Global variation in diabetes diagnosis and prevalence based on fasting glucose and hemoglobin A1c

This is a self-archived version of an original article. This version may differ from the original in pagination and typographic details. Author(s): Title: Year: Version: Copyright: Rights: Rights url: Please cite the original version: CC BY 4.0 https://creativecommons.org/licenses/by/4.0/ Global variation in diabetes diagnosis and prevalence based on fasting glucose and hemoglobin A1c © The Author(s) 2023 Published version NCD Risk Factor Collaboration (NCD-RisC) NCD Risk Factor Collaboration (NCD-RisC). (2023). Global variation in diabetes diagnosis and prevalence based on fasting glucose and hemoglobin A1c. Nature Medicine, 29, 2885-2901. https://doi.org/10.1038/s41591-023-02610-2 2023 Nature Medicine | Volume 29 | November 2023 | 2885–2901 2885 nature medicine https://doi.org/10.1038/s41591-023-02610-2Article Global variation in diabetes diagnosis and prevalence based on fasting glucose and hemoglobin A1c NCD Risk Factor Collaboration (NCD-RisC)* Fasting plasma glucose (FPG) and hemoglobin A1c (HbA1c) are both used to diagnose diabetes, but these measurements can identify different people as having diabetes. We used data from 117 population-based studies and quantified, in different world regions, the prevalence of diagnosed diabetes, and whether those who were previously undiagnosed and detected as having diabetes in survey screening, had elevated FPG, HbA1c or both. We developed prediction equations for estimating the probability that a person without previously diagnosed diabetes, and at a specific level of FPG, had elevated HbA1c, and vice versa. The age-standardized proportion of diabetes that was previously undiagnosed and detected in survey screening ranged from 30% in the high-income western region to 66% in south Asia. Among those with screen-detected diabetes with either test, the age-standardized proportion who had elevated levels of both FPG and HbA1c was 29–39% across regions; the remainder had discordant elevation of FPG or HbA1c. In most lowand middle-income regions, isolated elevated HbA1c was more common than isolated elevated FPG. In these regions, the use of FPG alone may delay diabetes diagnosis and underestimate diabetes prevalence. Our prediction equations help allocate finite resources for measuring HbA1c to reduce the global shortfall in diabetes diagnosis and surveillance. Diabetes is associated with debilitating complications such as amputation, vision loss and renal failure, and with increased risk of cardiovascular events, dementia, some cancers and infectious diseases such as severe COVID-19 and tuberculosis 1–6 . The diagnostic criteria for diabetes have evolved over time to incorporate hemoglobin A1c (HbA1c), which is a measure of long-term glycemic status and more convenient to measure for patients than fasting glucose or the 2-h oral glucose tolerance test (OGTT)7–10. In contemporary guidelines, any one or the combination of fasting plasma glucose (FPG), OGTT and HbA1c may be used to diagnose diabetes 10–14 . With the exception of diagnosis of gestational diabetes, OGTT is now rarely used in clinical practice or population surveillance because of the inconvenience related to the glucose load, 2-h time frame and the two blood draws required for the test 15,16 . FPG and HbA1c, which are both used in clinical practice and epidemiological research and surveillance, measure different glycemic features, namely basal glucose level (FPG) and average glucose level in the previous 2–3 months (HbA1c)17. Therefore, individuals may have elevated levels of one or both biomarkers, and FPG and HbA1c may classify different people as having diabetes 9,10 . Diabetes also has a long subclinical period defined by hyperglycemia and can remain undiagnosed without screening or other mechanisms for early identification 18 . Some studies have assessed sensitivity and specificity of diabetes diagnosis using either FPG or HbA1c relative to the OGTT or have compared diabetes prevalence based on these different glycemic biomarkers, but most did not provide a direct comparison of HbA1c and FPG19–21. Most population-based studies on the concordance and discordance Received: 15 March 2023 Accepted: 25 September 2023 Published online: 9 November 2023 Check for updates *A list of authors and their affiliations appears at the end of the paper. e-mail: m aj id .e zz at i@ im pe ri al.ac.uk Nature Medicine | Volume 29 | November 2023 | 2885–2901 2886 Article https://doi.org/10.1038/s41591-023-02610-2 seven of eight world regions (Extended Data Table 1). We had no study that measured both FPG and HbA1c from the region of Oceania, which consists of Pacific island nations. The number of studies in other regions ranged from seven in sub-Saharan Africa to 48 in the high-income western region (Table 1). The mean age of study participants was 50 years and 56% of participants were women. Of the 117 studies with data on glycemic variables, 113 (97%) with 351,270 participants (96% of all participants) also had data on body-mass index (BMI); the remaining four studies either did not collect anthropometric information or only had self-reported height and weight data. Screen-detected diabetes by FPG and HbA1c Across all studies, 16% of participants had diagnosed or previously undiagnosed screen-detected diabetes. Diagnosed diabetes was calculated based on reporting a previous diagnosis and screen-detected diabetes as having FPG and/or HbA1c levels at or above the thresholds of 7.0 mmol l −1 and 6.5% (refs. 10–13) (Fig. 2). After age-standardization, the total prevalence of diabetes became 12%. The age-standardized prevalence of diagnosed and screen-detected diabetes were 7% and 5%, respectively. Those without a previous diabetes diagnosis had a lower BMI than those with a previous diagnosis in every region, by an average of 2.9 kg m −2 across all studies (Table 1). Among those without a previous diagnosis, participants with screen-detected diabetes (FPG ≥7.0 mmol l −1 and/or HbA1c ≥ 6.5%) had a mean BMI that was higher than those who did not have diabetes (FPG < 7.0 mmol l −1 and HbA1c < 6.5%) by an average of 2.4 kg m−2. In most regions, age-standardized diabetes prevalence was slightly lower than crude prevalence, except south Asia where the participants were on average younger than in other regions (Table 1). Regionally, the age-standardized total diabetes prevalence (the combination of diagnosed and screen-detected diabetes) ranged from ~9% in the high-income western region to ~21% in south Asia and sub-Saharan Africa. The age-standardized proportion of diabetes that was previously undiagnosed, and was detected in the screening via the survey, was highest (66%) in studies from south Asia, and lowest (<35%) in studies from the high-income western region, central and eastern Europe, and the region of central Asia, Middle East and north Africa. Two studies in sub-Saharan Africa were from Mauritius, a country that is different demographically and economically from most other countries in the region. When these studies were removed, total age-standardized diabetes prevalence in sub-Saharan Africa declined from 21% to 13% and the proportion who were previously undiagnosed increased from 46% to 53% (Extended Data Fig. 2). Across all studies together, 29% of participants with screendetected diabetes had isolated elevated FPG, 37% had isolated elevated HbA1c and 34% had elevated levels of both. These global proportions were the same before and after age-standardization. There was substantial variation across regions in the composition of screen-detected diabetes across these three groups, both in terms of whether both biomarkers were elevated or only one, and in the case of the latter, whether the elevated biomarker was FPG or HbA1c (Fig. 2). Regionally, the shares of participants in these three groups changed little after age-standardization, and we report the age-standardized results here. The age-standardized proportion of those with screen-detected diabetes who had elevated levels of both FPG and HbA1c ranged from 29–39% across regions. The remaining 61–71% of participants with screen-detected diabetes had discordant FPG and HbA1c elevations. Isolated elevated HbA1c made up 54% of participants with screen-detected diabetes in sub-Saharan Africa, and 47% in the region of central Asia, Middle East and north Africa. In these regions, isolated elevated FPG accounted for <17% of all screen-detected diabetes. In contrast, 55% of participants with screen-detected diabetes in central and eastern Europe, and 46% in high-income western region, had isolated elevated FPG. The correlation coefficient between FPG and HbA1c among participants without previous diagnosis of diabetes ranged of diabetes diagnosis using FPG versus HbA1c have been conducted in a single country or region 14,22–42 and the only multi-country study 43 used data largely from high-income western countries. Therefore, there are scant data on how the concordance and discordance of FPG and HbA1c in classifying diabetes vary across regions in the world, and on the factors associated with this variation. The lack of data on the regional variation in diabetes identified based on FPG versus HbA1c means that we cannot quantify the full extent of the global diabetes epidemic and its regional variation, because diabetes prevalence is measured and reported using a single glycemic biomarker in most population-based surveys and analyses44–46. For example, in the latest global analysis44, only ~15% of surveys had measured both FPG and HbA1c. We assembled a global database of population-based studies that had measured both FPG and HbA1c. Using these data, we quantified the regional variation in the extent of diabetes diagnosis, with diabetes defined as in the Methods. We also quantified, among those who were previously undiagnosed and were detected as having diabetes through screening in the survey, the concordance and discordance of having FPG and HbA1c above common diagnostic thresholds (7.0 mmol l−1 for FPG and 6.5% for HbA1c). We refer to this group as screen-detected diabetes, which is an epidemiological definition, because many clinical guidelines recommend two measurements for diabetes diagnosis 10–13 . We then used regression analysis to examine what individual and study-level factors were associated with whether participants with screen-detected diabetes were identified by elevated FPG, elevated HbA1c or elevated levels of both. It has been shown that having elevated levels of both biomarkers has high positive predictive value for subsequent clinical diagnosis and risk of complications 14,47 , and hence this group is similar to clinically diagnosed diabetes. Finally, we leveraged the global coverage of the dataset and its large sample size to develop prediction equations that estimate, for any given FPG level, the probability that a person without previously diagnosed diabetes would have HbA1c above the clinical threshold for diabetes had it been measured, and vice versa. We aimed to develop and validate global and generalizable prediction equations that account for both personal characteristics and regional differences. These equations serve three purposes. First, they allow more efficient use of finite diagnostic resources, by identifying some people with belowor near-threshold level for one biomarker (for example, FPG) for measurement of another (for example, HbA1c). Second, they allow the estimation of the probability that a person with a screen-detected elevated level of one biomarker would also have an elevated level of the other, as a confirmation of diabetes status14,47. Finally, the prediction equations could improve diabetes surveillance by allowing estimation of prevalence of diabetes based on both FPG and HbA1c in health surveys that have measured only one of these biomarkers. Results Data sources We used data collated by the NCD Risk Factor Collaboration (NCD-RisC), a global consortium of population-based health examination surveys and studies with measurement of both FPG and HbA1c, and with data on previous diagnosis of diabetes, as described in the Methods. The criteria for including and excluding studies are stated in Methods. Within each study, we excluded participants who had missing data or were pregnant, under 18 years of age or from follow-up rounds of studies that had multiple measurements of the same cohort over time (Fig. 1). After exclusions, we used data on 601,307 participants aged 18 years and older with information on whether they had been previously diagnosed with diabetes, of whom 364,825 participants also had measured FPG and HbA1c. The difference between the number of participants with data on previous diagnosis and with biomarker data is mostly because many studies do blood tests on a subsample of those with questionnaire data. These participants were from 117 studies whose mid-year was from 2000 to 2021 in 45 countries from Nature Medicine | Volume 29 | November 2023 | 2885–2901 2887 Article https://doi.org/10.1038/s41591-023-02610-2 from 0.51 in central and eastern Europe to 0.76 in sub-Saharan Africa (Extended Data Fig. 3). Association with individual and study characteristics Some participant and study-level characteristics were associated with whether screen-detected diabetes was manifested as elevated levels of FPG, HbA1c or both (Table 2). Among those with screen-detected diabetes, male sex was associated with a higher probability of having elevated FPG, either alone (prevalence ratio (PR) = 1.10; 95% credible interval (CrI) 1.07–1.14) or together with elevated HbA1c (1.07; 1.03–1.11), and with a lower probability of having isolated elevated HbA1c (0.86; 0.83–0.89). Older age was associated with a lower probability of having elevated FPG, alone (PR = 0.97 per decade of age; 0.96–0.98) or together with elevated HbA1c (PR = 0.97; 0.96–0.99) and a higher probability of having isolated elevated HbA1c (1.05; 1.04–1.06). Higher BMI was associated with a higher probability of having concordant elevation of FPG and HbA1c (PR = 1.07 per 5 units; 1.06–1.08) and a lower probability of having isolated elevated FPG (PR = 0.92; 0.90–0.93). At the study level, in studies that used a portable device to measure HbA1c, the composition of screen-detected diabetes was shifted toward more isolated elevated HbA1c, but the estimates for this association had wide confidence intervals because the great majority of studies in our analysis had measured glucose and HbA1c in a laboratory. Neither the year of study nor the percentage of participants with diabetes who had reported previous diagnosis were associated with the composition of screen-detected diabetes. After adjustment for participant and study characteristics, regional differences remained in the composition of screen-detected diabetes (Table 2). After adjustment for these factors, the composition of screen-detected diabetes, in terms of having elevated FPG and HbA1c in isolation or together, was statistically indistinguishable between the high-income western region and central and eastern 727,588 participants from 131 studies 11,470 participants without data on BMIg 294,150 participants who had FPG <7 mmol l–1 and HbA1c <6.5%h 327,554 participants from 117 studies without previous diagnosis of diabetes and with complete data on FPG and HbA1c Used to examine the prevalence and biomarker composition of screen-detected diabetes 316,084 participants from 113 studies without previously diagnosed diabetes who also had BMI measurement Used to develop prediction equations 21,934 participants from 109 studies without previously diagnosed diabetes who also had BMI measurement and had FPG ≥7 mmol l–1 and/or HbA1c ≥6.5% Used to examine the predictors of FPGHbA1c concordance versus discordance 601,307 participants from 117 studies with data on previous diagnosis of diabetes Used to calculate the prevalence of diagnosed diabetes 4,830 participants who were pregnanta participants with missing sex or age participants aged <18 years participants from follow-up rounds of studies that had 1,183 23,001 diabetes 32,326 64,941 multiple measurements of the same cohort over timeb participants with missing information on previous diagnosis of 57,785 participants who had been previously diagnosed with diabetes participants from one specific area in one study in Pakistand participants with FPG <2 or >30 mmol l–1 or HbA1c <3% or >18%e 203,604 1,216 11,107 participants with implausible combinations of FPG and HbA1cf 41 participants whose FPG or HbA1c were not measured by design or missingc Fig. 1 | Flowchart of data cleaning and use. aExcluded because glucose metabolism changes during pregnancy. bData from the first available measurement were used for these participants. cSome surveys only measured glycemic biomarker on a subset of participants for logistic or budget reasons. dExcluded because glycemic measurements in these participants were systematically different from the rest from the same study, possibly because the specific area had high prevalence of thalassemia94. eExcluded because such values are more likely to be due to data recording error than values within the range. fWe removed participants for implausible pairs of FPG and HbA1c using the method of local outlier factor (LOF)95. This approach detects data combinations that are extremes in the joint density of the variable pairs (for example, a participant with FPG of 5 mmol l−1 and HbA1c of 17%, or with FPG of 28 mmol l−1 and HbA1c of 5%). We identified extremes as those measurements whose measure of local density by LOF method is less than half of the average of their 100 nearest neighbors. gIncluding all 2,436 participants from four studies that did not measure BMI. hIncluding all 3,455 participants from four studies in which all individuals without previously diagnosed diabetes had FPG < 7.0 mmol l−1 and HbA1c < 6.5%. Nature Medicine | Volume 29 | November 2023 | 2885–2901 2888 Article https://doi.org/10.1038/s41591-023-02610-2 Europe. In other regions, elevated HbA1c was a more common form of screen-detected diabetes than in the high-income western region, in isolation (PR ranging 1.42–2.20 across these regions) or together with elevated FPG (PR ranging 1.31–1.52 in east and southeast Asia and the Pacific; south Asia; sub-Saharan Africa). In all regions, isolated elevated FPG was less common than in the high-income western region (PR ranging 0.24–0.51). Prediction equations We developed nine prediction equations (Extended Data Table 2) that estimate, for any given FPG level, the probability that a person without previously diagnosed diabetes would have HbA1c above the clinical threshold for diabetes had it been measured, and vice versa. The variables in the prediction equations included FPG as well as sex, age, BMI, whether FPG was measured in a laboratory or using a portable device, and region. We assessed the performance of the models in predicting (1) individual participants’ status of having HbA1c ≥ 6.5% based on their FPG and (2) the prevalence of HbA1c ≥ 6.5% for an entire study. We used the same method for predicting the probability of having FPG ≥ 7.0 mmol l−1 based on HbA1c. The performance at the individual level reflects how well the prediction equation works for triaging patients for further measurement for diabetes, and the performance at study (or population) level assesses how well the prediction equation works for diabetes surveillance. Most of the prediction equations had acceptable performance for estimating the probability that a person without diagnosed diabetes at a specific level of one glycemic biomarker (FPG or HbA1c) was above the clinical threshold for the other (Extended Data Tables 3 and 4). Specifically, the C-statistic ranged 0.85–0.90 for prediction equations that used either biomarker to predict the elevated level of the other. The mean errors were between −0.18 and −0.65 percentage points and the mean absolute errors were between 2.32 and 3.30 percentage points. The best-performing models for predicting whether participants had HbA1c ≥ 6.5% using FPG measurement included BMI and region-specific terms for FPG, referred to Table 1 | Characteristics of studies and participants included in the analysis: all participants, participants without diagnosed diabetes, and participants without diagnosed diabetes who had FPG ≥7.0 mmol l−1 and/or HbA1c ≥6.5% Number of studies Number of countries (% of all countries in the region or world) Median year of studies Number of participants Percent female (%) Mean (s.d.) age (years) Mean FPG (mmol l−1)Mean HbA1c (%) Mean BMI (kg m−2) All participants Central and eastern Europe 84 (20%) 2012 51,352 55.6 55 (11) 5.8 5.5 28.2 Central Asia, Middle East and north Africa 10 5 (18%) 2015 73,109 54.4 47 (15) 5.7 5.9 27.7 High-income western 48 11 (41%) 2010 190,276 53.2 53 (18) 5.6 5.5 27.8 Latin America and the Caribbean 17 11 (31%) 2016 75,257 62.3 48 (18) 5.7 5.7 28.3 South Asia 82 (29%) 2012 87,404 54.4 42 (14) 5.9 6.0 23.1 East and southeast Asia and the Pacific 19 7 (41%) 2012 112,854 56.2 52 (16) 5.6 5.7 24.0 Sub-Saharan Africa 75 (10%) 2014 11,055 62.6 49 (14) 6.1 6.2 26.3 All studies 117 45 (22%) 2012 601,307 55.6 50 (17) 5.7 5.7 26.4 Participants without diagnosed diabetes Central and eastern Europe 84 (20%) 2012 12,086 52.2 49 (14) 5.4 5.4 27.4 Central Asia, Middle East and north Africa 10 5 (18%) 2015 46,886 55.1 46 (14) 5.3 5.6 27.5 High-income western 48 11 (41%) 2010 100,140 53.9 52 (16) 5.4 5.3 27.4 Latin America and the Caribbean 17 11 (31%) 2016 38,524 60.8 48 (17) 5.3 5.4 28.0 South Asia 82 (29%) 2012 28,554 52.7 41 (14) 5.6 5.7 24.0 East and southeast Asia and the Pacific 19 7 (41%) 2012 92,900 56.6 51 (16) 5.4 5.6 23.9 Sub-Saharan Africa 75 (10%) 2014 8,464 62.2 48 (14) 5.6 5.8 26.2 All studies 117 45 (22%) 2012 327,554 55.7 49 (16) 5.4 5.5 26.2 Participants without diagnosed diabetes who had FPG ≥ 7.0 mmol l−1 and/or HbA1c ≥ 6.5% Central and eastern Europe 84 (20%) 2012 551 41.7 58 (11) 8.0 6.4 31.3 Central Asia, Middle East and north Africa 10 5 (18%) 2015 3,328 52.0 55 (13) 7.7 7.3 30.2 High-income western 44 11 (41%) 2009 4,422 43.1 62 (13) 7.9 6.7 31.0 Latin America and the Caribbean 17 11 (31%) 2016 2,718 63.0 55 (15) 8.4 7.3 30.4 South Asia 82 (29%) 2012 4,612 51.7 47 (13) 8.0 7.4 26.0 East and southeast Asia and the Pacific 19 7 (41%) 2012 6,157 52.0 58 (13) 8.1 7.0 26.1 Sub-Saharan Africa 75 (10%) 2014 1,257 60.5 55 (11) 7.5 7.2 28.7 All studies 113 45 (22%) 2013 23,045 51.7 56 (14) 8.0 7.1 28.4 Nature Medicine | Volume 29 | November 2023 | 2885–2901 2889 Article https://doi.org/10.1038/s41591-023-02610-2 as models 5 and 8 in Extended Data Tables 2 and 3. These two models had similar C-statistic. Model 5 had the smallest deviation and model 8 had the smallest bias. The addition of sex interaction terms did not improve model performance. The best models for predicting whether participants had FPG ≥ 7.0 mmol l −1 using HbA1c measurement were also models 5 and 8 (Extended Data Tables 2 and 4). The coefficients of these models are shown in Extended Data Tables 5 and 6. In Fig. 3, the coefficients from model 8 were used to calculate the probability that a person without a history of diabetes diagnosis, based on measurement of a single glycemic biomarker that is below the clinical threshold, would have elevated level of the other (elevated HbA1c at a specific FPG and BMI level (Fig. 3a) or elevated FPG at a specific HbA1c and BMI level (Fig. 3b)). For example, in south Asia, people aged 55 years and older, without a previous diabetes diagnosis, with obesity (BMI ≥ 30 kg m−2), whose FPG is 6.5–6.9 mmol l−1 have a 29–63% probability of having elevated HbA1c. In contrast, the probability of having elevated HbA1c remained no higher than 17% for men and women of the same age and FPG level in the high-income western region and central and eastern Europe, which means that screen-detected diabetes that is manifested as isolated elevated HbA1c is relatively rare in these two regions. For those whose HbA1c was measured, the probability of having elevated FPG was below 30% in every region except central and eastern Europe; the probability surpassed 20% only in those with high BMI and HbA1c levels. In Fig. 4, the coefficients from model 8 were used to calculate the probability that a person without a history of diabetes diagnosis, based on measurement of a single glycemic biomarker that is above the clinical threshold, would have elevated level of the other (elevated HbA1c at a specific FPG and BMI level (Fig. 4a) or elevated FPG at a specific HbA1c and BMI level (Fig. 4b)). These results show that people without a previous diagnosis who had an elevated level of one diabetes biomarker had varying probabilities of also being elevated for the other depending on region, age, sex and BMI. In particular, for those with screen-detected elevated HbA1c, the probability of also a Crude Age-standardized 10 All studies High-income western Central and eastern Europe East and southeast Asia and the Pacific Latin America and the Caribbean Central Asia, Middle East and north Africa South Asia Sub-Saharan Africa 0 10 20 Proportion of all participants (%) Diagnosed diabetes Without previous diagnosis of diabetes Diagnosed diabetes Elevated levels of both FPG and HbA1c Isolated elevated HbA1c Isolated elevated FPG b Crude Age-standardized 0 25 50 75 100 0 25 50 75 100 All studies Central Asia, Middle East and north Africa East and southeast Asia and the Pacific Latin America and the Caribbean Sub-Saharan Africa South Asia High-income western Central and eastern Europe Proportion of participants with screen-detected diabetes (%) All studies High-income western Central and eastern Europe East and southeast Asia and the Pacific Latin America and the Caribbean Central Asia, Middle East and north Africa South Asia Sub-Saharan Africa Fig. 2 | Extent and composition of diagnosed and screen-detected diabetes by region. a, Crude and age-standardized proportion of participants with diagnosed or screen-detected diabetes and, for those without previous diagnosis, whether they had isolated elevated FPG (FPG ≥ 7.0 mmol l−1 and HbA1c < 6.5%), isolated elevated HbA1c (HbA1c ≥ 6.5% and FPG < 7.0 mmol l−1) or elevated levels of both. b, Crude and age-standardized proportion of participants with screen-detected diabetes who had isolated elevated FPG, isolated elevated HbA1c or elevated levels of both, by region. The contents in b are the same as the segment of a that is below the zero line, scaled to 100% so that the composition of screen-detected diabetes can be compared across regions, regardless of its total prevalence. Having elevated levels of both biomarkers has high positive predictive value for subsequent clinical diagnosis and risk of complications14,47 and hence this group is similar to clinically diagnosed diabetes. In a, regions are ordered by the total proportion of participants who had diagnosed and screen-detected diabetes. In b, regions are ordered by the crude proportion of participants with screen-detected diabetes who had elevated levels of both FPG and HbA1c. Extended Data Fig. 1 provides sex-specific results. Nature Medicine | Volume 29 | November 2023 | 2885–2901 2890 Article https://doi.org/10.1038/s41591-023-02610-2 having FPG ≥ 7.0 mmol l −1 surpassed 90% in some region-age-BMI combinations. The exceptions were south Asia and Latin America and the Caribbean, where isolated elevated HbA1c and isolated elevated FPG were both common and hence only partially predicted one another. Discussion Our analysis of pooled global data showed that the use of either FPG or HbA1c alone might substantially underestimate the burden of diabetes relative to the number of people who would have elevated levels of either glycemic measure, especially in lowand middle-income countries where diagnosis rates are currently low. We also presented prediction equations to help allocate finite resources for measurement of HbA1c in settings where FPG (but not HbA1c) is routinely measured due to logistic or cost constraints. The prediction equations can also be used to enhance diabetes surveillance, to adjust the estimated prevalence in the majority of population-based health surveys which measure only one biomarker. Our results, based on a large number of studies from different regions of the world, are consistent with a previous smaller study with data from mostly high-income western countries43 and with the collective results from studies done in individual countries 22–42 in identifying substantial variation in diabetes classified by FPG versus HbA1c across regions. None of the previous studies had sufficient geographical coverage or participants to robustly quantify regional differences in how those with previously undiagnosed diabetes that were identified based on elevation of FPG and HbA1c, in isolation or together, as we did. A study using baseline data from the ORIGIN trial 48 , which covered people with diabetes or prediabetes from 40 countries, did not quantify the concordance and discordance of diabetes based on different biomarkers but its graphical results indicated smaller differences in FPG-HbA1c relationship between Europe and north America than between these regions and Asia or south America. We found that sex, age and BMI were predictors of having concordant versus discordant elevated FPG and elevated HbA1c, which is consistent with results from studies in individual countries 22,32,34,40,49 . Finally, to our knowledge, our prediction equations are the only global and generalizable tool for predicting the probability of being classified as having diabetes based on one glycemic biomarker based on measurement of another. A previous regression related HbA1c to average glucose50 (but not fasting glucose). This relationship is currently used by the American Diabetes Association for assessing glycemic control51 and not for inferring new diagnosis of diabetes. It used data from only 507 individuals, 422 of whom were non-Hispanic White. The data came from ten centers, of which nine were in the United States and Europe. Over half (268) had type 1 diabetes, which is the less common form of diabetes in adults. The conversions did not account for other traits such as BMI and age, nor was the performance of the prediction equation validated in data that were not used in its derivation. Table 2 | Association of whether screen-detected diabetes is manifested as isolated elevated FPG, isolated elevated HbA1c or elevated levels of both with individual and study characteristics Isolated elevated FPG Isolated elevated HbA1c Elevated levels of both PR CrI Posterior probability PR CrI Posterior probability PR CrI Posterior probability Region High-income western Reference Reference Reference Central and eastern Europe 1.16 0.73–1.86 0.259 0.62 0.35–1.09 0.049 0.83 0.61–1.12 0.115 Latin America and the Caribbean 0.48 0.32–0.72 <0.001 1.42 0.93–2.16 0.053 1.16 0.91–1.46 0.109 East and southeast Asia and the Pacific 0.51 0.35–0.73 <0.001 1.53 1.04–2.25 0.015 1.35 1.10–1.67 0.002 South Asia 0.24 0.13–0.44 <0.001 1.65 0.89–3.10 0.056 1.52 1.08–2.15 0.009 Central Asia, Middle East and north Africa 0.33 0.20– 0.54 <0.001 2.20 1.31–3.67 0.001 1.06 0.80–1.40 0.342 Sub-Saharan Africa 0.33 0.19–0.57 <0.001 1.65 0.92–2.94 0.045 1.31 0.96–1.79 0.045 Sex Women Reference Reference Reference Men 1.10 1.07–1.14 <0.001 0.86 0.83– 0.89 <0.001 1.07 1.03-1.11 <0.001 Age (per 10 years of age) 0.97 0.96– 0.98 <0.001 1.05 1.04–1.06 <0.001 0.97 0.96–0.99 <0.001 BMI (per 5 kg m−2)0.92 0.90– 0.93 <0.001 0.99 0.98–1.01 0.137 1.07 1.06–1.08 <0.001 Study year (per 5 years of time) 1.01 0.89–1.14 0.447 1.05 0.92–1.20 0.240 1.06 0.99–1.14 0.048 Percent people with diabetes who had been diagnosed before (per 10 percentage points) 0.98 0.89–1.09 0.380 0.98 0.88–1.09 0.354 1.05 0.99–1.11 0.046 Measurement of FPG Laboratory Reference Reference Reference Portable device 1.71 1.00–2.91 0.025 0.89 0.51–1.56 0.338 0.87 0.64–1.16 0.169 Measurement of HbA1c Laboratory Reference Reference Reference Portable device 0.33 0.16–0.68 0.001 2.13 1.05–4.20 0.018 0.54 0.35–0.81 0.002 The association with each variable is reported as prevalence ratios (PRs), adjusted for all other variables in the table, in the regression models described in the Methods, in which data from individual participants with screen-detected diabetes were used. Extended Data Table 7 shows results excluding studies that had measured FPG in capillary whole blood using a portable device. CrI, credible interval. Nature Medicine | Volume 29 | November 2023 | 2885–2901 2891 Article https://doi.org/10.1038/s41591-023-02610-2 a 5.3 5.7 6.1 6.5 6.9 5.3 5.7 6.1 6.5 6.9 5.3 5.7 6.1 6.5 6.9 5.3 5.7 6.1 6.5 6.9 5.3 5.7 6.1 6.5 6.9 5.3 5.7 6.1 6.5 6.9 FPG (nmol l–1) High-income western Central and eastern Europe Latin America and the Caribbean East and southeast Asia and the Pacific Central Asia, Middle East and north Africa South Asia Sub-Saharan Africa Women High-income western Central and eastern Europe Latin America and the Caribbean East and southeast Asia and the Pacific Central Asia, Middle East and north Africa South Asia Sub-Saharan Africa Men 20253035 20253035 20253035 20253035 20253035 20253035 20253035 20253035 20253035 20253035 20253035 20253035 20253035 20253035 BMI (kg m–2)BMI (kg m–2) 75 years 65 years 55 years 45 years 35 years 25 years Predicted probability of having HbA1c ≥6.5% <5% 5–9% 10–19% 20–29% 30–39% 40–49% <5% 5–9% 10–19% 20–29% 30–39% 40–49% 50–63% b 5.4 5.6 5.8 6.0 6.2 6.4 5.4 5.6 5.8 6.0 6.2 6.4 5.4 5.6 5.8 6.0 6.2 6.4 5.4 5.6 5.8 6.0 6.2 6.4 5.4 5.6 5.8 6.0 6.2 6.4 5.4 5.6 5.8 6.0 6.2 6.4 HbA1c (%) 75 years 65 years 55 years 45 years 35 years 25 years Predicted probability of having FPG ≥7.0 mmol l–1 High-income western Central and eastern Europe Latin America and the Caribbean East and southeast Asia and the Pacific Central Asia, Middle East and north Africa South Asia Sub-Saharan Africa Women High-income western Central and eastern Europe Latin America and the Caribbean East and southeast Asia and the Pacific Central Asia, Middle East and north Africa South Asia Sub-Saharan Africa Men 20253035 20253035 20253035 20253035 20253035 20253035 20253035 20253035 20253035 20253035 20253035 20253035 20253035 20253035 BMI (kg m–2) BMI (kg m–2) Fig. 3 | The predicted probability of having screen-detected diabetes with isolated elevated HbA1c or FPG. a,b, The probability, by sex, age and region, of participants who did not have previous diagnosis of diabetes of having elevated HbA1c (≥6.5%) at different FPG and BMI levels (a) and elevated FPG (≥7.0 mmol l−1) at different HbA1c and BMI levels (b). The probabilities were calculated using coefficients of prediction equation model 8, with measurement method set to laboratory for prediction. These results show the probability of having screendetected diabetes if the second biomarker had been measured, for a person whose first biomarker was below the clinical threshold for diabetes diagnosis. Nature Medicine | Volume 29 | November 2023 | 2885–2901 2892 Article https://doi.org/10.1038/s41591-023-02610-2 a 7.0 7.3 7.6 7.9 8.2 8.5 FPG (mmol l–1) High-income western Central and eastern Europe Latin America and the Caribbean East and southeast Asia and the Pacific Central Asia, Middle East and north Africa South Asia Sub-Saharan Africa Women High-income western Central and eastern Europe Latin America and the Caribbean East and southeast Asia and the Pacific Central Asia, Middle East and north Africa South Asia Sub-Saharan Africa Men 20253035 20 253035 20253035 20253035 20253035 20253035 20253035 20253035 20 253035 20253035 20253035 20253035 20253035 20253035 BMI (kg m–2)BMI (kg m–2) 75 years 65 years 55 years 45 years 35 years 25 years Predicted probability of having HbA1c ≥6.5% <5% 5–9% 10–29% 30–49% 50–59% 60–69% 70–79% 80–89% ≥90% Predicted probability of having FPG ≥7.0 mmol l–1 <5% 5–9% 10–29% 30–49% 50–59% 60–69% 70–79% 80–89% ≥90% b 6.5 6.7 6.9 7.1 7.3 7.5 HbA1c (%) 75 years 65 years 55 years 45 years 35 years 25 years High-income western Central and eastern Europe Latin America and the Caribbean East and southeast Asia and the Pacific Central Asia, Middle East and north Africa South Asia Sub-Saharan Africa Women High-income western Central and eastern Europe Latin America and the Caribbean East and southeast Asia and the Pacific Central Asia, Middle East and north Africa South Asia Sub-Saharan Africa Men 20253035 20 253035 20253035 20253035 20253035 20253035 20253035 20253035 20 253035 20253035 20253035 20253035 20253035 20253035 BMI (kg m–2) BMI (kg m–2) 7.0 7.3 7.6 7.9 8.2 8.5 7.0 7.3 7.6 7.9 8.2 8.5 7.0 7.3 7.6 7.9 8.2 8.5 7.0 7.3 7.6 7.9 8.2 8.5 7.0 7.3 7.6 7.9 8.2 8.5 6.5 6.7 6.9 7.1 7.3 7.5 6.5 6.7 6.9 7.1 7.3 7.5 6.5 6.7 6.9 7.1 7.3 7.5 6.5 6.7 6.9 7.1 7.3 7.5 6.5 6.7 6.9 7.1 7.3 7.5 Fig. 4 | The predicted probability of having screen-detected diabetes with elevated levels of both FPG and HbA1c. a,b, The probability by sex, age and region of participants who did not have a previous diagnosis of diabetes of having elevated HbA1c (≥6.5%) at different FPG and BMI levels (a) and elevated FPG (≥7.0 mmol l−1) at different HbA1c and BMI levels (b). The probabilities were calculated using coefficients of prediction equation model 8, with measurement method set to laboratory for prediction. These results show the probability that the second biomarker, had it been measured, would be above the clinical threshold for diabetes diagnosis, for a person whose first biomarker was above the clinical threshold for diabetes diagnosis. Having elevated levels of both biomarkers has high positive predictive value for subsequent clinical diagnosis and risk of complications14,47. Nature Medicine | Volume 29 | November 2023 | 2885–2901 2899 Article https://doi.org/10.1038/s41591-023-02610-2 Oslo, Norway. 51Universidad de Cuenca, Cuenca, Ecuador. 52Zahedan University of Medical Sciences, Zahedan, Iran. 53National Institute of Public Health, Tunis, Tunisia. 54University of Bergen, Bergen, Norway. 55Oulu University Hospital, Oulu, Finland. 56University of Oulu, Oulu, Finland. 57Regional Authority of Public Health, Banska Bystrica, Slovakia. 58Diabetic Association of Bangladesh, Dhaka, Bangladesh. 59Neyshabur University of Medical Sciences, Neyshabur, Iran. 60Research Institute for Endocrine Sciences, Tehran, Iran. 61National and Kapodistrian University of Athens, Athens, Greece. 62University of Science and Technology, Sana’a, Yemen. 63Medical University of Lodz, Lodz, Poland. 64Universidad Autónoma de Madrid CIBERESP, Madrid, Spain. 65University of Palermo, Palermo, Italy. 66University of Miami, Miami, FL, USA. 67University Hospital Centre Zagreb, Zagreb, Croatia. 68Universidad del Valle, Cali, Colombia. 69Baqai Institute of Diabetology and Endocrinology, Karachi, Pakistan. 70Jordan University of Science and Technology, Irbid, Jordan. 71Universidade Federal de Ouro Preto, Ouro Preto, Brazil. 72University of Sydney, Sydney, New South Wales, Australia. 73Christian Medical College Vellore, Vellore, India. 74University Tunis El Manar, Tunis, Tunisia. 75Cafam University Foundation, Bogotá, Colombia. 76Kazakh National Medical University, Almaty, Kazakhstan. 77Universidad Peruana Cayetano Heredia, Lima, Peru. 78Pontificia Universidad Católica de Chile, Santiago, Chile. 79University of São Paulo, São Paulo, Brazil. 80Sunder Lal Jain Hospital, Delhi, India. 81Institute of Medical Research and Medicinal Plant Studies, Yaoundé, Cameroon. 82Ufa Eye Research Institute, Ufa, Russia. 83Nepal Health Research Council, Kathmandu, Nepal. 84University of Southern Denmark, Copenhagen, Denmark. 85University of Gothenburg, Gothenburg, Sweden. 86Universidade Federal do Rio de Janeiro, Rio de Janeiro, Brazil. 87National Institute for Public Health and the Environment, Bilthoven, The Netherlands. 88University of Turin, Turin, Italy. 89University College London, London, UK. 90Universidad de la República, Montevideo, Uruguay. 91IRCCS Neuromed, Pozzilli, Italy. 92Caja Costarricense de Seguro Social, San José, Costa Rica. 93KU Leuven, Leuven, Belgium. 94Ministry of Health, Victoria, Seychelles. 95Unisanté, Lausanne, Switzerland. 96Universidad Central de Venezuela, Caracas, Venezuela. 97German Cancer Research Center, Heidelberg, Germany. 98The Fred Hollows Foundation, Auckland, New Zealand. 99University of the Andes, Mérida, Venezuela. 100Instituto Politécnico de Lisboa, Lisbon, Portugal. 101University College Copenhagen, Copenhagen, Denmark. 102Universidad de La Laguna, Tenerife, Spain. 103Pan American Health Organization, Washington, DC, USA. 104Istanbul University - Cerrahpasa, Istanbul, Türkiye. 105Universidade Federal de Juiz de Fora, Juiz de Fora, Brazil. 106National Institute of Public Health, Prague, Czech Republic. 107Gaetano Fucito Hospital, Mercato San Severino, Italy. 108Karolinska Institutet, Huddinge, Sweden. 109Santiago de Compostela University, Santiago de Compostela, Spain. 110Council for Agricultural Research and Economics, Rome, Italy. 111Sanpasitthiprasong Regional Hospital, Ubon Ratchathani, Thailand. 112Federation University Australia, Ballarat, Victoria, Australia. 113Xiangtan University, Xiangtan, China. 114Ahvaz Jundishapur University of Medical Sciences, Ahvaz, Iran. 115CIBERESP, Madrid, Spain. 116Medical University of Silesia, Katowice, Poland. 117Charles University, Prague, Czech Republic. 118Thomayer University Hospital, Prague, Czech Republic. 119University of Salerno, Fisciano, Italy. 120UMR CNRS-MNHN 7206, Paris, France. 121Agency for Preventive and Social Medicine, Bregenz, Austria. 122University of Southampton, Southampton, UK. 123CIBEROBN, Madrid, Spain. 124Universidade Federal do Rio Grande do Norte, Natal, Brazil. 125University of Malta, Msida, Malta. 126National Research Council, Reggio Calabria, Italy. 127Federal University of Santa Catarina, Florianópolis, Brazil. 128Institut Pasteur de Lille, Lille, France. 129Eduardo Mondlane University, Maputo, Mozambique. 130Tabriz Health Services Management Research Center, Tabriz, Iran. 131Ghent University, Ghent, Belgium. 132Innovating Health International, Port-au-Prince, Haiti. 133Sciensano, Brussels, Belgium. 134French Public Health Agency, St Maurice, France. 135University of Zagreb, Zagreb, Croatia. 136Ministry of Health and Medical Education, Tehran, Iran. 137Istituto Superiore di Sanità, Rome, Italy. 138Sun Yat-sen University, Guangzhou, China. 139Carol Davila University of Medicine and Pharmacy, Bucharest, Romania. 140University Medicine Greifswald, Greifswald, Germany. 141University Hospital Düsseldorf, Düsseldorf, Germany. 142Lazarski University, Warsaw, Poland. 143Robert Koch Institute, Berlin, Germany. 144IRL 3189 ESS, Marseille, France. 145Scuola Superiore Sant’Anna, Pisa, Italy. 146Ministry of Health and Medical Services, Gizo, Solomon Islands. 147Hormozgan University of Medical Sciences, Bandar Abbas, Iran. 148University of Benin, Benin City, Nigeria. 149National Institute of Nutrition and Food Technology, Tunis, Tunisia. 150The University of the West Indies, Kingston, Jamaica. 151Institut Hospital del Mar d’Investigacions Mèdiques, Barcelona, Spain. 152CIBERCV, Barcelona, Spain. 153University of Calabar, Calabar, Nigeria. 154University of Stellenbosch, Cape Town, South Africa. 155University of Duisburg-Essen, Essen, Germany. 156Karadeniz Technical University, Trabzon, Türkiye. 157Dokuz Eylul University, Izmir, Türkiye. 158University of Helsinki, Helsinki, Finland. 159Rafsanjan University of Medical Sciences, Rafsanjan, Iran. 160Monash University, Melbourne, Victoria, Australia. 161Fasa University of Medical Sciences, Fasa, Iran. 162Shiraz University of Medical Sciences, Shiraz, Iran. 163Baqai Medical University, Karachi, Pakistan. 164Centro de Salud Villanueva Norte, Badajoz, Spain. 165Hospital Don Benito-Villanueva de la Serena, Badajoz, Spain. 166Federal University of Alagoas, Maceió, Brazil. 167Wageningen University, Wageningen, The Netherlands. 168Wuqu’ Kawoq, Tecpan, Guatemala. 169Umeå University, Umeå, Sweden. 170Hospital Universitario Son Espases, Palma, Spain. 171Kyoto University, Kyoto, Japan. 172Medical University of Warsaw, Warsaw, Poland. 173Universidade Federal de Minas Gerais, Belo Horizonte, Brazil. 174Utrecht University, Utrecht, The Netherlands. 175Kurdistan University of Medical Sciences, Sanandaj, Iran. 176B. P. Koirala Institute of Health Sciences, Dharan, Nepal. 177University of Insubria, Varese, Italy. 178Mediterranea Cardiocentro, Naples, Italy. 179University of Adelaide, Adelaide, South Australia, Australia. 180Lund University, Lund, Sweden. 181McGill University, Montreal, Québec, Canada. 182PASs Hirszfeld Institute of Immunology and Experimental Therapy, Wroclaw, Poland. 183Federal University of Pelotas, Pelotas, Brazil. 184University Agostinho Neto, Luanda, Angola. 185Universidad Politécnica de Madrid, Madrid, Spain. 186International Clinical Research Center, Brno, Czech Republic. 187Centro de Estudios en Diabetes A.C, Mexico City, Mexico. 188Universidad Autónoma de Santo Domingo, Santo Domingo, Dominican Republic. 189University of Lille, Lille, France. 190Institute for Clinical and Experimental Medicine, Prague, Czech Republic. 191Jagiellonian University Medical College, Kraków, Poland. 192University of Southern Denmark, Odense, Denmark. 193Universidad Icesi, Cali, Colombia. 194Eternal Heart Care Centre and Research Institute, Jaipur, India. 195Institute for Clinical Effectiveness and Health Policy, Buenos Aires, Argentina. 196National University of Singapore, Singapore, Singapore. 197The University of the West Indies, Cave Hill, Barbados. 198Kermanshah University of Medical Sciences, Kermanshah, Iran. 199Africa Health Research Institute, Durban, South Africa. 200University of Eastern Finland, Kuopio, Finland. 201Capital Medical University, Beijing, China. 202Yasuj University of Medical Sciences, Yasuj, Iran. 203Kyushu University, Fukuoka, Japan. 204Federal University of Pernambuco, Recife, Brazil. 205Chronic Diseases Research Center, Tehran, Iran. 206University of Hong Kong, Hong Kong, China. 207French National Research Institute for Sustainable Development, Montpellier, France. 208Shahid Beheshti University of Medical Sciences, Tehran, Iran. 209Kingston Health Sciences Centre, Kingston, Ontario, Canada. 210Universidad Autónoma de Bucaramanga, Bucaramanga, Colombia. 211University Oran 1, Oran, Algeria. 212Independent Public Health Specialist, Nay Pyi Taw, Myanmar. 213Ministry of Health and Sports, Nay Pyi Taw, Myanmar. 214VU University Medical Center, Amsterdam, The Netherlands. 215International Agency for Research on Cancer, Lyon, France. 216College of Medicine, University of Nigeria, Ituku-Ozalla, Enugu, Nigeria. 217The University of Tokyo, Tokyo, Japan. 218Alex Ekwueme Federal University Teaching Hospital, Abakaliki, Nigeria. 219Deakin University, Geelong, Victoria, Australia. 220Hokkaido University, Sapporo, Japan. 221Hadassah University Medical Center, Jerusalem, Israel. 222Université Catholique de Louvain, Brussels, Belgium. 223Gambia National Nutrition Agency, Banjul, The Gambia. 224Kuwait Institute for Scientific Research, Safat, Kuwait. 225University of Melbourne, Melbourne, Victoria, Australia. 226Heart Foundation, Melbourne, Victoria, Australia. 227Universidad Eugenio Maria de Hostos, Santo Domingo, Dominican Republic. 228Institute of Molecular and Clinical Ophthalmology Basel, Basel, Switzerland. 229World Health Organization Country Office, Delhi, India. 230Guilan University of Medical Sciences, Rasht, Iran. 231University of Opole, Opole, Poland. 232University of Crete, Heraklion, Greece. 233Maharajgunj Medical Campus, Kathmandu, Nepal. 234Aarhus University, Aarhus, Denmark. 235University of Toronto, Toronto, Nature Medicine | Volume 29 | November 2023 | 2885–2901 2900 Article https://doi.org/10.1038/s41591-023-02610-2 Ontario, Canada. 236Research Institute for Primordial Prevention of Non-communicable Disease, Isfahan, Iran. 237Mashhad University of Medical Sciences, Mashhad, Iran. 238Research Institute of Child Nutrition, Dortmund, Germany. 239Shahrekord University of Medical Sciences, Shahrekord, Iran. 240Mazandaran University of Medical Sciences, Sari, Iran. 241Hypertension Research Center, Isfahan, Iran. 242Medical University of Innsbruck, Innsbruck, Austria. 243VASCage - Research Centre on Vascular Ageing and Stroke, Innsbruck, Austria. 244Newcastle University, Newcastle, UK. 245University College South Denmark, Haderslev, Denmark. 246Masaryk University, Brno, Czech Republic. 247University of Vienna, Vienna, Austria. 248Tartu University Clinics, Tartu, Estonia. 249Ministry of Health and Wellness, Port Louis, Mauritius. 250University of Zurich, Zurich, Switzerland. 251University of Groningen, Groningen, The Netherlands. 252University of Jyväskylä, Jyväskylä, Finland. 253National Institute of Cardiology, Warsaw, Poland. 254African Population and Health Research Center, Nairobi, Kenya. 255Hanoi University of Public Health, Hanoi, Vietnam. 256University of Limpopo, Polokwane, South Africa. 257Stellennbosch University, Polokwane, South Africa. 258Ministry of Health, Algiers, Algeria. 259Ministry of Health, Georgetown, Guyana. 260Oulu Deaconess Institute Foundation, Oulu, Finland. 261Sahlgrenska Academy, Gothenburg, Sweden. 262Endocrinology and Metabolism Research Center, Tehran, Iran. 263University of Public Health, Yangon, Myanmar. 264Centro Studi Epidemiologici di Gubbio, Gubbio, Italy. 265Tampere University Hospital, Tampere, Finland. 266Tampere University, Tampere, Finland. 267University of Douala, Douala, Cameroon. 268Oswaldo Cruz Foundation Rene Rachou Research Institute, Belo Horizonte, Brazil. 269National Taiwan University, Taipei, Taiwan. 270Uppsala University, Uppsala, Sweden. 271Zhengzhou University, Zhengzhou, China. 272Universidad San Martín de Porres, Lima, Peru. 273Consejería de Sanidad Junta de Castilla y León, Valladolid, Spain. 274Lithuanian University of Health Sciences, Kaunas, Lithuania. 275University of Porto, Porto, Portugal. 276University of Coimbra, Coimbra, Portugal. 277Coimbra University Hospital Center, Coimbra, Portugal. 278University of Texas Rio Grande Valley, Harlingen, TX, USA. 279Institute of Neuroscience of the National Research Council, Padua, Italy. 280Agricultural University of Athens, Athens, Greece. 281Academia VBHC, São Paulo, Brazil. 282Institute of Internal and Preventive Medicine, Novosibirsk, Russia. 283Harokopio University, Athens, Greece. 284Université Catholique de Bukavu, Bukavu, Democratic Republic of the Congo. 285University of Padua, Padua, Italy. 286Secretaria de Estado da Saúde de Santa Catarina, Florianópolis, Brazil. 287Universidade Estadual do Centro-Oeste, Guarapuava, Brazil. 288UiT The Arctic University of Norway, Tromsø, Norway. 289Sefako Makgatho Health Sciences University, Pretoria, South Africa. 290Instituto Conmemorativo Gorgas de Estudios de la Salud, Panama City, Panama. 291Brown University, Providence, RI, USA. 292University of Abidjan, Abidjan, Côte d’Ivoire. 293Universidade de Lisboa, Lisbon, Portugal. 294Saveetha Institute of Medical and Technical Sciences, Chennai, India. 295Università degli Studi di Firenze, Florence, Italy. 296Ain Shams University, Cairo, Egypt. 297Psychiatry and Psychology Research Center, Tehran, Iran. 298Isfahan Cardiovascular Research Center, Isfahan, Iran. 299University of Pécs, Pécs, Hungary. 300Mulago Hospital, Kampala, Uganda. 301Gorgas Memorial Institute for Studies of Health, Panama City, Panama. 302University of Medical Sciences of Cienfuegos, Cienfuegos, Cuba. 303University of Zaragoza, Zaragoza, Spain. 304Sabzevar University of Medical Sciences, Sabzevar, Iran. 305International Institute of Molecular and Cell Biology, Warsaw, Poland. 306World Health Organization Country Office, Lilongwe, Malawi. 307Department of Public Health, Nay Pyi Taw, Myanmar. 308University of Brescia, Brescia, Italy. 309Universiti Sains Malaysia, Kelantan, Malaysia. 310Universiti Kebangsaan Malaysia, Kuala Lumpur, Malaysia. 311University de Kinshasa, Kinshasa, Democratic Republic of the Congo. 312Bushehr University of Medical Sciences, Bushehr, Iran. 313Ulm University, Ulm, Germany. 314Department of Statistics, Kuala Lumpur, Malaysia. 315Suraj Eye Institute, Nagpur, India. 316Ministry of Health, Apia, Samoa. 317Mahidol University, Bangkok, Thailand. 318National Institute of Hygiene and Epidemiology, Hanoi, Vietnam. 319Hanoi Medical University, Hanoi, Vietnam. 320Xi’an Jiaotong University, Xi’an, China. 321Precision Care Clinic Corp, St. Cloud, FL, USA. 322Eastern Mediterranean Public Health Network, Amman, Jordan. 323University of Manchester, Manchester, UK. 324University of Abuja College of Health Sciences, Abuja, Nigeria. 325Korea Disease Control and Prevention Agency, Cheongju-si, Republic of Korea. 326Japan Wildlife Research Center, Tokyo, Japan. 327Istanbul University, Istanbul, Türkiye. 328Ministry of Health, Bandar Seri Begawan, Brunei. 329University of Madeira, Funchal, Portugal. 330Osteoporosis Research Center, Tehran, Iran. 331Universidad de Santander, Bucaramanga, Colombia. 332Kwame Nkrumah University of Science and Technology, Kumasi, Ghana. 333Academia Sinica, Taipei, Taiwan. 334Privatpraxis Prof Jonas und Dr Panda-Jonas, Heidelberg, Germany. 335IRCCS Ente Ospedaliero Specializzato in Gastroenterologia S. de Bellis, Bari, Italy. 336Jivandeep Hospital, Anand, India. 337Centro de Investigação em Saúde de Angola, Caxito, Angola. 338Vietnam National Heart Institute, Hanoi, Vietnam. 339National Hospital of Endocrinology, Hanoi, Vietnam. 340Clínica de Medicina Avanzada Dr. Abel González, Santo Domingo, Dominican Republic. 341University of Sarajevo, Sarajevo, Bosnia and Herzegovina. 342Ministry of Health and Medical Services, Honiara, Solomon Islands. 343Public Health Agency of Catalonia, Barcelona, Spain. 344Observatorio de Salud Pública de Santander, Bucaramanga, Colombia. 345Ardabil University of Medical Sciences, Ardabil, Iran. 346Lausanne University Hospital, Lausanne, Switzerland. 347Alborz University of Medical Sciences, Karaj, Iran. 348Ministry of Health, Hanoi, Vietnam. 349University of Turku, Turku, Finland. 350Institute of Nutrition of Central America and Panama, Guatemala City, Guatemala. 351Institut Universitari d’Investigació en Atenció Primària Jordi Gol, Girona, Spain. 352Universiti Putra Malaysia, Serdang, Malaysia. 353University of Malaya, Kuala Lumpur, Malaysia. 354University of Valencia, Valencia, Spain. 355University of Santa Cruz do Sul, Santa Cruz do Sul, Brazil. 356CS S. Agustín Ibsalut, Palma, Spain. 357Ministerio de Salud, Panama City, Panama. 358Canarian Health Service, Tenerife, Spain. 359Universidad Industrial de Santander, Bucaramanga, Colombia. 360Ministery of Health and Social Protection, Bogotá, Colombia. 361Associazione Calabrese di Epatologia, Reggio Calabria, Italy. 362Sahlgrenska University Hospital, Gothenburg, Sweden. 363Sitaram Bhartia Institute of Science and Research, New Delhi, India. 364University or Zagreb, Zagreb, Croatia. 365National Institute of Health, Lima, Peru. 366Wellbeing Services County of South Karelia, Lappeenranta, Finland. 367Universidade Federal do Rio Grande do Sul, Porto Alegre, Brazil. 368University of São Paulo Clinics Hospital, São Paulo, Brazil. 369Human Sciences Research Council, Cape Town, South Africa. 370Academy of Preventive Medicine, Almaty, Kazakhstan. 371Teikyo University, Tokyo, Japan. 372Finnish Institute of Occupational Health, Helsinki, Finland. 373Public Health Promotion and Development Organization, Kathmandu, Nepal. 374St Vincent’s Hospital, Sydney, New South Wales, Australia. 375University of New South Wales, Sydney, New South Wales, Australia. 376Karolinska Institutet, Stockholm, Sweden. 377London School of Hygiene & Tropical Medicine, London, UK. 378Diponegoro University, Semarang, Indonesia. 379University of Bari, Bari, Italy. 380University of Bordeaux, Bordeaux, France. 381University of Hohenheim, Stuttgart, Germany. 382Bonn University, Bonn, Germany. 383National Institute of Public Health - National Institute of Hygiene, Warsaw, Poland. 384Pontificia Universidad Javeriana Seccional Cali, Cali, Colombia. 385Ubon Ratchathani University, Ubon Ratchathani, Thailand. 386National Statistical Office, Praia, Cabo Verde. 387Johns Hopkins Bloomberg School of Public Health, Baltimore, MD, USA. 388Ministry of Health, Amman, Jordan. 389Amrita Institute of Medical Sciences, Kochi, India. 390Aristotle University of Thessaloniki, Thessaloniki, Greece. 391University Medical Center Utrecht, Utrecht, The Netherlands. 392National Research and Innovation Agency, Jakarta, Indonesia. 393Universidad Miguel Hernandez, Madrid, Spain. 394Department of Health, Faga’alu, American Samoa. 395LBJ Hospital, Faga’alu, American Samoa. 396Universidad Centro-Occidental Lisandro Alvarado, Barquisimeto, Venezuela. 397Meharry Medical College, Nashville, TN, USA. 398University of Tampere Tays Eye Center, Tampere, Finland. 399MONICA-FRIULI Study Group, Udine, Italy. 400Institute of Tropical Medicine, Antwerp, Belgium. 401CIBERESP, Alicante, Spain. 402Institute for Medical Research, Kuala Lumpur, Malaysia. 403Capital Medical University Beijing Tongren Hospital, Beijing, China. 404Xinjiang Medical University, Urumqi, China. 405Ministry of Health and Welfare, Taipei, Taiwan. 406The Ministry of Health and Wellness, Kingston, Jamaica. 407St George’s, University of London, London, UK. 408Medical University of Vienna, Vienna, Austria. 409Universitas Indonesia, Jakarta, Indonesia. 410Institute of Food and Nutrition Development of Ministry of Agriculture and Rural Affairs, Beijing, China. 411Beijing Institute of Ophthalmology, Beijing, China. Nature Medicine | Volume 29 | November 2023 | 2885–2901 2901 Article https://doi.org/10.1038/s41591-023-02610-2 412Children’s Hospital of Fudan University, Shanghai, China. 413Niigata University, Niigata, Japan. 414South China Institute of Environmental Sciences, Guangzhou, China. 415Iran University of Medical Sciences, Tehran, Iran. 416Peking University, Beijing, China. 417Duke University, Durham, NC, USA. 418West Kazakhstan Medical University, Aktobe, Kazakhstan. 419University of Ghana, Accra, Ghana. 420Deceased: Mostafa K. Mohamed. 421Deceased: Altan Onat. 422Deceased: Michael Sjöström. 423Deceased: Agustinus Soemantri. ✉ e-mail: m aj id .e zz at i@ im pe ri al.ac.uk Nature Medicine Article https://doi.org/10.1038/s41591-023-02610-2 M et ho ds The pooled analysis was approved by Imperial College London Research Ethics Committee and complies with all relevant ethical regulations. The participating studies followed their institutional approval process at the time of data collection. Data We used data collated by the NCD-RisC. The data sources included national and multi-country measurement surveys that were either publicly available or identified and accessed through contacts with relevant government or academic partners. Additionally, we searched and reviewed published studies as detailed previously 44 and invited eligible studies to join NCD-RisC, as we did with participating studies in previous pooled analyses of cardiometabolic risk factors 96–99 . The NCD-RisC database is continuously updated through the above routes and through periodic requests to NCD-RisC members to suggest additional sources in their countries. The inclusion criteria for this analysis were (1) data were collected using a probabilistic sampling method with a defined sampling frame; (2) data were from population samples at the national, subnational (defined as covering one or more subnational regions, more than three urban communities or more than five rural communities) or community level (defined as having up to three urban communities or up to five rural communities); and (3) both FPG and HbA1c were measured. Studies were excluded if they had (1) enrolled participants based on health status or cardiovascular risk; (2) were conducted only among ethnic minorities or specific educational, occupational or other socioeconomic subgroups; (3) recruited participants through health facilities, except studies based on the primary care system in high-income and central European countries with universal insurance; (4) had not measured either FPG or HbA1c; (5) had not instructed participants to fast for at least 6 h before FPG measurement; (6) had only measured FPG or HbA1c in the subset of participants who had known diabetes; (7) had measured HbA1c only in a subset of participants selected based on their levels of FPG and vice versa; (8) had not collected information on a previous diagnosis of diabetes; and (9) their mid-year was before 2000, before HbA1c assays were widely standardized100. At least two independent people ascertained that each data source met the inclusion criteria. All NCD-RisC members were asked to review the list of data sources from their country, to verify that the included data met the inclusion criteria and were not duplicates. When FPG and/or HbA1c data were missing for more than 10% of participants in a survey, we checked the study design documentation to verify missingness at random so that the above inclusion criteria were met. Questions and clarifications were discussed with NCD-RisC members and resolved before data were incorporated in the database. For each data source, we recorded the study population, sampling approach, years of measurement and measurement methods, including whether FPG and HbA1c were measured in a laboratory or using a portable point-of-care device. In 11 studies, fasting glucose was measured in capillary whole blood; four of these used equipment that reported plasma-equivalent values. We converted the measurements from the other seven studies to plasma-equivalent using the relationship in a study that compared different types of specimens101. In a sensitivity analysis, we excluded these 11 studies from the analysis. We established whether a participant had diagnosed diabetes using questions worded as variations of ‘Have you ever been told by a doctor or other health professional that you had diabetes, also called high blood sugar?’ In some surveys, the question on previous diabetes diagnosis was asked only if a participant had answered ‘yes’ to an earlier question, usually worded as ‘Have you ever been screened for diabetes?’ or ‘Have you ever had your blood glucose measured?’. In these cases, participants who answered ‘no’ to the first question were coded as not having been diagnosed with diabetes. We also considered participants who used diabetes medication such as metformin or insulin as having diabetes. Survey data typically do not separate type 1 and type 2 diabetes in adults, but studies that had data on these subtypes show that most (85–95%) cases of diabetes in adults are type 2 diabetes102. The data cleaning and use process is summarized in Fig. 1 and the list of data sources and their characteristics are stated in Supplementary Table 1. Statistical analysis We divided the participants into those who had a previous diagno - sis of diabetes (hereafter referred to as diagnosed diabetes), those without a previous diagnosis of diabetes who had elevated FPG (FPG ≥ 7.0 mmol l −1 ) and/or elevated HbA1c (HbA1c ≥ 6.5%) (referred to as screen-detected diabetes) and the remainder who did not have a previous diagnosis, elevated FPG, or elevated HbA1c. We conducted the following three analyses. Screen-detected diabetes by FPG and HbA1c. We graphically presented how total diabetes is divided into diagnosed and screendetected diabetes, and how screen-detected diabetes is further divided into those manifested as only elevated FPG (FPG ≥ 7.0 mmol l −1 and HbA1c < 6.5%, referred to as isolated elevated FPG), only elevated HbA1c (HbA1c ≥ 6.5% and FPG < 7.0 mmol l−1, referred to as isolated elevated HbA1c) or elevated levels of both FPG and HbA1c. We report crude and age-standardized prevalence. We calculated crude prevalence using data from all participants regardless of age. We calculated age-standardized prevalence as the weighted mean of the age-specific values using the World Health Organization standard population 103 . We also graphically described the relationship of FPG and HbA1c among people without diagnosed diabetes. Association with individual and study characteristics. We fitted regression models to examine what individual and study-level factors were associated with whether participants with screen-detected diabetes were identified by elevated FPG, elevated HbA1c or elevated levels of both. We fitted three separate log-binomial regressions, with each of the three outcomes (isolated elevated FPG, isolated elevated HbA1c and elevated levels of both) as a distinct dependent variable. A log-binomial regression estimates the association of each independent variable with the probability of a participant falling in each of the three categories as PR. The individual level independent variables were sex, age and BMI; the study-level variables were region, study year, whether FPG and HbA1c were measured in a laboratory or using a portable device (to account for differences in measurement between them53,54) and percentage of participants with diabetes who had been diagnosed before in each study. The regressions also included a study-level random effect to account for unobserved factors that led to systematic differences in each study compared to others104,105. We fitted the log-binomial regressions using Bayesian model fitting implemented in MultiBUGS (v.2.0) 106 . Bayesian model fitting has better estimation performance for log-binomial model than a frequentist approach 107 . We used a normal distribution with mean of zero and s.d. of 0.01 as the prior for the regression coefficients and a uniform distribution on 0.01–2.00 as the prior for the s.d. of study-level random effects. We ran four chains and assessed convergence visually using trace plots. After burn-in and thinning, we kept 50,000 draws to represent the posterior distributions of the PRs. We report PRs and their 95% CrIs as the mean and the 2.5th and 97.5th percentiles of their posterior distributions. We report the posterior probability that a PR with posterior mean estimate >1.0 is less than one and vice versa for PRs <1.0; the posterior probabilities are analogous to P values in a frequentist analysis. Nature Medicine Article https://doi.org/10.1038/s41591-023-02610-2 Prediction equations. We tested nine logistic regression models for estimating the probability that a person without diagnosed diabetes at a specific level of FPG had an HbA1c over the clinical threshold for diabetes (HbA1c ≥ 6.5%). The variables in the models were selected based on clinical and epidemiological relevance and data availability. The variables included FPG as well as sex, age, BMI, glycemic measurement method (laboratory based or via a portable device) and region. The nine prediction models (Extended Data Table 2) differed by the predictors included and whether the coefficient of the FPG term was allowed to vary by sex and region. In all models, we included a study-level random effect to account for unobserved factors that led to systematic differences in each study compared to others 104,105 . We also tested the inclusion of nonlinear (square and cubic) terms of FPG, year of data collection and other interaction terms; these models performed worse than those without the additional terms as evaluated by the metrics below and are not presented. We did not interact age, which is a continuous variable, with FPG and other terms, to avoid overfitting. We fitted and evaluated all prediction models in R (v.4.2.1)108. We assessed the performance of the models in predicting (1) individual participants’ status of having HbA1c ≥ 6.5% based on their FPG and (2) the prevalence of HbA1c ≥ 6.5% for an entire study. The performance at the individual level reflects how well the prediction equation works for triaging patients for further measurement for diabetes, and the performance at study (or population) level assesses how well it works for diabetes surveillance. We used the C-statistic to assess individual-level performance and mean error and mean absolute error between the predicted and observed prevalence for population-level performance. The C-statistic measures how well a prediction equation distinguishes individuals with higher risk from those with lower risk. Mean error assesses whether there is systematic difference (bias) in the predicted prevalence compared to the observed one and mean absolute error assesses any deviation of the predicted prevalence from the observed prevalence. We calculated error by study, sex and age group (18–39 years, 40–59 years and 60 years and older). We evaluated the performance of the models in 20 rounds of tenfold cross-validation109. In each fold of each round, we held out all data from a random 10% of studies, fitted the model to the data from the remaining 90% of studies and made estimates for the held-out observations. We repeated this process ten times, each time holding out a different 10% of studies so that each study was held out exactly once. We calculated the above individual-level and population-level performance metrics for all held-out observations. We repeated the tenfold cross-validation 20 times and report the means and ranges of the performance metrics from all 20 rounds. We repeated the same process for predicting the probability of having FPG ≥ 7.0 mmol l−1 based on HbA1c. Ethics and inclusion This research followed the recommendations set out in the Global Code of Conduct for Research in Resource-Poor Settings. Reporting summary Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article. Data availability Data used in this research are governed by data-sharing protocols of participating studies. Contact information for data providers can be obtained from www.ncdrisc.org and https://doi.org/10.5281/ zenodo.8169145. Code availability The computer code for the log-binomial regression in this work is available at www.ncdrisc.org and https://doi.org/10.5281/zenodo.8169145. References 96. Farzadfar, F. et al. National, regional, and global trends in serum total cholesterol since 1980: systematic analysis of health examination surveys and epidemiological studies with 321 countryyears and 3.0 million participants. Lancet 377, 578–586 (2011). 97. Finucane, M. M. et al. National, regional, and global trends in body-mass index since 1980: systematic analysis of health examination surveys and epidemiological studies with 960 countryyears and 9.1 million participants. Lancet 377, 557–567 (2011). 98. Danaei, G. et al. National, regional, and global trends in systolic blood pressure since 1980: systematic analysis of health examination surveys and epidemiological studies with 786 countryyears and 5.4 million participants. Lancet 377, 568–577 (2011). 99. Danaei, G. et al. National, regional, and global trends in fasting plasma glucose and diabetes prevalence since 1980: systematic analysis of health examination surveys and epidemiological studies with 370 country-years and 2.7 million participants. Lancet 378, 31–40 (2011). 100. American Diabetes Association. Implications of the Diabetes Control and Complications Trial. Diabetes Care 23, S24–S26 (2000). 101. Carstensen, B. et al. Measurement of blood glucose: comparison between different types of specimens. Ann. Clin. Biochem. 45, 140–148 (2008). 102. Bullard, K. M. et al. Prevalence of diagnosed diabetes in adults by diabetes type — United States, 2016. MMWR Morb. Mortal. Wkly. Rep. 67, 359–361 (2018). 103. Ahmad, O. B. et al. Age standardization of rates: a new WHO standard. GPE Discussion Paper Series: No.31 (2001). 104. Laird, N. M. & Ware, J. H. Random-effects models for longitudinal data. Biometrics 38, 963–974 (1982). 105. Feller, A. & Gelman, A. Hierarchical Models for Causal Effects. in Emerging Trends in the Social and Behavioral Sciences (eds Scott, R. A. & Kosslyn, S. M.) 1–16 (2015). 106. Goudie, R. J. B., Turner, R. M., De Angelis, D. & Thomas, A. MultiBUGS: a parallel implementation of the BUGS modelling framework for faster Bayesian inference. J. Stat. Softw. 95, 1–20 (2020). 107. Torman, V. B. & Camey, S. A. Bayesian models as a unified approach to estimate relative risk (or prevalence ratio) in binary and polytomous outcomes. Emerg. Themes Epidemiol. 12, 8 (2015). 108. R Core Team. R: a language and environment for statistical computing (2022). 109. Borra, S. & Di Ciaccio, A. Measuring the prediction error. A comparison of cross-validation, bootstrap and covariance penalty methods. Comput. Stat. Data Anal. 54, 2976–2989 (2010). Acknowledgements This study was funded by the UK Medical Research Council (grant number MR/V034057/1 to M.E.), the UK Research and Innovation (Research England Policy Support Fund to M.E.) and the US Centers for Disease Control and Prevention (to E.W.G.). B. Zhou is supported by a fellowship from the Abdul Latif Jameel Institute for Disease and Emergency Analytics, funded by a donation from Community Jameel, at Imperial College London. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. For the purpose of open access, the author has applied a Creative Commons Attribution license to the Author Accepted Manuscript version arising from this submission. Author contributions B. Zhou, K.E.S. and R.K.S. led the data collection and management. B. Zhou, J.E.B., A. Mishra, C.J.P., S.V.H. and M.E. developed the statistical method. B. Zhou coded the statistical method, conducted analyses and prepared results. The other authors contributed to the Nature Medicine Article https://doi.org/10.1038/s41591-023-02610-2 study design and collected, reanalyzed, checked and pooled the data. B. Zhou and M.E. wrote the first draft of the report. All other authors reviewed and commented on the draft report. Competing interests A.N.W. reports an honorarium from Sanofi for serving as a panel member at an educational event on thyroid cancer. The authors are responsible for the views expressed in this article and they do not necessarily represent the views, decisions or policies of the institutions with which they are affiliated. Additional information Extended data is available for this paper at https://doi.org/10.1038/s41591-023-02610-2. Supplementary information The online version contains supplementary material available at https://doi.org/10.1038/s41591-023-02610-2. Correspondence and requests for materials should be addressed to Majid Ezzati. Peer review information Nature Medicine thanks Sarah Wild and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. Primary Handling Editor: Jennifer Sargent, in collaboration with the Nature Medicine team. Reprints and permissions information is available at www.nature.com/reprints. Nature Medicine Article https://doi.org/10.1038/s41591-023-02610-2 A 10% 0% 10% 20% Proportion of all participants (%) Women Men Crude Age-standardised Crude Age-standardised All studies High-income western Central and eastern Europe East and southeast Asia and the Pacific Latin America and the Caribbean Central Asia, Middle East and north Africa South Asia Sub-Saharan Africa All studies High-income western Central and eastern Europe East and southeast Asia and the Pacific Latin America and the Caribbean Central Asia, Middle East and north Africa South Asia Sub-Saharan Africa All studies High-income western Central and eastern Europe East and southeast Asia and the Pacific Latin America and the Caribbean Central Asia, Middle East and north Africa South Asia Sub-Saharan Africa All studies High-income western Central and eastern Europe East and southeast Asia and the Pacific Latin America and the Caribbean Central Asia, Middle East and north Africa South Asia Sub-Saharan Africa Diagnosed diabetes Elevated levels of both FPG and HbA1c Isolated elevated HbA1c Isolated elevated FPG B Crude Age-standardised Women Crude Age-standardised Men 0% 25% 50% 75% 100% 0% 25% 50% 75% 100% Proportion of participants with screen-detected diabetes (%) 0% 25% 50% 75% 100% 0% 25% 50% 75% 100% Proportion of participants with screen-detected diabetes (%) All studies Latin America and the Caribbean East and southeast Asia and the Pacific Central Asia, Middle East and north Africa Sub-Saharan Africa South Asia High-income western Central and eastern Europe Extended Data Fig. 1 | Extent and composition of diagnosed and screendetected diabetes by region and sex. (a) Crude and age-standardized proportion of participants with diagnosed or screen-detected diabetes, and, for those without prior diagnosis, whether they had isolated elevated FPG (FPG ≥7.0 mmol/L and HbA1c < 6.5%), isolated elevated HbA1c (HbA1c ≥6.5% and FPG < 7.0 mmol/L) or elevated levels of both, and (b) crude and age-standardized proportion of participants with screen-detected diabetes who had isolated elevated FPG, isolated elevated HbA1c or elevated levels of both, by region and sex. Its contents are the same as the segment of Panel A that is below the zero line, scaled to 100% so that the composition of screen-detected diabetes can be compared across regions, regardless of its total prevalence. Having elevated levels of both biomarkers has high positive predictive value for subsequent clinical diagnosis and risk of complications14,47, and hence this group is similar to clinically-diagnosed diabetes. In panel A, regions are ordered by the total proportion of participants who had diagnosed and screen-detected diabetes. In panel B, regions are ordered by the crude proportion of participants with screendetected diabetes who had elevated levels of both FPG and HbA1c. Nature Medicine Article https://doi.org/10.1038/s41591-023-02610-2 A Crude Age-standardised All studies High-income western Central and eastern Europe East and southeast Asia and the Pacific Latin America and the Caribbean Central Asia, Middle East and north Africa South Asia Sub-Saharan Africa All studies High-income western Central and eastern Europe East and southeast Asia and the Pacific Latin America and the Caribbean Central Asia, Middle East and north Africa South Asia Sub-Saharan Africa 10% 0% 10% 20% Proportion of all participants (%) Diagnosed diabetes Elevated levels of both FPG and HbA1c Isolated elevated HbA1c Isolated elevated FPG B Crude Age-standardised 0% 25% 50% 75% 100% 0% 25% 50% 75% 100% All studies Central Asia, Middle East and north Africa East and southeast Asia and the Pacific Latin America and the Caribbean Sub-Saharan Africa South Asia High-income western Central and eastern Europe Proportion of participants with screen-detected diabetes (%) Extended Data Fig. 2 | Extent and composition of diagnosed and screen-detected diabetes by region, after removing two studies in Mauritius from sub-Saharan Africa. (a) Crude and age-standardized proportion of participants with diagnosed or screen-detected diabetes, and, for those without prior diagnosis, whether they had isolated elevated FPG (FPG ≥7.0 mmol/L and HbA1c < 6.5%), isolated elevated HbA1c (HbA1c ≥6.5% and FPG < 7.0 mmol/L) or elevated levels of both, and (b) crude and age-standardized proportion of participants with screen-detected diabetes who had isolated elevated FPG, isolated elevated HbA1c or elevated levels of both, by region. Its contents are the same as the segment of Panel A that is below the zero line, scaled to 100% so that the composition of screen-detected diabetes can be compared across regions, regardless of its total prevalence. Having elevated levels of both biomarkers has high positive predictive value for subsequent clinical diagnosis and risk of complications14,47, and hence this group is similar to clinically-diagnosed diabetes. In panel A, regions are ordered by the total proportion of participants who had diagnosed and screen-detected diabetes. In panel B, regions are ordered by the crude proportion of participants with screen-detected diabetes who had elevated levels of both FPG and HbA1c. Regions are in the same order as in Fig. 2. Nature Medicine Article https://doi.org/10.1038/s41591-023-02610-2 4 6 8 10 12 4 6 8 10 12 4 6 8 10 12 4 6 8 10 12 HbA1c (%) 4 6 8 10 12 4 6 8 10 12 4 6 8 10 12 High−income western Central and eastern Europe Latin America and the Caribbean East and southeast Asia and the Pacific Central Asia, Middle East and north Africa Sub−Saharan Africa South Asia r = 0.56 r = 0.51 r = 0.65 r = 0.64 r = 0.67 r = 0.76 r = 0.61 4812 16 4812 16 4812 16 4812 16 4812 16 4812 16 4812 16 FPG (mmol/L) Extended Data Fig. 3 | Relationship between FPG and HbA1c, among participants who had not been previously diagnosed with diabetes, by region. The shading indicates the density of participants in each region, with darker shades corresponding to more participants and vice versa. The dotted lines are placed at FPG of 7.0 mmol/L and HbA1c of 6.5%, which are common clinical thresholds for diabetes10–13. The numbers on the panels indicate the Pearson correlation coefficient between FPG and HbA1c in each region. A total of 623 (0.2%) participants with FPG of 19-28 mmol/L and/or HbA1c of 12-17% are not shown in the figure so that the axes have sufficient resolution in ranges where the great majority of participants were. Nature Medicine Article https://doi.org/10.1038/s41591-023-02610-2 Extended Data Table 1 | List of analysis regions and countries in each region. The data used in the analysis came from countries shown in bold Region Country Central and eastern Europe Albania, Belarus, Bosnia and Herzegovina, Bulgaria, Croatia, Czech Republic, Estonia, Hungary, Latvia, Lithuania, Moldova, Montenegro, North Macedonia, Poland, Romania, Russian Federation, Serbia, Slovakia, Slovenia, Ukraine Central Asia, Middle East and north Africa Algeria, Armenia, Azerbaijan, Bahrain, Egypt, Georgia, Iran, Iraq, Jordan, Kazakhstan, Kuwait, Kyrgyzstan, Lebanon, Libya, Mongolia, Morocco, Occupied Palestinian Territory, Oman, Qatar, Saudi Arabia, Syrian Arab Republic, Tajikistan, Tunisia, Turkey, Turkmenistan, United Arab Emirates, Uzbekistan, Yemen High-income western Andorra, Australia, Austria, Belgium, Canada, Cyprus, Denmark, Finland, France, Germany, Greece, Greenland, Iceland, Ireland, Israel, Italy, Luxembourg, Malta, Netherlands, New Zealand, Norway, Portugal, Spain, Sweden, Switzerland, United Kingdom, United States of America Latin America and the Caribbean Antigua and Barbuda, Argentina, Bahamas, Barbados, Belize, Bermuda, Bolivia, Brazil, Chile, Colombia, Costa Rica, Cuba, Dominica, Dominican Republic, Ecuador, El Salvador, Grenada, Guatemala, Guyana, Haiti, Honduras, Jamaica, Mexico, Nicaragua, Panama, Paraguay, Peru, Puerto Rico, Saint Kitts and Nevis, Saint Lucia, Saint Vincent and the Grenadines, Suriname, Trinidad and Tobago, Uruguay, Venezuela Oceania American Samoa, Cook Islands, Federated States of Micronesia, Fiji, French Polynesia, Kiribati, Marshall Islands, Nauru, Niue, Palau, Papua New Guinea, Samoa, Solomon Islands, Tokelau, Tonga, Tuvalu, Vanuatu South Asia Afghanistan, Bangladesh, Bhutan, India, Nepal, Pakistan, Sri Lanka East and southeast Asia and the Pacific Brunei Darussalam, Cambodia, China, Indonesia, Japan, Lao PDR, Malaysia, Maldives, Myanmar, North Korea, Philippines, Singapore, South Korea, Taiwan, Thailand, Timor-Leste, Viet Nam Sub-Saharan Africa Angola, Benin, Botswana, Burkina Faso, Burundi, Cabo Verde, Cameroon, Central African Republic, Chad, Comoros, Congo, Cote d'Ivoire, Djibouti, DR Congo, Equatorial Guinea, Eritrea, Eswatini, Ethiopia, Gabon, Gambia, Ghana, Guinea, Guinea Bissau, Kenya, Lesotho, Liberia, Madagascar, Malawi, Mali, Mauritania, Mauritius, Mozambique, Namibia, Niger, Nigeria, Rwanda, Sao Tome and Principe, Senegal, Seychelles, Sierra Leone, Somalia, South Africa, South Sudan, Sudan, Tanzania, Togo, Uganda, Zambia, Zimbabwe 1 nature portfolio | reporting summary March 2021 Corresponding author(s): Majid Ezzati Last updated by author(s): Sep 20, 2023 Reporting Summary Nature Portfolio wishes to improve the reproducibility of the work that we publish. This form provides structure for consistency and transparency in reporting. For further information on Nature Portfolio policies, see our Editorial Policies and the Editorial Policy Checklist. Statistics For all statistical analyses, confirm that the following items are present in the figure legend, table legend, main text, or Methods section. n/a Confirmed The exact sample size (n) for each experimental group/condition, given as a discrete number and unit of measurement A statement on whether measurements were taken from distinct samples or whether the same sample was measured repeatedly The statistical test(s) used AND whether they are oneor two-sided Only common tests should be described solely by name; describe more complex techniques in the Methods section. A description of all covariates tested A description of any assumptions or corrections, such as tests of normality and adjustment for multiple comparisons A full description of the statistical parameters including central tendency (e.g. means) or other basic estimates (e.g. regression coefficient) AND variation (e.g. standard deviation) or associated estimates of uncertainty (e.g. confidence intervals) For null hypothesis testing, the test statistic (e.g. F, t, r) with confidence intervals, effect sizes, degrees of freedom and P value noted Give P values as exact values whenever suitable. For Bayesian analysis, information on the choice of priors and Markov chain Monte Carlo settings For hierarchical and complex designs, identification of the appropriate level for tests and full reporting of outcomes Estimates of effect sizes (e.g. Cohen's d, Pearson's r), indicating how they were calculated Our web collection on statistics for biologists contains articles on many of the points above. Software and code Policy information about availability of computer code Data collection Processing of secondary data was conducted using the statistical software R (version 4.2.1). Data analysis Analyses were conducting using the statistical software R (version 4.2.1) and MultiBUGS (version 2.0). Code for log-binomial model is provided at www.ncdrisc.org and https://doi.org/10.5281/zenodo.8169146. For manuscripts utilizing custom algorithms or software that are central to the research but not yet described in published literature, software must be made available to editors and reviewers. We strongly encourage code deposition in a community repository (e.g. GitHub). See the Nature Portfolio guidelines for submitting code & software for further information. Data Policy information about availability of data All manuscripts must include a data availability statement. This statement should provide the following information, where applicable: - Accession codes, unique identifiers, or web links for publicly available datasets - A description of any restrictions on data availability - For clinical datasets or third party data, please ensure that the statement adheres to our policy This is data-pooling study that brings together 117 data sources. Data used in this research are governed by data sharing protocols of participating studies. Contact information for data providers can be obtained from www.ncdrisc.org and https://doi.org/10.5281/zenodo.8169146. 2 nature portfolio | reporting summary March 2021 Human research participants Policy information about studies involving human research participants and Sex and Gender in Research. Reporting on sex and gender Use the terms sex (biological attribute) and gender (shaped by social and cultural circumstances) carefully in order to avoid confusing both terms. Indicate if findings apply to only one sex or gender; describe whether sex and gender were considered in study design whether sex and/or gender was determined based on self-reporting or assigned and methods used. Provide in the source data disaggregated sex and gender data where this information has been collected, and consent has been obtained for sharing of individual-level data; provide overall numbers in this Reporting Summary. Please state if this information has not been collected. Report sexand gender-based analyses where performed, justify reasons for lack of sexand gender-based analysis. Population characteristics Describe the covariate-relevant population characteristics of the human research participants (e.g. age, genotypic information, past and current diagnosis and treatment categories). If you filled out the behavioural & social sciences study design questions and have nothing to add here, write "See above." Recruitment Describe how participants were recruited. Outline any potential self-selection bias or other biases that may be present and how these are likely to impact results. Ethics oversight Identify the organization(s) that approved the study protocol. Note that full information on the approval of the study protocol must also be provided in the manuscript. Field-specific reporting Please select the one below that is the best fit for your research. If you are not sure, read the appropriate sections before making your selection. Life sciences Behavioural & social sciences Ecological, evolutionary & environmental sciences For a reference copy of the document with all sections, see nature.com/documents/nr-reporting-summary-flat.pdf Behavioural & social sciences study design All studies must disclose on these points even when the disclosure is negative. Study description We pooled and analysed data from population-based studies that had measured FPG and HbA1c (quantitative data) and collected information on prior diagnosis of diabetes (qualitative data) for adults aged 18 years and over. We reported the proportions of participants who had diagnosed diabetes, and for those without diagnosed diabetes, whether they had elevated FPG (FPG ≥7.0 mmol/L), elevated HbA1c (HbA1c ≥6.5%) or both. We examined the individual-level and study-level factors associated with whether participants with screen-detected diabetes were identified by elevated FPG, elevated HbA1c or elevated levels of both. We tested prediction equations for estimating the probability that a person without diagnosed diabetes at a specific level of FPG had an HbA1c over the clinical threshold for diabetes (HbA1c ≥6.5%), and vice versa. Research sample We used all studies collated by the NCD Risk Factor Collaboration that had collected information on whether participants had been previously diagnosed with diabetes, and measured both FPG and HbA1c. In total, we used 117 population-based studies that had data on 601,000 participants aged 18 years or over in 45 countries, of whom 365,000 also had measurements of both FPG and HbA1c. Sampling strategy We included studies that had collected data using a probabilistic sampling method with a defined sampling frame. Hence, we included studies with simple random and complex survey designs, and excluded convenience samples and studies whose participants were selected based on factors that might be associated with their diabetes status. Data collection We used participant-level data for 601,000 participants from 117 studies. This is an observational study and there was no experiment. Timing We used data from surveys with mid-point of data collection period from 2000 to 2021. Data exclusions Studies were excluded if they (1) enrolled participants based on health status or cardiovascular risk; (2) were conducted only among ethnic minorities or specific educational, occupational, or other socioeconomic subgroups; (3) recruited participants through health facilities, except studies based on primary care system in high-income and central European countries with universal insurance; (4) had not measured either FPG or HbA1c; (5) had not instructed participants to fast at least for 6 hours prior to FPG measurement; (6) had only measured FPG or HbA1c in the subset of participants who had known diabetes; (7) had measured HbA1c only in a subset of participants selected based on their levels of FPG, and vice versa; (8) had not collected information on prior diagnosis of diabetes; and (9) their mid-year was prior to 2000, before HbA1c assays were widely standardised. Participants were excluded if they (1) were pregnant at the time of measurement; (2) had missing sex or age; (3) had missing 3 nature portfolio | reporting summary March 2021 information on prior diagnosis of diabetes; (4) were 18 years of age or younger; (5) had not been measured for FPG or HbA1c by design or data were missing; (6) were from one specific area in one study in Pakistan with high prevalence of thalassemia; (7) were from follow-up rounds of studies that had multiple measurements of the same cohort over time; (8) had FPG <2 or >30 mmol/L or HbA1c <3% or >18%; (9) had implausible combinations of FPG and HbA1c as determined by the method of local outlier factor. Non-participation We used all studies that met our inclusion criteria, which were designed to ensure participants of the surveys included were representative of the general population from which each sample was drawn. Information on response rate from individual participating studies is not available to us. Randomization Our study is observational, and we did not carry out experiments. Reporting for specific materials, systems and methods We require information from authors about some types of materials, experimental systems and methods used in many studies. Here, indicate whether each material, system or method listed is relevant to your study. If you are not sure if a list item applies to your research, read the appropriate section before selecting a response. Materials & experimental systems n/a Involved in the study Antibodies Eukaryotic cell lines Palaeontology and archaeology Animals and other organisms Clinical data Dual use research of concern Methods n/a Involved in the study ChIP-seq Flow cytometry MRI-based neuroimaging