Fig. 1
Diagram of the workflow: using electronic health record data to assess the effects of yoga on blood pressure in the real world. The analysis plan is illustrated in three parts, preprocessing, stratification, and modeling. 1Preprocessing steps included: within patient imputation of covariates (covariates selected to represent biological, behavioral, environmental, and social factors that may affect blood pressure or use of yoga) and the application of inclusion criteria (3 years of medical history, encounters with primary care providers, age 18–79 years) and exclusion criteria (missing BMI, pregnancy or end-stage renal disease, blood pressure and weight thresholds), see Methods. 2CEM stratifies the data set based on categorical or coarsened values of the covariates, then removes any strata that do not contain at least one case and one control. This ensures there is at least one near match for each observation and each observation is weighted based on the number of cases and controls within its stratum. The symbols in the diagram are colored to indicate different weights. 3Mixed effects linear and logistic regression models were fit to the balanced data set. Models included all covariates and CEM derived weights