From emergency admission to ICU triage — how AI-assisted risk stratification fits the acute pancreatitis patient journey.
Six key steps from ED presentation to early targeted intervention. PenuX integrates at step 3, adding an automated risk score with no additional tests required.
PenuX result available within 4 hours of admission — before Ranson (48 h), BISAP (24 h), and APACHE-II (24 h). No CT scan, no mental-status exam, no second blood draw required.
PenuX returns one of three risk tiers based on the predicted probability of severe AP, aligned with the 2012 Revised Atlanta Classification.
Threshold 0.535 achieves 96.8% sensitivity (TP=566, FN=19) with 38.7% specificity on 5-fold CV (n=722). Clinical context always takes precedence over the score.
Comparing timing, data requirements, and key differentiators of PenuX against established clinical scoring systems for acute pancreatitis.
| Score | When Available | Data Required | Key Limitation | PenuX Advantage |
|---|---|---|---|---|
| Ranson Criteria | 48 hours | 11 criteria across two time-points (admission + 48 h) | Not usable at presentation; requires a second data set at 48 h | PenuX result available at admission |
| BISAP | 24 hours | BUN, SIRS criteria, mental status, age, pleural effusion | Subjective mental-status component; 24 h wait | PenuX needs no clinical exam findings |
| APACHE-II | 24 hours | 12 physiological parameters + Glasgow Coma Scale | Complex, time-consuming; GCS unreliable at admission | PenuX uses routine labs only; fully automated |
| CTSI | CT required | Contrast-enhanced CT (Balthazar grade + % necrosis) | Radiation, contrast risk, costly, delayed availability | PenuX requires no imaging |
| Marshall Score | 24 hours | Respiratory, renal, cardiovascular parameters | Organ-failure focus; less predictive at early presentation | PenuX predicts severity, not only organ failure |
| PenuX-AP-Severity This work | < 4 hours | Routine admission labs (59 features from 19 parameters) | Research prototype; single Chinese cohort n=722; external validation pending | Earliest available · Fully automated · No imaging · No extra tests |
* Comparative figures are from published meta-analyses. PenuX AUROC 0.877 from 5-fold stratified cross-validation on n=722; no external test cohort has been evaluated.
PenuX is designed to augment clinical judgment — never to replace it. These principles are non-negotiable by design.
Patient date-of-birth, MRN, and name are stripped before the API call is formed. The PenuX server never receives or retains identifiable patient data.
Lab values transmitted to the API are used solely for inference. No request payload is persisted. GDPR Article 5(1)(e) storage-limitation principle is upheld by design.
PenuX provides a probability score and a suggested risk tier. All clinical decisions — admission, treatment, escalation — remain exclusively with the treating physician.
The API response lists which fields were used and which are missing, so clinicians can gauge confidence. Missing critical labs are flagged explicitly in every response.
PenuX has not received CE mark, FDA 510(k) clearance, or any other regulatory approval. It must not be used as the sole basis for any clinical decision.
Training data is a single-centre Chinese cohort (n=722). Performance may differ across other ethnic or geographic populations. External validation is actively sought.
"The model augments clinical judgment; it does not replace it. A high PenuX score is a prompt to look harder — not a substitute for the physician's examination, imaging, and holistic evaluation of the patient."
PenuX slots into your existing clinical infrastructure via FHIR R4 or HL7 v2 — no proprietary connector required.
{
"severe_ap_probability": 0.782,
"risk_group": "High",
"recommendation": "Consider ICU/HDU admission. Early aggressive fluid resuscitation.",
"atlanta_classification": "Potentially Severe",
"fields_used": ["wbc", "crp", "creatinine", "glucose", "ldh", "ast"],
"missing_fields": ["bun", "calcium", "albumin"],
"model_version": "1.0.0"
}
We are currently in Stage 0 — building hospital data partnerships before formal validation begins. We are looking for gastroenterology or acute-medicine departments willing to provide de-identified retrospective SAP data, co-develop an IRB protocol, and serve as the first external validation site. Co-authorship offered. Data stays on-site — PenuX supplies the model, pipeline, and analysis.
PI: Netanel Stern — PenuX Research Initiative