The Road Ahead
Our Research & Product Roadmap
We advance the Marqi Index along two tracks running in parallel: rigorous external validation of our prediction layer, and prospective validation of the workflow execution layer that turns risk scores into prevented readmissions. The science ships either way.
80%
Internal validation complete
≥0.80
Target AUC (MIMIC-IV)
3 EHRs
External transportability goal
2
Peer-reviewed papers shipped
The Foundation
Two Layers of the Marqi Index
Everything on this roadmap maps to one of two layers. The first predicts who is at risk. The second acts on that prediction—and proves the action works.
Marqi Index v1
Prediction Layer
The validated risk engine. It scores every patient for 30-day and 365-day readmission risk—across prediction and prevention—so care teams know who is at risk before discharge.
- Trained on 1.2M+ real patient encounters
- Prediction and prevention risk scores
- 30-day and 365-day windows
- Proven generalizability across EHRs (Track 1)
Marqi Index v2
Workflow Execution Layer
The action engine. It turns a risk score into a recommended, trackable intervention inside the clinical workflow—then measures whether that action actually prevents the readmission.
- Risk scores become recommended interventions
- Embedded in the live clinical workflow
- Outcomes measured against controls
- Proven to reduce readmissions (Track 2)
Two Tracks, One Goal
Running in Parallel
Track one proves the model generalizes. Track two proves the model changes outcomes. A self-funding backstop sits behind both—the same model that delivered our first two papers.
Track 1
External Validation
Marqi Index v1 · Prediction Layer
Funding
Self-funded + Bridge
Deliverable
Paper 2 + Paper 3
Goal
3 EHRs
Track 2
Workflow Validation
Marqi Index v2 · Workflow Execution Layer
Funding
NHLBI SBIR
Deliverable
v2 + Efficacy
Goal
Fewer readmits
Track 1 — External Validation Timeline
Internal Validation
In progress80% complete with a 60-day target to finish. Anticipated AUC ≥ 0.80 on the MIMIC-IV all-cause cohort across 30-day and 365-day windows, including prediction and prevention risk scores.
External (AMI & HF) Validation Begins
Retrospective external validation begins with leading academic medical centers and health systems, extending the evidence base to acute myocardial infarction and heart failure cohorts.
All-Cause External Validation Complete
Multi-site transportability story locked across three independent EHRs. The external validation paper ships, demonstrating generalizability beyond the development environment.
Track 2 — Workflow Validation Timeline
Target Trial Emulation
Retrospective data paired with a hypothetical v2 intervention and causal inference methods—propensity scoring and marginal structural models—to estimate counterfactual outcomes.
Prospective Study
Live deployment of the v2 workflow at a partner site using a stepped-wedge or parallel-group design, measuring real-time outcomes against controls.
Commercial Deployment
Integrated payvider design partners, Epic Showroom listing, and a defined revenue model bring validated workflow execution to scale.
Self-funding backstop: if grant funding does not land on schedule, Marqi self-funds the target trial emulation and prospective study. Either way, the science ships.
How We Validate
Data & Methodology
The roadmap is only as credible as the methods behind it. Here is how we keep results rigorous, reproducible, and comparable across institutions.
OMOP-Standardized Data
Validation runs against OMOP common data model extracts, so cohorts are harmonized across institutions and results are comparable site to site.
Defined Cohort Scope
30-day all-cause readmission plus an AMI subcohort spanning 2022–2025, with heart-failure cohorts layered in during external validation.
Causal Inference Methods
Propensity scoring and marginal structural models power the target trial emulation, estimating the counterfactual outcomes of the v2 intervention.
Governed & IRB-Aligned
Full data use agreements and standard IRB pathways govern every engagement—rigor and patient-data stewardship come first.
What Success Looks Like
Milestones & Success Criteria
Concrete checkpoints mark progress along the way—each one a measurable step from internal benchmark to validated impact at the bedside.
Internal validation finished
MIMIC-IV all-cause benchmark locked at target AUC.
External validation underway
Retrospective AMI & HF cohorts open at partner sites.
Transportability proven
Three independent EHRs, external validation paper shipped.
Workflow efficacy modeled
Target trial emulation estimates v2 impact on readmissions.
Validated at the bedside
Prospective deployment and a path to commercial scale.
How We Partner
A Two-Layer Partnership Model
Dollars are attached to both layers. Partners gain both scientific credit and funded scope—mutual wins that compound over the lifetime of the collaboration.
Layer 1
Bridge SOW with Full DUA
Marqi-funded · now through Q1 2027
- Scoped statement of work with attached funding
- Full data use agreement for the OMOP extract
- IRB pathway proceeds normally
- Retrospective external validation in 3–6 months
- Feeds the Phase II application as preliminary data
Layer 2
SBIR Phase II Subaward
Grant-funded · Q1 2027 forward
- 18-month subaward from the Phase II award
- Named co-investigator role on the submission
- Continues and expands the bridge work
- Funds the target trial emulation
- Self-fund backstop if the grant does not land
Mutual Wins
What Partners Walk Away With
Two streams of value—scientific and financial—that compound over the lifetime of the partnership.
Scientific Value
- Co-authorship on the upcoming external validation papers
- Co-first authorship for substantive methodology contributions
- Named co-investigator roles on grant submissions
- OMOP harmonization work publicly credited
- Continuation into prospective Phase III studies
Financial Value
- Direct bridge funding to start validation now
- Multi-year subaward scope as grant funding lands
- Legal, IRB, and data-engineering time covered in scope
- Two distinct funded workstreams, not one
- Self-fund backstop preserves value if a grant slips
How We Operate
Guiding Principles
The convictions that shape every decision on this roadmap.
Rigor over hype
Every claim is anchored to peer-reviewed evidence and pre-registered methodology—never a marketing number.
Transportability first
A model only matters if it works outside the lab. We prove generalizability across independent EHRs before we scale.
The science ships either way
A self-funding backstop sits behind both tracks. Grant timing can move; the evidence base does not stall.
Outcomes, not just scores
Prediction is table stakes. We measure whether the workflow actually prevents readmissions in the real world.
Roadmap Questions
Common questions about how the research and product roadmap comes together.
Why run two tracks at the same time?
Track 1 proves the prediction model generalizes across health systems, while Track 2 proves the workflow layer changes outcomes. Running them in parallel means we build the efficacy story while the transportability evidence matures—neither waits on the other.
What happens if grant funding is delayed?
A self-funding backstop sits behind both tracks. If grant funding does not land on schedule, Marqi self-funds the target trial emulation and prospective study so the research continues uninterrupted.
How do you ensure results hold across different hospitals?
We validate against OMOP-standardized data extracts and lock a multi-site transportability story across three independent EHRs before publishing the external validation paper. Harmonized cohorts make site-to-site comparisons valid.
What does partnering look like in practice?
Most partnerships start with a scoped, funded bridge engagement under a full data use agreement, then expand into a multi-year subaward as grant funding lands. Partners receive both scientific credit and funded scope.
Build the evidence base with us
We're partnering with academic medical centers, health systems, and payers to validate the Marqi Index across diverse populations and prove it reduces readmissions in the real world.