Clinical risk scores improved by separating unknowns from decision rules
Unknown is not normal: separating language-model extraction from rule-based decision logic for clinical risk scores
Computation and LanguageArtificial Intelligence
Summary
Clinical risk scores help doctors decide how serious a patient’s condition might be based on medical notes. The authors found that if an AI treats missing information as if it’s normal, it can make mistakes and miss patients who need urgent help. They showed that using AI to flag unknown information separately and then having clear rules to handle those unknowns leads to more accurate scores with fewer unnecessary questions. This approach works well even with small local models and noisy data.
What this means in practice
- •For hospital data teams: Build clinical decision support tools that explicitly handle unknown data to improve risk scoring accuracy from patient notes.
- •For medical software developers: Create interactive clinical note systems that minimize irrelevant questions by separating data extraction from scoring rules.
Authors
Nicolás Vera Zúñiga
Abstract
Large language models (LLMs) are increasingly used to compute clinical risk scores from free-text notes. Notes are often incomplete, and treating undocumented findings as normal can silently misclassify patients. We test whether separating three-state extraction (present, absent or unknown, by an LLM) from decision logic (deterministic code computing score bounds over unknown inputs) lets a system ask only questions that can change the decision. On 1,200 synthetic emergency cases across six calculators (HEART, CURB-65, qSOFA, PERC, Wells, Cockcroft-Gault), with a simulated clinician answering questions, we compared this bounds policy with asking for every missing input, a missing-equals-normal schema, and an end-to-end LLM agent (Claude Opus 5.5). With Claude Haiku 4.5 as extractor, the bounds policy matched ask-all accuracy (99.4% vs 99.4%) with half the questions (0.92 vs 1.78 per case) and no irrelevant ones. Treating missing as normal dropped accuracy to 91.2% and under-triaged 8.5% of patients (95% CI 7.1-10.2), and under-triage persisted under messy notes and a noisy clinician. The agent was equally accurate under ideal conditions (99.6%) but 9.5% of its questions were irrelevant; with a noisy clinician it was less accurate than the bounds policy (83.5% vs 87.0%, p<0.001) and committed prematurely in 2.7% of cases (bounds: 0%). A 9B local model as extractor reached oracle-level accuracy (99.8%). In 584 real case reports from MedCalc-Bench, only 52% contained enough information to determine the category (HEART 13%). Routing decisions through code that reasons explicitly about unknowns avoids premature commitment and irrelevant questions, halves the questions asked, and works with small local models.