The Doctor, Sir Luke Fildes, 1891
Psychiatry sets its diagnostic thresholds by diagnosis, not by clinical utility, and that’s a problem worth fixing
STUDY: Kotelnikova Y et al, JAMA Psychiatry 2026
STUDY TYPE: Review
FUNDING: Independent
Background
Psychiatry has dozens of validated rating scales, but the cutoffs that define “mild,” “moderate,” or “severe” are largely borrowed from diagnostic thresholds, which were never designed to tell us who needs treatment, who is at risk, or when is a medication worth the cost.
This review from a multisite team lays out four alternative ways to clarify when symptoms become a diagnosis:
- Statistical deviance
- Functional impairment
- Prediction of future outcomes
- Risk-benefit analysis
The most common approach, aligning a scale cutoff with a diagnosis, inherits the semi-arbitrary symptom counts that define the diagnosis and works well only for screening. Three better options exist for other purposes.
Statistical Deviance
This approach sets cutoffs relative to the general population: a T score above 65 marks the top 7% of the population (moderate elevation), above 70 the top 2.5% (marked), and above 75 the top 1% (severe). This is already standard in cognitive testing and personality assessment.
Functional Impairment
This approach sets cutoffs at points where scores predict real-world disability, using instruments like the Social and Occupational Functioning Assessment Scale (SOFAS). It’s useful for determining eligibility for services but misses patients who are not impaired on paper but not yet functionally affected.
Illness Prediction
This approach, borrowed from internal medicine, sets cutoffs based on risk of future harm: hospitalization, suicide attempt, or psychosis onset. For example, a self-harm scale cutoff at or below 4 on the SNAP-2 Self-Harm subscale identified 26% of outpatients as low-risk for suicide attempt with 94% sensitivity, functioning as a useful negative screen.
Risk-Benefit
This approach, rare in psychiatry but common in cardiology and endocrinology, identifies the symptom severity level at which treatment benefits outweigh costs. For example, a meta-analysis cited here found that computer-assisted cognitive therapy had a treatment effect size of Hedges g = 0.87 for high levels of depressive symptoms vs low levels, illustrating that who you treat matters as much as how you treat.
Pulling it Together
This review makes a clear case for applying different thresholds to different questions: use statistical deviance to communicate severity, functional impairment to determine service eligibility, prediction scores for suicide risk screening, and cost-benefit thresholds to decide who is likely to benefit from a specific treatment.
Practice Implications
- In a world where depression scales (PHQ-9) are embedded into every EMR, we need to think critically about when treatment is needed and worth the risks.
- Have you found other models to inform that decision?
— Chris Aiken, MD
Director, Psych Partners
Editor in Chief, Carlat Psychiatry Report
