CRC Screening Research Article Appendix

E L E C T R O N I C S U P P L E M E N T A R Y M A T E R I A L S Health economic evaluation of colorectal cancer screening policy in Flanders 1. Electronic Supplementary Material 1.1. Epidemiological estimates We used the data from the yearly epidemiological reports from both the Belgian Cancer Registry (SKR) and the Centre for Cancer Detection (CvKO) to construct an adjusted full prevalence of CRC in the population of Flanders. This is analogous to calculating the expected distribution of cancers “if everybody was screened with a perfect test”. This takes into account that the number of clinically presenting cancers is not increasing with this hypothetical test. The graph below shows the age distribution of reported cases, and the smoothed adjustment. Table 1: Distribution of reported cancer stages between participation groups Distribution of reported CRC by stage and participation type Participants Interval cancers nonParticipants CRC St. IV 4.60% 16.90% 23.80% CRC St. III 11.40% 15.20% 21.40% CRC St. II 9.70% 14.40% 19.90% CRC St. I and 0 74.30% 53.50% 34.90% Figure 1: Resulting prevalence curves before and after adjustments. Figure 2: Diagram of transition probability estimation method. To estimate the initial transition probabilities (within the undetected group) to progress through the states, we applied a methodology as described in the diagram above, and in section 2.1 in the article. Using the estimated prevalences per year and state (χi,j in the diagram) we can solve backwards from Stage IV for the required progression rates from the previous state (θi,j for progressions, γi,j for the remainder). Building on the assumption that the adjusted prevalence as above is correct, the diagram can be filled in through a backwards induction process starting with CRC St. IV. Because there are only known outflows from this state (deaths, detections, treatments) we can solve the no-transition rate (1-outflows), and the inflow from CRC St. III to make up the difference. At this point the set of outflows of CRC St. III are full, and the steps can be traced back. We do this for multiple years to take an average transition rate. This process is somewhat naïve, in that it expects transition rates to remain stagnant over time. This could be altered with drift parameters, should enough information be available to estimate these. A benefit of this method is that the resulting transition rates give the model a sense of gravity towards the current state of CRC in the population. I.e. These rates tend to produce a long-run state not too dissimilar from the current state, if no changes in screening occur. Another benefit stems from the ease in which this process could be automated with yearly reported data, meaning that the government could easily use this tool to make a new evaluation with future epidemiological data. The resulting transition rates are shown in the table below. Table 2: resulting transition probabilities within the undetected group. Males Females From Healthy Polyp. Ad. CRC1 CRC2 CRC3 Healthy Polyp. Ad. CRC1 CRC2 CRC3 To Polyp. Ad. CRC1 CRC2 CRC3 CRC4 Polyp. Ad. CRC1 CRC2 CRC3 CRC4 Age 45 0.002 0.070 0.273 0.857 0.710 0.002 0.069 0.268 0.840 0.696 46 0.002 0.067 0.275 0.866 0.718 0.002 0.065 0.268 0.843 0.699 47 0.002 0.065 0.277 0.874 0.726 0.002 0.062 0.268 0.846 0.703 48 0.002 0.063 0.279 0.883 0.735 0.002 0.060 0.268 0.848 0.706 49 0.003 0.062 0.281 0.891 0.744 0.003 0.059 0.268 0.850 0.710 50 0.003 0.061 0.283 0.899 0.753 0.003 0.058 0.268 0.852 0.713 51 0.003 0.061 0.285 0.907 0.761 0.003 0.057 0.268 0.853 0.716 52 0.003 0.061 0.287 0.914 0.769 0.003 0.057 0.268 0.854 0.719 53 0.003 0.061 0.288 0.920 0.777 0.003 0.057 0.268 0.855 0.722 54 0.004 0.062 0.290 0.926 0.785 0.004 0.057 0.267 0.854 0.724 55 0.004 0.063 0.291 0.932 0.792 0.004 0.057 0.267 0.854 0.726 56 0.004 0.064 0.293 0.936 0.799 0.004 0.058 0.266 0.852 0.727 57 0.005 0.065 0.294 0.941 0.806 0.004 0.059 0.266 0.850 0.729 58 0.005 0.066 0.295 0.944 0.812 0.005 0.060 0.265 0.848 0.730 59 0.005 0.068 0.296 0.948 0.819 0.005 0.061 0.264 0.845 0.730 60 0.005 0.070 0.297 0.951 0.825 0.005 0.062 0.263 0.842 0.731 61 0.006 0.072 0.298 0.953 0.830 0.005 0.063 0.262 0.839 0.731 62 0.006 0.074 0.298 0.956 0.836 0.006 0.064 0.261 0.836 0.731 63 0.007 0.076 0.299 0.958 0.841 0.006 0.066 0.260 0.832 0.731 64 0.007 0.078 0.300 0.962 0.854 0.006 0.067 0.259 0.830 0.737 65 0.007 0.080 0.301 0.965 0.866 0.007 0.069 0.258 0.827 0.743 66 0.008 0.083 0.301 0.968 0.877 0.007 0.071 0.257 0.824 0.747 67 0.008 0.085 0.302 0.971 0.887 0.008 0.072 0.255 0.821 0.750 68 0.009 0.088 0.303 0.973 0.896 0.008 0.074 0.254 0.818 0.753 69 0.009 0.091 0.303 0.975 0.904 0.008 0.076 0.253 0.814 0.755 70 0.010 0.094 0.304 0.977 0.912 0.009 0.078 0.252 0.810 0.757 71 0.010 0.097 0.304 0.978 0.919 0.009 0.080 0.251 0.806 0.758 72 0.011 0.100 0.305 0.980 0.926 0.010 0.082 0.250 0.802 0.758 73 0.011 0.103 0.305 0.981 0.933 0.010 0.083 0.248 0.798 0.759 74 0.012 0.106 0.306 0.982 0.939 0.011 0.085 0.247 0.793 0.759 75 0.012 0.102 0.286 0.919 0.879 0.011 0.082 0.230 0.738 0.705 1.2. Screening rates The screening participation is an important dynamic in this model, so we took care to create a methodology that accounted for sufficient real-world complexity. This started with designing two schedule matrices, as shown below. Figure 3: Screen Wave Matrices for the Current (top) and Expanded (bottom) screening scenarios. Fig. 2. These matrices show the logic behind the screening participation for the current and expanded screening scenarios. Blue indicates that this age does not participate in state screening. Green indicates a year-age combination where most people are invited to participate. Red means a year-age combination where people are less likely to be invited. A ripple can be seen in this structure, as a result of previous staggered expansions in the target screening demographic. Age 75 Year 2023 SP SP SP SP SP High Low High High Low Low Low High High High Low Low High Low High Low High Low High Low High Low High Low High SP 2024 SP SP SP SP SP High Low High Low Low High High High Low Low Low High High Low High Low High Low High Low High Low High Low High SP 2025 SP SP SP SP SP High Low High Low High High Low Low Low High High High Low Low High Low High Low High Low High Low High Low High SP 2026 SP SP SP SP SP High Low High Low High Low Low High High High Low Low Low High High Low High Low High Low High Low High Low High SP 2027 SP SP SP SP SP High Low High Low High Low High High Low Low Low High High High Low Low High Low High Low High Low High Low High SP 2028 SP SP SP SP SP High Low High Low High Low High Low Low High High High Low Low Low High High Low High Low High Low High Low High SP 2029 SP SP SP SP SP High Low High Low High Low High Low High High Low Low Low High High High Low Low High Low High Low High Low High SP 2030 SP SP SP SP SP High Low High Low High Low High Low High Low Low High High High Low Low Low High High Low High Low High Low High SP 2031 SP SP SP SP SP High Low High Low High Low High Low High Low High High Low Low Low High High High Low Low High Low High Low High SP 2032 SP SP SP SP SP High Low High Low High Low High Low High Low High Low Low High High High Low Low Low High High Low High Low High SP 2033 SP SP SP SP SP High Low High Low High Low High Low High Low High Low High High Low Low Low High High High Low Low High Low High SP 2034 SP SP SP SP SP High Low High Low High Low High Low High Low High Low High Low Low High High High Low Low Low High High Low High SP 2035 SP SP SP SP SP High Low High Low High Low High Low High Low High Low High Low High High Low Low Low High High High Low Low High SP 2036 SP SP SP SP SP High Low High Low High Low High Low High Low High Low High Low High Low Low High High High Low Low Low High High SP 2037 SP SP SP SP SP High Low High Low High Low High Low High Low High Low High Low High Low High High Low Low Low High High High Low SP 2038 SP SP SP SP SP High Low High Low High Low High Low High Low High Low High Low High Low High Low Low High High High Low Low Low SP 2039 SP SP SP SP SP High Low High Low High Low High Low High Low High Low High Low High Low High Low High High Low Low Low High High SP 2040 SP SP SP SP SP High Low High Low High Low High Low High Low High Low High Low High Low High Low High Low Low High High High Low SP 2041 SP SP SP SP SP High Low High Low High Low High Low High Low High Low High Low High Low High Low High Low High High Low Low Low SP 2042 SP SP SP SP SP High Low High Low High Low High Low High Low High Low High Low High Low High Low High Low High Low Low High High SP Legend: Spontaneous screening only (blue), scheduled invite year (green), unscheduled invite year (red) 70 -74 65 - 69 60 -64 55 - 59 50 - 54 45 - 49 Age 75 Year 2023 SP SP SP SP High High Low High High Low Low Low High High High Low Low High Low High Low High Low High Low High Low High Low High SP 2024 SP SP High High High Low Low High Low Low High High High Low Low Low High High Low High Low High Low High Low High Low High Low High SP 2025 High High High Low Low Low High High Low High High Low Low Low High High High Low Low High Low High Low High Low High Low High Low High SP 2026 High Low Low Low High High High Low Low High Low Low High High High Low Low Low High High Low High Low High Low High Low High Low High SP 2027 High Low High High High Low Low Low High High Low High High Low Low Low High High High Low Low High Low High Low High Low High Low High SP 2028 High Low High Low Low Low High High High Low Low High Low Low High High High Low Low Low High High Low High Low High Low High Low High SP 2029 High Low High Low High High High Low Low Low High High Low High High Low Low Low High High High Low Low High Low High Low High Low High SP 2030 High Low High Low High Low Low Low High High High Low Low High Low Low High High High Low Low Low High High Low High Low High Low High SP 2031 High Low High Low High Low High High High Low Low Low High High Low High High Low Low Low High High High Low Low High Low High Low High SP 2032 High Low High Low High Low High Low Low Low High High High Low Low High Low Low High High High Low Low Low High High Low High Low High SP 2033 High Low High Low High Low High Low High High High Low Low Low High High Low High High Low Low Low High High High Low Low High Low High SP 2034 High Low High Low High Low High Low High Low Low Low High High High Low Low High Low Low High High High Low Low Low High High Low High SP 2035 High Low High Low High Low High Low High Low High High High Low Low Low High High Low High High Low Low Low High High High Low Low High SP 2036 High Low High Low High Low High Low High Low High Low Low Low High High High Low Low High Low Low High High High Low Low Low High High SP 2037 High Low High Low High Low High Low High Low High Low High High High Low Low Low High High Low High High Low Low Low High High High Low SP 2038 High Low High Low High Low High Low High Low High Low High Low Low Low High High High Low Low High Low Low High High High Low Low Low SP 2039 High Low High Low High Low High Low High Low High Low High Low High High High Low Low Low High High Low High High Low Low Low High High SP 2040 High Low High Low High Low High Low High Low High Low High Low High Low Low Low High High High Low Low High Low Low High High High Low SP 2041 High Low High Low High Low High Low High Low High Low High Low High Low High High High Low Low Low High High Low High High Low Low Low SP 2042 High Low High Low High Low High Low High Low High Low High Low High Low High Low Low Low High High High Low Low High Low Low High High SP Legend: Spontaneous screening only (blue), scheduled invite year (green), unscheduled invite year (red) 45 - 49 50 - 54 55 - 59 60 -64 65 - 69 70 -74 The next step was to estimate the invitation rates that correspond to the red/green year-age combinations as seen in the matrices above. To be able to freely change model any changes in the screening schedule, we estimated an invitation rate for scheduled years (green) and unscheduled years (red). Because a notinsignificant percentage is invited outside scheduled years, we incorporated this is the model. We fitted 3rd order polynomials to the data, stratified by age and schedule. We applied a ceiling to the maximum rates to never fall above or below observed rates. This same process was performed on the probability to participate in screening, conditional on being invited. Figure 4: Invitation and Participation Rates This two-step differentiation between invitation and participation allowed us to separate out two different dynamics. First, that invitations decrease with age (as more people opt out or are excluded for medical reasons), and also become less stratified according to the schedule. Second, that intrinsically people become more likely to participate as they age. Using these three principles together (wave matrices, invitation rate, and participation rate) on past years allows to compare the resulting number of tests versus the reported number of tests (averaged over 4 years), as shown in the figure below. Figure 5: Number of tests, predicted and reported. Overall, the “retrodiction” tends to follow the actual pattern well. It tends to underestimate the peaks and overestimate the troughs, but to only a small amount. The net result is a slight underestimation of the number of tests of less than 5% on the short run. This results primarily from the overestimation of exclusions in the younger age groups, which is likely to become more accurate in the long run. 1.3. Input Costs and Disutilities in the Model The following table shows the disutilities that have been applied to each health state. These disutilities are applied relative to the healthy utility score by age-year. The healthcare costs aggregated by specific health state are also shown, as is a comparison to the cost applied in the prior model [1]. The sources for these specified below the table. All costs are indexed to 2022 using the Healthcare costs index. Table 3: Costs and disutilities Disutilities and costs per health state, aggregated values Health State Disutility Ref. Healthcare cost Prior Values* Ref. Ad. Polyp 0.065 [1] n/a n/a Unidentified CRC St. I 0.085 [1] n/a n/a Unidentified CRC St. II 0.085 [1] n/a n/a Unidentified CRC St. III 0.140 [1] n/a n/a Unidentified CRC St. IV 0.655 [1] n/a n/a Treatment CRC St. I 0.170 [1] € 15,263 € 12,808 [1, 2] Treatment CRC St. II 0.170 [1] € 27,071 € 22,716 [1, 2] Treatment CRC St. III 0.280 [1] € 33,251 € 31,351 [1, 2] Treatment CRC St. IV 0.655 [1] € 39,634 € 37,369 [1, 2] Follow-up years 1-4 CRC St. I 0.128 [1] € 9,730 € 7,795 [1, 2] Follow-up years 1-4 CRC St. II 0.128 [1] € 7,349 € 5,888 [1, 2] Follow-up years 1-4 CRC St. III 0.210 [1] € 4,733 € 3,992 [1, 2] Follow-up years 1-4 CRC St. IV 0.655 [1] € 13,837 € 11,669 [1, 2] Follow-up years >4 CRC St. I 0.064 [1] € 241 € 202 [1, 2, 3] Follow-up years >4 CRC St. II 0.064 [1] € 241 € 202 [1, 2, 3] Follow-up years >4 CRC St. III 0.105 [1] € 4,733 € 3,992 [1, 2] Follow-up years >4 CRC St. IV 0.328 [1] € 13,837 € 11,669 [1, 2] False positive result 0.021 [1] n/a n/a Colonoscopy 0.003 [1] € 241 € 202 [3] Polypectomy 0.014 [1] € 540 € 393 [3] [1] Pil, L., Fobelets, M., Putman, K., Trybou, J., & Annemans, L. (2016). Cost-effectiveness and budget impact analysis of a population-based screening program for colorectal cancer. European journal of internal medicine, 32, 72-78. DOI: 10.1016/j.ejim.2016.03.031 [2] Pacolet, J., De Coninck, A., Hedebouw, G., Cabus, S., & Spruytte, N. (2011). De medische en niet-medische kosten van kankerpatiënten. ISSN: 978-90-5550-481-7 [3] RIZIV Rijksinstituut voor ziekteen invaliditeitsverzekering. NomenSoft. Available at: https://www.riziv.fgov.be/webprd/appl/pnomen/Search.aspx?lg=N. 1.4. Results: Total costs and benefits The following tables show the resulting total costs and benefits between the three scenarios. Costs are separated between “screening costs” (programme costs), healthcare costs for all resulting treatments, and the societal costs (from lost labour production). These are shown in the deterministic case, the probabilistic average, and the standard error. Finally, these are also shown as a proportion to the person years in the model. The final table also shows the general breakdown of healthcare costs per scenario, into costs made for screening (FIT and fixed costs), colonoscopies (and polypectomies), cancer treatments, and follow-up costs. The last two categories are the main drivers of total costs. Table 4: Total costs and benefits per scenario Costs per scenario Total (20 year, in Millions) Average per year (in Thousands) Det. Prob. Avg. St. Err. Det. Prob. Avg. St. Err. Avg. p.p.y. Screening costs No Screening 0 M 0 M (0.0) 0 K 0 K (0.0) 0.0 Current Screening 75 M 79 M (7.6) 3,404 K 3,781 K (364.2) 0.6 Expanded Screening 79 M 84 M (8.4) 3,594 K 3,991 K (401.1) 0.6 Healthcare costs No Screening 4,817 M 5,102 M (243.9) 219,049 K 242,973 K (11615.7) 35.7 Current Screening 4,723 M 4,999 M (2595) 214,703 K 238,034 K (12359.4) 34.9 Expanded Screening 4,719 M 4,994 M (260.4) 214,507 K 237,823 K (12401.3) 34.8 Labour costs No Screening 13,509 M 19,418 M (0.9) 614,046 K 924,668 K (42.9) 135.7 Current Screening 13,456 M 19,337 M (0.9) 611,618 K 920,813 K (42.9) 134.9 Expanded Screening 13,453 M 19,335 M (0.9) 611,518 K 920,705 K (42.9) 134.8 Table 5: Breakdown of Healthcare cost per scenario and category Cost breakdown per scenario No Screening Current Screening Expanded Screening Screening Costs 0.6% 2.0% 2.1% Colonoscopies and polypectomies 2.6% 5.2% 5.3% Treatment costs 49.5% 44.2% 43.9% Post-Treatment costs 47.3% 48.7% 48.7% Benefits per scenario QALYs (total) Per person-year Life years (total) Det. Prob. Avg. St. Err. Det. Prob. Avg. Det. Prob. Avg. St. Err. No Screening 79,073K 79,007K (5175.4) 0.5474 0.5469 113,183 K 113,196 K (5309) Current Screening 79,266K 79,197K (5184.3) 0.5487 0.5482 113,384 K 113,395 K (5319) Expanded Screening 79,274K 79,206K (5184.7) 0.5488 0.5483 113,396 K 113,407 K (5320)