Neuraminidase inhibitors for preventing and treating influenza in healthy adults

Reason for withdrawal from publication

This review was withdrawn from The Cochrane Library Issue 3, 2011 because the results are now out‐of‐date and we realise that the information therein (which has the potential for being used for policy decisions) could be used inappropriately. Please refer to the BMJ article www.bmj.com/content/342/bmj.c7258.full?sid=dc37372f‐dde0‐455c‐8725‐dd188f1fba4d for more information.

The editorial group responsible for this previously published document have withdrawn it from publication.

Feedback

Neuraminidase inhibitors for preventing and treating influenza in healthy adults, 16 July 2009

Summary

Dear Mr Jefferson  
 We have some questions on the conclusion in your Oseltamivir review especially about the prevention of complication.  You described that Oseltamivir 150 mg daily prevented lower respiratory tract complications (OR 0.32, 95% CI 0.18 to 0.57). 
 However, we have found that this conclusion is based on the other review (Kaiser 2003) and not on your own data analysis. The authors of the review were four employees of F. Hoffman‐La Roche Ltd, one paid consultant to F. Hoffman‐La Roche Ltd and Kaiser. We cannot find any raw data about this conclusion from your review.  Kaiser's review included 10 RCTs; two RCTs (Nicholson 2000 and Treanor 2003) were published as articles in the peer‐reviewed medical journal (JAMA and Lancet), but other 8 RCTs were proceedings of congress (5 RCTs), abstracts of the congress (one RCT) and meeting (one RCT) and data on file, Hoffmann‐La Roche, Inc, Nutley, NJ (one RCT).  The lower respiratory tract complication rates of these articles were summarized on table: there was no significant difference between Oseltamivir and placebo, and their Odds Ratio's (ORs) were1.81. But ORs of other 8 RCTs were 4.37. We strongly suppose that the reviewer's conclusion about the complications was mainly determined by these 8 RCTs, we should appraise the 8 trials rigidly. Without this process it's difficult to conclude that Osltamivir can prevent lower respiratory tract complications.  

Table: All lower respiratory tract complications (influenza case only) 

Nicholson 2000 + Treanor 2003 
 Complications  Placebo  Oseltamivir 150 mg 
 +              13       7 
 ‐              277  　 270   

Other 8 RCTs 
 Complications  Placebo  Oseltamivir 150mg 
 +               22         10 
 ‐              350         695 

Kaiser (Cochrane) 
 Complications  Placebo  Oseltamivir 150mg 
 +               35          17 
 ‐              627        965

Submitter agrees with default conflict of interest statement: I certify that I have no affiliations with or involvement in any organization or entity with a financial interest in the subject matter of my feedback.

Reply

Response to Hayashi's Feedback comment: critical analysis of Kaiser et al (2003)

Kaiser et al (2003) combined 10 randomised control trials (RCTs) comparing oseltamivir with placebo in the treatment of influenza. They focused on risk of complications leading to antibiotic use. A limitation of their analysis was the combining of bronchitis, pneumonia and lower respiratory tract infections which they labelled as lower respiratory tract complications (LRTC). In the original trials complications studied also included sinusitis and otitis media however these were ignored in the Kaiser et al study. In addition bronchitis is a very general diagnosis whereas pneumonia is more specific and a much more serious condition. Combining of these two outcomes is questionable. Another limitation of the Kaiser et al study involves their choice of analysis strategy: Fishers exact test. This analysis does not stratify by trial but treats the whole 10 trials as one study. Therefore the benefit of randomisation is lost, resulting in a non‐randomised comparison. To confirm that they did indeed use Fishers exact test two analyses where the actual P value is reported can be checked.

 

Hospitalisations:                              18/1063 versus 9/1350 P = 0.019 (Kaiser et al report P = 0.02)

LRTC in high risk patients:             74/401 versus 45/368 P = 0.021 (Kaiser et al report P = 0.02)

 

The resulting P values are the same (to two decimal places) therefore it is highly likely that they did indeed use Fishers exact test to compare the overall groups (without stratification). Normally in a meta‐analysis of individual RCTs, separate comparisons by trial are made and then combined in an appropriate way to obtain the overall effect of treatment. A “correct” analysis is especially important in this case because of the following facts reported in Kaiser et al:

1. The populations studied in each trial are different: healthy adults in four studies; elderly patients in four studies, and adults with chronic obstructive airways disease (COAD) in two studies.

2. Overall there are more oseltamivir patients compared to placebo patients (2023 versus 1541) hence at least one trial did not have a 1:1 allocation ratio.

3. The trials had different proportions of influenza infected patients (ranging from 50% to 73%).

4. Overall there were more high risk patients in the placebo group compared to the oseltamivir group (38% versus 27%) hence (overall) groups are not comparable.

The Kaiser et al study did not report the numbers of patients randomised to the two groups for each of the 10 trials; they just reported overall numbers. The following hypothetical meta‐analysis of two trials illustrates why a correct analysis is critical.

Table of proportions of adverse events by (hypothetical) trial

Trial number	Adverse Events
	Treatment	Placebo	Total
Trial 1 (high risk patients)	50/100 (50%)	50/100 (50%)	100/200 (50%)
Trial 2 (low risk patients)	10/200 (5%)	4/80 (5%)	14/280 (5%)
Total	60/300 (20%)	54/180 (30%)

 

This meta‐analysis shows two trials with no effect. However, the two trials have recruited much different patient groups (e.g. elderly patients in trial 1 and the general population in trial 2). Also trial 2 has not allocated with a ratio of 1:1 (as in at least one of the Kaiser et al trials). Like the Kaiser et al study there is a higher proportion of high risk patients overall in the placebo group (56% versus 33%). A naïve analysis that does not stratify by trial (Fishers exact test) shows a significant difference between treatment (20% events) and placebo (30% events) with P = 0.01 (odds ratio = 0.58). Conversely an analysis that stratifies for trial (logistic regression) shows no difference (P = 1.0, odds ratio = 1.0). In the case of the Kaiser et al data, a random‐effects meta‐analysis that takes into account heterogeneity between trials may be most appropriate.

Note that the hypothetical example shown above is “extreme”. However, it does illustrate what could happen with a naïve analysis that does not stratify by trial. The important point is that with a naïve analysis there is no guarantee of an unbiased estimate of treatment effect or a realistic 95% confidence interval and P value.  

Tom Jefferson, Mark Jones, Peter Doshi, Chris Del Mar, Liz Dooley 
 Date of inclusion: 10 November 2009

Contributors

Keiji Hayashi 
 Date of inclusion: 16 July 2009

Neuraminidase inhibitors for preventing and treating influenza in healthy adults, 30 July 2009

Summary

The last sentence under Results, preceding Discussion is: 'Finally, use of relief medications and antibiotics is unaffected by assumption of NIs (OR 0.81, 95% CI 0.59 to 1.12).'  Here 'assumption' makes no sense, so should the words in bold be 'consumption of an NI'? 

Submitter agrees with default conflict of interest statement: I certify that I have no affiliations with or involvement in any organization or entity with a financial interest in the subject matter of my feedback.

Reply

Thanks you, we have re‐written this part of the review to make it clearer.

Tom Jefferson, Mark Jones, Peter Doshi, Chris Del Mar, Liz Dooley 
 Date of inclusion: 15 November 2009

Contributors

Andrew Herxheimer 
 Date of inclusion: 30 July 2009

What's new

Date	Event	Description
12 November 2009	New citation required and conclusions have changed	1. We excluded two new studies (Blumentals 2007 and Toovey 2008). 2. We now study pharmacovigilance data. 3. We excluded a previously included study (Kaiser 2003) as we could not answer the Hayashi comment by reconstructing the Kaiser 2003 data set. Hayashi prompted us to more carefully evaluate the Kaiser 2003 study. More critical evaluation of it leads essentially to a retraction of our 2006 and 2009 updates of this review. It results in changed conclusions: excluding the Kaiser 2003 data, and failing to identify sufficient toxicity data from pharmacovigilance sources, we conclude that there is insufficient evidence to describe the effects of oseltamivir on complications of influenza and its toxicity. 4. There is a change in authors of the review team. 5. The review was published in a print journal in a shortened form, December 2009 (Jefferson 2009e).
7 August 2009	New search has been performed	Safety/adverse effects searches conducted.
30 July 2009	Feedback has been incorporated	Feedback added to review.
16 July 2009	Feedback has been incorporated	Feedback added to review.
14 July 2009	New search has been performed	Effectivess searches conducted.

History

Protocol first published: Issue 1, 1999
 Review first published: Issue 2, 1999

Date	Event	Description
20 May 2008	New search has been performed	Searches conducted in May 2008. For this update we assessed 688 possible studies, retrieved 17 and excluded all of them. Our conclusion did not change but we found non‐comparative phase IV evidence from a thorough review of the evidence on harms by Hama which we mentioned in the Discussion section. Updated review published in issue 2, 2009.
29 April 2008	Amended	Converted to new review format.
19 May 2006	New citation required and conclusions have changed	Substantive amendment published in Issue 3, 2006.
13 October 2005	New search has been performed	Searches conducted in October 2005. We completely revised the text and added a section on evidence from an avian influenza epidemic that took place in the Netherlands in 2003 and claimed one life. We also added a section on post‐exposure prophylaxis (PEP). We dropped studies looking at the effects of neuraminidase inhibitors (NIs) on experimental influenza cases (that is to say, on subjects who had been deliberately infected as part of an experiment) and concentrated on the now numerous studies of naturally‐acquired influenza cases. The terms "laboratory‐confirmed influenza" and "clinically confirmed influenza" have been changed for the more correct terms "influenza" and "influenza‐like‐illness" (ILI). We believe these words to reflect the difference between real influenza (caused by influenza A and B viruses) and what is colloquially known as "the flu". The two are rarely clinically distinguishable in real‐time unless a very good surveillance apparatus is in place, as in most of the trials in our review. Updated review published in Issue 3, 2006.
24 February 1999	New search has been performed	Review first published in Issue 2, 1999.

Sources of support

Internal sources

• The authors' own institutions (2005 update), Italy.

• The author's own institutions (2005 update), Australia.

External sources

• National Instiutue of Health Research (NIHR), UK.

  Competitive grant awarded through the Cochrane Collaboration

• National Health and Medical Research Council (NHMRC), Australia.

  Competitive grant awarded to Chris Del Mar and Tom Jefferson, 2009

Withdrawn from publication for reasons stated in the review, comment added to review