Individualized Prediction of Patient-Defined Treatment Success to Support Patient-Centered Clinical Decision-Making.

OBJECTIVE: Clinical prediction models often target generic thresholds for treatment success, such as the Minimal Important Change. The Personal Meaningful Gain was introduced as a personalized threshold, capturing the improvement an individual seeks to achieve in a self-chosen domain to become satisfied with the treatment results. This study developed and internally validated a treatment-conditional prediction model estimating the probability of achieving the PMG within 3 months in patients with common hand and wrist conditions. METHODS: This study used real-world data from 25 specialized hand and wrist care clinics in the Netherlands. We included 6263 patients (training data: n=5010; test data: n=1253) treated for common hand and wrist conditions. The prediction target was the PMG achievement at 3 months post-treatment. Model performance was evaluated using discrimination (area under the curve [AUC]) and calibration. The final model was selected based on AUC, calibration, and the number of model features. RESULTS: We selected a gradient boosting machine algorithm with thirteen predictors. The model had an AUC of 0.73 (95% CI: 0.70 to 0.76, indicating acceptable-to-good discrimination) and excellent calibration in the test set. Following a user-centered design, we built a Shiny app visualizing the probability of reaching the PMG for different treatment options (i.e., hand therapy ± injection, injection, or surgery) (https://doi.org/10.6084/m9.figshare.31084375). CONCLUSIONS: A treatment-conditional prediction model using a patient-defined threshold for treatment success can provide individualized estimates of the probability of achieving outcomes that matter to patients. This generalizable approach may enhance shared decision-making and expectation management in preference-sensitive care.