A tutorial comparing different covariate balancing methods with an application evaluating the causal effect of exercise on the progression of Huntington’s disease

Randomized controlled trials are the gold standard for measuring the causal effects of treatments on clinical outcomes. However, randomized trials are not always feasible, and causal treatment effects must, therefore, often be inferred from observational data. Observational study designs do not allow conclusions about causal relationships to be drawn unless statistical techniques are used to account for the imbalance of confounders across groups while key assumptions hold. Propensity score (PS) and balance weighting are two useful techniques that aim to reduce the imbalances between treatment groups by weighting the groups to look alike on the observed confounders. There are many methods available to estimate PSand balancing weights. However, it is unclear a priori which will achieve the best trade-off between covariate balance and effective sample size. Weighted analyses are further complicated by small studies with limited sample sizes, which is common when studying rare diseases. To address these issues, we present a step-by-step guide to covariate balancing strategies, including how to evaluate overlap, obtain estimates of PS and balancing weights, check for covariate balance, and assess sensitivity to unobserved confounding. We compare the performance of a number of commonly used estimation methods on a synthetic data set based on the Physical Activity and Exercise Outcomes in Huntington Disease (PACE-HD) study, which explored whether enhanced physical activity affects the progression and severity of the disease. We provide general guidelines for the choice of method for estimation of PS and balancing weights, interpretation, and sensitivity analysis of results. We also present R code for implementing the different methods and assessing balanc