Jeremy A. Rassen

High-dimensional propensity score adjustment in studies of treatment effects using health care claims data

Background: Adjusting for large numbers of covariates ascertained from patients’ health care claims data may improve control of confounding, as these variables may collectively be proxies for unobserved factors. Here, we develop and test an algorithm that empirically identifies candidate covariates, prioritizes covariates, and integrates them into a propensity-score-based confounder adjustment model. Methods: We developed a multistep algorithm to implement high-dimensional proxy adjustment in claims data. Steps include (1) identifying data dimensions, eg, diagnoses, procedures, and medications; (2) empirically identifying candidate covariates; (3) assessing recurrence of codes; (4) prioritizing covariates; (5) selecting covariates for adjustment; (6) estimating the exposure propensity score; and (7) estimating an outcome model. This algorithm was tested in Medicare claims data, including a study on the effect of Cox-2 inhibitors on reduced gastric toxicity compared with nonselective nonsteroidal anti-inflammatory drugs (NSAIDs). Results: In a population of 49,653 new users of Cox-2 inhibitors or nonselective NSAIDs, a crude relative risk (RR) for upper GI toxicity (RR = 1.09 [95% confidence interval = 0.91–1.30]) was initially observed. Adjusting for 15 predefined covariates resulted in a possible gastroprotective effect (0.94 [0.78–1.12]). A gastroprotective effect became stronger when adjusting for an additional 500 algorithm-derived covariates (0.88 [0.73–1.06]). Results of a study on the effect of statin on reduced mortality were similar. Using the algorithm adjustment confirmed a null finding between influenza vaccination and hip fracture (1.02 [0.85–1.21]). Conclusions: In typical pharmacoepidemiologic studies, the proposed high-dimensional propensity score resulted in improved effect estimates compared with adjustment limited to predefined covariates, when benchmarked against results expected from randomized trials.

The Comparative Safety of Analgesics in Older Adults With Arthritis

BACKGROUND The safety of alternative analgesics is unclear. We examined the comparative safety of nonselective NSAIDs (nsNSAIDs), selective cyclooxygenase 2 inhibitors (coxibs), and opioids. METHODS Medicare beneficiaries from Pennsylvania and New Jersey who initiated therapy with an nsNSAID, a coxib, or an opioid from January 1, 1999, through December 31, 2005, were matched on propensity scores. We studied the risk of adverse events related to analgesics using incidence rates and adjusted hazard ratios (HRs) from Cox proportional hazards regression. RESULTS The mean age of participants was 80.0 years, and almost 85% were female. After propensity score matching, the 3 analgesic cohorts were well balanced on baseline covariates. Compared with nsNSAIDs, coxibs (HR, 1.28; 95% confidence interval [CI], 1.01-1.62) and opioids (1.77; 1.39-2.24) exhibited elevated relative risk for cardiovascular events. Gastrointestinal tract bleeding risk was reduced for coxib users (HR, 0.60; 95% CI, 0.35-1.00) but was similar for opioid users. Use of coxibs and nsNSAIDs resulted in a similar risk for fracture; however, fracture risk was elevated with opioid use (HR, 4.47; 95% CI, 3.12-6.41). Use of opioids (HR, 1.68; 95% CI, 1.37-2.07) but not coxibs was associated with an increased risk for safety events requiring hospitalization compared with use of nsNSAIDs. In addition, use of opioids (HR, 1.87; 95 CI, 1.39-2.53) but not coxibs raised the risk of all-cause mortality compared with use of nsNSAIDs. CONCLUSIONS The comparative safety of analgesics varies depending on the safety event studied. Opioid use exhibits an increased relative risk of many safety events compared with nsNSAIDs.

Cardiovascular Outcomes and Mortality in Patients Using Clopidogrel With Proton Pump Inhibitors After Percutaneous Coronary Intervention or Acute Coronary Syndrome

Background— Recent studies have raised concerns about the reduced efficacy of clopidogrel when used concurrently with proton pump inhibitors (PPIs), but those studies may have overestimated the risk. Methods and Results— We studied the potential for increased risk of adverse cardiovascular events among users of clopidogrel with versus without concurrent use of PPIs in 3 large cohorts of patients ≥65 years of age, treated between 2001 and 2005. All patients had undergone percutaneous coronary intervention or had been hospitalized for acute coronary syndrome in Pennsylvania, New Jersey, or British Columbia, and subsequently had initiated treatment with clopidogrel. We recorded myocardial infarction hospitalization, death, and revascularization among PPI users and nonusers. We assessed our primary end point of myocardial infarction hospitalization or death using cohort-specific and pooled regression analyses. We entered 18 565 clopidogrel users into our analysis. On a pooled basis, 2.6% of those who also initiated a PPI versus 2.1% of PPI nonusers had a myocardial infarction hospitalization; 1.5% versus 0.9% died; and 3.4% versus 3.1% underwent revascularization. The propensity score–adjusted rate ratio for the primary end point of myocardial infarction or death was 1.22 (95% confidence interval, 0.99 to 1.51); for death, 1.20 (95% confidence interval, 0.84 to 1.70); and for revascularization, 0.97 (95% confidence interval, 0.79 to 1.21). Matched analyses generally yielded similar results. Conclusions— Although point estimates indicated a slightly increased risk of myocardial infarction hospitalization or death in older patients initiating both clopidogrel and a PPI, we did not observe conclusive evidence of a clopidogrel-PPI interaction of major clinical relevance. Our data suggest that if this effect exists, it is unlikely to exceed a 20% risk increase.

Confounding control in healthcare database research: challenges and potential approaches

Epidemiologic studies are increasingly used to investigate the safety and effectiveness of medical products and interventions. Appropriate adjustment for confounding in such studies is challenging because exposure is determined by a complex interaction of patient, physician, and healthcare system factors. The challenges of confounding control are particularly acute in studies using healthcare utilization databases where information on many potential confounding factors is lacking and the meaning of variables is often unclear. We discuss advantages and disadvantages of different approaches to confounder control in healthcare databases. In settings where considerable uncertainty surrounds the data or the causal mechanisms underlying the treatment assignment and outcome process, we suggest that researchers report a panel of results under various specifications of statistical models. Such reporting allows the reader to assess the sensitivity of the results to model assumptions that are often not supported by strong subject-matter knowledge.

Instrumental variable methods in comparative safety and effectiveness research

Instrumental variable (IV) methods have been proposed as a potential approach to the common problem of uncontrolled confounding in comparative studies of medical interventions, but IV methods are unfamiliar to many researchers. The goal of this article is to provide a non‐technical, practical introduction to IV methods for comparative safety and effectiveness research. We outline the principles and basic assumptions necessary for valid IV estimation, discuss how to interpret the results of an IV study, provide a review of instruments that have been used in comparative effectiveness research, and suggest some minimal reporting standards for an IV analysis. Finally, we offer our perspective of the role of IV estimation vis‐à‐vis more traditional approaches based on statistical modeling of the exposure or outcome. We anticipate that IV methods will be often underpowered for drug safety studies of very rare outcomes, but may be potentially useful in studies of intended effects where uncontrolled confounding may be substantial. Copyright

The Comparative Safety of Opioids for Nonmalignant Pain in Older Adults

BACKGROUND Severe nonmalignant pain affects a large proportion of adults. Optimal treatment is not clear, and opioids are an important option for analgesia. However, there is relatively little information about the comparative safety of opioids. Therefore, we sought to compare the safety of opioids commonly used for nonmalignant pain. METHODS We devised a propensity-matched cohort analysis that used health care utilization data collected from January 1, 1996, through December 31, 2005. Study participants were Medicare beneficiaries from 2 US states who were new initiators of opioid therapy for nonmalignant pain, including codeine phosphate, hydrocodone bitartrate, oxycodone hydrochloride, propoxyphene hydrochloride, and tramadol hydrochloride; none had a cancer diagnosis, and none were using hospice or nursing home care. Our main outcome measures were incidence rates and rate ratios (RRs) with 95% confidence intervals (CIs) for cardiovascular events, fractures, gastrointestinal events, and several composite end points. RESULTS We matched 6275 subjects in each of the 5 opioid groups. The groups were well matched on baseline characteristics. The risk of cardiovascular events was similar across opioid groups 30 days after the start of opioid therapy, but it was elevated for codeine (RR, 1.62; 95% CI, 1.27-2.06) after 180 days. Compared with hydrocodone, after 30 days of opioid exposure the risk of fracture was significantly reduced for tramadol (RR, 0.21; 95% CI, 0.16-0.28) and propoxyphene (0.54; 0.44-0.66) users. The risk of gastrointestinal safety events did not differ across opioid groups. All-cause mortality was elevated after 30 days for oxycodone (RR, 2.43; 95% CI, 1.47-4.00) and codeine (2.05; 1.22-3.45) users compared with hydrocodone users. CONCLUSIONS The rates of safety events among older adults using opioids for nonmalignant pain vary significantly by agent. Causal inference requires experimental designs, but these results should prompt caution and further study.

One-to-many propensity score matching in cohort studies.

Among the large number of cohort studies that employ propensity score matching, most match patients 1:1. Increasing the matching ratio is thought to improve precision but may come with a trade‐off with respect to bias.

Assessing the comparative effectiveness of newly marketed medications: methodological challenges and implications for drug development.

Comparative‐effectiveness research (CER) aims to produce actionable evidence regarding the effectiveness and safety of medical products and interventions as they are used outside of controlled research settings. Although CER evidence regarding medications is particularly needed shortly after market approval, key methodological challenges include (i) potential bias due to channeling of patients to the newly marketed medication because of various patient‐, physician‐, and system‐related factors; (ii) rapid changes in the characteristics of the user population during the early phase of marketing; and (iii) lack of timely data and the often small number of users in the first few months of marketing. We propose a mix of approaches to generate comparative‐effectiveness data in the early marketing period, including sequential cohort monitoring with secondary health‐care data and propensity score (PS) balancing, as well as extended follow‐up of phase III and phase IV trials, indirect comparisons of placebo‐controlled trials, and modeling and simulation of virtual trials.

Effects of Adjusting for Instrumental Variables on Bias and Precision of Effect Estimates

Recent theoretical studies have shown that conditioning on an instrumental variable (IV), a variable that is associated with exposure but not associated with outcome except through exposure, can increase both bias and variance of exposure effect estimates. Although these findings have obvious implications in cases of known IVs, their meaning remains unclear in the more common scenario where investigators are uncertain whether a measured covariate meets the criteria for an IV or rather a confounder. The authors present results from two simulation studies designed to provide insight into the problem of conditioning on potential IVs in routine epidemiologic practice. The simulations explored the effects of conditioning on IVs, near-IVs (predictors of exposure that are weakly associated with outcome), and confounders on the bias and variance of a binary exposure effect estimate. The results indicate that effect estimates which are conditional on a perfect IV or near-IV may have larger bias and variance than the unconditional estimate. However, in most scenarios considered, the increases in error due to conditioning were small compared with the total estimation error. In these cases, minimizing unmeasured confounding should be the priority when selecting variables for adjustment, even at the risk of conditioning on IVs.

Instrumental Variable Analysis for Estimation of Treatment Effects With Dichotomous Outcomes

Instrumental variable analyses are increasingly used in epidemiologic studies. For dichotomous exposures and outcomes, the typical 2-stage least squares approach produces risk difference estimates rather than relative risk estimates and is criticized for assuming normally distributed errors. Using 2 example drug safety studies evaluated in 3 cohorts from Pennsylvania (1994-2003) and British Columbia, Canada (1996-2004), the authors compared instrumental variable techniques that yield relative risk and risk difference estimates and that are appropriate for dichotomous exposures and outcomes. Methods considered include probit structural equation models, 2-stage logistic models, and generalized method of moments estimators. Employing these methods, in the first study the authors observed relative risks ranging from 0.41 to 0.58 and risk differences ranging from -1.41 per 100 to -1.28 per 100; in the second, they observed relative risks of 1.38-2.07 and risk differences of 7.53-8.94; and in the third, they observed relative risks of 1.45-1.59 and risk differences of 3.88-4.84. The 2-stage logistic models showed standard errors up to 40% larger than those of the instrumental variable probit model. Generalized method of moments estimation produced substantially the same results as the 2-stage logistic method. Few substantive differences among the methods were observed, despite their reliance on distinct assumptions.