TREND (Transparent Reporting of Evaluations with Nonrandomized Designs)

What it is

— TREND (Transparent Reporting of Evaluations with Nonrandomized Designs) is a 22-item reporting checklist developed by the Centers for Disease Control and Prevention (CDC) HIV/AIDS Prevention Research Synthesis (PRS) team and published as the TREND Statement (Des Jarlais, Lyles, Crepaz et al., American Journal of Public Health, 2004). It is maintained and indexed within the EQUATOR Network reporting-guideline library. TREND was created to fill a specific gap: CONSORT governs the reporting of randomized trials, but a large share of behavioral and public-health intervention evaluations cannot randomize (ethical, logistical, or community-level constraints) and are evaluated with non-randomized designs. TREND is the nonrandomized-design companion to CONSORT for primary intervention-evaluation reports. Its purpose is transparency of how the intervention, comparison condition, assignment mechanism, and analysis were actually carried out, so readers can judge the threats to internal validity that randomization would otherwise have addressed.

When to use

— Apply TREND when you are reporting a primary evaluation of a behavioral, social, educational, or public-health intervention that used a non-randomized design — quasi-experimental, pre-post, controlled-before-after, interrupted time series, or otherwise non-randomly assigned groups. Its native habitat is the peer-reviewed public-health/behavioral-science literature (HIV/STI prevention, harm reduction, vaccination and screening uptake, school- or community-based programs, health-promotion campaigns). Decision rule for picking the right checklist: if the intervention evaluation was randomized, use CONSORT (or its extensions: cluster, pragmatic, PRO); if it is a non-randomized intervention evaluation, use TREND; if you are reporting an observational etiologic/comparative study with no investigator-assigned intervention (a drug-vs-drug claims cohort, a case-control study, a registry analysis), use STROBE and, when the data are routinely-collected health data, RECORD / RECORD-PE; for a pharmacoepidemiology protocol use HARPER or the ENCePP checklist; for an intervention protocol, use SPIRIT. TREND governs the report, not the protocol, and the intervention evaluation, not the etiologic observational study.

What it requires

— TREND's 22 items mirror the CONSORT skeleton but add domains that matter precisely because there was no randomization. They compel reporting of: the title/abstract structured to flag the non-randomized design; the theory or behavioral model underpinning the intervention; eligibility and the setting/location of recruitment; a detailed description of the intervention and the comparison condition (content, delivery, provider, dose/intensity, fidelity), which is the heart of TREND and the item most often done badly; explicitly stated objectives, hypotheses, and outcomes with how and when each was measured; sample size justification; the unit of assignment and the method by which groups were formed (the non-random assignment mechanism — the item that replaces CONSORT's randomization items and is where confounding/selection threats live); blinding where feasible; the analytic methods, including methods used to control for confounding introduced by non-random assignment; participant flow, recruitment, and losses/exclusions at each stage (attrition); baseline group comparability; numbers analyzed and the basis (e.g., intention-to-treat-like vs as-treated); estimated effects with precision; ancillary analyses; and a discussion that interprets results in light of the non-randomized design and its specific biases. The substantive emphasis is therefore: intervention/comparator description and fidelity, the assignment mechanism and consequent confounding control, time-zero/intervention-start alignment, attrition across the participant-flow stages, and an honest accounting of internal-validity threats.

When NOT to use — limitations and common misapplications

— (1) TREND is a reporting checklist, not a risk-of-bias instrument and not a quality score. Completing all 22 items does not certify that the evaluation is internally valid or that the estimate is causal; appraisal of a non-randomized study is done with ROBINS-I, and a fully TREND-compliant paper can still report a hopelessly confounded comparison. (2) Wrong checklist for drug/device pharmacoepidemiology — the single most common misapplication is reaching for TREND (or its scaffold framing as a generic "non-randomized RWE" guideline) when reporting a claims/EHR comparative-effectiveness or safety cohort. Those are observational etiologic studies of routinely-collected data, not investigator-assigned behavioral-intervention evaluations; they require STROBE + RECORD / RECORD-PE, with HARPER/ENCePP at the protocol stage — not TREND. (3) Wrong sibling for a randomized evaluation — if the behavioral intervention was in fact randomized (including cluster- or stepped-wedge designs), use CONSORT/CONSORT-Cluster, not TREND. (4) Checklist-as-theater — ticking the intervention-description item while omitting dose, fidelity, or the actual assignment mechanism defeats the entire purpose, which is to expose exactly the design features randomization would have neutralized. (5) It does not replace the analysis — TREND requires that confounding control be reported; it does not tell you which method to use, and reporting transparency is necessary but not sufficient for a valid causal claim.

How it maps to this catalog

— TREND is an intervention-evaluation reporting checklist, so only a subset of the catalog's pharmacoepidemiology concepts genuinely implement its items; the honest mapping is narrow on purpose: - Intervention/comparator and assignment, attrition, time-zero (items that DO apply): estimands-ate-att-intercurrent-events-rwe sharpens the objective/outcome and intention-to-treat-vs-as-treated reporting TREND demands; time-zero-index-date-alignment-rwe operationalizes alignment of follow-up at intervention start; attrition-and-loss-to-follow-up-rwe implements the stage-by-stage participant-flow and losses items; generalizability-transportability-external-validity-rwe supports the discussion of how far the setting-bound estimate transports. - Confounding from non-random assignment: when the evaluation has a comparison group formed non-randomly, difference-in-differences-staggered-adoption-rwe is the design most aligned with TREND's quasi-experimental territory (pre-post with a comparison series), and target-trial-emulation is the explicit framework for making the non-random assignment defensible. - What does NOT map (deliberately): high-dimensional-propensity-score-hdps-rwe, active-comparator-new-user, diagnosis-phenotype-algorithm-1ip-2op-time-window-rwe, and claims-analysis are STROBE/RECORD-PE territory. If you find yourself needing those concepts, you are reporting a database pharmacoepidemiology study and TREND is the wrong checklist — route to strobe, record, record-pe, or harper.

Applied note (claims/EHR/registry RWE)

Behavioral and public-health intervention evaluations increasingly read outcomes from routinely-collected data — e.g., a community vaccination-uptake program measured against HEDIS claims, or a screening intervention evaluated through EHR-captured test orders and downstream ED visits. In that hybrid case TREND still governs the intervention/comparator description, the non-random assignment mechanism, and participant flow, but it does not cover the database-specific items those data require: the diagnosis/outcome algorithm and its validity (PPV/sensitivity), the data-source/linkage and completeness reporting, and the time-window definitions. Use TREND together with RECORD / RECORD-PE for those items; do not assume TREND's 22 items suffice for an analysis whose outcomes are algorithm-defined in claims or EHR.