Bias measurement in, bias results out: how an assumption free height adjusted weight model outperforms body mass index

Objective Body mass index (BMI) is the most commonly used predictor of weight-related comorbidities and outcomes. However, the presumed relationship between height and weight intrinsic to BMI may introduce bias with respect to prediction of clinical outcomes. Using Vanderbilt University Medical Center’s deidentified electronic health records and landmark methodology, we performed a series of analyses comparing the performance of models representing weight and height as separate interacting variables to models using BMI. Methods Model prediction was evaluated with respect to established weight-related cardiometabolic traits, metabolic syndrome and its components hypertension, diabetes mellitus, low high-density lipoprotein, and elevated triglycerides, as well as cardiovascular outcomes, atrial fibrillation, coronary artery disease, heart failure, and peripheral artery disease. Model performance was evaluated using likelihood ratio, R 2 , and Somers’ Dxy rank correlation. Differences in model predictions were visualized using heatmaps. Results Regardless of outcome, the maximally flexible model had a higher likelihood ratio, R 2 , and Somers’ Dxy rank correlation for event-free prediction probability compared to the BMI model. Performance differed based on the outcome and across the height and weight range. Conclusions Compared to BMI, modeling height and weight as independent, interacting variables results in less bias and improved predictive accuracy for all tested traits. Study Importance Questions What is already known about this subject? Body mass index, derived from collected height and weight measures, is an imperfect proxy measure of body fat composition often used in medical research. What are the new findings in your manuscript? We demonstrate how BMI introduces complex non-uniform biases across outcome and height-weight space. How might your results change the direction of research or the focus of clinical practice? Modeling height and weight as separate, non-linear, interacting variables improves clinical prediction across the complete spectrum of heights and weights for all clinical out-comes.