Evaluating Short-Term Drug Effects Using a Physician-Specific Prescribing Preference as an Instrumental Variable

Abstract

The instrumental variable method that we have proposed appears to have substantially reduced the bias due to unobserved confounding. However, more work needs to be done to understand the sensitivity of this approach to possible violations of the instrumental variable assumptions.

The risk of infection associated with tumor necrosis factor α antagonists: Making sense of epidemiologic evidence

“…An instrumental variable refers to a variable independently related to treatment choice (i.e., TNF␣ antagonists), but unrelated to confounders or the outcome, other than through the actual treatment (18). Once these assumptions are fulfilled, an instrumental variable serves as an unconfounded substitute for the actual treatment, which results in an unbiased treatment effect estimate, even if confounders remain unmeasured.…”

Generating Evidence From Computerized Healthcare Utilization Databases

“…Finally, new analytic techniques, such as sensitivity analysis, 69 instrumental variable methods, 70 or propensity score calibration, 71 are increasingly applied to HCU databases to account for residual confounding. Although a description of their rationale and functioning is beyond the scope of the present review, an example of the advantages of the use of these methods may help the reader to penetrate this complex area.…”

Bias Analysis to Guide New Data Collection

“…Table 2: Stratified cross-tabulation of VKA therapy recorded in the county prescription registries (rows, the assumed gold-standard) versus history of heart valve replacement (columns, the IV assumed susceptible to misclassification of function), in the validation subset (n=24,647 IV analyses commonly use two-stage linear regression to estimate associations. In the first stage, the outcome is regressed on the instrument-as we have done-which yields an unconfounded estimate of the exposure-outcome association that is distorted by misclassification (Brookhart et al, 2006b;Brookhart et al, 2006a). In the second stage, the instrument-outcome association is scaled by the instrument-exposure association, yielding an unconfounded estimate of the exposure-outcome risk difference, adjusted for misclassification of the target exposure (Brookhart et al, 2006b).…”

“…IV analyses commonly use two-stage linear regression to estimate associations. In the primary stage, the outcome is regressed on the instrument, yielding an estimate of the exposureoutcome association that is unconfounded but distorted by misclassification (Brookhart et al, 2006a;Brookhart et al, 2006b). In the secondary stage, the instrument-outcome association is scaled by the instrument-exposure association, yielding an unconfounded estimate of the exposure-outcome risk difference, adjusted for misclassification of the target exposure (Brookhart et al, 2006b).…”