Predictive models of adverse outcomes of overweight or obesity: a nationwide cohort validation

Obesity and overweight have a wide range of adverse clinical outcomes. Predictive models could help clinicians identify individuals with obesity and overweight at greater risk for obesity-related complications, enabling personalized shared decision-making about obesity treatment. We previously developed a set of predictive models of adverse clinical outcomes of obesity and overweight using electronic health record (EHR) data of a large health system in Massachusetts. However, these models have not been validated on data from other sources.

We validated 12 predictive models for cardiometabolic, oncologic and musculoskeletal complications of excess weight using a linked EHR and claims dataset of a national patient cohort (Optum^® Market Clarity) that included adults 18-80 years of age with body mass index (BMI) ≥ 25 kg/m² between 2007 and 2023.

A 5% randomly selected validation dataset included 437,223 patients (51.9% women) with a mean age of 51.4 years and a mean BMI of 31.9 kg/m² who were followed for a mean of 5.7 years. Most (11/12) models achieved Harrell C-index ≥ 0.70. Most (11/12) models had either 5-year or 10-year calibration slope within 0.3 of the ideal value of 1.0; 7 out of 12 had either 5-year or 10-year calibration slope that was not significantly different from 1.0; all had intercepts very close to 0.

A set of EHR-based predictive models of adverse outcomes of obesity and overweight was generalizable to a large national cohort. These findings suggest potential clinical and population health applications for predictive models in treatment of individual patients with obesity / overweight, as well as in population management and quality improvement.

Not applicable.