Why Design, Setting, and Time Belong in the Report
A treatment comparison does not take place in an abstract population. It is assembled under a particular design, in particular care settings and locations, during a particular period. Those details help readers understand who could enter the study, what care was available, and whether treatment groups were observed under comparable conditions. They also help readers decide how far the findings may apply beyond the study’s own context.
A label such as “retrospective cohort study” is useful but insufficient. It does not tell readers which institutions contributed patients, where those institutions were, when patients were recruited or identified, or whether clinical practice changed during the study. Reporting those features is part of explaining the comparison—not a substitute for evaluating its design or causal assumptions.
The introductory material on STROBE establishes the value of reporting observational study methods transparently. Here the emphasis is on making design, setting, and calendar time specific enough that readers can interpret the target estimand already stated and understand the context in which it is estimated. A reporting description should state what happened, not imply that a design label alone guarantees an unbiased result.
What to Report
Name the design using a standard term and briefly explain how it operated in this study. For example, say whether investigators assembled a cohort and compared groups according to observed treatment, sampled participants based on outcome status in a case-control design, or measured exposure and outcome at one time in a cross-sectional design. When useful, clarify whether the study was prospective or retrospective and whether it used a single site or multiple sites. Those modifiers should describe the actual conduct, not serve as decorative labels.
Then describe the setting at a useful level of detail. Depending on the question, this may include the type of care (such as primary care, hospital, or specialist outpatient care), the participating facility or health-system structure, and the country, region, or other geographic area. Mention characteristics such as referral status, urban or rural location, or a restricted catchment area when they help explain who receives care there. Balance interpretability with privacy: a small location need not be named if doing so risks identifying participants.
Dates need equally careful treatment. “The study ran from 2021 to 2023” is ambiguous unless readers know what those dates refer to. Recruitment or cohort-entry dates, treatment-initiation dates, outcome-observation dates, and the dates on which study records were collected can be different. State the relevant date ranges separately. If only years are available, say so rather than implying precise start and end dates.
Finally, report calendar-time features that could alter the comparison’s meaning. Examples include a change in clinical guidance, a new treatment becoming available, a formulary or reimbursement change, a service reorganization, a major disruption to care, or strong seasonality. A feature is worth describing when it could plausibly affect treatment availability, who receives each treatment, or the outcome being compared. The report should give its timing and relevance, not merely list historical events.
| Reporting element | Useful description | Commonly insufficient description |
|---|---|---|
| Design | Retrospective cohort assembled from treatment initiations, with outcomes observed afterward | “Observational study” |
| Care setting | Adult medical wards at three regional hospitals | “Hospital patients” |
| Geography | Hospitals serving a defined rural region in one country | “A regional sample” |
| Calendar time | Entry from March 2019 through February 2022; a guideline change occurred in July 2020 | “Data from 2019–2022” |
Calendar Time Can Change the Comparison
Calendar time matters because treatment use and outcomes can both change over time. If one treatment is used mostly in an earlier era and another mostly in a later era, the groups may differ not only in treatment but also in the care standards, patient mix, or background risks of those eras. A pooled comparison can therefore combine treatment differences with differences in when patients were treated. A report that gives only an overall study period conceals whether the two groups had similar calendar-time distributions.
A short justification makes the issue clear. Suppose the average observed outcome in each treatment group is formed by combining outcomes from two calendar periods. Each group’s overall average depends both on the period-specific averages and on the fraction of that group observed in each period. If background outcomes differ between periods, unequal fractions can produce different overall averages even when the period-specific averages are the same. Reporting the dates and any important era changes lets readers recognize this possibility; it does not by itself remove time-related confounding.
This is why a description should distinguish dates from explanations. Give the calendar boundaries, then state which relevant practice or contextual changes occurred within them and when. Do not assert that a change caused the observed treatment difference unless the analysis supports that conclusion.
Worked Example: A Retrospective Cohort in Regional Hospitals
Consider a hypothetical study comparing two medication strategies among adults admitted to four hospitals in one province. Investigators identify eligible treatment initiations from hospital records between April 1, 2018, and March 31, 2022, and observe outcomes after initiation.
A concise, informative report would say: “We conducted a retrospective cohort study of adults initiating either medication strategy during hospital admission at four acute-care hospitals in the province. Cohort entry occurred from April 1, 2018, through March 31, 2022. The hospitals serve both urban and rural communities. A provincial prescribing guideline changed in October 2020, during the cohort-entry period.”
The design description explains that participants were grouped by observed treatment initiation and followed for outcomes, rather than sampled based on outcome status. The setting gives the care level, number of facilities, and geographic scope. The entry dates have explicit boundaries, and the guideline change is placed on the calendar. These details are supported by the hypothetical study description; they do not assume that all hospitals changed practice immediately or that the guideline affected every patient. If the investigators know whether implementation varied by site, they should report that fact as well.
The report should not say merely “a four-year study” or “a provincial cohort.” Those phrases omit exact boundaries and the clinical setting. Nor should it call the cohort “representative of the province” without evidence about how patients and facilities were selected.
Worked Example: Calendar-Time Imbalance Across Treatment Groups
In a hypothetical two-era comparison, outcome risks are as follows. The earlier era has 800 patients receiving Treatment A, with 80 outcomes, and 200 receiving Treatment B, with 20 outcomes. The later era has 200 patients receiving Treatment A, with 40 outcomes, and 800 receiving Treatment B, with 160 outcomes.
| Calendar period | Treatment A | Treatment B |
|---|---|---|
| Earlier era | 80/800 = 10% | 20/200 = 10% |
| Later era | 40/200 = 20% | 160/800 = 20% |
Within each era, the hypothetical risks are equal. Pooled across eras, Treatment A has \(80+40=120\) outcomes among \(800+200=1{,}000\) patients, or \(120/1{,}000=12\%\). Treatment B also has \(20+160=180\) outcomes among \(200+800=1{,}000\) patients, or \(180/1{,}000=18\%\). The pooled difference is \(12\%-18\%=-6\) percentage points.
The overall contrast arises here because the treatment groups have different distributions across eras, while the outcome risk differs between eras. This arithmetic is an illustration, not evidence about a real treatment or proof that treatment has no effect in other settings. In a report, the necessary first step is to make the relevant dates, group-specific calendar distributions, and era changes visible. The example also shows why a single start and end year may not be enough to interpret a pooled result.
Design Labels and Alternatives
Different observational designs organize evidence differently, so the report should identify the actual approach rather than use a broad label as a substitute for description. A cohort design starts with a defined group and compares outcomes across observed treatment groups. A case-control design selects participants according to outcome status and looks back to compare prior treatment or exposure. A cross-sectional design measures exposure and outcome in a defined time frame, often without establishing which came first. These labels do not themselves establish causal validity; the details of participant selection, timing, and measurement remain important.
A prospective design may allow investigators to define procedures before outcomes occur, while a retrospective design uses information already recorded. Neither label alone ensures better measurement or less bias. A single-site study may describe its care processes precisely but have limited transportability; a multicenter study may cover a broader range of practice while requiring clear reporting of differences among sites. The appropriate description is factual and allows readers to judge relevance to their own setting.
In a comparative treatment study, study design and setting descriptions complement—not replace—the later reporting of eligibility, treatment strategies, covariates, follow-up, matching or weighting, and effect estimates. Keep the description focused here: name how the comparison was organized, where it occurred, and when relevant events took place.
Diagnostics and Common Reporting Failures
Before finalizing the report, check whether a reader could reconstruct the study’s broad setting and calendar frame without guessing. Look for vague terms such as “recent,” “routine care,” “during the study period,” or “across several centers.” Replace them with dates, setting details, and an explanation of what the period describes. Check that dates are consistent across the abstract, methods, tables, and flow of participants. If recruitment dates differ from outcome observation or record collection, do not collapse them into one range.
- Design label without an operational description. “Retrospective cohort” does not say how the cohort was assembled. Add the basic sequence of identification, treatment grouping, and outcome observation.
- Setting described too broadly. “Hospitals in the region” may hide major differences in care level or referral population. Give relevant facility and geographic details.
- Calendar dates reported as a vague span. State whether dates refer to cohort entry, recruitment, observation, or data collection, and use clear boundaries.
- Important era changes ignored. If care guidance or treatment availability changed during the period, identify when and why it could matter.
- Overclaiming from context. A multisite or population-wide setting does not automatically make a sample representative, and a prospective design does not automatically eliminate confounding.
A limitation remains even with excellent reporting: readers may still lack information about unrecorded changes, site-specific implementation, or settings not represented in the study. Transparent context makes these limits easier to assess; it cannot make the observed study cover every population or era.
Worked Example: Revising a Vague Multicenter Description
A hypothetical draft says: “We performed an observational study in several clinics during 2020–2022.” The investigators clarify that it was a prospective cohort at six outpatient clinics in two neighboring regions. Adults entered between January 15, 2020, and June 30, 2022. A temporary clinic-service disruption occurred from March through May 2020, and clinics resumed usual appointment capacity by June.
A stronger description is: “We conducted a prospective cohort study at six outpatient clinics in two neighboring regions. Adult cohort entry took place from January 15, 2020, through June 30, 2022. Clinic services were temporarily disrupted from March through May 2020 and had returned to usual appointment capacity by June 2020.”
This revision identifies the design, care setting, geographic scope, and exact entry dates. It also reports the disruption and its timing without asserting how it affected treatment selection or outcomes. If the disruption differed across clinics or affected enrollment, those additional facts belong in the description because they would help interpret the comparison.
Check Your Understanding
Answer each question, then compare your reasoning with the solutions.
- A paper says, “We conducted a retrospective study in hospitals from 2019 to 2021.” Name three details needed to make this description more informative.
- Why should a report distinguish cohort-entry dates from dates of outcome observation?
- In the two-era example, calculate the pooled outcome risk for each treatment and explain why the pooled comparison differs from the within-era comparisons.
- A multisite study spans a change in treatment guidance. What calendar-time information should the report include, and what should it avoid claiming without supporting analysis?
- Does naming a study “prospective cohort” establish that its treatment comparison is unbiased? Explain briefly.
Solutions. (1) State how the retrospective study was organized—for example, whether it assembled a cohort or sampled on outcome status—describe the type and location of hospitals, and give exact dates with clarification of what the dates refer to. Also report relevant changes in care during the period. (2) Entry dates define when participants joined the comparison; outcome-observation dates indicate when outcomes were assessed. They can differ, and combining them obscures the study’s calendar frame. (3) Treatment A has \(120/1{,}000=12\%\); Treatment B has \(180/1{,}000=18\%\). Each era has equal risks across treatments, but the groups have different proportions in the earlier and later eras, where risks differ. (4) Report when the guidance changed and whether its implementation differed across sites or time if known. Do not claim it caused a treatment difference without analysis supporting that conclusion. (5) No. The label describes how the study was organized; it does not guarantee comparable treatment groups, valid measurement, or absence of confounding.