Obesity, clinical, and genetic predictors for glycemic progression in Chinese patients with type 2 diabetes: A cohort study using the Hong Kong Diabetes Register and Hong Kong Diabetes Biobank

Background Type 2 diabetes (T2D) is a progressive disease whereby there is often deterioration in glucose control despite escalation in treatment. There is significant heterogeneity to this progression of glycemia after onset of diabetes, yet the factors that influence glycemic progression are not well understood. Given the tremendous burden of diabetes in the Chinese population, and limited knowledge on factors that influence glycemia, we aim to identify the clinical and genetic predictors for glycemic progression in Chinese patients with T2D. Methods and findings In 1995–2007, 7,091 insulin-naïve Chinese patients (mean age 56.8 ± 13.3 [SD] years; mean age of T2D onset 51.1 ± 12.7 years; 47% men; 28.4% current or ex-smokers; median duration of diabetes 4 [IQR: 1–9] years; mean HbA1c 7.4% ± 1.7%; mean body mass index [BMI] 25.3 ± 4.0 kg/m2) were followed prospectively in the Hong Kong Diabetes Register. We examined associations of BMI and other clinical and genetic factors with glycemic progression defined as requirement of continuous insulin treatment, or 2 consecutive HbA1c ≥8.5% while on ≥2 oral glucose-lowering drugs (OGLDs), with validation in another multicenter cohort of Hong Kong Diabetes Biobank. During a median follow-up period of 8.8 (IQR: 4.8–13.3) years, incidence of glycemic progression was 48.0 (95% confidence interval [CI] 46.3–49.8) per 1,000 person-years with 2,519 patients started on insulin. Among the latter, 33.2% had a lag period of 1.3 years before insulin was initiated. Risk of progression was associated with extremes of BMI and high HbA1c. On multivariate Cox analysis, early age at diagnosis, microvascular complications, high triglyceride levels, and tobacco use were additional independent predictors for glycemic progression. A polygenic risk score (PRS) including 123 known risk variants for T2D also predicted rapid progression to insulin therapy (hazard ratio [HR]: 1.07 [95% CI 1.03–1.12] per SD; P = 0.001), with validation in the replication cohort (HR: 1.24 [95% CI 1.06–1.46] per SD; P = 0.008). A PRS using 63 BMI-related variants predicted BMI (beta [SE] = 0.312 [0.057] per SD; P = 5.84 × 10−8) but not glycemic progression (HR: 1.01 [95% CI 0.96–1.05] per SD; P = 0.747). Limitations of this study include potential misdiagnosis of T2D and lack of detailed data of drug use during follow-up in the replication cohort. Conclusions Our results show that approximately 5% of patients with T2D failed OGLDs annually in this clinic-based cohort. The independent associations of modifiable and genetic risk factors allow more precise identification of high-risk patients for early intensive control of multiple risk factors to prevent glycemic progression.


Conclusions
Our results show that approximately 5% of patients with T2D failed OGLDs annually in this clinic-based cohort. The independent associations of modifiable and genetic risk factors allow more precise identification of high-risk patients for early intensive control of multiple risk factors to prevent glycemic progression.

Author summary
Why was this study done?
• There is marked heterogeneity in the rate of progression to insulin requirement in patients with type 2 diabetes (T2D). Whereas some patients can maintain optimal glycemic control with oral glucose-lowering drugs for prolonged periods, others experience rapid deterioration in glycemia, requiring early insulin treatment.
• It is important to gain insight into what factors are associated with glycemic progression. Previous studies identified baseline HbA1c, young age, and weight gain as independent clinical predictors for glycemic progression in White populations, but there is a paucity of data in Asians.
• Besides environmental and lifestyle factors including obesity, it is well known that genetic factors play a pivotal role in the onset of diabetes. However, it remains unclear whether genetic variants of body mass index (BMI) and T2D predicts glycemic Introduction Type 2 diabetes (T2D) is a progressive and multisystemic disease with hyperglycemia as a major risk factor for complications [1]. Delayed intervention, patient nonadherence, heterogeneity of phenotypes and treatment responses can lead to different trajectories in treatment escalation. Whereas some patients can be controlled with oral glucose-lowering drugs (OGLDs) for decades, others experience rapid deterioration in glycemia, requiring early insulin treatment [2]. In Whites with T2D, baseline HbA1c, young age, and weight gain independently predicted glycemic progression [3]. In the IMI-DIRECT study, researchers used time to insulin initiation as an index of glycemic progression and reported that patients with low body mass index (BMI) <24 kg/m 2 and BMI >30 kg/m 2 were susceptible to rapid glycemic progression [2]. Genetic factors play a pivotal role in the onset of diabetes although this is strongly influenced by environmental and lifestyle factors, notably obesity. The heritability of T2D ranges from 30% to 70% in different populations with the discovery of a large number of genetic variants mainly implicated in beta-cell biology [4]. Although genetic analysis also supports the causal role of obesity in T2D [5], it is unclear whether genetic variants of BMI and T2D predict glycemic progression once T2D develops. In the IMI-DIRECT study, a high polygenic risk score (PRS) comprising 61 risk variants of T2D was associated with a young age at diagnosis but not time to insulin requirement in European patients [2]. Furthermore, pharmacogenetic studies have identified variants associated with poor response to OGLD, such as metformin [6][7][8][9][10][11][12][13], sulphonylurea (SU) [14][15][16][17][18], and thiazolidinediones (TZD) [19][20][21].

PLOS MEDICINE
We hypothesize that clinical and genetic variants associated with T2D and response to OGLDs predict glycemic progression in Asian patients with T2D. These predictors may identify high-risk patients for early intensification and individualization of treatment to prevent glycemic progression and development of complications. In a prospective cohort with documentation of clinical and genetic profiles and treatment progression, we explored the associations of clinical variables and different PRSs of obesity, T2D, and response to OGLDs with glycemic progression in insulin-naïve Chinese patients with T2D.

Study population
This study utilized data from the Hong Kong Diabetes Register (HKDR) established in 1995 at the Prince of Wales Hospital (PWH), the teaching hospital of the Chinese University of Hong Kong [22,23]. The HKDR consecutively enrolled patients referred to the Diabetes Mellitus and Endocrine Centre for comprehensive assessment of complications and metabolic control. Referral sources included hospital-based specialty clinics, community clinics, and general practitioners. Once registered, patients were observed to the time of death. Ethical approval was obtained from the Clinical Research Ethics Committees of the Chinese University of Hong Kong. Written informed consent was obtained from all participants at the time of enrollment for collection of clinical information and biosamples for archival and research purposes. The recruitment methods, definitions, and biochemical investigations have been described [22,24].
Between 1995 and December 31, 2007, a consecutive cohort consisting of 10,129 patients with diabetes was assessed. After sequentially excluding 61 patients with non-Chinese or unknown ethnicity, 400 with type 1 diabetes (T1D) or missing data on type of diabetes, 2,544 with insulin use or considered having OGLD failure at baseline (see definition below), and 33 requiring insulin within 1 year, we prospectively curated a primary cohort of 7,091 Chinese patients with detailed information on risk factors, complications, drug use, and clinical outcomes for further analysis (S1 Fig).
For validation of genetic association, we used an independent prospective cohort from Hong Kong Diabetes Biobank (HKDB), which recruited patients from 11 diabetes centers at major public hospitals across Hong Kong since 2014. Recruitment and assessment methods were similar to that in HKDR. A total of 6,914 unrelated patients from HKDB were genotyped. In this replication cohort, after excluding 162 patients with non-Chinese or non-T2D, 2,675 with insulin use at baseline, 3 requiring insulin within 1 year, and 135 patients who overlapped with HKDR, we included 3,939 patients with T2D for this replication study (S2 Fig). All patients gave written informed consent for DNA collection and data analysis for research purposes. Genetic data from HKDB became available for replication analysis around April 2019.

Baseline clinical and laboratory measurements
Clinical assessment was performed based on a modified European DiabCare protocol [25]. Details of assessment methods and definitions of clinical outcomes have been described [24,26]. In brief, all patients in the HKDR and HKDB underwent clinical assessments and laboratory investigations after 8-hour overnight fast, including eye, feet, urine, and blood examinations. Eye examination included visual acuity and fundoscopy through dilated pupils or retinal photography. Retinopathy was defined by typical changes due to diabetes, laser scars, or a history of vitrectomy. Foot examination was performed using Doppler ultrasound scan and monofilament and graduated tuning fork. Fasting blood was sampled for measurement of plasma glucose, HbA1c, and lipid profile (total cholesterol, high-density lipoprotein cholesterol [HDL-C], triglycerides, and calculated low-density lipoprotein cholesterol [LDL-C]), and random spot urinary sample was used to assess albumin-to-creatinine ratio (ACR). The Chronic Kidney Disease Epidemiology Collaboration equation was used to estimate glomerular filtration rate (eGFR) [27].

Definition of clinical outcomes
Clinical outcomes were defined using hospital discharge diagnoses based on the International Classification of Diseases, Ninth Revision (ICD-9) and mortality as censored on or before June 30, 2014. The Hong Kong Hospital Authority Central Computer System records admissions to all public hospitals, which provides about 95% of inpatient bed-days in Hong Kong. All hospitalization records were retrieved from this system using a unique identifier number. Given the frequent delay in starting insulin due to patients' reluctance or clinicians' delayed decision, we used a composite end point for "requirement of insulin treatment" to define glycemic progression, (1) progression to continuous insulin treatment (more than 6 months' duration), or (2) failure of OGLDs (2 consecutive HbA1c �8.5%, more than 3 months apart during treatment with �2 OGLDs [metformin, SU, or TZD]), in line with the definition used in the IMI-DIR-ECT study [2]. Follow-up time was defined as the period from baseline visit to the date of the first clinical end point or the censored dates, whichever came first. Since the OGLD data during follow-up in the HKDB were not available for this analysis, we used continuing insulin treatment as the end point in the replication cohort.
We extracted these SNPs from imputed genotyping data using the Illumina Omni 2.5 + exome array (Illumina, San Diego, CA). We selected independent common SNPs (linkage disequilibrium [LD] coefficient r 2 < 0.5; minor allele frequency [MAF] > 0.01) available in our dataset with good genotyping quality to construct the final PRSs. Standard quality control (MAF > 0.01; call rate > 97%; P > 0.0001 in Hardy-Weinberg equilibrium) was performed and genotype data were imputed using minimac3 with the 1000 Genomes Project phase 3 v5 as reference panel [23]. Finally, we obtained 123, 48, 63, 8, 7, and 3 independent common SNPs to develop the European-T2D, Asian-T2D, BMI, metformin, SU, and TZD PRSs, respectively (S1-S6 Tables). Because of the small number of SNPs associated with each OGLD, we created a combined PRS using all 18 SNPs for drug responses. Each PRS was first constructed by summing the score of reported risk allele for each SNP based on an additive genetic model and then rescaled to a score to express the standard deviation (SD) using the following formula: (individual PRS value − population mean PRS) � population SD of PRS. The final analysis is based on (1) the genetic risk per SD of the standardized PRS and (2) tertile analysis of the PRS for glycemic progression (tertile 1: low; tertile 2: intermediate, and tertile 3: high).
In the replication cohort, we performed genotyping using the Illumina Infinium Global Screening Array (GSA) and conducted quality control and imputation using similar criteria. A proxy SNP with r 2 >0.6 (according to the 1000 Genome CHB panel) was selected when the index SNP was imputed with poor quality. One European-T2D variant (rs17405722 near STAT3 gene) with poor proxies was dropped, and 1 SU variant (rs1801278 in IRS1 gene) was replaced with a proxy variant (rs56171406; r 2 = 1).

Statistical analysis
We used Cox proportional hazards regression model for prospective analysis. We examined the Cox proportional hazards assumption by plotting the scaled Schoenfeld residuals against log-transformed time for each covariate and further tested the assumption via the test proposed by Grambsch and Therneau [31]. As suggested by reviewers, we used 3-knot restricted cubic spline to plot the relationship between BMI as a continuous variable (rather than BMI categories) and glycemic progression. Because of the nonlinear relationship between BMI and rate of progression to insulin requirement, we divided BMI into 4 groups using the World Health Organization (WHO) Asian definition for obesity-underweight (BMI < 18.5 kg/m 2 ), normal (�18.5-23.0 kg/m 2 ), overweight (�23.0-25.0 kg/m 2 ), and obese (�25.0 kg/m 2 )-for the survival analysis [32]. We assessed crude survival (for non-insulin requirement) with the Kaplan-Meier estimator, stratified for BMI categories. Since the proportional hazard assumption for baseline HbA1c was not met, we stratified the cohort into 3 groups (HbA1c < 7%, � 7-9%, and � 9%) to allow for a different hazard function in each group. Both BMI and HbA1c categories were included as strata variables in Cox models, whereas other covariates were considered to have same effect across strata. A procedure of stepwise model selection by the Akaike information criterion (AIC) was employed to identify clinical predictors for glycemic progression in multivariate analysis. Linear regression was used to examine associations between PRSs and quantitative traits.
All data were expressed as percentages; means and SDs; or medians and interquartile ranges (IQRs) as appropriate. All statistical tests were 2-sided, and a P value <0.05 was considered to be significant. Analyses were performed using R (version 3.1.0; http://www.R-project.org). Among the progressors, 2,519 were actually started on insulin, with 33.2% of them (n = 836) experiencing a median lag period of 1.3 (IQR: 0.4-3.0) years before insulin was initiated. Compared with nonprogressors, patients with glycemic progression had younger age of diagnosis; longer duration of diabetes; and higher BMI, HbA1c, triglyceride, LDL-C, and urinary ACR and were more likely to be smokers and have microvascular complications (Table 1). In the replication cohort comprising 3,939 patients, 172 (4.4%) had glycemic progression during a  Table).

Association of clinical factors with diabetes progression
There was a U-shape association between BMI and glycemic progression (P < 0.001 for nonlinearity; Fig 1)

Association of PRS with glycemic progression
After adjusting for clinical predictors (Fig 3 and S8 Table). No significant associations were found for other PRSs or individual risk variant (S1-S6 Tables). On multivariate linear regression with adjustment for confounding factors, the European-T2D PRS and Asian-T2D PRS were associated with earlier age at diagnosis (beta  Table).

Sensitivity analysis
We performed the following sensitivity analyses to assess the robustness of our findings. First, we repeated the analysis by defining the subcomponent of glycemic progression using the end  per SD, P = 0.013, respectively) (S10 Table). No significant associations were found for the other PRSs (S10 Table). Second, since HDL-C was independently associated with progression to insulin in European patients [2], we replaced triglyceride and LDL-C with HDL-C in multivariate Cox analysis and confirmed its association with glycemic progression in our cohort (HR: 0.85 [95% CI 0.74-0.98], P = 0.025; S11 Table). We weighted the European-T2D and Asian-T2D PRS using the logarithm of the allelic odds ratio of the SNP as reported in the meta-analyses and examined the associations with glycemic progression. The results were similar to that from the unweighted PRS (HR: 1.06 [95% CI 1.01-1.11] per SD; P = 0.013 for the European-T2D PRS).
Third, taking into consideration of the relationship between BMI and T2D, some T2Drelated SNPs may be associated with BMI. Therefore, following suggestions from reviewers, we updated the European-T2D PRS and Asian-T2D PRS by excluding those BMI-related SNPs and examined the associations between the updated PRSs and glycemic progression. We first checked for LD between the European T2D-related SNPs, or Asian T2D-related SNPs, and the list of SNPs associated with BMI using data from 1000 Genome project and data for Han Chinese population (CHB). We found that in the European T2D-related SNPs, a total of 7 SNPs (rs2867125, rs2972144, rs2307111, rs72892910, rs7903146, rs7185735, and rs12970134) had r 2 >0.6 with any one of the BMI-related SNPs on the same chromosome. For the Asian T2D-related SNPs, we identified 4 SNPs (rs2943641, rs7903146, rs9939609, and rs12970134) that had r 2 >0.6 with any one of the BMI SNPs on the same chromosome. After exclusion of these SNPs with high LD with BMI-related SNPs, we reconstructed the European-T2D PRS and Asian-T2D PRS and repeated the analysis. The results showed that the association between the European-T2D PRS was similar to the results of our original analysis (S12 Table). Furthermore, we identified 17 SNPs from the European T2D-related SNPs, which showed significant association at level of 0.05 with baseline BMI in our primary cohort of HKDR. The identities of these SNPs are rs780094, rs12617659, rs13389219, rs4865796, rs459193, rs10077431, rs7756992, rs853974, rs12681990, rs516946, rs10811661, rs2796441, rs2237892, rs576674, rs963740, rs982077, and rs8108269. Likewise, 11 SNPs from the list of Asian T2D-related SNPs showed association with baseline BMI (rs780094, rs3923113, rs1470579, rs459193, rs7756992, rs791595, rs7041847, rs10811661, rs2796441, rs2237892, and rs7178572). We further removed these SNPs from the respective European-T2D PRS and Asian-T2D PRS, in addition to removing the SNPs with high LD with BMI-related SNPs, and repeated the association analysis. Interestingly, the association between the European-T2D PRS and diabetes progression became stronger, whereas the results for the Asian-T2D PRS and diabetes progression remained similar, with no evidence of association (S12 Table). Replication analyses in the HKDB showed similar results (S13 Table).
Fourth, in order to explore the impact of the different drug response PRS stratified according to prescribed drugs, we investigated the associations of metformin PRS and SU PRS with glycemic progression by stratifying the exposure time of each oral drug among new users of oral drug by dividing into tertiles according to duration of drug exposure during follow-up. The interaction term between duration of metformin exposure and metformin PRS was not significantly associated with glycemic progression. However, the HR for metformin PRS increased with increasing exposure to metformin during the follow-up period, with nominal significance for the high-exposure subgroup (HR: 1.16 [95% CI 1.01-1.33]; P = 0.04), with adjustment for confounding factors (S14 Table). However, no significant association was observed for the metformin PRS if stratified according to duration of exposure to SU. Although the sample size was significantly reduced because of exclusion of prevalent drug users in each group and the smaller subgroups, this provides suggestive evidence that the association between metformin PRS and glycemic progression in users of metformin is mainly driven by those with longer duration of exposure to the drug.
Fifth, as the replication cohort, HKDB, included patients with a longer duration of diabetes and worse renal function compared with those in HKDR, we investigated the potential impact of CKD on the associations of PRS and glycemic progression. After excluding 688 with history of CKD, 3,251 patients from HKDB were included in this analysis (mean age 59.5 ± 10.

Discussion
There is considerable heterogeneity in glycemic progression in T2D. In this prospective cohort of the HKDR, we analyzed time to requirement or initiation of insulin in patients with T2D followed for a median duration of 8.8 years. Our main findings were 3-fold. First, there was an approximately 1.3-year delay of insulin use after OGLD failure in real-world practice. Second, young age of diagnosis, high HbA1c and triglyceride, extremes of BMI, use of tobacco, and microvascular complications predicted rapid glycemic progression [2,3]. Third, genetic variants associated with T2D in European populations predicted early age of diagnosis and rapid glycemic progression. These findings emphasize the importance of both genetic and modifiable risk factors, notably lipo-glucotoxicity, obesity, and tobacco use on glycemic progression, which enables identification of high-risk patients for optimal control of risk factors to improve glycemic durability.
Among progressors to insulin requirement, 33.2% experienced delay of 1.3 years before actual initiation of insulin. Compared with nonprogressors or those with more prompt insulin initiation, patients with delayed insulin treatment were younger and had shorter disease duration, lower blood pressure, and higher BMI. In a recent study from the United States, approximately 30% of patients with T2D had delayed insulin therapy, with an average lag period of 2 years [33]. In our analysis, the presence of microvascular complications was associated with earlier start of insulin, raising the possibility that delayed insulin and increased glycemic burden might contribute to increased risk of complications especially in young patients [34]. Inconvenience, fear of injection, insufficient empowerment, physicians' concerns of patient nonadherence, and lack of supporting system might delay insulin initiation [35,36]. These barriers can be overcome by the use of team-based care with focus on education/self-care, task shifting (e.g., using nurse educator), and providing feedback to promote physician-patient communication, especially in low-resource settings [37].
It is well established that smoking is an independent risk factor for T2D [38], but our study further revealed that smoking can also influence the glycemic progression once T2D is diagnosed, suggesting that cessation of smoking is a key factor for prevention and management of T2D. Similar to reports in European populations [2], extremes of BMI predicted glycemic progression in our population, but not BMI PRS. The association of high BMI might be more related to lifestyle factors, whereas low BMI might be associated with atypical forms of diabetes due to autoimmunity or uncommon genetic variants. Indeed, in a subgroup analysis, 1 in 5 patients with BMI <18 kg/m 2 had GADA. Although our sample size had 87% power to detect an association between BMI PRS and glycemic progression at α-level of 0.05, assuming an HR of 1.07, only 0.5% variance of BMI was explained by the PRS, comprising 63 risk variants.
There are relatively few GWASs on obesity in Asians, and more BMI-related genetic variants are needed to validate the genetic association of obesity and glycemic progression.
In the IMI-DIRECT study including 5,250 patients with T2D, a high PRS derived from 61 T2D risk variants predicted young age at diagnosis but not glycemic progression [2]. In this larger cohort including over 7,000 Chinese patients with T2D, we derived a European-T2D PRS using 123 variants that predicted age at diagnosis and glycemic progression. The Asian-T2D PRS also predicted age at diagnosis but not glycemic progression. The considerably fewer Asian-T2D SNPs compared with the European-T2D SNPs might have reduced the power of the study. However, using progression to insulin as an end point in a subgroup analysis, the association with Asian-T2D PRS was validated. Given the small effect size of these common SNPs, the use of genome-wide PRS based on thousands of SNPs may further improve the discriminative performance in predicting complex traits [39]. For genetic risk of drug responses, only the metformin PRS predicted rapid glycemic progression, albeit not replicated in the HKDB cohort. We did not find any associations with SU PRS and TZD PRS. These negative associations may be due to secular changes in drug use between the primary and replication cohorts. Over 90% of patients were treated with metformin-based therapy in the primary cohort established since 1997, whereas newer drugs (e.g., dipeptidyl peptidase-4 [DPP-4] inhibitors) were more frequently prescribed in the replication cohort established since 2014. Small sample size, long disease duration, and short period of follow-up in the replication cohort may also reduce its power to detect the association. There are few pharmacogenetic studies in Asian population on OGLDs, calling for more large-scale GWASs to fill this knowledge gap.
One of the important applications from the use of such PRS in our study is to identify potentially important pathways involved in glycemic progression and generate new etiological insights. SNPs associated with T2D are most commonly associated with beta-cell function, highlighting the potential central role of beta-cell dysfunction in both the pathogenesis of T2D and glycemic progression following onset of T2D. The lack of association between the PRS for BMI and glycemic progression suggests that weight loss alone, although useful to improve metabolic control, may not have sustained effects on slowing glycemic deterioration. Recent studies in mainly European cohorts have generated phenotyped-based clusters, pathophysiologybased clusters, and PRSs that reflect different pathophysiological processes ranging from betacell dysfunction to adiposity to lipodystrophy [40][41][42]. Although beyond the scope of the current manuscript, we plan to explore the importance of these pathophysiological defects in relation to glycemic progression using these pathway-based PRSs in future analyses. The fact that over 90% of patients in HKDR were treated with metformin at some stage, and the association between metformin PRS and glycemic progression, highlights the potential utility for precision medicine in diabetes, whereby genetic determinants of response to key glucose-lowering drugs have important modulating effects on disease progression, and therefore can be used to guide treatment escalation. On the basis of findings from the current study, we have already embarked on translational studies to evaluate the use of such biomarker information to guide clinical treatment decisions.
We acknowledge several limitations in our study. First, despite excluding patients with T1D and those requiring insulin within 1 year of diagnosis, it remains possible that some patients were misdiagnosed as T2D, especially for those with low BMI. In a sensitivity analysis of a subgroup of patients, predominantly those with young-onset diabetes who had available GADA data, approximately 7% had positive GADA [43]. Although there were more patients with positive GADA in the low-BMI group compared with the other groups (19% versus 5%-10%), only 25% of patients had low BMI, and thus the impact of misclassification was only modest. However, measurement of these biomarkers, especially in young and/or lean patients, may avoid undue delay in insulin treatment. In the United Kingdom Prospective Diabetes Survey involving 3,672 patients with T2D, GADA (defined as titer >20 U/L by a radioimmunoprecipitation assay) was present in 9.8% of adult patients, who were more likely to require insulin within 6 years of diagnosis than those without GADA [44]. Second, since different definitions of phenotypes, especially for drug response, were used in previous studies together with some variants not being identified in our GWAS dataset, we did not weigh the PRSs using the reported effect sizes of these risk alleles. However, we used weighted European-T2D PRS and found similar results in the sensitivity analysis. Third, we did not have a replication cohort with detailed follow-up data of drug use as the primary cohort. However, the associations between clinical variables and end point were similar in both discovery/validation cohorts and consistent with previous studies.
In conclusion, in Chinese patients with T2D, extremes of BMI were associated with rapid glycemic progression, suggesting heterogeneous etiologies and effects of obesity in T2D. Young age at diagnosis, long disease duration, suboptimal lipid and glucose control, tobacco use, microvascular complications, and genetic variants of T2D all independently predicted glycemic progression or requirement of insulin treatment. Our study suggests potential overlap between the pathogenesis of T2D and glycemic progression. The contributions of both modifiable and nonmodifiable risk factors enable the precise identification of high-risk individuals for close monitoring and early treatment intensification to maintain glycemic durability.