Drosophila

Clinical trial matching for oncology, powered by SMART on FHIR.

Drosophila automatically identifies clinical trials a patient may be eligible for by extracting structured data from the EHR and comparing it against trial eligibility criteria from ClinicalTrials.gov.

How It Works

Drosophila follows a two-stage EHR integration pattern:

1. CDS Hooks — A patient-view hook fires when a clinician opens a patient chart. Drosophila returns a CDS card summarizing potential trial matches with a SMART App Link to launch the full app.

2. SMART on FHIR App — The clinician clicks the card to launch the Drosophila app inside the EHR. The app authenticates via SMART, extracts patient data, searches ClinicalTrials.gov, and displays ranked trial matches.

EHR opens chart
    │
    ▼
CDS Hooks patient-view
    │
    ▼
CDS Card: "3 trials found, 2 likely eligible"
    │  [click: SMART App Link]
    ▼
SMART on FHIR Launch
    │
    ├─► Extract patient data (FHIR R4)
    │     Patient, Condition, Observation,
    │     MedicationRequest, Procedure
    │
    ├─► Search ClinicalTrials.gov (API v2)
    │     Condition-based query
    │
    ├─► Parse eligibility criteria
    │     Free text → structured criteria
    │
    ├─► Match patient to trials
    │     Each criterion: met / not met / indeterminate
    │
    └─► Display ranked results

Architecture

Drosophila is split into two packages:

Package	Description
drosophila_core	Pure Dart. Criteria parsing, patient matching, trial ingestion. No Flutter dependency.
drosophila_flutter	Flutter UI. Screens, widgets, SMART launch service, CDS Hooks service, theme.

Patient Data Extraction

On SMART launch, Drosophila requests these scopes and extracts:

FHIR Resource	Extracted Data
Patient	Age, sex
Condition	Primary cancer diagnosis (ICD-10 / SNOMED), comorbidities
Observation	Staging (TNM), biomarkers (HER2, ER, PR, EGFR, ALK, BRAF, KRAS, PD-L1, etc.), labs, ECOG/Karnofsky
MedicationRequest	Current and past medications
Procedure	Surgical and treatment history

Cancer detection uses ICD-10-CM ranges (C00-C96, D00-D09, D37-D48) and a curated set of ~200 mCODE SNOMED codes covering all major cancer types.

Eligibility Criteria Pipeline

ClinicalTrials.gov returns eligibility criteria as free text. Drosophila parses this into structured, matchable criteria:

1. Split — Separate inclusion and exclusion sections, extract individual criteria from bullet points
2. Classify — Categorize each criterion (demographics, staging, biomarker, lab, performance status, prior therapy, etc.)
3. Extract — Pull structured requirements (e.g., "age ≥ 18" → AgeRange(min: 18))
4. Match — Compare against patient profile, producing a three-state verdict:
   • Met — Patient data satisfies the criterion
   • Not Met — Patient data contradicts the criterion
   • Indeterminate — Insufficient data to determine

Scoring

Each trial receives a match score (0–100%) based on the proportion of criteria met, with exclusion criteria weighted more heavily. Trials are ranked by score and categorized:

• Eligible — All evaluable criteria met, no exclusions triggered
• Needs Review — Some criteria indeterminate (missing data)
• Not Eligible — One or more exclusion criteria met (hidden by default)

Trial Data

Drosophila fetches from the ClinicalTrials.gov API v2 using format=json. For each trial it extracts:

• NCT ID, title, status, phase
• Brief description
• Target conditions
• Interventions
• Sponsor
• Enrollment count
• Study locations
• Eligibility criteria (free text, parsed into structured form)

Dependencies

Drosophila builds on several FHIR-FLI packages:

• fhir_r4 — FHIR R4 resource models
• fhir_r4_at_rest — FHIR REST client for patient data extraction
• fhir_r4_auth — SMART on FHIR authentication
• fhir_r4_cds_hooks — CDS Hooks service definition and response building