Module 5Centralized multimodal scoringv1.0 · cal 2026-05-22

Risk Engine

Every patient gets a 0–100 score built from explainable feature contributions: demographics, symptoms, comorbidities, imaging, procedures, pathology, treatment, response, and follow-up. Configure thresholds per use case to route patients into clinic, imaging, research, urgent review, or tumor board workflows.

Total scored: 0 patients.

Use-case thresholds

Clinic referral

Surface to clinic for new-patient triage

300.0%

0patients

Imaging escalation

Order pancreas-protocol CT/MRI for further workup

450.0%

0patients

Research inclusion

Eligible for research cohorts

500.0%

0patients

Urgent review

Same-week physician review

600.0%

0patients

Tumor board

Auto-routed to multidisciplinary review

800.0%

0patients

Score distribution

Patient counts by 10-point risk-score bucket. Tier zones are color-shaded (light → strong) and dashed lines mark each configured threshold — drag the sliders to move them.

Zone key:Clinic referral (≥ 30)Imaging escalation (≥ 45)Research inclusion (≥ 50)Urgent review (≥ 60)Tumor board (≥ 80)

Top-risk patients

Sampled top-25 across groups. Click into any patient for full attribution. Hover any group chip to see its definition.

PatientScoreGroupTop driver

PANDA risk-group key

AHealthy
BComorbidities
CGI symptoms ≥50
DElevated risk
EImaging ordered
FCyst / lesion confirmed
GEUS surveillance
HTumor confirmed
IOn treatment
JStage IV / progressing

Scoring engine is deterministic and decomposable. Weights are inline insrc/lib/data/risk-engine-queries.tsand intended to be calibrated against Research Hub outcomes — that feedback loop is the next major build. Per spec module 5 §"Continuous score recalibration using downstream outcomes".