For comparisons including only one trial, we used the "treat as one trial" method of calculating the NNT [ 7].
There is also little difference between the results using standard meta-analysis and the treat-as-one-trial method, but as we noted above, although use of the treat as one trial method increases the risk of bias, bias will not always be seen.
When more than one trial was included in the comparison, we calculated NNT from pooled odds ratios and risk ratios reported in the included meta-analyses using the Visual Rx, version 2; this method is less prone to bias than the "treat as one trial" method of NNT calculation [ 8, 9].
While the standard approach indeed gives a pooled risk difference of 6%, the treat-as-one-trial method pools two trials each with a risk difference of 6% and gets an estimated effect of 3%.
Shifting cultivators unanimously applied mixed cropping as a trial and error method for finding the spots where crops perform best, while semi-permanent cultivators predominantly planted crops in rows (P < 0.01).
For example, it would be misleading not to report the use of the treat-as-one-trial method.
The treat-as-one-trial method of calculating numbers needed to treat should not be used as it is prone to bias.
Adding numbers may have some value for simple descriptive purposes, but the treat-as-one-trial method should not be used for substantive analysis.
The 'treat-as-one-trial' method was found to be susceptible to bias when there were imbalances between groups within one or more trials in the meta-analysis (Simpson's paradox).
We also note that, compared to standard meta-analysis, the treat-as-one-trial method gives greater weight to large trials and will tend to give narrower confidence intervals.
We will refer to two stage meta-analysis (as recommended by Cates) as standard meta-analysis, and adding up the numbers (as recommended by Moore et al) as the 'treat-as-one-trial' method.
