Guidelines for Good Pharmacoepidemiology Practices (GPP)
ISPE's overarching good-practice guidance for the planning, conduct, analysis, archiving, and communication of pharmacoepidemiologic and non-interventional real-world studies. It is substantive practice guidance, not a line-by-line reporting checklist or a risk-of-bias instrument.
What it is
— The Guidelines for Good Pharmacoepidemiology Practices (GPP) are the profession-defining good-practice statement maintained by the International Society for Pharmacoepidemiology (ISPE) through its Public Policy Committee. First issued in 1996 and revised in 2004, 2007, and 2015 (the current "fourth version"), GPP is substantive practice guidance spanning the entire study lifecycle: protocol development, responsibilities and qualifications of personnel, study conduct, data-quality assurance, statistical analysis, documentation and archiving, privacy/ethics, conflict-of-interest disclosure, adverse-event reporting, and communication of results. It is deliberately design-agnostic — it governs cohort, case-control, self-controlled, drug-utilization, and database studies alike — and sits alongside ENCePP's Guide on Methodological Standards as one of the two foundational good-practice references for the field. GPP is not a reporting checklist (that is STROBE/RECORD-PE) and not a risk-of-bias tool (that is ROBINS-I); it is the operating standard for how the work is done.
When to use
— Treat GPP as the default standard of conduct for any non-interventional or hybrid pharmacoepidemiologic study, regardless of data source or submission target: FDA or EMA regulatory submissions and post-authorization safety/effectiveness studies (PASS), HTA/payer dossiers, and peer-reviewed publication. Apply it from the moment a question is framed — before data access — to discipline protocol pre-specification, personnel qualifications, and a data-management/archiving plan. Decision rule for siblings: use GPP for overarching conduct and quality-system expectations; reach for ENCePP's Guide/Checklist when the work is an EU-regulated PASS or you need the agency-facing methodological-standards mapping; reach for HARPER / StaRT-RWE when you need a structured protocol template with study-design diagrams and pre-specified parameter tables; and reach for STROBE/RECORD-PE at the reporting stage for the manuscript checklist and attrition flow diagram. These are complementary, not interchangeable: GPP tells you to pre-specify, archive, and quality-assure; the templates tell you exactly which fields to fill; the reporting checklists tell you what to disclose in print.
What it requires
— GPP enforces good practice across domains that, for real-world data, map directly onto the hardest design decisions: (1) design transparency — an a-priori written protocol stating objectives, the causal/descriptive question, design, and analysis plan, amended with version control rather than rewritten; (2) data fitness-for-use — documented assessment of whether the source (claims, EHR, registry, linked) can capture the exposures, outcomes, and covariates required, including provenance, completeness, lags, and validation; (3) operational definitions and phenotype/algorithm validation — explicit, reproducible exposure and outcome algorithms with reported performance (e.g., PPV) where feasible; (4) time-zero alignment and follow-up rules that avoid immortal time; (5) estimands and the analytic contrast, with pre-specified handling of intercurrent events (switching, discontinuation, death); (6) confounding control strategy declared a priori; (7) attrition and missing-data handling with a transparent flow from source to analytic cohort; (8) sensitivity and quantitative bias analysis to probe key assumptions; and (9) documentation, code/algorithm versioning, and long-term archiving so the study is auditable and reproducible. GPP frames these as quality-system obligations — who is responsible, what is documented, and what is retained — rather than as a manuscript checklist.
When NOT to use — limitations and common misapplications
— The dominant reviewer-facing error is category confusion. GPP is not a reporting checklist: you cannot "complete GPP" item by item to satisfy a journal's reporting requirements — that is STROBE or RECORD-PE, and submitting a "GPP checklist" in place of a RECORD-PE flow diagram will be rejected. GPP is also not a risk-of-bias / critical-appraisal instrument: it does not score study quality the way ROBINS-I rates confounding, selection, and measurement bias, so it cannot be cited as the appraisal tool in an evidence synthesis. Nor is GPP a protocol template — pointing to GPP does not relieve you of producing the structured design and parameter tables that HARPER / StaRT-RWE (or the ENCePP protocol shell for EU PASS) provide. Adherence to GPP does not make a study causal or unbiased: a beautifully documented, fully archived study can still be wrecked by confounding by indication, immortal-time bias, or an unvalidated outcome algorithm — GPP governs process and transparency, not the validity of any single design choice. Finally, avoid GPP-as-theater: a generic statement that the study "followed GPP" with no protocol version, no data-fitness assessment, no phenotype validation, and no archiving plan is the failure mode senior regulatory and HTA reviewers flag first.
How it maps to this catalog
— GPP's good-practice domains are implemented by specific concepts in this repository. Design transparency and pre-specification → picots-framework-rwe and estimands-ate-att-intercurrent-events-rwe (the estimand and intercurrent-event handling GPP demands a priori). Data fitness-for-use → fit-for-purpose-data-assessment-rwe, with source-specific nuance in claims-analysis and medicare-ffs-ma-commercial-claims-differences-rwe. Phenotype/algorithm validation → diagnosis-phenotype-algorithm-1ip-2op-time-window-rwe. Time-zero alignment that avoids immortal time → time-zero-index-date-alignment-rwe and the drug-free lookback in washout-clean-lookback-period-rwe. Confounding control → the active-comparator-new-user design and high-dimensional-propensity-score-hdps-rwe, with the trial-emulation scaffold in target-trial-emulation. Attrition/missing data → the source-to-analytic flow in attrition-and-loss-to-follow-up-rwe. Sensitivity / quantitative bias analysis → e-value-sensitivity-analysis and quantitative-bias-analysis-toolkit-rwe. In practice for a claims/EHR/registry study, satisfying GPP means: write the protocol and lock PICOTS and the estimand before pulling data; run a fit-for-purpose assessment of the source; define and (where possible) validate exposure and outcome phenotypes; set time zero at first qualifying fill after a documented washout; pre-specify the confounding-control approach; report the full attrition funnel; run pre-planned sensitivity and bias analyses; and archive code, code lists, and protocol versions so an auditor can reproduce the result.