TREND Statement: Reporting Non-Randomised Behavioural Interventions

The TREND statement (Transparent Reporting of Evaluations with Nonrandomized Designs) is a 22-item reporting checklist for evaluations of behavioural and public-health interventions that use a nonrandomised design — quasi-experimental, cohort, before-after, or other designs where participants were not randomly assigned to condition. It was developed by Don C. Des Jarlais, Cynthia Lyles, Nicole Crepaz and the TREND Group, working through the CDC’s HIV Prevention Research Synthesis Project, and published in the American Journal of Public Health in March 2004 (Des Jarlais et al., AJPH 94(3):361–366).

TREND is explicitly built on the CONSORT statement for randomised controlled trials, extending and modifying it for the specific reporting problems a nonrandomised design creates: how the assignment method actually worked, whether the units being compared are equivalent at baseline, and how the analysis accounts for a unit-of-assignment/unit-of-analysis mismatch. Three items in particular — assignment method, unit-of-analysis handling, and baseline-equivalence adjustment — are, in the original paper’s own framing, “extremely relevant to nonrandomized designs” and have no direct CONSORT counterpart.

The 22-item TREND checklist

The checklist is organised by the same section structure as a standard paper — title/abstract, introduction, methods, results, discussion — so it can be worked through alongside the manuscript itself.

#	Section	What to report
1	Title/Abstract	How participants were allocated to conditions, and the target population.
2	Introduction	The theory or theories used to design the behavioural intervention.
3	Methods – Participants	Eligibility criteria, recruitment method, sampling approach, and recruitment setting.
4	Methods – Interventions	Content, delivery method, unit of delivery, who delivered it, setting, exposure quantity/duration, time span, and any activities to increase compliance.
5	Methods – Objectives	Specific objectives and hypotheses.
6	Methods – Outcomes	Primary and secondary outcome measures, how and when they were collected, and information on any instrument’s validity/reliability.
7	Methods – Sample size	How sample size was determined, and any interim analyses or stopping rules.
8	Methods – Assignment method	The unit of assignment (individual, group, community) and the method used to assign units to conditions — the first of the three items with no direct CONSORT counterpart.
9	Methods – Blinding/masking	Whether outcome assessors, participants, or those delivering the intervention were blinded, and how.
10	Methods – Unit of analysis	The unit of analysis, and how any mismatch between the unit of assignment and the unit of analysis was handled — the second nonrandomised-specific item.
11	Methods – Analytical methods	Statistical methods for primary and secondary outcomes, how missing data were handled, and software used.
12	Results – Participant flow	A flow diagram of participants through each stage, and any protocol deviations with reasons.
13	Results – Recruitment	Dates defining recruitment and follow-up periods.
14	Results – Baseline data	Baseline demographic and clinical/behavioural characteristics of each condition.
15	Results – Baseline equivalence	Baseline comparability across conditions and, where groups differ, the statistical methods used to adjust for it — the third nonrandomised-specific item, since there is no randomisation to make this a formality.
16	Results – Numbers analysed	The number of participants included in each analysis, for each outcome.
17	Results – Outcomes and estimation	Results for each outcome, effect size and confidence interval, and results of any test of the intervention’s theorised causal pathway.
18	Results – Ancillary analyses	Any other analyses performed, including subgroup and exploratory analyses, identified as such.
19	Results – Adverse events	Important adverse events or unintended effects in each condition.
20	Discussion – Interpretation	Interpretation of the results accounting for possible sources of bias, imprecision, and how faithfully the intervention was actually implemented.
21	Discussion – Generalisability	The external validity of the findings — who and what settings the results can reasonably be generalised to.
22	Discussion – Overall evidence	General interpretation of the results in the context of the current evidence.

Item groupings verified directly against the primary publication record (Des Jarlais, Lyles, Crepaz and the TREND Group, “Improving the Reporting Quality of Nonrandomized Evaluations of Behavioral and Public Health Interventions: The TREND Statement,” American Journal of Public Health 94(3), March 2004, pp. 361–366), cross-checked 2026-08-31.

Why a separate checklist for nonrandomised designs

CONSORT assumes random assignment did the work of balancing the comparison groups, so its checklist doesn’t need an item asking authors to demonstrate baseline equivalence — randomisation is supposed to make that a formality (subject to CONSORT’s own reporting of the randomisation process itself). A nonrandomised evaluation has no such guarantee: participants who select or are selected into a programme condition can differ systematically from a comparison group before the intervention ever starts. TREND’s items 8, 10 and 15 exist specifically to force that gap into the open — how units were assigned, whether the unit used for assignment matches the unit used for analysis, and what the two groups actually looked like at baseline, adjusted or not. A reader can only judge whether an observed effect reflects the intervention or a pre-existing difference between groups if all three are reported explicitly.

This matters for the same reason CASRAI’s broader reporting-guideline coverage matters: a checklist is only useful if it is applied to the design it was built for. Using CONSORT language (“randomisation”) to describe a quasi-experimental assignment process, or omitting the baseline-equivalence item because a CONSORT-trained reviewer doesn’t expect to see it, is a common and avoidable reporting gap in nonrandomised public-health and behavioural evaluation literature.

How TREND relates to other reporting guidelines

Guideline	Scope	Relationship to TREND
CONSORT	Randomised controlled trials	TREND is explicitly adapted from CONSORT; most items map directly, with items 8, 10 and 15 added or substantially reworked for the absence of randomisation.
STROBE	Observational studies (cohort, case-control, cross-sectional)	Overlapping territory — a nonrandomised programme evaluation and an observational cohort study can look similar. Use TREND when the paper is reporting an evaluation of a deliberately delivered intervention; use STROBE when it’s reporting an association study with no intervention being evaluated.
PRISMA 2020	Systematic reviews and meta-analyses	A systematic review that pools nonrandomised intervention evaluations reports itself under PRISMA, while using TREND as a lens for assessing how completely each included study was reported.

For a fuller map of which reporting guideline applies to which study design, see CASRAI’s EQUATOR Network guide. TREND is listed in the EQUATOR Network’s own library of reporting guidelines. Related design background: Experimental Design.

Using TREND when writing up a nonrandomised evaluation

1. Work through all 22 items, not just the ones that feel novel. Most of the checklist parallels CONSORT closely enough that authors already familiar with trial reporting can move through it quickly — the value is in not skipping the three assignment/analysis/baseline items that CONSORT experience doesn’t prepare you for.
2. Name the assignment method precisely (item 8). “Non-randomised” is not itself an assignment method. State how units actually ended up in each condition — self-selection, staged rollout by site, administrative assignment, matching — since the credibility of the whole evaluation turns on this.
3. State the unit of assignment and unit of analysis explicitly, even when they’re the same (item 10). When a community, clinic, or classroom is the unit assigned to condition but individuals within it are the unit analysed, say so and describe how clustering was handled statistically; when they match, a one-line confirmation is enough.
4. Report baseline equivalence as data, not as an assertion (item 15). A table comparing conditions on the characteristics most likely to confound the outcome, plus whichever adjustment method was used for anything that differed, does the work a randomisation flow diagram does in a CONSORT paper.
5. Report items 12 and 19 (participant flow and adverse events) even when nothing unusual happened. Explicitly stating “no adverse events were identified” or that flow matched the sampling frame is still the complete, transparent answer TREND asks for.

Frequently asked questions

Does TREND replace CONSORT?

No. TREND is for evaluations that used a nonrandomised design. If participants were randomly assigned to condition, report under CONSORT instead — TREND’s own basis is CONSORT, not a substitute for it.

Is TREND only for HIV/public-health research?

It originated from CDC’s HIV Prevention Research Synthesis Project, but the checklist itself is written generally for nonrandomised evaluations of behavioural and public-health interventions, and is used well beyond HIV research — in health-services, education, and community-intervention evaluation literature.

What’s the difference between TREND and STROBE for a nonrandomised study?

The distinction is whether the paper is evaluating a delivered intervention. TREND applies when a specific programme or intervention was implemented and evaluated without randomisation; STROBE applies to observational studies more broadly, including ones with no intervention being evaluated at all (e.g. a pure exposure-outcome association study).

Do journals require TREND the way many require CONSORT?

Journal policy varies. Check your target journal’s Instructions for Authors — some public-health and behavioural-science journals name TREND explicitly for nonrandomised evaluation submissions, following the same pattern as CONSORT/STROBE/PRISMA adoption elsewhere.