HARPER

What it is

HARPER — the HARmonized Protocol Template to Enhance Reproducibility of hypothesis-evaluating real-world evidence (RWE) studies on treatment effects — is a structured protocol template developed by a joint ISPE/ISPOR (International Society for Pharmacoepidemiology / International Society for Pharmacoeconomics and Outcomes Research) task force and published as a Good Practices report in 2022 (concurrently in Pharmacoepidemiology and Drug Safety and Value in Health). It is not a brief reporting checklist; it is a fill-in protocol skeleton combining narrative text, standardized tables, and a study-design diagram that compels authors to state every parameter a reader needs to reproduce the study: design choice, eligibility, exposure and comparator operationalization, outcome algorithms, time-zero (index) alignment, covariate windows, the estimand and its intercurrent-event handling, the analysis specification, and the sensitivity analyses. HARPER is the protocol-stage successor to and harmonization of STaRT-RWE (the Structured Template for plAnning and Reporting on the implementation of RWE studies, Wang et al., BMJ 2021); the two share the same DNA and the task force intends HARPER to be the single template for both planning and reporting transparency. It is maintained by the ISPE/ISPOR task force authorship and promoted through both societies and, in the US policy context, referenced alongside CMS/FDA expectations for transparent RWE.

When to use

Use HARPER whenever you are designing or registering a hypothesis-evaluating (confirmatory, comparative effectiveness or safety) non-interventional study of a treatment effect using routinely collected data — claims, EHR, registry, or linked sources — and the study is destined for a regulatory submission (FDA RWE program, EMA, an imposed or voluntary PASS), an HTA/payer dossier, or a high-impact peer-reviewed journal. It is the right instrument at the protocol stage, ideally before data access and certainly before any outcome-dependent analytic choices are made; registering a completed HARPER protocol is the strongest available defense against accusations of data-driven specification. Decision rules for choosing HARPER over a sibling: use HARPER (or its predecessor STaRT-RWE) when the study estimates a treatment effect and you need a full protocol template; use STROBE/RECORD/RECORD-PE instead when your task is the final reporting of an observational study in a manuscript (those are reporting checklists, not protocol templates); use the ENCePP Checklist for Study Protocols in parallel when an EU PASS or ENCePP seal is in scope (HARPER organizes the science, ENCePP confirms regulatory governance); use SPIRIT for interventional trial protocols and CHEERS for economic evaluations — HARPER is for descriptive-of-effect, non-interventional designs, not RCTs or cost-effectiveness models. HARPER is generally not the tool for purely descriptive epidemiology (incidence, prevalence, utilization), where a lighter design description suffices.

What it requires

HARPER's tables and narrative force substantive content, mapped here to real-world-data realities: (1) Design transparency — explicit design label, a study-design diagram on a calendar timeline (assessment, washout, baseline, and follow-up windows drawn relative to time zero), and a PICOTS-style framing of the question. (2) Data fitness-for-use — naming the data source(s), provenance, capture mechanism, known coverage gaps, and a justification that the data can actually measure the exposure, outcome, and confounders required. (3) Exposure/comparator operationalization — code lists, grace periods, stockpiling rules, and (for comparative work) a defensible active comparator. (4) Outcome phenotype/algorithm validation — the operational definition and its validation metrics (PPV/sensitivity) or a plan to obtain them. (5) Time-zero alignment — index-date definition that aligns eligibility, treatment assignment, and start of follow-up to avoid immortal time. (6) Estimand and intercurrent events — the target population, treatment strategies being contrasted, and the pre-specified strategy for intercurrent events (treatment switching, discontinuation, death). (7) Confounding control — covariate measurement windows and the adjustment method (e.g., propensity or high-dimensional propensity scores). (8) Attrition and missing data — an attrition (CONSORT-style) accounting and a missing-data plan. (9) Sensitivity / quantitative bias analysis — pre-specified robustness checks (washout/grace-period variants, negative controls, E-value or other bias analysis) and versioned code lists so the analysis can be reproduced exactly.

When NOT to use — limitations and common misapplications

HARPER is a transparency and reproducibility instrument, not a risk-of-bias tool and not a quality score: a fully completed HARPER protocol can describe a badly confounded study with great clarity. Do not treat a filled-in template as evidence the study is valid, nor as a substitute for a formal bias assessment (ROBINS-I) or for the design thinking of target-trial emulation — completing the template does not make an observational estimate causal; it only makes the (possibly biased) design legible. Common failure modes: template-as-theater — pasting boilerplate into the cells without genuine pre-specification, then changing analyses after seeing results; using a reporting checklist (STROBE) where a protocol template (HARPER/STaRT-RWE) was needed, or vice-versa; using HARPER for an interventional trial (use SPIRIT) or an economic model (use CHEERS); and omitting the data-fitness and phenotype-validation cells, which are exactly the parts regulators and HTA reviewers scrutinize. HARPER also does not itself govern EU PASS regulatory process — pair it with the ENCePP checklist rather than assuming HARPER satisfies that requirement.

How it maps to this catalog

HARPER is an organizing frame; the concepts in this catalog implement each cell. Its design-and-estimand spine is implemented by target-trial-emulation (pre-specify the hypothetical trial before emulating it), active-comparator-new-user (the new-user + active-comparator + time-zero structure that fills the eligibility/exposure/index tables), and estimands-ate-att-intercurrent-events-rwe (the estimand and intercurrent-event cells). The data-fitness cell is implemented by fit-for-purpose-data-assessment-rwe and, for US claims nuance, claims-analysis and medicare-ffs-ma-commercial-claims-differences-rwe (FFS vs MA capture and coding-intensity differences that determine whether "no prior claim" is a true washout). Outcome/exposure operationalization maps to diagnosis-phenotype-algorithm-1ip-2op-time-window-rwe (1-inpatient / 2-outpatient rules, time windows, PPV) and time-zero-index-date-alignment-rwe. Confounding control maps to high-dimensional-propensity-score-hdps-rwe and propensity-score-methods-psm-iptw. Attrition and robustness map to attrition-and-loss-to-follow-up-rwe, e-value-sensitivity-analysis, and quantitative-bias-analysis-toolkit-rwe. For claims/EHR/registry RWE specifically: a defensible HARPER protocol pre-registers the diagnosis/outcome algorithm with its validation metrics, requires continuous-enrollment and data-capture conditions so absence-of-fill is observed rather than missing, draws the calendar-timeline design diagram so immortal time is visibly excluded, and versions every code list — so the eventual STROBE/RECORD-PE manuscript can be checked back against a protocol that was fixed before the analyst saw an effect estimate.