Direct Empirical Comparison of CHR (Clinical-High-Risk) Samples versus Population Samples

What the evidence says

Two independent lines of evidence bear on how CHR (clinical-high-risk) samples differ from comparison populations at the level of observable behavior: automated speech analysis and psychosocial/trauma history. On speech, one prospective cohort study found that automated classification of speech features predicted psychosis onset over a 2.5-year follow-up with 100% accuracy — correctly identifying all 5 individuals who transitioned to psychosis (CHR+) and all 29 who did not (CHR−) — and that this outperformed prediction from clinical diagnostic interviews (SIPS/SOPS), which reached 79% accuracy and misclassified 3 of 5 CHR+ and 4 of 29 CHR− participants. Within the same sample, speech coherence features (words per phrase, semantic coherence, determiner frequency) correlated with negative symptom severity (r = 0.57, P = 0.046). Both findings come from a single study with only 5 individuals who converted, a limitation the authors themselves flag as requiring replication in a larger sample.

On social cognition, a meta-analysis found CHR samples show small-to-moderate impairment on mentalizing tasks using static illustrations relative to healthy controls (g = −0.408, 95% CI −0.677 to −0.139), with the impairment concentrated specifically in intentionality tasks rather than mentalizing broadly. Schizophrenia and early schizophrenia show greater impairment than CHR on intentionality tasks, consistent with mentalizing deficits scaling with proximity to full psychosis.

On psychosocial stress and trauma, a systematic review of 16 studies found CHR samples report higher overall psychosocial stress than controls (13 of 16 studies), and — with more precision — that CHR individuals were 5.5 times as likely to report emotional abuse, 2.5 times as likely to report physical abuse, and 3.1 times as likely to report bullying victimization, with all 10 studies examining trauma finding significantly higher levels in CHR versus controls. Daily stressors and early or recent trauma occurred more frequently in CHR status, while significant life events did not differ between groups (2 studies, null). CHR samples also report higher social withdrawal (7 of 8 studies) and higher interpersonal sensitivity (5 of 5 studies) than controls.

Some of these psychosocial variables carry predictive, not just associational, weight: emotional abuse increased risk of transition to psychosis 3.8-fold in CHR individuals, perceived discrimination increased transition risk by 52.4% for every unit increase in lifetime discrimination score, and social withdrawal was associated with subsequent conversion in two studies. The relationship between psychosocial stress and symptom severity is less settled — five studies found increased stress associated with greater symptom severity, four found no relationship.

Digital-phenotyping evidence is thinner but points in a specific direction: CHR-p samples do not differ from controls in total smartphone use for social communication, but they report less daily passive social media use, and decreased text message reciprocity predicted 1- and 2-year conversion risk. None of these effects are reported with numerical magnitude in the available extracts.

Where the evidence splits

The relationship between psychosocial stress and symptom severity in CHR is contested: five studies found greater stress associated with greater symptom severity, four studies found no relationship. The sources on this page do not state what distinguishes the studies that found an association from those that did not — no difference in design, population, or period is given, so the split is unexplained by the sources here.

Social withdrawal shows a near-unanimous but not complete pattern: 7 of 8 studies found higher social withdrawal in CHR versus controls, one did not. The review does not explain what differed about the outlier study, so this split is also unexplained by the sources on this page.

Thresholds and dose-response

Perceived discrimination shows a stated dose-response relationship: each one-unit increase in lifetime perceived discrimination score is associated with a 52.4% increase in risk of transition to psychosis in CHR individuals. No other source below establishes a level, rate, or cutoff at which an effect appears, reverses, or changes size.

Open empirical questions

• Does the 100% speech-classification accuracy for psychosis onset replicate in a larger CHR+ sample, as the original authors themselves flag as needed?
• Does the CHR mentalizing deficit on intentionality tasks generalize beyond static illustration paradigms to other mentalizing task formats?
• What distinguishes the studies that found an association between psychosocial stress and symptom severity from the four that did not?
• What is the numerical magnitude of the difference in passive social media use and the strength of the text-message-reciprocity prediction of conversion risk in CHR-p samples?