970-P: Prediction of Cardiovascular Disease Risk among Patients with Type 2 Diabetes Using Repeated Measurements of Blood Pressure, Cholesterol, and Hemoglobin A1c

Abstract

Background: Cardiovascular disease (CVD) risk prediction models among people with type 2 diabetes mellitus (T2DM) are mostly based on single measures of risk predictors. Few have considered incorporating repeated risk predictor measures. We aimed to investigate whether using repeated measurements of risk predictors improves CVD risk prediction in T2DM patients. Methods: We analyzed primary care data (2004-2017) from the UK Clinical Practice Research Datalink for 83,910 individuals aged 40-85 years, with confirmed T2DM but without CVD before study entry. In a 2/3 derivation sample, we derived age- and sex-specific prediction models with predictors including systolic blood pressure (SBP), total and HDL cholesterol, hemoglobin A1c (HbA1c), duration of T2DM, hypertension treatment, smoking, atrial fibrillation, and ethnicity. In stage 1, repeated measures of SBP, cholesterol, and HbA1c were summarized using: a) last observation carried forward; b) cumulative mean; c) longitudinal models with random intercept and slope. In stage 2, we fitted Cox regression models to predict 10-year CVD risk using summarized predictor values from stage 1 and last observed values for categorical predictors. Model performance was evaluated in a 1/3 validation sample. Results: Compared with the model using last observation carried forward values, the cumulative means and the longitudinal model led to no substantial improvement in risk discrimination with changes in C-index by -0.0016 (95% CI: -0.0030, -0.0003) and 0.0011 (95% CI: -0.0001, 0.0023), respectively; the overall net reclassification was improved by 0.035 (95% CI: 0.009, 0.061) for the cumulative means model and 0.062 (95% CI: 0.033, 0.090) for the longitudinal model. Conclusion: Incorporating repeated measurements of blood pressure, cholesterol, and HbA1c into CVD risk prediction models for T2DM patients slightly improves risk reclassification but not discrimination. Disclosure Z. Xu: None. A. M. Wood: None. Funding Alan Turing Institute/British Heart Foundation; UK National Institute for Health Research Cambridge Biomedical Research Centre