CORPUS
Population risk management · Payers / Providers
Caritas: the predictive layerbetween your cohort and the decision.
Caritas reads the clinical data your institution already holds and returns a calibrated AMI probability per patient, with its operating band. Built for payers, providers, and insurers running prevention at scale, where prioritizing well is the difference between preventing an event and reacting to one.
01 / individual view One patient
Eight clinical variables,three numbers, one band.
This patient exists in the cohort. The eight variables come from a routine visit, no advanced imaging, no special biomarkers. Caritas reads them together and produces a calibrated probability.
14.7% AMI risk · 6 months
22.4% AMI risk · 12 months
8.4 m Median time to event
Top factors · relative contribution
- HbA1c
- LDL
- Age
High risk top 12% of the cohort
02 / population view One cohort
15% of the cohortconcentrates 55% of the events.
The engine separates the cohort into four strata. Critical (3%) + High (12%) = 15%, and that's where 55% of the expected AMIs land. That is what justifies prioritizing.
Stratum
% Cohort
% Expected events
Critical
% Cohort 3%
% Expected events 17%
High
% Cohort 12%
% Expected events 38%
Medium
% Cohort 30%
% Expected events 24%
Low
% Cohort 55%
% Expected events 21%
Top 15% (Critical + High) → 55% of expected events. Operationally: if the payer or provider only has capacity to intervene on 15% of the cohort, it would catch more than half of the infarctions.
03 / intervention Modifiable factors
Four levers,−19% in acute events.
The model identifies the modifiable factors of each patient. When all four are addressed in a High-band cohort, the expected reduction in acute AMI is 19 percentage points.
Preliminary figure · primary outcome expected at 3 months · current cut from an in-flight prospective cohort.
The levers are not arithmetically additive: 7 + 4 + 4 + 4 ≠ 19. The weights are relative contributions calibrated on the cohort, the combined effect is measured empirically, not computed.
04 / trajectory Risk over time
Three possible futuresfor the same patient.
A High-band patient's 10-year trajectory changes drastically depending on how much of the modifiable risk is addressed. The model quantifies the three paths.
- No intervention 85%
Cumulative risk if no modifiable factor is addressed. The progression is what the model predicts from the initial profile.
- Key factors 58%
Targeted intervention on the two or three highest-weight levers for this patient, typically LDL and blood pressure. Significant reduction, not full.
- Full intervention 38%
All four levers active simultaneously: LDL, hypertension, glycemia, and clinical follow-up. The clinically achievable floor with the available information.
Trajectories derived from the predictive model and the intervention response observed in the LatAm cohort. Individual patients may vary by adherence, uncaptured comorbidities, and intercurrent events.
05 / evidence What the numbers say
Near-perfect calibrationand robust discrimination.
Most cardiovascular models report AUC and ignore calibration. Caritas inverts the emphasis: calibrating a probability well is what makes it a usable clinical decision. Ordering correctly is necessary but not sufficient.
When the model says 22%, it is 22% in reality.
O/E (observed over expected) compares how many events actually occurred against how many the model predicted. A value near 1.0 means the model's probabilities are numerically correct: when it says 22% risk at 12 months, the patient really does have a 22% chance of AMI in the actual cohort, not 5%, not 60%. That is what lets you use the probability as a clinical threshold, share it with the patient, and define operating bands. Discrimination (AUC, C-index) tells you the order is right; calibration tells you the numbers are real.
O/E
0.998
Near-perfect calibration. The numerical probability the model reports matches the observed rate in the cohort.
AUC
0.869
Discrimination: the model separates patients who will have an AMI from those who will not in 86.9% of random pairs.
C-index 6m
0.836
Time-aware discrimination at 6 months: the priority order matches the actual order of event occurrence.
C-index 12m
0.846
Time-aware discrimination at 12 months. Concordance holds when the prediction horizon is extended.
External validation on an insurer never seen during training. The separation between bands holds, calibration survives the population shift.
5,602
Patients
147.6
F statistic
<.001
p value
Observed rate per band
High 54.6%
Medium 38.1%
Low 21.0%
Classic models (Framingham, SCORE2, PROCAM, PREVENT) measure 10-year risk on European or American cohorts. Applied as-is in LATAM they overpredict: O/E < 1. The system ends up prioritizing people who won't have the event.
Systematic LATAM pattern: Framingham and SCORE2 overpredict (O/E < 1). Sex-adjusted PROCAM partially closes the calibration gap in Colombia. Caritas calibrates almost perfectly on the real cohort.
Modelo
AUC
O/E
Scope
Corpus AI
0.869
0.998
Colombia · 1990–2025 · AMI 6–12 m
Framingham Fuente: Muñoz 2014 · n=1,013
0.658
0.76
Colombia · primary prevention · 10 years
PROCAM (sex-adj) Fuente: Muñoz 2014 · n=1,013
0.744
0.94
Colombia · best localized model · 10 years
SCORE2 Fuente: López-López 2025 · n=2,022
0.68 – 0.72
22–42% overprediction
Colombia · PURE cohort · 12.3 years
Framingham Fuente: Camargos 2024 · n=12,155
0.77
0.38
Brazil · ELSA · 4.2 years
SCORE2 Fuente: Camargos 2024 · n=12,155
0.76
0.63
Brazil · ELSA · recalibrated low-risk
PREVENT (AHA) Fuente: Mancini 2024 · Scheuermann 2024
,
,
No external LATAM validation with incident outcomes as of May 2026
06 / precision screening From model to procedure
Five times more efficient,the same clinical yield.
100 → 100 → 5
Patients screened → Scans ordered → Clinical findings
5%
Yield per scan yield per scan: 5%
100 → 20 → 5
Patients screened → Scans ordered → Clinical findings
25%
Yield per scan yield per scan: 25%
The model does not replace the intervention. It makes it financially viable: screening the top 10% by risk captures the same clinical yield with 5× less advanced imaging.
07 / integration How we connect
Four input channels,four output channels.
Caritas integrates with what your institution already runs and delivers results through the channels the clinical team already uses.
How we receive data
REST API
Programmatic integration. The model processes clinical events as they arrive. Built for systems with in-house development or middleware that already emits JSON.
Payload example
POST /v1/score
{
"patient_id": "...",
"vitals": { "sbp": 148, "dbp": 92 },
"labs": { "hba1c": 8.2, "ldl": 142, "egfr": 58 },
"history":{ "smoking": "active", "bmi": 31 }
}
SFTP / flat files
The fastest path to start. One CSV per week with the cohort; Caritas returns the same format enriched with score and band. No client-side development.
Payload example
patient_id,age,hba1c,ldl,egfr,sbp,smoking,bmi,last_visit_months P0001,64,8.2,142,58,148,active,31,11 P0002,57,6.1,135,72,128,never,28,4
HL7 / FHIR
For institutions with FHIR-native EHRs. Caritas consumes Bundles, maps resources (Observation, Condition, MedicationStatement), and returns a FHIR RiskAssessment.
Payload example
{
"resourceType": "Bundle",
"type": "transaction",
"entry": [
{ "resource": { "resourceType": "Observation",
"code": { "coding": [{ "system": "http://loinc.org",
"code": "4548-4" }] },
"valueQuantity": { "value": 8.2, "unit": "%" } } }
]
}
Direct EHR connector
When the institution prefers not to send data outside. Caritas connects via VPN to a read-only view and pulls the agreed fields. Output returns via dashboard or webhook.
How we deliver results
Embedded web dashboard
Browser access, no install. Prioritized cohort table, individual profile, band filters. Built for clinical managers and lead physicians.
Payload example
/dashboard?cohort=cardiometabolic&band=high
EHR-native alerts
Native banner inside the clinician's workflow when they open a High-band patient. Suggested action, not mandatory. No screen switch.
Payload example
context.alert {
"severity": "high",
"patient": "...",
"band": "high",
"action": "CCTA / CT-FFR · interventional consult"
}
Automatic reports
Weekly or monthly PDF for clinical management: band distribution, current cohort, changes since the last cut. Built for management meetings.
Payload example
GET /v1/reports/weekly?format=pdf
API / Webhook
For integrating with call center, scheduling, or care-gap closure systems. When a patient changes band, a POST event fires to the client's URL.
Payload example
POST <client_webhook>
{
"event": "band_change",
"patient_id": "...",
"from": "medium",
"to": "high",
"actions": ["schedule_ccta", "outreach_call"]
}
Operating bands and clinical action
standard primary prevention · annual check-up
pharmacological optimization · targeted follow-up · 6-month re-evaluation
CCTA / CT-FFR · interventional consult · selective catheterization
08 / data required What we need to start
Five variables are enough.Nine is the model's ceiling.
Minimal data 5 variables
0.78 AUC
Variables - Age and sex
- Active comorbidities (hypertension, diabetes, CKD)
- Lipid profile (LDL / HDL)
- HbA1c + renal function
- Blood pressure + BMI
What a basic EHR already holds. AUC 0.78. Enough to start prioritizing and seeing return before investing in additional infrastructure.
Complete data 9 variables
0.869 AUC
Variables - Minimum dataset (5 variables)
- Active / past smoking
- Family history of early AMI
- Physical activity record
- Time since last visit
The four additional variables push calibration and discrimination to the model's ceiling. If you already capture them in the visit, there's no extra work; if not, we can add them progressively.
Want to see Caritason your own data?
30 minutes with our clinical-technical team. We map your available data subset, estimate the AUC you can expect in the first iteration, and lay out the integration plan. We respond within 2 business days.