Chen (2026) Development of an interpretable machine-learning model to predict insufficient functional improvement after aquatic therapy in stroke

In short: This 2026 study used data from 1,320 stroke patients to build machine-learning models that predict who is unlikely to achieve meaningful motor improvement after aquatic therapy. The best model, using only information available before the first pool session, predicted this reasonably well in a later group of patients.

About this study

Chen and colleagues carried out a single-center retrospective study of stroke patients who received a routine aquatic rehabilitation program between 2022 and 2025. The outcome was insufficient improvement on the Fugl-Meyer motor assessment (less than 10 points) about 56 days after the start of aquatic therapy. Patients from 2022 to 2024 were used to develop the models and patients from 2025 to validate them. Several algorithms were compared, and Shapley values (SHAP) were used to identify the most important predictors.

Main findings

• 409 of the 1,320 patients had insufficient improvement after aquatic therapy.
• A Random Forest model based on baseline data and the initial treatment plan reached an AUC of 0.818 in the validation group.
• Adding treatment data from the first 14 days did not improve prediction (AUC 0.814).
• The most important predictors were the planned duration and weekly frequency of aquatic therapy, baseline lower- and upper-extremity Fugl-Meyer scores and time from stroke onset to rehabilitation.

The authors conclude

The authors conclude that an interpretable model using routine clinical data available before the first aquatic session can help identify patients at risk of insufficient improvement, but that external validation and prospective studies are needed before clinical use.

Limitations

The study was retrospective and carried out in one center. It does not show that adjusting treatment based on the model improves recovery.

Source

Chen Y, Zhang J, Qie S. Development of an interpretable machine-learning model to predict insufficient functional improvement after aquatic therapy in stroke. Frontiers in Neurology. 2026;17:1871732. doi: 10.3389/fneur.2026.1871732. Published under a Creative Commons Attribution (CC BY) licence.