When we say that medicine use is a cause of an improved health outcome we mean that it is a contributory cause of the improved health outcome in the sense that use of the medicine is one of a complex set of conditions that jointly produced the improvement in health. In order to infer that medicine use is a contributory cause of an improved health outcome we need evidence that: (1) medicine use and the health outcome covary; and (2) evidence that makes other explanations of the relationship implausible, leaving medicine use as a plausible contributory explanation of the improved health outcome [6-8].
Assessing covariation
We can assess whether medicine use and a health outcome covary in experiments (such as randomised controlled trials) or in observational studies (e.g. ecological, case-control, cohort, time series and cross-sectional studies).
Excluding alternative explanations
A and B may be correlated without being causally related. Hence, in order to make a case for a causal relationship we need to exclude plausible alternative explanations of the relationship [7-9]. Experiments provide the strongest basis for excluding alternative explanations of covariation but they are expensive and difficult to conduct. Observational designs are easier to enact but provide a weaker warrant for causal inferences because of their limitations in excluding the following alternative explanations [8,10].
Chance?
We can assess the plausibility of chance by constructing a confidence interval around the measure of covariation between medicine use and the health outcome. If the confidence interval does not include the value consistent with the absence of a relationship between medicine use and the health outcome, then we can infer that medicine use and a health outcome covary [11,12].
Cause or consequence?
If medicine use is a cause of an improved health outcome, then medicine use should occur before the improvement. Experiments and cohort studies (which measure medicine use before assessing the health outcome) provide the best basis for deciding which is cause and which is consequence [3].
A common cause?
If medicine use and the health outcome covary, and medicine use precedes the improved health outcome, we then have to exclude the possibility that a common cause explains the relationship between the two. Experiments provide the best evidence against a common cause because randomisation to an active medicine or a placebo ensures that subjects differ only in whether or not they have been exposed to the medicine [12,13]. When subjects are randomly assigned to a medicine or a placebo then all other causal factors will be equally distributed between the two groups [14] and hence, any difference between the two groups can be attributed to medicine use.
Randomisation is not infallible because there are "threats to validity" that may arise after random assignment that may undermine the equivalence of the two groups. There may, for example, be differential rates of drop-out from the two treatments, or subjects who have been assigned to the control treatment may obtain active treatment elsewhere [12,15]. If there are no such threats to validity, then experiments provide a stronger warrant for causal inferences than observational studies because the former exclude more alternative explanations than the latter.
Answers to all your Biology Questions
