The effect of exercise on outcomes for hospitalised older acute medical patients: an individual patient data meta-analysis

Appendix 1: Statistical Methods for each regression model. Regression modeling techniques with trial covariates were performed for each outcome. Age, gender, prior place of residence, Charlson score, admission BI, TUG and FAC scores were reported for both trials and were assessed as potential covariates for each regression model, as recommended by Tabachnick and Fidell. Each regression model was stratified by “hospital”. A likelihood ratio test was performed to assess interaction between treatment effect and hospital. Potential sources of heterogeneity in patient outcomes were investigated if significant differences were detected between hospital sites.