What is a level of evidence?
A level of evidence measures how much confidence can be placed in the result of a study, or a body of studies, to answer a clinical question. It depends first on the study design (a randomized trial controls bias better than a case series), then on how well the study was conducted: sample size, consistency of results, relevance of the outcomes measured. Clinical practice guidelines rely on the level of evidence to set the strength of each recommendation.
The hierarchy of studies, in practice
Not all studies carry the same weight for a given question. To assess the effect of an intervention, they are usually ranked from most to least reliable:
- Systematic reviews and meta-analyses of randomized trials: they gather and pool all available trials on a question.
- Randomized controlled trials: random allocation spreads confounding factors evenly across groups.
- Cohort studies: people exposed or not exposed to a factor are followed over time.
- Case-control studies: people with and without a condition are compared retrospectively.
- Case series and expert opinion: useful to raise a hypothesis, not enough to demonstrate it.
This pyramid is a simplification. A poorly conducted randomized trial can carry less weight than a large, rigorous cohort, and some questions (rare adverse effects, prognosis) are better answered by observational studies.
The French National Authority for Health (HAS) grades
In France, the HAS links the level of evidence of studies to a grade of recommendation. The grid below is the one presented in its methodological review published in 2013.
| Grade | Meaning | Supporting studies |
|---|---|---|
| A | Established scientific evidence | Level 1: high-powered randomized comparative trials, meta-analyses of randomized comparative trials, decision analyses based on well-conducted studies |
| B | Scientific presumption | Level 2: low-powered randomized comparative trials, well-conducted non-randomized comparative studies, cohort studies |
| C | Low level of evidence | Level 3: case-control studies. Level 4: comparative studies with major bias, retrospective studies, case series, descriptive epidemiological studies |
| AE | Expert agreement | In the absence of scientific data: approval by at least 80% of the working group members |
A point often misunderstood: a grade C or an expert agreement does not mean that an intervention is ineffective. It means the available data do not yet allow a firm conclusion. The HAS itself notes that a grade C or expert-agreement recommendation can be strong despite the lack of scientific support.
The GRADE approach
The GRADE approach, adopted as early as 2008 by more than 25 organizations, including the World Health Organization and the Cochrane Collaboration, separates two notions that the French grid links directly: the certainty of evidence and the strength of the recommendation.
- Certainty of evidence: high, moderate, low or very low. Randomized trials start at high certainty, observational studies at low certainty.
- Factors that lower certainty: study limitations (risk of bias), inconsistency of results, indirectness of the data, imprecision, publication bias.
- Factors that raise it, mainly for observational studies: a very large effect, a dose-response gradient, plausible biases that would all reduce the observed effect.
- Strength of recommendation: strong or weak (also called conditional). It is strong when desirable effects clearly outweigh undesirable ones, or clearly do not; it is weak when this balance is uncertain, because of low-certainty evidence or closely balanced effects. Patient values and preferences are explicitly taken into account.
Why therapeutic exercise is harder to evaluate
Movement-based interventions raise specific methodological challenges, which explain sometimes modest levels of evidence despite real clinical benefits.
- Blinding is difficult or impossible: patients know whether they exercise or not, which increases the risk of bias.
- Protocols vary widely: type of exercise, intensity, frequency, duration and progression differ from one study to the next, which complicates meta-analyses.
- Adherence shapes the result: a program that is poorly followed looks less effective than it really is.
- Interventions are often incompletely described, which hampers interpretation and replication of results. The TIDieR checklist (2014, 12 items) and the CERT checklist (2016, 16 items, specific to exercise programs) were created to address this.
Certainty is also judged outcome by outcome. In coronary heart disease, the 2021 Cochrane review of exercise-based cardiac rehabilitation rates certainty as high for the reduction in myocardial infarction at 6 to 12 months, moderate for the reduction in all-cause hospitalization, and low for cardiovascular hospitalization.
Reading a guideline: four questions to ask
What is the grade?
Look for the grade (A, B, C, AE) or the GRADE strength attached to each recommendation, not just to the document.
Based on which studies?
Check the type and number of studies, and the population included: does it resemble my patients?
Which outcome?
A benefit on pain, function or quality of life does not carry the same weight as a change in a biological marker.
How recent?
An older guideline may have been overtaken by newer trials: check the date of the last literature review.
This reading is what turns a guideline into a program: for adapted physical activity as for any functional rehabilitation, the level of evidence guides the choice of exercise, dose and progression.
Frequently asked questions
What is the difference between level of evidence and grade of recommendation?
The level of evidence describes how reliable the available studies are. The grade of recommendation describes how strong the resulting advice is. In France, the HAS links the two: level 1 studies support a grade A recommendation.
What does an HAS grade A mean?
A grade A means established scientific evidence, based on level 1 studies: high-powered randomized comparative trials, meta-analyses of randomized comparative trials or decision analyses based on well-conducted studies. It is the highest level of confidence in the French grid.
Does a low level of evidence mean a treatment does not work?
No. A low level of evidence means the available data do not yet allow a firm conclusion. The intervention may be effective, but studies are lacking, small or at risk of bias.
What is the GRADE approach?
GRADE is an international method for rating evidence, adopted by organizations including the WHO and Cochrane. It classifies certainty of evidence into four levels (high, moderate, low, very low) and distinguishes strong from weak recommendations, also called conditional.
Why are exercise studies harder to evaluate?
Because blinding is difficult or impossible, protocols vary widely between studies and participants' adherence influences the result. Programs are also often incompletely described, which the TIDieR and CERT checklists aim to correct.
Where can I find the level of evidence of a French guideline?
In the guideline itself: HAS clinical practice guidelines state the grade of each recommendation. The method and references are detailed in the supporting document.
Editorial review in progress by MYA's editorial committee.
Editorial content intended to inform health professionals. It does not constitute individual medical advice and does not replace assessment by a professional.