01 · The Question
Compared with what?
Statements such as “the intervention improved performance,” “exposed participants had greater risk,” or “scores increased significantly” are incomplete until you know the comparison behind them. Improvement compared with baseline? Compared with no treatment? Compared with usual care? Greater risk than which exposure category?
The comparator is part of the question the study actually answers. Even labels such as “control group” can conceal important differences. A control condition might involve no intervention, placebo, usual care, attention from researchers, an alternative treatment, or another dose of the same intervention. Identifying what was actually examined therefore requires identifying what it was compared against.
03 · What You Need to Know
How to reconstruct the comparison the researchers actually made
Every comparative result has a reference point
A difference cannot exist in isolation. If a study reports that one group had lower anxiety, greater survival, higher achievement, or increased risk, there must be another condition, group, value, or time point against which that statement is defined.
Sometimes the comparator is obvious because the study has two clearly labeled groups. In other cases, it is embedded in a regression model, defined through a reference category, or created by comparing measurements within the same participants over time.
Ask a literal question: What two things are being contrasted in this estimate?
In a randomized trial, identify what each assigned group was supposed to receive
A two-arm randomized trial might compare a new intervention with placebo, usual care, an active treatment, a wait-list condition, or no intervention. These are not interchangeable comparisons.
Current CONSORT guidance emphasizes describing interventions in each group sufficiently for replication. When the control group receives usual care, that care should be described because usual care can vary substantially across settings.
| Comparator |
What the contrast may address |
| No intervention |
Intervention versus receiving no study intervention |
| Placebo or sham |
Intervention versus a condition designed to resemble aspects of treatment without its hypothesized active component |
| Usual care |
Intervention versus the care ordinarily provided in the study setting |
| Active comparator |
One intervention versus another active intervention |
| Wait-list |
Immediate intervention versus delayed access during the comparison period |
| Alternative dose or format |
Different versions or intensities of an intervention |
The appropriate interpretation follows from the actual contrast. Evidence that a treatment performs better than no treatment does not automatically establish that it performs better than an established alternative treatment.
“Usual care” is not a complete description
Usual care can differ between hospitals, regions, countries, clinicians, and historical periods. It may also contain substantial treatment rather than functioning as an absence of intervention.
If the comparison group received usual care, determine what that actually meant in the study. CONSORT 2025 specifically recommends describing usual care so readers can assess whether the comparator differs from usual care in their own setting.
This matters because an intervention's estimated effect is relative to its comparator. The same intervention could produce different relative effects when compared with minimal care versus an intensive existing program.
In observational studies, the comparator may be an exposure category
Suppose researchers investigate whether weekly working hours are associated with cardiovascular risk. They might compare people working 55 or more hours per week with those working 35 to 40 hours. Alternatively, they could model working hours continuously.
Those are different analytical questions. In the categorical version, one group becomes the reference category. In the continuous version, the model may estimate the change in outcome associated with a specified increase in working hours.
STROBE asks observational-study authors to clearly define outcomes, exposures, predictors, potential confounders, effect modifiers, and diagnostic criteria where applicable. It also asks authors to explain how quantitative variables were handled in analyses, including any groupings chosen.
The reference category can substantially affect how a result reads
Imagine three exposure groups: low, moderate, and high. If moderate exposure is the reference, results describe low and high exposure relative to moderate exposure. If low exposure becomes the reference, the numerical contrasts and verbal interpretation change accordingly.
The underlying data have not necessarily changed. The reference point has.
Whenever you see a risk ratio, odds ratio, hazard ratio, regression coefficient based on categorical predictors, or another relative estimate, identify the reference category before interpreting its direction or magnitude.
A comparison can occur within the same participants
Not every comparison involves separate groups. Researchers may measure the same participants before and after an intervention, under multiple experimental conditions, or at several time points.
A single-group pre-post study, for example, may compare participants' outcomes after an intervention with their own baseline measurements. That is a real comparison, but it is not equivalent to comparing changes against a concurrent control group.
Within-group change
Compares observations within the same group or participants across conditions or time.
Between-group difference
Compares outcomes or changes between distinct groups or assigned conditions.
This distinction becomes especially important when researchers say that an intervention “worked” because outcomes improved significantly from baseline. Without an appropriate counterfactual comparison, the observed change might also reflect secular trends, maturation, regression to the mean, concurrent events, measurement effects, or other processes.
Do not confuse two significant results with a significant difference between them
Suppose an intervention group's outcome improves significantly from baseline while the control group's outcome does not. That pattern alone does not establish that the groups changed by significantly different amounts.
The appropriate question is generally whether the relevant between-group contrast itself supports a difference, not whether one within-group test crosses a significance threshold while another does not.
Inspect the primary analysis to see what comparison the statistical model actually estimated.
Adjusted comparisons may differ from crude comparisons
An observational paper may report an unadjusted association and then a model adjusted for age, baseline values, socioeconomic variables, or other covariates. The adjusted estimate represents a conditional comparison defined by the model and its assumptions, not simply the raw difference between observed groups.
Do not describe an adjusted estimate as though researchers merely compared two group averages. Identify the variables included in the model and understand what contrast the resulting estimate represents.
The comparison may vary across analyses
A single paper can contain several comparators. The primary analysis might compare treatment A with usual care, while subgroup analyses compare effects across demographic categories. A secondary dose-response analysis might use the lowest exposure category as its reference.
Therefore, do not ask only “What was the study's control group?” Ask “What was the comparison for this particular result?”
Some studies do not have a meaningful comparator
Purely descriptive research may estimate the prevalence of a condition without comparing groups. Qualitative research may investigate experiences without establishing a formal comparator. Case series may describe characteristics without a control group.
Do not manufacture a comparison merely because familiar appraisal frameworks contain a “C.” Whether a comparator is necessary depends on the study design that was actually used and the question being asked.
06 · What This Means for You
Complete every effect statement with “compared with what?”
Whenever you encounter a claim about an increase, decrease, benefit, harm, association, or effect, mentally finish the sentence with “compared with what?” If you cannot answer that precisely, you are not yet ready to interpret the result.
A simple comparison framework
If the study has intervention and control groups
Describe exactly what participants in both groups were assigned and what care or activities each condition contained.
If the study compares exposure categories
Identify how the categories were defined and which category serves as the reference.
If the result concerns change over time
Determine whether the estimate is a within-group change, a between-group difference, or a comparison of changes between groups.
If several analyses are reported
Identify the comparator separately for the specific result you intend to interpret.
Once the comparison is explicit, many apparently simple claims become more precise. “Treatment A reduced symptoms” becomes “participants assigned to treatment A had lower symptom scores than participants assigned to usual care at the specified time point.” That is longer, but it tells you what evidence actually exists.
07 · A Quick Checklist
Before interpreting a difference or effect, identify its reference point
For the result you are reading, check:
Identify exactly which groups, conditions, exposure levels, values, or time points were compared.
For intervention studies, determine what participants in every relevant group actually received or were assigned to receive.
If the comparator is usual care, determine what usual care consisted of in that setting.
For categorical exposures, identify the reference category and how all categories were defined.
Distinguish within-group changes from between-group comparisons.
Do not infer a between-group difference merely because significance differs across separate within-group tests.
Determine whether the reported estimate is crude or adjusted and what the adjusted comparison represents.
Verify that the comparator for the specific result matches the comparison you describe in your interpretation.
