Supplementary Material for the Paper "Opening the black box - Effects of decision transparency on therapists' trust in and intended use of AI-based decision support systems in ICBTs"

This study investigated the impact of local SHAP (SHapley Additive exPlanation) values on therapists’ perceptions of a Machine Learning-based Decision Support Tools (DST) in Internet-based Cognitive Behavioural Therapy (ICBT). The randomised experiment included 35 Swedish ICBT therapists who were each presented with a DST with six exemplary patient cases with SHAP values and six without them. The study measured therapists' understanding, trust, and perception of clinical usefulness through self-ratings at multiple stages. The primary hypothesis that adding SHAP values increased self-reported trust was confirmed by the findings (p=0.01, d=0.43 [0.07-0.76]). Further, the results suggested increased understanding, agreeance with, and the perceived usefulness of the DST predictions through SHAP values.