跳轉至

trial — 病人對臨床試驗配對

TrialGPT 三階段流程的端到端簡化實作(Jin et al. Nat Commun 15, 9074, 2024):

  1. Retrieval(檢索) — 自病人摘要對候選試驗進行關鍵字比對。
  2. Matching(配對) — 逐條條件評估收錄與否,並附理由說明。
  3. Ranking(排名) — 將各條件分數彙整為單一的試驗等級排名。

用法

mrnavax trial --patient mrnavax/examples/patient_summary.txt \
              --trials mrnavax/examples/trials.jsonl \
              --top-k 5 --backend mock

輸入

  • 病人摘要 — 自由文字,英文臨床筆記效果最佳。
  • 臨床試驗 — JSONL,每行一筆試驗:
{
  "nct_id": "NCT05933577",
  "title": "INTerpath-001: Personalized mRNA-4157 + Pembrolizumab",
  "condition": "Stage IIB-IV melanoma",
  "phase": "3",
  "inclusion": ["Completely resected melanoma", "ECOG 0 or 1"],
  "exclusion": ["Active autoimmune disease"],
  "biomarkers": ["BRAF V600E", "BRAF V600K"]
}

輸出 schema

{
  "ranked": [
    {
      "nct_id": "NCT05933577",
      "title": "INTerpath-001: ...",
      "score": 1.0,
      "eligibility_pct": 100.0,
      "n_met": 4,
      "n_total": 4,
      "reasons": ["Completely resected melanoma", "ECOG 0 or 1"]
    }
  ],
  "n_candidates_screened": 5
}

score = eligibility_pct / 100 − 0.2 × (符合的排除條件數)。分數最高的試驗即為建議配對。