Do You Really Need to See the Doctor? Developing a Machine Learning Approach for Optimal Assignment Rules of Physician Visits for Patients After Knee Replacement
Speaker(s)
Cordier J1, Langenberger B2, Salvi I3, Kuklinski D4, Geissler A1
1University of St. Gallen, St. Gallen, SG, Switzerland, 2Technical University Berlin, Berlin, Berlin, Germany, 3University of St. Gallen, St. Gallen, St. Gallen, Switzerland, 4University of St. Gallen, Zürich, Switzerland
OBJECTIVES: Knee replacement patients regularly see their physician after surgery to receive feedback on their recovery progress, and on potential alteration of their post-surgery treatment protocol. In a world with increasing healthcare costs and limited availability of healthcare personnel, only patients that require a visit should see their physician. Analytics can help to steer patients to visit their physician only if it benefits their recovery progress. With this paper we aim to develop an algorithm that optimally assigns rules for physician visits at three-, six- and twelve-months post-surgery based on patient characteristics and recovery pathways. Optimal assignment rules are built to maximize functional improvement of the patient, measured by the Knee injury and Osteoarthritis Outcome Score (KOOS), a disease specific patient-reported outcome measure (PROM). METHODS: We use patient-level data for 3’110 cases that underwent surgery in nine German hospitals in 2020 and 2021. Data was collected in the “PROMoting Quality” trial and contains patient characteristics, the post-treatment path, physician visits, and the KOOS at three-, six- and twelve-months post-surgery. We use a causal forest to estimate the double-robust treatment effects of visiting a physician, controlling for patient characteristics. Subsequently, we build a policy tree to develop the optimal treatment assignment rules, whether a patient should see her physician. RESULTS: We expect that on average physician visits do not significantly lead to higher KOOS improvements. However, we assume to see heterogeneous effects allowing the model to optimally assign patient profiles to physician visits, and in turn to see overall welfare increases as only patients in need see their physician. CONCLUSIONS: We present a novel approach to determine the optimal assignment to physician visits for patients after knee replacement. Selecting the right patients for a physician visit increases the average KOOS improvement, and therefore has a positive effect on overall welfare.
Code
MSR135
Topic
Economic Evaluation, Methodological & Statistical Research
Topic Subcategory
Artificial Intelligence, Machine Learning, Predictive Analytics, Thresholds & Opportunity Cost
Disease
Musculoskeletal Disorders (Arthritis, Bone Disorders, Osteoporosis, Other Musculoskeletal), No Additional Disease & Conditions/Specialized Treatment Areas