🎯 Multi-Center Retrospective External Validation

Validating Phase 1 findings across diverse US patient populations and healthcare settings

Total Target Enrollment
5,000
Patients across 5 cohorts
Study Sites
12
Multi-center US locations
Data Elements
179
Comprehensive data collection
Endpoints
39
19 Primary + 20 Secondary
📊 Overall Enrollment Progress
0%
Current: 0 patients enrolled Target: 5,000 patients Remaining: 5,000 patients
🔬
Five Validation Cohorts
1. Echocardiography Cohort
AI echo analysis validation across 12 sites
Target
1,500
Current
0
Data Elements
27
Endpoints
7
0.0%
2. ECG Analysis Cohort
12-lead ECG rhythm and STEMI detection validation
Target
2,000
Current
0
Data Elements
30
Endpoints
8
0.0%
3. CAD Detection Cohort
Coronary artery disease detection vs angiography
Target
600
Current
0
Data Elements
41
Endpoints
8
0.0%
4. Risk Stratification Cohort
Primary prevention with 12-month MACE outcomes
Target
1,000
Current
0
Data Elements
41
Endpoints
8
0.0%
5. Longitudinal Monitoring Cohort
Time-series event prediction in high-risk patients
Target
1,000
Current
0
Data Elements
40
Endpoints
10
0.0%
📋
Data Elements Summary
179
Total Data Elements
128
Required Elements
51
Optional Elements
87.6%
Avg. Completeness
Cohort Demographics Clinical History Diagnostic Laboratory Imaging Outcomes Total
Echocardiography 6 6 0 3 10 2 27
ECG Analysis 4 6 13 3 0 4 30
CAD Detection 5 9 6 10 0 11 41
Risk Stratification 5 10 7 9 0 10 41
Longitudinal 5 8 8 7 0 12 40
TOTAL 25 39 34 32 10 39 179
🎯
Detailed Cohort Specifications

Echocardiography Cohort (1,500 patients)

Retrospective validation of AI echocardiography analysis across 12 sites. Comparing AI-calculated ejection fraction and wall motion assessment against expert cardiologist interpretations.

Primary Endpoints

1. EF Correlation
Correlation between AI-calculated EF and expert-measured EF using Pearson correlation coefficient
Phase 1
r = 0.96
Phase 2 Target
r ≄ 0.90
Success Criteria:
r ≄ 0.90 (non-inferiority to Phase 1: r=0.96)
2. EF Mean Absolute Error
Mean absolute error in ejection fraction calculation
Phase 1
±3.2%
Phase 2 Target
≀5.0%
Success Criteria:
MAE ≀ 5% (non-inferiority margin of -1.8%)
3. Clinical Agreement
Agreement on EF category (normal ≄50%, reduced <40%, borderline 40-49%)
Phase 1
Îș = 0.91
Phase 2 Target
Îș ≄ 0.85
Success Criteria:
Cohen's kappa ≄ 0.85

Key Data Elements (27 total)

Category Key Elements Required Completeness
Demographics (6) Age, Gender, Race, BMI, Height, Weight Required 95-100%
Imaging (10) LVEF (expert), Wall motion, LV volumes, LA size, RV function Required 85-98%
Clinical History (6) Prior MI, HF, HTN, DM, CAD, Valvular disease Required 90-95%
Laboratory (3) BNP, Troponin, Creatinine Optional 50-80%

ECG Analysis Cohort (2,000 patients)

Multi-center validation of AI ECG interpretation for rhythm classification and STEMI detection across diverse US populations.

Primary Endpoints

1. Overall Accuracy
Diagnostic accuracy across all rhythm classifications
Phase 1
97.3%
Phase 2 Target
≄95.0%
Success Criteria:
≄95% accuracy (non-inferiority margin: -2.3%)
2. STEMI Sensitivity
Critical: Sensitivity for detecting ST-elevation myocardial infarction
Phase 1
94.7%
Phase 2 Target
≄92.0%
Success Criteria:
≄92% sensitivity (life-threatening condition)
3. Atrial Fibrillation Sensitivity
Sensitivity for detecting atrial fibrillation
Phase 1
96.4%
Phase 2 Target
≄94.0%
Success Criteria:
≄94% sensitivity

Condition Distribution (2,000 ECGs)

Condition Target N Percentage Phase 1 Performance Phase 2 Target
Normal Sinus Rhythm 400 20% 98.1% ≄95%
Atrial Fibrillation 300 15% 96.4% ≄94%
STEMI 200 10% 94.7% ≄92%
Bundle Branch Block 200 10% 95.2% ≄93%
Ventricular Tachycardia 100 5% 95.8% ≄93%
Other Conditions 800 40% Variable ≄90%

CAD Detection Cohort (600 patients)

Validation of AI-based coronary artery disease detection compared against invasive coronary angiography gold standard.

Primary Endpoints (4)

Endpoint Phase 1 Performance → Phase 2 Target Success Criteria
Overall Accuracy 94.2% → ≄92.0% Non-inferiority margin: -2.2%
Sensitivity 92.8% → ≄90.0% Detect significant CAD (≄50% stenosis)
Specificity 95.1% → ≄93.0% Rule out CAD accurately
AUC-ROC 0.96 → ≄0.93 Excellent discrimination

Risk Stratification Distribution

200
Low Risk (33%)
250
Intermediate (42%)
150
High Risk (25%)

Key Data Elements (41 total)

Comprehensive clinical, laboratory, and diagnostic data collection:

  • Clinical History (9): Chest pain characteristics, exercise-induced symptoms, cardiac risk factors
  • Diagnostic Data (6): Baseline ECG, stress test results, METs achieved, exercise-induced angina
  • Laboratory (10): Complete lipid panel, glucose, renal function, biomarkers
  • Gold Standard (11): Angiography results, stenosis severity, SYNTAX score, CAD-RADS

Risk Stratification Cohort (1,000 patients)

Primary prevention cardiovascular risk prediction with prospective 12-month MACE outcome assessment.

Primary Endpoints

1. AUC-ROC for MACE Prediction
Discrimination for 12-month major adverse cardiac events
Phase 1
0.89
Phase 2 Target
≄0.85
Success Criteria:
AUC ≄0.85 (excellent discrimination maintained)
2. Model Calibration
Agreement between predicted and observed event rates
Success Criteria:
Hosmer-Lemeshow test p > 0.05 (well-calibrated model)

12-Month Outcome Events

Composite MACE definition includes:

  • Cardiovascular death
  • Myocardial infarction
  • Stroke
  • Heart failure hospitalization
  • Coronary revascularization (PCI/CABG)
  • Unstable angina requiring hospitalization

Risk Category Distribution (1,000 patients)

Risk Category Target N Percentage Expected 12-mo MACE Rate
Very Low 150 15% <1%
Low 250 25% 1-3%
Intermediate 300 30% 3-7%
High 200 20% 7-12%
Very High 100 10% >12%

Longitudinal Monitoring Cohort (1,000 patients)

Time-series prediction of cardiac events in high-risk patients with continuous monitoring data.

Primary Endpoints (4)

Endpoint Phase 1 Phase 2 Target Clinical Significance
HF Exacerbation Sensitivity (7-day) 86.7% ≄84.0% Early intervention opportunity
Specificity N/A ≄80.0% Minimize false alarms
Positive Predictive Value N/A ≄30.0% Actionable predictions
Prediction Lead Time N/A ≄3 days Time for clinical intervention

Prediction Windows Performance

89.4%
24-Hour Prediction
86.7%
7-Day Prediction (Primary)
84.3%
30-Day Prediction
81.8%
90-Day Prediction

Patient Population Distribution

Cohort N Percentage Monitoring Focus
Heart Failure (Any Class) 500 50% HF exacerbation, decompensation
Post-MI (within 6 months) 300 30% Recurrent MI, arrhythmia
High-Risk Pregnancy 200 20% Peripartum cardiomyopathy
📊
Phase 1 → Phase 2 Performance Targets
Cohort Metric Phase 1 (Single-Center) → Phase 2 Target (Multi-Center) Non-Inferiority Margin
Echo EF Correlation r = 0.96 → ≄0.90 -0.06
MAE ±3.2% → ≀5.0% +1.8%
Agreement (Îș) 0.91 → ≄0.85 -0.06
ECG Overall Accuracy 97.3% → ≄95.0% -2.3%
STEMI Sensitivity 94.7% → ≄92.0% -2.7%
AF Sensitivity 96.4% → ≄94.0% -2.4%
CAD Accuracy 94.2% → ≄92.0% -2.2%
Sensitivity 92.8% → ≄90.0% -2.8%
Specificity 95.1% → ≄93.0% -2.1%
AUC-ROC 0.96 → ≄0.93 -0.03
Risk AUC-ROC (MACE) 0.89 → ≄0.85 -0.04
Accuracy 88.4% → ≄85.0% -3.4%
Longitudinal HF Exacerbation (7d) 86.7% → ≄84.0% -2.7%
MI Risk (30d AUC) 84.3% → ≄82.0% -2.3%

* All Phase 2 targets set with conservative -3% non-inferiority margins to account for multi-center variability while maintaining clinical significance

📅
Study Timeline & Milestones
⏳
Q1 2026: Study Initiation (Current)
IRB approvals, site activation, protocol finalization, data collection planning
○
Q2 2026: Data Extraction Begins
First site data extraction initiation, establish data flow protocols
○
Q2-Q3 2026: Enrollment Progress
Continuous retrospective data acquisition across all 12 sites: Target 50% by Q3 2026
○
Q4 2026: 100% Data Acquisition
Complete all 5,000 records, finalize data extraction and quality checks
○
Q1 2027: Database Lock & Analysis
Data validation complete, lock database, begin statistical analysis
○
Q2 2027: Results & Manuscript
Complete analysis, prepare manuscript, conference presentation
○
Q3-Q4 2027: FDA Submission & Publication
FDA 510(k) submission with Phase 1+2 data, manuscript publication
✅
Overall Success Criteria

Primary Success Criteria

All 5 Must Be Met:

  • Overall accuracy ≄85% across all conditions
  • Critical diagnosis recall ≄90%
  • Enrollment ≄4,500/5,000 patients (90%)
  • Data completeness ≄90% for primary endpoints
  • Zero serious adverse events from AI

Secondary Success Criteria

≄3 of 5 Must Be Met:

  • Superiority to standard care on diagnostic accuracy
  • Processing time <5 minutes (workflow integration)
  • Clinician satisfaction score >7.5/10
  • Positive cost-effectiveness ratio
  • No clinically significant subgroup disparities

Go/No-Go Decision

Upon meeting all primary criteria and ≄3 secondary criteria:
PROCEED TO FDA 510(k) SUBMISSION (Q2 2026)

đŸ„
12 Multi-Center Study Sites
Academic Centers
6
Community Hospitals
4
Women's Health Centers
2

Participating sites represent diverse geographic regions, patient demographics, and practice settings across the United States, ensuring external validity and generalizability of Phase 2 findings.