Key points are not available for this paper at this time.
The one-step (Peto) method for obtaining pooled effect estimates can yield extremely biased results when applied to unbalanced data. Even for balanced studies, the one-step estimate may incorporate an unacceptable degree of bias. In place of the one-step estimate, we recommend use of ordinary Mantel-Haenszel, weighted least squares, or maximum likelihood estimates whenever the total number of events is adequate for such methods. If the total number of events is small, we recommend exact methods.
Greenland et al. (1990) studied this question.