1. Unit of assessment
Every rating should identify the intervention or marker, specific indication, population, comparator, outcome, time horizon, and relevant harms. ‘Magnesium has strong evidence’ is therefore incomplete; magnesium for a defined outcome in a defined group may have one rating while another claim has a different one.
2. Source hierarchy is question-dependent
- Guidelines, systematic reviews, meta-analyses, and randomized trials usually receive priority for benefits and harms.
- Official regulatory and public-health sources govern labeling, safety alerts, diagnostic criteria, and legal context.
- Observational evidence may be the best available source for rare harms, long latency, prognosis, or exposures that cannot ethically be randomized.
- Mechanistic, animal, cell, surrogate, and uncontrolled evidence may explain plausibility but cannot be silently upgraded into proven patient benefit.
3. Ratings
- Strong: consistent, high-confidence support for the defined claim.
- Moderate: meaningful support with relevant uncertainty.
- Limited: suggestive findings with important limitations.
- Insufficient: evidence cannot support a reliable conclusion.
- Benefit not demonstrated: adequate evidence has not shown the claimed benefit.
- Potential for harm: clinically relevant safety evidence or signals alter the decision.
4. Rating process
Ratings consider risk of bias, consistency, directness, precision, magnitude, clinical relevance, publication bias, dose and formulation, follow-up, applicability, and the balance of benefits and harms. A rating is not a prescription and may differ from the certainty label used by a guideline.
5. Localization and change
The scientific core is shared across markets, but units, diagnostic thresholds, product regulation, approved claims, standard drinks, screening guidance, and legal language are localized. Every article must either display the material Brazil–United States differences relevant to the decision or explicitly state that no material difference was identified in the core evidence and recommendation. Silence is never treated as proof of equivalence. Ratings may change when better evidence emerges; the current review date and material corrections should be visible.
Official sources