What is Inverse Probability of Treatment Weighting (IPTW)?

What Is Inverse Probability of Treatment Weighting (IPTW)?

Inverse Probability of Treatment Weighting (IPTW) is a method used to control for confounding in observational data. IPTW methods use propensity scores to assign weights to each subject creating a pseudo-population where covariates are balanced across exposure groups. 

Understanding Confounding in Observational Research

In a typical study, we want to measure the association between exposure and an outcome; for example: alcohol consumption and cancer. We cannot simply compare rates of cancer between drinkers and non-drinkers since they are likely to differ by other characteristics (covariates).  

If a covariate is independently associated with the outcome and differs between the exposure groups, it will confound the association between the exposure and the outcome. For example, smoking is a risk factor for cancer and drinkers may be more likely to smoke.  

How Propensity Scores Help Simulate Randomization

Randomized controlled trials are considered the gold standard of research because randomization balances covariates across treatment groups resulting in unbiased estimates. In an observational study, exposure status cannot be assigned but we can mimic randomization using propensity scores. The propensity score is the probability of exposure conditional on a subject’s observed covariates.  

Propensity scores can be used in a variety of analytic methods including stratification, regression, matching, or weighting. Stratification and regression have been shown to be less effective at controlling differences between exposure groups. Matching is an effective approach that provides intuitive results, but it can be challenging to implement and discards data from unmatched subjects. Weighting methods are versatile, straightforward to implement, and use all available data. 

Illustrating an IPTW Weighted Population 

In this example, the propensity score (probability of drinking) for smokers is .75 (3/4). The inverse probability weight among drinkers is 1/.75 = 1.33. and 1/(1-.75) = 4 among non-drinkers. For non-smokers the propensity score is .25 (1/4). The inverse probability weight among drinkers is 1/.25 = 4 and 1/(1-.25) = 1.33 among non-drinkers. 

In the original sample, smoking is imbalanced across the exposure groups, but after creating a weighted sample, the proportion of smokers is the same in each group (4/8). The observed association between exposure and the outcome will not be confounded by smoking in a weighted population. 

Study Design Considerations for Using IPTW Effectively

In an actual study, our propensity score would be calculated using a logistic regression model with many covariates and covariate balance would be assessed using standardized mean differences. If imbalances remained after weighting, we could re-fit the propensity model or adjust for those covariates in the analysis.  

Depending on the study question, we may want to use the inverse probability for weighting (as shown in this example) or the inverse of the odds of exposure. The former estimates the average treatment effect (ATE) across the entire population. The latter estimates the average treatment effect on the treated (ATT) and may be more relevant when assessing pharmaceutical products. 

Whatever weighting method we choose, the population should be assessed for extreme weights. Stabilized weights and truncation of weights can be applied to limit the influence of extreme values. 

Partner with Magnolia Market Access for IPTW-Supported Analysis

Our team can work with you to design and execute an analysis using IPTW or other methods to control for confounding. Contact us to support your research planning and ensure your analyses are grounded in robust methodologies.