“Observational data are useful to answer many questions, such as whether treatments are harmful or not. However, the fact that experimental data from randomised trials are less prone to error from bias than are observational studies is not debatable.”

Del Mar C, Doshi P, Hama R, Jones M, Jefferson T, Heneghan C, et al . Neuraminidase inhibitors for influenza complications. The Lancet 2014; 384: 1260-61.

In their Comment (Aug 2, p 386), Jonathan Nguyen-Van-Tam and colleagues suggest that findings from our Cochrane review² and a study of observational data³ are consistent. In our review, which was based on full clinical study reports of all manufacturer-sponsored randomised trials, we did not find evidence that neuraminidase inhibitors improve important outcomes of influenza, whereas the Roche-funded individual analysis of a subset of retrospective case reports suggested that neuraminidase inhibitors do have some beneficial effect. These observational studies were of patients admitted to hospital for influenza, some of whom apparently benefited from neuraminidase inhibitors.³ Most treated patients received oseltamivir, with a minority receiving zanamivir. Similar evidence was cited in a statement from Roche.⁴ Our Cochrane review did not include similar patients, but was based on typical patients with influenza-like illness.

However, the hypothesis that oseltamivir protects against complications of influenza proposed by Nguyen-Van-Tam and colleagues, together with the observational studies that they cite in support of their hypothesis,³ can be dismissed by our finding that neuraminidase inhibitors did not seem to reduce severe complications of influenza, and oseltamivir specifically did not seem to reduce risk of admission to hospital (admission to hospital was not reported for zanamivir), but seemed to suppress production of antibody to the virus and cause potentially serious side effects.² Furthermore, oseltamivir reduced time to alleviation of symptoms the least for trials of patients with co-morbidities such as asthma and chronic obstructive airways disease.²

The appeal to observational data in support of oseltamivir is curious because the original notion that oseltamivir was effective for treatment of influenza complications came from a meta-analysis⁵ of a subset of randomised trials—both published and unpublished—which was in contrast to the US Food and Drug Administration’s requirement that the drug label made no claims for effectiveness for influenza complications. ⁶ This absence of effectiveness is supported by our Cochrane review (based on a full set of clinical study reports), which found no protective benefit of oseltamivir. Now drug manufacturers and Nguyen-Van-Tam and colleagues are claiming that randomised trial data are no longer adequate. Yet, when the results of reviews of randomised trials agreed with their views, Nguyen-Van-Tam and colleagues accepted the reviews. Now that the Cochrane review does not support their hypothesis, they appeal to contradictory real-world (observational) data.³ This inconsistency needs to be addressed.

Observational data are useful to answer many questions, such as whether treatments are harmful or not. However, the fact that experimental data from randomised trials are less prone to error from bias than are observational studies is not debatable. The randomised trial data should therefore be trusted—insufficient evidence exists to justify conclusions about oseltamivir’s effectiveness for treatment of influenza. Nguyen-Van-Tam and colleagues seem to suggest that pandemic planners should act to minimise public outcry rather than accept the best available evidence. This advice could risk scarce health-care resources stockpiling a probably ineffective drug.