Augmented Inverse Probability Weighting and the Double Robustness Property

This article discusses the augmented inverse propensity weighted (AIPW) estimator as an estimator for average treatment effects. The AIPW combines both the properties of the regression-based estimator and the inverse probability weighted (IPW) estimator and is therefore a “doubly robust” method in that it requires only either the propensity or outcome model to be correctly specified but not both. Even though this estimator has been known for years, it is rarely used in practice. After explaining the estimator and proving the double robustness property, I conduct a simulation study to compare the AIPW efficiency with IPW and regression under different scenarios of misspecification. In 2 real-world examples, I provide a step-by-step guide on implementing the AIPW estimator in practice. I show that it is an easily usable method that extends the IPW to reduce variability and improve estimation accuracy. [Box: see text]

Abstract WP114: Effect of Prestroke Aspirin on Infarct Volume

Objective: To investigate if prestroke aspirin use affects infarct volume, generally and by subtype. Background: Prestroke aspirin use may reduce initial stroke severity and improve functional outcome at discharge, especially in the large artery atherosclerosis (LAA) stroke. We investigated whether prestroke aspirin use is associated with infarct volume and the effect of prestroke aspirin on infarct volume is different between LAA versus non-LAA strokes. Methods: A total of 4427 patients were included. The association between infarct volume on DWI and prior aspirin use was assessed by multiple linear regression analysis. To adjust a significant imbalance between aspirin users vs. non-aspirin users, the augmented inverse-probability weighting (AIPW) and propensity score matching were used. Results: Mean age was 67.7(SD 12.4) years and 58.7% were male. 19.6% (n=869) took aspirin before stroke. Prestroke aspirin had an inverse relation with log-infarct volume (P = 0.007) and the effect was significantly modified by LAA versus non-LAA strokes (P for interaction = 0.02). In LAA stroke (n=2336), prestroke aspirin use was independently associated with log-infarct volume (standardized beta = -0.047; P = 0.032). In non-LAA stroke (n=2091), prestroke aspirin was not associated with infarct volume (P = 0.27). Using AIPW and propensity score matching, the mean difference of log-infarct volume between prestroke aspirin user versus non-aspirin uses was -0.28 (95% CI -0.52 to -0.04, P = 0.009) and -0.39 (95% CI -0.67 to -0.11, P = 0.02) in LAA strokes. In non-LAA strokes, AIPW and propensity score matching showed that prestroke aspirin use was not associated with infarct volume (P = 0.27 and P = 0.85, respectively). Conclusions: Our results showed that prestroke aspirin use is negatively associated with initial infarct volume on DWI in LAA strokes but not in non-LAA strokes.

“Robust-Squared” Imputation Models Using Bart

Examples of “doubly robust” estimators for missing data include augmented inverse probability weighting (AIPWT) and penalized splines of propensity prediction (PSPP). Doubly robust estimators have the property that, if either the response propensity or the mean is modeled correctly, a consistent estimator of the population mean is obtained. However, doubly robust estimators can perform poorly when modest misspecification is present in both models. Here we consider extensions of the AIPWT and PSPP that use Bayesian additive regression trees (BART) to provide highly robust propensity and mean model estimation. We term these “robust-squared” in the sense that the propensity score, the means, or both can be estimated with minimal model misspecification, and applied to the doubly robust estimator. We consider their behavior via simulations where propensities and/or mean models are misspecified. We apply our proposed method to impute missing instantaneous velocity (delta-v) values from the 2014 National Automotive Sampling System Crashworthiness Data System dataset and missing Blood Alcohol Concentration values from the 2015 Fatality Analysis Reporting System dataset. We found that BART, applied to PSPP and AIPWT, provides a more robust estimate compared with PSPP and AIPWT.