Finding the Best Match — a Case Study on the (Text‑) Feature and Model Choice in Digital Mental Health Interventions

Research output: Journal contributionsJournal articlesResearchpeer-review

Standard

Finding the Best Match — a Case Study on the (Text‑) Feature and Model Choice in Digital Mental Health Interventions. / Zantvoort, Kirsten; Scharfenberger, Jonas; Boß, Leif et al.
In: Journal of Healthcare Informatics Research, Vol. 7, No. 4, 00148, 12.2023, p. 447-479.

Research output: Journal contributionsJournal articlesResearchpeer-review

Harvard

APA

Vancouver

Bibtex

@article{c90d7531bb5b4c94911b929e62aade9b,
title = "Finding the Best Match — a Case Study on the (Text‑) Feature and Model Choice in Digital Mental Health Interventions",
abstract = "With the need for psychological help long exceeding the supply, finding ways ofscaling, and better allocating mental health support is a necessity. This paper contributes by investigating how to best predict intervention dropout and failure to allow for a need-based adaptation of treatment. We systematically compare the predictive power of different text representation methods (metadata, TF-IDF, sentiment and topic analysis, and word embeddings) in combination with supplementary numerical inputs (socio-demographic, evaluation, and closed-question data). Additionally, we address the research gap of which ML model types — ranging from linear to sophisticated deep learning models — are best suited for different features and outcome variables. To this end, we analyze nearly 16.000 open-text answers from 849 German-speaking users in a Digital Mental Health Intervention (DMHI) for stress. Our research proves that — contrary to previous findings — there is great promise in using neural network approaches on DMHI text data. We propose a task-specific LSTM-based model architecture to tackle the challenge of long input sequences and thereby demonstrate the potential of word embeddings (AUC scores of up to 0.7) forpredictions in DMHIs. Despite the relatively small data set, sequential deep learning models, on average, outperform simpler features such as metadata and bag-of-words approaches when predicting dropout. The conclusion is that user-generated text of the first two sessions carries predictive power regarding patients{\textquoteright} dropout and intervention failure risk. Furthermore, the match between the sophistication of features and models needs to be closely considered to optimize results, and additional nontext features increase prediction results.",
keywords = "E-mental health, Health care analytics, Machine learning, Natural language processing, Precision psychiatry",
author = "Kirsten Zantvoort and Jonas Scharfenberger and Leif Bo{\ss} and Dirk Lehr and Burkhardt Funk",
note = "Funding Information: Open Access funding enabled and organized by Projekt DEAL. The present study has been funded by Leuphana University. The original RCTs were funded by the European Union (project EFRE: CCI 2007DE161PR001). Publisher Copyright: {\textcopyright} 2023, The Author(s).",
year = "2023",
month = dec,
doi = "10.1007/s41666-023-00148-z",
language = "English",
volume = "7",
pages = "447--479",
journal = "Journal of Healthcare Informatics Research",
issn = "2509-498X",
publisher = "Springer Nature Switzerland AG",
number = "4",

}

RIS

TY - JOUR

T1 - Finding the Best Match — a Case Study on the (Text‑) Feature and Model Choice in Digital Mental Health Interventions

AU - Zantvoort, Kirsten

AU - Scharfenberger, Jonas

AU - Boß, Leif

AU - Lehr, Dirk

AU - Funk, Burkhardt

N1 - Funding Information: Open Access funding enabled and organized by Projekt DEAL. The present study has been funded by Leuphana University. The original RCTs were funded by the European Union (project EFRE: CCI 2007DE161PR001). Publisher Copyright: © 2023, The Author(s).

PY - 2023/12

Y1 - 2023/12

N2 - With the need for psychological help long exceeding the supply, finding ways ofscaling, and better allocating mental health support is a necessity. This paper contributes by investigating how to best predict intervention dropout and failure to allow for a need-based adaptation of treatment. We systematically compare the predictive power of different text representation methods (metadata, TF-IDF, sentiment and topic analysis, and word embeddings) in combination with supplementary numerical inputs (socio-demographic, evaluation, and closed-question data). Additionally, we address the research gap of which ML model types — ranging from linear to sophisticated deep learning models — are best suited for different features and outcome variables. To this end, we analyze nearly 16.000 open-text answers from 849 German-speaking users in a Digital Mental Health Intervention (DMHI) for stress. Our research proves that — contrary to previous findings — there is great promise in using neural network approaches on DMHI text data. We propose a task-specific LSTM-based model architecture to tackle the challenge of long input sequences and thereby demonstrate the potential of word embeddings (AUC scores of up to 0.7) forpredictions in DMHIs. Despite the relatively small data set, sequential deep learning models, on average, outperform simpler features such as metadata and bag-of-words approaches when predicting dropout. The conclusion is that user-generated text of the first two sessions carries predictive power regarding patients’ dropout and intervention failure risk. Furthermore, the match between the sophistication of features and models needs to be closely considered to optimize results, and additional nontext features increase prediction results.

AB - With the need for psychological help long exceeding the supply, finding ways ofscaling, and better allocating mental health support is a necessity. This paper contributes by investigating how to best predict intervention dropout and failure to allow for a need-based adaptation of treatment. We systematically compare the predictive power of different text representation methods (metadata, TF-IDF, sentiment and topic analysis, and word embeddings) in combination with supplementary numerical inputs (socio-demographic, evaluation, and closed-question data). Additionally, we address the research gap of which ML model types — ranging from linear to sophisticated deep learning models — are best suited for different features and outcome variables. To this end, we analyze nearly 16.000 open-text answers from 849 German-speaking users in a Digital Mental Health Intervention (DMHI) for stress. Our research proves that — contrary to previous findings — there is great promise in using neural network approaches on DMHI text data. We propose a task-specific LSTM-based model architecture to tackle the challenge of long input sequences and thereby demonstrate the potential of word embeddings (AUC scores of up to 0.7) forpredictions in DMHIs. Despite the relatively small data set, sequential deep learning models, on average, outperform simpler features such as metadata and bag-of-words approaches when predicting dropout. The conclusion is that user-generated text of the first two sessions carries predictive power regarding patients’ dropout and intervention failure risk. Furthermore, the match between the sophistication of features and models needs to be closely considered to optimize results, and additional nontext features increase prediction results.

KW - E-mental health

KW - Health care analytics

KW - Machine learning

KW - Natural language processing

KW - Precision psychiatry

UR - http://www.scopus.com/inward/record.url?scp=85171458668&partnerID=8YFLogxK

UR - https://www.mendeley.com/catalogue/c1ccce9b-85a7-3e08-9249-6abbcf3b5023/

U2 - 10.1007/s41666-023-00148-z

DO - 10.1007/s41666-023-00148-z

M3 - Journal articles

C2 - 37927375

VL - 7

SP - 447

EP - 479

JO - Journal of Healthcare Informatics Research

JF - Journal of Healthcare Informatics Research

SN - 2509-498X

IS - 4

M1 - 00148

ER -