Table 1 Approaches to studying the effects of medicines on health outcomes

Randomised Controlled Trials and meta-analyses of such trials

• Gold standard evidence for causal relationship by virtue of randomisation to treatment

May not predict effects of medicines on health outcomes because:

• May be too small to detect rare adverse events

• May be too short to detect long term adverse effects

• May exclude high risk patients e.g. those with comorbidity

• May involve optimal treatment and compliance

Linked data on individuals

• Links data on medicine use and health outcomes in individuals

• Closer to routine clinical practice than evidence from RCTs

• Cheap and quick to do retrospectively

• Confounding by indication: patients who use medicines are at a higher risk of a disease

• Limited assessment of confounders e.g. comorbidity, OTC drugs, alcohol & tobacco

• Often uses treated morbidity as a proxy for comorbidity

Ecological studies

• Simple and cheap to do because use existing data on medicines and health outcomes

• Directly examine relationships between population medicine use and health outcomes

• Use aggregate rather than individual level data

• Crude measures of medicine use e.g. drug sales or scripts

• Limited capacity to exclude alternative explanations such as changes in risk factors, and increased use of other treatments

Epidemiological modelling

• Mathematical synthesis of epidemiological data on the disease and clinical trial data on safety and efficacy of medicines

• Simplifications of complex natural history of disease

• Uncertainties about long term effects of medicines (addressed by sensitivity analyses)

• Underdeveloped in studies of effects of medicines on health outcomes