Analytic Sample. This study employed a pooled MCBS panel data from 2006 to 2012, creating an unbalanced panel sample of unique person-year observations among Medicare FFS beneficiaries with ADRD. I appended all the available panels over the seven-year time span between 2006 and 2012. Due to the survey design of MCBS, each participant in the MCBS could be present in the appended MCBS data for one to four consecutive years. The selection of analytic sample is shown in Figure 2: Flowchart for the Selection of Analytic Sample. The analytic sample was first limited to the population with ADRD using the ICD9-CM codes (based on the CMS Chronic Conditions Data Warehouse) as well as survey questions (“whether the respondent has ever been told by a doctor that they have AD”) to identify patients with ADRD. I further limited the study population to older adults (ages 65 years or older) who have reported available address zip codes, have not changed address, have enrolled in the Medicare FFS plan for 12 months and were not institutionalized at baseline. Since I allowed for a one-year time lag between independent variables and functional outcomes in the following year, the analytic sample was further limited to participants with at least two consecutive years of enrollment in the MCBS. Due to the survey design of MCBS, study members could have two, three, or even four years of data in the analytic sample. The calculation for COC requires that the patient has made at least two evaluation and management (E&M) visits, Medicare FFS claims data was utilized to further exclude those who had less than two E&M visits. Finally, study members with missing confounders were excluded from the analytic sample, and the final analytic sample is 2,009 observations. Figure 2: Flowchart for the Selection of Analytic Sample
