How Do You Handle Evidence When an Intervention Helps One Outcome but Harms Another?

What if the intervention works for one outcome but makes another worse?

An intervention reduces the primary outcome exactly as intended. At the same time, another important outcome deteriorates.

Does the intervention work?

That question may no longer have a meaningful yes-or-no answer. An intervention can have several causal effects at once, and those effects do not have to point in the same direction. Improving one outcome does not erase deterioration in another, just as an adverse effect does not automatically negate every benefit.

The synthesis should therefore preserve the competing outcomes and help readers understand the trade-off rather than compressing a multidimensional result into a single declaration of success or failure.

Preserve each important outcome before interpreting the trade-off

Opposing outcomes are not necessarily contradictory findings

An intervention can cause several effects simultaneously

Researchers sometimes speak about “the effect” of an intervention as though it were a single property. In practice, an intervention can affect numerous outcomes through different pathways.

A treatment might reduce pain while increasing fatigue. A teaching intervention might improve examination performance while increasing student workload. A workplace policy might improve productivity while worsening employee satisfaction. None of these combinations is logically inconsistent.

The intervention has different effects on different outcomes.

This distinction matters. If one outcome improves and another worsens, the solution is not necessarily to investigate which study is “right.” Both findings may be correct.

The primary outcome should not automatically dominate the interpretation

Trials commonly designate a primary outcome for sound methodological reasons, including sample-size planning and control of multiplicity. But “primary” does not mean “the only outcome that matters to a decision.”

Suppose an intervention substantially improves the designated primary outcome while causing a serious adverse effect. Reporting only that the trial met its primary endpoint would provide an incomplete picture of its consequences.

Evidence synthesis should therefore identify the outcomes that are critical to the decision, including important benefits and harms, rather than allowing the hierarchy used for statistical testing within individual studies to become an automatic hierarchy of human importance.

Do not cancel different outcomes mathematically without a defensible common scale

A common temptation is to treat a benefit and harm as though they can simply be subtracted.

Suppose an intervention produces a five-point improvement in one scale and a three-point deterioration in another. The “net effect” is not two points. The scales may measure entirely different constructs, and a point on one scale may have no meaningful equivalence to a point on the other.

Even when both outcomes are binary, subtracting event counts can embed hidden assumptions about their relative importance. Preventing one mild event is not necessarily equivalent to causing one severe irreversible event.

Compare absolute effects when they are meaningful

Relative effects can be useful, but absolute effects often make competing outcomes easier to understand.

Imagine that an intervention reduces the risk of outcome A from 20% to 15% but increases outcome B from 1% to 2%. Another intervention might reduce A from 4% to 3% while producing the same relative increase in B. The relative effects could look similar even though the practical trade-offs differ considerably because baseline risks differ.

Presenting expected numbers of events per 100, 1,000, or another suitable population size can help readers compare consequences, provided the baseline risks used are appropriate to the target population.

Magnitude alone does not determine importance

A frequent mild inconvenience and a rare catastrophic harm cannot be compared solely according to event frequency. Several dimensions may matter:

• how large the beneficial and harmful effects are;
• how severe the outcomes are;
• whether effects are temporary or persistent;
• whether harms are reversible;
• when benefits and harms occur;
• how certain the evidence is;
• how people value the outcomes.

These dimensions should inform interpretation without being silently converted into an unsupported ranking.

The certainty of evidence may differ by outcome

You may have high-certainty evidence that an intervention improves one outcome and much weaker evidence suggesting harm on another.

The reverse can also occur.

Do not transfer certainty from the primary outcome to every other outcome. Each important outcome has its own evidence base, precision, risk of bias, consistency, directness, and other considerations affecting confidence in the estimate.

This is particularly important when benefits and harms come from different or uneven evidence bases.

Timing can change the trade-off

A benefit may occur immediately while a harm appears only after prolonged exposure. Alternatively, an intervention may cause short-term discomfort in exchange for a durable benefit.

A single cross-sectional summary of both outcomes can therefore misrepresent the decision.

When the relevant consequences occur at different stages, preserve the timing of the outcome effects. Readers may need to know not merely what is gained and lost, but when each consequence occurs and how long it persists.

Average outcomes can conceal who experiences the benefit and who experiences the harm

Suppose an intervention produces a modest average benefit and a modest average harm. Those averages do not tell you whether everyone experiences a little of both.

Perhaps one group benefits substantially with little harm while another receives little benefit and experiences most of the adverse effect. If credible effect modification exists, the overall trade-off may differ across populations.

That possibility connects directly to the problem of average effects hiding important subgroup differences.

Values and preferences become unavoidable when outcomes compete

Evidence can estimate how often outcomes occur and how interventions change them. Evidence alone cannot always determine how much one outcome should be valued relative to another.

Two people presented with exactly the same effect estimates may make different choices because they attach different importance to the outcomes.

One person may accept a substantial burden for a modest improvement in an outcome they consider crucial. Another may decide that the same improvement is not worth the burden. Neither preference changes the effect estimate.

The role of evidence synthesis is therefore to make the consequences and uncertainty clear enough that such judgments are informed rather than hidden inside the reviewer's conclusion.

How should you interpret an intervention with a clear benefit and a clear disadvantage?

The conclusion is not indecisive. It is more informative than declaring that the intervention simply “works,” because it tells readers what they gain and what accompanies that gain.

How competing outcomes get flattened into misleading conclusions

Make the trade-off explicit instead of deciding it invisibly

Begin by identifying all outcomes that are critical to the decision. Estimate each separately using the most appropriate evidence, then place those estimates alongside one another.

A useful conclusion might therefore state that an intervention probably improves outcome A but also increases outcome B, with uncertainty remaining about outcome C. That sentence preserves more decision-relevant information than “the intervention is effective.”

When the evidence supports a genuine trade-off, ambiguity in the final decision is not necessarily a failure of synthesis. Sometimes the evidence has done its job precisely when it reveals that the choice depends on what consequences matter most.

When important outcomes point in opposite directions, check:

Common questions when an intervention has competing effects

Opposing outcomes should remain visible in the conclusion

How to Cite This Guide

This guide is intended to be read, shared, and used in research, teaching, and academic work. If you draw on its ideas, explanations, or other content, please acknowledge the source by citing the guide. Doing so gives appropriate credit and helps your readers locate the original resource.