Tamiflu and adverse effects

For reasons not revealed to anyone I know, WHO is saying the Tamiflu resistance in a Danish swine flu isolate is "isolated case." Could be, but I'd sure like to know why they think so, other than they don't have any other examples. Meanwhile WHO and CDC continue to advise prompt use of Tamiflu (oral oseltamivir) for treatment of swine flu in high risk patients. Which brings up the question of side effects.

The most common Tamiflu side effect is nausea and vomiting. When my daughter was seen at the local hospital ER for swine flu symptoms and a positive rapid test for influenza A at a time when essentially the only circulating flu virus was swine flu, she was offered a course of Tamiflu with the additional counsel that it made many people feel worse, not better. Since she had just gotten 4 liters of IV fluids because of a night of vomiting, Tamiflu-induced nausea wasn't too high on her list (unlist she held the list upside down). She declined and recovered after a few days. But if she'd been sicker or had an underlying medical condition, a course of Tamiflu might have made sense. Is nausea and vomiting the only thing to worry about?

As with most drugs, there is the rare but nasty allergic reactions, including serious skin reactions like Stevens-Johnson syndrome. But the side effect that has caught the most attention are reports that young people taking Tamiflu have had hallucinations and suicidal behaviors. 75% of all Tamiflu use in the world until the swine flu outbreak was in Japan and most, but not all, of the reports are from that country. In November 2006 the US FDA ordered Tamiflu carry a "black box warning" of the possibility of this adverse event:

The FDA knows of 103 reported cases -- 95 from Japan, five from the U.S., and three from other countries. Sixty of the 103 cases featured delirium with prominent behavioral disturbance such as panic attack, hallucinations, and convulsions; only one of those cases occurred in the U.S.

[snip]

"People with the flu, particularly children, may be at an increased risk of self-injury and confusion shortly after taking Tamiflu and should be closely monitored for signs of unusual behavior," states Tamiflu's new warning.

"Patients with should be closely monitored for signs of abnormal behavior throughout the treatment period," the warning also notes.

"A healthcare professional should be contacted immediately if the patient taking Tamiflu shows any signs of unusual behavior," states the warning.(WebMD)

A black box warning (named for the color of the box border surrounding the text) is the strongest warning FDA has that the drug might present serious adverse effects and is placed in the package patient insert accompanying the drug. Whether Tamiflu actually presents such a risk has been debated. Japanese authorities have funded two epidemiological studies in children. The first and smaller study suggested an increased risk of abnormal behaviors on day 1 of infection, but additional study was recommended. An interim report of a much larger study of more than 10,000 children conducted during the 2006-2007 flu season was said to show no association, it was subsequently criticized for a flawed analysis. This week a corrected analysis was published in the journal, Epidemiology.

The analytic problem is of interest to show how doing these studies is not always straightforward. Here's the basic cohort study design:

In the second and larger study, sets of 10-20 successive patients infected with influenza virus were recruited from 692 institutions, for a total of 11,661 patients. Data regarding the prescription and intake of the drug, as well as adverse behaviors, were obtained from questionnaires completed by both the doctors and the patients' families.11 Families recorded adverse behaviors during the first 4 days from the onset of fever, and the questionnaires were submitted to their doctors. Doctors evaluated these questionnaires and completed their own questionnaires. The adverse behaviors consisted of hallucinations, delirious speech, frightening episodes, abrupt anger, abnormal activities leading to accidents, and putting anything unusual into the mouth. The accuracy of adverse behaviors in the doctors' questionnaires was checked by pediatricians in the research group who were unaware of prescription status.

At first blush it would seem that we could just compare the number of abnormal behaviors in the treated group with abnormal behaviors in the untreated group. 7586 children diagnosed with flu got Tamiflu and there were 988 recorded adverse behaviors (13%). In the 2129 children who didn't get Tamiflu, 187 had adverse behaviors (9%). This was a statistically significant difference, suggesting Tamiflu was associated with adverse behavior. But it turned out that the adverse behavior of 99 of the children in the Tamiflu group started before Tamiflu treatment. In effect, these children were really "untreated" at that point. So for the analysis the original research report moved the 99 to the untreated group, adding their adverse behaviors to that total and subtracting them from the total in the treated group. This produced a slight deficit in adverse behaviors in the Tamiflu group (12% versus 13%). That's where things stood in the interim report in July 2008.

So what was wrong with this? The 99 adverse events counted in the untreated group contributed a significant number of "person-days at risk" that were still counted as part of the treated group but should have been transferred, along with the numerator, to the untreated group:

Since patients were followed for 4 days from the onset of fever, the maximum possible person-time would be 30,344 person-days in the prescribed group and 8516 person-days in the nonprescribed group. However, the person-days in the prescribed group consist of 2 parts, namely the person-days before oseltamivir intake and the person-days after oseltamivir intake. In the nonprescribed group, 187 subjects manifested adverse behaviors from 8516 person-days, which is 0.022 incidences per person-day. Assuming that this incidence is homogeneous over the 4 days; we can assume that the rate of adverse behaviors in the nonprescribed group is equal to the corresponding rate in the prescribed group before oseltamivir intake. From this, we can estimate the number of person-days in A by calculating 99 incidences divided by 0.022, leading to 4500 person-days. This leaves 25,844 person-days after oseltamivir intake.

Using these figures, there are 2 analytic options described by Rothman: exclusion of the patients manifesting adverse behaviors and their corresponding person-days before oseltamivir intake in the prescribed group, or reassignment of these to the nonprescribed group. With the first method, the rate of adverse behaviors in the prescribed group is 0.034 (889 incidences divided by 25,844 person-days) and the corresponding rate in the nonprescribed group is 0.022 as described above, leading to a rate ratio of 1.57 (95% CI = 1.34-1.83). With the second method, the rate of adverse behaviors in the prescribed group is 0.034 (as in the first method) and the corresponding rate in the nonprescribed group is still 0.022 (187 plus 99 incidences divided by 8516 plus 4500 person-days), leading again to a rate ratio of 1.57 (1.37-1.79). Thus, using either method, it is clear that when children are infected with influenza, oseltamivir intake increases the rate of adverse behaviors by 1.5-fold compared with nonusers. (The 95% CIs for the rate ratios were estimated using the formula provided in Rothman.) (Yorifuji et al., Epidemiology, Volume 20(4), July 2009, pp 619-621; cites and figure references omitted)

This sounds complicated but it isn't. The bottom line is that the original analysis deflated the risk of the treated group while inflating the risk of the untreated group, making it appear there was no difference when there was once the correction was made. Is this the final word? Not really. Because there was insufficient information available in the report about the comparability of the two groups on essential factors like age, fever, other drugs, severity, etc., it is possible that the differences would have to be adjusted to take them into account. Moreover the 4 day window for observing adverse behaviors is fairly long compared to the peak plasma values of oseltamivir which are attained in 4 to 5 hours. What we can say at this point is that the best data at hand seem to confirm the impression that Tamiflu can induce bizarre and at times dangerous behavior in adolescents also infected with influenza.

Given the fact that in a pandemic millions of people could be receiving this drug, it is hardly an unimportant question. Important or not, however, it is not an easy one to answer.