Trial Pre-Screening API

The API accepts a minimal, de-identified patient profile containing key demographic and diagnostic data points without exposing Protected Health Information (PHI). This initial submission typically includes age, sex, primary diagnosis (ICD-10-CM), and high-level genomic markers. The API acts as a privacy firewall, ensuring that full medical records are never exposed during the initial screening phase. This architecture supports HIPAA compliance by design, allowing institutions to query external trial databases without risking a data breach.

Internally, the API translates structured trial eligibility criteria into executable database queries. This engine parses complex inclusion and exclusion logic—such as 'HbA1c < 8.0% within the last 6 months'—and converts it into FHIR API calls or SQL statements that run against the submitting institution's clinical data warehouse. The translation layer handles temporal constraints, unit normalization, and ontology alignment to ensure that a criterion written in human language is accurately represented as a machine-executable filter.

Beyond deterministic rule-based filtering, the API generates a dense patient vector embedding from the submitted profile. This numerical representation captures the semantic essence of the patient's clinical state and is compared against pre-computed trial requirement vectors using cosine similarity. This hybrid approach catches eligible patients who might be missed by strict criteria matching—for example, a patient with a synonymous diagnosis not explicitly listed in the protocol. The result is a ranked list of trials with a quantitative match score.

When a patient does not match a trial, the API returns structured screen failure reasons rather than a simple null result. The response payload includes a breakdown of which specific criteria caused the exclusion, enabling downstream systems to log the failure for cohort analysis and protocol feasibility assessment. This feedback loop is critical for sites to understand their patient population's alignment with a sponsor's requirements and for sponsors to identify overly restrictive criteria that may be hindering recruitment.

The API integrates a drug interaction and exclusion checker that cross-references the patient's active medication list against a trial's prohibited concomitant medications. Using RxNorm normalized drug codes, the system identifies exclusionary medications, including those within a specified washout period. This automated check prevents the common manual error of overlooking a disallowed medication buried in a patient's lengthy active prescription list, a leading cause of late-stage screen failures.

For basket and umbrella trials, the API evaluates a single patient profile against all sub-study arms simultaneously. The response returns a list of matching arms within a master protocol, each with its own eligibility score. This capability is essential for precision oncology trials where a single biomarker test can qualify a patient for multiple targeted therapy arms, maximizing the value of each screened patient and accelerating enrollment across complex trial designs.

The core computational engine that compares a structured patient profile against a trial's inclusion and exclusion criteria. It goes beyond simple keyword matching by using semantic similarity and temporal reasoning to determine eligibility. The algorithm typically operates in two phases: a high-recall pre-screening step to filter out obvious mismatches, followed by a high-precision deep screening against the full protocol.

The automated NLP process of transforming free-text clinical trial protocols into a machine-readable, structured format. This involves:

• Entity extraction: Identifying conditions, medications, lab values, and procedures.
• Constraint normalization: Converting 'HbA1c > 7.0%' into a structured operator, value, and unit.
• Logical decomposition: Breaking down complex criteria like 'Must have failed at least 2 prior lines of therapy' into atomic, evaluable rules.

A technique that transforms a patient's entire clinical profile—diagnoses, medications, procedures, and lab results—into a dense numerical vector. This embedding captures the semantic essence of the patient's medical state. The Pre-Screening API can use these vectors to perform similarity searches against trial embeddings, finding matches even when exact terminology differs between the patient record and the protocol.

A machine-processable definition of a clinical condition, expressed as a set of logical expressions and data queries. For trial screening, computable phenotypes are used to define both the target disease and the eligibility constraints. A phenotype for 'Type 2 Diabetes with Neuropathy' might include ICD-10-CM codes, HbA1c thresholds, and medication logic that the Pre-Screening API executes against a patient's structured data.

The AI capability to interpret and validate time-dependent clinical constraints against a patient's longitudinal record. Examples include:

• Washout periods: 'No investigational drug within 30 days.'
• Disease progression: 'Documented progression within the last 6 months.'
• Sequence logic: 'Surgery must have occurred after completion of neoadjuvant chemotherapy.' This ensures the Pre-Screening API doesn't just match facts, but their correct chronological order.

An analysis that uses the Pre-Screening API to estimate the number of potentially eligible subjects at a specific research site. By running the API against a site's de-identified patient registry, sponsors can rank sites by recruitment potential before activation. This data-driven approach replaces guesswork with quantitative feasibility metrics, reducing the number of non-enrolling sites that delay trials.