PRISMA-IPD

What it is

PRISMA-IPD (Preferred Reporting Items for a Systematic Review and Meta-Analysis of Individual Participant Data) is the reporting guideline for the specific class of evidence synthesis in which the reviewers obtain the individual participant data — the raw, line-level records for each randomized or enrolled subject — from the original investigators, harmonize and re-check those data, and analyze them centrally, instead of extracting summary effect estimates from publications. Published by Stewart and colleagues for the IPD Meta-analysis Methods Group as the PRISMA-IPD Statement (JAMA, 2015), it extends the parent PRISMA checklist with items unique to the IPD process: documenting which studies provided data and which did not, how the supplied data were integrity-checked, how variables were harmonized across datasets, and how the meta-analysis modeled within- and between-study structure. It is maintained within the EQUATOR Network / PRISMA family alongside the parent PRISMA 2020 statement and the protocol extension PRISMA-P; the checklist and flow diagram are hosted at prisma-statement.org. PRISMA-IPD is a reporting standard — it governs the transparency and completeness of what authors write, not the methodological quality of the review itself.

When to use

Use PRISMA-IPD whenever the synthesis is built on individual participant data — whether the underlying studies are randomized trials, observational cohorts, or a mix — and the deliverable is a peer-reviewed manuscript, an HTA/payer evidence dossier, a regulatory (FDA/EMA) submission relying on pooled patient-level evidence, or a registered protocol. Decision rule for choosing within the PRISMA family: if you are pooling published aggregate results, report under parent PRISMA 2020; if you are writing the protocol for any systematic review, use PRISMA-P; if you are obtaining and re-analyzing patient-level records, PRISMA-IPD is the correct extension and PRISMA 2020 alone is insufficient because it has no items for data solicitation, integrity checking, or harmonization. For network or diagnostic-accuracy IPD syntheses, layer the relevant topic extension (PRISMA-NMA, PRISMA-DTA) on top of the IPD items. In RWE, PRISMA-IPD increasingly applies to federated or pooled analyses of claims, EHR, and registry cohorts where partner sites contribute patient-level or common-data-model (e.g., OMOP) extracts to a central analysis.

What it requires

Beyond the standard PRISMA reporting domains (eligibility criteria, search strategy, selection process, synthesis methods, results, certainty of evidence), PRISMA-IPD enforces IPD-specific items that map directly onto RWD fitness-for-use concerns: (1) which data were sought and obtained — an explicit accounting of studies/sites approached, those that provided IPD, those that did not, and the participants represented versus missing, so the reader can judge availability bias; (2) data integrity and harmonization — how supplied datasets were checked (range/logic checks, randomization integrity, baseline balance, follow-up completeness) and how heterogeneous variable definitions, coding systems, and outcome ascertainment were reconciled into a common analysis-ready structure; (3) the analytic model — whether a one-stage (pooled with study as a factor/random effect) or two-stage (study-specific estimates then meta-analyzed) approach was used, how clustering by study was preserved, and how heterogeneity and subgroup/interaction effects were modeled; (4) risk of bias at the study and outcome level, assessed using the line-level data; and (5) handling of missing data and participants across contributing datasets. For real-world data these items force exactly the disclosures regulators and HTA bodies demand: transparent design, data-source fitness-for-use, phenotype/outcome-algorithm definitions reconciled across sites, time-zero/cohort-entry alignment, the target estimand and its handling of intercurrent events, and missing-data/attrition accounting.

When NOT to use — limitations and common misapplications

PRISMA-IPD is a reporting checklist, so the most damaging misuses confuse reporting completeness with scientific validity. (a) It is not a risk-of-bias instrument and not a quality score — completing every PRISMA-IPD item does not certify the review is unbiased; risk of bias still requires a dedicated tool (e.g., RoB 2 for trials, ROBINS-I for non-randomized studies, ROBIS for the review). (b) A high checklist score does not make pooled observational data causal — central re-analysis of patient-level claims/EHR records inherits confounding, selection, and misclassification that no reporting standard removes; the estimand and confounding strategy must be argued on their own merits. (c) Wrong member of the family: using parent PRISMA 2020 (or, worse, a non-IPD checklist) for an IPD synthesis hides the availability, integrity, and harmonization decisions that determine whether the pooled result is trustworthy; using PRISMA-IPD for an aggregate-data review imports items the study never engaged. (d) Checklist-as-theater: appending a filled grid at submission without the methods text actually describing data solicitation, integrity checks, and the one-/two-stage choice satisfies a journal field but informs no reader. (e) It does not replace protocol registration (use PRISMA-P), nor does it adjudicate certainty of evidence (use GRADE).

How it maps to this catalog

PRISMA-IPD's reporting items are implemented by concrete RWE methods in this repository: the data-availability/fitness items map to fit-for-purpose-data-assessment-rwe, database-feasibility-attrition-funnel-rwe, and multi-database federated analysis; harmonized outcome/exposure definitions map to diagnosis-phenotype-algorithm-1ip-2op-time-window-rwe and outcome-algorithm-construction-rwe (with OMOP common-data-model patterns via omop-cdm-method-patterns-rwe); cohort-entry/time-zero alignment across contributing studies maps to time-zero-index-date-alignment-rwe; the estimand and intercurrent-event reporting maps to estimands-ate-att-intercurrent-events-rwe; the pooled-cohort confounding-control strategy that the reanalysis must still defend maps to active-comparator-new-user, high-dimensional-propensity-score-hdps-rwe, and target-trial-emulation; missing-data and participant-flow items map to attrition-and-loss-to-follow-up-rwe and missing-data-pattern-table-rwe; and the one-stage vs two-stage modeling choice and study-level pooling map to ipd-meta-analysis, mixed-effects-models-longitudinal-rwe, and meta-analysis-rct / meta-analysis-obs. Applied note for claims/EHR/registry RWE: in a federated pooled analysis where partner sites map data to OMOP and return patient-level (or privacy-preserving site-level) extracts, PRISMA-IPD is the natural reporting backbone — report the sites solicited versus contributing (availability bias), the integrity/logic checks run on each site's extract, how phenotype algorithms and time-zero were harmonized across data sources of differing completeness (e.g., Medicare FFS vs Medicare Advantage vs commercial claims), and whether effects were pooled one-stage or two-stage — while remembering the checklist documents these choices but does not validate them.