A General Propensity Score for Signal Identification Using Tree-Based Scan Statistics.

The tree-based scan statistic (TreeScan; Martin Kulldorff, Harvard Medical School, Boston, Massachusetts) is a data-mining method that adjusts for multiple testing of correlated hypotheses when screening thousands of potential adverse events for signal identification. Simulation has demonstrated the promise of TreeScan with a propensity score (PS)-matched cohort design. However, it is unclear which variables to include in a PS for applied signal identification studies to simultaneously adjust for confounding across potential outcomes. We selected 4 pairs of medications with well-understood safety profiles. For each pair, we evaluated 5 candidate PSs with different combinations of 1) predefined general covariates (comorbidity, frailty, utilization), 2) empirically selected (data-driven) covariates, and 3) covariates tailored to the drug pair. For each pair, statistical alerting patterns were similar with alternative PSs (≤11 alerts in 7,996 outcomes scanned). Inclusion of covariates tailored to exposure did not appreciably affect screening results. Inclusion of empirically selected covariates can provide better proxy coverage for confounders but can also decrease statistical power. Unlike tailored covariates, empirical and predefined general covariates can be applied "out of the box" for signal identification. The choice of PS depends on the level of concern about residual confounding versus loss of power. Potential signals should be followed by pharmacoepidemiologic assessment where confounding control is tailored to the specific outcome(s) under investigation.

Risk of Psychiatric Adverse Events Among Montelukast Users.

BACKGROUND: There have been conflicting results from observational studies regarding the risk of psychiatric adverse events (PAEs) with montelukast use. OBJECTIVE: To determine whether there are associations of depressive disorders, self-harm, and suicide with use of montelukast compared with inhaled corticosteroid (ICS) use. METHODS: Using data from the Sentinel Distributed Database from January 1, 2000, to September 30, 2015, patients (n = 457,377) exposed to montelukast or ICS, aged 6 years and older with a diagnosis of asthma, were matched 1:1 on propensity scores. Hazard ratios (HRs) and 95% CIs were estimated for each study outcome overall and by age, sex, psychiatric history, and pre-/post-2008 labeling updates using Cox proportional hazards regression models. RESULTS: Exposure to montelukast was associated with a lower risk of treated outpatient depressive disorder (HR, 0.91; 95% CI, 0.89-0.93). No increased risks of inpatient depressive disorder (HR, 1.06; 95% CI, 0.90-1.24), self-harm (HR, 0.92; 95% CI, 0.69-1.21), or self-harm using a modified algorithm (HR, 0.81; 95% CI, 0.63-1.05) were observed with montelukast use compared with ICS use. Most PAEs occurred in the roughly one-third of patients having a past psychiatric history. CONCLUSIONS: When compared with use of ICS, we did not find associations between montelukast use and hospitalizations for depression or self-harm events. Our findings should be interpreted considering the study's limitations. Psychiatric comorbidity was common, and most PAEs occurred in patients with a past psychiatric history.

Quantifying how small variations in design elements affect risk in an incident cohort study in claims.

BACKGROUND: Epidemiological study reporting is improving but is not transparent enough for easy evaluation or replication. One barrier is insufficient details about design elements in published studies. METHODS: Using a previously conducted drug safety evaluation in claims as a test case, we investigated the impact of small changes in five key design elements on risk estimation. These elements are index day of incident exposure's determination of look-back or follow-up periods, exposure duration algorithms, heparin exposure exclusion, propensity score model variables, and Cox proportional hazard model stratification. We covaried these elements using a fractional factorial design, resulting in 24 risk estimates for one outcome. We repeated eight of these combinations for two additional outcomes. We measured design effects on cohort sizes, follow-up time, and risk estimates. RESULTS: Small changes in specifications of index day and exposure algorithm affected the risk estimation process the most. They affected cohort size on average by 8 to 10%, follow-up time by up to 31%, and magnitude of log hazard ratios by up to 0.22. Other elements affected cohort before matching or risk estimate's precision but not its magnitude. Any change in design substantially altered the matched control-group subjects in 1:1 matching. CONCLUSIONS: Exposure-related design elements require attention from investigators initiating, evaluating, or wishing to replicate a study or from analysts standardizing definitions. The methods we developed, using factorial design and mapping design effect on causal estimation process, are applicable to planning of sensitivity analyses in similar studies.