@traumatrial/match

Real-time trauma trial eligibility matching. TypeScript port of the Python engine — same operators, same trace format, same confidence rubric. Includes a NEMSIS v3.5 ePCR adapter.

npm install @traumatrial/match

The matching engine has zero runtime dependencies. The optional NEMSIS adapter at @traumatrial/match/nemsis requires fast-xml-parser (declared as a peer dep).

Match a patient

import { match, matchAll, type Patient, type Trial } from "@traumatrial/match";

const patient: Patient = {
  patient_id: "P-001",
  age_years: 34, sex: "M",
  gcs: 7, sbp_mmhg: 82, hr_bpm: 128,
  mechanism: "blunt_mvc",
  trauma_activation_level: 1, eta_minutes: 4,
  pregnancy_status: "not_applicable",
  anticoagulant_use: false,
  presumed_tbi: true, presumed_hemorrhage: true,
  presumed_intracranial_hemorrhage: false,
  spinal_injury_suspected: false,
};

const trial: Trial = {
  trial_id: "NCT05638581",
  short_name: "TROOP",
  title: "Trauma Resuscitation With Low-Titer Group O Whole Blood",
  requires_efic: true, phase: "3",
  inclusion: [
    { field: "age_years", op: "gte", value: 15, hard: true },
    { field: "presumed_hemorrhage", op: "eq", value: true, hard: true },
  ],
  exclusion: [
    { field: "pregnancy_status", op: "in",
      value: ["pregnant", "unknown_could_be_pregnant"], hard: true },
  ],
};

const result = match(patient, trial);
// { eligible: true, confidence: 1.0, trace: [...] }

const ranked = matchAll(patient, [trial, ...]);
// eligible-first, then confidence desc

Eight operators: eq, ne, gte, lte, gt, lt, in, not_in. in/not_in take a list value; the others take a scalar.

Confidence rubric. Any hard inclusion missed or hard exclusion hit → eligible=false, confidence=0.0. Otherwise eligible; confidence = soft_inclusion_hits / soft_inclusion_total (or 1.0 if no soft inclusions).

Convert a NEMSIS v3.5 ePCR

npm install @traumatrial/match fast-xml-parser

import { fromNemsisXml } from "@traumatrial/match/nemsis";
import { matchAll } from "@traumatrial/match";

const xml = `<?xml version="1.0"?>
<EMSDataSet xmlns="http://www.nemsis.org"> … </EMSDataSet>`;

const { patient, trace } = fromNemsisXml(xml);
// trace.extractions: [{ field, source: "extracted"|"inferred"|"defaulted"|"skipped",
//                       value, nemsis_path, raw, notes }, ...]

const results = matchAll(patient, myTrials);

The adapter pulls ~10 high-signal eFields (ePatient.13, ePatient.15, eVitals.06/10/23, eSituation.02, eHistory.06/16) and surfaces every Patient field with a one-line trace explaining where it came from. Fields without a clean NEMSIS source (e.g. presumed_tbi) are inferred from physiology + mechanism rules; the trace records the rule.

v0 mapping; not a clinical-grade extractor. Core v3.5 only, no state extensions, no XSD validation.

What this is NOT

• Not a clinical decision-support system.
• Not regulated, certified, or BAA-able.
• Not a substitute for a research coordinator's clinical judgment.

It is a structured, testable, transparent starting point for talking about how trauma trial enrollment could be automated.

License

MIT — see LICENSE.