Primary Outcome

The primary outcome is the main endpoint that a study is designed and statistically powered to evaluate, specified in the study protocol before data collection begins. It serves as the basis for determining whether the study has achieved its objective and is the outcome around which the sample size calculation, statistical analysis plan, and regulatory conclusions are structured.

The primary outcome anchors the entire study design and dictates the study's interpretation. Selecting the right primary outcome ensures that the study addresses the most clinically or scientifically relevant question. Because the sample size is calculated based on the expected effect on the primary outcome, it is usually the only outcome for which the study is deliberately powered, and even that power holds only if the assumptions behind the calculation are right. Changing the primary outcome after data collection, or reporting a different primary outcome from the one registered without saying so, is known as outcome switching; it is a form of outcome reporting bias that can produce misleading results. Trial registries and ethics applications ask for the primary outcome in advance, SPIRIT 2025 asks protocols to define it, and CONSORT 2025 asks trial reports to state the prespecified outcomes and any changes made to them after the trial began; together these make selective reporting easier to detect and improve transparency.

During protocol development, researchers select a primary outcome that is directly relevant to the research question, clinically meaningful, reliably measurable, and sensitive enough to detect the expected treatment effect. The primary outcome is then used as the basis for the power analysis, which determines the minimum sample size needed. In most studies, there is a single primary outcome. When a trial has more than one primary outcome, the success rule decides whether a multiplicity adjustment is needed. If the treatment can be declared effective when either outcome succeeds, alpha must be split or otherwise controlled to maintain the overall Type I error rate; if both must succeed, as with true co primary outcomes, no such adjustment is needed, although power falls because both tests have to succeed.

The primary outcome must be defined with precision, including the measurement instrument, the time point at which it will be assessed, and the method of analysis. For example, specifying "depression severity" is insufficient; a properly defined primary outcome would read "change from baseline in Hamilton Depression Rating Scale total score at 12 weeks." This level of detail prevents ambiguity and post hoc manipulation.

In regulatory contexts, such as pivotal phase III trials for drug approval, the primary outcome determines whether a treatment meets the threshold for regulatory approval. The U.S. Food and Drug Administration and the European Medicines Agency base their efficacy evaluations primarily on the prespecified primary endpoint. Other endpoints are usually classified as secondary or exploratory outcomes, which provide supplementary information but are not the basis for the study's primary conclusions.

The distinction between primary and secondary outcomes also affects how multiple testing is handled. A single primary outcome is tested at the full significance level (typically a two sided alpha of 0.05). Secondary outcomes may be tested without formal multiplicity adjustment (and interpreted as exploratory) or subjected to hierarchical testing procedures in which testing starts only after the primary outcome is significant and each subsequent outcome is tested only if the preceding one is significant.

Every confirmatory study should have a clearly defined primary outcome specified in the protocol. This includes randomized controlled trials, pragmatic trials, and adequately powered observational studies. In pilot or feasibility studies, the primary outcome is often a process measure (such as recruitment rate or retention) rather than a clinical endpoint. For systematic reviews, defining the primary outcome in the protocol ensures consistency in data extraction and synthesis. Always prespecify the primary outcome before data collection; selecting it afterward undermines the study's credibility.

A serious mistake is switching the primary outcome after reviewing the data, selecting whichever outcome produced the most favorable result. This practice, detectable by comparing publications against trial registry entries, is a well documented form of reporting bias. In the COMPare project, which checked every trial published over six weeks in five major medical journals against its protocol or registry entry, only 9 of the 67 trials assessed reported their prespecified outcomes without discrepancies. Another common error is selecting a primary outcome that is not sensitive to the intervention, leading to a null result despite genuine clinical effects on other outcomes. Researchers also sometimes specify several primary outcomes, any one of which would count as success, without adjusting for multiplicity, inflating the risk of false positive findings. Finally, choosing a surrogate marker as the primary outcome when patient centered outcomes would be more appropriate can result in regulatory approval for treatments that do not improve outcomes that matter to patients.

A phase III multicenter trial of a new biologic therapy for rheumatoid arthritis specifies the primary outcome as the proportion of patients achieving ACR50 response (50% improvement in American College of Rheumatology criteria) at 24 weeks. The sample size of 420 patients (210 per arm) is calculated to detect a 15 percentage point difference between groups (for example, 25% versus 40% responding) with 90% power at a two sided alpha of 0.05; about 200 per arm gives that power, and the rest allows for dropout. Secondary outcomes include ACR20 and ACR70 response rates, change in Disease Activity Score 28 (DAS28), and radiographic progression at 52 weeks. The regulatory submission hinges on whether the trial demonstrates a statistically significant difference in ACR50 at 24 weeks.