Causal effects of time-dependent treatments in older patients with non-small cell lung cancer.

Abstract

BACKGROUND:Treatment selection for elderly patients with lung cancer must balance the benefits of curative/life-prolonging therapy and the risks of increased mortality due to comorbidities. Lung cancer trials generally exclude patients with comorbidities and current treatment guidelines do not specifically consider comorbidities, so treatment decisions are usually made on subjective individual-case basis. METHODS:Impacts of surgery, radiation, and chemotherapy mono-treatment as well as combined chemo/radiation on one-year overall survival (compared to no-treatment) are studied for stage-specific lung cancer in 65+ y.o. patients. Methods of causal inference such as propensity score with inverse probability weighting (IPW) for time-independent and marginal structural model (MSM) for time-dependent treatments are applied to SEER-Medicare data considering the presence of comorbid diseases. RESULTS:122,822 patients with stage I (26.8%), II (4.5%), IIIa (11.5%), IIIb (19.9%), and IV (37.4%) lung cancer were selected. Younger age, smaller tumor size, and fewer baseline comorbidities predict better survival. Impacts of radio- and chemotherapy increased and impact of surgery decreased with more advanced cancer stages. The effects of all therapies became weaker after adjustment for selection bias, however, the changes in the effects were minor likely due to the weak selection bias or incompleteness of the list of predictors that impacted treatment choice. MSM provides more realistic estimates of treatment effects than the IPW approach for time-independent treatment. CONCLUSIONS:Causal inference methods provide substantive results on treatment choice and survival of older lung cancer patients with realistic expectations of potential benefits of specific treatments. Applications of these models to specific subsets of patients can aid in the development of practical guidelines that help optimize lung cancer treatment based on individual patient characteristics.