Error in meta-analysis data
Dr. Jefferson et al have produced a provocative report in conjunction
with updating the Cochrane review and there will be many questions about
public availabilty of data and which data to use when conducting
Meanwhile we still have a key question of analysis of efficacy for
the data we choose to use.
Assuming we limit our analysis of the effect of oseltamivir on
complications to 3 trials (Nicholson 2000, Treanor 2000, Li 2003) the
results as shown in Figure 6 of the new meta-analysis reports no
significant difference with an odds ratio of 0.55 and 95% confience
interval of 0.22 to 1.35.
Unfortunately this data appears to be incorrect. Instead of entering
the data as number of events/total number of patients as implied by the
figure labeling and processed in the statistical analysis, the numbers
actually reported appear to be number of events/number of people without
For example, when reviewing the full-text of the Treanor 2003 article
there are 129 patients in the placebo group and 14 of these patients had
antibiotic use for secondary complications. In Figure 6 this data is
reported as 14/115 instead of 14/129.
This difference in data entry may be sufficient to change the
statistical significance and overall conclusion.
Editor-in-Chief of DynaMed (www.ebscohost.com/dynamed)
Competing interests: No competing interests