AHRQ Registries for Evaluating Patient Outcomes: A User's Guide

What it is

Registries for Evaluating Patient Outcomes: A User's Guide (the "AHRQ Registries Guide," currently the 4th edition, Gliklich, Leavy & Dreyer, eds., AHRQ, 2020) is the field's most comprehensive operating manual for patient registries. It is not a one-page reporting checklist; it is a multi-chapter best-practice handbook that walks an investigator through the full life cycle of a registry — defining purpose and stakeholders, governance and oversight, sample-size and design, selection of data elements, patient identification and recruitment, data sources and linkage, data collection and quality assurance, analysis, interpretation, and long-term operation, transition, or closure. It is maintained and published by the U.S. Agency for Healthcare Research and Quality (AHRQ) through its Effective Health Care Program, with successive editions and topic-specific addenda (e.g., registries assessing the safety and effectiveness of medical products, linking registries to other data, 21st-century interoperability). It is constructive ("how to build and run one well") rather than purely evaluative, which is what distinguishes it from registry critical-appraisal checklists.

When to use

Reach for the AHRQ Registries Guide whenever the deliverable is a registry as an evidence-generation instrument — a disease/condition registry, a product (drug or device) registry, a pregnancy/exposure registry, a post-authorization safety study (PASS) or post-marketing requirement run as a registry, or a registry-based study feeding an FDA/EMA submission, an HTA/payer dossier, or a peer-reviewed manuscript. Use it at the design and build stage, before data collection, when you must justify governance, data-element definitions, outcome ascertainment, and quality processes prospectively. Decision rules versus siblings: choose the AHRQ Guide when the task is to plan, build, or operate a registry end-to-end; choose ENCePP Guide/Checklist when you are conducting or reporting a (often EU) pharmacoepidemiologic study more broadly; choose the FDA RWE framework / FDA non-interventional guidances or CIOMS RWD/RWE when the task is to align registry output with a specific regulatory submission (those frameworks consume what the AHRQ Guide helps you produce); choose ISPE GPP / SCOPE or GRACE when you need to appraise the quality of a finished study rather than construct a registry; and choose STROBE / RECORD / RECORD-PE for the reporting manuscript that comes out of the registry. In practice the AHRQ Guide is used alongside, not instead of, those documents.

What it requires

The Guide enforces substantive, registry-specific domains, each of which has a concrete real-world-data implementation in this catalog: (1) Purpose, stakeholders, and governance — an explicit primary question, oversight structure, and a transparent plan (links to study protocol/SAP elements). (2) Design and fitness-for-purpose of data — registry design choices and an honest assessment of whether the captured data can answer the question (data fitness-for-use). (3) Patient identification, enrollment, and selection — eligibility, recruitment, representativeness, and time-zero alignment so follow-up begins at a defined, unbiased index. (4) Data elements, phenotypes, and outcome ascertainment — standardized, validated condition/exposure/outcome definitions and adjudicated or validated endpoints rather than ad hoc code lists. (5) Data quality, completeness, and follow-up — source-data verification, monitoring, and pre-specified handling of attrition, loss to follow-up, and missing data. (6) Analysis and confounding control — appropriate comparison strategies (often active-comparator, new-user, or target-trial-emulation logic), confounding adjustment, and clearly specified estimands and intercurrent events. (7) Interpretation, sensitivity, and quantitative bias analysis — assessment of residual confounding, selection, and misclassification, plus transportability/generalizability of the registry population. (8) Quality assessment of the registry itself — the Guide's evaluation chapter provides a structured framework to grade a registry's rigor.

When NOT to use — limitations and common misapplications

(a) The AHRQ Registries Guide is a best-practice manual, not a validated risk-of-bias instrument and not a numeric quality score; do not report "AHRQ score = X" — its evaluation chapter is a structured appraisal aid, not a weighted scale like a STROBE-derived score or a ROBINS-I rating. For formal risk-of-bias use ROBINS-I/ROBINS-E; for reporting completeness use STROBE/RECORD. (b) Following the Guide does not make a registry causal. A well-governed, high-quality registry that lacks a sound comparator, time-zero alignment, and confounding control still cannot support a causal claim; the operational excellence the Guide demands is necessary, not sufficient. (c) Wrong document for the stage: using the AHRQ Guide as your manuscript reporting checklist (STROBE/RECORD is correct), or using a reporting checklist to design a registry (the AHRQ Guide is correct), is a category error. (d) Checklist/manual-as-theater: citing the Guide in a protocol while none of its data-quality, phenotype-validation, or governance practices were actually implemented. (e) Edition/scope drift: applying generic registry advice to a regulatory product/safety registry without consulting the relevant addendum (medical-product safety and effectiveness, or registry-to-data linkage) that adds the regulatory-grade requirements.

How it maps to this catalog

Each Guide requirement is implemented by a concrete concept here: fitness-for-purpose of data -> `fit-for-purpose-data-assessment-rwe`; phenotype/algorithm definition and validation -> `diagnosis-phenotype-algorithm-1ip-2op-time-window-rwe` and `algorithm-validation`; outcome ascertainment/adjudication -> `endpoint-adjudication-chart-review-rwe`; time-zero/index alignment -> `time-zero-index-date-alignment-rwe`; estimands and intercurrent events -> `estimands-ate-att-intercurrent-events-rwe`; attrition, follow-up, and missing data -> `attrition-and-loss-to-follow-up-rwe` and `missing-data-pattern-table-rwe`; comparison and confounding control -> `active-comparator-new-user`, `propensity-score-methods-psm-iptw`, and `target-trial-emulation`; sensitivity and residual bias -> `quantitative-bias-analysis-toolkit-rwe`; external validity of the registry population -> `generalizability-transportability-external-validity-rwe`. The registry study-type definitions are `disease-registry`, `product-registry`, and `pregnancy-registry` (the latter typically paired with `mother-infant-linkage-rwe`), and registry analyses frequently lean on `claims-analysis` for linkage to complete utilization and mortality data.

Applied note (claims/EHR/registry RWE)

Registries are rarely self-contained. A disease or product registry usually captures clinical detail and adjudicated outcomes well but is incomplete for full medication exposure, healthcare utilization, and death; linking to administrative claims (with continuous-enrollment requirements) and a death index is the standard remedy and is exactly where the AHRQ Guide's data-source, linkage, and quality chapters earn their keep. For a regulatory product/safety registry, treat the Guide's medical-product addendum as the floor: pre-specify validated phenotypes and adjudicated endpoints, document time-zero and the comparator strategy, quantify attrition and missingness, and carry residual-confounding sensitivity analyses — the same machinery a reviewer expects from any defensible non-interventional RWE study.