Federated Cohort Discovery

A Distributed Query Engine receives a single analytical request and decomposes it into site-specific sub-queries. Each sub-query is dispatched to a remote node where it executes locally against the institution's native data model.

• Translates logical cohort criteria into local dialect (e.g., OMOP, i2b2)
• Executes aggregate-only operations: COUNT, SUM
• Assembles partial results into a harmonized final answer
• Never requests or retrieves patient-level rows

A Computable Phenotype is a machine-executable algorithm that defines cohort membership using structured clinical logic. Unlike free-text descriptions, it eliminates human ambiguity.

• Composed of inclusion and exclusion criteria
• References standardized terminologies: SNOMED CT, LOINC, RxNorm
• Combines diagnosis codes, lab values, medication orders, and temporal constraints
• Example: HbA1c > 9.0 within 6 months AND Type 2 Diabetes diagnosis AND no insulin prescription

The cardinal rule of federated discovery: nodes return only aggregate statistics, never individual records. This is enforced cryptographically or architecturally.

• Responses are limited to contingency tables and count vectors
• Prevents reconstruction of individual patient trajectories
• Satisfies Data Use Agreements and regulatory requirements
• Enables honest-broker architectures where a neutral third party validates query safety

Before pooling results, the system performs a Heterogeneity Assessment to determine if site-specific counts are statistically compatible. Significant variability may indicate differences in coding practices or population demographics.

• Quantified via I-squared statistic or Cochran's Q test
• High heterogeneity triggers a review of local phenotype execution
• Prevents misleading global counts that mask site-specific divergence
• Outputs often visualized as a forest plot of per-site cohort sizes

Federated discovery achieves semantic interoperability by mapping local data to the OMOP Common Data Model maintained by OHDSI. This standardization ensures a concept like 'Myocardial Infarction' maps identically across sites.

• Standardized vocabularies eliminate lexical mismatches
• Enables the same computable phenotype to run unmodified at every node
• Transforms heterogeneous EHR schemas into a unified analytics surface
• Facilitates cross-institutional Real-World Evidence generation

A Secure Aggregation Protocol cryptographically ensures that the central coordinator learns only the sum of all site counts, not any individual site's contribution. This protects against honest-but-curious intermediaries.

• Uses secure multi-party computation or additive secret sharing
• Each site's count is masked before transmission
• Masks cancel out during aggregation, revealing only the total
• Critical for competitive or sensitive feasibility assessments

A machine-executable algorithm that identifies a patient cohort based on specific clinical characteristics. It translates complex inclusion criteria into a deterministic query logic using structured codes, lab values, and medication orders.

• Key Components: ICD-10/SNOMED CT codes, LOINC lab results, and RxNorm drug orders
• Validation: Requires iterative refinement against manual chart review to ensure sensitivity and specificity
• Federated Context: Phenotypes must be harmonized across sites with different coding practices to ensure consistent cohort identification

An open community data standard maintained by OHDSI that transforms heterogeneous clinical data into a consistent, standardized format. It maps native EHR schemas to a unified set of tables and vocabularies, enabling systematic observational research across institutions.

• Core Tables: PERSON, OBSERVATION, DRUG_EXPOSURE, MEASUREMENT, CONDITION_OCCURRENCE
• Vocabulary Mapping: Normalizes local codes (ICD-10, SNOMED) to standard concept IDs
• Federated Role: Provides the semantic interoperability layer that allows a single cohort definition to execute identically across disparate hospital systems

A software component that decomposes a single analytical query into sub-queries, executes them locally at each participating institution, and assembles the partial results into a final, harmonized answer without moving patient-level data.

• Execution Model: Push-down computation where only aggregate counts leave the site
• Optimization: Query planning must account for heterogeneous database backends (PostgreSQL, SQL Server, BigQuery)
• Example: A query for 'diabetic patients on metformin with HbA1c > 7%' returns only the count per site, not the individual records

A cryptographic method that allows a central server to compute the sum of model updates or statistics from multiple clients while ensuring that individual contributions remain private and unreadable. In cohort discovery, it protects the exact count from a single site before global summation.

• Mechanism: Uses secret sharing or pairwise masking to blind individual inputs
• Threat Model: Protects against an honest-but-curious aggregator; can be extended for malicious adversaries
• Integration: Often combined with differential privacy noise addition to prevent inference from the final aggregate

The specific clinical, demographic, and temporal characteristics that a patient must possess to be eligible for enrollment in a clinical study or cohort definition. These criteria form the logical backbone of a computable phenotype.

• Temporal Logic: Requires defining time windows (e.g., 'diagnosis within 6 months of first prescription')
• Boolean Operators: Combines criteria using AND, OR, NOT, and temporal sequencing (e.g., 'A occurred before B')
• Granularity: Can range from simple age/gender filters to complex genomic variant and biomarker thresholds

A cross-tabulation matrix that displays the frequency distribution of two categorical variables. In federated cohort discovery, it is the foundational output structure for calculating odds ratios and performing chi-squared tests without exposing raw data.

• Structure: A 2x2 table showing exposure status vs. outcome status
• Federated Computation: Each site computes its local 2x2 table; the global table is the element-wise sum
• Statistical Power: The aggregated table enables meta-analysis techniques like inverse variance weighting to compute pooled effect estimates