Micronutrient Article Error

One of the figures in our micronutrient article is wrong–and in fact, it is perhaps the most important one of the paper. The correct figure is attached below and is the one that we originally submitted. The error doesn’t change our findings–these figures have the effect of showing what we found in a graphic. The wrong one tells the story that we found nothing important! The correct one shows how robust our findings actually are.

We are in the process of correcting the error with PLoSOne and will be preparing a more formal response. In the process of re-sizing the figure for editing purposes this past February, an earlier figure was used. This version appears to show no effect from the intervention.

Some might presume that this means we merely cherry-picked the data to find a palatable solution. That would be incorrect as the initial analysis was faulty. Further, the frequentist analysis underscores by comparison our findings of a robust effect.

Our team had, in setting up the analysis, initially used a VERY flat prior probability. This had the effect of squashing the data and distorting such that it no longer reflected the original findings of the studies we were analysing. The reason we can be sure this is the case is that the result did not resemble the original data. I.e., the original data was in fact, for example, Fawzi 0.56 (0.32, 0.98). You can see that clearly in the frequentist analysis (supplementary figure 1).

Upon adjusting to a more accurate scale factor, the result was a strong effect for the impact of a simple multi on the rate of HIV disease progression.

S1 Fig. Forest plot of rate of HIV disease progression, frequentist analysis.

A frequentist analysis yielded a similar 40% reduction in the rate of progression to clinical disease stages (RR 0.60, 95% CI 0.46, 0.78; p = 0.00008) for subjects on MNS, when including supplement arms that included a MNS alone or MNS plus either zinc or selenium.


This underscores also the importance of scrutinizing closely the way data are handled and analyzed. As I have learned over the years in doing this work, a preferred answer may indeed be produced. But in this case, the answer reflects the actual data of the original studies.

George M. Carter

Fig. 2 Bayesian Analysis


Leave a Reply

Please log in using one of these methods to post your comment:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s