Table 1

Common criteria for causation and their assessment through systematic reviews

Causal criteria	Evaluative systematic review methods
Strength of association	Comparing the relationship between presumed aetiological factor and condition under investigation, generating measures of association for individual studies, for example, ORs, and pooling results in meta-analysis if appropriate,
Consistency	Consistency of individual results across different studies (settings, tests for condition under investigation, study designs), examined graphically by L’Abbe and Forrest plots and statistically by χ² test and I² statistic.
Temporality	What comes first – condition under investigation or presumed aetiological factor? Information needs to be sought from studies with observations prior to exposure to the presumed aetiological factor.
Specificity	Often there are many purported aetiological causes for the condition under investigation. Attempts should be made to study the contribution made by the presumed aetiological factor over and above that made by any other pathologies using subgroup and meta-regression analyses.
Biological gradient	Does an increase in the presumed aetiological factor worsen the condition under investigation? This question can be studied using subgroup and meta-regression analyses for example, using test result as an explanatory variable for symptom severity.
Plausibility, coherence and analogy	Is the causation of the condition under investigation by the presumed aetiological factor biologically plausible? Does it conflict with generally known facts of the natural history of the condition? Are there analogous causal relationships between the aetiological factor and the condition? Answers may lie in the discussion sections of selected articles, which likely include reference to basic scientific studies on pathogenesis. It is useful to explore these referenced studies quantitatively and draw up a biological pathway. If appropriate a formal literature search may be undertaken to capture this literature comprehensively.
Experimental evidence	Does removing the presumed aetiological factor (eg, through treatment) eliminate the condition under investigation? This question can be addressed in randomised controlled trials. In the absence of such trials (whether or not randomised) uncontrolled comparisons may have to be included instead.