Die Rehabilitation · 2026
[Potential of Machine Learning for Predicting Individual Treatment Success in Inpatient Psychosomatic Rehabilitation].
Аннотация на английскомСтатьяНазвание и аннотация: Перевод в очереди
Обработка статьи · этапы, ошибки и управление
Загрузка состояния…
Ошибки и ограничения
История · было → стало
Состояние обновляется каждые 5 секунд, пока блок открыт. Процент показывается только там, где известен объём работы. Пауза останавливает продолжение этапа; текущий запрос к источнику может завершиться.
Velthuysen Paul-Gerrit, Kröger Christoph, Kobelt-Poenicke Axel
Аннотация
PURPOSE: Accurate prediction models of individual psychosomatic rehabilitation success based on patient characteristics might help to identify potential non-responders and to optimize the fit between patients and treatment. The aim of the present study was to investigate the potential of machine learning in this context.
METHODS: Random forest models were trained to predict treatment success in terms of the dimensions of symptom severity, health-related quality of life, interpersonal relationship building, and self-efficacy. In addition, relevant predictors were analyzed. Since random forest allows using both scale values and individual items as independent predictors, the present study utilized 555 potential predictors from 16 self-assessment instruments and 11 personal characteristics. This is a secondary data analysis of a dataset collected in a naturalistic single-group pre-post design. The random forest models were implemented using nested cross-validation. In the outer cross-validation, the final model was validated, and the included predictors were analyzed for their respective permutation importance.
RESULTS: Depending on the dimension, between 30.02% and 38.94% of the variance in individual treatment success could be explained. In external cross-validation, between 29.33% and 37.83% of the variance was explained. Across all dimensions, a total of 45 predictors - 17 scale scores and 28 single items - were included. Only two predictors were relevant for all dimensions.
CONCLUSION: Using random forest models, large proportions of variance in individual treatment success could be explained for all investigated dimensions. In cross-validation, the large effect sizes and generalizability regarding new observations could be confirmed for all investigated dimensions. The analysis of relevant predictors provides additional evidence for predictors already associated with treatment success, while emphasizing the need of a nuanced consideration of psychosomatic rehabilitation success. A practical implementation of personalized treatments in the sense of personalized medicine does not seem feasible in the near future due to numerous limitations.
METHODS: Random forest models were trained to predict treatment success in terms of the dimensions of symptom severity, health-related quality of life, interpersonal relationship building, and self-efficacy. In addition, relevant predictors were analyzed. Since random forest allows using both scale values and individual items as independent predictors, the present study utilized 555 potential predictors from 16 self-assessment instruments and 11 personal characteristics. This is a secondary data analysis of a dataset collected in a naturalistic single-group pre-post design. The random forest models were implemented using nested cross-validation. In the outer cross-validation, the final model was validated, and the included predictors were analyzed for their respective permutation importance.
RESULTS: Depending on the dimension, between 30.02% and 38.94% of the variance in individual treatment success could be explained. In external cross-validation, between 29.33% and 37.83% of the variance was explained. Across all dimensions, a total of 45 predictors - 17 scale scores and 28 single items - were included. Only two predictors were relevant for all dimensions.
CONCLUSION: Using random forest models, large proportions of variance in individual treatment success could be explained for all investigated dimensions. In cross-validation, the large effect sizes and generalizability regarding new observations could be confirmed for all investigated dimensions. The analysis of relevant predictors provides additional evidence for predictors already associated with treatment success, while emphasizing the need of a nuanced consideration of psychosomatic rehabilitation success. A practical implementation of personalized treatments in the sense of personalized medicine does not seem feasible in the near future due to numerous limitations.