The use of prescription and non-prescription drugs for prevention of chronic diseases is a central feature of the care of older adults. Randomized controlled trials assessing the effectiveness of pharmacoprevention of chronic diseases in older adults 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 changes in therapies that occur over the course of the trial. Nonexperimental studies on beneficial and harmful effects of drugs, including studies based on administrative data that often are the only available data, however, have frequently been criticized as producing biased results. It is therefore vital to develop and apply adequate methods to reduce bias in observational studies that assess the preventive effects of medication use in older patients. Propensity scores (PSs) are an increasingly used technique to address confounding in nonexperimental research. PSs are often perceived as superior to conventional analyses in nonexperimental studies assessing the effectiveness of drugs and medical procedures. Funded by RO1 AG023178, we have assessed limitations and advantages of propensity scores (PS) in real datasets and extensive simulations and have developed novel analytic techniques based on PS since 2005. We have disseminated our results by means of oral presentations (28), posters (16), and workshops/symposia (4) at major international epidemiologic, pharmacoepidemiologic, biostatistics, and statistics meetings and in a series of 27 publications, including 9 in the highet ranked epidemiologic journal (the American Journal of Epidemiology). We have focused our research on one of the major problems of nonexperimental research of beneficial and harmful effects of drug treatments, i.e., the problem of unmeasured confounding. We previously had developed an innovative method to include additional information on confounders not measured in the main study from an external validation study combining PS and regression calibration (Propensity Score Calibration). We recently published another innovative method to deal with confounding bias by unmeasured confounders. This method is based on the assumption that patients treated contrary to prediction are more likely to have unmeasured confounders, like frailty, leading the physician to override the predicted treatment. This is the first paper ever showing a clear advantage of PS methods over conventional multivariable outcome models with respect to unmeasured confounders. The second competing continuation will build on our work over the last 7 years and extend it in the same domain - limitations and value of PSs to assess the preventive effects of medication use in older patients with a slightly changed team of researchers while keeping the same core investigators. Using new data sources, including a large random national sample of 3 million Medicare beneficiaries each for the years 2007, 20008, and 2009, we will focus on several unresolved new and distinct topics regarding the implementation of PSs and the effects of preventive drug use in the elderly.
Drugs are a mainstay of today's healthcare. Universal access to affordable drugs with the optimal benefit to harm balance for a given individual is a major global, national, and local public and individual health goal. Despite pre-marketing proof of efficacy in randomized trials, there is little known about harms at the time of drug approval (e.g. Vioxx) and data on benefits are usually lacking for the majority of patients who will ultimately be treated, e.g., older adults with co-morbidities and co-medications. Nonexperimental post-approval studies can fill this knowledge gap on drug benefits and harms needed to make optimal treatment decisions but need to be designed carefully to be valid. Funded by NIA we have successfully developed and evaluated advanced and novel methods to increase the validity of nonexperimental post- approval studies over the last 7 years and propose to continue to do so mainly based on data on prescription drug use and diagnoses from a representative sample of approximately 3 million Medicare beneficiaries per year for the years 2007 to 2009.
Showing the most recent 10 out of 133 publications
