Discussion
This study is among the few to identify multiple subphenotypes of individuals with pre-diabetes who are highly susceptible to type 2 diabetes and CVDs. We examined the associations between these subphenotypes and the incidence of type 2 diabetes, vascular complications and mortality. Our findings suggest that six commonly measured clinical parameters—age, BMI, HbA1c, FPG, HDL-C and ALT—can classify individuals with pre-diabetes into six clusters. Each of these clusters exhibits a different trajectory regarding diabetes progression, vascular complications, and mortality.
The clusters fall into two main subgroups: low risk and high risk. The low-risk subgroup comprises cluster 1 (low risk), cluster 2 (mild dysglycemia elderly) and cluster 4 (mild dysglycemia obese). Clusters 1 and 2 are associated with less progression to type 2 diabetes, whereas cluster 4 is associated with reduced risks of developing type 2 diabetes, microvascular complications and mortality. The high-risk subgroup encompasses cluster 3 (severe dysglycemia obese), cluster 5 (severe dysmetabolic obese) and cluster 6 (severe dysglycemia elderly). These clusters demonstrate a heightened risk of progressing to type 2 diabetes, with cluster 6 being particularly prone to composite macrovascular complications, CeVD and increased mortality.
Cluster 6 represents the highest risk phenotype, characterized by significantly elevated risks of CeVD, macrovascular complications and type 2 diabetes, with marginal increases in CAD and mortality. Conversely, cluster 4 emerged as the lowest risk phenotype, showing significantly decreased risks of CKD, composite microvascular complications and mortality, alongside a slight increase in type 2 diabetes incidence.
A prior meta-analysis of nine cross-sectional studies on retinopathy revealed an overall prevalence of 3.6% detected by fundus photography, highlighting a significantly higher prevalence in patients with pre-diabetes than in individuals with normoglycemia.36 Our study revealed a retinopathy incidence of 0.2% over a median follow-up period of 8.8 years, which is notably lower than the results reported in the previous meta-analysis. Regarding neuropathy, previous research has indicated that 11–25% of individuals with pre-diabetes might exhibit peripheral neuropathy and that 13–21% could experience neuropathic pain.37 Our analysis revealed a 1.6% neuropathy incidence rate. However, our figures for both retinopathy and neuropathy may be underestimated, as they were based on medical records documented by physicians in cases where related symptoms and diseases were established. Consequently, standard diagnostic procedures such as fundus examinations and nerve conduction studies may not have been performed in all cases.
Elderly individuals with mild dysglycemia (cluster 2) presented an increased risk of CeVD. However, elderly patients in cluster 6, which is marked by progressively worsening glucose abnormalities, greater proportion of baseline HT and statin use, experienced significantly poorer cardiovascular outcomes and higher overall mortality over the follow-up period. Previous studies have elucidated the relationship between chronic hyperglycemia and the onset of vascular complications, including both macrovascular and microvascular damage, in individuals with pre-diabetes.2 24 38–40 Notably, microvascular complications have been reported to occur before conventional diagnostic criteria for diabetes are met.41
A recent meta-analysis examining the associations between pre-diabetes and all-cause mortality, CVD and stroke over a median follow-up period of 9.8 years revealed that various glycemic statuses and definitions influenced the risk of vascular outcomes.6 42 While our analysis focused primarily on high-risk clusters and adverse CVD outcomes, we also identified the lowest risk cluster: cluster 4 (mild dysglycemia obese). This cluster presented low risks for CKD, composite microvascular outcomes and total mortality. Several mechanisms might underlie these cardiovascular findings, with age and the severity of glycemia playing pivotal roles in exacerbating cardiovascular outcomes in individuals with pre-diabetes.
The study employed cluster analysis to identify distinct common risk profiles in individuals without diabetes and demonstrated important similarities to profiles reported previously. For example, the phenotypes defined by Wagner et al27 showed comparable characteristics. We identified six characteristic clusters, each with varying levels of risk for developing type 2 diabetes, vascular complications and mortality. The clinical characteristics and features of these clusters in individuals without diabetes align with previous reports.27
Prior studies investigating the progression from pre-diabetes to type 2 diabetes have found that individuals with pre-diabetes have varying risks of developing type 2 diabetes, depending on their glycemic phenotypes within the pre-diabetes range.24 43–45 Our analysis revealed differing type 2 diabetes incidences among the clusters. Each cluster represents a population with distinct clinical characteristics and glycemic parameter levels. Our findings suggest that clinical characteristics other than glycemic phenotypes may influence the incidence of type 2 diabetes.
The participants in clusters 3 and 5 presented the highest incidence rates and aHRs for diabetes, independent of sex, statin use and current HT. This finding aligns with previous clinical studies that reported that a similar cluster phenotype—characterized by high BMI, high cardiometabolic risk and high liver fat—was associated with an increased risk for type 2 diabetes.27 46
A previous 9-year prospective Korean cohort study explored the associations between pre-diabetes and the development of CKD.47 It found that individuals with pre-diabetes defined by impaired glucose tolerance and/or HbA1c levels—but not by impaired fasting glucose alone—had a significantly greater risk of developing CKD after considering traditional CKD risk factors. Our study’s overall CKD incidence (9.4%) was slightly lower than that of the Korean cohort (11.3%). However, this observation is nuanced by the presence of cluster 6 in our study, which included a high proportion of patients with CKD.
Cluster 4 includes obese individuals with minimal glycemic deterioration, corresponding to metabolically healthy obesity. This cluster is associated with lower risks of type 2 diabetes, composite microvascular complications and overall mortality, regardless of sex, current HT status and statin treatment. This association may be explained by a more metabolically favorable fat distribution in cluster 4 individuals.48 49
Several randomized controlled trials have demonstrated that lifestyle and pharmacological interventions improve outcomes for individuals at risk of type 2 diabetes.9 50 51 However, such interventions may not be necessary or cost-effective for all individuals with pre-diabetes. Identifying subphenotypes before they develop type 2 diabetes and complications could prove valuable. These subphenotypes offer the potential to enhance our understanding of pathophysiological heterogeneity, improve screening practices and refine treatment and prevention strategies.46
Our analysis identified three distinct risk profiles. Individuals in clusters 3 and 5, who are at risk of developing type 2 diabetes, could benefit from intensive lifestyle interventions focused on weight management and possibly diabetes-preventive medications. Individuals in cluster 6 are at increased risk of developing CKD and CVD and experiencing higher overall mortality. For this group, preventive measures should emphasize lipid and blood pressure control, albuminuria management and smoking cessation, with glycemic control potentially less prioritized. Individuals in clusters 1, 2 and 4 appear to be at lower risk; annual monitoring and risk factor screening may suffice for them.
These findings suggest that a precision medicine approach, especially in primary care settings for individuals with pre-diabetes, can yield significant clinical benefits and inform the development of more targeted health policies.
Our analysis has both strengths and limitations. Few studies have explored how pre-diabetes phenotypes—characterized by common clinical risk factor clustering—are related to diabetes, vascular complications and all-cause mortality. This is the first such study in Thailand. It drew on a high-risk Thai clinical sample with a median follow-up of 8.8 years, making the study findings particularly pertinent for practicing clinicians. Previous phenotype research by Wagner et al predominantly involved a white cohort and a UK occupational cohort, both enriched with individuals prone to diabetes. The phenotypic outcomes in these cohorts were validated and found to have similar associations with clinical outcomes as those observed in our study. The findings from our investigation and Wagner’s study revealed similar associations between pre-diabetes clusters and clinical outcomes. Our analysis used a representative sample set and included all outcomes related to individuals with pre-diabetes, including type 2 diabetes, vascular complications and all-cause mortality.
However, our study has several limitations that restrict the generalizability of our findings. The primary limitation is the retrospective nature of the cohort, with data retrieved from electronic medical records and other databases. Health records may not capture all relevant risk factors (eg, family history of diabetes in the first-degree relative), and diagnoses might be incompletely or accurately recorded. Outcome diagnostics were sourced from diverse reports and records, with data accuracy contingent on the responsible physicians’ expertise and the equipment and procedures employed. This may result in a limited number of cardiovascular events as outcomes, which could potentially affect the accuracy of the relationship between the outcomes and the phenotypes. Next, a relatively short period may limit the ability to fully capture the long-term development of vascular complications, particularly in the pre-diabetes population, emphasizing the need for future studies with longer follow-up periods and incorporating sex-stratified analyses to validate our findings and further elucidate the long-term implications of pre-diabetes subphenotypes. Moreover, our cohort may not fully represent the broader pre-diabetes population. Any of these factors may have introduced bias into the observed associations. Nonetheless, our analysis documented exposure to all relevant risk factors before type 2 diabetes onset and the related outcomes, thus establishing the temporal sequence between risk factors and outcomes. Diagnoses and risk definitions adhered to stringent standard references and pertinent record parameters, including medication records, ICD-10 codes and laboratory findings. To minimize data retrieval errors, trained research doctors and staff meticulously reviewed and double-checked all diagnoses and risk factors. We also included all-cause mortality as an outcome measure, although the specific causes of death were undetermined.
Another limitation is the dynamic nature of phenotype clustering, with participants potentially shifting between clusters over time owing to changes in age, body composition and laboratory findings. However, this dynamic model reflects the nature of pre-diabetes, where risk factors and prognoses can vary across an individual’s lifespan. Targeted management strategies and screening and preventive programs should adapt to the different life stages of patients with pre-diabetes. Lastly, our analytical model excluded variables such as impaired glucose tolerance, insulin sensitivity, insulin resistance, genetic variations and fat composition measurements. In Thailand, these risk factors are not routinely measured in clinical practice, resulting in limited data on these variables in our dataset. Furthermore, future studies involving more diverse and representative pre-diabetes prospective cohorts with comprehensively evaluation, are necessary to confirm the findings.
Leave a Reply