临床试验设计可行性研究

WORKFLOW_DETAILS.md

EGFR L858R+ NSCLC Phase 1/2 Trial

Full Python example using ToolUniverse to assess trial feasibility across all 6 research paths.

from tooluniverse import ToolUniverse

tu = ToolUniverse()
tu.load_tools()

# ============================================================================
# PATH 1: PATIENT POPULATION SIZING
# ============================================================================

# Step 1.1: Get disease prevalence
disease_info = tu.tools.OpenTargets_get_disease_id_description_by_name(
    diseaseName="non-small cell lung cancer"
)
efo_id = disease_info['data']['search']['hits'][0]['id']

# Get HPO phenotype annotations (no prevalence figures)
phenotypes = tu.tools.OpenTargets_get_associated_phenotypes_by_disease_efoId(
    efoId=efo_id
)
# Note: Prevalence must come from literature (PubMed)

# Step 1.2: Estimate EGFR mutation prevalence
egfr_variants = tu.tools.ClinVar_search_variants(
    gene="EGFR",
    significance="pathogenic,likely_pathogenic"
)

# Filter to L858R specifically
l858r_variants = [v for v in egfr_variants['data']
                  if 'L858R' in v.get('name', '')]

# Also check population databases for allele frequency
# gnomad_search_variants takes a variant ID / rsID / ClinVar ID (not a gene symbol).
# L858R is rs121434568; a somatic hotspot, so "No data returned" means absent from gnomAD.
gnomad_l858r = tu.tools.gnomad_search_variants(
    query="rs121434568",
    dataset="gnomad_r4"
)
# If found, pass the returned variant_id to gnomad_get_variant for allele frequencies

# Step 1.3: Search literature for epidemiology
epi_papers = tu.tools.PubMed_search_articles(
    query="EGFR L858R prevalence non-small cell lung cancer epidemiology",
    max_results=20
)
# Extract prevalence estimates from recent papers

# ============================================================================
# PATH 2: BIOMARKER PREVALENCE & TESTING
# ============================================================================

# Step 2.1: Find FDA-approved CDx tests
cdx_search = tu.tools.PubMed_search_articles(
    query="FDA approved companion diagnostic EGFR L858R",
    max_results=10
)

# Step 2.2: Literature on EGFR testing in clinical practice
testing_papers = tu.tools.PubMed_search_articles(
    query="EGFR mutation testing guidelines NCCN turnaround time",
    max_results=15
)

# ============================================================================
# PATH 3: COMPARATOR SELECTION
# ============================================================================

# Step 3.1: Find current standard of care (osimertinib)
soc_drug = "osimertinib"

soc_info = tu.tools.drugbank_get_drug_basic_info_by_drug_name_or_id(
    query=soc_drug
)

soc_indications = tu.tools.drugbank_get_indications_by_drug_name_or_drugbank_id(
    query=soc_drug
)

soc_pharmacology = tu.tools.drugbank_get_pharmacology_by_drug_name_or_drugbank_id(
    query=soc_drug
)

# Step 3.2: Check FDA Orange Book for approved generics
orange_book = tu.tools.FDA_OrangeBook_search_drug(
    generic_name=soc_drug
)

# Step 3.3: Find FDA approval details
fda_approval = tu.tools.OpenFDA_get_approval_history(
    drug_name=soc_drug
)

# ============================================================================
# PATH 4: ENDPOINT SELECTION
# ============================================================================

# Step 4.1: Search for precedent Phase 2 trials in EGFR+ NSCLC
precedent_trials = tu.tools.ClinicalTrials_search_studies(
    condition="EGFR positive non-small cell lung cancer",
    filter_phase="PHASE2",
    filter_status="COMPLETED",
    page_size=20
)
precedent_studies = precedent_trials['data']['studies']

# Analyze which primary endpoints were used (ORR, PFS, etc.).
# Search results carry no outcome fields; primary outcomes come from ClinicalTrials_get_study.
orr_trials = []
for t in precedent_studies:
    details = tu.tools.ClinicalTrials_get_study(nct_id=t['nct_id'])
    measures = ' '.join(o.get('measure', '') for o in details['data'].get('primary_outcomes', []))
    if 'response rate' in measures.lower():
        orr_trials.append(t)

# Step 4.2: Find FDA approvals using ORR as primary endpoint
orr_approvals = tu.tools.PubMed_search_articles(
    query="FDA approval objective response rate NSCLC accelerated approval",
    max_results=30
)

# Step 4.3: Get detailed trial results for sample size justification
for trial in precedent_studies[:5]:
    nct_id = trial.get('nct_id')
    trial_details = tu.tools.ClinicalTrials_get_study(
        nct_id=nct_id
    )
    # Extract: ORR, n, confidence intervals

# ============================================================================
# PATH 5: SAFETY ENDPOINTS & MONITORING
# ============================================================================

# Step 5.1: Get mechanism-based toxicity from drug class
class_drug = "erlotinib"  # Example EGFR TKI for class effect reference

class_safety = tu.tools.drugbank_get_pharmacology_by_drug_name_or_drugbank_id(
    query=class_drug
)

class_warnings = tu.tools.FDA_get_warnings_and_cautions_by_drug_name(
    drug_name=class_drug
)

# Step 5.2: FAERS data for real-world adverse events
faers_egfr_tki = tu.tools.FAERS_search_adverse_event_reports(
    medicinalproduct="ERLOTINIB",
    limit=100  # capped at 100 per request; read total_available for the full count
)

# Summarize top adverse events
ae_summary = tu.tools.FAERS_count_reactions_by_drug_event(
    medicinalproduct="ERLOTINIB"
)

# Step 5.3: Search for DLT definitions in similar trials
dlt_papers = tu.tools.PubMed_search_articles(
    query="dose limiting toxicity Phase 1 EGFR inhibitor definition",
    max_results=20
)

# ============================================================================
# PATH 6: REGULATORY PATHWAY
# ============================================================================

# Step 6.1: Search for breakthrough therapy designations in NSCLC
breakthrough_search = tu.tools.PubMed_search_articles(
    query="FDA breakthrough therapy designation NSCLC EGFR mutation",
    max_results=20
)

# Step 6.2: Check if indication qualifies for orphan drug status
us_nsclc_annual = 200000  # From epidemiology data
l858r_prevalence = 0.45 * 0.15  # 45% of EGFR+ (15% of NSCLC)
l858r_annual_us = us_nsclc_annual * l858r_prevalence  # ~13,500/year
# Note: Orphan requires <200,000 total prevalence; may not qualify if prevalent

# Step 6.3: Find relevant FDA guidance documents
fda_guidance_search = tu.tools.PubMed_search_articles(
    query="FDA guidance clinical trial endpoints oncology non-small cell lung cancer",
    max_results=15
)

# ============================================================================
# COMPILE FEASIBILITY REPORT
# ============================================================================

feasibility_scores = {
    'patient_availability': 8,  # 8/10 based on 13,500 patients/year, good access
    'endpoint_precedent': 9,    # 9/10 ORR widely accepted
    'regulatory_clarity': 7,    # 7/10 breakthrough possible, single-arm needs FDA input
    'comparator_feasibility': 9, # 9/10 osimertinib available, efficacy data clear
    'safety_monitoring': 8      # 8/10 EGFR TKI class effects well-characterized
}

weights = {
    'patient_availability': 0.30,
    'endpoint_precedent': 0.25,
    'regulatory_clarity': 0.20,
    'comparator_feasibility': 0.15,
    'safety_monitoring': 0.10
}

overall_score = sum(feasibility_scores[k] * weights[k] * 10 for k in weights.keys())
# overall_score = 81/100 -> HIGH feasibility

print(f"Feasibility Score: {overall_score}/100 - HIGH")
print("Recommendation: RECOMMEND PROCEED to protocol development")

---

Example Use Cases

Use Case 1: Biomarker-Selected Oncology Trial

Query: "Assess feasibility of Phase 2 trial for EGFR L858R+ NSCLC, ORR primary endpoint"

Workflow:

1. Disease prevalence: 200K NSCLC/year x 15% EGFR+ = 30K
2. Biomarker: L858R is 45% of EGFR+ -> 13.5K/year
3. Eligible: 60% -> 8K/year
4. Endpoint: ORR accepted (osimertinib precedent)
5. Comparator: Osimertinib (ORR 57%, generic available)
6. Feasibility: HIGH (82/100) -> RECOMMEND PROCEED

Use Case 2: Rare Disease Trial

Query: "Feasibility of trial in Niemann-Pick Type C (prevalence 1:120,000)"

Workflow:

1. US prevalence: ~2,750 patients total, ~25 new cases/year
2. Endpoint challenge: No validated clinical outcome
3. Orphan drug: QUALIFIED (7-year exclusivity)
4. Comparator: No approved drugs -> single-arm feasible
5. Enrollment: Multi-year, need ALL US centers
6. Feasibility: MODERATE (58/100) -> CONDITIONAL GO (requires patient registry partnership)

Use Case 3: Superiority Trial vs. Standard of Care

Query: "Phase 2b design for new checkpoint inhibitor vs. pembrolizumab in PD-L1 high NSCLC"

Workflow:

1. Patient availability: 40K PD-L1 high NSCLC/year (HIGH)
2. Endpoint: ORR for Phase 2b, plan OS for Phase 3
3. Comparator: Pembrolizumab (ORR 45%, PFS 10mo) - readily available
4. Design: Randomized 1:1, N=120 (60/arm) for 20% ORR improvement
5. Feasibility: HIGH (78/100) -> RECOMMEND PROCEED

Use Case 4: Non-Inferiority Trial

Query: "Non-inferiority trial for oral anticoagulant vs. warfarin"

Workflow:

1. Patient availability: 2M AFib patients, 600K on warfarin (HIGH)
2. Endpoint: Stroke/SE (FDA-accepted, but requires large N)
3. Non-inferiority margin: HR <1.5 (FDA guidance)
4. Sample size: N=5,000+ for 90% power -> LARGE trial
5. Comparator: Warfarin generic, INR monitoring standard
6. Feasibility: MODERATE (65/100) - large N drives cost and timeline

Use Case 5: Basket Trial (Multiple Cancers, One Biomarker)

Query: "Basket trial for NTRK fusion+ solid tumors (15 histologies)"

Workflow:

1. Patient availability: NTRK fusions rare (<1% across cancers) -> Broad screening
2. Biomarker testing: NGS required (FDA-approved FoundationOne CDx)
3. Endpoint: ORR (precedent: larotrectinib approval, ORR 75%, n=55)
4. Design: Single-arm, N=15-20 per histology x 5-10 histologies
5. Regulatory: Tissue-agnostic approval precedent (pembrolizumab MSI-H)
6. Feasibility: MODERATE (62/100) - enrollment slow but feasible with broad screening