Predicting early patient dropout from a 12-week DMHI for depression and anxiety: a case study

Thumbnail Image

URL

Journal Title

Journal ISSN

Volume Title

School of Business | Master's thesis

Authors

Date

2024

Department

Tieto- ja palvelujohtamisen laitos

Major/Subject

Mcode

Degree programme

Information and Service Management (ISM)

Language

en

Pages

65 + 34

Series

Abstract

More than 1 in 5 people suffered from mental illness in the U.S in 2021, with major depression and anxiety disorders ranking among the most prevalent mental illnesses. Meanwhile less than half of these people were treated for their illness. The imbalance in supply and demand has caused digital mental health interventions (DMHIs) to emerge as a scalable and effective alternative to traditional interventions. However, dropout is still a major issue with up to 50% dropout rates reported for guided depression interventions. At the same time, early engagement has been shown to improve symptoms the most, but early dropout has not been studied. Dropout predictors have been studied extensively for both traditional and digital interventions, but prediction of dropout in DMHIs is less studied. In this thesis we used tree-based machine learning ensemble methods to predict dropout from weeks 2,3,4, and 5 onwards in a 12-week DMHI for depression and anxiety using sociodemographic, clinical baseline, program details, week 1 activity, and therapist information of 22,849 patients. Furthermore, we analysed SHAP values for feature importance to understand which datapoints predicted early dropout. As findings, we were able to predict early dropout from week 5 onward with Bal-anced Accuracy (BACC) of 0.76, and from week 2 onward with BACC of 0.83 using a Gradient Boosting Classifier. All models beat chance (0.5) and the benchmark BACC rate (0.67) necessary for iCBT treatment adaption. Week 1 activity data significantly improved prediction performance. The number of ac-tive days in week 1, total seconds spent doing meditation in week 1, the therapists past completer rate, and the patient cost estimate were significant predictors of dropout throughout different dropout variants. For the case company our prediction model can enable timely intervention which can increase adherence and consequently treatment outcomes. Identifying factors that predict dropout can help the company understand patterns that signal increased risk of dropout.

Description

Thesis advisor

Malo, Pekka

Keywords

mental health, machine learning, dropout, churn, prediction, DMHI

Other note

Citation

Permanent link to this item

https://urn.fi/URN:NBN:fi:aalto-202408185516

Collections

[gradu] Kauppakorkeakoulu / BIZ