BACKGROUND
Precision psychiatry is data-driven personalisation of mental healthcare. One major aim is to identify the specific forms of therapy that will be most clinically meaningful for each patient. Before major progress towards this aim is possible however, tools that objectively assess the mental wellbeing of a patient must be validated in this context.
OBJECTIVE
To compare the fidelity of a therapy recommendation algorithm when trained with an objective quantification of psychological stress versus subjective ecological momentary assessments of stress and mood.
METHODS
From 2,786 unique individuals engaging between March 2015 and December 2022 in English language psychotherapy sessions and providing pre- and post-session self-report and facial biometric data via the AmDTx mental health platform (Mobio Interactive Pte Ltd, Singapore), analysis was conducted on 67 “super users” that completed at least 28 sessions with all pre- and post-session measures. AmDTx is a clinically validated mental health platform that provides patients with audio recordings supporting mental wellbeing (asynchronous and on-demand psychotherapy). AmDTx also contains easy to use tools that rapidly assess mental wellbeing, including an objective measure of psychological stress derived from AI analysis of facial biomarkers (Objective Stress Level; ∆OSL), and ecological momentary assessments (EMAs). Two commonly used EMAs within AmDTx are self-reported stress (∆SRS) and self-reported mood (∆SRM). These three data sources were used to independently train an algorithm designed to predict what future therapy sessions would prove most efficacious for each individual. Algorithm predictions were compared against the efficacy of the individual’s self-selected sessions.
RESULTS
EMAs demonstrated significant convergence with each other (r=0.53, P<.01), and divergence from the objective measure of psychological stress (r<0.03). The recommendation algorithm selected increasingly efficacious therapy sessions as a function of training data, when trained with the pre- and post-session ∆OSL (F(1, 16)=5.37, P=.034) or ∆SRM data (F(1, 16) = 3.69, P=.048). Using the first 15 therapy sessions, the algorithm trained using ∆OSL data predicted future sessions with a statistically higher efficacy relative to self-selected sessions (P<.01). Training the algorithm with self-reported data showed potential trends that did not reach significance (∆SRS: P=.092; ∆SRM: P=.12). Self-selections were over-represented amongst the algorithmically recommended sessions, an effect most pronounced for ∆OSL (F(1, 14)=30.94, P<.001).
CONCLUSIONS
A rapid, scalable, and objective measure of psychological stress, when combined with a robust recommendation algorithm, may improve clinical predictiveness within precision psychiatry and thereby support the capacity for healthcare professionals to meet global demand.
CLINICALTRIAL
clinicaltrials.gov