Current practice in analysing and reporting binary outcome data—a review of randomised controlled trial reports

Table 2 Principal analysis method reported for the primary binary outcome used in the 200 included studies

	Reported in the main text (n = 200)	Reported in the abstract (n = 200)
Chi-squared-style tests¹	127 (64%)	13 (7%)
Logistic regression	22 (11%)	3 (2%)
Looking at confidence interval limits²	8 (4%)	3 (2%)
Binomial regression	7 (4%)	1 (1%)
Others³	10 (5%)	0 (0%)
Not reported	24 (12%)	180 (90%)
No analysis⁴	2 (1%)	0 (0%)

¹Including Fisher’s exact and Mantel-Haenszel test
²‘Looking at confidence intervals’ refers to where the assessment of non-inferiority was made by comparing the upper or lower limits of the confidence interval, as appropriate, to the non-interiority margin
³These include Poisson models, exact binomial test, tests for non-inferiority (including Farrington-Manning), and Newcombe’s method
⁴One study reported no events and therefore did not perform the planned principal analysis. One study described a composite primary endpoint, which was not reported in the paper; the components of the composite endpoint were reported separately

ISSN: 1741-7015