Holmusk MaST

The model is the engine of MaST, and the tool carries other value. MaST's Risk of Crisis algorithm is a model fitted to electronic health record data, derived on one NHS trust's patients and validated on five more, that scores each person's risk of needing crisis services and sorts the caseload by risk and complexity. Around it sit dashboards that show each patient's journey and track team caseloads and performance, which have value without the score.

Holmusk's NeuroBlu research data business uses its own language models to structure clinician notes, and is described on this page for context.

Ask which parts of MaST depend on the risk score, and whether the score is recalibrated for each new trust or customer.

The score informs a team's decisions, and the structure is described without its threshold. MaST groups a caseload by crisis risk and complexity, and teams use its dashboards in caseload meetings, supervision and resource planning to decide who needs outreach and who can step down, so a clinician decides what happens to each person.

What is not published is the risk level at which a team is expected to act, how a person the model ranks low but a clinician is worried about is handled, and what happens after a crisis the model did not flag.

Ask what threshold triggers outreach, how clinicians override the score, and how missed crises are reviewed.

The approach and its data are described, without version or update discipline. Holmusk describes the Risk of Crisis algorithm as propensity scores built from an ordered list of clinical and sociodemographic predictors, including previous crises, recent mental health service contacts, Mental Health Act section, accommodation status and employment status, derived on 2,620 patients in one trust and validated on 107,879 across five more. No model version is published, and nothing describes how the model is updated or recalibrated for a new customer, or how customers learn of changes.

Ask for the model version in use, its update history, and whether it was refit for your population.

The model chain is in house and short, and the rest of the chain is unnamed. Holmusk built the Risk of Crisis algorithm itself from NHS trust records, with no outside model provider named or implied. The MaST agreement does not name hosting providers or subcontractors, and no subprocessor list is published.

Ask for the list of parties that host or process MaST data, including any cloud provider, and the terms that bind them.

Published validation of the risk score, and the outcome claims rest on less. The Risk of Crisis algorithm was derived on 2,620 patients in one NHS trust and externally validated on 107,879 patients across five more, identifying 64 to 80 percent of crises within its top quintile of risk, as reported in BMJ Health and Care Informatics and in a BJPsych Open abstract (2022), with authors from King's College London and Holmusk. No area under the curve, calibration or positive predictive value is reported in the published summary.

Holmusk's headline claims are fewer crises, shorter inpatient stays and lower cost. The support it cites is an NHS study estimating 1.7 million pounds saved in one trust over six months, which Holmusk presents as an 18 times annual return, and a systems efficiency audit reporting shorter crises and fewer admissions after implementation, neither with a published method.

Ask for calibration and positive predictive value at the risk level your teams act on, and for the method behind the savings study.

The contract covers deletion at termination and leaves the use of data broad. Under the MaST agreement Holmusk destroys all copies of customer content within 10 days of termination, and keeps a perpetual and irrevocable right to collect and analyze data relating to the provision, use and performance of the services and to use it to improve them. Whether that reaches patient records, and whether customer data trains or recalibrates the risk model, is not stated, and no retention period during the contract or de identification method is published for MaST.

Safety engineering, meaning how a faulty risk score is detected and handled, is not published.

Ask whether patient data is used to retrain or recalibrate the model, what is kept during the contract, and how any reuse is de identified.

HIPAA compliance is committed to in the contract, without business associate status stated. Holmusk's MaST software as a service agreement for US customers says Holmusk will meet HIPAA requirements and undertake risk assessments and audits when the customer requests them, and that customer content remains the customer's property. No business associate agreement is published or attached, and the website privacy statement excludes Holmusk's software and services. For UK trusts, no data protection position is published.

Ask for the business associate agreement, whether it binds Holmusk's subcontractors, and the UK data processing terms.

Controls are asserted with nothing independent behind them. Holmusk's website privacy statement says it maintains commercially reasonable administrative, technical and physical safeguards, and the MaST agreement lets customers request risk assessments and audits. No SOC 2 report, ISO/IEC 27001 certificate, HITRUST certification, Cyber Essentials certificate or trust center is published on Holmusk's site.

MaST works on behavioral health records from NHS trusts and US providers.

Ask for a current security attestation with its scope and date, and the NHS Data Security and Protection Toolkit status for UK deployments.

No regulatory position is published, and the score reaches clinicians with its basis visible. MaST ranks community mental health caseloads by risk of crisis, a prioritization of patients by clinical urgency. Neither the product page, the US software agreement nor Holmusk's news states a position under FDA rules in the United States or UK medical device regulation, or a clinical safety standard such as DCB0129. The factors the algorithm draws on, previous crises, recent service contacts, Mental Health Act status, accommodation and employment, are record data a clinician can open.

Ask how Holmusk classifies MaST for regulatory purposes in each market, and for its clinical safety case.

Validation across sites is published, and nothing by group. The Risk of Crisis algorithm was tested on five NHS trusts beyond the one it was built on, but no results are published by age, sex, ethnicity or diagnosis, and no governance process for the model is described. The predictors include accommodation and employment status, so a person's social circumstances move their score, and a model learning from who used crisis services in the past can rank people who never reached those services lower than their need.

Ask for performance by ethnicity, age and diagnosis in each validation trust, and for the process that reviews the model before release.

Clinicians can check the record behind a score, and the contract stands behind nothing. MaST shows the patient's journey and service history beside the risk grouping, so a team can test a score against the record. The US software agreement provides the services as is, disclaims fitness for a particular purpose, caps Holmusk's liability at fees paid in the six months before a claim, and offers an indemnity only for intellectual property infringement. No error rate at an operating threshold, warranty or correction route for a wrong score is published.

Ask what Holmusk does when a high risk person is missed, and for the false negative rate at the threshold your teams use.

A named record system with read access, and trust records behind the UK deployments. In the United States MaST integrates with Streamline Healthcare's SmartCare behavioral health record, announced in June 2023 with Mental Health Partners in Colorado as the first site. In NHS trusts it draws on multiple clinical sources within the trust's electronic health record, and at Mersey Care single sign on was added through the trust's business intelligence portal. Nothing published says MaST writes scores back into the record.

Ask which record systems MaST reads from today, whether risk scores can be written back to the patient's record, and how often data refreshes.

A single hosted service, with location implied. Under its US agreement Holmusk hosts the MaST software and provides it as a service. Where patient data is stored and processed for US or UK customers, which cloud carries it, and whether any processing crosses a border are not stated, and Holmusk operates from New York, Singapore, London and Basel.

Ask where MaST data is hosted for your country, by which provider, and whether staff in other countries can access it.

No price is published and how the product is sold can be established. Holmusk publishes its MaST software as a service agreement for US customers, last updated March 2023, under which fees and service levels are set in an order form. In the United Kingdom MaST is contracted by NHS trusts, and in the United States it is offered with Streamline Healthcare's SmartCare record. No price, unit of charge or implementation fee is published.

Ask for the price per team, per clinician or per patient on the caseload, and what integration with your record system costs.

Coverage is named, with validation behind the UK part of it. MaST is built for community mental health services. It has launched across eight NHS foundation trusts, including Mersey Care and Greater Manchester Mental Health, and its risk algorithm was validated on five trusts' records. In the United States it is offered to behavioral health providers on Streamline Healthcare's SmartCare record, with Mental Health Partners in Colorado as the first site, and no validation on US data is published.

Ask whether the risk score has been validated on your population and record system, and how it performs for children and older adults.