gms | German Medical Science

In this talk we give a brief overview of how IPTW can be used to analyze causal inference in an observational study. Real data is presented and visualized for each step of the analysis. For demonstration purposes, we use data from the Heinz Nixdorf Recall Study, a population-based prospective cohort study, to investigate the association between regularity and duration of midday napping and the occurrence of strokes, taking into account several cardiovascular confounders. A total of 4,636 participants aged 45-76 years were included in the analysis, 175 of whom suffered a stroke during a median follow-up period of 13.4 years.

First, we use standard logistic regression to estimate the propensity score for each patient and discuss the graphical representation of the probabilities within the exposure groups. We show how the positivity assumption can be checked using the PS. After calculating the weights, we give examples of how to visualize the balancing of the covariates in the weighted pseudo-population by examining standardized mean differences. We further explain how balance optimization can be achieved by changing the treatment model. In presenting the weight distribution, we illustrate that a few subjects can have very large weights and would therefore dominate the weighted analysis. We explain of how to deal with extreme weights using trimming of observations and by stabilizing the weights.

Finally, we apply weighted crude analysis to estimate the causal effect between midday napping and the incidence of stroke. In sensitivity analyses, we compare this result with effect estimates obtained after trimming the dataset and using stabilized weights, and show double-robust estimates after combining the exposure weighting model with the outcome regression model.

The authors declare that they have no competing interests.

The authors declare that an ethics committee vote is not required.