CMD-1 cuts crisis triage from 9 hours to 8–13 minutes

AI system

System: CMD-1 (Crisis Message Detector 1): key-word filter followed by a logistic-regression classifier on chat messages
Model: A logistic-regression classifier following a keyword filter.
Release: Exact release not established.
Application developer: Cerebral Inc. with Stanford University School of Medicine collaborators
Role: The classifier selected potentially urgent messages for crisis specialists to review.
Configuration: The deployment used CMD-1.
Inputs and outputs: Not assessed
Autonomy: Not assessed
Workflow: Thousands of daily patient chat messages on a telehealth platform with a crisis response team already in place. A key-word filter followed by a logistic-regression classifier trained on 721 labeled messages. A Slack-based workflow in which crisis specialists review and label surfaced messages.
Human oversight: Not assessed
Relation to AI: The classifier decides which messages are surfaced to crisis specialists, who then triage them. Humans labeled the outcomes. False negatives still reached specialists through usual channels. The effect is reported by the company that built and runs the system, with Stanford co-authors. The pre-deployment baseline is an internal figure. Stanford HAI's summary gives slightly different numbers (10 hours to under 10 minutes; 97%).
AI contribution: The classifier decides which messages are surfaced to crisis specialists, who then triage them. Humans labeled the outcomes. False negatives still reached specialists through usual channels. The effect is reported by the company that built and runs the system, with Stanford co-authors. The pre-deployment baseline is an internal figure. Stanford HAI's summary gives slightly different numbers (10 hours to under 10 minutes; 97%).

Benefits

Outcome identifiers: Not assessed
Benefit: daily median receipt-to-triage time 8–13 min versus 9 h before
Category: Reduced Task Time
Status: Not assessed
Beneficiary groups: People. Organizations.
Beneficiaries: Not assessed
Reach: Not assessed
Distribution and access: Not assessed
Duration: Not assessed
Expected?: Yes
Costs and harms: Not assessed

Measurement

Studies: Not assessed
Study design: Not assessed
Measurement: Not assessed
Setting and sample: Not assessed
Measures: daily median receipt-to-triage time 8–13 min versus 9 h before
Comparison: response times to patient chat messages before CMD-1 was deployed, per Cerebral's internal analysis
Results: Reference: over 9 h on average. Result: 8 to 13 min.
Additional value from AI: Not assessed
Limitations: Not assessed
Denominators: Not assessed
Materials for replication: Not assessed

Money

Kind of value: Not assessed
Amount: Not assessed
Currency and prices: Not assessed
Period and unit: Not assessed
Realization: Not assessed
Recipients and payers: Not assessed
Gross or net: Not assessed
Included costs: Not assessed
Comparison: Not assessed
Method: Not assessed
Return and payback: Not assessed

People and places

Participants: Developer: Cerebral Inc. with Stanford University School of Medicine collaborators. Operator: Cerebral Inc.
Jurisdictions: Benefit: United States. Developer: United States. Operator: United States.
Ownership and control: Not assessed
Dependencies: Not assessed
Cross-border connections: Not assessed
Geopolitical implications: Not assessed

Follow-up

Conditions for replication: Crisis specialists available to act on surfaced messages within minutes. A high-recall filter stage, since messages that do not pass it are classified as non-crises automatically. Message volume and language similar to the training data. The authors note the prospective PPV was 0.66.
Replication attempts: Not assessed
Later findings: No source dated after the October 2022 prospective deployment reports whether CMD-1 kept running or was adopted elsewhere; the January 2024 Stanford HAI piece only restates the paper's results.
Availability: Not assessed
Institutional decision: Not assessed
Grounds: Not assessed
Action and result: Not assessed
Success criterion: Not assessed

Evidence

Source numbers: AIBD-0311-source-0001–0003
Citations: Numbered references below.
Language: Not assessed
Source relationships: Developer
Original evidence: Not assessed
Access and versions: Not assessed
Supported findings: Not assessed
Evidence links: Not assessed
Evidence strength: Not assessed
Missing evidence: Not assessed

Grades

Assessed benefits: People. Organizations.
Depth: Not assessed
Reasons: People: Patients in crisis were triaged within minutes rather than hours on median. The paper reports no clinical outcome for any patient. Organizations: The 9-hour baseline comes from Cerebral's internal analysis. The 8–13 minute figure is from the company-authored evaluation. No independent measurement exists.
Observed or intended: Not assessed
Rubric: Not assessed
Assessor and date: Not assessed
Alternative assessments: Not assessed
Fill: The makers' claim

Entry details

Kind: Not assessed
Related entries: Not assessed
Field: Mental health crisis response / telehealth. Tele-mental-health chat message triage.
Purpose: Built to detect and surface crisis chat messages so that crisis specialists triage them sooner. The measured triage-time reduction is that purpose.
Dates: Date basis unassessed: 2022 (precision unassessed). Updated: 22 October 2026.
Stage: Not assessed
Version: SHA-256: b99c2969249e…

References

Other identifiers: Not assessed
Category mappings: Not assessed
Brief: Not assessed
Revisions: 7 revisions. Latest: 22 October 2026.

Sources

Peer-reviewed study Developer

Swaminathan A, López I, Garcia Mar RA, Heist T, McClintock T, Caoili K, Grace M, Rubashkin M, Boggs M, Chen JH, Gevaert O, Lorigo D, Nock MK, Williams PA, Ahuja M, et al. (2023). Natural language processing system for rapid detection and intervention of mental health crisis chat messages. npj Digital Medicine 6 (PMC10663535). Lead authors employed by Cerebral Inc.
Press release Developer

Kannan P (2024). Using NLP to Detect Mental Health Crises. Stanford HAI, 8 January 2024 (Stanford researchers co-authored the paper).