PRECIS-2 (Pragmatic-Explanatory Continuum Indicator Summary 2)

What it is

PRECIS-2 (Pragmatic-Explanatory Continuum Indicator Summary 2) is a structured design-and-appraisal tool that characterizes how pragmatic (oriented to usual-care decision-making across diverse settings) or how explanatory (oriented to maximizing the chance of detecting a mechanistic effect under ideal conditions) a trial is, across nine domains: (1) Eligibility — how closely participants resemble those who would receive the intervention in usual care; (2) Recruitment — how participants are identified and enrolled (routine appointments vs. dedicated recruitment drives); (3) Setting — how closely the care setting matches usual practice; (4) Organisation — the expertise and resources delivering the intervention; (5) Flexibility: delivery — how much latitude practitioners have in how the intervention is delivered; (6) Flexibility: adherence — how much latitude participants have, and what is done to enforce adherence; (7) Follow-up — how closely follow-up intensity matches usual care; (8) Primary outcome — how directly relevant the outcome is to participants; (9) Primary analysis — the extent to which all data are included (e.g., intention-to-treat) versus restricted (per-protocol). Each domain is rated 1 (very explanatory) to 5 (very pragmatic) and plotted on a nine-spoke wheel, forcing designers to justify each score against the trial's purpose. It is not a reporting checklist and not a risk-of-bias instrument. PRECIS-2 was developed by Loudon, Treweek, Zwarenstein, and colleagues (Loudon et al., BMJ 2015), revising the original PRECIS (Thorpe et al., 2009); guidance, the scoring wheel, and worked exemplars are maintained at the open www.precis-2.org toolkit.

When to use

Use PRECIS-2 prospectively at the design stage to align each design choice with the decision the trial is meant to inform: a regulatory efficacy trial supporting marketing authorization sits toward the explanatory end, whereas a comparative-effectiveness or implementation trial intended to inform HTA, payer, and routine-practice decisions should be deliberately pragmatic on the domains that matter. It is equally useful retrospectively to appraise where a published trial sits and to judge whether its conditions support transporting the result to a real-world target population. Decision rule for picking the right tool: PRECIS-2 answers "how applicable to usual care is this trial's design?" For reporting a pragmatic trial, use the CONSORT extension for pragmatic trials instead; for randomized-trial risk of bias, use RoB 2; for non-randomized/observational RWE bias, use ROBINS-I. PRECIS-2 complements but does not replace any of these. In RWE work it is most relevant to pragmatic randomized trials, hybrid effectiveness-implementation trials, and registry-based randomized trials (RRCTs) that randomize within a claims/EHR/registry infrastructure.

What it requires

PRECIS-2 requires the design team to (a) fix the intended applicability question first — name the population, setting, and decision-maker the results must serve, because a domain score is only meaningful relative to that purpose; (b) score all nine domains independently, with a written rationale per domain, ideally by several raters to expose disagreement; (c) plot the wheel and inspect for unintended explanatory pull (e.g., a pragmatic eligibility criterion undercut by an explanatory, resource-intensive follow-up schedule that no real clinic would run); and (d) iterate the protocol so the design profile matches intent. In a registry/EHR-embedded trial the domains map onto operational RWD decisions: Eligibility and Recruitment depend on the phenotype algorithm and the database feasibility funnel; Setting and Organisation depend on the data partners; Follow-up depends on routinely captured encounters rather than study visits; Primary outcome depends on an outcome algorithm with known PPV/sensitivity; and Primary analysis depends on the estimand (intention-to-treat vs. per-protocol) and how intercurrent events are handled.

When NOT to use — limitations and common misapplications

(1) It is not a quality or validity score. A high pragmatic score is not "better" — pragmatism is a design target, not a virtue; a high overall rating does not mean the trial is well conducted or low-risk-of-bias. Treating the wheel as a quality grade is the most common abuse. (2) It does not assess bias. A pragmatic trial can still be confounded, unblinded inappropriately, or underpowered; do not substitute PRECIS-2 for RoB 2 or ROBINS-I. (3) It is not a reporting checklist — completing the wheel does not satisfy CONSORT or its pragmatic extension. (4) It applies to trials, not to purely observational designs. PRECIS-2 presumes a randomized comparison; do not apply it to a single-arm or non-randomized cohort study as if the domain scores conferred causal credibility. (5) Scores are judgment-based and rater-dependent; a single unblinded rater rationalizing a funded protocol produces theater. (6) It does not, by itself, establish transportability — a pragmatic profile makes generalization more plausible but does not replace formal effect-transport or external-validity analysis to a named target population.

How it maps to this catalog

When PRECIS-2 is used to design or appraise a registry/EHR-embedded pragmatic or hybrid trial, the catalog concepts below implement the operational requirements behind each domain: - Eligibility / Recruitment → implemented by diagnosis-phenotype-algorithm-1ip-2op-time-window-rwe (who the RWD identifies as eligible) and database-feasibility-attrition-funnel-rwe (how many survive each criterion); active-comparator-new-user disciplines the comparison when the pragmatic trial randomizes between two active treatments. - Follow-up / attrition → attrition-and-loss-to-follow-up-rwe quantifies how pragmatic, routine-care follow-up trades completeness for realism. - Primary outcome → claims-outcome-algorithm-ppv-sensitivity-rwe establishes whether the routinely captured outcome is measured well enough to support the pragmatic intent. - Primary analysis → estimands-ate-att-intercurrent-events-rwe and estimand-analysis-traceability-rwe formalize the intention-to-treat vs. per-protocol choice the Primary Analysis domain scores, and how intercurrent events (switching, discontinuation) are handled. - Design backbone / transportability → target-trial-emulation provides the protocol scaffold a registry trial mirrors, and medicare-ffs-ma-commercial-claims-differences-rwe flags when the data source's population limits the Setting/Eligibility pragmatism that can honestly be claimed; residual confounding in pragmatic comparisons can be probed with empirical-calibration-negative-controls-rwe.

Applied note (registry-based RCT in claims/EHR). For a pragmatic RRCT embedded in linked claims-EHR data, score Eligibility/Recruitment against the actual phenotype-derived screening funnel (not the protocol's aspirational criteria), score Follow-up against routinely captured encounters rather than study visits, and score Primary Outcome against the measured PPV/sensitivity of the outcome algorithm. A trial can advertise pragmatic eligibility yet collapse to an explanatory profile once an outcome algorithm requires confirmatory testing only done at academic centers — the wheel makes that contradiction visible before the protocol is locked.