Precision Psychiatry
Identifying - in real time - what are the most efficacious forms of psychotherapy for each individual.
Psychiatry, despite the great and global need for progress, has notprogressed like other fields of medicine. Moreover, the disparities in psychiatric care, both globally and between socioeconomic groups within any given region, are extreme.
Our primary discovery from a prospective real world study is that Mobio's objective measure of psychologicalstress is superior to ecological momentary assessments for psychotherapy recommendation algorithm training. This result is exciting because unlike ecological momentary assessments, Mobio's objective psychological stress measure is not affected by cultural (e.g., language, perception of stress) or physiological variation (e.g., skin tone, age, gender), and only requires the common smartphone to enable data collection.
Wang et al. (2024). Scalable precision psychiatry with an objective measure of psychological stress: Prospective real world study. Journal of Medical Internet Research. (In press)
Psychological scale limitations and an overviewof data used in psychiatry. (A) Three “buckets”of limitations inherent to psychological scales. While validated scales will remain integral to the diagnosis of mental health conditions, they nonetheless have profound limitations for use in precision psychiatry. (B) Venn diagram depicting the overlap between quantifiable data and desireddata in precision psychiatry. The objective measures often utilized inthe current precision psychiatry, such as neuroimaging and genotyping, have little overlap with assessments that are needed to assess treatment efficacy. We propose an AI prediction of psychological stress based on heart rate variability may represent data that is both objective and required.
Performance of the recommendation algorithmtrained with three separate measures of wellbeing. A-C:The differential of AS (algorithmically selected) and US (user selected) sessions as a function of the training set size. For stress measures (A & B), values below zero (dark orange) indicate the AS sessions are more efficacious than US sessions. For the measure of mood (C), values above zero (dark orange) indicate AS sessions are more efficacious than US sessions. D-F: Comparisons of the efficacy of AS and US sessions when 15 sessions were included in the training set. G-I: Cross-validation of D-F via 10-fold bootstrapping, taking 80% of users for each sample. Error bars represent standard error of the mean. *p<.05. ∆OSL = objectivestress level, ∆SRS = self-reported stress, ∆SRM = self-reported mood, AS = algorithmically selected, US = user selected
Overlap of the algorithmically selected (AS) anduser selected (US) sessions in the testing data. Expected (by chance) and actual number of users with one, two, or three or more AS sessions in the testing dataset, as the training dataset increased from 1 to 25 sessions. In all three cases, and most pronounced for ∆OSL, more overlap was observed than what would be expected by chance. ∆OSL = objective stress level, ∆SRS = self-reported stress, ∆SRM = self-reported mood, AS = algorithmically selected, US = user selected
