What STROBE Does—and Does Not Do
A comparative observational report needs to do more than present two outcome estimates. Readers need enough information to understand what was compared, how participants entered the study, which variables were measured, how the analysis was conducted, and what the results can reasonably support. STROBE provides a structured way to check whether those parts of a report are described.
STROBE stands for Strengthening the Reporting of Observational Studies in Epidemiology. Its checklist covers cohort, case-control and cross-sectional studies. It organizes reporting expectations into title and abstract, introduction, methods, results, discussion and other information. For a comparative study, the checklist helps a reviewer ask whether the report makes the comparison understandable and sufficiently transparent.
The distinction matters. A report may satisfy many checklist items while its study remains vulnerable to confounding, selection bias, measurement error or imprecision. Conversely, an item may not apply to a particular design; the report should make that clear rather than supplying irrelevant material. The goal is transparent reporting, not a high checklist score.
STROBE has 22 numbered items, some with subitems specific to study design. An author or reviewer should first identify the design and then consider the applicable items. The checklist is a prompt for reporting, not a substitute for judgment about what readers need in the specific setting.
A Practical Map of the Checklist
For a comparative report, it is useful to read the checklist as a sequence of questions. The methods should explain how the study was set up and analyzed; the results should show who contributed data and what was observed; the discussion should interpret the findings while acknowledging the study’s limits. The table summarizes this structure without treating each item as a standalone box to tick.
| Report section | STROBE asks the report to make clear | Study-specific elaboration may include |
|---|---|---|
| Title, abstract and introduction | That the study is observational; the scientific context; and the objectives | Which comparison and question make the study relevant |
| Methods | Design, setting, participants, variables, data sources, bias considerations, study size and statistical methods | Operational definitions, analytic decisions and explanations needed to reproduce or interpret this particular analysis |
| Results | Participant numbers, descriptive information, outcome data, main results and additional analyses | Counts, denominators, estimates and precision needed to understand the reported comparison |
| Discussion and other information | Key results, limitations, interpretation, generalisability and funding | How the study’s specific data, assumptions and setting qualify its conclusions |
The checklist is broad by design. For example, it asks authors to describe participants, but the exact flow diagram or table that best does so depends on the study. It asks for variables and measurement methods, but the detail required depends on whether a variable is a routine record, a survey response or a derived measure. A useful report supplies enough context to interpret its choices, rather than repeating generic statements.
Some checklist items are often overlooked because they do not look like statistical results. The report should describe relevant sources of bias, explain how study size was determined, and explain how quantitative variables were handled. The results should distinguish descriptive data from outcome data and report estimates with appropriate precision. The discussion should connect limitations to interpretation rather than merely list them.
Worked Example: Mapping a Hypothetical Report to STROBE
Suppose a hypothetical hospital cohort compares 30-day readmission among adults who received a new discharge-support service with readmission among adults who did not. A reviewer is asked whether the report is sufficiently transparent. The reviewer should map the report to the relevant checklist questions, not infer quality from whether it contains a heading called “STROBE.”
Title and introduction. Check whether the title identifies the work as observational and whether the introduction explains the question and objective. If the report only says “an evaluation of discharge support,” the reader may not know that it is a comparative observational cohort.
Methods. Check that the report names the design, setting and calendar period; explains who could enter each group; defines the service and readmission outcome; describes data sources and measurement; and explains the statistical methods. For this hypothetical study, a sentence that says “we compared the groups” would not explain how the comparison was calculated or what variables entered an adjusted analysis.
Results. Check that the report gives the number of people assessed, included and analyzed in each group, describes relevant participant characteristics, and reports the readmission outcome with its group-specific denominator. If it reports only “readmission was lower,” a reader cannot verify the scale of the comparison.
Discussion and other information. Check for a discussion of the main findings, study-specific limitations, cautious interpretation, relevance beyond the hospital, and funding. A general statement that “more research is needed” does not explain which limitations qualify this particular comparison.
This mapping identifies what the reviewer needs to look for. It does not establish whether the service caused a difference, because reporting completeness alone cannot establish that the groups were exchangeable or that the analysis removed bias.
From Checklist Item to Useful Reporting
A checklist item is a minimum prompt, not a complete sentence to paste into a paper. To apply it well, trace the comparison through the report: identify what was compared, determine which people and measurements contribute to it, then check whether the analysis and results are described in enough detail to understand the reported estimate. The exact level of detail should reflect the study’s data and decisions.
Locate the groups, outcome and study design as stated in the report. If a reader cannot tell what the groups represent, later statistical details will not repair the ambiguity.
Check the participant account and descriptive information. Counts and denominators should make the analyzed sample understandable, including relevant exclusions or unavailable data.
Check that the analysis described in the methods matches the estimates presented. The report should identify the variables and procedures needed to interpret those estimates, with details appropriate to its design.
Check whether limitations, interpretation and generalisability are discussed in relation to the actual study. A conclusion should not make a stronger claim than the design and analysis support.
This sequence also gives a short justification for why the checklist is more than a list of headings. A comparative estimate is interpretable only if its groups, outcome and underlying analyzed data are identifiable. If any of these links is missing, readers cannot reliably connect the reported result to the study described. Reporting each link makes the result assessable; it does not, by itself, make the result unbiased.
Worked Example: Reporting Counts and a Comparative Outcome
Consider a second hypothetical report of the same general kind. It states that 420 service recipients and 460 nonrecipients were eligible. It then reports that 20 recipients and 60 nonrecipients were excluded before analysis, leaving 400 people in each group. At 30 days, 32 recipients and 48 nonrecipients had been readmitted. Check the arithmetic and identify what a clear results paragraph could report.
Participant counts. The analyzed counts follow from the reported exclusions:
The flow is internally consistent. A report should also explain the relevant reasons for exclusion; arithmetic consistency alone does not tell readers whether the exclusions could affect the comparison.
Group-specific observed proportions. For recipients, the observed proportion readmitted is \(32/400=0.080\), or 8.0%. For nonrecipients, it is \(48/400=0.120\), or 12.0%. The observed risk difference, defined here as recipient proportion minus nonrecipient proportion, is:
Thus, the observed 30-day readmission proportion was 4.0 percentage points lower among recipients in this hypothetical sample. This is a descriptive comparison of the reported groups. Without further design and analysis assumptions, it is not a causal effect of the service.
A transparent results statement would give both event counts and denominators, the two proportions and the difference, with any interval estimate and its method if reported. For example: “Readmission occurred in 32 of 400 recipients (8.0%) and 48 of 400 nonrecipients (12.0%); the unadjusted difference in observed proportions was −4.0 percentage points.” This wording makes the arithmetic checkable and does not turn an observed association into a causal claim.
Study-Specific Detail, Diagnostics and Failure Modes
STROBE leaves room for judgment because the same reporting prompt may require different elaboration in different studies. A variable created by linking records, for instance, may need an explanation of how those records were measured and combined. A complex analysis may need enough detail for readers to understand its inputs and decisions. The relevant question is not “How many words did the authors spend?” but “Can a reader understand what was done and what the reported comparison represents?”
Several warning signs suggest that a report needs closer review:
- Unclear groups or design. The report labels the analysis a comparison but does not explain the observational design or what group membership means.
- Untraceable participant counts. The reported number analyzed cannot be reconciled with the numbers eligible, excluded or otherwise unavailable.
- Results without denominators. A percentage or effect estimate is presented without enough group-specific counts or sample information to interpret it.
- Methods that do not match results. The report describes one analysis, but the presented estimate appears to come from another, with no explanation.
- Limitations without consequences. The discussion names a possible limitation but does not explain how it might qualify interpretation, or treats an uncertain direction of bias as known.
The checklist should also be used with appropriate alternatives and extensions. CONSORT is designed for randomized trials, whereas STROBE addresses observational designs; one should not substitute for the other merely because both contain reporting checklists. For studies based on routinely collected health data, a relevant extension such as RECORD may add reporting prompts. An extension supplements the applicable guideline; it does not make the underlying study design stronger.
The central limitation of checklist use is that a report can be complete yet wrong, or incomplete yet contain a useful analysis that is difficult to assess. STROBE supports transparent assessment; it does not certify measurement validity, eliminate confounding, guarantee reproducibility, or replace a careful substantive and statistical review. In particular, checking that an author described an adjustment procedure does not establish that the procedure was appropriate.
Worked Example: Auditing Missing-Data Reporting and a Limitation
In a third hypothetical report, a baseline covariate is unavailable for 36 of 400 recipients and 44 of 400 nonrecipients. The methods say only, “missing values were addressed,” and the discussion says, “missing data may be a limitation.” Use the reported counts to identify what readers still need.
Calculate the proportions with unavailable values. Among recipients, \(36/400=0.090\), or 9.0%. Among nonrecipients, \(44/400=0.110\), or 11.0%. These computations describe the reported data; they do not reveal whether the missingness could bias the comparison.
The methods statement is too vague to show how the analysis handled the unavailable values. A more useful report would identify the variable, state how missing values were handled in the analysis, and describe any relevant examination of their pattern. The results should make clear how many people or observations contributed to the reported analysis. The precise details depend on the method actually used; authors should not claim a method they did not use.
The limitation statement also needs study-specific elaboration. The report should explain why missingness matters to this comparison and what uncertainty it creates. The different observed percentages (9.0% and 11.0%) are a reason to describe the pattern, not proof that missingness caused bias in a particular direction. Without stronger information, the responsible conclusion is that the impact is uncertain.
Common Mistake: Treating STROBE as a Quality Score
Wrong reasoning: “The report addresses every STROBE item, so its comparison is valid and its conclusion is causal.” This confuses the quality of reporting with the quality of the design and analysis. STROBE asks whether relevant information is reported; it does not verify that the information is correct or that assumptions needed for a causal interpretation hold.
Correction: Use the checklist to find information and omissions, then assess the study’s methods and claims on their merits. A well-reported observational result can still be confounded, imprecise or limited to a narrow setting. A report should make those concerns assessable, and its interpretation should reflect them.
Check Your Understanding
Answer each question using the reporting principles in this tutorial.
- What is STROBE designed to improve, and what does a completed checklist not establish?
- A report gives 25 events among 250 people in one group and 30 among 300 in another. What additional information makes these outcome data interpretable, and what are the two observed proportions?
- Why is “missing values were addressed” usually insufficient as a methods description?
- A reviewer finds that a report describes every checklist item but gives no explanation for its exclusions. What should the reviewer conclude?
- Which is the more appropriate general reporting guideline for an observational cohort: STROBE or CONSORT? Briefly explain.
Solutions. (1) STROBE supports transparent reporting of observational studies; completion does not prove validity, absence of bias, or a causal effect. (2) The report should give group labels and denominators, as well as enough context to identify the outcome and analysis. The proportions are \(25/250=0.100\), or 10.0%, and \(30/300=0.100\), or 10.0%. (3) The phrase does not identify what was missing or how the analysis handled it, so readers cannot understand which observations contributed to the results. (4) The report has an important transparency gap: the reviewer should request or note the reasons for exclusion and assess whether the reported participant account is sufficient. The omission does not alone prove bias, but it limits assessment. (5) STROBE is the relevant general guideline for an observational cohort; CONSORT is designed for randomized trials.