoses,
4 and optimizing the most effective treatment for individual patients (Awwalu et al., 2015). The overarching goal is to usher in a new era of healthcare that is not only more personalized but also more effective in addressing the unique needs of each patient. PM is commonly defined as a methodology that customizes treatments according to the distinct needs of individual patients. This customization is rooted in considerations such as their genetic composition, biomarkers, phenotypic traits, or psychosocial characteristics, which set apart one patient from another despite sharing highlighted clinical symptoms(König et al., 2017), (Jameson & Longo, 2015). Widely embraced definitions of PM span several dimensions. Alternatively, PM is perceived as an allencompassing, data-driven paradigm that underscores the significance of integrating diverse types of data, incorporating clinical information and other pertinent data sources. This data integration enables the categorization of patients into distinct subgroups, with the expectation that these subgroups share a common foundation of disease susceptibility and presentation. Ultimately, this data-driven approach aims to facilitate more accurate and personalized therapeutic solutions. These dual aspects of PM collectively underscore its commitment to tailoring medical care to individual patients while leveraging advanced data analysis and integration techniques for enhanced diagnosis and treatment (Grady, 2016) While a precise definition of PM may remain elusive, it is widely acknowledged that the PM approach transcends the traditional symptom-based method in clinical decision-making, emphasizing the consideration of individual variabilities. As a result, this groundbreaking concept embraces a diverse range of individual data, including biomarkers, lifestyle factors, environmental influences, and genetic information, thereby underlining the comprehensive and personalized nature of PM (Sanchez-Pinto et al., 2023), (König et al., 2017), (Pelter & Druz, 2024). Inspired by the evolving landscape of data-driven decision-making in PM,this study stands as a dedicated endeavor to contribute significantly to this dynamic protocol. Going beyond exploration, our focus is to intricately detail how data-driven insights can not only shape the framework but also drive the development of applications in PM. We envision this research as a catalyst, actively advancing the integration of data-driven methodologies, thereby fostering a transformative shift in clinical decision-making for the betterment of precision healthcare. The personal connection and motivation behind choosing this topic for this PhD journey stem from a profound experience during my childhood. At that time, my health was jeopardized due to the potential risks associated with an incorrect medical prescription. Despite being young, the memories of that challenging period, which not only impacted my health but also profoundly affected my family, have served as a powerful inspiration.
5 This personal experience ignited a passion within me to embark on an academic journey. The motivation lies in leveraging the advancements in technology and utilizing the available tools to contribute to this area. The goal is to enhance precision in clinical decision-making, ensuring that others do not face the same risks and challenges that I encountered in my early years. This journey is a testament to my commitment to making a meaningful and positive impact on healthcare practices, drawing strength from my narrative. 2.3 Research Question And Objectives In the ever-evolving landscape of healthcare, the integration of Intelligent Decision Support Systems (IDSS) has become increasingly pivotal. This research seeks to address the fundamental question: "How can data-driven insights shape the framework of Intelligent Decision Support Systems and advance optimal clinical decision-making?" To achieve this overarching goal, a set of interconnected objectives guided the investigation. Firstly, an exploration of existing IDSS frameworks will be conducted to discern their strengths and limitations. Subsequently, the research will delve into the realm of data-driven insights, investigating methodologies for extracting meaningful information from clinical data. Building upon this, the study will propose methods for seamlessly integrating these insights into the design and functionality of an IDSS. Furthermore, objective includes the dissemination of the study's outcomes within academic platforms, aiming to validate both the performance and scientific merit of our work. By sharing our findings in scholarly settings, we seek to contribute to the academic discourse, foster collaboration and invite constructive feedback from the scientific community. This commitment to transparent communication ensures the robustness and credibility of our research, aligning with the rigorous standards of scholarly inquiry. In conclusion, this research endeavors to furnish exhaustive recommendations for future works, accompanied by a thoughtful discussion on identified limitations. Through this research, we aim to contribute to the ongoing discourse surrounding the optimization of healthcare decision-support systems and their transformative potential. Based on the points, our summarized objectives are delineated as follows: a. To Investigate Current Practices Objective: To analyze existing clinical decision-making protocols in precision medicine.
12 decision-makers, leading to satisfying decisions rather than optimizing ones. Simon addressed human limitations affecting decision-making accuracy and argued for a more practical model to achieve more rational decisions (Jean-Charles Pomerol, 1997), (Kalantari, 2010),(Michael A. Eierman, Fred Niederman, 1995). Simon also made a pivotal distinction between enterprise-wide DSS and desktop DSS. An enterprise DSS accesses a data repository, making it accessible to multiple individuals, while a desktop DSS is a standalone system (Felsberger et al., 2017). Since the inception of Decision Support Systems (DSS), researchers have adopted various approaches to delve into the diverse types of DSS, each embodying distinct philosophies of support and potential. Notably, in the 1970s, Personal Decision Support Systems (PDSS) and Group Decision Support Systems (GDSS) gained prominence. PDSSs, being the earliest type of DSS, focused on addressing singular decision tasks for individual managers or independent users. However, as the 1980s unfolded, the focus shifted towards Group-oriented DSS (GDSS) and Organizational Decision Support Systems (ODSS), where multiple individuals collaboratively contribute to decision-making processes. Executive Information Systems (EIS), a data-oriented DSS providing organizational reports for management, Online Analytical Processing Systems (OLAP) for multidimensional queries, data warehousing (a repository of raw data for decision-making), and Business Intelligence (BI), encompassing applications for data collection, categorization, analysis, and presentation as DSS inputs, are additional categories within the DSS landscape(Arnott & Pervan, 2008), (Dulcic et al., 2012), (P. Bernus, P. Nemes, 1946). Furthermore, in the mid-1980s, the integration of expert systems into DSS marked a significant development. Keen and Morton (1978) identified AI as a potential support to enhance decision-making. The fusion of AI and DSS has given rise to knowledge-based decision support systems, often referred to as Intelligent DSS (IDSS). The marriage of AI and decision science emerged from a vast literature of problem-solving methods, with the advent of computers in the 1940s facilitating automated reasoning (Horvitz et al., 1988). In the 1990s, the globalized world economy prompted organizations to navigate integrated business platforms, emphasizing the generation of data to support managerial decision-making (Arnott & Pervan, 2008). As the twenty-first century unfolded, Business Intelligence (BI), communication, and knowledge management (KM) became pivotal contributors to Decision Support Systems (DSS) through web platforms, ushering in a new framework for solving organizational problems(Abubakar et al., 2019),(D. J. Power, 2000, 2002), highlighted the significance of communication driven DSS, acknowledging its role in considering alternatives beyond operational scopes, standards, and existing domains for problem-
13 solving(Fentahun Moges Kasie, Glen Bright, 2017), (Shim J. P. et al., 2002). This shift broadened the perspective, accommodating innovative approaches to decision-making outside the conventional boundaries (Asemi et al., 2011),(Felsberger et al., 2017), (Reza Kheirandish, 2019). Moreover, the evolution from traditional databases to the implementation of data warehousing and online analytical processing (OLAP), coupled with the shift from mainframe to client-server architecture, presented both an opportunity and a necessity. This transition compelled the exploration of Business Intelligence (BI) and Intelligent Decision Support System (IDSS) solutions to effectively navigate the increasing complexity of business operations (Kirs et al., 2006), (S. Liu et al., 2010b), (Shim J. P. et al., 2002). Figure 1, the literature review map, provides an overview of the evolution of Decision Support Systems (DSS) within a specific timeframe. Figure 1: Literature-Map 1960s to 2000s Adapted from (Arnott & Pervan, 2005)
14 According to Figure 2, The development phases of Decision Support Systems (DSS) from 2000 to 2019 can be summarized as follows: 2000-2005: - Integration of data warehousing and business intelligence technologies. - Emergence of web-based DSS, enabling remote access and collaboration. - Advancements in data mining techniques for predictive analytics. 2005-2010: - Growth of real-time decision support systems, facilitating faster decision-making. - Expansion of DSS into mobile platforms, enhancing accessibility. - Integration of social media data for decision support purposes. 2010-2015: - Adoption of cloud computing for DSS, enabling scalability and cost-effectiveness. - Focus on big data analytics, handling large volumes of structured and unstructured data. - Emphasis on personalized decision support, utilizing machine learning algorithms. 2015-2019: - Rise of prescriptive analytics in DSS, providing recommendations for optimal actions. - Integration of artificial intelligence and cognitive computing technologies. - Enhancement of visualization techniques for better data representation and interpretation. Throughout these phases, DSS continued to evolve to meet the changing needs of organizations, leveraging advancements in technology and data analytics to provide more powerful and effective decision-support capabilities (Delen, 2020),(Sprague, 1980), Figure 2: Literature-Map 2000s to 2019s 2000-2005 •Web-Based Integration: Transforming Decision Support 2005-2010 •Real-Time Revolution: Mobile Expansion and Social Integration 2010-2015 Cloud Empowerment: Big Data and Personalization Era 2015-2019 •AI Integration: Prescriptive Analytics and Cognitive Advancements
15 In summary, in the subsequent sections, we explore the significant evolution of Decision Support Systems (DSS) spanning the years 1980 to 2020, as depicted in Figure 3. The progression began with the inception of DSS in the 1980s, followed by the emergence of Enterprise Data Warehousing in the 1990s. The early 2000s saw the rise of real-time data warehousing, which continued to evolve until 2010. From 2010 to 2019, a notable shift occurred with the introduction of "Adaptive Business Intelligence" (ABI), challenging traditional paradigms of "Business Intelligence" (BI). Concurrently, the concept of "Business Analytics" (BA) emerged, ushering in a transformative approach to decision-making. The landscape of problem-solving, heavily influenced by analytics, has prompted a shift in terminology. Terms like "intelligence," "mining," and "discovery" are gradually being replaced by the more comprehensive term "analytics." This linguistic evolution is evident in the transition from "Business Intelligence" to the broader scope of "Business Analytics," and from "Knowledge Discovery" to the encompassing term "Data Analytics" (Delen, 2020). This reflects the evolving nature of decision support systems and their reliance on advanced analytical methodologies for nuanced insights and informed decision-making (Praful Bharadiya, 2023) Finally, in 2019, we witnessed the advancement of cognitive computing, which seamlessly integrated with the adoption of AI in decision-making processes, further enhancing the capabilities of DSS (Lytras et al., 2020) Data analysis by humans can be a labor-intensive process, consuming significant time and resources. Utilizing sophisticated cognitive systems can alleviate this burden by efficiently processing vast amounts of data. Cognitive computing presents a promising solution to mitigate the challenges associated with handling big data (S. Gupta et al., 2018).
16 Figure 3: Maturity model of DSS Adapted from (Watson, 2017b) The cognitive aspect involves the integration of cognitive computing technologies into DSS, significantly influencing its development during this period in several ways, including: - Cognitive computing involves the use of AI techniques, such as natural language processing (NLP), machine learning, and neural networks, to mimic human-like cognitive abilities in processing and analyzing data (S. Gupta et al., 2018; Lytras et al., 2020). - AI-powered DSS can interpret and understand unstructured data sources, such as text and images, enabling more comprehensive decision support capabilities. Enhanced Data Interpretation: - Cognitive DSS can analyze complex datasets more effectively, extracting meaningful insights and patterns that may not be apparent through traditional analytics methods. - AI algorithms in DSS can learn from past data and adapt to changing environments, improving decision-making accuracy over time (Behera et al., 2019). - Cognitive computing enables DSS to provide personalized decision support tailored to individual user preferences and behaviors.
17 - By analyzing user interactions and feedback, cognitive DSS can refine recommendations and insights to better meet user needs. Real-time Decision Making - Cognitive DSS can process and analyze data in real time, enabling organizations to make faster and more informed decisions(Tarafdar et al., 2017). - By continuously monitoring data streams and detecting patterns or anomalies, cognitive DSS can alert users to potential opportunities or risks in real time (Hugh J. Watson,2028). Overall, the integration of cognitive computing technologies into DSS has significantly advanced the field, empowering organizations to harness the full potential of their data for better decision-making(Baxter et al., 2008; Bini, 2018; Srivani et al., 2023). 2.1.1 Types of Decision Support Systems The taxonomy of DSS delineates a classification based on the inherent operations it performs, independent of variables such as the type of problem, function, or situation. DSS can assume varied orientations, being either completely model-oriented or data-oriented. Due to the inherent heterogeneity within the DSS category, as noted by Steven Alter in 1977, there hasn't been a universally agreed-upon taxonomy for DSS. Haettenschwiler introduced a categorization scheme, classifying DSS into passive, active, and cooperative types. Passive DSS cannot provide suggestions, while active DSS can proactively propose solutions. Cooperative DSS leverages feedback mechanisms to enhance the decision-making process. Additionally, Sprague and Carlson in 1982 characterized DSS as a system that collaborates with other components of the information system to facilitate decision-making tasks. While no specific classification system for DSS exists, Power suggests a categorization based on variables like the frequency of decision-making and the complexity of the situation, distinguishing between structured, semi-structured, and unstructured problems. Figure 4 illustrates this classification, showcasing six categories of DSS for semi-structured and unstructured decision-making situations. These categories include Communication-Driven, Data-Driven, Document-Driven, Knowledge-Driven, and Model-Driven DSS. In particular, represents a form of Group Decision Support System (GDSS) centered around facilitating communication, information sharing, and coordination among two or more individuals. An exemplary instance of this type is web conferencing, illustrating the collaborative nature of Communication-Driven
18 DSS (Stanek et al., 2014). Data-Driven Decision Support Systems (DSS) are characterized by their reliance on business analytics, leveraging substantial datasets for analysis. Within this category, Business Intelligence (BI) systems stand out, encompassing both operational and strategic functionalities (D. J. Power, 2008),(Wong & Wang, 2003). In addition, Document-Driven DSS, exemplified by Correspondence or Document Management Systems (DMS), falls under the umbrella of Knowledge Management Systems (KMS). These systems utilize document processing technologies, including retrieval, versioning, and analysis, to support the decisionmaking process. Knowledge-driven systems represent another type of DSS designed to assist managers in decision-making through recommendations and suggestions. These systems are typically structured around Artificial Intelligence and statistical techniques, shaping their architectural framework. Lastly, Model-Driven systems employ analytical tools rooted in statistical, mathematical, optimization, and simulation models to dissect and understand the situation. In essence, quantitative models play a pivotal role in shaping the functionality of these DSS. Moreover, automated decision-making is most suitable for well-structured situations and frequent decision-making scenarios. In contrast, computerized special studies are designed for rational decisionmaking, utilizing analytical processes when faced with unstructured and infrequent situations. Automated decision processes in this context are devoid of human involvement and rely on programmed algorithms, AI, or statistical and mathematical modeling for execution (D. Power, 2017), (Felsberger et al., 2017), (Prakash & Sarkar, 2015), (Reza Kheirandish, 2019),(Richard & Averweg, n.d.).
19 Figure 4: Types of DSS Adapted from (Ciara Heavin, Daniel J. Power,2017) 2.1.2 Intelligent Decision Support Systems The architecture of an IDSS follows a structured process encompassing input, processing, and output, all integrated with a feedback loop. Input sources may include data, knowledge, advice, algorithms, or models. Processing involves tasks such as forecasting, generating recommendations, or providing explanations based on the organized input, leading to the generation of output—the outcome of the processing. Importantly, the system utilizes this output as the new input for subsequent analyses (PHILLIPS-WREN, 2012). In essence, an intelligent system derives the most logical actions by thoroughly analyzing data and available information. Expert systems, for instance, represent a category of intelligent systems that rely on a knowledge base as input and store rules for decision-making. Additionally, IDSS leverages data mining techniques to acquire intelligence and knowledge critical to the decision-making process. The application of data mining technology serves as a knowledge discovery tool, extracting trends, patterns,
20 and insights from databases to inform decision-making through recommendations and predictions (Aristodemou & Tietze, 2018), (Phillips-Wren, 2008), Jean-(Jean-Charles Pomerol, 1997), S. Liu, Duffy, (S. Liu et al., 2010a), (Skulimowski, 2016). IDSS exemplifies intelligent behavior, encompassing self-learning from experience, swift and adept responses to novel situations, knowledge processing, operational functions based on probabilistic data, value generation, filtering, reasoning for problem-solving, and the ability to generate answers with incomplete input variables(Kuceba & Kieltyka, 2009), (PHILLIPS-WREN, 2012). The multifaceted nature of IDSS has led researchers to conceptualize it in various ways, each emphasizing specific aspects. For instance, the concept of an Intelligent, Interactive, and Integrated Decision Support System (3IDSS) is particularly beneficial for large-scale enterprises dealing with economic and social development strategies. In scenarios where high levels of management involvement and intricate environmental complexities are present, the integration of all influential factors becomes crucial. In such cases, a DSS with a heightened level of machine-human interaction proves to be essential. Another notable example is the Intelligent Decision Support System based on Natural Language Understanding. This variant of IDSS operates on natural language understanding, addressing scenarios where a substantial amount of unstructured information is presented in text format. Consequently, this type of IDSS employs intelligent techniques to navigate and extract insights from natural language text efficiently. Furthermore,(Zhou et al., 2008) the IDSS based on Knowledge Discovery (IDSSKD) plays a pivotal role in facilitating decision-making processes in commerce and finance. This form of intelligent support relies on predictive analytics and in-depth analysis to enhance decision-making outcomes (Zhou, Yang, Li, & Chen, 2008). 2.2 Precision Medicine as a New Approach in Clinical Decision Making 2.2.1 Precision Medicine: an Evolving Paradigm in Clinical Decision-Making In 2011, the National Research Council, through its Toward Precision Medicine initiative, elucidated precision medicine as the personalized tailoring of medical interventions to the unique characteristics of each patient. This involves classifying individuals into distinct subpopulations based on varying susceptibilities to specific diseases or responses to treatments. The essence of precision medicine lies in its ability to guide healthcare decisions, directing them toward the most effective treatments for individual patients. Doing so, not only enhances the overall quality of care but also diminishes the need for
21 unnecessary diagnostic tests and therapies. This strategic approach, as highlighted by Geoffrey S. Ginsburg and Kathryn A. Phillips in 2018, ensures that preventative or therapeutic interventions are concentrated where they will be most beneficial, thus sparing both expenses and potential side effects for those who would not derive significant benefits. On January 20, 2015, former President Barack Obama heralded a groundbreaking advancement in healthcare with the introduction of the "Precision Medicine" initiative. This initiative, a transformative leap in medical practice, was envisioned to significantly enhance public health. President Obama emphasized its potential, stating, "Tonight, I'm launching a new Precision Medicine Initiative to bring us closer to curing diseases like cancer and diabetes and to give all of us access to the personalized information we need to keep ourselves and our families healthier" (Maglaveras et al., 2016). The innovative approach to medical practice, as outlined by Francis S. Collins in 2015, encompasses two key components: a short-term focus on cancer diseases and a long-term perspective addressing a spectrum of other illnesses. According to the U.S. National Library of Medicine, "precision medicine" is an emerging paradigm that accounts for individual differences, including genes, environment, and lifestyle, to prevent and treat specific diseases for everyone (Mesko, 2017b), (Francis S. Collins, 2015),(Haque et al., 2020b). Additionally, the U.S. National Cancer Institute defines Precision Medicine as "a form of medicine that uses information about a person’s genes, proteins, and environment to prevent, diagnose, and treat disease" (Haque et al., 2020b). This progressive methodology seeks to replace the antiquated “one size fits all” model with a patientcentric “patient like me” approach. Its core mission is to address issues like inefficient treatments and medical errors, ultimately reducing the burden of overtreatment and hospitalization while saving more lives (Glen H. Murata, Albuquerque, 2014). Moreover, this approach represents a pivotal shift in medical science, where the core objectives include predicting the likelihood of developing a disease, achieving precise diagnoses, and optimizing the most effective treatment for individual patients(Awwalu et al., 2015). The overarching goal is to usher in a new era of healthcare that is not only more personalized but also more effective in addressing the unique needs of each patient. For over a decade, Hood and colleagues have championed the concept of P4 Medicine—Personalized, Preventive, Predictive, and Participatory (Chen & Snyder, 2013), (Mesko, 2017c).This forward-looking paradigm encompasses "Predictive" medicine, leveraging longitudinal biological information for comprehensive health insights. The "Preventive" aspect focuses on early disease detection, contributing significantly to population health. "Personalized Medicine," a widely used term until 2015, saw a transition
28 Figure 8: Healthcare Analytics To underscore the implications of each analytics approach on clinical decision-making, we turn to the succinct and practical model of rational decision-making known as “Intelligence-Design-Choice” by Herbert Simon. Simon introduced the concept of bounded rationality to address the reality that human decision-making often falls short of achieving truly optimal outcomes. He highlighted that human cognition is inherently limited, making it impossible for individuals to consider and analyze every conceivable alternative and its consequences during problem-solving. Consequently, decisions may be influenced by individual preferences rather than being strictly rational. In response to these limitations, Simon devised a systematic and logical framework, initially consisting of “Intelligence-Design-Choice” to guide rational decision-making. Later, he added an “Implementation” phase as a fourth step, incorporating monitoring and feedback from each preceding phase to validate, verify, and refine the decision-making process (Delen, 2020), (Woodside et al., 2016). The “Intelligence” phase encompasses activities such as information gathering and scanning. This stage also involves the critical task of problem identification and definition. Moving to the “Design” phase, the focus shifts to conceptualizing the problem, defining assumptions, generating alternative solutions, and Data Analytics Descriptive & Diagnostic observe & analyse Ex: vital condition Information Focused Predictive Predict Ex: patient at risk Information Focused Prescriptive recommend Ex: the best possible treatment Decision Focused Clinical Data
29 constructing models. Finally, the “Choice” phase pertains to goal attainment, involving the evaluation and selection of the best possible course of action (Simon, 1959; Woodside et al., 2016). Considering the role of analytics in rational decision-making, it becomes evident that each type of analytics corresponds to specific phases within this framework. Descriptive and diagnostic analytics are instrumental in data collection and analysis, problem definition, and knowledge extraction, aligning with the “Intelligence” phase. Predictive analytics, which involves formulating and generating alternatives, is well-suited to the “Design” phase, given its capacity to provide insights and options (Ramesh Sharda, Dursun Delen, 2014). Additionally, because predictive analytics offers informative capabilities, it also aligns with the “Intelligence” phase. Lastly, prescriptive analytics, with its focus on providing actionable recommendations and achieving optimal outcomes, naturally corresponds to the “Choice” phase. In summary, each analytics approach serves distinct functions within the decision-making process, contributing to a more informed and rational approach to clinical decision-making(Mosavi & Santos, 2020; Palanisamy & Thirunavukarasu, 2019). d. Other influential factors references A considerable number of individuals encounter the inefficacy of medications in their daily treatment, emphasizing the imperative need for a more precise approach(Schork, 2015). The rising aging population, coupled with the diverse responses of patients to treatments, imposes global economic (K. B. Johnson et al., 2021; McDonald et al., 2000). As expectations for enhanced care and service quality rise, there is a simultaneous decline in the economic capacity to deliver high-quality services, posing a formidable challenge for healthcare providers. In addition, rapid progress in genomics, particularly with techniques like whole genome sequencing, has analyzed individual genetic information more accessible and affordable. This progress enhances our understanding of the genetic basis of various diseases and opens avenues for tailored treatments. Pharmaceutical companies are increasingly developing targeted therapies designed for individuals or specific patient subgroups based on their genetic profiles. This approach has shown promise in cancer treatment and beyond. Another point of view is patient-centered care, precision medicine places a strong emphasis on patientcentric care, integrating individual preferences, values, and goals into the treatment decision-making process. Its application extends across medical specialties, including oncology, cardiology, psychology, and rare diseases, with potential applications in numerous other areas of medicine(Kosorok & Laber, 2019a; Latimer et al., 2017).
30 Furthermore, governments and healthcare organizations globally are investing in precision medicine research and infrastructure to foster its development and adoption. The growing emphasis on educating healthcare providers, researchers, and the public about the potential benefits of precision medicine and its integration into clinical practice is evident. As our understanding of genetics, data analysis, and medical technology expands, the concept of precision medicine is poised to evolve further. This evolution holds the promise of offering more targeted and effective treatments for a wide range of diseases, marking a significant shift toward personalized and data-driven healthcare, a trend expected to persist in the years to come(Milewa, 2009). 2.2.3 IDSS4PM; Gaps and Limitations The evolution of intelligent decision support systems for precision medicine is a dynamic and rapidly advancing field. Despite extensive research at the application level, numerous challenges and scientific gaps persist in the development of data-driven clinical decision support systems. This section delves into the key obstacles and gaps encountered in this progressive domain which has been summarised In Figure 9.
31 Figure 9:IDSS4PM; gaps and limitations IDSS4PM; gaps & limitations Data-Driven aspects data processing & managment Diverse data resources Heterogeneity of Data data integration & interopratability data quality & standardization Infrequent data generation Dimentionality of data Accessability to Data Data Privacy and Security ML & Computing aspcts Other aspects Interdisciplinary Collaboration Integration into Healthcare Workflow Ethical and Regulatory Issues Clinical Validation Patient Engagement and Education Limited Knowledge and Research Gaps Cost and Reimbursement Scalability
32 a. Data processing, and management: Patient data in precision medicine is highly heterogeneous, including genetic, clinical, lifestyle, and imaging data(Ahmed et al., 2020). In the realm of digitized healthcare platforms, a prevalent scenario involves the continuous monitoring and storage of various variables of patients, resulting in a vast amount of collected data. The integration of data is crucial for studying its correlation with other information generated during a patient's treatment journey and other clinical aspects. Ensuring that these data sources can work together seamlessly is essential for precision medicine but remains a technical challenge. Moreover, the presence of outliers and abnormal data introduces bias, necessitating the filtration of such data from modeling and study scenarios (Chang et al., 2019; Fenning et al., 2019). Despite the promising strides in Big Data analysis in clinical research, marked by a growing number of peer-reviewed articles, addressing these challenges remains limited. A concerted effort is needed to validate the knowledge extracted from clinical Big Data and seamlessly implement it in clinical practice(Carra et al., 2020). Numerous studies have contributed fragmented solutions to address these challenges. Table 2 highlights a few major efforts that present promising frameworks. For instance, the "attention scores" technique for feature importance in time series clinical data is a complex yet applicable method suitable for nonlinear data(N. Johnson et al., 2021). Other research endeavors leverage summaries of patient time series data from the ICU, focusing on the early prediction of in-hospital mortality. This approach involves static observations and physiological data, including labs and vital signs, aggregated based on hourly circumstances, thereby addressing data extraction and integration within the context of data aggregation (N. Johnson et al., 2021). Another significant limitation faced by clinical data for processing pertains to storage and computing, particularly concerning high-frequency data such as physiological indicators. The "Electron" framework has emerged as a solution designed to store and analyze longitudinal physiologic monitoring data (McPadden et al., 2019). Moreover, the Topological Data Analysis (TDA) approach proves effective for large-scale datasets, utilizing algebraic topology to analyze Big Data by reducing dimensionality, especially for geometric representations to extract patterns and gain insights into them. To cope with the velocity of data, the introduction of "anytime algorithms" to learn from data streaming has proven useful for time series data, contingent upon the computation capacity they can handle. Additionally, addressing heterogeneous data (data variety), the Generalized Nonnegative Matrix Tri-Factorization (GNMTF) framework is an efficient solution for data integration, although its competency complexity increases with
33 the number of data types to be integrated(Gligorijević et al., 2016), (Hulsen et al., 2019), (N. Johnson et al., 2021). As discussed above, to overcome these limitations, researchers and practitioners have delved into diverse solutions, where those existing approaches have provided solutions to manage specific challenges such as feature importance, feature selection, dimensionality reduction, and addressing issues related to data processing tasks. Table 2: Techniques to deal with limitations associated with data processing Technique Focused area Limitations attention scores feature importance feature selection, time series data, nonlinear features post-hoc explanation techniques Modeling and features relations Interpretationblack box summaries of patient time series data Data extractiontime-series clinical data vital signs and lab data Filtering outliers Data cleaning Time series datavital signs Electron Data extraction Physiologic Signal Monitoring and Analysis Topological Data Analysis (TDAs) Dimensionality Reduction anytime algorithms Velocity learn from streaming data GNMTF Variety Complex when data types increase The other approaches include the implementation of standardized data formats, centralization of data, utilization of ontology and semantic interoperability, adoption of indexing techniques, application of Natural Language Processing (NLP), and exploration of blockchain technology. In this regard, it’s important to note that the effectiveness of these solutions may vary depending on the specific context and the nature of the clinical data in terms of the characteristics of the data sources and the objectives of the integration. Additionally, each way to solve those challenges carries limitations and considerations. i. Standardized data format: Implementing standardized data formats and coding systems can enhance interoperability and facilitate the integration of clinical data from different sources. The use of standardized vocabularies, such as SNOMED or LOINC, helps in achieving consistency. However, achieving universal adoption of standardized formats can be challenging. Furthermore, variations in the implementation and interpretation of standards may persist, leading to interoperability issues(Martínez-García & Hernández-Lemus, 2022). ii. Ontology and Semantic Interoperability: Developing ontologies that define relationships between different clinical concepts can improve semantic interoperability. This approach enables a more
34 comprehensive understanding of the data, facilitating the integration of events across diverse sources (Kraus et al., 2018). iii. Master Patient Index (MPI): Implementing an MPI helps in uniquely identifying and linking patient records across various systems. This is crucial for integrating clinical events related to the same individual, even if the data comes from different sources. The limitation associated with this method is Building and maintaining a reliable MPI requires accurate patient identification, which can be difficult in cases of variations in patient data or discrepancies. Privacy concerns and data security are also critical issues(Lintz, 2020). iv. Data Integration Platforms: Although, utilizing data integration platforms and tools can streamline the process of combining data from various sources. Complex data integration processes may require skilled personnel and resources. Handling heterogeneous data formats and ensuring data quality during the integration process still has remained a challenge. These platforms often include features for data cleansing, transformation, and loading (ETL) to ensure data quality (Martínez-García & Hernández-Lemus, 2022). v. Data Warehousing: Creating a centralized data warehouse that consolidates information from disparate sources can provide a unified platform for analysis. Data warehouses are designed to handle diverse data structures and can offer a consolidated view of clinical events. This solution carries major challenges and limitations in terms of in handling real-time data updates and ensuring data quality over time. Because developing and maintaining a comprehensive data warehouse can be resource-intensive (Bae et al., 2015; Diouf et al., 2018; Kherdekar & Metkewar, 2016; Khnaisser et al., 2015; Nacional, 2003; Nakache, 2003; Yang et al., 2024) vi. Blockchain Technology: Some studies explore the use of blockchain to create a secure and transparent environment for sharing and integrating clinical data. Blockchain can enhance data integrity and traceability. However, While blockchain enhances security and transparency, its scalability for handling large volumes of clinical data is a concern. Regulatory and legal challenges, as well as potential resistance to adopting blockchain in healthcare, are also limitations(Al-Nbhany et al., 2024; Hasan et al., 2022; Karafiloski & Mishev, 2017; Velmovitsky et al., 2021)
35 vii. Machine Learning aspects: Applying machine learning algorithms and NLP techniques can aid in the identification and extraction of relevant information from unstructured clinical text. This can contribute to a more comprehensive view of patient records The effectiveness of such techniques is dependent on the quality of input data which is a significant limitation. NLP models may struggle with understanding context and may require manual validation. Generalizability across different healthcare settings and languages can be a challenge(Bzdok & MeyerLindenberg, 2018; Naik et al., 2023; Quazi, 2022; Republic Of Korea AI Strategy, 2019; Wilkinson et al., 2020). In addition, overfitting and underfitting are common challenges in ML. Overfitting occurs when a model learns to fit noise in training data, leading to poor performance on new data. Conversely, underfitting arises when a model is too simplistic and fails to capture patterns in the data. These issues can be addressed by adjusting training data or optimizing model parameters. Successful model development requires large datasets, informatics expertise, domain knowledge, and proper validation. Collaboration between medical and informatics experts is crucial to effectively utilize big data in healthcare (Tran et al., 2021). b. Data Quality and Standardization: ensuring data quality verification is a pivotal step in the data processing journey. The quality of data is significantly influenced by key factors, including the clinical team's assessment of the patient's condition, potential misinterpretations of original documents, and errors during data entry(Brown, 2016). Additionally, medical waveforms, such as electrocardiograms and electroencephalograms, commonly employed in physiological examinations, may introduce random noise and gaps (Khadanga et al., 2019). Addressing missing values is another crucial aspect to be handled during data cleaning and preparation(Adiba et al., 2021). Thus, Lack of standardized protocols for data collection and quality control across different healthcare systems and institutions. c. Data Privacy and Security: Precision medicine relies on large amounts of sensitive patient data. Protecting patient privacy and ensuring data security while making this information accessible for research and treatment is a significant challenge. Concerns about data breaches, unauthorized access, and patient consent may limit the availability of comprehensive datasets for decision support(Sadat Mosavi & Filipe Santos, 2021). d. Ethical and Regulatory Issues:
36 The ethical use of patient data, informed consent, and the potential for discrimination based on genetic information are all important concerns. Regulatory frameworks must evolve to address these ethical issues. Therefore, ensuring that decision support systems are ethically designed and implemented can be challenging, and a lack of ethical guidelines may hinder progress (Maceachern & Forkert, 2021). e. Interdisciplinary Collaboration: Precision medicine requires collaboration between healthcare professionals, data scientists, bioinformaticians, and information technology experts. Thus, Gaps in interdisciplinary communication and collaboration may impede the development and implementation of effective decision-support systems (Tran et al., 2021). f. Scalability: Precision medicine solutions need to be scalable to handle large and diverse datasets. Therefore, Lack of scalable infrastructure and resources may limit the ability to analyze and process data efficiently. g. Longitudinal Data Challenges: Precision medicine often requires longitudinal data to track changes in health over time. The limited availability of long-term, comprehensive datasets may hinder the development of accurate predictive models. h. Clinical Validation: Transitioning from research findings to clinically validated and actionable insights can be slow and challenging. Furthermore, the maturity of existing clinical practices in this domain requires careful consideration. The establishment of new policies, regulations, and collaborative pipelines between stakeholders is essential to expedite the evolution of PM. Successful and valid projects on a larger scale have a positive impact not only on overall performance quality but also indirectly contribute to enhancing best practices, particularly in problem-solving aspects associated with technical areas such as data processing (Blasimme, Fadda, Schneider, & Vayena, 2018). Robust clinical trials and validation studies are required to prove the effectiveness of precision medicine approaches. i. Integration into Healthcare Workflow: Integrating precision medicine into routine clinical practice can be complex. Healthcare providers may need training, and electronic health record systems must accommodate new data sources and decisionsupport tools (Wagner, 2019) (Ahmed et al., 2020).
37 In addition to the data-driven aspects, other aspects impacted the successful development of such a framework such as: a. Access to High-Quality Genomic and Molecular Data: Many patient populations, especially in underserved areas, may lack access to the necessary genetic testing and data. Disparities in access can limit the application of PM. b. Cost and Reimbursement: Precision medicine often involves expensive genetic tests and targeted therapies. Ensuring that these are affordable and that reimbursement models support precision medicine can be difficult(Wagner, 2019). c. Limited Knowledge and Research Gaps: There are still many gaps in our understanding of the genetic basis of many diseases. Research is ongoing to identify new biomarkers and therapeutic targets. d. Rare and Undiagnosed Diseases: Precision medicine is particularly challenging for rare and undiagnosed diseases, where limited data and treatment options exist. e. Patient Engagement and Education: Patients need to be educated about precision medicine and engaged in the decision-making process. Informed consent and understanding the implications of genetic testing and treatments are critical. Addressing these challenges and scientific gaps requires a collaborative effort involving researchers, healthcare providers, policymakers, and patients. Ongoing research and innovation are essential to advancing precision medicine and realizing its full potential in improving patient outcomes and healthcare delivery (Duffy, 2016), (X. Liu et al., 2019),(Cazzola et al., 2017),(Sadat Mosavi & Filipe Santos, 2021).
44 3.4 Literature Review Strategy Summarized in Figure 13, in this study, we conducted an extensive search using key terms such as "Intelligent Decision Support," "Clinical Decision Making," "Precision Medicine," "AI applied in healthcare," and "Optimal Decision Making." We utilized various libraries including Scopus, ScienceDirect, and Google Scholar. The sources of papers were explored across two main areas: Information Systems and Technology, and Decision Support and Health Informatics. Additionally, we delved into interdisciplinary sources from the healthcare domain, including biotechnology, medicine, annual reviews, and statistics, spanning the period from 1997 to 2024. It's noteworthy that foundational concepts such as decision science and theoretical aspects have been extensively cited and date back further in time. Figure 13: Literature review strategy Year: from 1997 to 2024. Source •Scopus •Google Scholar •ScienceDirect KeyWords •Intelligent Decsion Support •Clinical decision making •Precision Medicine •AI applied in healthcare •Optimal decision making •Data processing and management in healthcare Type of source •Journals ;hight quality •Journal of Information Technology, applied clinical informatics, Procedia Computer Science, studies in Health Technology and informatics and annual Review of Statistics and Its ApplicationJournal of Management Information Systems, International Journal on Artificial Intelligence Tools, IEEE International Conference on Computational Intelligence and Computing Research, etc) •Conference papers;core A, B,C •Books
45 4. DATA-DRIVEN PSYCHOTHERAPY DECISION SUPPORT (DD_PT_DS) By employing data mining techniques, the study seeks to unveil hidden patterns, relationships, and trends within the data, thereby advancing the comprehension of therapeutic alliance and potentially guiding therapeutic practices. Our investigation centered on two primary questions: 1. How can analytical insight be effectively applied to investigate and understand the therapy factors influencing therapeutic alliance? 2. How can diverse datasets encompassing physiological and psychological factors from both clients and therapists be utilized to accurately predict therapeutic alliance? the pivotal components of the project's framework encompass a diverse range of physiological and psychological variables associated with both clients and therapists. The process initiates with knowledge discovery, manifested in the form of interconnected clinical events. In this phase, we conduct a comprehensive analysis, presenting insights that unveil patterns and relationships within an integrated platform. Following knowledge discovery, predictive modeling takes center stage. Through the adoption of optimization techniques, we rank algorithms to pinpoint the most effective one. In this pivotal phase, we meticulously identify the best-performing elements, and consequently, the most significant predictors emerge. This insight contributes profoundly to the interpretation of results, influencing various aspects of tailoring treatment. By intricately considering patient data, clinical backgrounds, and the information about their therapists, we achieve a holistic approach to treatment customization. 4.1 Comprehensive Exploration of Dataset The sample included 18 women and 8 men, aged between 18 and 52 years. The therapists involved in the study comprised 5 women and 1 man, with clinical experience spanning from 4 to 22 years. The initial dataset comprised 24,525 rows and 45 variables, featuring 13,676 missing cells. Figure 14 provides a detailed description of each variable. Unique identifiers for clients and therapists are represented by the client’s ID and therapist’s ID, with six unique therapist IDs and twenty-two unique client IDs. Each client was exclusively assigned to one therapist for all therapy sessions, while therapists could attend to different clients. The variable “Sex” was categorized as male (sex = 0) or female (sex = 1). The “Diagnosis” variable indicated whether the record was associated with anxiety (diagnostic = 0) or depression (diagnostic = 1).
46 The “Outcome” variable identified the result of each therapy process, categorized as poor (outcome = 0) or good (outcome = 1). Additionally, the “Termination” variable indicated whether the client experienced dropout (termination = 0) or completion (termination = 1). The number of therapy sessions is denoted as “Session,” fragmented into epochs with a fixed period of one minute. The “Time” variable represents the period in seconds, approximately equal to sixty epochs/session numbers. The value of the “Condition” variable is constant as “1,” indicating undertreatment status, and “0,” representing the baseline before the session. Biological variables for both clients and therapists included Heart Rate (mean, mean_baseline, SD_baseline, standardized) and EDA (mean, mean_baseline, SD_baseline, standardized). The values of EDA and HR reported in the context of _baseline, SD baseline, and standardized are repetitive for different epochs within the same session at baseline. However, the HR (mean) and EDA (mean) values differ in each epoch, referring to the in-session period. The WAI-total score which will be discussed as “WAI” in technical machine learning represents the value of TA for the client and therapist (client’s WAI, and therapist’s WAI, respectively). The therapeutic alliance was assessed using the Portuguese version of the Working Alliance InventoryShort Revised (WAI-SR). This inventory evaluates the quality of the therapeutic alliance across three dimensions: (1) agreement on tasks, (2) agreement on goals, and (3) development of a bond. Each dimension contributes to the total score of the WAI-SR. The WAI-SR (client) comprises 12 items rated on a 5-point Likert scale (1-5), with higher scores indicating a stronger alliance (total scores range from 12 to 60). Similarly, the therapist's WAI-SR consists of 10 items on the same Likert scale, with total scores ranging from 10 to 50. In this study, special care was taken to prioritize privacy and ethical concerns. To address these issues, the data underwent rigorous de-identification and coding processes to ensure complete anonymity and prevent any association with real patients.
47 Figure 14: Data exploration (Psychotherapy) Dataset of Psychotherapy 24,525 Rows 45 variables Demographics **id identification number **sex man=0/woman= 1 Psychological variables diagnostic anxiety=0, depression=1 outcome poor=0, good=1 termination dropout=0, completed=1 *session therapy session number condition baseline=0, under treatment=1 *epoch 1 minuetes time interval *time time duration in sec **Physiological Variables Heart Rate mean value mean baseline SD baseline standardized EDA mean mean baseline sd baseline standardized **WAI total value of TA
48 4.2 Analytical Insight Through Interrelated Clinical Events In this phase, to comprehend the intricate relationship within the data and extract meaningful patterns, we conducted a comprehensive analysis. We systematically observed and analyzed three significant relationships: Therapeutic Alliance (TA) categorized by session, gender, and individual IDs (specific therapist and client). Additionally, we scrutinized the distribution of the number of therapy sessions attended by clients. As a concluding step, we delved into the exploration of potential linear relationships between Heart Rate (HR) and Electrodermal Activity (EDA) with Therapeutic Alliance (TA). 4.2.1 Therapeutic Alliance by Session-Client In Figure 15, we illustrate the correlation between the number of therapy sessions and Therapeutic Alliance (TA) for clients who completed the therapy (termination = 1). Each data point on the scatter chart represents the client ID (legend), session number (x-axis), and the corresponding TA value for that specific session (yaxis). Examining the scatter chart reveals variations in TA values among clients in initial sessions, with some displaying high TA values, while others, after more sessions, exhibit lower TA values. For instance, client ID 20 had a TA value of 44 in session 2, which decreased to 38 by session 18. Notably, client ID 23 consistently demonstrated the maximum TA value (60) across all therapy sessions, while client ID 13 exhibited an upward trend in WAI with an increasing number of sessions.
49 Figure 15: Relationship between session and WAI score (TA) for clients; termination=1 Figure 16 presents a parallel analysis focusing on clients who terminated therapy prematurely (termination = 0). As depicted in the scatter chart, only two clients opted to discontinue therapy – one with 8 sessions (ID 7) and the other with 11 sessions (ID 9).
50 Figure 16: Relationship between session and WAI score (TA) for clients; termination=0 4.2.2 Therapeutic Alliance by Session-Therapist To explore the correlation between the therapist's Therapeutic Alliance (TA) and the number of therapy sessions, we factored in the client ID, recognizing that each therapist may have multiple clients. Figure 17 illustrates a scatter chart featuring the therapist's ID, the count of clients, and the respective session numbers.
51 Figure 17: Relationship between session and WAI score (TA) for therapists; termination=1 As depicted in Figure 18, two therapists (ID = 14, 31) have conducted sessions with clients who eventually dropped out.
52 Figure 18: Relationship between session and WAI score (TA) for therapists; termination=0 4.2.3 Therapeutic Alliance by Sex-Client In Figure 19, we present the average value and standard deviation of Therapeutic Alliance (TA) categorized by the client's sex ("client_sex" equals zero for male clients and one for female clients). Through this analysis, we observed that out of the twenty-two clients, seven were male (client ID = 3, 5, 12, 19, 20, 22, 23), and the remaining fifteen were female (client ID = 2, 4, 7, 9, 10, 11, 13, 14, 15, 16, 17, 18, 21, 25, 26). In Figure 6, the mean TA value for female clients was 51.74, while for male clients, it was 51.60. Additionally, the standard deviation of TA for female clients equaled 7.38, and for male clients, it was 7.10.
53 Figure 19: Average of WAI score (TA) by sex – client 4.2.4 Therapeutic Alliance by Sex-Therapist In the TA by sex analysis (Figure 20), it's notable that only therapist ID=19 is male (therapist sex = 0), and the average TA value associated with this therapist is 43.70. Contrarily, the average TA value for other female therapists (therapist id = 3, 5, 14, 21, 31) stands at 45.07. Hence, female therapists exhibit a higher TA value compared to their male counterparts. Referring to Figure 6, the standard deviation of TA for therapist ID equal to 19 is 4.33, whereas for female therapists, it is 4.17.
60 In Figure 30, we delved into the connection between Therapeutic Alliance (TA) and the average Electrodermal Activity (EDA) for clients with termination status set to one. The x-axis signifies the TA value, while the y-axis illustrates the average EDA for each client across therapy sessions. Session numbers are elegantly introduced through colorful triangles in the legend, where the smallest triangle signifies session one, and the largest corresponds to session eighteen. To offer a frame of reference, a vertical average line for TA at 51.64 and a horizontal average line for EDA at 5.16 are incorporated, enriching the visual interpretation of the correlation dynamics. Figure 30: Relationship between EDA and WAI score (TA) for clients; termination=0 4.2.12 Investigating The Relation of EDA of the Therapeutic Alliance - Therapist Exploring the association between Electrodermal Activity (EDA) and Therapeutic Alliance (TA) for therapists is visualized in Figure 31. The highlighted data point spotlights therapist ID = 21, assigned to client ID = 14 in session 16, with an average EDA value of 13.65 and a corresponding TA value of 43. This focused illustration provides insight into the interplay between EDA and TA for therapists across specific therapy sessions. .
61 Figure 31: Relationship between EDA and WAI score (TA) for therapists; termination=1 In Figure 32, the scatter chart replicates the analysis, this time with a termination status equal to 0. A specific data point is highlighted, corresponding to session 10 for therapist ID = 14 and client ID = 9. In this instance, the therapist's average Electrodermal Activity (EDA) is 5.44, while the Therapeutic Alliance (TA) value stands at 38. This focused example provides a glimpse into the dynamics of EDA and TA in sessions where termination did not occur. Figure 32: Relationship between EDA and WAI score (TA) for therapists; termination=0 4.2.13 Results and Discussion Performing analytical insights contributes to tailoring treatment and therapy plans for each client by understanding the intricate links between indicators and the relationship between physiological and psychological factors to TA. Such analysis and data observation are vital for monitoring client and therapist
62 status, enabling early detection, and enhancing patient engagement, ultimately leading to improved outcomes. Additionally, conducting statistical analysis provides a better understanding of data quality and boundaries, while visualization techniques offer insights into variable relationships and pattern identification, aligning with study objectives. Data processing, including grouping and aggregation at the session level, along with integration based on therapist-client session linkage, supports this investigation by preparing effective data for modeling. This facilitates and advances predictive analysis for the next phase. Figure 33 provides a summary of the contribution of this phase (Analytical Insight) to Intelligent Decision Support Systems for Precision Medicine (IDSS4PM) from a data-driven perspective. It also outlines its application in precision medicine, bearing in mind that the project is an exploratory investigation. Figure 33:Contributions IDSS4PM; Analytical Insights Contribuations Data Driven prespective Data analysis, processing & integration Interrelating clinical events Pioneering Analytical Insights Identify patterns Interdiciplinary collaboration Application in PM Detection and Prevention Continuous Patient Monitoring Treatment Effectiveness Precise and customosed treatment Improved Patient Engagement
63 4.3 Predictive Modeling for Therapeutic Alliance 4.3.1 Comprehensive Data Preparation For Modelling This phase encompasses a series of tasks such as feature selection, feature construction, data cleaning, and formatting. In the data preparation for modeling, we systematically eliminated records with missing values. Employing Pearson correlation analysis (denoted as "Pearson’s r"), we quantified the linear relationship between variables to discern correlation coefficients among features. The results of the correlation analysis are presented in Figure 34. This step is pivotal for selecting influential predictors and excluding features with a negative impact on modeling results. Recognizing that a high correlation between features is redundant and doesn't enhance model accuracy, we excluded those with less influence on predicting the target variable (WAI-SR score). The constant feature "condition" with a value of 1 was also removed. The variable "ID," deemed non-contributory and potentially causing overfitting, was likewise excluded. Additionally, noting the substantial skewness (Y1 = 20.54) in Client_HR (Standardized), we omitted this feature from modeling. To enhance model efficiency, we adjusted variable types to the most suitable ones, categorizing some variables and converting others to float type. Figure 34 illustrates the correlation matrix. By considering Pearson correlation coefficients with a range of 0.1 <= x <= -0.1, we identified effective indicators for predicting the WAI value. Furthermore, a correlation level of 0.4 <= x <= -0.4 was employed to examine relationships among variables. This comprehensive approach ensures a refined dataset, optimized for subsequent modeling endeavours.
64 Figure 34: Pearson Correlation Coefficients Matrix Table 3 highlights that "WAI_Therapist" stands out as the most influential indicator with a positive impact on predicting the target, boasting a correlation coefficient score of 0.44. Additionally, variables such as "Session," "Outcome," "HR-Client(SD_Baseline)," "EDA_Client(mean_Baseline)," "Diagnostic,"
65 "EDA_Therapist(SD_Baseline)," "Therapist_sex," and "EDA_Client(mean)" were identified as other effective predictors for "WAI-Client," presented as correlated variables to "WAI_Client." Moreover, "WAI_Client" emerged as the most potent variable in predicting the value of "WAI_Therapist," holding a correlation of 0.44. Several other indicators, including "Session," "Diagnostic," "Outcome," "Termination," "HR_Therapist(mean_Baseline)," "HR_Client(mean)," "HR_Client(mean_Baseline)," and "HR_Therapist(mean)," were acknowledged as influential factors. Interestingly, EDA did not exhibit a significant association with "WAI_Therapist," while Heart Rate emerged as a crucial variable. As per Table 3, the therapist's HR positively influenced WAI, while the client's HR showed an inverse relation to "WAI_Therapist." Furthermore, "Diagnostic" and "Termination" exhibited a stronger association with "WAI_Therapist" than with "WAI_Client." Table 3: Pearson Correlation Between TA (WAI Scores) And Predictors # Predictors for WAI_Client score Predictors for WAI_Therapist score 1 WAI therapist 0.45 WAI-Client 0.45 2 Session 0.30 Session 0.39 3 Outcome 0.28 Diagnostic 0.33 4 HR-Client(SD_Baseline) (-) 0.18 Outcome 031 5 EDA_Client(mean_Baseline) 0.17 Termination 0.23 6 Diagnostic 0.15 HR_Therapist(mean_Baseline) 0.19 7 EDA_Therapist(SD_Baseline) (-) 0.11 HR_Client(mean) (-) 0.18 8 Therapist_sex (-) 0.12 HR_Client(mean_Baseline) (-) 0.16 9 EDA_Client(mean) 0.10 HR_Therapist(mean) 0.16 In the process of selecting the final predictors, we meticulously examined the correlation coefficients among the predictors. The details of the included and excluded variables for predicting "WAI_Client" are presented in Table 4. Notably, due to the robust correlation observed between "session" and "WAI_Therapist" (0.39), and between "Diagnostic" and "Outcome" (0.42), these variables were excluded from the modeling process. Additionally, considering the strong correlation (0.75) between "EDA_Client(mean)" and "EDA_Client(mean_Baseline)," the former was also excluded. As highlighted in the data understanding phase, where the values of all EDA and HR mean_Baseline and SD_Baseline were repeated for each epoch within the same session number, we aggregated the data based on the session number to predict "WAI_Client." This step ensures a refined selection of predictors for an effective and accurate modeling approach.
66 Table 4: Pearson Correlation Among” WAI_Client” Predictors # Predictors for WAI_Client Correlation Coefficient Score Included Description 1 WA therapist 0.44 YES 2 Session 0.30 NO high correlation with “WAI-Therapist” (0.39) 3 Outcome 0.28 YES 4 HR-Client(SD_Baseline) (-) 0.18 YES repeated value in each epoch for each session number 5 EDA_Client(mean_Baseline) 0.17 YES repeated value in each epoch for each session number 6 Diagnostic 0.15 NO high correlation with “Outcome” (0.42) 7 EDA_Therapist(SD_Baseline) (-) 0.11 YES repeated value in each epoch for each session number 8 Therapist_sex (-) 0.12 YES 9 EDA_Client(mean) 0.10 NO high correlation with “EDA_Client(mean_Baseline)” (0.75) Upon scrutinizing the predictors of "WAI_Therapist" as outlined in Table 5, it became apparent that "Outcome" exhibited a robust correlation with "Diagnostic" (0.42) and was consequently excluded from the modeling process. Moreover, in the decision between "HR_Therapist(mean_Baseline)" and "HR_Therapist(mean)," the latter was omitted due to its lesser influence on "WAI_Therapist" compared to the mean_baseline value. Regarding "HR_Client(mean_Baseline)," its noteworthy correlation score of 0.75 with "HR_Client(mean)" prompted a careful evaluation. In this context, we opted for the baseline value, considering its stronger correlation and impact on "WAI_Therapist." It's crucial to highlight that "HR_Therapist(mean_Baseline)" comprised repeated values for each epoch within the same session, necessitating the aggregation of data at the session level for accuracy. Lastly, the variable "Therapist_Sex," exhibiting a Pearson correlation score of 0.49 with "HR_Therapist(mean_Baseline)," was excluded from the modeling process. This decision ensures a streamlined and effective predictor selection process for robust modeling outcomes. Table 5: Pearson Correlation Among” WAI_Therapist” Predictors # Predictors for WAI_Therapist Correlation Coefficient Score Included Description 1 WAI-Client 0.44 YES 2 Session 0.39 NO Repeated in each epoch. 3 Diagnostic 0.33 YES 4 Outcome 031 NO High correlation with “Diagnostic” (0.42)
67 5 Termination 0.23 YES 6 HR_Therapist(mean_Baseline) 0.19 YES Repeated value in each epoch for each session number 7 HR_Client(mean) (-) 0.18 NO High correlation with “HR_Client(mean_Baseline)” (0.75) 8 HR_Client(mean_Baseline) (-) 0.16 YES Choose to aggregate repeated value in each epoch for each session number 9 HR_Therapist(mean) 0.16 NO High correlation with “HR_Therapist(mean_Baseline)” (0.73) 10 Therapist_Sex 0.13 NO High correlation with “HR_Therapist(mean_Baseline)” (0.49) Leveraging the results of correlation analysis, we strategically picked the most impactful variables for predicting the target which are listed in Table 6. To predict "WAI_Client," the top six highly correlated predictors were chosen, including "WAI_Therapist," "Outcome," "Therapist_sex," "HRClient(SD_Baseline)," "EDA_Client(mean_Baseline)," and "EDA_Therapist(SD_Baseline)." Similarly, for predicting "WAI_Therapist," we selected key predictors such as "WAI_Client," "Diagnostic", "Termination","HR_Therapist(mean_Baseline)," and "HR Client(mean_baseline)." Table 6: List of selected predictors Target Predictors WAI_Client 1 WAI_Therapist 2 Outcome 3 Therapist_sex 4 HR-Client(SD_Baseline 5 EDA_Client(mean_Baseline) 6 EDA_Therapist(SD_Baseline) WAI_Therapist 1 WAI_Client 2 Diagnostic 3 Termination 4 HR_Therapist(mean_Baseline 5 HR Client(mean_baseline) 4.3.2 Modelling And Evaluation In this phase, we tailored our approach based on the problem type and target variable, employing various machine learning (ML) techniques. In addressing the research question on predicting therapeutic alliance using physiological and psychological factors from both clients and therapists, we utilized a range of popular machine learning methods including Artificial Neural Network (ANN), Decision Tree (DT), Random Forest (RF), Linear Regression (LR), and Support Vector Regression (SVR). This approach enabled us to explore the relationships between independent and dependent variables.
68 Evaluation techniques involved Nested cross-validation (CV=5) and GridSearchCV for model ranking and hyper-parameter tuning. Nested cross-validation, incorporating both inner and outer loops, facilitated unbiased performance estimates on unseen data while considering hyper-parameter tuning and model evaluation, mitigating overfitting risks. Although computationally demanding, we opted for CV=5 folds for both inner and outer folds to balance reliability and computational efficiency. Performance evaluation metrics included Mean Squared Error (MSE) and Coefficient of Determination (R2). A lower MSE indicates better model performance, while a higher R2 demonstrates improved goodness of fit. These metrics collectively gauged the model's overall performance, optimization through hyper-parameter tuning, and predictive accuracy in capturing the relationship between features and the target value. According to Table 7, in our comprehensive evaluation of various algorithms, Linear Regression (LR) and Random Forest (RF) emerged as the most adept techniques for predicting Therapeutic Alliance (WAI) for both clients and therapists. The LR utilizes a coefficient implementation, relying on weighted sums for predictions, where the coefficients serve as a crude feature importance score. Similarly, RF calculates the feature importance, offering insights into the relative significance of each predictor. Table 7: Ranking algorithms Technique Client’sWAI Therapist’s WAI MSE R2 MSE R2 Artificial Neural Network (ANN) 6.42 0.46 8.51 0.14 Random Forest (RF) 5.49 0.59 9.66 0.16 Decision Tree (DT) 8.04 0.33 16.48 (-) 0.23 K-Nearest Neighbors (KNN) 7.59 0.42 9.88 0.20 Support Vector Regression (SVR) 5.79 0.54 9.15 0.17 Linear Regression (LR) 6.06 0.53 7.90 0.30 4.3.3 Significance of Predictors Table 8 shows the predictive outcomes of Logistic Regression (LR) and Random Forest (RF) models for the therapist and client Working Alliance Inventory (WAI). Notably, for therapist WAI, "Diagnostic" (score = 3.62) and "Termination" (score = 2.93) ranked as the most impactful variables with a positive influence. Conversely, "HR_Client(Mean_Baseline)" exhibited less influence, while "HR_therapist(mean_baseline)" and "HR_client(mean_baseline)" played significant roles. Specifically, a lower HR(mean_baseline) in therapists correlated with higher therapist WAI, with HR_therapist(mean_baseline) showing importance as a physiological indicator (coefficient score of (-) 0.15).
69 For client WAI, the therapist's WAI (score = 0.40) and "Outcome" (score = 0.31) emerged as substantial predictors. In the realm of physiological indicators, EDA_client(mean baseline) (score = 0.13) stood out as a significant factor. In summary, concerning physiological indicators, while Heart Rate(mean_baseline) proved crucial for predicting therapist WAI (with a negative impact), EDA(mean_baseline) emerged as a noteworthy indicator for predicting client WAI. Table 8: Significance of predictor top predictor for " client’s WAI" by RF top predictors for “therapist’s WAI” by LR physiological Score NOT physiological Score physiological Score NOT physiological Score EDA_Client (mean Baseline) 0.13 therapist’s WAI 0.41 HR_Therapist (mean_baseline) (-)0.15 Diagnostic 3.62 ----- --- Outcome 0.31 ------- ----- Termination 2.93 4.3.4 Results and Discussion To conduct predictive analysis, we studied correlations among features and between features and targets, selecting relevant features and grouping repeated physiological indicators per session to prepare effective data for modeling. In the modeling phase, we employed various regression algorithms and ranked their performance through parameter tuning to identify the most effective algorithm for predicting the target. During the evaluation phase, we utilized different metrics to assess performance. Finally, we identified the significance of predictors to discuss outcomes at the application level. While this study is exploratory in nature, we successfully achieved the defined objectives and provided answers to the research question. Importantly, this investigation was conducted through interdisciplinary collaboration, closely working with the Department of Psychology at the University of Minho to support and validate our study, gain domain knowledge, and facilitate data collection. At application level, this study offers crucial insights into the application of data mining in psychotherapy. data mining and ML techniques are increasingly employed in psychotherapy to aid clinical decisionmaking, considering factors such as patients' characteristics and treatment history. These algorithms excel at capturing intricate patterns and relationships, leading to more accurate predictions of therapeutic alliance. Despite the rarity of such technologies in the field of therapeutic alliance, we have demonstrated their potential to strengthen it. By training prediction models on complex relationships identified in the data,
76 Figure 38: Steps to perform analytical insight 5.2.1 Data Processing and Preparation This phase encompasses a series of pivotal data processing tasks, aimed at enhancing the quality and applicability of the collected data. These tasks revolve around critical aspects such as: a. Feature Selection: Employing a meticulous approach to feature selection, we leveraged domain knowledge, statistical analyses, and data quality assessments. In instances like the vital sign dataset, variables with over 90% missing data were systematically excluded. b. Feature Engineering: In several datasets, we introduced novel variables to enhance analytical insight and harness valuable information. A prime illustration involves deriving the length of hospital-ICU stay by utilizing admission and discharge dates, thus transforming raw data into meaningful metrics. c. Extraction and Processing: The effectiveness of analytical performance was significantly amplified through rigorous data extraction. An illustrative example is the extraction of demographic details such as age and gender from diverse datasets, culminating in the creation of a consolidated dataset that streamlined subsequent data processing. d. Conditional Columns: Our transformative approach extended to the creation of conditional columns, converting raw data into actionable information. An example is our utilization of laboratory references for each exam, enabling a systematic comparison with exam results to ascertain their normal or abnormal status. e. Grouping and aggregating time-series data for vital indicators, thus addressing sporadic data registration issues encountered with biological sensors in the ICU. By adopting an hourly aggregation approach, we effectively mitigated the challenge of infrequent data updates. f. Cleaning-Missing cells: In addressing gaps within the vital sign dataset, a meticulous strategy was employed: we judiciously populated missing cells by computing the average value from neighboring cells preceding and following the voids. Similarly, within another Analytical insight Interrelating Clinical Events Crafting CEid Data Preparation
77 dataset, a pragmatic approach was taken by eliminating missing cells. As an integral part of our preparations, we meticulously fine-tuned all datasets to ensure they boasted fitting data types and pertinent features. Table 9: Key Transformations # Key data processing Description|example 1 Exploratory data analysis Identify mean, max, min, missing cells, and data quality 2 Feature Selection Using Knowledge of domain and missing cells 3 Feature engineering Construct new variable such as length of stay 4 Data extraction Demography table; Extract age and gender 5 Conditional feature Create Laboratory result status 6 The correct type of data Considering categorical, numerical, and also time series data 7 Group and aggregation For handling infrequent data generation, particularly from biological sensors 8 Missing cells Imputation method and elimination. 5.2.2 Encapsulating Diverse Clinical Events with CEid This phase harnessed a formula to craft a distinct key, uniquely pointing to each clinical event for individual patients. Illustrated in Figure 39, this key exhibits a structured composition: the initial number designates the event's day, followed by an abbreviation denoting the data type. Additionally, the third and fourth components encompass the patient's process number and episode number, respectively. The sequential order of events during a specific period is encapsulated within the event sequence, an ascending value ranging from one to n, culminating in the count of parallel events transpiring on the same day. Figure 39 serves as an illustrative example, underscoring an event linked to an episode number (20016701) attributed to a patient identified by process number 859785. This informative data point delineates the event's occurrence on the fifth day within the ICU context (vital sign data), constituting the eleventh clinical transaction. Creating this new variable holds significance in establishing connections among clinical transactions through a unique timeline based on the day of the clinical event. This process aims to facilitate the development of an analytical dashboard, wherein this code serves as a pointer to the key information of a patient. Also, to facilitate the clustering analysis.
78 Figure 39: Structure of CEid 5.3 Analytical Insight Through Interrelated Clinical Events Conducting analytical insight for the temporal analysis of interrelated clinical events stands as a potent data visualization tool in clinical decision-making. The paramount goal is to shed light on the chronological sequence and interconnections among a myriad of clinical events, providing invaluable insights into the temporal facets of patients' medical trajectories. This process enhances our understanding of the intricate web of events, contributing to a more nuanced comprehension of patients' healthcare experiences. The innovative integration of CEid further enhances the interrelation of these clinical events, providing a foundation for the development of advanced analytical insight. This integration using the multidimensional model of tables, provides a holistic perspective on patients' medical trajectories. Moreover, introduces a range of filters, empowering decision-makers to meticulously observe and analyze the distinctive clinical journey of each patient over a timeline, especially during their hospital stay. Figure 40 offers a visual representation of the access to analytical insights facilitated by the utilization of CEid, highlighting its capacity to provide a comprehensive view of patient's medical journeys. Through the utilization of various filters such as "gender," "process number," "episode number," "day of a clinical event," "age," and "type of clinical transaction," decision-makers can access and analyze patients' clinical trajectories. This allows for the identification of trends and patterns within the data. day of event type of event process# episode# event sequance parallel event 5. vts.859785, 20016701.3.11
79 Figure 40: Access the Patient’s analytical insight 5.3.1 Length of Stay in ICU In Figure 41, the bar chart illustrates the length of stay in the ICU for each patient, with the process number serving as the identifier. The legend distinguishes between genders, indicating whether the patient is female or male. Additionally, the presentation includes pre-operation and post-operation distinctions in the scatter chart. Furthermore, a scatter chart showcases the relationship between the length of stay and age. This diverse set of information can be accessed and presented using various filters, providing insights into demographics alongside the length of stay for each patient. Figure 41: Length of stay in ICU To present a comprehensive view of diverse clinical information on an integrated platform based on the "day of a clinical event," we focused on a specific patient with a process number equal to "1000025," Overview Days of stay in ICU clinical events diognostic interventions vital signs Laboratory results sepsis medications
80 as depicted in Figure 42. The total number of clinical transactions (TS) associated with this patient is 7798, with 5977 distinct transactions recorded. Notably, the patient is a 45-year-old male with the pathology code "N979." Their ICU status indicates "post-operation," with a total ICU stay of 22 days. Figure 42: Analysing Clinical Data for Patients with process number:”1000025” 5.3.2 Total Number of Clinical Events In Figure 43, the x-axis effectively chronicles the progression of days throughout the patient's tenure in the hospital, while the y-axis aptly quantifies the tally of clinical events, neatly categorized by transaction type. This informative bar chart represents a filtered view, concentrating on three pivotal event types for patients sharing a common process number of “1000025”: diagnostics (dig), interventions (int), and laboratory exams (lab). A discerning glance at this chart reveals that diagnostic events were notably concentrated on the first two days, with just a few transactions recorded. Furthermore, intervention actions persevered for a remarkable 27-day duration. And 125 laboratory exams on day one were reduced to 24 exams on day two. Figure 43: Temporal analysis of diverse clinical transactions
81 5.3.3 Diagnostics Figure 44 illustrates the number and types of diagnoses for the selected patient. It reveals that COVID-19 was diagnosed on day one, followed by a diagnosis of SARS on day two. This observation and analysis provide valuable insights into the patient's health status over a distinct period, facilitated by the unique periodic platform of the day of the clinical event. Figure 44: Temporal analysis of Diagnostic transactions 5.3.4 Intervention Figure 45 displays two types of interventions repeated from day one to day 21 and day 30. The yaxis represents the total number of each specific type of intervention.
82 Figure 45: Temporal analysis of intervention transactions 5.3.5 Vital Signs The vital sign dataset includes seven biological indicators registered by sensors in ICU, after hourly aggregation and transformation, all data points are placed in a timeline regardless of the time and date of data acquisition. Thus, there possible to monitor the average, minimum, and maximum value of each vital sign during the days of stay in the ICU. Figure 46 presents the fluctuation of the oxygen saturation average value (light blue), the maximum value (orange line), and the minimum value (blue line). The minimum value was observed on day three with a value of 83.9, and the maximum value of oxygen saturation was registered on days one and two with a value of 100. During four days of monitoring, fluctuations in the average value were observed to range between 100 and 97."
83 Figure 46: Temporal analysis of oxygen saturation Figure 47 illustrates a line chart that provides a detailed view of the fluctuation in Heart Rate (HR) values over the course of the patient's stay in the ICU. The y-axis of the chart represents the HR values, showcasing the minimum, maximum, and average HR values recorded each day. This visualization offers valuable insights into the patient's HR trends during their ICU stay. On the second day of the patient's admission, the chart indicates a minimum HR value of 56.88 beats per minute (bpm). In contrast, the HR reached its highest point on the fourth day, recording a maximum value of 148.87 bpm. Figure 47: Temporal analysis of Heart Rate 5.3.6 Laboratory Result Figure 48 shows the results of laboratory exams under the specific category. We transformed the minimum and maximum references associated with each test to define a conditional column. This feature construction identifies the condition of the result. Based on that, “Lr” means lower than minimum, “mr” is more than maximum, and “nr” means normal. According to the line chart, the result of “Neutrophiles” from 81 dropped to 6 on day one. It means that the value decreased from more than the maximum to a normal level. This method extracts significant information on the condition of a specific patient regarding the particular laboratory exam, and analyzing such information helps clinical decision-makers observe necessary laboratory results over days of stay in the hospital.
84 Figure 48: Temporal analysis of laboratory exams 5.3.7 Sepsis Figure 49 presents a graphical representation illustrating the occurrence of four distinct sepsis events over four days. Each instance is denoted by a recorded value, which remained constant at 53 for the first, second, and third days. However, on the fourth day, a noticeable increase is observed, with the recorded value rising to 60. This chart provides a clear and concise visualization of the timeline of sepsis events, offering valuable insight into the progression and severity of these occurrences over the observed period. Figure 49: Temporal analysis of sepsis 5.3.8 Medication Prescription
85 Figure 50 provides insights into four different types of prescribed medications, complete with average dosages and their respective units of measurement. Notably, on day 26, a prescription for 500 milliliters (ml) of 'glucose 10%' was administered. 'Brometo de ipratropium' was consistently prescribed from day 14 to day 40 at a fixed dose of 500 milligrams (mg). 'Colloredo de sodio' was recommended for a duration of 40 days, with varying dosage levels. Lastly, 'hydrocortisone' was prescribed from day 13 to day 26, with a notable decrease in dosage from 200mg to 50mg during this period. Figure 50: Temporal analysis of medications 5.3.9 Results and Discussion Figure 51 illustrates the significant contributions derived from analytical insight. Introducing the CEid and establishing connections among clinical events for individual patients through an analytical dashboard constitutes a significant scientific contribution to data-driven clinical decision-making, particularly within the realm of precision medicine. This phase marks a substantial advancement in the evolution of intelligent decision support systems tailored for precision medicine. The impact of the analytical dashboard, facilitated by CEid, is comprehensively explored from both a data-driven perspective and practical application. Furthermore, we systematically address the encountered limitations during this phase, ensuring a comprehensive understanding of the challenges and opportunities in implementing this innovative approach. a. The Transformative Impact of interrelating clinical events via CEid
92 admission/discharge timestamps, and date details were incorporated by merging with the admissiondischarge dataset. The resulting consolidated dataset, comprising 1,705,174 rows, encompassing 29 variables, and associated with 70 patients (process numbers), was meticulously prepared for subsequent clustering analysis. Figure 53: Pipeline to unify datasets 5.4.2 Clustering In preparation for the k-means clustering approach, we focused exclusively on numerical variables, omitting the "Process Number" too. Our data preparation encompassed several crucial steps, including handling missing cells, discarding columns with more than 50% missing values (Glasgow Coma, Sepsis), and eliminating corresponding rows with missing data. This involved meticulous selection of numerical variables, rectifying variable types, and ensuring the dataset's integrity by addressing duplicates. Consequently, the refined dataset earmarked for clustering comprises 692 rows and 9 variables, each intricately linked to 70 patients across a span of 39 days of clinical events. To choose the optimal number of clusters we performed the elbow method. The elbow method is a heuristic approach employed to determine the optimal number of clusters (k) in a dataset. It involves running the k-means clustering algorithm on the dataset for a range of values of k (e.g., from 1 to a certain maximum). For each value of k, the Within-Cluster Sum of Squares (WCSS) or the sum of squared distances between data points and their assigned cluster centroid is calculated. The WCSS is then plotted against the number of clusters, and the "elbow" of the curve represents the point i. Read each dataset. ii extract "day of event" from CEid (Process #- event dayvalue) group & aggregate (numerical mean value) iiii. generate custom code. extract the "day of event" from CEid (Process #, event day, code) Group and aggregate iv. Merge datasets two by two v. Merge with Admission-Discharge vi. Unified dataset for clustering
93 where the rate of decrease in WCSS slows down significantly. The idea is to choose the value of k at the elbow, as it signifies a good balance between model complexity (number of clusters) and the fit to the data. Figure 54 depicts the application of the elbow method to determine the optimal number of clusters. In addition to the elbow method, we explored a cluster range from n=2 to n=7, utilizing the average silhouette score for each cluster configuration. Through this analysis, we identified n=4 as the optimal number of clusters, ensuring a balance between cohesion within clusters and separation between clusters. The silhouette score is a valuable metric for assessing the quality of clustering, and our approach aimed to leverage it effectively in determining the most suitable cluster count for the given dataset. Figure 54: Elbow method for the optimal number of clusters 5.4.3 Identifying Patterns and Cluster References To analyze the characteristics of each cluster we extracted the average and standard deviation (std) of each variable in each cluster. The presented Table 10 outlines key numerical indicators across four distinct clusters (CLUSTER 0, CLUSTER 1, CLUSTER 2, CLUSTER 3), each characterized by a Silhouette Score denoting the cohesion and separation of data points within the cluster. The Silhouette Score ranges from 0.12 to 0.36, providing an assessment of the clustering quality, with higher scores indicating betterdefined clusters. Additionally, the table includes data distribution statistics, specifying the number of data points (rows) in each cluster. a. The key observation of Cluster 0 with the Silhouette Score equal to 0.23 and Data Distribution equal to 271 rows shows: - The average pulse rate of the arterial blood pressure is 100.66 with a standard deviation of 12.20. - Diastolic arterial blood pressure is around 64.18 with a standard deviation of 9.6. - Systolic arterial blood pressure is notably higher at 118.81 with a standard deviation of 15.46.
94 - Mean arterial blood pressure is 82.65 with a standard deviation of 11.14. - Heart rate is 99.18 with a standard deviation of 17.49. - The pulse oximetry oxygen saturation level is relatively high at 94.21 with a small standard deviation of 4.44. - Body temperature is around 36.17 with a standard deviation of 1.27. - The day of the clinical event is on average 10.60 with a relatively high standard deviation of 8.00. b. Cluster 1 with a Silhouette Score equal to 0.12 and Data Distribution of 44 has a lower silhouette score, indicating less cohesion and separation. - Variables exhibit higher standard deviations, suggesting greater variability within this cluster. - Body temperature has a relatively high standard deviation of 2.72, indicating variability in this aspect. Cluster 2 Silhouette has a higher silhouette score, suggesting better-defined clusters (Score: 0.26) and the Data Distribution equal to 285 rows. In This cluster - Variables like systolic arterial blood pressure and mean arterial blood pressure show significant variability with standard deviations of 15.13 and 12.20, respectively. - The pulse oximetry oxygen saturation level has a low standard deviation, indicating more consistency. c. Furthermore, Cluster 3 has the highest silhouette score, indicating well-defined clusters (Silhouette Score: 0.36) and Data Distribution equal to 92 rows. The key Observations show: - Systolic arterial blood pressure is notably high at 129.29 with a standard deviation of 15.13. The day of the clinical event has the lowest average at 4.98, suggesting events are concentrated around this day. In terms of any possible Trends and Patterns: - Cluster 3 stands out with the highest silhouette score, indicating the most distinct cluster. - Systolic arterial blood pressure is a key differentiator among clusters, with Cluster 3 having the highest values. The day of the clinical event shows variation, with Cluster 2 having the highest average and Cluster 3 having the lowest. Table 10: Characteristics of each cluster Indicator – Numerical (692 rows, 8 columns) CLUSTER 0 CLUSTER 1 CLUSTER 2 CLUSTER 3 Silhouette Score: 0.23 Silhouette Score: 0.12 Silhouette Score 0.26 Silhouette Score: 0.36 Data Distribution: 271 Data Distribution: Data Distribution: 285 Data Distribution: 92
95 44 Ave std Aver std Ave std Ave std The pulse rate of the arterial blood pressure 100.66 12.20 98.44 11.81 73.32 9.59 77.15 13.93 to diastolic arterial blood pressure 64.18 9.6 27.89 14.46 63.76 10.08 59.01 7.93 systolic arterial blood pressure 118.81 15.46 56.73 12.23 129.29 15.13 113.38 16.02 mean arterial blood pressure 82.65 11.14 43.09 7.26 86.90 12.20 77.21 9.62 Heart Rate 99.18 17.49 98.50 21.69 74.47 10.79 83.76 17.76 the pulse oximetry oxygen saturation level 94.21 4.44 93.11 6.2 95.52 2.73 94.27 3.93 Body Temperature 36.17 1.27 35.90 2.72 36.09 1.16 28.83 3.80 Day of clinical event 10.60 8.00 16.45 11.22 7.94 6.52 4.98 4.15 In this phase, we extracted the minimum and maximum values for each variable within every cluster. These extremal values serve as proximate references to be applied to the original dataset. By filtering the original dataset based on these boundaries, each data point is assigned to its respective cluster. Table 11 presents the minimum and maximum values of each variable across all clusters, offering a comprehensive overview of the distribution and range within each cluster. Based on that, the subsequent section of the table details the minimum (min) and maximum (max) values for various physiological parameters within each cluster. For instance, parameters such as pulse rate, arterial blood pressure, heart rate, oxygen saturation level, body temperature, and the day of the clinical event are captured. These min-max values offer a comprehensive view of the range and variability of each parameter within the respective clusters. For example, in CLUSTER 0, the pulse rate of arterial blood pressure ranges from 70.54 to 144.32, providing insights into the dispersion of this physiological metric within that specific cluster. This detailed breakdown facilitates a nuanced understanding of how clusters differ in terms of physiological characteristics. Also, to facilitate mapping and cluster assignments (next phase). Table 11: Cluster references Indicator – Numerical (692 rows, 8 columns) CLUSTER 0 CLUSTER 1 CLUSTER 2 CLUSTER 3 Silhouette Score: 0.23 Silhouette Score: 0.12 Silhouette Score 0.26 Silhouette Score: 0.36 Data Distribution: 271 Data Distribution: 44 Data Distribution: 285 Data Distribution: 92 min max min max min max min max the pulse rate of the arterial blood pressure 70.54 144.32 70.34 123.46 42.69 101.43 55.27 124.39 diastolic arterial blood pressure 48.80 138.37 10.83 53.50 37.01 101.60 40.51 88.03 systolic arterial blood pressure 82.88 173.66 45.15 96.03 98.27 179.98 74.60 164.71
96 mean arterial blood pressure 51.20 143.86 31.65 61.22 67.61 165.66 58.62 108.86 heart rate 47.71 144.65 57.38 149.02 49.63 116.32 54.46 133.74 pulse oximetry oxygen saturation level 49.18 99.46 71.31 98.75 77.85 99.54 65.00 99.03 body temperature 30.65 39.00 23.53 38.67 29.64 37.91 16.74 32.63 day of the clinical event 1 37 1 39 1 29 1 19 5.4.4 Results and Discussion This performance represents a significant stride in the continuous development of the IDSS4PM framework. Based on Figure 55, a crucial aspect of adopting a data-driven approach was the introduction of standardization, particularly through the creation of custom codes for categories lacking clinical codes. This transformative step not only enhanced data quality but also facilitated data integration and clustering techniques. Furthermore, leveraging the CEid to extract the day of the event proved instrumental in grouping, aggregating, and implementing temporal clustering. By incorporating process numbers and aggregating based on the day of the clinical event, we successfully integrated datasets to apply clustering techniques. The extraction of homogeneous data points enabled a comprehensive study of the behavior of similar data, allowing for the identification of potential patterns. Additionally, extracting the minimum and maximum values of each variable in each cluster provided us with cluster references crucial for subsequent analysis. This experimental phase marked a significant advancement in health profiling. The utilization of cluster references and their mapping onto the main dataset in the next phase will further enhance our understanding of the intricacies within the data, particularly in identifying anomalies or outliers beyond established boundaries.
97 Figure 55: Contributions to IDSS4PM; Clustering 5.5 Cluster-Based Data Mapping and Assignment This phase encompasses crucial steps aimed at grouping similar data points, both categorically and numerically. As illustrated in Figure 56, our initial approach involves leveraging cluster references for data extraction, followed by assigning clusters and data mapping. Subsequently, we meticulously analyze each cluster to gain insights and draw meaningful conclusions. Figure 56: Cluster-based data mapping and cluster assignment 5.5.1 Leveraging Cluster References, Extraction and Mapping In this pivotal phase, our focus was on translating the identified clusters into actionable insights within t he original dataset. Leveraging the minimum and maximum values established for each numerical varia Contributions Standardization Integration Identify homogenious data Cluster reference Identification of patterns Quality Control and Data Integrity Data Driven prespective Limitations & challenges Mixed data types Data qualilty Size of data Cluster overview Identify homogeneous data. Data Mapping &Assignment Cluster-Based Data Extraction Leveraging Cluster Reference
98 ble in every cluster, we meticulously mapped and assigned rows to their corresponding clusters. By emp loying these cluster references, we extracted pertinent data points from the original dataset and systema tically tagged each entry with its designated cluster number. For the critical task of mapping clusters onto the original dataset—comprising both numerical and categorical data—we employed a unique approach. By determining cluster references through the minimum and maximum values of each variable within each cluster, we not only surmounted the challenge of clustering mixed data types but also successfully addressed the grouping of columns with 50% missing data, previously omitted during clustering. This pioneering approach emerged as a cornerstone, seamlessly accommodating the diversity inherent in clinical data types—both categorical and numerical. The final step involved a thorough observation and analysis of each profile, offering a comprehensive understanding of the distinctive characteristics encapsulated within every cluster. 5.5.2 Identify Homogeneous Data In the current phase of our study, we have successfully mapped and assigned clusters to a dataset comprising 1,631,242 records, encompassing 19 variables. This comprehensive analysis allows us to understand the distribution of data across different clusters, providing valuable insights into distinct health patterns. Based on that the total size of mapped clusters is 1,631,242 rows, 19 variables. The individual Cluster Breakdown shows: - Cluster 0: 598,266 rows - Cluster 1: 110,042 rows - Cluster 2: 593,192 rows - Cluster 3: 325,289 rows It's noteworthy that, within the total dataset, there are 4,453 unclustered data points. These unclustered data points represent instances that may not align with the identified patterns in the current clustering approach or could have missing values that need further investigation. This cluster distribution summary provides a clear overview of the sizes and composition of each cluster, setting the stage for more detailed analyses and insights into the specific characteristics associated with each health pattern. Additionally, attention to unclustered data points allows for a more comprehensive understanding of the dataset and potential areas for refinement in future analyses.
99 5.5.3 Cluster Overview and Analysis The bar chart, in Figure 57, shows the average value of each indicator based on cluster number. The key observation shows that: Cluster 0 - Pulse Rate: High (111.04 beats per minute) - Diastolic Arterial Blood Pressure: Moderate (64.57 mmHg) - Mean Arterial Blood Pressure: Moderate (77.92 mmHg) - Systolic Arterial Blood Pressure: High (108.65 mmHg) - Pulse Oximetry Oxygen Saturation Level: Normal (96.04%) - Body Temperature: Elevated (36.52°C) - Heart Rate: High (107.05 beats per minute) - Glasgow Coma: Low (12.27) - Sepsis: Absent - Day of Clinical Event: 1.06 Cluster 1: - Pulse Rate: Moderate (90.43 beats per minute) - Diastolic Arterial Blood Pressure: Low (35.65 mmHg) - Mean Arterial Blood Pressure: Low (42.80 mmHg) - Systolic Arterial Blood Pressure: Low (55.58 mmHg) - Pulse Oximetry Oxygen Saturation Level: Normal (73.15%) - Body Temperature: Low (27.02°C) - Heart Rate: Moderate (77.68 beats per minute) - Glasgow Coma: Moderate (5.33) - Sepsis: Absent - Day of Clinical Event: 1.09 Cluster 2: - Pulse Rate: Low (74.88 beats per minute) - Diastolic Arterial Blood Pressure: Moderate (64.98 mmHg) - Mean Arterial Blood Pressure: Moderate (85.31 mmHg) - Systolic Arterial Blood Pressure: High (128.07 mmHg) - Pulse Oximetry Oxygen Saturation Level: Normal (95.41%) - Body Temperature: Elevated (36.53°C)
100 - Heart Rate: Moderate (75.21 beats per minute) - Glasgow Coma: High (14.52) - Sepsis: Present (49.97% probability) - Day of Clinical Event: 1.22 Cluster 3: - Pulse Rate: Moderate (79.01 beats per minute) - Diastolic Arterial Blood Pressure: Moderate (56.27 mmHg) - Mean Arterial Blood Pressure: Moderate (70.16 mmHg) - Systolic Arterial Blood Pressure: High (101.39 mmHg) - Pulse Oximetry Oxygen Saturation Level: Normal (93.86%) - Body Temperature: Elevated (30.01°C) - Heart Rate: Moderate (80.84 beats per minute) - Glasgow Coma: Moderate (14.11) - Sepsis: Present (49.97% probability) - Day of Clinical Event: 1.09 The analysis reveals pivotal observations (presented in Figure 57) indicating that each cluster exhibits unique patterns in health parameters, indicating potentially distinct health profiles. Cluster 0 has a high pulse rate, elevated body temperature, and high systolic arterial blood pressure, suggesting a more intense clinical event. Moreover, cluster 1 shows signs of lower blood pressure, lower heart rate, and lower body temperature, indicating a different health state compared to Cluster 0. Clusters 2 and 3 have relatively similar pulse rates but differ in other parameters, such as systolic arterial blood pressure and the presence of sepsis. Figure 57: Behaviour of data in each cluster
101 5.5.4 Exploring Health Profiles Analyzing the minimum, maximum, and mean variables in each cluster enhances the ability to understand the distribution of health parameters, identify characteristic patterns, and glean insights into the diverse health profiles represented within your dataset. These insights, in turn, contribute to more informed decision-making and the development of precise healthcare strategies. Cluster 0: - The day of the clinical event: Ranges from 1 to 37, with a mean of 1.06. Events are spread across the observed period. - Glasgow Coma Scale: Varies between 3 and 15, with a mean of 12.27. Indicates a range of consciousness levels, potentially reflecting diverse patient conditions. - the sepsis from 61 to 63, with a mean of 61.13. This variable shows minimal variability within the cluster. - arterial blood pressure ranges from 70.54 to 136.56, with a mean of 111.04. Suggests a spectrum of blood pressure levels, potentially indicating varying severity of cardiovascular conditions. - Systolic arterial blood pressure varies between 77.92 and 114.91, with a mean of 108.65. Indicates a range of systolic blood pressure levels. - Mean arterial blood pressure ranges from 48.81 to 104.75, with a mean of 96.04. Indicates variations in overall blood pressure within the cluster. - Diastolic arterial blood pressure varies from 31.27 to 38.01, with a mean of 64.57. Shows a range of diastolic blood pressure levels. - Temperature: Consistent at a mean of 36.52, reflecting stability in body temperature within the cluster. - Oxygen saturation levels range from 49.18 to 99.13, with a mean of 96.04. Indicates variability in oxygen saturation. - Heart rate varies between 47.72 and 137.94, with a mean of 107.05. Shows a range of heart rates within the cluster. - The laboratory results display a wide range, spanning from -0.6 to 88033, with an average value of 109.55. This considerable variability suggests a spectrum of diverse clinical conditions. However, for a more precise discussion, it is essential to consider the specific category of the laboratory test being analyzed, as will be further presented in subsequent discussions. • Cluster 1 - Event Day: Similar to Cluster 0, with a mean of 1.09, indicating events spread across the observed period.
108 Figure 65: Behaviour of body temperature in each cluster 5.6.6 Sepsis From Figure 66, it's evident that cluster 0 displays minimal variability in sepsis (ranging from 61 to 63), hinting at a relatively uniform response to sepsis within this group. On the other hand, cluster 2 exhibits moderate variability (ranging from 28 to 66), indicating differing degrees of sepsis conditions within the cluster. Additionally, cluster 3 demonstrates the highest maximum value of sepsis at 97, while the lowest minimum value is found in cluster 2 at 28. Figure 66: Behavior of sepsis in each cluster 5.6.7 Days of Clinical Events The average day of clinical events across all clusters is approximately 1.1, suggesting a relatively even distribution of events throughout the observed period. However, as depicted in Figure 67, cluster 1 exhibits the highest number of maximum days associated with clinical transactions (39 days), while cluster 3 has the least (19 days).
109 Figure 67: Analysing the total days of clinical transactions in each cluste In summary, considering the vital signs, Clusters 0 and 3 appear to represent groups with more critical conditions, as indicated by higher GCS values, elevated blood pressure, and a wider range of laboratory results. Moreover, Cluster 1 signifies a group with consistently lower GCS, moderate blood pressure, and wider variability in heart rate and laboratory results. Additionally, Cluster 2 reflects a mix of conditions, with higher GCS, moderate blood pressure, and variability in oxygen saturation and sepsis indicators. These trends provide a foundation for further investigation and highlight the need for domain-specific knowledge to interpret the clinical significance of the observed patterns. Additionally, outlier detection and data validation are crucial for a more accurate understanding of the dataset. 5.6.8 Temporal Distribution Of Clusters One of the paramount advantages of temporal analysis lies in its capability to delve into the behavioral patterns of clusters over time. In this context, proposing the utilization of CEid facilitates the extraction of clinical event data daily during the clustering phase-processing. This step significantly advances the identification of patterns and trends over time, offering a more comprehensive understanding of temporal dynamics. Figure 68 illustrates the temporal distribution of each cluster. For instance, cluster 2 is observed on all days except days 37 and 39. Another noteworthy observation is that after day 34, a singular cluster is exclusively assigned to that specific day. Additionally, cluster 3 is assigned only during the initial 19 days. In summary, temporal analysis in the context of cluster distribution provides a dynamic view of patients' health trajectories, aiding precision medicine by enabling timely interventions, enhancing predictive modeling, and fostering a nuanced understanding of the temporal aspects of clinical data.
110 Figure 68: Temporal distribution of each cluster 5.6.9 Laboratory Exams and Results The bar chart displayed in Figure 69 offers a comprehensive glimpse into the variety of laboratory examinations within each cluster, detailing the results as either normal or abnormal. The legends conveniently provide the codes for each examination. Hence, this visualization and analysis offer the opportunity to explore and examine the union, intersection, and distinctions of each cluster in terms of laboratory codes. This analytical insight holds paramount importance for decision-makers seeking to discern the distinct behavioral patterns of each cluster. For instance, within this analysis, cluster 1 exhibits the lowest count of examinations yielding abnormal results, suggesting a relatively healthier profile. On the other hand, Cluster Two stands out with the highest number of examinations producing both abnormal and normal results. Such granular insights enable decision-makers to better understand and interpret the diverse dynamics within each cluster, facilitating more informed and targeted decision-making processes.
111 Figure 69: Behaviour of laboratory exams & results in each cluster 5.6.10 Interventions Figure 70 provides a visual representation of the exploration into the union, intersection, and distinctions of interventions within each cluster. Remarkably, 'int9' stands out as the most recurrent intervention, observed across all clusters. In detail, the x-axis corresponds to the intervention codes, the y-axis signifies the number of interventions, and the legend elucidates the cluster types under consideration. This graphical representation enhances our understanding of the intervention landscape, emphasizing the prevalence and distribution of intervention types across various clusters. For instance, the “int9” is observed in all clusters.
112 Figure 70: Distribution of interventions (codes) in each cluster Likewise, the chart depicted in Figure 71 showcases diagnostic codes associated with their respective diagnostic types (x-axis) and their distribution across clusters. Notably, the diagnostic code 8271 is exclusively linked to cluster Zero, indicating a specific association within this cluster. 5.6.11 Diagnostic Figure 71: Distribution of diagnostic (codes) in each cluster 5.6.12 Medications In the concurrent analysis, as portrayed in Figure 72, the visual representation delineates the distribution of medications across each cluster. While the bulk of medications is allocated to all four clusters,
113 discernible differences emerge in the total count of each medication type (on the x-axis) assigned to individual clusters. Figure 72: Distribution of medications (codes) in each cluster 5.6.13 Procedures (local, Zona) The two Figures, 73 and 74, depict the categorization of procedures into local and Zona types across each cluster. Notably, "prcl1" is exclusively associated with clusters zero and two, while "prcl10" and "112" are present solely in cluster zero. Likewise, concerning the Zona procedures, "prcz3" is specifically assigned to cluster 3.
114 Figure 73: Distribution of procedure-local (codes) in each cluster Figure 74: Distribution of procedure-zona (codes) in each cluster 5.6.14 Results and Discussion This phase marks a crucial advancement in the clustering process. Figure 76 illustrates the contributions to the clinical decision support system by identifying similar data behaviors. Leveraging the cluster references acquired from the preceding step, we extract data from the original dataset, which encompasses diverse types of clinical data. This extraction facilitates the assignment of the whole data to the relevant cluster. In this experimental context, our approach yields grouped categorical and numerical data. Consequently, the identification of trends and analysis of data behavior in a temporal context represent sophisticated strides toward data-driven decision-making.
115 The temporal analysis of cluster distribution is crucial for precision medicine for several reasons. By examining how specific clusters manifest over time, we gain insights into the temporal dynamics of clinical events and their associations. This temporal perspective allows us to identify patterns, trends, and potentially critical transitions in patients' health journeys. In precision medicine, understanding the evolution of clusters helps healthcare professionals tailor interventions based on the timing of specific clinical events. For example, if a particular cluster is prevalent during a specific time window, it may indicate a critical phase in a patient's condition that requires targeted and timely interventions. Moreover, temporal analysis enhances the accuracy of predicting and assigning new data to relevant clusters. It enables healthcare providers to recognize patterns that may be indicative of disease progression, treatment effectiveness, or other important factors influencing patient outcomes. This, in turn, contributes to more informed decision-making and personalized treatment strategies. The analysis presented above marks a significant stride in advancing clinical decision-making, leveraging both a data-driven perspective and its application in precision medicine. The exhaustive breakdown of categorical data, including medications, procedures, laboratory results, diagnostics, and more, linked to each cluster, imparts a nuanced understanding of the diverse clinical pathways embedded in the data. This level of granularity equips healthcare professionals with insights to comprehend the intricacies of patient conditions and treatment patterns. Furthermore, visualizing the relationships between the union, intersection, and differences of codes across clusters unveils distinct patterns and trends within the data. This visualization aids in more informed decision-making, drawing on historical data to understand the associations between various factors. Moreover, the identification of specific clinical types unique to certain clusters facilitates tailored and targeted interventions. Healthcare providers can utilize this detailed information to craft personalized treatment plans aligned with the specific needs and characteristics of each patient subgroup. In essence, this analytical approach empowers clinicians with a data-driven comprehension of the intricate relationships between available historical data and patient clusters. By providing insights into both historical trends and the specific needs of patient subgroups, this method aligns closely with the principles of data-driven healthcare and precision medicine, fostering a more refined and personalized approach to clinical decision-making. Building on these insights, the clinical decision-maker gains the ability to allocate new data into specific clusters, utilizing the outcomes of this clustering process. Furthermore, in this phase, we advanced
116 predictive modeling techniques to forecast the most suitable cluster for new data, ensuring a more accurate and tailored approach to decision support. Figure 75: Contributions to IDSS4PM; Analytical Insight Contributions Data Driven prespective Data Exteraction Quality Control and Data Integrity advance classification phase Anomaly detection temporal anaysis Identify catgorical assigned to each cluster Identify homogenious data Mapping with cluster references Application in PM Identify patients at risk Navigating intricate patient trajectories adaptive treatment Health profiling
117 5.7 Predicting Modeling through Classification 5.7.1 Classification In this phase, we implement predictive health profiling by leveraging behavior-based classification on new data, building upon the insights gained from the temporal clustering performed in the previous phase. The objective is to apply a classification approach to train the model using 258,054 rows (after excluding missing data) and 18 predictors—a mix of categorical and numerical variables. The predictors encompass a comprehensive range of clinical indicators, providing the capability to forecast the cluster number over the course of a clinical day. By training the classification algorithm on this dataset, we empower it to discern patterns and associations within the data, enabling the prediction of cluster numbers for each day of clinical performance. This predictive health profiling methodology offers a dynamic and proactive approach to understanding patient trajectories, aiding healthcare professionals in anticipating and responding to potential developments in patient conditions. As new data becomes available, the classification algorithm can be applied to furnish timely predictions of cluster numbers, thereby contributing to a more informed and personalized healthcare decision-making process. For example, we applied a random forest classifier with cross-validation techniques on a dataset of 258,054 records with 19 features. The goal was to predict the cluster using historical data and to discern the significance of each variable in the prediction process. Following the analysis, with cross-validation k=5, we obtained a mean accuracy of 0.97 and it was determined that "Temperature” was the most influential factor in predicting the cluster according to the random forest model. The results obtained indicate the performance of your Random Forest classifier using cross-validation. Here's a summary: a. Mean Accuracy: 0.946, with a standard deviation of 0.078. This indicates that, on average, the model correctly predicts the target class around 94.6% of the time. Accuracy mentioned in equation 1, measures the proportion of correctly predicted instances among all instances. 𝐸𝑞𝑢𝑎𝑡𝑖𝑜𝑛 1:𝐴𝑐𝑐𝑢𝑟𝑎𝑐𝑦 = 𝑇𝑟𝑢𝑒 𝑃𝑜𝑠𝑖𝑡𝑖𝑣𝑒𝑠 + 𝑇𝑟𝑢𝑒 𝑁𝑒𝑔𝑎𝑡𝑖𝑣𝑒𝑠 𝑇𝑜𝑡𝑎𝑙 𝑃𝑟𝑒𝑑𝑖𝑐𝑡𝑖𝑜𝑛𝑠 Equation 1.Accuracy
124 Figure 78: Theoretical reference; Project I vi. At application level, Analytical insight, as described in the context of examining relationships between therapists and clients in psychotherapy sessions, adds significant value to applications in precision medicine in several ways: The examination of relationships between therapists and clients provides insights into the dynamics of the therapeutic alliance. Analyzing these interactions allows for the identification of factors that positively or negatively impact the alliance. This knowledge is valuable in refining and optimizing the therapeutic relationship, which is crucial for successful precision medicine interventions. Moreover, identifying significant behavioral trends early on provides an opportunity for early intervention and preventive measures. This proactive approach aligns with the goals of precision medicine by addressing issues before they escalate, potentially improving treatment outcomes and minimizing adverse effects. Additionally, analytical insight supports continuous monitoring of patient data, allowing for dynamic adjustments to treatment plans. As patterns evolve or new trends emerge, healthcare providers can adapt interventions in real time, ensuring that the precision medicine approach remains aligned with the sub-optimal outcome IDSS4PM'--exploratory study Techniques role Analytics Intelligence-Design-Choice reference Analytics Descriptive observe and analyse data analysis data integration & Analytical insight intelligence Predictive predicting the next performance AI/machine learning Predictive analysis intelligence & design Prescriptive recommending the best possible solution simulation & optimization Advancing recommendation intelligence & choice
125 patient's evolving needs. By customizing treatment approaches based on observed patterns, analytical insight supports a patientcentered care model. Precision medicine aims to consider individual variations, and the personalized insights derived from data analysis contribute to a more patient-centric and tailored healthcare approach. In terms of decision-making, the meticulous analysis of interrelated factors contributes to more informed decision-making in precision medicine. Healthcare professionals can make evidence-based choices, considering the nuanced relationships between physiological and psychological indicators, leading to more effective and targeted interventions. Finally, predictive analysis and Feature importance help in assessing the risk associated with different factors. For example, in predicting the response to a particular treatment/medication, the algorithm can identify which patient-specific features are most indicative of positive or negative outcomes. This aids healthcare providers in making informed decisions about the potential risks and benefits for a specific patient. This allows for timely interventions or preventive measures, contributing to better health outcomes. Furthermore, by understanding feature importance, precision medicine can optimize the allocation of diagnostic and therapeutic resources. Focus can be directed toward the most influential factors, improving efficiency and reducing unnecessary interventions. Another key contribution Is about understanding complex relationships. In precision medicine, factors influencing health outcomes are often interconnected and complex. Machine learning algorithms can unravel these intricate relationships by assigning importance to each factor. This helps healthcare professionals comprehend the multifaceted nature of diseases and treatments.
126 Figure 79: Contributions to IDSS4PM; Project I Contributions DataDriven Analytical insight Interrelated clinical events Identify patterns Uncocver links between data Predictive Modelling Correlation analysis Feature selection Data processing and preparation Model ranking and tunning Discover complex relationships Significance of Predictors Psychological indicators Physiological Indicators Application in PM Continious monitoring Early detection, prevention Improve patient engagement Customise therapy plan Evidence based decision Identify risk factors Limitations &Challenges Ethical considerations Sample limitations Theoritical Reference Interdiciplinary collaboration
127 6.2 Data-DrivenIntensive Care Medicine Decision Support i. CEid formulation and its transformative Impacts. In the realm of data-driven aspects, the CEid formula stands out for its multifaceted and impactful advantages. These transformative endeavors resonate across datasets, introducing a variable that amalgamates key information. This identity plays a pivotal role in facilitating data processing and integration, essential for predictive modeling. The sequential arrangement of events within a specific period, as captured by the event sequence, provides a comprehensive view of temporal order. This addresses challenges related to infrequent data generation and proved impactful for information exchange as a standard approach. Additionally, the CEid formula introduces a standardized method for creating keys for clinical events, ensuring uniformity and consistency in data representation. This standardization is crucial for interoperability and compatibility across healthcare systems, supporting the analysis and interpretation of temporal relationships. Moreover, the crafted key serves as a standardized identifier for clinical events, promoting consistency and ease of information exchange. This common temporal reference facilitates smooth information exchange across different systems and platforms, providing a standardized framework for understanding and interpreting temporal relationships in clinical data. The event sequence, part of the crafted key, furnishes a chronological order of clinical events, enabling temporal clustering. This sequential information allows for the identification of patterns or clusters of events occurring closely in time, enhancing temporal clustering by encapsulating diverse clinical events within a unified temporal framework. This aids in the recognition of meaningful clusters. In terms of predictive modeling, the structured key, with detailed components such as the day of the event, data type, process number, episode number, and other clinical indicators, serves as a rich source for features in predictive modeling. This enables the development of predictive models that account for the sequential nature of clinical events, contributing to more accurate predictions based on temporal relationships. The strength of the formula lies not only in its technical and data-driven aspects but also in its practical application in precision medicine. The CEid offers a dynamic and flexible means to analyze and interpret complex patient data, supporting informed decision-making and significantly enhancing the quality of patient care by providing a nuanced understanding of individual medical journeys. This approach equips medical professionals with a versatile toolset, complete with filters facilitating a comprehensive review of diverse clinical events and their associated details, all within a unified temporal
128 framework. This innovative approach provides practitioners with a holistic lens through which to navigate and discern intricate patient trajectories, fostering informed decision-making and enhanced patient care. The meticulous examination of clinical data plays a pivotal role in continuous patient monitoring. By keenly observing data fluctuations over time, healthcare providers can promptly identify deviations from a patient's baseline health and intervene early to mitigate potential complications. The unification of clinical events further empowers decision-makers, offering a comprehensive understanding of each patient's unique health journey. Armed with this information, they can tailor interventions to address the individual needs of the patient, thereby making treatment decisions more personalized. Moreover, this holistic approach to decision-making is instrumental in providing a panoramic view of a patient's health history. It not only facilitates proactive care but also contributes to an overall enhancement of patient well-being. The capability to track and analyze clinical events longitudinally enables decision-makers to pinpoint trends and patterns that may signify the onset of health issues. Early detection through this method leads to proactive interventions, curbing the severity and cost of treatment. In terms of treatment effectiveness, clinical decision-makers leverage their ability to assess the impact of treatment strategies by analyzing patient responses over time. This valuable information guides the adjustment of treatment plans, aiming for improved outcomes and optimized resource allocation. Lastly, the focus extends beyond mere disease treatment to encompass preventive and proactive health management. The emphasis lies not only in addressing existing conditions but also in identifying risk factors and taking pre-emptive measures to prevent or delay the onset of diseases. ii. Analytical insight Clinical Event Identification (CEid) serves as a cornerstone in linking clinical events across a unified platform. By integrating each patient's clinical background, it enables thorough observation and analysis. This step is pivotal in identifying patterns within an integrated platform, providing decision-makers with comprehensive insights crucial for informed, evidence-based decisions. Ultimately, this approach enhances the likelihood of successful treatment outcomes and ensures patient satisfaction. iii. Cluster analysis As previously mentioned, the CEid plays a pivotal role in seamlessly integrating diverse data sources and types by extracting the day of the clinical event and the process number. Consequently, employing clustering techniques on the integrated data, independent of the process number, establishes an optimal platform to investigate the homogeneity of physiological data. This process involves creating cluster references to assign a combination of categorical and numerical data to each cluster. Moreover, the
129 standardization applied to generate custom codes for categorical variables enhances the overall performance of this procedure. In addition, the phase of mapping and cluster assignment, with analyzing the minimum, maximum, and mean variables in each cluster provides valuable insights and benefits in several ways within the context of this work on health profiling and data-driven aspects. Here are the advantages: a. Identification of Extreme Values: Min and Max Values: Examining the minimum and maximum values within each cluster helps identify extreme values or outliers. These outliers may represent unique cases or anomalies that can provide valuable information about exceptional health conditions or atypical patient responses. b. Understanding Range and Variability: Mean and Variability: Calculating the mean and understanding the variability (min to max range) of variables within each cluster provides insights into the overall distribution of health parameters. This information is crucial for understanding the typical range of values associated with different health profiles. c. Differential Health Characteristics: Cluster-Specific Patterns: By comparing the minimum, maximum, and mean values across clusters, we can identify cluster-specific patterns. This helps in distinguishing the characteristic health parameters that define each cluster and contributes to the creation of more nuanced health profiles. d. Clinical Relevance: Identification of Clinical Significance: Extreme values or specific patterns in minimum, maximum, or mean values may have clinical significance. For instance, unusually high or low values could signal potential health risks or specific medical conditions that warrant closer attention. e. Feature Importance in Predictive Modelling: Variable Importance: Understanding the importance of variables based on their range and variability in different clusters can aid in feature selection for predictive modeling. Variables with substantial variations across clusters may play a crucial role in predicting health outcomes. f. Quality Control and Data Integrity:
130 Detecting Data Anomalies: Discrepancies or unexpected patterns in minimum, maximum, or mean values may indicate data anomalies or errors. Regularly monitoring these statistics helps ensure the quality and integrity of the dataset. At the end of each phase support clinical decision makers to tailoring Interventions: Insights derived from the analysis of variable ranges can contribute to the development of more precise medicine strategies. Tailoring interventions based on the specific health characteristics within each cluster allows for more targeted and effective healthcare practices. iv. Classification: Predictive health profiling through behavior-based classification, coupled with the utilization of temporal clustering, holds significant potential to propel the field of precision medicine forward. Here are key aspects that underscore its potential impact: From a Data-Driven Decision-Making perspective, the incorporation of 18 predictors, encompassing a mix of categorical and numerical variables, reflects a comprehensive approach to capturing diverse facets of patient health. This rich dataset empowers healthcare professionals with the tools for more informed and data-driven decision-making, thereby elevating the precision and accuracy of medical interventions and treatment. In Application-Level Insights: Leveraging behavior-based classification and temporal clustering enhances the model's capability to discern individual variations in patient health trajectories. This nuanced profiling contributes to a more personalized understanding of patient conditions, facilitating tailored medical interventions for better outcomes (Individualized Patient Profiling). Moreover, resulted in early detection and Intervention. The model's capacity to predict cluster numbers throughout the clinical day signifies an augmented ability to detect subtle changes in a patient's health status early on. This early detection lays the groundwork for timely interventions, potentially averting adverse health events and improving overall patient outcomes. In addition, demonstrating real-time adaptability, the model's capacity to work with new data and provide daily predictions is crucial in a clinical setting where patient conditions may evolve rapidly. This feature enables continuous monitoring and the adaptation of treatment plans as needed. In conclusion, the synergy between behavior-based classification, temporal clustering, and advanced data analytics holds transformative potential for precision medicine, promising more personalized, timely, and effective healthcare interventions. v. Limitations and Challenges
131 We encountered significant limitations and challenges in advancing our analytical insights, including issues with data quality, diverse datasets, and infrequent data collection from sensors. Leveraging the CEid, we were able to address some of these challenges partially. To create the artifact, we faced limitations in accessing specific types of data. For instance, datasets like SOAP contained predominantly unstructured text data, lacking the structured format required for predictive modeling. Additionally, ethical concerns regarding the accessibility of genetic profiles posed constraints on sample availability, further limiting our ability to gather diverse and comprehensive data for analysis. vi. Theoretical reference - Suboptimal Decision-Making Figure 80 illustrates the pivotal role of analytics and applied techniques on clinical data to achieve suboptimal outcomes. As discussed earlier, the complexity of individual patient data arises from diverse sources and formats, leading to dynamic, semi/unstructured, multi-dimensional, and fragmented data (Kaur & Mann, 2018). In the context of healthcare, suboptimal decision-making refers to instances where the chosen course of action does not yield the best possible outcome for the patient. This can occur due to various factors, including incomplete information, cognitive biases, or limitations in analytical tools. Therefore, advanced analytics, particularly predictive and prescriptive analytics, become essential for optimizing treatment performance (Palaniappan & Awang, 2008). By leveraging past patient data and sophisticated machine learning algorithms, predictive analytics can anticipate treatment features and potential outcomes. However, despite the predictive capabilities of these techniques, decision-makers may still encounter challenges in identifying the most appropriate course of action. Prescriptive analytics, on the other hand, offers a decision-focused approach to healthcare decisionmaking. By simulating various treatment scenarios and optimizing for the desired outcome, prescriptive analytics can guide clinicians toward more effective interventions. Nonetheless, the application of prescriptive analytics in real-world settings may be constrained by factors such as resource limitations or organizational constraints. Overall, the integration of advanced analytics into clinical decision-making processes holds significant promise for improving patient outcomes and enhancing healthcare delivery. However, it is essential to recognize the inherent challenges and limitations associated with suboptimal decision-making in healthcare settings and to continue refining analytical approaches to address these complexities. To advance the sub-optimal model of clinical decision-making and develop the IDSS4PM artifact, we introduced the CEid and analytical insights to grasp the underlying problem. This involved employing descriptive analytics to glean valuable knowledge, thus representing the intelligent phase. Furthermore,
132 predictive modeling contributed to both the intelligence and design phases. In Simon's model, decisionmakers design alternative courses of action based on predictions—a process mirrored in predictive analytics, which assists in devising treatment plans by suggesting strategies informed by projected outcomes. This iterative approach aligns with the iterative process of generating and evaluating multiple treatment scenarios based on predictive models. Finally, Choice Phase: Prescriptive analytics corresponds to the choice phase of Simon's model, where decision-makers select the most optimal course of action among the alternatives generated in the design phase. Prescriptive analytics guides decision-makers by recommending specific treatment options or interventions that are expected to yield the best outcomes based on simulation and optimization techniques. Furthermore, prescriptive analytics supports intelligent adaptation by continuously refining treatment recommendations based on new data or changing patient conditions. This reflects Simon's notion of intelligent adaptation, where decision-makers adjust their strategies in response to feedback and environmental changes to achieve better outcomes over time. In this context, building upon the preceding discussion, our contribution was directed towards enhancing intelligence, refining the design process, and advancing the decision-making choices.
133 Figure 80: Theoretical reference; Project II I. Application in PM: In addition to our contribution to data-driven aspects, including the development of an intelligent decision support system for precision medicine with implications on the application level, the outcomes indirectly lead to: Patients witnessing their healthcare providers making data-informed decisions based on a unified view of their clinical events are more likely to actively engage in their care. This increased engagement often results in better adherence to treatment plans and subsequently leads to improved health outcomes (Improved Patient Engagement). Moreover, Decision-makers gain the ability to allocate healthcare resources more efficiently by prioritizing patients who require immediate attention, this strategic allocation enhances resource utilization, ultimately leading to improved patient outcomes. (Optimized Resource Allocation) and Understanding the progression of a patient's health condition and the impact of various clinical events empowers decisionsub-optimal outcome IDSS4PM Techniques role Analytics Intelligence-Design-Choice reference Analytics Descriptive observe and analyse data analysis CUid & Analytical insight intelligence Predictive predicting the next performanc AI/machine learning Clustering and classification intelligence & design Prescriptive recommending the best possible solution simulation & optimization Advancing recommendation intelligence & choice