HEDIS Measure Data Architecture: Building a Compliant Quality Reporting Pipeline

Introduction

[HEDIS](/terms/HEDIS) measure calculation is not a data warehouse query problem. It is a data integration problem, a logic specification problem, and a source data quality problem — all combined. NCQA's technical specifications define denominator criteria, numerator criteria, exclusion logic, and allowable value sets for over 90 measures. Each measure specification runs 5–15 pages. One misunderstood logic clause can incorrectly exclude thousands of members from a denominator, producing a measure rate that bears no resemblance to reality.

NCQA Compliance Audits and Health Plan Accreditation reviews are not forgiving of measure logic errors. This guide covers the architecture and data model for a defensible HEDIS pipeline.

---

Source Data Domains

HEDIS measures draw from four primary data sources, each with different availability and quality characteristics:

---

Administrative vs Hybrid Measures

Administrative measures are calculated entirely from claims data. No supplemental data sources required. Faster to calculate, but limited to what appears on a claim.

Hybrid measures use a combination of administrative data and medical record review. For a random sample of members, NCQA requires medical record abstraction to validate administrative data or supplement it.

Electronic Clinical Data System (ECDS) measures use EHR data (clinical quality measures) directly.

Your data architecture must support all three. Hybrid measures require a sampling workflow and a medical record abstraction data model alongside the claims-based logic.

---

Measure Denominator Continuous Enrollment Logic

Every HEDIS measure has a continuous enrollment requirement — the member must have been continuously enrolled for a specified period (typically 1 year for the measurement year) to be eligible for the denominator. This is the most commonly misimplemented logic in HEDIS pipelines.

-- Continuous enrollment check: member must have no gap >45 days during measurement year
-- (NCQA standard is no gaps, or 1 gap ≤45 days depending on measure)
WITH enrollment_gaps AS (
  SELECT
    enterprise_member_id,
    coverage_start_date,
    coverage_end_date,
    LAG(coverage_end_date) OVER (
      PARTITION BY enterprise_member_id ORDER BY coverage_start_date
    ) AS prev_coverage_end,
    coverage_start_date - LAG(coverage_end_date) OVER (
      PARTITION BY enterprise_member_id ORDER BY coverage_start_date
    ) AS gap_days
  FROM member.coverage_periods
  WHERE coverage_start_date <= '2024-12-31'
    AND coverage_end_date   >= '2024-01-01'
),
member_gap_summary AS (
  SELECT
    enterprise_member_id,
    MAX(COALESCE(gap_days, 0)) AS max_gap_days,
    COUNT(CASE WHEN gap_days > 0 THEN 1 END) AS gap_count
  FROM enrollment_gaps
  GROUP BY enterprise_member_id
)
SELECT enterprise_member_id
FROM member_gap_summary
WHERE max_gap_days <= 45  -- NCQA threshold; varies by measure
  AND gap_count <= 1;     -- at most 1 gap allowed; varies by measure

---

Breast Cancer Screening (BCS) Measure Logic

Breast Cancer Screening is one of the most commonly reported HEDIS measures. Here is the logic structure:

Denominator: Women aged 52–74 as of December 31 of the measurement year who were continuously enrolled in the measurement year and the year prior.

Numerator: A mammogram (bilateral or unilateral) on or between October 1 of the year prior to the measurement year and December 31 of the measurement year (27-month window).

-- BCS Denominator
WITH bcs_denominator AS (
  SELECT m.enterprise_member_id
  FROM mdm.member_master m
  JOIN analytics.member_continuous_enrollment ce
    ON m.enterprise_member_id = ce.enterprise_member_id
    AND ce.measurement_year = 2024
    AND ce.continuously_enrolled_flag = TRUE
  WHERE m.gender_code = 'F'
    AND DATEDIFF('year', m.date_of_birth, '2024-12-31') BETWEEN 52 AND 74
),

-- BCS Numerator: mammogram in 27-month window
bcs_numerator AS (
  SELECT DISTINCT c.enterprise_member_id
  FROM claims.claim_header c
  JOIN claims.claim_line cl ON c.claim_id = cl.claim_id
  WHERE cl.cpt_code IN (
    '77055','77056','77057','77061','77062','77063',  -- [CPT](/terms/CPT) mammography codes
    'G0202','G0204','G0206'                           -- HCPCS mammography codes
  )
  AND c.service_from_date BETWEEN '2022-10-01' AND '2024-12-31'  -- 27-month window
  AND c.adjudication_status = 'PAID'
)

-- BCS Measure Result
SELECT
  d.enterprise_member_id,
  CASE WHEN n.enterprise_member_id IS NOT NULL THEN TRUE ELSE FALSE END AS in_numerator,
  TRUE AS in_denominator
FROM bcs_denominator d
LEFT JOIN bcs_numerator n ON d.enterprise_member_id = n.enterprise_member_id;

---

HEDIS Data Model

CREATE TABLE quality.hedis_measure_result (
  result_id             BIGINT GENERATED ALWAYS AS IDENTITY,
  measurement_year      INT           NOT NULL,
  measure_id            VARCHAR(20)   NOT NULL,  -- BCS, CBP, CDC, COL, etc.
  measure_name          VARCHAR(200),
  enterprise_member_id  VARCHAR(36)   NOT NULL,
  in_denominator        BOOLEAN       NOT NULL,
  denominator_exclusion BOOLEAN,
  exclusion_reason      VARCHAR(100),
  in_numerator          BOOLEAN,
  data_source_flag      VARCHAR(20),             -- ADMINISTRATIVE, HYBRID, ECDS
  numerator_date        DATE,                    -- date of qualifying event
  numerator_code        VARCHAR(20),             -- CPT/HCPCS/LOINC that satisfied numerator
  gap_open_flag         BOOLEAN,                 -- denominator-eligible, not in numerator
  computed_at           TIMESTAMP     NOT NULL,
  logic_version         VARCHAR(20),             -- NCQA spec version used
  PRIMARY KEY (result_id)
);

CREATE UNIQUE INDEX ON quality.hedis_measure_result
  (measurement_year, measure_id, enterprise_member_id);

---

Value Set Management

NCQA publishes HEDIS value sets — the CPT, HCPCS, ICD-10, LOINC, and NDC code lists that define qualifying events for each measure. These update annually and are licensed from NCQA.

Store value sets in a dedicated reference table:

CREATE TABLE reference.hedis_value_set (
  value_set_id          VARCHAR(50)   NOT NULL,
  value_set_name        VARCHAR(200)  NOT NULL,
  code_system           VARCHAR(20)   NOT NULL,  -- CPT, HCPCS, ICD10CM, LOINC, NDC, RXNORM
  code                  VARCHAR(20)   NOT NULL,
  code_desc      VARCHAR(500),
  ncqa_release_year     INT           NOT NULL,
  effective_date        DATE,
  PRIMARY KEY (value_set_id, code_system, code)
);

Join your claims and lab data against reference.hedis_value_set rather than hardcoding CPT lists in your measure SQL. This makes annual value set updates a data load, not a code change.

---

Key Takeaways

• Continuous enrollment logic is the most commonly misimplemented HEDIS requirement. Implement it as a reusable utility table consumed by all measures.
• Value sets must be stored in a reference table, not hardcoded in measure SQL — NCQA updates them annually and a hardcoded value set will produce incorrect results after the update.
• Hybrid measures require a sampling and medical record abstraction workflow alongside the administrative calculation pipeline.
• The logic_version column in your measure result table is not optional — NCQA auditors will ask which specification year your logic implements.
• Use the [HCC Calculator](/tools/hcc-calculator) to validate risk adjustment data that feeds into HEDIS measure denominator eligibility for risk-stratified measures.