Eligibility Criteria Normalization

The process of aligning free-text clinical terms to standardized concept identifiers within reference terminologies.

• Maps 'high blood pressure' → SNOMED CT 38341003
• Maps 'heart attack' → ICD-10-CM I21.9
• Resolves brand names to generics: 'Lipitor' → RxNorm 617310

This step ensures that synonymous expressions are treated as the same logical entity during automated screening, eliminating false negatives caused by vocabulary mismatches.

The conversion of disparate laboratory value representations into a single canonical unit system for consistent numeric comparison.

• Converts 'mg/dL' to mmol/L for glucose
• Normalizes 'lbs' to kg for weight-based criteria
• Standardizes date formats to ISO 8601 for temporal reasoning

Without this step, a criterion requiring 'HbA1c < 7.0%' would fail to match a patient record expressing the value as '0.07 proportion'.

The algorithmic generation of a comprehensive set of codes that satisfy a single clinical concept referenced in a trial criterion.

• A criterion for 'type 2 diabetes' expands to include ICD-10-CM E11.0 through E11.9
• 'Moderate to severe pain' expands to a numeric pain scale range of 4-10
• 'Beta blockers' expands to the full RxNorm descendant hierarchy

This ensures that the screening engine captures all possible coded representations of a condition, not just the most common one.

The standardization of how absent or negated clinical findings are represented in structured logic.

• Resolves 'absence of metastasis' to a negated SNOMED concept
• Normalizes 'no history of stroke' to a logical NOT operator on the stroke concept
• Distinguishes 'never smoked' from 'former smoker' using distinct status codes

Inconsistent negation handling is a primary source of false-positive screening results, where patients are incorrectly matched to trials for which they are ineligible.

The conversion of relative and ambiguous time references into precise, machine-computable intervals.

• 'Within the last 6 months' → duration ≤ 180 days from index date
• 'Newly diagnosed' → first occurrence timestamp within 90 days
• 'Stable disease for 4 weeks' → no progression event in prior 28 days

This normalization enables the temporal reasoning engine to accurately evaluate time-window constraints against a patient's longitudinal record.

The use of embedding-based similarity to identify clinical terms that are semantically equivalent but lexically distinct, beyond simple synonym matching.

• Identifies 'impaired renal function' as equivalent to 'chronic kidney disease'
• Recognizes 'elevated liver enzymes' as a manifestation of 'hepatic injury'
• Clusters 'shortness of breath' with 'dyspnea' and 'breathlessness'

This layer catches the edge cases that deterministic ontology mapping misses, using dense vector representations trained on clinical corpora.

The process of mapping synonymous clinical terms across disparate coding systems to a unified standard. Eligibility Criteria Normalization depends on this to resolve semantic equivalence—for example, mapping 'hypertension' (SNOMED CT 38341003) to 'I10' (ICD-10-CM) and 'high blood pressure' (lay term) into a single computable concept. Without robust ontology alignment, automated screening engines fail to recognize that different terms refer to the same condition, leading to false exclusions.

The automated extraction and structuring of free-text inclusion and exclusion requirements from clinical trial protocols into machine-readable formats. This upstream process feeds directly into normalization pipelines by isolating atomic criteria—such as 'HbA1c > 7.0% within the last 3 months'—that must then be mapped to standard lab codes (LOINC) and units. Parsing handles the syntactic decomposition; normalization handles the semantic standardization.

The conversion of normalized, structured eligibility criteria into executable database queries (SQL, FHIR API calls, or SPARQL). This downstream process consumes the output of Eligibility Criteria Normalization—standardized concepts with resolved units and value ranges—to generate precise retrieval logic. For example, a normalized criterion for 'serum creatinine ≤ 1.5 mg/dL' becomes a parameterized query against a lab results table with the correct LOINC code and unit conversion factor applied.

A machine-processable definition of a clinical condition expressed as logical expressions and data queries. Eligibility Criteria Normalization is the critical bridge between narrative trial criteria and computable phenotypes—it transforms 'moderate to severe rheumatoid arthritis' into a formal definition combining diagnosis codes, lab thresholds, and temporal constraints that a phenotype execution engine can evaluate against patient records.

The process of grounding ambiguous medical mentions to unique identifiers in standardized knowledge bases (UMLS, SNOMED CT, RxNorm). While Eligibility Criteria Normalization handles the mapping of terms and units, entity linking resolves the specific identity of a mention—disambiguating whether 'cold' refers to a temperature sensation, a viral infection, or chronic obstructive lung disease in the context of a trial criterion. Both work in tandem to eliminate semantic ambiguity.

The NLP capability to distinguish between affirmed, negated, and uncertain clinical findings in narrative text. During Eligibility Criteria Normalization, this is essential for correctly interpreting exclusion criteria—'patients without evidence of metastasis' must be normalized to a negated concept, while 'suspected but unconfirmed diagnosis' requires an uncertainty qualifier. Misclassifying negation or uncertainty directly corrupts the normalized criteria and produces invalid screening results.