Estimating the Causal Effect of the Sword Health Program on Healthcare Procedures

Abstract

In recent years, demographic and economic changes around the world have created a huge burden on the healthcare systems. The insufficiency of resources, affecting everything from infrastructures to healthcare professionals, demands new strategies and approaches. Musculoskeletal (MSK) conditions represent a major contributor to this problem, making them a critical focus for modern healthcare solutions. This dissertation is the result of an internship at Sword Health, a company that provides AI-driven digital care solutions for MSK disorders. Therefore, the main objective of this work is to estimate the causal impact of Sword solutions on the reduction of healthcare procedures. Evaluating these programs is crucial for measuring their validity and effectiveness and, since the study is conducted using real-world observational data, the presence of a selection bias and confounding factors increases the complexity of the task. To overcome these challenges, several advanced causal inference methods were explored. Beyond traditional Propensity Score and Difference-in-Difference, the data characteristics demanded the implementation of a more sophisticated methodology. Consequently, the Callaway & Sant'Anna (C&S) framework was adopted for its ability to handle heterogeneous treatment effects over time. Furthermore, due to high data dimensionality, Double Machine Learning was combined with the C&S, resulting in the DML-CSDID method. This combination, apart from dealing with staggered data, also reduces bias and captures nonlinear confounder relationships.