OncologyNeurosymbolicTrialMatching Agent

<!--
# COPYRIGHT NOTICE
# This file is part of the "Universal AI Agentic Skills" project.
# Copyright (c) 2026 MD BABU MIA, PhD <md.babu.mia@mssm.edu>
# All Rights Reserved.
#
# This code is proprietary and confidential.
# Unauthorized copying of this file, via any medium is strictly prohibited.
#
# Provenance: Authenticated by MD BABU MIA
-->



<!-- AUTHOR_SIGNATURE: 9a7f3c2e-MD-BABU-MIA-2026-MSSM-SECURE -->

---
name: 'oncology-neurosymbolic-trial-matching'
description: 'Match oncology patients to clinical trials using knowledge-graph retrieval, symbolic eligibility reasoning, specialized agents, conflict resolution, and clinician-auditable evidence.'
measurable_outcome: 'Execute skill workflow successfully with valid output within 15 minutes.'
allowed-tools:
  - read_file
  - run_shell_command
  - web_fetch
---

# Oncology Neuro-Symbolic Trial Matching

## Overview

This skill produces explainable oncology clinical trial candidate matches by combining oncology-specific knowledge representation, symbolic eligibility evaluation, and coordinated specialist-agent review. It is grounded in a prospective evaluation involving 3,804 patients and is designed to support clinician review rather than replace clinical judgment.

## When to Use This Skill

- A clinician or trial-navigation team needs candidate trials for a patient with cancer.
- Eligibility criteria must be evaluated against structured and unstructured patient data.
- Trial retrieval must account for disease ontology, biomarkers, prior therapies, stage, and geography.
- Multiple agents or matching methods disagree and require explicit conflict resolution.
- A trial shortlist needs criterion-level evidence, uncertainty labels, and an audit trail.
- Prospective matching operations need a reproducible human-in-the-loop workflow.

## Core Capabilities

1. **Patient representation:** Normalize diagnosis, histology, stage, molecular findings, treatment history, performance status, organ function, comorbidities, age, and location while preserving source provenance and dates.
2. **Oncology knowledge-graph retrieval:** Connect patient concepts to cancer types, variants, therapies, trial interventions, sites, and eligibility concepts using ontology-aware expansion without treating inferred relationships as confirmed facts.
3. **Symbolic eligibility reasoning:** Translate inclusion and exclusion criteria into explicit rules and classify each criterion as met, not met, unknown, or not applicable.
4. **Specialist-agent decomposition:** Assign focused review roles for disease fit, biomarker fit, treatment history, laboratory and clinical constraints, temporal logic, and geographic feasibility.
5. **Evidence-grounded synthesis:** Require every match or exclusion claim to cite the relevant patient datum and trial criterion; never fill missing clinical facts by assumption.
6. **Conflict resolution:** Compare agent conclusions, expose contradictory interpretations, apply precedence to directly documented evidence, and route unresolved conflicts for clinician review.
7. **Ranking with uncertainty:** Rank candidates by eligibility confidence, clinical relevance, recruiting status, location, and unresolved-data burden without presenting the ranking as a treatment recommendation.
8. **Clinician review package:** Produce a compact, auditable report containing candidate trials, criterion-level decisions, missing-data requests, conflicts, provenance, and review timestamps.
9. **Prospective 3,804-patient benchmark case:** Use the 2026 PubMed-indexed prospective evaluation in 3,804 oncology patients as a benchmark case for neuro-symbolic multi-agent oncology trial matching grounded in an oncology-specific knowledge graph, with criterion-level reasoning, explicit conflict resolution, clinician-auditable evidence, and prospective performance reporting; do not infer unstated benchmark or performance values.
10. **Prospective evaluation workflow controls:** Normalize trial eligibility concepts through the oncology knowledge graph before symbolic evaluation, use multi-agent adjudication with documented conflict resolution, preserve criterion-level evidence for every match decision, and report trial-matching metrics without inventing benchmarks not stated in the cited prospective study.
11. **Clinician-auditable prospective outputs:** For prospective oncology trial matching, ground candidate retrieval in the oncology-specific knowledge graph, run neuro-symbolic eligibility checks, resolve multi-agent conflicts with documented rationale, and emit patient-level evidence trails with `MET`, `NOT`, or `UNKNOWN` outputs for clinician audit.
12. **Prospective evaluation audit workflow:** For workflows modeled on the 3,804-patient prospective evaluation, display knowledge-graph-grounded retrieval paths, criterion-level reasoning, conflict-resolution outcomes, throughput fields such as patient volume and reviewed match counts, and evidence links in a clinician audit view; leave any throughput value unknown unless it is explicitly reported by the source data.
13. **Prospective validation pattern:** Treat the 2026 3,804-patient evaluation as a validation pattern for combining oncology-specific knowledge graph retrieval, neuro-symbolic eligibility checks, multi-agent review roles, criterion-level audit trails, conflict handling, and clinician-facing evidence summaries in trial matching; do not invent performance claims beyond the source.
14. **Prospective monitoring loop:** For 3,804-patient prospective-style deployments, construct and refresh the oncology-specific knowledge graph as the retrieval and harmonization layer, keep extraction, reasoning, prioritization, and review agents role-separated, apply criterion-level symbolic rules, adjudicate conflicts explicitly, display supporting evidence for each decision, and monitor performance only with source-supported metrics.
15. **Prospective evidence anchor:** Use the PubMed-indexed 2026 prospective evaluation of neuro-symbolic, multi-agent oncology trial matching in 3,804 patients as evidence supporting KG-backed eligibility reasoning, criterion-level auditability, explicit conflict resolution, clinician review workflows, and prospective performance monitoring; do not extrapolate performance claims beyond the reported finding.
16. **Prospective human-review checkpoints:** In workflows modeled on the 3,804-patient prospective evaluation, keep retrieval, eligibility reasoning, conflict adjudication, and clinical review roles separate; show knowledge-graph eligibility paths, criterion-level evidence, source-supported prospective metric fields, and explicit human review checkpoints before any match is treated as actionable.
17. **Prospective clinician-review metrics guardrail:** When adapting the 3,804-patient prospective evaluation pattern, report only source-supported clinician review metrics, preserve the oncology knowledge graph grounding behind each candidate match, separate retrieval, eligibility reasoning, adjudication, and review agents, document conflict-resolution outcomes, and attach auditable evidence for every neuro-symbolic eligibility conclusion.
18. **Prospective neuro-symbolic evidence model:** Apply the 3,804-patient prospective evaluation pattern by grounding patient-trial concept alignment in the oncology-specific knowledge graph, running multi-agent symbolic eligibility reasoning, preserving criterion-level evidence for each met, not met, or unknown decision, documenting conflict resolution, and keeping the final review clinician-auditable.
19. **Prospective confidence reporting:** Use PubMed 42004487 as evidence that neuro-symbolic, multi-agent oncology trial matching can pair oncology-specific knowledge graph retrieval with criterion-level reasoning, conflict resolution, confidence reporting, and clinician-auditable explanations in a prospective 3,804-patient evaluation; do not infer unstated accuracy, enrollment, or outcome benchmarks.
20. **Prospective clinician-review evaluation anchor:** For PubMed 42004487-style prospective workflows, align retrieval through the oncology-specific knowledge graph, run criterion-level symbolic eligibility checks, resolve specialist-agent conflicts with documented rationale, preserve audit trails for each patient-trial decision, and report clinician review metrics only when the source data explicitly supports them.
21. **KG-backed prospective reference pattern:** Use PubMed 42004487's 2026 prospective evaluation of 3,804 oncology patients as a reference pattern for oncology-specific KG-backed eligibility reasoning, criterion-level evidence capture, specialist-agent conflict resolution, prospective performance measurement, and clinician-auditable outputs; keep quantitative performance values limited to what the source explicitly reports.
22. **Human-in-the-loop prospective adjudication:** For prospective oncology trial matching modeled on PubMed 42004487, use the oncology knowledge graph for patient-trial retrieval and concept normalization, apply symbolic eligibility logic at the criterion level, reconcile specialist-agent disagreements through documented conflict resolution, preserve supporting evidence, and require qualified human final adjudication before treating a match as actionable.
23. **Prospective operational coverage review:** In 3,804-patient prospective-style neuro-symbolic multi-agent trial matching, document knowledge-graph grounding, criterion-level reasoning, conflict-resolution rationale, coverage gaps from missing patient data, unmapped concepts, or unavailable trial details, clinician audit-trail fields, and operational metrics only when those metrics are explicitly source-supported.
24. **Prospective 3,804-patient validation pattern:** Model validation on the PubMed 42004487 prospective evaluation by keeping neuro-symbolic agents role-separated for retrieval, eligibility reasoning, conflict adjudication, and clinician review; ground patient-trial alignment in the oncology knowledge graph, preserve criterion-level eligibility rationales and audit trails, close unresolved cases through clinician review loops, and report only source-supported prospective performance fields.
25. **Prospective evidence case:** Use PubMed 42004487 as the core evidence case for oncology neuro-symbolic multi-agent trial matching in 3,804 patients: ground retrieval in the oncology-specific knowledge graph, apply symbolic eligibility reasoning, expose criterion-level explanations, document conflict resolution and audit trails, keep clinician review in the loop, and report only prospective matching metrics present in the source.

## Inputs / Outputs

### Inputs

- Patient oncology summary or structured record with dates and source references.
- Current trial records containing identifiers, recruiting status, sites, interventions, and full eligibility criteria.
- Terminology mappings or an oncology knowledge graph for diagnoses, biomarkers, drugs, and procedures.
- Operational constraints such as travel radius, age range, trial phase, and preferred institutions.
- Optional clinician-provided interpretations or rules for ambiguous eligibility language.

### Outputs

- A ranked candidate-trial table with trial identifier, title, site, recruiting status, and match rationale.
- A criterion-level eligibility matrix labeled met, not met, unknown, or not applicable.
- Evidence links from each decision to patient facts and source trial text.
- A list of missing or stale data that could change eligibility.
- An agent-conflict log with the resolution, rationale, and items requiring clinician adjudication.
- A final status for each trial: likely eligible, potentially eligible pending data, likely ineligible, or insufficient evidence.
- An auditable run summary recording data versions, retrieval time, rules applied, and reviewer sign-off status.

Do not contact trial sites, alter patient care, or represent eligibility as confirmed. Verify current recruiting status and protocol details from authoritative trial sources, and require qualified clinician or trial-site confirmation before enrollment decisions.

## References

- Loaiza-Bonilla A, Yost C, Kurnaz S, Tuysuz E, Thaker NG. “Transforming oncology clinical trial matching through neuro-symbolic, multi-agent AI and an oncology-specific knowledge graph: a prospective evaluation in 3804 patients.” *ESMO Real World Data and Digital Oncology*. 2026 Jun. PubMed: https://pubmed.ncbi.nlm.nih.gov/42004487/