Heterogeneity in a Nursing Review: Work from Study Differences to a Defensible Conclusion

Start with the claim you want the review to support

Suppose a nursing assignment asks whether a structured discharge teach-back session, compared with usual discharge education, reduces all-cause readmission within 30 days among adults leaving a general medical ward. That sentence defines the population, the intervention comparison, the event and the time window. Keep it beside the review while reading. “Discharge education helps” is much broader and could conceal a change of intervention or outcome.

The four study cards below are invented teaching examples. They are not clinical evidence or a recommendation to change discharge practice. Their purpose is to show how apparently similar studies can answer different questions, and how a student can document that distinction before borrowing a pooled estimate for a rationale.

Begin with the review’s eligibility criteria and the studies included in the specific analysis, not only its title. A review may cover several education strategies but report separate analyses for each. The article you cite should make clear which comparison and outcome your statement comes from. If the necessary detail is absent, record the question you need to resolve from the review or primary report.

Put the study differences where you can see them

In this example, Study A and Study B both compare one teach-back session with usual discharge education and report 30-day all-cause readmission. Their event counts allow a simple illustration of a risk ratio. Study C adds weekly calls and measures readmission at 90 days. Study D reports recall of discharge instructions rather than readmission. All four concern discharge education, but that shared topic does not make their results interchangeable.

The event counts are complete for this teaching calculation. In a real report, verify the denominator, losses to follow-up, analysis population and whether the unit of allocation requires an adjusted analysis. Do not silently convert an adjusted effect into an unadjusted one or treat repeated events as the number of people with at least one event.

The matrix should describe differences concretely. “Different interventions” is too vague to explain a decision. “Study C adds four weekly telephone contacts, while A and B provide one predischarge session” tells the reader what changed and why the result cannot be attributed to the session alone.

Separate comparability from the size of the result

For Study A, the two fictional risks are 20/200 = 10% and 30/200 = 15%. The risk ratio is 0.10/0.15, approximately 0.67. For Study B, the risks are 15% and 20%, giving a risk ratio of 0.75. These are point estimates only. The example does not supply confidence intervals or a complete risk-of-bias assessment, so it cannot establish a precise intervention effect or the certainty of evidence.

Study C also has a risk ratio of approximately 0.67: 20/150 divided by 30/150. That numerical match with Study A does not repair the different follow-up period or remove the extra telephone intervention. Conversely, A and B need not have identical point estimates to be relevant to the same question. A sensible comparison begins with what was studied, then examines what the estimates and their uncertainty show.

The practical decision in this teaching exercise is to consider A and B for the 30-day, single-session comparison, while keeping C in a separate intervention/time-window group and D in the recall-outcome group. This grouping is an application of the stated question, not a general rule that every difference requires exclusion from an entire review. A broader protocol could legitimately ask additional questions.

Read the heterogeneity statistics in context

Cochrane distinguishes differences in participants, interventions and outcomes from differences in methods, and from variation in estimated effects beyond sampling variation. I-squared describes the proportion of observed variation attributed to heterogeneity under its calculation; it is not the percentage of studies that are unreliable. Its interpretation and uncertainty depend on the analysis. With few studies, a low estimate cannot establish that the underlying effects are identical.

For the teaching matrix, start by checking which rows entered the forest plot. If the plot combines A, B and C under a label such as “education and readmission,” return to the comparison definition. A reassuring I-squared value would not make C a direct answer to the narrower question. If A and B alone are analysed, inspect their confidence intervals and the review’s methods before deciding what their average represents.

Check the direction of the scale as well. For an undesirable event such as readmission, a risk ratio below one corresponds to fewer events in the intervention group. That reading does not transfer unchanged to an outcome such as correct recall, where more events may be desirable. A forest plot label and an outcome definition belong together; do not infer benefit from the position of a square without reading them.

• Cochrane Handbook, chapter 10: Meta-analysis and heterogeneity

Check what the review did about the differences

A random-effects model can incorporate between-study variation; it does not turn different clinical questions into one question or identify the cause of disagreement. Cochrane advises investigating heterogeneity and considering the interpretation of the average. Decisions about subgroup and sensitivity analyses should be assessed alongside the protocol, study characteristics and available data, rather than selected only because they make a result more favourable.

In the example, separating C is justified by the intervention and follow-up definitions already written in the question. It is not justified by preferring C’s numerical result or by wanting a smaller heterogeneity statistic. If the review added that separation only after seeing the results, report the analysis as exploratory where appropriate. Do not describe a post hoc explanation as a demonstrated mechanism.

Imagine that A and B later prove to differ in whether usual education already included a structured comprehension check. That would be a reason to examine the comparator descriptions and consider the interpretation of the contrast. It would not be enough to announce that comparator intensity “caused the heterogeneity.” With only a small number of studies, many characteristics may differ together. Keep a plausible explanation separate from an established explanation.

When the review presents a sensitivity analysis, write down what changed and why. Excluding a study because its design violates eligibility is different from removing it because its result is inconvenient. Compare the relevant conclusions across the stated analyses, including uncertainty, rather than choosing whichever single estimate best fits the assignment’s intended argument.

Write the conclusion with its comparison intact

A useful evidence summary preserves the path from the question to the claim. It names the actual comparison and outcome, identifies the important differences, and explains what those differences mean for use of the result. Cochrane’s interpretation guidance also separates effect estimates from the certainty and applicability of the evidence. A precise-looking average cannot replace those judgments.

For this exercise, a defensible note would read: “Studies A and B address a single discharge teach-back session versus usual education for 30-day all-cause readmission. Their fictional risk-ratio point estimates are approximately 0.67 and 0.75. Study C evaluates an additional telephone component and a 90-day outcome, while Study D measures recall. I would not quote a pooled result combining these four records as an estimate for the narrower 30-day question. Confidence intervals, study validity and the review’s synthesis methods still need appraisal before drawing an effect conclusion.”

The note does not say the intervention is effective, ineffective or recommended. The teaching data do not support that decision. It also does not reject every study merely because the studies differ. It explains which result belongs to which question and what information is missing. That is more useful than appending “high heterogeneity” to an otherwise unchanged recommendation.

• Cochrane Handbook, chapter 15: Interpreting results and drawing conclusions

Keep a short appraisal record beside the citation

The final record should make the interpretation easy to check later. Save the review citation, analysis name or figure number, outcome, follow-up, included study IDs and the exact sentence the source is being used to support. If a study report or appendix resolved an ambiguity, record that location rather than leaving the reasoning in an untraceable note.

Use the blank fields below for a real assignment. The filled entries refer only to the fictional exercise. When a field is unknown, state what is unknown and the next source to inspect. Do not populate an empty confidence-interval or heterogeneity field from a different outcome merely because it appears in the same review.

Before submission, read the rationale beside this record. If the rationale names a single session but the record describes a multicomponent programme, revise the claim or choose a better-matched analysis. If the assignment concerns an individual case, additional professional and course-specific judgments remain necessary; this worksheet evaluates how a source supports a statement, rather than selecting care for a patient.