Methods Projects

This study compares covariate selection strategies for confounding adjustment in secondary database analyses via Plasmode simulation: high-dimensional propensity score, machine learning algorithms: lasso, elastic net, random forest, etc.

We introduce an inverse probability weighted approach to rebalance covariates across treatment groups while mitigating the influence of differential misclassification bias.

This large, nationally representative study of opioid prescription patterns suggest that the introduction of a tamper-deterrent formulation of long-acting oxycodone, against a background of changes in public drug benefit policy, was associated with statistically significant, sustained changes in selection of long-acting opioids but only modest changes in the quantities of long-acting opioids dispensed.

Among venous thromboembolism patients, treatment with DOACs when compared to warfarin is not associated with an increased risk for major bleeding or all-cause mortality.

Using bisacodyl in combination with commonly prescribed colonoscopy bowel cleansers was not associated with increased risk of ischemic colitis, major adverse renal outcomes, or death.

Through simulations, we show that, in the presence of weight model misspecification in a marginal structural Cox model context, with a rich and diverse set of candidate algorithms, 'Super Learner' can generally offer a better alternative to the commonly used logistic regression or statistical learning approaches.