Analysis of switching insurance plan type: Comparison of two statistical methods

PURPOSE: To compare results of 2 statistical methods for identifying factors in claims data that are associated with switching insurance plans between managed care (MC) and indemnity (IN). METHODS: Using claims data from 2 insurance providers in a northeastern city, we analyzed patients aged 18+ with diabetes, asthma, or congestive heart failure (CHF) who were covered any time in 1993–1997 (N = 88,917). Stratifying by initial plan type, we examined predictors of switching from the initial plan type using logistic regression and survival analysis. Covariates included age, time in study (for logistic models), gender, diabetes (yes/no), CHF (yes/no), and asthma (yes/no). Survival analysis accounted for time to switch and allowed time-varying covariates. RESULTS: In logistic regression models, older individuals who were in IN were much less likely to switch into MC. Those in MC were more likely to switch to IN, with the greatest likelihood of switching in ages 60-69 (OR = 4.00, 95% CI = 3.32–4.83). Females were less likely to switch from IN to MC (OR = 0.92, 95% CI = 0.87–0.98), CHF patients were less likely to switch from IN to MC (OR = 0.75, 95% CI = 0.68–0.83), and diabetes patients were less likely to switch from MC to IN (OR = 0.77, 95% CI = 0.62–0.96). Hazard ratios calculated using Cox regression were similar to odds ratios for most covariates. However, some coefficients for diseases were significant in Cox models but not in the logistic models. Cox models took 45 times longer in CPU time than logistic regression models. CONCLUSIONS: Logistic regression was a good approximation to Cox regression in identifying many of the factors in switching insurance plan in these data, at a fraction of the computing time. However, Cox models allowed diseases to be time-varying, and so was more sensitive to identifying significant relationships with disease.