Prescription drugs play a crucial role in the prevention and treatment of chronic disease in older adults. The evidence base for the benefit and harm of these treatments comes from experimental (randomized controlled trials, RCTs) and nonexperimental (observational, epidemiologic) studies. For older adults, and especially those with multiple comorbidities, critically important evidence comes from nonexperimental studies, because these individuals ? particularly those who use the most drugs ? are often excluded from RCTs. Unfortunately, nonexperimental studies often suffer from confounding by frailty, as recently evinced by two RCTs on statins that failed to confirm nonexperimental findings of protective effects in sepsis-associated acute respiratory distress syndrome and in chronic obstructive pulmonary disease. Confounding by frailty can lead to suboptimal or even harmful treatment decisions (if nonexperimental studies are the only evidence available) or the conduct of costly RCTs that fail to replicate the findings of nonexperimental research. For a timely assessment of drug benefit and harm in older adults and real-world settings, it is therefore vital to develop and apply improved methods to reduce confounding in nonexperimental studies. Funded by R01 (and now R56) AG023178 since 2005, we have achieved substantial advances in knowledge about methods to improve the validity of nonexperimental research. Using both empirical data and extensive simulations, we have developed several novel analytic techniques to reduce confounding, including propensity score calibration and the exclusion of patients treated contrary to prediction. Over the last 11 years, we have disseminated our results by means of oral presentations, posters, and workshops/symposia and in a series of 67 publications (i.e., > 6 per year), including 15 in the top epidemiologic journals (AJE and Epidemiology), 14 in the top pharmacoepidemiologic journal (PDS), 7 in Medical Care, and several in top medical journals (JAMA, JNCI, Diabetes Care, and JAGS). We propose to continue building effective tools to address the most significant problem hindering nonexperimental study designs from generating valid answers about beneficial and harmful effects of treatments, i.e., the problem of unmeasured confounding. The proposed study will build on our work over the last decade and extend it in the same domain ? through increasing the validity of nonexperimental methods to assess the preventive effects of treatments in older adults. We will focus on better defining study populations at equipoise between treatments compared, as well as six additional aims, using empirical motivating examples and a variety of simulations to improve the validity of nonexperimental Comparative Effectiveness Research (CER). Improved validity of nonexperimental CER will result in more robust evidence about drug benefit and harm in older adults. This work will directly inform clinically relevant treatment decisions, provide timely evidence unavailable from RCTs, and ultimately improve individual and overall health of older adults.
At the time of drug approval there is little known about potential harm of drugs used to prevent and treat chronic conditions (e.g., Vioxx, Avandia), and data on benefit are lacking for the majority of patients who will ultimately be treated, e.g., older adults with multiple conditions and medications. Nonexperimental post- approval studies can fill this knowledge gap to support optimal treatment decisions but must be designed and analyzed appropriately to minimize the potential for biased findings and erroneous conclusions. We propose to develop and evaluate novel methods to increase the validity of nonexperimental post-approval studies based on a large database of Medicare beneficiaries currently available at UNC for the years 2007 through 2014.
