The use of prescription and non-prescription drugs for prevention of chronic diseases is a central feature of the care of the elderly. Growing attention has been paid recently to the evidence of efficacy of several drugs in this context, including non-steroidal anti-inflammatory drugs, statins, and hormone therapy in postmeno- pausal women. Randomized controlled trials assessing the effectiveness of pharmacoprevention of chronic diseases in the elderly are very costly, take a long time to design and implement, and may be of limited generalizability due to selection of participants, shorter duration of treatment, and competing outcomes. However, observational studies of these questions, including ones based on administrative data that often are the only ones available, have frequently been criticized as producing biased results. It is therefore vital to develop and apply innovative techniques to improve causal inference from non-experimental research assessing the preventive effects of medication use in older patients. Funded by RO1 AG023178, we previously assessed limitations and advantages of propensity scores (PSs) in real datasets and extensive simulations. We disseminated our results by means of oral presentations (12), posters (9), and symposia (2) at the highest ranked international epidemiologic and pharmacoepidemiologic meetings and in a series of 9 publications, including 4 in the highest ranked epidemiologic journal. The competing continuation will allow us to continue and expand our previous work on the limitations and value of propensity scores to assess the preventive effects of medication use in older patients. The team of researchers will continue to focus on several unresolved topics regarding the use of validation studies to adjust for confounding unmeasured in the main study and PSs. We propose to 1. compare methods for missing data and measurement error correction to adjust for confounding not measured in a main study using validation study data, 2. apply these methods to address unmeasured confounding in the prolongation of life with cholinesterase inhibitors, 3. assess the value of excluding patients with very low and very high PSs from the analysis, 4. assess the applicability and use of propensity scores in the setting of non-dichotomous exposures, and 5. assess the value empirical Bayes correction of the PS compared with variable selection. Dissemination of our results will increase correct application of PS methods including methods to adjust for unmeasured confounding.
Showing the most recent 10 out of 133 publications
