Risk prediction of complications and mortality in Chinese patients with type 2 diabetes mellitus

Abstract

Given the increasing prevalence and substantial heterogeneity of patients with type 2 diabetes mellitus (T2DM), to facilitate more cost-effective healthcare delivery and service planning, it is imperative to develop a risk prediction model to identify patients at high risk of diabetic complications such as cardiovascular disease (CVD) and end-stage renal disease (ESRD). However, available risk prediction models were primarily developed from Western general populations and thus may not be fully transferable to the Chinese T2DM population. Therefore, in this thesis I aimed to (1) develop and validate risk prediction models for CVD, ESRD, and all-cause mortality and (2) develop a classification rule for CVD risk among Chinese primary care T2DM patients. I conducted a population-based retrospective cohort study of Chinese adult T2DM patients who had received public primary care services for their DM during 2010. The Hong Kong Hospital Authority extracted the data from their information system of 149,333 T2DM patients who satisfied the inclusion criteria. I selected the relevant patient demographics, medical history, treatment modalities, clinical parameters, diagnosis, and death data. The cohort was randomly split on a 2:1 basis into derivation and validation cohorts, used respectively for deriving and validating the risk prediction models. Cox regression analysis with a forward stepwise approach on routinely available predictors was used to derive the models, with stratification by sex. Common predictors for CVD, ESRD, and all-cause mortality included age, smoking status, use of anti-hypertensive drugs or insulin, glycosylated hemoglobin (HbA1c), systolic blood pressure, urine albumin-to-creatinine ratio (ACR), and estimated glomerular filtration rate (eGFR). Specifically, the model for CVD was the first to include both urine ACR and eGFR. Smoking was an additional predictor in men, while predictors specific to women included T2DM duration, oral anti-diabetic drug use, body mass index, and diastolic blood pressure. The curvilinear association between outcome events and predictors, and also the interaction factors with age, were first incorporated in the risk prediction models. Based on the validation cohort, my derived models outperformed existing models in terms of prediction accuracy. I also translated these new risk prediction models into a web calculator as well as a color-coded chart to facilitate their use in primary care. Moreover, I conducted a classification tree analysis to develop a first risk classification rule for CVD to categorize patients into five 5-year CVD risk groups (<5%; 5−9%; 10−14%; 15−19% and ≥ 20%). Using validation cohort, the classification rule had better agreement between observed and predicted risk strata than those derived from other existing models that use Cox regression analysis. In conclusion, my study has developed new risk prediction and classification models for Chinese primary care patients with T2DM that will facilitate accurate complication risk estimation and stratification, which can enhance empowerment and motivation, enable better prioritization of health-care resources, and promote more cost-effective interventions by policy makers, clinicians, patients, and researchers.