Secondary Outcome

A secondary outcome is a prespecified endpoint that provides supplementary information about the effects of an intervention beyond what the primary outcome captures. Secondary outcomes are defined in the study protocol but are not the primary basis for determining the study's conclusions or statistical power. They may address different aspects of the treatment's impact, such as safety, quality of life, functional status, or biological mechanisms.

While the primary outcome determines whether the study met its main objective, secondary outcomes provide a more comprehensive picture of the intervention's effects. They can reveal benefits or harms not captured by the primary endpoint, inform secondary research questions, and generate hypotheses for future studies. However, because studies are not typically powered for secondary outcomes, their results should be interpreted with caution and presented as supplementary rather than definitive evidence. Transparent prespecification of secondary outcomes in the protocol discourages post hoc outcome selection and makes it detectable.

Secondary outcomes are selected during protocol development and listed in the trial registration and statistical analysis plan. They complement the primary outcome by measuring different dimensions of the intervention's impact. A trial of an antihypertensive drug might have systolic blood pressure reduction as the primary outcome, with secondary outcomes including cardiovascular events, renal function, quality of life, and adverse events. The study's sample size is calculated based on the primary outcome, meaning secondary outcomes may be underpowered.

The analysis of secondary outcomes raises multiplicity concerns. Testing many secondary outcomes increases the risk of finding spurious significant results. Hierarchical (fixed sequence) testing procedures, where secondary outcomes are tested in a prespecified order, starting only if the primary outcome is significant and continuing only while each preceding outcome is significant, maintain the overall Type I error rate. Alternatively, secondary outcomes can be presented as descriptive or exploratory, with confidence intervals reported but formal hypothesis testing deemphasized.

Include secondary outcomes in any study protocol to provide a comprehensive assessment of the intervention's effects. Prespecify them before data collection and describe them in the protocol and trial registration. Interpret secondary outcome results in the context of the primary outcome findings and with appropriate acknowledgment of multiplicity and potential underpowering. Do not elevate a positive secondary outcome to the status of a primary finding if the primary outcome was null, as this risks misleading interpretation.

A common error is switching the focus from a null primary outcome to a significant secondary outcome, presenting the secondary finding as the study's main result. This practice, a form of spin, distorts the evidence and misleads readers; when the secondary outcome is relabeled as the primary, it becomes outcome switching. Another mistake is not prespecifying secondary outcomes, allowing post hoc selection of whichever outcomes showed favorable results. Failing to acknowledge that the study was not powered for secondary outcomes leads readers to overinterpret significant findings and to read null results as evidence of no effect, when they may reflect low power. Finally, testing many secondary outcomes without any multiplicity adjustment inflates the false positive rate.

A phase III trial of a new monoclonal antibody for asthma specifies the primary outcome as the annualized rate of severe exacerbations over 48 weeks. Secondary outcomes include forced expiratory volume in one second (FEV1), Asthma Control Questionnaire (ACQ) scores, rescue medication use, and patient rated quality of life. The primary outcome shows a 35% reduction in exacerbations (p < 0.001). A prespecified hierarchical testing procedure is applied to secondary outcomes in order: FEV1 improvement is significant (p = 0.002), ACQ improvement is significant (p = 0.008), but the reduction in rescue medication use does not reach significance (p = 0.07), halting the hierarchy. Quality of life data are reported descriptively.