The efficacy of alternative treatment strategies for lung cancer patients aged 65 years and older can be evaluated using observational heath care data such as SEER-Medicare. Statistical techniques for observational data such as marginal structural model (MSM) allow elucidating the causality between time- dependent treatments and patients'survival. In this project, we apply MSM to SEER-Medicare data to study the causal effects of treatment (such as surgery, radiation, and/or chemotherapy) on survival of lung cancer patients given individual information on tumor characteristics, comorbidities, and demographic and socio- economic factors. This model considers both time-dependent treatment and comorbidity--two important survival factors that are commonly interrelated. However, MSM is still based on strong assumptions such as the absence of unobserved covariates related to treatment assignment and survival, and also does not involve disease-specific clinical information. Therefore, we further develop the generalized approach based on recent stochastic process model advances which can address MSM limitations to more precisely evaluate studied effects. The results of this study will provide i) further insight into the theory of causal inferece of time- dependent treatment in the presence of other time dependent covariates such as comorbiditiy, ii) a new model for effects analysis with SAS-based software that addresses MSM limitations, and iii) substantive results of the treatment effects on lung cancer survival by stage and different comorbidity states. Successful completion of this project will assist clinicians in choosing the optimal lung cancer treatment that balances benefits and risks based on individual patient characteristics.
The proposed study addresses the problem of estimating treatment effect and optimizing care for elderly lung cancer patients with different comorbidities. This study uses a large national-level Medicare-based administrative dataset, making the results generalizable to the US elderly population. Successfully completing this study will give clinicians a new computational tool that helps guide the choice of lung cancer treatment to optimally balance treatment benefits and risks for a patient based on their individual characteristics.