Defense Date


Document Type


Degree Name

Doctor of Philosophy



First Advisor

Juan Lu


Background: Inverse-probability-of-treatment-weighted estimation (IPTW) of marginal structural models was proposed to adjust for time-varying confounders that are influenced by prior treatment use. It is unknown whether pharmacoepidemiologic studies that applied IPTW conformed to the recommendations proposed by methodological studies. In addition, no previous study has compared the performance of different analytic strategies adopted in IPTW analyses. Objectives: This project aims 1) to review the reporting practice of pharmacoepidemiologic studies that applied IPTW, 2) to compare the validity and precision of several approaches to constructing weight, 3) to use IPTW to estimate the effectiveness of glucosamine and chondroitin in treating osteoarthritis. Methods: We systematically retrieved pharmacoepidemiologic studies that were published in 2012 and applied IPTW to estimate the effect of a time-varying treatment. Under a variety of simulated scenarios, we assessed the performance of four analytic approaches what were commonly used in studies conducting IPTW analyses. Finally, using data from Osteoarthritis Initiative, we applied IPTW to estimate the long-term effectiveness of glucosamine and chondroitin on treating knee osteoarthritis. Results: The practice of reporting use of IPTW in pharmacoepidemiologic studies was suboptimal. The majority of reviewed studies did not report that the positivity assumption was assessed, and several studies used unstablized weights or did not report that the stabilized weights were used. With data simulation, we found that intention-to-treat analyses underestimated the actual treatment effect when there was non-null treatment effect and treatment non-adherence. This underestimation was linearly correlated with adherence levels. As-treated analyses that took into account the complex mechanism of treatment use generated approximately unbiased estimates without sacrificing the estimate precision when the treatment effect was non-null. Finally, after adjustment for potential confounders with marginal structural models, we found no clinically meaningful benefits of glucosamine/chondroitin in relieving knee pain, stiffness and physical function or slowing joint space narrowing. Conclusions: It may be prudent to develop best practices of reporting the use of IPTW. Studies performing intention-to-treat analyses should report the levels of adherence after treatment initiation, and studies performing as-treated analyses should take into the complex mechanism of treatment use in weight construction.


© The Author

Is Part Of

VCU University Archives

Is Part Of

VCU Theses and Dissertations

Date of Submission


Included in

Diseases Commons