Built for the systems that already deployed LACE and need more.
Marqi Index is peer-reviewed AI that predicts 30-day readmission risk from the structured EHR and closed claims data your system already has, with discrimination 8 to 19 points of AUC above the most widely deployed readmission scores in published literature.
Trusted by healthcare organizations
HIPAA Compliant
Full BAA coverage
SOC 2 Type II
In development
Peer Reviewed
npj Digital Medicine
Validated
0.807 AUC
The penalty math is public. The miss rate is not.
Under the CMS Hospital Readmissions Reduction Program, approximately 75 percent of eligible hospitals are penalized in any given fiscal year, with a maximum payment reduction of 3 percent of Medicare base payments. Across the program, CMS withholds roughly $300 million in payments annually, applied to the six measured conditions: acute myocardial infarction, heart failure, pneumonia, COPD, CABG, and elective hip or knee arthroplasty.
For a 350-bed hospital averaging 20,000 Medicare discharges a year, a 1.5 percent payment reduction translates into seven figures of recurring revenue loss before any quality scorecard impact, payer rate impact, or reputational impact is counted.
The risk scores most systems use to triage these patients, LACE and HOSPITAL, were built more than a decade ago against data far narrower than what your EHR holds today. In published validation studies, both score in the AUC 0.62 to 0.73 range for 30-day readmission. That performance ceiling is the gap Marqi Index was built to close.
Sources: CMS Hospital Readmissions Reduction Program (FY 2024 IPPS Final Rule), peer-reviewed LACE and HOSPITAL validation literature.
Predicting the readmissions that drive the penalty.
Marqi Index is a 30-day readmission risk model built and validated for the conditions CMS measures under HRRP. The current product covers acute myocardial infarction, with peer-reviewed validation in submission at npj Digital Medicine. Additional HRRP conditions are in the development queue on a published roadmap.
Three near-term capabilities, in the order they ship:
30-day all-cause readmission
A single model trained across HRRP and non-HRRP conditions, in active validation with academic medical center co-authors. Targeted submission Q3 2026.
External validation across health systems
Independent recomputation of the AMI model on validation partner data, with full calibration and discrimination metrics reported. Targeted submission Q4 2026.
Cross-note longitudinal extraction
The Marqi Index v2 architecture, currently in research, adds structured signals extracted from clinical notes alongside the structured EHR signals already used. Targeted submission Q1 2027.
The capability your team starts with today is the 30-day AMI readmission model. The capability you grow into through a Marqi partnership is the full HRRP and all-cause suite, with cross-note signal extraction in v2.
A direct comparison against the scores in your chart today.
| Dimension | LACE | HOSPITAL | Marqi Index |
|---|---|---|---|
| Published discrimination (AUC, 30-day readmission) | 0.62 to 0.70 | 0.65 to 0.73 | 0.807 on temporal validation, AMI cohort (95% CI 0.771 to 0.838) |
| Method generation | Logistic regression, 4 variables | Logistic regression, 7 variables | 15-seed transformer ensemble with XGBoost out-of-fold meta-learner |
| Data the score reads | Discharge disposition, age, comorbidity index, ED visits | Hemoglobin, sodium, oncology, procedure, prior admissions, index admission type, length of stay | Hundreds of structured EHR and closed claims features, calibrated and audited per build |
| Calibration reporting | Variable across literature | Variable across literature | Calibration plot, intercept, slope, and Brier score reported per validation |
| Limitations disclosure | Limited in original publications | Limited in original publications | Full limitations section in Paper 1, including the data integrity event we caught and disclosed |
| Integration into the EHR | Manual scoring or static rules | Manual scoring or static rules | API delivery to discharge planning, case management, and care coordination workflows. Active integration work for Epic, Cerner, and FHIR-compatible environments |
| Independent peer review | Yes, decade-old | Yes, decade-old | Yes, manuscript under submission at npj Digital Medicine (Nature Portfolio) |
Published AUC
Method
EHR Integration
Peer Review
Published LACE and HOSPITAL ranges drawn from peer-reviewed validation studies in JAMA Internal Medicine, BMJ Open, and the Journal of Hospital Medicine. Marqi Index numbers drawn from the Paper 1 manuscript currently in submission. Full citations available on the Evidence page.
From signed DUA to scored discharges in 90 to 120 days.
Marqi Index is not a rip-and-replace EHR product. It is a risk score delivered to the workflows your case managers and care coordinators already run. The deployment path is the same one your team has used for any other clinical decision support integration, with a Marqi-led validation period built in.
Data Use Agreement
Weeks 1-6Your privacy office and ours align on a HIPAA-compliant DUA. Marqi provides a template that has already cleared multiple academic medical center privacy reviews. Data sent under the DUA is limited to the structured EHR and closed claims fields the model uses. No clinical notes are required for the current product.
Data integration and validation
Weeks 4-12Your data engineering team or ours establishes the data feed. Marqi runs the AMI model on a temporal validation cohort from your environment and reports back discrimination, calibration, and subgroup performance, with limitations disclosed in writing.
Workflow integration
Weeks 8-16Risk scores are delivered to your discharge planning, case management, or care coordination workflow through standard interfaces. Active integration work is underway for Epic, Cerner, and FHIR-compatible environments. Custom integrations supported for non-standard environments.
Live scoring
Week 16 onwardDischarges are scored in production, with a 6-month review period that produces a written validation report and a recommendation on expanded deployment.
What is and is not required from your team is captured below. The dependency that drives timeline most is the DUA, not the technical integration. Customers that have a HIPAA DUA template ready to redline move fastest.
Validated by the systems closest to the use case.
Marqi Index is in active validation conversation with multiple academic medical centers and integrated delivery networks across the United States. Co-authors of Paper 1 sit at institutions whose population mix, EHR environment, and quality program structure closely mirror the systems Marqi is built to serve. Specific institutions and names will be added to this page as public co-listing permission is granted.
For the buyer who wants to see the validation evidence today, the manuscript currently under submission at npj Digital Medicine contains the full cohort, methodology, discrimination, calibration, and limitations. Request the evidence brief through the link below and we will share it on call.
Request the validation evidence briefA real partnership, structured around what we both need.
Marqi Index validation works best when both sides come in with shared expectations. Here is what a partner health system commits to and what Marqi commits to in return.
Your system commits to
- An executive sponsor, typically a CMO, VP of Quality, or Chief Population Health Officer
- A HIPAA-compliant DUA, redlined against the Marqi template
- A data engineering point of contact for the structured EHR and closed claims feed
- A 6-month validation window, with the option to extend, expand, or end on transparent terms
- A clinical workflow owner for case management or care coordination integration
- A point of view on which HRRP conditions matter most in your population
Marqi commits to
- A named technical and clinical lead from Marqi, available for the duration of the engagement
- A DUA template that has already cleared multiple academic medical center privacy reviews
- The data engineering, validation, and reporting infrastructure required to score your population
- A written validation report at the end of the period, with discrimination, calibration, subgroup performance, and limitations disclosed
- Standard integrations for Epic, Cerner, and FHIR-compatible environments. Custom support where needed
- Roadmap visibility and influence on the order of the all-cause and cross-note expansion
The product, the evidence, and the methodology behind it.
A production-grade risk score
Marqi Index delivered into your discharge planning, case management, or care coordination workflow through standard interfaces. Scored in production on your population, with calibration and discrimination monitored continuously.
A written validation report
At the end of the 6-month validation period, a report covering cohort, discrimination, calibration, subgroup performance, and limitations on your specific population. Suitable for sharing with your quality committee, your board, and your payer partners.
The methodology behind the model
Full disclosure of the architecture, the data sources, the feature engineering approach, and the retraining cadence. We do not treat the methodology as a proprietary black box, because the buyers we serve evaluate AI vendors on whether the methodology can survive a careful read.
The publication co-authorship if you want it
Validation partners interested in academic publication are invited to co-author the Paper 3 external validation manuscript, on terms aligned with academic norms.
The questions every CMO has asked us.
If you are accountable for the readmission number on your scorecard, we want a conversation.
A 30-minute call covers Paper 1 evidence in detail, the Marqi Index product, the deployment approach in your environment, and what a 6-month validation partnership looks like. No deck dump, no sales theater, no asks before you have seen the evidence.