DissertationFirst publication DOI: 10.48548/pubdata-1596

Machine Learning Dropout Predictions for Personalizing Digital Mental Health Interventions

Chronological data

Date of first publication2025-01-24
Date of publication in PubData 2025-01-24
Date of defense2025-01-10

Language of the resource

English

Related external resources

Related part DOI: 10.1038/s41746-024-01360-w
Zantvoort, K., Nacke, B., Görlich, D., Hornstein, S., Jacobi, C., Funk, B. (2024). Estimation of minimal data sets sizes for machine learning predictions in digital mental health interventions. npj Digital Medicine, 7(1), Article 361.
Related part DOI: 10.1177/20552076241248920
Zantvoort, K., Hentati Isacsson, N., Funk, B., Kaldo, V. (2024). Dataset size versus homogeneity: A machine learning study on pooling intervention data in e-mental health dropout predictions. Digital Health, 10.
Related part DOI: 10.1007/s41666-023-00148-z
Zantvoort, K., Scharfenberger, J., Boß, L., Lehr, D., Funk, B. (2023). Finding the Best Match — a Case Study on the (Text‑) Feature and Model Choice in Digital Mental Health Interventions. Journal of Healthcare Informatics Research, 7(4), 447-479.

Related PubData resources

Supplemented by

Resource
Dataset
Zantvoort, Kirsten; Nacke, Barbara; Görlich, Dennis; Hornstein, Silvan; Jacobi, Corinna; Funk, Burkhardt

Abstract

With the global need for psychological help long exceeding the supply, finding ways of increasing and better allocating mental health support is paramount. Digital Mental Health Interventions (DMHIs) are an effective way of complementing current mental healthcare measures. However, their effectiveness depends on engagement with the interventions and users prematurely dropping out are a significant problem across DMHIs. While it is known that measures such as human guidance can lower dropout, it remains unclear which user would most benefit from such measures. With health staff’s time being scarce and costly, the question arises of how to best allocate it. Machine Learning (ML) models are a powerful tool for predicting individual behaviour and have the potential to identify those patients most in need of support. However, translating ML methods into the field of DMHIs has presented challenges, as 1) datasets tend to be small, 2) the gathered data is noisy, and 3) little insight into how to translate predictions into care exists. This thesis addresses these problems by systematically exploring these hurdles across six papers. In cooperation with clinical psychologists from three universities, a total of >11.000 individual users’ data are analysed in four Machine Learning papers. Further, a randomised trial investigates therapists’ opinions on explainable predictions, and a systematic review defines and quantifies the prevalence of personalization in DMHIs. The contributions 1) quantify and address the limitations of small data sets in DMHIs, 2) provide proofs-of-concept for common and innovative data types and model combinations, and 3) explore ways to increase trust and actionability of predictions to facilitate the adaption of ML-based Decision Support Tools in DMHIs. Beyond that, this PhD contributes to the very limited body of research on intervention dropout predictions by analysing a variety of large datasets and, hence, advancing a highly relevant academic and practical topic.

Keywords

Digital Health; Machine Learning; Intervention; Mental Health; Personalization; Dropout

Grantor

Leuphana University Lüneburg

Study programme

More information

DDC

613 :: Persönliche Gesundheit und Sicherheit
006.31

Creation Context

Research