Prediction modelling of 1-year outcomes to a personalized lifestyle intervention for Canadians with metabolic syndrome

Abstract

Metabolic syndrome (MetS) comprises a cluster of risk factors that includes central obesity, hypertension, dyslipidemia, and impaired glucose homeostasis. Although lifestyle interventions reduce MetS risk, not everyone responds to the same extent. The primary objective of this study was to identify variables that could predict 1-year changes in MetS risk in individuals participating in the Canadian Health Advanced by Nutrition and Graded Exercise (CHANGE) program. Participants were allocated into training (n = 157) and test (n = 29) datasets by availability of genetic data. A linear mixed-effect model revealed that age, medication, fasting glucose, triglycerides, high-density lipoprotein cholesterol, waist circumference, systolic blood pressure, and fibre intake were associated with continuous MetS (cMetS) score across all time points. Multiple linear regressions were then used to build 2 prediction models using 1-year cMetS score as the outcome variable. Model 1 included only baseline variables and was 38% accurate for predicting cMetS score. Model 2 included both baseline variables and the 3-month change in cMetS score and was 86% accurate. As a secondary objective, we also examined if we could build a model to predict a person’s categorical response bin (i.e., positive responder, nonresponder, or adverse responder) at 1 year using the same variables. We found 72% concordance between predicted and observed outcomes. These various prediction models need to be further tested in independent cohorts but provide a potentially promising new tool to project patient outcomes during lifestyle interventions for MetS. Novelty Short-term changes in cMetS score improve prediction model performance compared with only baseline variables. Predictive models could potentially facilitate clinical decision-making for personalized treatment plans.