Data-Driven Approaches in Physiotherapy Clinics: Improving Clinical Outcomes and Operational Performance

• AFM Tanvir Anjum

  Author

• Priyanka Patel

  Author

• Azhar Uddin

  Author

• Dr Mahzabin Mahmud

  Author

• Shaiful Mahmud

  Author

• Israt Jahan

  Author

Missed rehabilitation appointments disrupt treatment continuity, waste limited clinical capacity, and may reduce access for patients requiring repeated therapy. Machine learning has been proposed to identify appointments for targeted reminders or flexible scheduling, yet many no-show studies rely on random resampling, outcome-proximal variables, or accuracy measures that obscure temporal drift and poor calibration. This study assessed whether routinely available scheduling data could support a deployable no-show model in a rehabilitation service including physiotherapy. We analysed 49,593 appointments recorded at the Centro de Especialização em Reabilitação, southern Brazil, from June 2016 to February 2022. A deployment-safe feature set included service, sex/gender, disability category, municipality, appointment timing, calendar variables, age, proxy lead time, and companion requirement. Post-outcome, unavailable, incomplete, or redundant variables were excluded. Penalised logistic regression, random forest, and histogram gradient boosting were trained on appointments through December 2021 and evaluated on a January-February 2022 temporal holdout using discrimination, average precision, calibration, Brier score, and decision-curve net benefit. Overall non-attendance was 9.7% (4,832/49,593), increasing from 4.0% in 2016 to 17.2% in the 2022 holdout. Physiotherapy accounted for 9,074 appointments with a 10.4% no-show rate. Penalised logistic regression performed best, but discrimination was near chance (ROC-AUC 0.506; average precision 0.163; Brier score 0.146). At a 0.20 threshold, 81.9% of appointments were flagged while capturing 86.4% of no-shows, producing negative net benefit. Therefore, the available administrative variables do not justify automated patient-level targeting. Clinics should improve data capture, consider universal attendance supports, and prospectively test risk-guided workflows before claiming clinical or financial benefit.

• PDF

Copyright (c) 2026 Package Printing

This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License.

All articles published are made available under the Creative Commons Attribution–NonCommercial 4.0 International (CC BY-NC 4.0) licence.