In this guide
Build a six-field evidence card
| Field | Question |
|---|---|
| People | What diagnosis, age, activity level, and exclusions defined the participants? |
| Comparison | Was the programme compared with no treatment, advice, another exercise, or equivalent attention? |
| Outcome and timing | Did it measure pain, strength, function, participation, or a laboratory mechanism—and when? |
| Magnitude | How large was the between-group difference, in what units? |
| Uncertainty | How precise is the estimate, and how certain is the body of evidence? |
| Applicability | What matches the person and setting in front of you, and what differs? |
This structure prevents a striking exercise demonstration from outrunning the actual result. The most visually unusual component may not be the component responsible for the outcome.
Separate improvement from treatment effect
People can improve over time because of natural recovery, expectations, attention, changes in other activities, or the treatment being studied. A within-group change cannot isolate those explanations. A suitable comparison helps estimate what differs when one intervention is used rather than another.
Read both the comparator and the follow-up. A programme may show an advantage over minimal care but little additional benefit over a credible, supported alternative. A short-term symptom result may not persist or translate to participation.
Read the size and the units
A mean difference on a 0–100 pain scale is expressed in those score units. A standardized mean difference uses standard-deviation units so studies measuring the same outcome with different instruments can be combined. An SMD of 0.5 does not mean “50% better.” Relative risk describes a ratio; the absolute difference depends on baseline risk.[1]
Worked example: the same relative effect can mean different absolute benefits
In a hypothetical one-year comparison, reducing a risk from 40 in 100 to 30 in 100 is a 25% relative reduction and 10 fewer events per 100 people. Reducing 4 in 100 to 3 in 100 is also a 25% relative reduction, but only one fewer event per 100. These are invented examples, not estimates of an exercise treatment.
Keep uncertainty visible
A confidence interval expresses uncertainty in an estimate. It is not the range of individual responses, and it does not reveal which person will benefit. A wide interval may include both worthwhile benefit and little benefit. A result that fails a significance threshold does not establish that two treatments are equivalent.[2]
Ask whether the study was designed to test superiority, non-inferiority, or equivalence before using those terms. Likewise, an average benefit does not promise the same result to every participant.
Judge certainty by outcome
A systematic review can still contain biased, inconsistent, small, or indirectly relevant studies. Cochrane’s framework considers risk of bias, inconsistency, indirectness, imprecision, and publication bias when judging certainty.[3] A label such as “systematic review” cannot substitute for examining those limits.
One study may measure strength precisely while offering uncertain evidence about return to sport. Keep those conclusions separate. A plausible mechanism can help explain a hypothesis without demonstrating a patient-important benefit.
Translate cautiously into practice
Use the result to inform a choice among reasonable options, with the person’s goals, resources, response, and clinical circumstances included. Identify the tested dose, the adherence achieved, and adverse events. Where a guide gives a practical example beyond what the study tested, it should say so.
When reading this site, follow the source links and compare the claim with the study’s scope. If the source does not support a statement, send the URL and passage through Contact & corrections.
Follow the evidence
References & context
- Cochrane Handbook · Choosing effect measures
Relative risk, absolute effects, and standardized mean differences answer different questions.
- Cochrane Handbook · Interpreting results and drawing conclusions
Interpret magnitude, confidence intervals, harms, certainty, and applicability together.
- Cochrane Handbook · Summary of findings and certainty of evidence
Certainty is judged by outcome and includes bias, inconsistency, indirectness, imprecision, and publication bias.