Minimum number of clusters and comparison of analysis methods for cross sectional stepped wedge cluster randomised trials with binary outcomes: A simulation study

Table 1 Power to detect an intervention effect (OR = 2.25) in scenario A with different methods of analysis

ICC	k	n _jk	True time effect OR = 1				True time effect OR = 1.227
ICC	k	n _jk	GEE	GLMM	Cluster summaries method	Fixed effects model	GEE	GLMM	Cluster summaries method	Fixed effects model
0.01	3	100	0.806	0.765	0.657	0.735	0.858	0.828	0.670	0.801
	6	50	0.813	0.802	0.740	0.737	0.866	0.852	0.785	0.812
	6	100	0.971	0.965	0.929	0.955	0.979	0.980	0.945	0.969
	9	50	0.929	0.926	0.907	0.893	0.948	0.947	0.920	0.930
	9	100	0.998	0.998	0.993	0.996	0.999	0.999	0.995	0.998
	18	10	0.665	0.653	0.632	0.535	0.736	0.724	0.705	0.583
	18	50	0.997	0.998	0.995	0.989	0.999	0.999	0.999	0.999
	36	5	0.690	0.683	0.675	0.549	0.773	0.762	0.747	0.625
	36	10	0.918	0.916	0.908	0.806	0.953	0.953	0.947	0.874
0.05	3	100	0.698	0.690	0.464	0.684	0.734	0.733	0.401	0.750
	6	50	0.697	0.702	0.590	0.695	0.764	0.766	0.563	0.749
		100	0.925	0.929	0.810	0.923	0.962	0.966	0.758	0.961
	9	50	0.859	0.863	0.782	0.860	0.900	0.904	0.755	0.899
		100	0.984	0.983	0.951	0.982	0.996	0.995	0.934	0.995
	18	50	0.987	0.986	0.978	0.986	0.998	0.999	0.985	0.998
	36	10	0.826	0.820	0.807	0.778	0.874	0.871	0.848	0.832
0.1	3	100	0.597	0.619	0.320	0.621	0.659	0.685	0.229	0.692
	6	50	0.623	0.647	0.475	0.645	0.699	0.719	0.399	0.715
		100	0.871	0.888	0.673	0.892	0.919	0.936	0.534	0.932
	9	50	0.811	0.827	0.693	0.820	0.847	0.860	0.599	0.850
		100	0.968	0.971	0.865	0.970	0.979	0.984	0.765	0.984
	18	50	0.985	0.987	0.964	0.986	0.992	0.992	0.941	0.992
		100	1.000	1.000	0.999	1.000	1.000	1.000	0.986	1.000
	36	10	0.763	0.766	0.740	0.752	0.838	0.832	0.793	0.808

Only scenarios where at least one method had a power of between 0.7 and 1 are shown. Each estimate is based on 2000 simulations. All methods adjust for time in the model

ISSN: 1745-6215