Applying latent class analysis to risk stratification of incident diabetes among Chinese adults. (April 2021)

• Record Type: Journal Article
• Authors: Wu, Yang; Hu, Haofei; Cai, Jinlin; Chen, Runtian; Zuo, Xin; Cheng, Heng; Yan, Dewen
• Is Part Of: Diabetes research and clinical practice. Volume 174(2021)
• Publication Date: 2021-04
• Subjects: Latent class analysis -- Incident diabetes -- Subpopulation; Diabetes -- Periodicals; Diabetes Mellitus -- Periodicals; 616.462
• DOI: 10.1016/j.diabres.2021.108742
• Languages: English
• ISSNs: 0168-8227

Abstract

Highlights: Latent class analysis (LCA) is a well-validated robust probabilistic approach. We used LCA to identify two subpopulations with differential risk of diabetes. 3 and 5-year diabetes risk differ between the two classes by 5.451 and 5.264 times. Our findings help clinicians design clinical trials and make healthcare protocols. Abstract: Objective: To use latent class analysis to identify unobservable subpopulations amongst the heterogeneous population and explore the relationship between subpopulations and incident diabetes among Chinese adults. Methods: The retrospective study included 32, 312 Chinese adults without diabetes at baseline. Latent class indicators included demographic and clinical variables. The outcome was incident diabetes. The relationship between latent class and outcome was evaluated with Cox proportional hazard regression analysis. Results: After screening, the two-class latent class model best fits the population. Participants in class 2 are characterized by higher age, body mass index, systolic and diastolic blood pressure, fasting plasma glucose, total cholesterol, triglyceride, low-density lipoprotein cholesterol, serum creatinine, serum urea nitrogen, alanine aminotransferase, and a higher proportion of males, ever/current smokers and drinkers, but lower high-density lipoprotein cholesterol and a lower proportion of family history of diabetes. The risk of diabetes in class 2 was 5.451 times (HR: 6.451, 95%CI: 4.179–9.960, P < 0.00001) and 5.264 times (HR: 6.264, 95%CI: 4.680–8.385, P < 0.00001) higher than that in class 1 during 3-year and 5-year follow-up, respectively. Conclusions: We used latent class analysis to identify two distinct subpopulations with differential risk of diabetes during 3-year and 5-year follow-up.