Differential Associations of GAD Antibodies (GADA) and C-Peptide With Insulin Initiation, Glycemic Responses, and Severe Hypoglycemia in Patients Diagnosed With Type 2 Diabetes

Compared with Caucasians, East Asians, including Chinese, with type 2 diabetes (T2D) are leaner and have lower insulin secretory capacity to overcome obesity-associated insulin resistance (1,2). Apart from genetic/epigenetic factors, life-course events (e.g., low birth weight, childhood obesity) and environmental factors that may affect β-cell function are particularly relevant to Asian populations undergoing rapid societal transition. However, autoimmunity remains an important cause of β-cell dysfunction. GAD autoantibody (GADA) is often used to diagnose autoimmune type 1 diabetes (T1D) (3). Patients with latent autoimmune diabetes in adults (LADA) have islet autoantibodies with nonketotic presentation but a high risk of rapid glycemic deterioration and hypoglycemia (4,5). In European patients with T2D, 9.3–12.0% had LADA versus 3.8–5.9% in their East Asian counterparts (6–8). In Japanese patients with LADA, early insulin therapy slowed the rate of glycemic deterioration more than sulfonylureas (9). Chinese patients with T2D diagnosed before the age of 40 who were GADA+ had marked glucose-lowering response to insulin (10). C-peptide (CP), cosecreted with insulin, is a proxy of endogenous insulin secretion (4). Owing to the legacy effect of poor glycemic control, detecting insulin insufficiency and autoimmunity early for timely insulin treatment may improve clinical outcomes (11,12).

Some experts suggested incorporating GADA and random CP levels in LADA treatment algorithms, although this has yet to be incorporated into treatment guidelines (4,13). Owing to the importance of β-cell dysfunction in Asian populations (14), we retrospectively measured GADA and CP levels in stored serum samples of patients with physician-diagnosed T2D enrolled in the Hong Kong Diabetes Register (HKDR) in 1996–2012 and analyzed their associations with insulin initiation, glycated hemoglobin (HbA_1c) and severe hypoglycemia followed until 2019.

Hong Kong has 7.5 million residents, mainly southern Chinese. The HKDR was established in 1994 at the Prince of Wales Hospital (PWH) Diabetes and Endocrine Centre affiliated to The Chinese University of Hong Kong (CUHK). Patients with physician-diagnosed diabetes receiving routine care in hospital- and community-based clinics and private general practitioners could be referred to the PWH Centre to undergo structured assessment for quality improvement (15). Overnight fasting blood samples were collected, and written informed consent was obtained from patients at enrollment for using the data and stored samples for research purpose.

This analysis was reported according to the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines. Among 9,816 enrollees in 1996–2012, we retrospectively measured GADA and CP in stored serum samples in 5,354 patients observed prospectively until 2019 (Supplementary Fig. 1). Reasons for missing measurements included 1) lack of or insufficient samples and/or 2) lack of consent for sample use. Exclusion criteria included unknown type of diabetes or T1D with acute ketoacidosis or continuous requirement of insulin within 1 year of diagnosis. The study was approved by the CUHK Clinical Research Ethics Committee (CREC ref no.: 2009-421).

All participants underwent a structured assessment by trained nurses using a case report form and standard procedures. Data included demographic status, medical history (use of oral glucose-lowering drugs [OGLDs] and insulin), clinical (body weight, body height, blood pressure [BP], eye and feet examination) and laboratory measurements, including fasting plasma glucose (FPG), HbA_1c, lipids (fasting total cholesterol [TC], triglyceride [TG], HDL-cholesterol [C], and calculated LDL-C), kidney function, and random spot urinary albumin-to-creatinine ratio. Estimated glomerular filtration rate (eGFR) was calculated by the Chronic Kidney Disease Epidemiology Collaboration equation (16). Disease history was captured by self-report or the International Classification of Diseases, Ninth Revision (ICD-9) in the territory-wide electronic medical record system. All biochemical assays were performed at the PWH Department of Chemical Pathology, with external accreditation.

Collected blood samples were stored at −80°C. Serum CP levels and GADA were measured retrospectively for research purpose. CP was measured by Mercodia C-peptide ELISA kit, with a lower limit of detection of 25 pmol/L. The intraassay and interassay coefficients of variation (CV) were <6.8% and <4.8%, respectively. GADA was measured using the GAD Autoantibody ELISA kit (RSR Limited), with a lower limit of detection of 0.57 units/mL. The respective intraassay and interassay CVs were 3.5–7.3% and 5.2–6.4%, respectively, for a GADA range of 5.7–97 units/mL. All assays were performed at the CUHK Diabetes and Obesity Laboratory, Li Ka Shing Institute of Health Sciences at the PWH.

The Hospital Authority (HA) governs all public hospitals and clinics, which provided 95% of hospital bed-days and 80% of outpatient visits in Hong Kong. All HA facilities shared a territory-wide electronic medical record system consisting of admission records, laboratory results, and prescriptions. The latter were dispensed on site after the medical consultations. We retrieved the principal diagnosis of hospital discharges based on ICD-9, date of insulin prescription, and HbA_1c records from enrollment date to 31 December 2019.

We defined insulin initiation as the first insulin prescription lasting for ≥28 days during outpatient visits. Severe hypoglycemia was captured by hospitalization records (ICD-9: 250.3, 250.8, or 251.2) (17). In new insulin users, we defined glycemic response as changes in HbA_1c between baseline and the end point (months 6 and 12). HbA_1c was captured within 3 months before insulin initiation and at the end point after insulin initiation.

GADA+ was defined as GADA titers >5 IU/mL, and low CP was defined as CP <200 pmol/L. The latter was used as a cutoff value for autoimmune T1D (18). Adequate β-cell function was defined as CP ≥200 pmol/L. Obesity was defined as BMI ≥25 kg/m² using Asian criteria (19). Metabolic syndrome was defined by the NCEP-ATP III (National Cholesterol Education Program Adult Treatment Panel III) criterion using the Asian obesity definition with two or more components (diabetes being one component): waist circumference ≥90 cm (men) or ≥80 cm (women), plasma TG ≥1.7 mmol/L, HDL-C <1.03 mmol/L (men) or <1.3 mmol/L (women), and/or BP ≥130/85 mmHg (20).

Data are presented as mean (SD) or median (interquartile range [IQR]). The Student t test, Mann-Whitney U test, χ², Fisher exact test, or ANOVA were used for comparative analysis as appropriate. We examined the independent associations of CP and GADA and their interactions with 1) insulin initiation, 2) glycemic responses, and 3) severe hypoglycemia adjusted for covariables based on prior knowledge (Supplementary Methods). We used Schoenfeld residuals to assess the proportional hazard assumption. Variables that violated the proportional hazard assumption were included as strata in the Cox models. Patients were stratified by GADA and CP status for Kaplan-Meier analysis to estimate the median time to insulin initiation and severe hypoglycemia with the log-rank test for intergroup differences. The likelihood ratio test was used to estimate interaction effects. The proportions of missing data were <2% for most of the baseline variables, except for history of retinopathy (9.3%), baseline treatment of OGLDs (7.5%), and renin-angiotensin system inhibitors (9.3%).

We undertook the following sensitivity analyses: 1) we repeated the analyses using CP level <250 pmol/L to define poor β-cell function (21); 2) we repeated the analyses in patients with preserved eGFR >30 mL/min/1.73 m² since CP is mainly excreted by the kidney (22); 3) we excluded patients with FPG <4 mmol/L and/or hypoglycemia at baseline, which might affect CP levels; 4) we defined insulin requirement as the composite outcome of continuous insulin use or failure of OGLDs to address the time gap between the need versus actual use of insulin; 5) we analyzed HbA_1c responses to insulin in patients initiated with basal insulin only (96.5%); 6) we analyzed risks for severe hypoglycemia, adjusting for baseline insulin regimens; and 7) we calculated the glucose-to-CP ratio as a proxy of insulin resistance and performed subgroup analysis stratified by its median. All analyses were performed using R 3.6.3 software (R Foundation for Statistical Computing, Vienna, Austria).

Consent had not been obtained from study participants for data sharing in public domain, but summary data are available from the corresponding author upon reasonable request.

Among 9,816 patients enrolled in 1996–2012, GADA and CP were measured in 5,230 patients (mean [SD] age 56.5 [13.9] years, age at diagnosis of 48.4 [13.2] years, and median diabetes duration of 6 [IQR 1, 12] years). These patients were younger at enrollment and less likely to have poor glycemic control and cardiovascular-renal complications than those without GADA and CP measurements, although the differences were small (Supplementary Table 1).

At baseline, 28.4% of patients were treated with insulin, and after a median follow-up period of 7.5 (IQR 2.6, 13.1) years, 2,017 of 3,745 insulin-naïve patients (54%) were initiated with insulin (66.3 per 1,000 person-year). There were 257 GADA+ patients (4.9%), and 1,494 patients (28.6%) had a low CP level. In the GADA+ group, 119 (46.3%) had low CP, of whom 93 had CP below the detection limit. The respective values were 1,375 (27.6%) and 990 in the GADA− group. Patients with age of diagnosis of <40 were more likely to have GADA+ (8.5% vs. 3.5%) and low CP (52.7% vs. 18.9%) than those diagnosed after the age of 40 (Supplementary Fig. 2).

The GADA+ and GADA− groups had similar disease duration and frequency of family history (Supplementary Table 2). The GADA+ group had lower BMI and TG, higher FPG and HbA_1c, and was less likely to have microvascular complications or use lipid-lowering drugs or OGLDs, except for insulin, than the GADA− group. The low-CP group was younger and had longer diabetes duration, with similar frequency of family history versus the high-CP group (Supplementary Table 3). They had lower BMI, BP, and TG but higher HbA_1c with similar FPG and were less likely to have cardiovascular-renal complications than the high-CP group. Fasting CP was negatively associated with disease duration, with an explained variance of 0.03 (r = −0.17, P < 0.001) (Supplementary Fig. 3).

When stratified by GADA and CP status, the GADA+ plus low-CP (GADA+/low-CP) group was the youngest with the youngest age of diagnosis, while the GADA−/high-CP group was oldest, with the oldest age of diagnosis (Table 1). The GADA+/low-CP group had the lowest BMI and highest HbA_1c level. The GADA−/low-CP group had the longest diabetes duration and highest proportions of patients with retinopathy and neuropathy. The GADA−/high-CP group had the highest frequencies of chronic kidney disease and cardiovascular disease at baseline than the other groups.

The median frequency of HbA_1c records was 2.5 (IQR 1.8, 3.0) times per year. At baseline, the GADA+ group was more likely to be treated with insulin than the GADA− group (45.9% vs. 27.5%, P < 0.001) (Supplementary Table 2). In the insulin-naïve patients, the median time to insulin initiation (i.e., 50% of patients started on insulin) was 7.7 (IQR 4.9, 11.0) years in the GADA+ group versus 11.3 (10.9, 11.7) years in the GADA− group (P < 0.001) (Fig. 1B), with an adjusted hazard ratio [HR] of 1.46 (95% CI 1.15–1.84) in the GADA+ versus GADA− group (P = 0.002) (Table 2). When stratified by CP level, GADA+ individuals were more likely to start insulin in the high-CP group (HR 1.61; 95% CI 1.18–2.18; P = 0.003) but not in the low-CP group (1.23; 95% CI 0.79–1.91; P = 0.356). Subgroup analysis did not show heterogeneity stratified by age, sex, age of diagnosis, diabetes duration, and metabolic indexes with the associations most evident in nonobese patients and those with low TG/HDL-C (Supplementary Fig. 4).

The GADA+ group had similar adjusted HbA_1c decrement to the GADA− group at month 6 (−1.4% [95% CI −1.7% to −1.0%] vs. −1.0% [95% CI −1.1% to −1.0%], P = 0.070) and month 12 (−1.3% [95% CI −1.6% to −0.9%] vs. −1.0% [95% CI −1.1% to −1.0%], P = 0.187) (Fig. 2). However, when stratified by CP status, the GADA+ group had greater adjusted HbA_1c reduction than the GADA− group in those with low CP (−1.9% [95% CI −2.4% to −1.4%) vs. −1.0% [95% CI −1.2% to −0.8%], P = 0.006) but not in those with high CP (−1.0% [95% CI −1.4% to −0.6%] vs. −1.1% [95% CI −1.1% to −1.0%], P = 0.746) at month 6, with no difference at month 12 (Fig. 2).

At baseline, the low-CP group was more likely to be treated with insulin than the high-CP group (40.3% vs. 23.6%, P < 0.001) (Supplementary Table 3). Among insulin-naïve patients, the median time to insulin initiation was 12.6 (IQR 11.3, 13.5) years in the low-CP group and 10.9 (IQR 10.5, 11.3) years in the high-CP group (Fig. 1A). The low-CP group had a HR of 0.88 (95% CI 0.77–1.00) for insulin initiation relative to the high-CP group just short of significance (P = 0.051) (Table 2). When stratified by GADA status, low CP was associated with a lower likelihood of insulin initiation in the GADA− group (HR 0.86; 95% CI 0.76–0.99; P = 0.032), whereas no association was observed in the GADA+ group (HR 1.40; 95% CI 0.43–4.56; P = 0.575) (Supplementary Fig. 5). Among new insulin users, the low-CP and high-CP groups had similar HbA_1c reduction at month 6 (−1.1% [95% CI −1.3% to −0.9%] vs. −1.1% [95% CI −1.1% to −1.0%], P = 0.674) and month 12 (−1.1% [95% CI −1.3% to −0.9%] vs. −1.0% [95% CI −1.1% to −0.9%], P = 0.451) (Fig. 2). Glycemia response to insulin between low CP and high CP stratified by GADA status was similar at month 6 and month 12 (Fig. 2).

The GADA+ group had higher adjusted HR of 1.38 (95% CI 1.04–1.83; P = 0.024) for severe hypoglycemia (Table 2). The associations were most evident in patients without obesity/metabolic syndrome and those with low CP, young age, short diabetes duration, or low TG/HDL-C (Supplementary Fig. 6). The low-CP group was more likely to have severe hypoglycemia (HR 1.29; 95% CI 1.10–1.52; P = 0.002) than the high-CP group (Table 2), with associations more evident in patients with low TG/HDL-C (Supplementary Fig. 7).

The associations of GADA positivity and CP status with insulin initiation, glycemic response, and severe hypoglycemia: 1) using a cutoff value of CP of 250 pmol/L to define poor or preserved β-cell function; or 2) excluding patients with eGFR ≤30 mL/min/1.73 m² at baseline; or 3) excluding patients with baseline FPG <4 mmol/L or history of severe hypoglycemia, were similar to the main analysis (Supplementary Tables 6–8). Similar effect sizes for GADA+ and CP status were found between associations with insulin initiation and insulin requirement (Supplementary Table 9). In 1,231 of 1,275 insulin-naïve patients (96.5%) initiated with basal insulin only, we found similar results for glycemic response (Supplementary Fig. 12). We also found similar associations with severe hypoglycemia after adjusting baseline insulin regimens (Supplementary Table 10). The associations of GADA+ with insulin initiation and severe hypoglycemia did not differ in analysis stratified by the median of the glucose-to-CP ratio. Associations between CP and insulin initiation did not differ by the glucose-to-CP ratio (P-interaction = 0.326), although lower CP was associated with lower likelihood of insulin initiation in patients with a below-median glucose-to-CP ratio. CP tended to negatively correlate with severe hypoglycemia in the above median group (P-interaction = 0.036) (Supplementary Table 11).

Diabetes is highly heterogeneous in phenotypes, causes, treatment requirement, and outcomes. Low CP and GADA+ may be used to assess β-cell function and insulin requirement. In this largest prospective cohort of patients diagnosed with T2D, retrospective measurement of stored samples indicated that 28.6% had low CP (<200 pmol/L), 4.9% had GADA+, and 2.3% had both low CP and GADA+. The GADA+ group was most likely to be treated with insulin at baseline, with 50% of them requiring insulin in 8 years. The GADA+ plus low-CP group had the largest HbA_1c reduction of 1.9% at month 6 and 1.5% at month 12 versus 1% in other groups. Low CP and GADA+ alone did not predict glycemic response, but both biomarkers were associated with a high risk of severe hypoglycemia. Counterintuitively, among GADA− patients, the high-CP group was more likely to initiate insulin than the low-CP group. Given this marked phenotypic heterogeneity, GADA and CP might be used to classify T2D for personalizing treatment.

This prospective cohort recruited since 1995 in a public health care setting did not have routine measurement of CP and GADA. Thus, the diagnosis of LADA was missed in at least 5% of patients. This figure was similar to the 5.9% reported in the LADA China Study that recruited newly diagnosed patients but lower than the 9–12% reported in European studies (6–8). European patients have higher frequency of genetic variations associated with autoimmune disease (e.g., LADA and rheumatoid arthritis) than Asians (23). In the HKDR, patients had a median duration of 5 years before referral to the PWH Centre for complication assessment. Thus, the prevalence of LADA might be underestimated given that GADA titers might decline with time (24).

In the LADA China study, which screened 4,880 ketosis-free newly diagnosed patients with diabetes, patients with LADA were leaner, had lower fasting CP, and lower frequency of metabolic syndrome (7). In line with other reports on the phenotypic heterogeneity of LADA (23), only half of our GADA+ patients had a low CP level. The GADA+ plus low-CP group resembled classical T1D with body leanness, few metabolic risk factors, and the highest HbA_1c. Compared with their counterparts with low CP, GADA+ patients with high CP had worse cardiometabolic risk factors but lower HbA_1c, suggesting that additional factors might contribute to these different phenotypes.

At baseline, the GADA+ plus low-CP group had the highest HbA_1c with only 50% treated with insulin. In an 8-year prospective study of 106 LADA patients from China, GADA titers were the most discriminable predictor for β-cell failure, defined as fasting CP <75 pmol/L (25). The HelseUndersøkelsen i Nord-Trøndelag (HUNT) study, which included 175 LADA patients and 2,331 T2D patients of European origin, reported similar findings (26). In our study, GADA+ patients initiated insulin earlier, with a robust glucose-lowering response especially in the presence of low CP, albeit also high risk of severe hypoglycemia. This might be due to autoimmune destruction of both insulin-secreting β-cells and glucagon-secreting α-cells (27). Our results suggested that GADA and CP can identify patients who require an insulin regimen more akin to a physiological pattern (e.g., basal-bolus insulin analogs or insulin pump) with frequent self-monitoring of blood glucose to optimize glycemic control and avoid severe hypoglycemia.

In this cohort, one in four patients with T2D had low CP closely associated with long diabetes duration and age of diagnosis. This is in line with the typical features of T2D in Asian, including a Chinese population, characterized by reduced β-cell function (14). In the HUNT study, CP <300 pmol/L predicted insulin dependency in both patients with LADA and T2D, with respective HRs of 6.4 and 5.0 (26). In the latest guideline for T1D, CP <200 pmol/L was recommended as an indicator of poor β-cell function and early insulin requirement. In our study, CP <200 pmol/L tended to predict early insulin requirement only in GADA+ patients but not in GADA− patients with T2D, suggesting other factors may modulate the association of CP status with insulin requirement.

In our study, GADA− patients with high CP (>200 pmol/L) were more likely to progress to insulin treatment. When stratified by indicators of insulin resistance, including TG/HDL-C and glucose-to-CP ratio, the associations tended to be more evident in those with a higher level of insulin resistance. Besides, the high-CP group had worse cardiometabolic risk factors. Taken together, high CP may indicate “β-cell overburden” due to insulin resistance, which may hasten β-cell failure, especially in those with low β-cell reserve, a feature particularly relevant to Asians (28). Since amylin is cosecreted with CP, we hypothesize that amylin misfolding, accumulation of amyloid polypeptide, and formation of amyloid deposits might also contribute to β-cell failure (29). Nevertheless, this hypothesis needs to be further investigated.

Despite their differential associations with insulin initiation and glycemic responses, GADA+ and low-CP status were both associated with severe hypoglycemia after adjustment for baseline insulin use with different regimens. Reduced counterregulation with loss of glucagon-secreting α-cells due to autoimmune and genetic causes, as well as long disease duration, might be important. In the Diabetes Control and Complications Trial/Epidemiology of Diabetes Interventions and Complications (DCCT/EDIC) Study, including patients with an average duration of 35 years of T1D, and the Danish Study Group for Childhood Diabetes (DanDiabKids) in children and adolescents with 3–6 years of T1D duration, adequate β-cell function was associated with a reduced risk of hypoglycemia (30,31).

There are interethnic differences in T2D regarding genetic profiles, pathophysiology, risk factors, clinical presentations, treatment responses, and outcome (14). Compared with Caucasians, T2D in East Asians, including Chinese, Japanese, and Koreans, was characterized by younger age of diagnosis, low BMI, reduced insulin secretory capacity, and insulin resistance (1). In these patients, minor perturbations in insulin sensitivity might alter glucose homeostasis (1). In countries/areas undergoing rapid changes in the ecosystem and acculturation, people with biological susceptibility might have a high risk of β-cell failure, further influenced by delayed diagnosis and suboptimal care. To this end, each person with T2D had a unique combination of causes and modifiers, including but not limited to genetic, perinatal, or epigenetic factors, hormonal profiles, or pancreatic disease, which may influence the trajectory of β-cell function (32–34). There is increasing advocacy of using biogenetic markers to increase the precision of prediction, prevention, prognostication, and personalized treatment in diabetes. While some researchers had used age, age of diagnosis, C-peptide, BMI, and HOMA values to assess insulin requirement and renal complications (33), our results suggested that the use of GADA and CP could identify Chinese patients with T2D in need of early insulin requirement with special attention to avoid risk of severe hypoglycemia.

Optimal control of glycemia, BP, lipids, and body weight might reduce glucolipotoxicity and inflammation, preserve islet function, and improve glycemic durability in patients with T2D (35,36). In an exploratory analysis, we had reported that patients with high CP treated with insulin had the highest mortality risk, although the risk was attenuated after adjusting for disease duration and comorbidities (37). In the latest Glycemia Reduction Approaches in Diabetes: A Comparative Effectiveness Study (GRADE) report, patients receiving metformin therapy treated with insulin and glucagon-like peptide 1 had the greatest reduction in HbA_1c compared with those treated with sulfonylurea and dipeptidyl peptidase 4 inhibitor (38). Clinical trials are needed to elucidate the interactions among phenotypes, biomarkers, and treatment regimens (e.g., insulin, incretin mimetics, metabolic surgery) in preserving islet function, optimizing glycemic control, and reducing risk of hypoglycemia.

This is the largest prospective cohort of patients with T2D observed for an average of 6 years, with comprehensive documentation of clinical profiles, laboratory results, and end points. This rich information allowed us to reveal differential associations of CP and GADA with insulin requirement and risk of severe hypoglycemia. Our study also had limitations. GADA and CP were measured in blood samples stored at −80°C. Although the sample quality had been ensured using a metabolomic platform, some sample degradation remained possible, with underestimation of GADA+ and high-CP status. However, we have adjusted for assessment year and diabetes duration. We did not measure glucagon-stimulated or random CP level as recommended. In this quality-improvement program with a structured protocol, we collected fasting blood samples and acknowledged that CP might be suppressed by hypoglycemia or during fasting (39). However, after excluding patients with baseline FPG <4 mmol/L and/or hypoglycemia, we found similar results. We used insulin requirement as an outcome to address the time gap between the need and actual use of insulin and found similar results. We did not measure other islet autoantibodies (e.g., antityrosine phosphatase and anti-zinc transporter 8 autoantibodies) although the additional yield for diagnosing LADA was only incremental (40). GADA persisted for the longest period in the postdiagnosis period and is the preferred biomarker in clinical practice. Severe hypoglycemia events occurring or treated outside the public health care system could not be captured. This was a single-center study, which might be biased by referral of atypical or advanced disease, although our analysis focused on risk association which is subject to less influence by such bias. There are few population-based studies with similar data granularity, and our results accorded with similar reports plus new findings.

Using CP and GADA, we identified considerable phenotypic heterogeneity in adult Chinese patients with T2D. Nearly 5% of patients had GADA+ who needed early insulin treatment. These patients had good glycemic response to insulin, especially in the presence of low CP, despite their high risk for severe hypoglycemia. One in four patients had low CP likely due to long disease duration and other causes such as genetic factors. The high insulin use in patients with high CP and metabolic syndrome suggested that insulin resistance might accelerate β-cell loss and drive insulin requirement. The large number of agents and strategies targeting insulin deficiency (e.g., insulin analogs and insulin secretagogues) and resistance (e.g., glucagon-like peptide 1 and insulin sensitizer) call for detailed phenotyping, including measurement of CP and GADA, to increase the precision of classification and treatment in T2D for better outcomes (34).

Acknowledgments. The authors thank all participants of the HKDR cohort for their contribution. Special thanks are extended to the medical and nursing staff of the Diabetes Mellitus and Endocrine Centre, Prince of Wales Hospital, for their professional input and support.

Funding. This study was supported by the Hong Kong Government Health and Medical Research Fund (CFS-CUHK2) and the Research Grants Council Research Impact Fund (R4012-18).

Duality of Interest. J.C.N.C. has received research grants and/or honoraria for consultancy and/or giving lectures from Applied Therapeutics, AstraZeneca, Bayer, Boehringer Ingelheim, Eli Lilly, Hua Medicine, Lee Powder, Merck Serono, Merck Sharp & Dohme, Pfizer, Sanofi and Viatris. A.O.Y.L. has served as an advisory committee member for AstraZeneca, Boehringer Ingelheim, Sanofi, and Amgen, and has received research grants and travel grants from AstraZeneca, Boehringer Ingelheim, MSD, Novartis, Novo Nordisk, Sanofi, and Amgen. R.C.W.M. has received research grants for clinical trials from AstraZeneca, Bayer, MSD, Novo Nordisk, Sanofi, and Tricida, and honoraria for consultancy or lectures from AstraZeneca and Boehringer Ingelheim. A.P.S.K. has received research grants and/or speaker honoraria from Abbott, AstraZeneca, Bayer, Boehringer Ingelheim, Eli Lilly, Merck Serono, Nestlé, Novo Nordisk, and Sanofi. M.H.T.W. is a shareholder of Beth Bioinformatics Co., Ltd. J.C.N.C. and R.C.W.M. hold patents for using biogenetic markers to predict risk of diabetes and its complications. J.C.N.C., C.K.P.L., and R.C.W.M. are co-founders of GemVCare, a biotech start up supported by the Technology Start-up Support Scheme for Universities of the Hong Kong Government Innovation and Technology Commission. No other potential conflicts of interest relevant to this article were reported.

Author Contributions. B.F. contributed to the statistical analysis, results interpretation, drafting and revising the manuscript critically, and approved the final version. C.K.P.L., E.W.M.P., E.S.H.L., H.W., A.Y., M.S., C.H.T.T., S.Y.S.W., E.K.-P.L., M.H.T.W., and N.H.S.C. contributed to results interpretation, revised the manuscript critically, and approved the final version. R.O., A.P.S.K., E.C., and R.C.W.M. revised the manuscript critically and approved the final version. A.O.Y.L. contributed to the statistical analysis, results interpretation, revised the manuscript critically, and approved the final version. J.C.N.C. contributed to the conception of the study, statistical analysis and results interpretation, drafting of the manuscript and revising the manuscript critically, and approved the final version. All authors contributed meaningfully to this manuscript and approved the final version. J.C.N.C. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Prior Publication. Parts of this study were presented in abstract form at the 82nd Scientific Sessions of the American Diabetes Association, virtual and at New Orleans, LA, 3–7 June 2022.