Eligibility Scoring

Assigns differential importance weights to individual inclusion and exclusion criteria based on protocol criticality. A hard exclusion (e.g., prior anaphylaxis) carries infinite negative weight, while a preferred but non-essential biomarker range contributes a fractional positive score. The final score is a weighted sum across all evaluated criteria, allowing a patient who perfectly matches critical factors but misses a minor preference to outrank a patient with mediocre performance across all dimensions.

Integrates the model's extraction confidence directly into the numerical score. When an NLP model extracts a key lab value with 99.8% confidence, it contributes fully to the score; when confidence drops to 65% due to ambiguous text, the contribution is proportionally discounted. This prevents low-certainty extractions from artificially inflating or deflating a patient's rank and provides a transparent signal for human-in-the-loop review prioritization.

Applies a time-decay function to clinical facts based on their recency relative to the trial's temporal constraints. A hemoglobin A1c value from 3 weeks ago contributes fully, while one from 11 months ago is penalized according to a configurable decay curve. This ensures the composite score reflects the clinical freshness of evidence, preventing outdated data from producing misleadingly high eligibility rankings.

Generates a rank-ordered list of patient-trial pairs across multiple active protocols simultaneously. Each patient receives a separate eligibility score for every trial they are screened against, enabling site coordinators to view a dashboard of top candidates sorted by match strength. This supports portfolio-level recruitment optimization, where a single patient can be directed to the trial where they represent the highest-value enrollment opportunity.

Every composite score is fully auditable and decomposable into its constituent parts. A reviewer can drill into a score of 87.3 to see exactly which criteria contributed positively, which contributed negatively, and which were unevaluable due to missing data. This criterion-level transparency is essential for regulatory compliance and builds trust with clinical reviewers who must validate automated rankings before patient contact.

Maps continuous numerical scores to discrete eligibility tiers using configurable thresholds. Patients scoring above 90 are classified as 'Highly Eligible' and fast-tracked for immediate coordinator review, while those in the 70-89 range enter a standard review queue. This stratification enables workflow automation, routing candidates to appropriate review pipelines based on match strength and reducing manual triage overhead for high-volume screening operations.

The assignment of relative importance scores to individual inclusion and exclusion criteria. Not all criteria are equal; a critical genomic biomarker (e.g., EGFR T790M mutation) carries more weight than a flexible age range.

• Weight derivation: Often assigned by clinical experts or learned from historical screen failure data.
• Impact: Directly determines the final eligibility score, preventing a perfect match on minor criteria from masking a failure on a major one.
• Example: A trial requiring PD-L1 expression > 50% might assign this criterion a weight of 0.4, while a prior therapy line count gets 0.1.

The process of breaking down a complex, multi-part clinical trial eligibility criterion into its atomic, independently evaluable logical components. A single sentence like "Patient must have histologically confirmed non-small cell lung cancer and have progressed on platinum-based chemotherapy" becomes multiple boolean checks.

• Atomic units: Each decomposed element maps to a single, queryable data point in the patient record.
• Logical operators: Preserves the original AND/OR/NOT relationships between components.
• Necessity: A scoring model cannot weight or evaluate a criterion it cannot isolate.

A technique that transforms a patient's entire clinical profile into a dense numerical vector, enabling semantic similarity comparisons with trial requirements. This moves beyond deterministic matching to capture latent clinical relevance.

• Representation: Encodes diagnoses, medications, procedures, and lab results into a fixed-length vector.
• Similarity scoring: The cosine distance between a patient vector and a trial vector contributes to the overall eligibility score.
• Advantage: Identifies candidates whose records describe a condition using synonymous or related terms not explicitly listed in the criteria.

The systematic review of reasons why pre-screened patients failed to meet trial eligibility. This is the feedback loop that refines the scoring model's accuracy over time.

• Root cause categorization: Failures are classified by criterion type (e.g., concomitant medication, incorrect staging).
• Model refinement: High failure rates on a specific criterion may indicate the parsing logic or weight assignment needs adjustment.
• Protocol feedback: Aggregated data can reveal overly restrictive criteria that are excluding the target population, informing protocol amendments.

The AI capability to interpret and validate time-dependent clinical constraints against a patient's longitudinal record. A criterion like "no chemotherapy within 28 days of enrollment" requires precise clinical event sequencing.

• Constraint types: Washout periods, disease progression timelines, and maximum intervals between diagnosis and treatment.
• Scoring integration: A temporal violation typically results in a hard exclusion (score of zero) rather than a partial penalty.
• Complexity: Requires accurate patient timeline reconstruction from timestamped, often inconsistent, source data.

A clinical trial screening system design that combines deterministic rule-based filtering with probabilistic semantic matching to maximize both precision and recall in the final eligibility score.

• Deterministic layer: Hard filters for binary criteria (e.g., confirmed diagnosis of NSCLC) using a rule engine.
• Probabilistic layer: Semantic vector search and weighted scoring for nuanced or text-heavy criteria.
• Orchestration: The final score is a composite, where a deterministic failure overrides any probabilistic match, ensuring patient safety and protocol adherence.