How to Read a Clinical Trial Results Table

Read the outcome, units, comparator, time point, and uncertainty before looking at the P value. A clinical trial results table describes a particular analysis. It does not automatically tell you how much an individual will benefit or whether a treatment is suitable for them.

This walkthrough uses the original 2021 SURPASS-2 trial to show how to read clinical trial results. It is a statistics example, not prescribing advice. Start with the evidence center for study-quality context and the trial-phase guide for what a phase label means.

First identify the question behind the table

SURPASS-2 randomized 1,879 adults with type 2 diabetes inadequately controlled with metformin. Mean starting HbA1c was 8.28%. It was an open-label, 40-week phase 3 comparison of three tirzepatide groups with a semaglutide group. Open-label means the treatment identity was not hidden in the same way as a fully blinded comparison. Its primary outcome was change in HbA1c, a blood test reflecting longer-term glucose levels.¹

The semaglutide comparator was a specific studied exposure, not every marketed semaglutide formulation. This trial therefore cannot answer every obesity-treatment comparison. Read the semaglutide versus tirzepatide evidence comparison for the clinical context.

An annotated results table

The entries below summarize the original paper's primary-outcome estimates. They identify the treatment groups by their studied maintenance exposure solely to make the source identifiable; they are not dosing instructions. HbA1c changes and between-group differences are percentage points, not relative percentage reductions.¹

What the columns tell you

Within-group change answers how much the measured outcome changed in that group. Between-group difference answers how the groups differed in the reported analysis. The latter is the comparison needed to assess an advantage over the alternative.

Notice that the displayed group means do not always subtract exactly to the reported treatment difference. Papers can use modeled estimates and rounding; report the published estimate rather than silently replacing it with a subtraction from rounded values. The table's footnotes and statistical methods explain the analysis.

Direction matters. A negative difference here means a greater HbA1c reduction with tirzepatide relative to the semaglutide comparator. For a different outcome or reversed group order, a negative number can have a different meaning.

Read uncertainty without turning it into a guarantee

A confidence interval communicates how precisely the analysis estimates the effect, under its assumptions. For a difference, zero represents no difference. For a ratio, the corresponding no-effect value is usually one. Do not use the wrong threshold.^2,3

The first row's interval excludes zero, but the possible difference is still relatively modest. Statistical evidence of a difference is not the same as a large or important benefit. Clinical importance requires the outcome, magnitude, harms, alternatives, and patient's circumstances.

A 95% confidence interval is not a promise that an individual response falls within it, and it is not a simple 95% probability statement about the true effect after observing the data. Cochrane's interpretation chapter explains the repeated-sampling meaning and its limits.²

Find the denominator and baseline

The randomized total is not necessarily the number included in every outcome analysis. Some participants may discontinue treatment, miss a measurement, or enter a separate safety analysis. Read participant flow and missing-data methods rather than assuming every table uses the same denominator.

Baseline tables show starting characteristics; outcome tables describe later changes. A difference between final values can be misleading when starting levels differ or when the analysis adjusts for baseline. For adverse-event percentages, record the actual number exposed and events, the observation period, and whether people or repeated events are being counted.

Check the analysis the headline leaves out

1. Was this a prespecified primary outcome or an exploratory subgroup?
2. Were the groups compared directly, or are numbers taken from separate trials?
3. Which analysis population and assumptions about missing measurements were used?
4. Does a small P value accompany a meaningful effect and a sufficiently precise interval?
5. What adverse events and withdrawals occurred, and what remains unknown beyond follow-up?

Read the full methods, table footnotes, and limitations. A secondary measurement should not be used as a substitute for an unsuccessful primary outcome. Use the ClinicalTrials.gov walkthrough to compare the published report with its registered plan.

For an annotated example of how a summary can differ from the full evidence, read why a study abstract can be misleading.