Digital therapeutics (DTx) show promise in bridging mental healthcare gaps. However, treatment selection often relies on availability and trial-and-error, prolonging suffering and increasing costs. Personalised prediction models could help identify individuals benefiting most from specific DTx.

The aim of this secondary analysis was to establish a machine learning-based prediction model for positive treatment outcomes in patients with depressive or anxiety symptoms after 8 weeks of internet-delivered cognitive behavioural therapy (iCBT).

We analysed a large real-world dataset of patients from the online therapy unit iCBT programme in Saskatchewan, Canada (2013–2021). Clinically significant changes in depressive symptoms or anxiety were measured using the Patient Health Questionnaire-9 (PHQ-9) and the Generalised Anxiety Disorder-7 (GAD-7). We trained six prediction models using sociodemographic and mental health-related factors at baseline, compared model performances and calculated Shapley values for feature importance.

Data from 4175 patients using 34 features for prediction, identified by least absolute shrinkage and selection operator regression, showed the Gradient Boosted Model (gbm) and logistic regression (log) performed best, with balanced accuracies of 0.76, 95% CI (0.70 to 0.83) and 0.70, 95% CI (0.63 to 0.77). Shapley values indicated GAD-7 scores at baseline as the most important predictor of clinically significant improvement, along with mental health history and sociodemographic variables.

The gbm and log models achieved comparable accuracy in predicting clinically significant improvement after iCBT, supporting the use of simpler, interpretable methods in clinical practice.

These findings could help improve mental health treatment selection, iCBT assignment, enhance effectiveness and optimise treatment for patients.

NCT05758285.