Figure 1.

Estimating effect size of surgery. Patients receive care over their lifetime, including gait analyses (●) and surgical interventions (■), to treat abnormal gait. (A) The true effect size of surgery is the difference in patient states with and without surgical intervention. Since, after a given gait visit, the patient can continue on only one of the “surgery” or “no surgery” paths, this effect size is unobservable and must be estimated using models built from patient history and gait visit data. (B) Data for each patient limb analyzed were first split into training and testing sets. The training data were used to estimate patient propensity for surgery at this center, select features, and build regression models to estimate outcome with and without surgical intervention. The held-out test data were used to evaluate the resultant regression models. Using the fixed training and testing data, these model-building and evaluation methods were repeated 1,000-fold to ensure results were robust to any stochastic variation in the training process.