Spaces:
Sleeping
Sleeping
File size: 12,739 Bytes
59abb4f | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 | """A2A (Agent-to-Agent) orchestration workflow β state machine for the recruitment pipeline.
Every inter-agent message carries a SHARP Extension Spec context envelope:
sharp_version, patient_context (id, fhir_ref, fhir_base, tenant_id, session_id),
data_classification, baa_in_scope, consent_status
"""
import uuid
import time
from datetime import datetime
from enum import Enum
from typing import Any
from fhir_adapter import get_patient_profile, get_mock_fhir_patient, build_patient_profile
from clinicaltrials_api import search_trials_sync, get_trial_details_sync
from matching_engine import get_criteria_for_trial, score_patient_for_trial, match_patient_to_trials
from llm_client import generate_outreach_message, summarize_trial
from fhir_server import build_sharp_context, get_live_patient_profile
import consent_agent
class WorkflowState(str, Enum):
PENDING = "PENDING"
INGESTING = "INGESTING"
PARSING_PROTOCOL = "PARSING_PROTOCOL"
MATCHING = "MATCHING"
SCORING = "SCORING"
RECRUITING = "RECRUITING"
COMPLETED = "COMPLETED"
FAILED = "FAILED"
# In-memory workflow store (production: use Redis or Neo4j)
_workflows: dict[str, dict] = {}
def _emit_event(workflow_id: str, state: WorkflowState, message: str, data: Any = None):
workflow = _workflows[workflow_id]
event = {
"state": state,
"message": message,
"timestamp": datetime.utcnow().isoformat(),
"data": data,
# SHARP envelope on every event so downstream agents have full context
"sharp_context": workflow.get("sharp_context", {}),
}
workflow["events"].append(event)
workflow["current_state"] = state
workflow["updated_at"] = datetime.utcnow().isoformat()
print(f"[A2A:{workflow_id[:8]}] {state} β {message}")
# ββ Sub-agents ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
def _agent_ingest_patient(workflow_id: str, patient_id: str) -> dict:
"""Sub-agent: Ingest and validate patient FHIR data."""
_emit_event(workflow_id, WorkflowState.INGESTING, f"Ingesting FHIR R4 data for patient {patient_id}")
time.sleep(0.3) # Simulate async data fetch
fhir_patient = get_mock_fhir_patient(patient_id)
if not fhir_patient:
raise ValueError(f"Patient {patient_id} not found in FHIR registry")
profile = build_patient_profile(fhir_patient)
_emit_event(workflow_id, WorkflowState.INGESTING,
f"FHIR data loaded: {len(fhir_patient.conditions)} conditions, {len(fhir_patient.medications)} medications",
{"profile": profile})
return profile
def _agent_parse_protocol(workflow_id: str, nct_id: str | None, condition: str) -> tuple[list[dict], dict]:
"""Sub-agent: Parse trial protocol and extract criteria."""
_emit_event(workflow_id, WorkflowState.PARSING_PROTOCOL,
f"Parsing trial protocols for condition: {condition}")
time.sleep(0.5)
if nct_id:
trials = [get_trial_details_sync(nct_id)]
trials = [t for t in trials if t]
else:
trials = search_trials_sync(condition, page_size=8)
if not trials:
raise ValueError(f"No trials found for condition: {condition}")
# Parse criteria for each trial using LLM
parsed_trials = []
for trial in trials[:5]: # Limit to avoid timeout
criteria = get_criteria_for_trial(trial)
parsed_trials.append({**trial, "parsed_criteria": criteria})
summary = summarize_trial(trials[0]) if trials else ""
_emit_event(workflow_id, WorkflowState.PARSING_PROTOCOL,
f"Parsed {len(parsed_trials)} trial protocols",
{"trial_count": len(parsed_trials), "protocol_summary": summary})
return parsed_trials, {"summary": summary}
def _agent_match(workflow_id: str, patient_profile: dict, trials: list[dict]) -> list[dict]:
"""Sub-agent: Semantic matching of patient to trials."""
_emit_event(workflow_id, WorkflowState.MATCHING,
f"Running semantic matching for patient {patient_profile['patient_id']} against {len(trials)} trials")
time.sleep(0.3)
candidates = []
for trial in trials:
score_result = score_patient_for_trial(patient_profile["patient_id"], trial)
candidates.append({
**trial,
"match_score": score_result.get("overall_score", 0.0),
"eligible": score_result.get("eligible", False),
"inclusion_results": score_result.get("inclusion_results", []),
"exclusion_results": score_result.get("exclusion_results", []),
"match_summary": score_result.get("summary", ""),
"risk_flags": score_result.get("risk_flags", []),
})
candidates.sort(key=lambda x: x["match_score"], reverse=True)
eligible = [c for c in candidates if c["eligible"]]
_emit_event(workflow_id, WorkflowState.MATCHING,
f"Matching complete: {len(eligible)}/{len(candidates)} trials eligible",
{"eligible_count": len(eligible), "top_score": candidates[0]["match_score"] if candidates else 0})
return candidates
def _agent_score(workflow_id: str, candidates: list[dict], patient_profile: dict) -> list[dict]:
"""Sub-agent: Predictive screening scoring with risk flags."""
_emit_event(workflow_id, WorkflowState.SCORING, "Running predictive screening analysis")
time.sleep(0.2)
for candidate in candidates:
flags = candidate.get("risk_flags", [])
# Add distance risk flag if no nearby sites
locs = candidate.get("locations", [])
if not locs:
flags.append("No site location data available")
# Add data completeness flag
if not patient_profile.get("biomarkers"):
flags.append("Biomarker data incomplete β may affect screening")
candidate["risk_flags"] = flags
candidate["screening_priority"] = (
"HIGH" if candidate["match_score"] >= 0.8
else "MEDIUM" if candidate["match_score"] >= 0.5
else "LOW"
)
_emit_event(workflow_id, WorkflowState.SCORING,
"Screening scoring complete",
{"high_priority": sum(1 for c in candidates if c.get("screening_priority") == "HIGH")})
return candidates
def _agent_recruit(workflow_id: str, candidates: list[dict], patient_profile: dict) -> list[dict]:
"""Sub-agent: Generate recruitment outreach for eligible candidates."""
_emit_event(workflow_id, WorkflowState.RECRUITING, "Generating personalized recruitment communications")
eligible = [c for c in candidates if c.get("eligible")][:3]
recruitment_records = []
for trial in eligible:
try:
outreach = generate_outreach_message(patient_profile, trial, "patient_email")
pcp_letter = generate_outreach_message(patient_profile, trial, "pcp_letter")
# A2A handoff β consent agent (SHARP envelope attached)
consent_task = {
"task_id": f"consent_{workflow_id}_{trial.get('nct_id','')}",
"type": "CONSENT_REQUEST",
"payload": {
"patient_id": patient_profile.get("patient_id", ""),
"nct_id": trial.get("nct_id", ""),
"trial_title": trial.get("title", ""),
"match_score": trial.get("match_score", 0.0),
},
"sharp_context": _workflows[workflow_id].get("sharp_context", {}),
}
consent_result = consent_agent.receive_a2a_task(consent_task)
recruitment_records.append({
"nct_id": trial.get("nct_id", ""),
"trial_title": trial.get("title", ""),
"match_score": trial.get("match_score", 0.0),
"patient_email": outreach,
"pcp_letter": pcp_letter,
"status": "PENDING",
"consent_id": consent_result.get("consent_id"),
"consent_status": consent_result.get("status", "PENDING"),
"created_at": datetime.utcnow().isoformat(),
})
except Exception as e:
recruitment_records.append({
"nct_id": trial.get("nct_id", ""),
"trial_title": trial.get("title", ""),
"error": str(e),
"status": "ERROR",
})
_emit_event(workflow_id, WorkflowState.RECRUITING,
f"Generated outreach for {len(recruitment_records)} trials",
{"record_count": len(recruitment_records)})
return recruitment_records
# ββ Public API βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
def start_pipeline(
patient_id: str,
nct_id: str | None = None,
condition: str | None = None,
fhir_token: str | None = None,
fhir_base_url: str | None = None,
session_id: str | None = None,
) -> str:
"""Start the A2A pipeline and return a workflow_id."""
workflow_id = str(uuid.uuid4())
sharp_ctx = build_sharp_context(
patient_id=patient_id,
fhir_ref=f"Patient/{patient_id}",
session_id=session_id or workflow_id,
)
if fhir_token:
sharp_ctx["fhir_token"] = fhir_token
if fhir_base_url:
sharp_ctx["patient_context"]["fhir_base"] = fhir_base_url
_workflows[workflow_id] = {
"workflow_id": workflow_id,
"patient_id": patient_id,
"nct_id": nct_id,
"condition": condition,
"current_state": WorkflowState.PENDING,
"events": [],
"result": None,
"sharp_context": sharp_ctx,
"created_at": datetime.utcnow().isoformat(),
"updated_at": datetime.utcnow().isoformat(),
}
return workflow_id
def run_pipeline(workflow_id: str) -> dict:
"""Execute the full A2A pipeline synchronously."""
workflow = _workflows.get(workflow_id)
if not workflow:
raise ValueError(f"Workflow {workflow_id} not found")
patient_id = workflow["patient_id"]
nct_id = workflow.get("nct_id")
condition = workflow.get("condition")
try:
# Agent 1: Ingest FHIR patient data
patient_profile = _agent_ingest_patient(workflow_id, patient_id)
# Infer condition
if not condition and patient_profile.get("diagnosis_names"):
condition = patient_profile["diagnosis_names"][0]
elif not condition:
condition = "cancer"
# Agent 2: Parse trial protocols
trials, protocol_meta = _agent_parse_protocol(workflow_id, nct_id, condition)
# Agent 3: Semantic matching
candidates = _agent_match(workflow_id, patient_profile, trials)
# Agent 4: Predictive scoring
candidates = _agent_score(workflow_id, candidates, patient_profile)
# Agent 5: Recruitment communication
recruitment_records = _agent_recruit(workflow_id, candidates, patient_profile)
result = {
"patient_profile": patient_profile,
"matched_trials": candidates,
"recruitment_records": recruitment_records,
"protocol_summary": protocol_meta.get("summary", ""),
"total_trials_evaluated": len(trials),
"eligible_trials": sum(1 for c in candidates if c.get("eligible")),
}
workflow["result"] = result
_emit_event(workflow_id, WorkflowState.COMPLETED,
f"Pipeline complete: {result['eligible_trials']} eligible trials found", result)
except Exception as e:
_emit_event(workflow_id, WorkflowState.FAILED, f"Pipeline failed: {str(e)}")
workflow["error"] = str(e)
return _workflows[workflow_id]
def get_workflow_status(workflow_id: str) -> dict:
workflow = _workflows.get(workflow_id)
if not workflow:
return {"error": "Workflow not found"}
return {
"workflow_id": workflow_id,
"current_state": workflow["current_state"],
"events": workflow["events"][-10:], # Last 10 events
"result": workflow.get("result"),
"error": workflow.get("error"),
"created_at": workflow["created_at"],
"updated_at": workflow["updated_at"],
}
def list_workflows() -> list[dict]:
return [
{
"workflow_id": wf["workflow_id"],
"patient_id": wf["patient_id"],
"current_state": wf["current_state"],
"created_at": wf["created_at"],
}
for wf in _workflows.values()
]
|