SPIRIT (Standard Protocol Items: Recommendations for Interventional Trials)

What it is

— SPIRIT (Standard Protocol Items: Recommendations for Interventional Trials) is a reporting guideline that defines the minimum set of items a protocol for a randomized or otherwise interventional clinical trial should contain, so that the trial's rationale, design, methods, and oversight are completely and transparently pre-specified before the trial begins. The original SPIRIT 2013 Statement (Chan, Tetzlaff, Altman et al.) provided a 33-item checklist plus a figure for the schedule of enrolment, interventions, and assessments; it was accompanied by an item-by-item explanation-and-elaboration paper, and the SPIRIT 2025 update (Chan et al.) modernized it to incorporate ICH E9(R1) estimands, structured harms reporting, open science (data/code/protocol sharing), equity/diversity of trial populations, and a new item set for AI interventions. SPIRIT is hosted and maintained as a reporting guideline within the EQUATOR Network and is the protocol-stage member of the trial-reporting family: SPIRIT governs what you commit to in advance, whereas CONSORT governs how you report the completed trial. Together they form a continuous chain of pre-specification-to-reporting that journals, registries, ethics committees, and regulators increasingly require.

When to use

— Apply SPIRIT whenever you are writing, registering, or appraising the protocol of an interventional study — a parallel-group, crossover, cluster, factorial, or adaptive randomized trial — across any setting: a regulatory IND/CTA package, an ethics/IRB submission, a funder application, a trial-registry record (e.g., ClinicalTrials.gov, EudraCT/CTIS), or a protocol manuscript. In the real-world-evidence space SPIRIT applies to the pragmatic randomized trial (where eligibility is broad and outcomes are often ascertained from EHR, claims, or registries) and the registry-based randomized trial / RRCT (where randomization is layered onto an existing registry data infrastructure); in externally-controlled or hybrid designs, SPIRIT governs the interventional arm's protocol while a separate observational protocol governs the external real-world comparator. Decision rule for the right family member: use SPIRIT for the trial protocol and CONSORT for the finished trial report; choose the matching SPIRIT extension by feature — SPIRIT-PRO when patient-reported outcomes are a focus, SPIRIT-AI for AI-based interventions, SPIRIT-Outcomes 2022 for rigorous outcome specification, and SPIRIT extensions for cluster/pilot designs where they exist. SPIRIT is not the checklist for a non-randomized observational study protocol — that is the province of HARPER, the ENCePP checklist, or STROBE/RECORD-PE — and it is not for systematic-review protocols, which use PRISMA-P.

What it requires

— SPIRIT compels pre-specification of the protocol elements that otherwise drift or get reconstructed post hoc: administrative information (title, trial registration, protocol version and amendment history, funding, roles, and committees); a scientific background and rationale with explicit objectives and hypotheses; the trial design (allocation ratio, framework — superiority/non-inferiority/equivalence); a participants/interventions/outcomes core (eligibility criteria, intervention description and adherence provisions, and a clearly defined primary outcome with measurement timing); sample size justification and recruitment plan; the methods that protect against bias — sequence generation, allocation concealment, and blinding; data collection, management, and analysis plans, including how missing data and intercurrent events are handled (SPIRIT 2025 frames the analysis around a formal estimand); harms monitoring; data monitoring committee arrangements and interim analyses; and the governance items — research ethics approval, consent, confidentiality, declaration of interests, and data-access/dissemination (including open-science commitments to share protocol, data, and code). For RWE-flavored trials these generic items carry specific weight: a pragmatic or registry-based trial that ascertains outcomes from EHR/claims/registry data must, in the protocol, specify the outcome/phenotype algorithm and its validity, the data source's fitness for use for the endpoint, and how attrition / loss to follow-up is defined and analyzed when "follow-up" is the continued presence of data — even though randomization, not design adjustment, is the engine of causal inference.

When NOT to use — limitations and common misapplications

— SPIRIT is a reporting checklist for a protocol; it is not a risk-of-bias instrument, not a quality score, and not a substitute for sound design. Concrete failure modes: (1) Wrong guideline entirely — applying SPIRIT to a non-randomized observational RWE study (a claims or EHR cohort/case-control study); those protocols follow HARPER, ENCePP, or RECORD-PE, and reporting follows STROBE/RECORD-PE — SPIRIT does not apply. (2) Protocol/report confusion — using SPIRIT to report a completed trial (use CONSORT) or citing CONSORT for the protocol. (3) Wrong extension — ignoring SPIRIT-PRO for a PRO-centered trial, or SPIRIT-AI for an AI intervention, leaving the special items unaddressed. (4) Mistaking completeness for validity — a fully SPIRIT-compliant protocol can still describe an underpowered, poorly randomized, or unblinded trial; SPIRIT documents what you plan, it does not certify that the plan is good. (5) Checklist-as-theater — ticking 33 items while leaving the estimand, the primary outcome, the blinding procedure, or the missing-data plan vague defeats the purpose; the value is genuine pre-specification, not page count. (6) Registration ≠ compliance — a registry record is necessary but not the same as a complete protocol; keep the registration synchronized with protocol amendments. (7) Importing observational machinery inappropriately — a randomized trial does not generally need propensity scores or active-comparator design framing; conflating SPIRIT's needs with those of a confounded observational study is a category error.

How it maps to this catalog

— In this repo, the SPIRIT-relevant concepts cluster around protocol discipline, estimands, and the narrow points where randomized trials touch real-world data: - The pre-specification discipline itself: study-protocol-or-sap-elements supplies the version-control, amendment, and pre-specified-analysis habits SPIRIT's administrative and methods items demand; picots-framework-rwe operationalizes the population/intervention/comparator/outcome/timing/setting frame SPIRIT requires for objectives and eligibility. - The analysis spine (SPIRIT 2025): estimands-ate-att-intercurrent-events-rwe implements the ICH E9(R1) estimand and intercurrent-event thinking the updated checklist now expects in the statistical-methods item. - The design contrast: target-trial-emulation is the conceptual mirror image — SPIRIT governs the protocol of an actual randomized trial, whereas target-trial emulation specifies an observational analysis to emulate such a trial when randomization is infeasible; reading them side by side clarifies which assumptions randomization buys for free. - The RWD-ascertainment caveats (only for pragmatic/registry trials that capture endpoints from EHR/claims): diagnosis-phenotype-algorithm-1ip-2op-time-window-rwe for how an algorithm-defined outcome is specified and validated, attrition-and-loss-to-follow-up-rwe for defining and analyzing dropout/data-availability, and claims-analysis for the operational realities of the underlying data source. These apply because the trial borrows a real-world data substrate, not because the trial is observational. Note what does not map: confounding-control concepts such as high-dimensional-propensity-score-hdps-rwe and active-comparator-new-user are observational-design tools; randomization, properly executed under SPIRIT, is what removes confounding, so those concepts are deliberately out of scope for a SPIRIT protocol.

Applied note (claims/EHR/registry RWE)

For a pragmatic or registry-based randomized trial whose endpoints are ascertained from administrative or registry data, a SPIRIT-compliant protocol should go beyond the generic items: name the data source and justify its fitness for use for the specific endpoint, pre-specify the outcome algorithm (e.g., a validated 1-inpatient-or-2-outpatient phenotype with its time window) and its expected sensitivity/positive predictive value, state how time alignment between randomization and data capture is handled, and define loss to follow-up as loss of data availability (disenrollment, registry exit) with a pre-specified, estimand-aligned analysis for the resulting missingness — so that the trial's real-world ascertainment layer is as transparent and auditable as its randomization.