There is no "gold standard" method for assessing the impact of medicine use on health outcomes. In its absence, a convergence of evidence from different types of studies using multiple methods of independent imperfection provides the best reason for attributing improvements in health outcomes to the use of medicines (see table 1).
The major requirement for being able to do so is good evidence that:
1. a safe and effective medicine is being appropriately prescribed in clinical practice;
2. there is covariation between medicine use and improved health outcomes;
3. we can discount alternative explanations of the covariation, leaving medicine use as a plausible explanation of the improved health outcomes.
The strongest possible evidence for an inference that the use of a medicine has improved population health outcomes would be provided by the coherence of the following types of evidence:
1. Individual linked data showing that patients are prescribed the medicine, there are reasonable levels of patient compliance, and there is a relationship between medicine use and health improvements that is not explained by other factors;
2. Evidence of aggregate improvements in these health outcomes in the population in which the medicine is used;
3. The replication of these results in comparable countries;
4. Consistent trends in population vital statistics in countries that have introduced the medicine;
5. Epidemiological modelling that changes observed in population health outcomes are plausible, given the epidemiology of the condition, and the clinical effectiveness of the medicines (after discounting for the decline in efficacy observed in RCTs to that expected in routine clinical practice).
In the absence of individual linked data the next best evidence comes from the coherence of the following types of evidence:
1. Trends in population health outcomes that covary with rates of exposure to the medicine in defined sub-populations (e.g. age, sex or geographic area);
2. Similar trends in comparable countries that have introduced the medicine;
3. The absence of improvements in mortality in comparable countries that have not introduced the medicine;
4. Evidence from epidemiological modelling that the changes observed in population health outcomes are plausible given the epidemiology of the condition and the clinical effectiveness of the medicines (after discounting for decline of efficacy in routine clinical practice).
Evidence from ecological and small observational studies warrants less confidence in causal inference than data from large-scale linked data sets. It may be the best that is available if the community is unprepared to allow record linkage in the absence of individual consent or governments are unprepared to invest in the necessary infrastructure to permit data linkage on medicines use and health outcomes. These limitations of such data provide a strong case for record linkage.
Answers to all your Biology Questions
