Eligibility Criteria Parsing

The process of breaking down a complex, multi-part clinical trial eligibility criterion into its atomic, independently evaluable logical components. This involves parsing narrative text to identify distinct clinical concepts, logical operators (AND, OR, NOT), and temporal constraints. For example, a single paragraph stating 'Patients must have histologically confirmed NSCLC with EGFR exon 19 deletion and no prior TKI therapy within 6 months' is decomposed into three separate evaluable facts: diagnosis, biomarker status, and a temporal therapy exclusion.

The ability to distinguish between affirmed, negated, and uncertain clinical findings within protocol text. This is critical because criteria often use phrases like 'absence of metastasis,' 'no active infection,' or 'suspected but unconfirmed diagnosis.' Advanced parsing systems use contextual embeddings and dependency parsing to correctly scope negation triggers. Misinterpreting 'patients without uncontrolled hypertension' as requiring hypertension would invert the eligibility logic and catastrophically compromise cohort identification.

The AI capability to interpret and validate time-dependent clinical constraints against a patient's longitudinal record. Trial protocols frequently specify windows such as 'within 28 days of enrollment,' 'no chemotherapy for 6 months prior,' or 'disease progression after 2 lines of therapy.' Parsing must extract not just the clinical event but the temporal operator (before, after, within) and the duration or anchor point. This structured output enables downstream temporal reasoning engines to sequence patient events correctly.

The process of converting parsed, structured eligibility criteria into executable database queries such as SQL, FHIR API calls, or SPARQL. Once atomic criteria are extracted and normalized, they must be translated into a syntax that can interrogate clinical data repositories. For instance, a parsed criterion for 'HbA1c > 7.0%' becomes a structured query selecting patients with a LOINC code 4548-4 and a numeric value exceeding 7.0. This translation layer bridges the gap between natural language protocols and operational screening infrastructure.

The mapping of extracted clinical terms to standardized terminologies such as SNOMED CT, LOINC, and RxNorm. Trial protocols use heterogeneous language—'heart attack,' 'myocardial infarction,' and 'MI' all refer to the same concept. Parsing systems must normalize these synonyms to a single concept ID to enable consistent querying across disparate data sources. This semantic harmonization is foundational for interoperability and prevents false negatives caused by terminological mismatches.

An automated process that cross-references a patient's active medication list against a trial's prohibited concomitant medications. Parsing must extract drug names, classes, and routes of administration from exclusion criteria, then normalize them to standard drug vocabularies like RxNorm. The system must also interpret dosage thresholds and temporal windows, such as 'systemic corticosteroids >10mg prednisone equivalent within 14 days.' This ensures patient safety and protocol compliance before enrollment.

A machine-processable definition of a clinical condition, expressed as a set of logical expressions and data queries, used to identify patient cohorts from electronic health records. Parsed eligibility criteria are essentially a collection of computable phenotypes.

• Often defined using standard vocabularies like SNOMED CT and ICD-10-CM
• Includes value sets for lab results, medications, and procedures
• Validated against gold-standard manual chart review

The AI capability to interpret and validate time-dependent clinical constraints against a patient's longitudinal record. Parsing must extract not just the condition, but the temporal logic.

• Washout periods: 'No investigational drug within 30 days'
• Disease progression: 'Documented progression after platinum-based chemotherapy'
• Chronicity: 'Diagnosis of hypertension at least 6 months prior to enrollment'

The application of natural language processing to identify and isolate specific eligibility conditions from narrative text in clinical trial protocols. This is the direct precursor to parsing.

• Handles complex sentence structures like 'Patients must have failed at least two prior lines of therapy, one of which must have included a taxane'
• Resolves anaphora and coreference (e.g., linking 'the drug' to a specific medication)
• Distinguishes between inclusion and exclusion criteria sections

The process of mapping synonymous clinical terms and varying units of measure within trial criteria to a standard ontology to ensure consistent automated interpretation. Without normalization, a parsed criterion for 'high blood pressure' would miss a patient record documenting 'essential hypertension'.

• Maps brand names to generics (e.g., 'Tylenol' to 'acetaminophen')
• Converts lab units (e.g., mg/dL to mmol/L)
• Aligns to reference terminologies like RxNorm and LOINC

The process of breaking down a complex, multi-part clinical trial eligibility criterion into its atomic, independently evaluable logical components. A single sentence like 'Histologically confirmed metastatic colorectal cancer with measurable disease per RECIST 1.1' decomposes into:

• Diagnosis: Colorectal Cancer (with morphology confirmation)
• Staging: Metastatic (Stage IV)
• Assessment: Measurable disease by RECIST 1.1 criteria