Evaluating drug toxicity signals: is a hierarchical classification of evidence useful or a hindrance?

Only randomized clinical trials (RCTs) are deemed suitable to determine causality, as epidemiological studies are considered not 'robust'. The objective of this study was to evaluate whether causality should remain the only consideration with drug toxicity signals. Cyclooxygenase-2 (Cox-2) inhibitors provided an example.

Our study population included patients aged 40+ years prescribed Cox-2 inhibitors in the General Practice Research Database (GPRD) (N = 150 000). We estimated the risks of upper gastrointestinal (GI) events and myocardial infarction (MI). Attributable risks were estimated using simulation methodology based on various hypothetical scenarios.

The risk-benefit profile was strongly related to the rate of GI events. With a RCT incidence, the GI benefits would exceed MI risks. With a 'real-life' GI incidence, the benefits did not exceed the risks substantially. The onset and offset of drug effects also predicted the magnitude of both risks and benefits. If risks and benefits occurred in different sub-groups, the risk-benefit profile varied substantially. Also, it was found that any restriction of use to patients at high risk of drug toxicity may not improve the risk-benefit profile when this restriction affected patients who would benefit most.

We conclude that rigid classification of evidence is not appropriate in the monitoring of risks and benefits and all valid study evidence--not only that derived from a RCT--needs to be included. The first priority should be to consider the potential impact of a drug toxicity signal.