Data-Driven Individualized Surgical Decision-making: Beyond “Better on Average” Clinical Trial Results

Randomized clinical trial (RCT) results guide clinical decision-making, but trial results usually only offer information on mean treatment effects, overall or by large subgroups. For example, a trial showing the superiority of a surgery over noninvasive management may provide evidence that patients meeting the trial enrollment criteria, on average, benefit from surgery. However, no single patient may match the typical patient characteristics or outcome. Meanwhile, there is likely heterogeneity in experience, with some people more prone to experience worse outcomes with the therapy shown to produce a superior outcome. Traditional subgroup analysis is often underpowered, and evaluating one clinical characteristic at a time may not inform the likely outcome of any individual patient.

To provide patients and clinicians information beyond the mean treatment effect, existing RCT data could be leveraged to develop prediction models specific to each treatment arm to estimate the risk of specific outcomes for treatment options in parallel, based on each patient’s characteristics. Comparing the individualized probabilities of outcome with either treatment option, interpreted with the overall treatment effect reported in the trial, could provide tangible and relevant information for patients. While this approach to estimating individual treatment effect has been used in other specialties, examples of its application to surgical decision-making are limited.

Commonly used surgical risk models do not provide the probability of outcome resulting from a counterfactual scenario of not undergoing surgery, providing only half of the information needed to make informed treatment decisions. Risk models developed from RCT data could inform comparative risks and potential benefits of undergoing alternative treatment options.