Reader Comments

Post a new comment on this article

Heterogeneity produces an unexpected result in the aggregated data

Posted by machta on 19 Mar 2014 at 15:03 GMT

The authors use a (random effects) weighted average of odds-ratios from different studies and find that this odds-ratio is significantly greater than unity. They therefore conclude that c-sections are correlated with and perhaps even cause overweight or obesity later in life.

If, instead of a weighted average, one simply takes the aggregated data found at the bottom of Figures 3 and 4 then one draws a qualitatively different conclusion. The aggregated odds-ratio for overweight following CS vs VD is 0.802. The odd-ratio for obesity is 0.876. Both numbers are "less" than unity suggesting that c-section is protective against overweight and obesity!

How can this be? It is because the odds of a c-section vary greatly between studies and the overweight and obesity rates are substantially higher in the studies with low odds of a c-section. This correlation makes c-sections appear to be protective in the aggregated data. Because of this heterogeneity, the weighted average odds-ratio used by the authors is the correct way to look for an effect.

Nonetheless, the fact that the two methods yield an opposite effect is troubling. It is indicative of serious confounding factors and I believe it calls into question the strength of the meta-analysis conclusions.

No competing interests declared.