Oseltamivir and Abnormal Behavior

To the Editors: In 2006–2007, the Japanese Ministry of Health, Labor and Welfare funded a large observational study to examine the relationship between oseltamivir exposure and risk of abnormal behavior. They did not find an association, either overall or in most subcategories.1 Yorifuji et al disagreed with the analytic approach in the original report and proposed that the person-time prior to drug exposure from 99 treated patients should have been shifted to the nontreated group. Their recalculation resulted in an estimated crude rate ratio (RR) of 1.57 (95% confidence interval = 1.37–1.79).2 Conceptually, we believe the approach suggested by Yorifuji et al is more appropriate than the risk (odds) ratio calculation in the original study. However, the approach by Yorifuji et al introduces additional important potential biases that could have considerable impact on the results. First, with a fixed 4-day follow-up period starting with fever onset in both groups, shifting person-time (prior to treatment) to the untreated group reduces the average follow-up in the treated group (to approximately 3.4 days). This difference in duration can result in a biased estimate of the RR, especially if abnormal behavior is associated with time since fever onset. Second, in both the original study and the reanalysis, 201 treated cases (20%) with indeterminate drug-event sequence were included among those who experienced abnormal behavior after drug intake.1,2 The resulting estimate may be exaggerated because some cases may actually have had the event prior to treatment. Third, Yorifuji et al assumed that the incidence of abnormal behavior is homogeneous over the 4-day window. However, if the incidence is actually higher in the hours immediately after fever onset (and before treatment initiation), then transferring 4500 person-days (as Yorifuji et al proposed) from the treated to untreated group would result in an overestimate of person-time in the nontreated group and bias the result toward a positive association. Finally, differential disease expression between the 2 groups may be associated with differences in prescription behavior (confounding by indication) and lead to bias in the estimate. Overall, the evidence from published literature supporting an increased risk of abnormal behaviors/neuropsychiatric adverse events in influenza patients treated with oseltamivir3,4 is relatively limited. We welcome additional research that can further clarify our understanding of this relationship. However, we believe potential biases introduced in this reanalysis may be substantial, making results difficult to interpret. Michael M. Peng Hoffman-La Roache, Inc. Nutley, NJ [email protected] Noah Jamie Robinson F. Hoffmann-La Roche, Ltd.