Mental illness can take up to 23 years to reach a clinician. A new reference standard sets out how technology can close the gap.

A new multi-institutional paper from Kivira Health and clinicians and researchers at Dell Medical School, the University of Miami, Yale, Duke, Kaiser Permanente and the University of Roehampton defines what technology-powered mental health assessment should be able to do — from structured screening to continuous monitoring and real-time risk detection.

Mental health and neurological disorders account for about 10% of disability-adjusted life years worldwide. Yet, as the paper documents, most people who need care are never identified, or are identified years after their disorder begins. When they are identified, diagnoses frequently change: reported rates of diagnostic revision for schizophrenia range from 36% to 51%.

Assessment is also typically a one-off event rather than a continuous process. In the paper's words, “a single PHQ-9 score during a 15-minute primary care visit cannot capture the longitudinal dynamics of symptom onset, progression, treatment response, and relapse.” Without measurement-based care, detection of treatment non-response is delayed. Misclassification and poorly matched treatment can erode trust, and patients may disengage until symptoms escalate to a crisis or emergency presentation.

The paper argues that this is largely an infrastructure problem — and one that can start to be solved with tools that already exist.

A standard, not a product

The paper's authors span the system the problem runs through. Charles Nemeroff, chair of psychiatry and behavioral sciences at Dell Medical School and a member of the National Academy of Medicine. Philip Harvey of the University of Miami, an expert in cognition, functioning and outcomes measurement in serious mental illness. Carol Alter of Dell Medical School, who as system lead for behavioral health at Baylor Scott and White Health oversaw services including a Collaborative Care Model delivered to over two million primary-care patients. Kathryn Erickson-Ridout of the Kaiser Permanente Division of Research. Ish Bhalla of Yale and Duke. Hank Capps, most recently chief information and digital officer at Wellstar Health System. And Kivira's Matthew Vowels and Roehampton's Laura Vowels.

Together they reviewed 21 condition domains, 17 assessment strategies scored across more than 30 dimensions, the treatment landscape, reimbursement, and 16 care settings from primary care and emergency departments to telehealth and correctional facilities. They then asked what a viable standard must satisfy at once: seamless workflow integration, reimbursability, low cost and high scalability, applicability across populations and settings, and honest acknowledgement of limits.

Their answer is deliberately vendor-neutral. It describes the capabilities any system in this class should support, and the evidentiary standards that should apply — not a specific product.

“Scalable, structured, longitudinal assessment grounded in validated instruments and governed evidence is the missing infrastructure layer required to improve mental health care and patient outcomes at scale.” — the paper's central claim

Seven capabilities the standard defines

1. Assess broadly, in minutes. Validated brief instruments across diagnostic domains, with adaptive branching to keep assessment time under five minutes on average.
2. Keep measuring. Scheduled reassessment, trend detection and clinician alerts, turning screening from a one-off event into continuous monitoring and enabling earlier detection of treatment non-response.
3. Flag risk in real time. Real-time stratification across suicide, self-harm, violence, substance use and functional decline, with clinician notification and recommended escalation pathways.
4. Support treatment decisions with traceable evidence. Recommendations incorporating guidelines, patient profile, contraindications, treatment history and preferences, grounded in an evidence library with strength-of-evidence grading and provenance tracking.
5. Bridge the gaps. Referral recommendations, wait-time support and between-session exercises, and follow-up through care transitions — points where reviews document substantial losses of patients.
6. Live inside the health record. Deep EHR integration, and every output logged with full provenance: input data, algorithm version, evidence sources, timestamp and clinician review status.
7. Keep checking itself. A governed evidence layer with continuous re-validation using real-world deployment data, including diagnostic accuracy, instrument comparison, subgroup fairness and treatment outcomes.

Throughout, all clinical outputs are advisory and the clinician retains diagnostic and treatment authority. The standard defines three levels of human oversight, and no system may autonomously initiate treatment, modify medication or discharge a patient.

What it isn't

The authors are explicit about limits. Digital phenotyping, speech analysis, neuroimaging and sensor fusion are promising but currently better seen as adjunctive, given incomplete validation and the absence of established clinical thresholds. Conversational systems may capture rich information but currently lack the standardisation and psychometric grounding needed for reliable measurement. Choosing the right treatment for a specific individual is described as a genuinely hard scientific problem, not an implementation gap. The standard uses validated digital self-report as the scalable foundation, with evidence-tier requirements for other modalities proportionate to their clinical risk.

“Structured measurement augments but does not replace the clinical interaction; rapport, contextual judgment, and the capacity to explore ambiguity and nuance in real time remain essential elements of competent care that no instrument can fully replicate.” — from the paper