Statistical methods for cost-effectiveness analysis using observational data
Observational studies are a useful resource for evaluating the cost and cost-effectiveness of medical treatments, but the results are subject to bias from measured and unmeasured confounding. Investigators must adjust for measured confounders using an appropriate model, and it is also advisable to assess the sensitivity of the results to potential unmeasured confounders. ^ In this dissertation, we develop a sensitivity analysis procedure for the treatment effect on cost. We show that, in some cases, closed-form relationships exist between the observed treatment effect and the treatment effect after adjustment for hypothetical confounders. We derive a general adjustment formula for log-linear cost models. We evaluate our method using simulations, and demonstrate the sensitivity analysis procedure by comparing two bladder cancer treatments using a cohort derived from SEER-Medicare. ^ Next we discuss the challenges of correctly modeling cost-effectiveness, including skewed outcomes, censoring, and correlations between costs and effects. We describe several methods for estimating the Net Monetary Benefit (NMB): linear regression, generalized linear models, parametric and semi-parametric survival methods, and non-parametric estimates with propensity score stratification. Using simulations, we compare the performance of the models for analysis of skewed and censored cost and survival data. We find that correctly specified non-linear parametric models provide the best estimates. Linear regression is insufficient for censored data, and semi-parametric and non-parametric approaches have improved bias and coverage over incorrectly specified parametric models. We illustrate the sensitivity of estimated NMB to model choice with a comparison of two prostate cancer treatments. ^ Finally, we propose a sensitivity analysis procedure for the NMB using a Gamma GLM for cost and a Weibull model for survival. We derive closed-form relationships between the expected values of cost and survival obtained from observed data, and the expected values which account for an unmeasured confounder. Our general formulas allow for any unmeasured confounder which can be characterized using a moment-generating function, and also allow for separate unmeasured confounders to influence cost and survival. We evaluate our formulas using simulations, and return to the bladder cancer example to demonstrate a sensitivity analysis for NMB.^
Elizabeth A Handorf,
"Statistical methods for cost-effectiveness analysis using observational data"
(January 1, 2012).
Dissertations available from ProQuest.