# Was the headline outcome chosen before the results were known?

Draft • Research checked 4 October 2026 • AI-assisted DomDNA editorial content. No independent clinical review or publication is claimed. General education, not individual medical advice.

A study can measure sleep, energy, mood, task speed, activity and several blood markers. That breadth may be useful. It also gives a selective summary many opportunities to find something that looks impressive.

Before deciding what a positive result means, ask which question the study was designed to answer first. The most memorable number in a press release is not necessarily the primary outcome in the original plan.

## Many questions change the interpretation

[FDA's final guidance on multiple endpoints](https://www.fda.gov/regulatory-information/search-fda-guidance-documents/multiple-endpoints-clinical-trials) explains why multiple statistical comparisons need attention. More opportunities to obtain an apparently positive finding can make a simple significance label misleading unless the analysis deals appropriately with multiplicity.

This is not a demand that every study measure only one thing. Secondary outcomes can explain an intervention's effects, harms or practical consequences. Exploratory analyses can generate useful hypotheses. Their role needs to remain visible.

The distinction is between a planned confirmatory claim and a promising observation found while looking around. Both can be reported, but they do not carry the same evidential weight.

## A fictional app evaluation

Imagine a hypothetical trial whose primary question is whether a scheduling app improves attendance at appointments over three months. Researchers also collect five weekly experience ratings and several app-use measures.

The attendance comparison is inconclusive. One experience rating improves in the final week, and that result becomes the headline: “App improves daily confidence.” The rating may be worth investigating, but it has not replaced the original research question simply because it produced a more appealing result.

Now imagine a different report that states the attendance finding clearly, describes the confidence result as exploratory and explains how the extra analyses were handled. The underlying measurements could be identical, yet the reader receives a more accurate picture.

Nothing in this example is a real app result. No probability is supplied for how often one result would appear by chance because that would depend on the tests, their relationships and the analysis plan.

## Plans are useful because they predate the temptation

The [SPIRIT 2025 statement](https://www.bmj.com/content/389/bmj-2024-081477) promotes transparent description of planned methods and accessible related documents. A protocol and statistical analysis plan give you something to compare with the finished report. They are documentation tools, not automatic guarantees of good design.

A primary outcome should be specific enough to identify the measurement and time point. “Wellbeing improved” is less informative than naming the actual instrument, comparison and follow-up period.

Plans can legitimately change. Recruitment difficulties, new information or measurement problems may require an amendment. What matters is whether the change is dated, explained and interpreted honestly, especially when knowledge of results could have influenced it.

## Selective reporting is a documented concern

A [2004 study by Chan and colleagues](https://pubmed.ncbi.nlm.nih.gov/15161896/) compared trial protocols with publications and found discrepancies in outcome reporting. It provides empirical support for examining the original plan. It does not establish the current frequency of the problem or prove misconduct in a trial you are reading today.

You need not reconstruct the entire dataset to notice a mismatch. A paper that originally focused on one outcome but promotes another should explain that change. If the material you can access does not explain it, record the limitation rather than inventing a reason.

Avoid treating every secondary finding as worthless. A coherent secondary pattern, appropriate analysis and subsequent replication can be informative. The point is to resist promoting an unexpected result to certainty without the supporting work.

## What you can do with this

When a headline interests you, find the named primary outcome in the methods or registry. Write it in one sentence before saving the headline finding. Then mark the headline as primary, secondary or exploratory, if the report makes that distinction available.

For the imaginary app, your note might read: “Attendance was the planned main question; the confidence result came from an additional rating.” That small sentence prevents the research question from disappearing during retelling.

If you cannot access the protocol, say so. “Primary status not verified” is more useful than silently assuming that the promoted result was planned. It also helps you judge later evidence: a follow-up study designed specifically around that question would answer something the original exploratory finding did not.

For an educational starting point, the [DomDNA lifestyle quiz](https://domdna.com/quiz) records lifestyle answers, not DNA analysis or a clinical diagnosis.