We used propensity score matching to create a matched control group such that the control andstudy groups were well matched on the basis of all measured baseline characteristics such as age,gender, body mass index (BMI), and each item on the Kihon checklist. We estimated the scores of theparticipation group for each subject by using a multivariable logistic regression model. We were ableto match 1620 pairs of intervention and control subjects who had similar propensity scores.