A Strategy for Imputing Missing Body Mass Index Values in a Large, Longitudinal DatabaseJames Bowen, Christiana Care Health System
Colin Kern, University of Delaware
*Paul Kolm, Christiana Care Health System
Keywords: Body Mass Index, missing data, imputation
Body Mass Index (BMI) is a recognized risk factor for a number of health problems including diabetes and cardiovascular disease as well as mortality, and is commonly included in databases of clinical trials, observational studies and patient registries. However, when BMI is obtained from medical records, electronic or paper charts, information to calculate BMI is often missing. For example, one or both of height and weight needed to calculate BMI is not recorded – most often height. In addition, physician reference to BMI issues in patient charts may include only a categorical description (e.g., “overweight”) without a specific BMI value. Although there have been very sophisticated developments in the last 20+ years of handling missing data, BMI poses a problem for many imputation of missing data methodologies because of the large degree of individual variability in height and weight. We present a strategy for imputing missing BMI values in a study that examines the relationship between BMI, chronic kidney disease and cardiovascular events involving more than 36,000 patients and half a million patient observations over a 10 year period.