Clustering in surgical trials - database of intracluster correlations

Table 4 Possible design effect for a continuous outcome based upon database

*Outcome*	*ICC†*	*Average cluster size*		Design effectχ
				*expertise-based design*	*stratified design*
EQ-5D 12 m	0.01	min(centre)*	9.4	1.08	0.99
	0.01	median(centre)*	32.4	1.31	0.99
	0.01	max(centre)*	51.1	1.50	0.99
	0.01	median adj.(centre)‡	64.4	1.63	0.99
	0.01	min(surgeon)*	2.8	1.02	0.99
	0.01	median(surgeon)*	6.1	1.05	0.99
	0.01	max(surgeon)*	29.9	1.29	0.99
	0.01	median adj.(surgeon)‡	35.3	1.34	0.99
Operation time	0.27	min(centre)*	9.4	3.27	0.73
	0.27	median(centre)*	32.4	9.48	0.73
	0.27	max(centre)*	51.1	14.53	0.73
	0.27	median adj.(centre)‡	64.4	18.12	0.73
	0.27	min(surgeon)*	2.8	1.49	0.73
	0.27	median(surgeon)*	6.1	2.38	0.73
	0.27	max(surgeon)*	29.9	8.80	0.73
	0.27	median adj.(surgeon)‡	35.3	10.25	0.73

ICC: Intracluster correlation coefficient.
† ICC was based upon median of observed values.
* Average cluster size based upon actual cluster sizes across trials in the database.
‡ Average cluster size calculated using the formula (∑n_i ²)/∑n_i) where n_i is the number of observations in the i^th cluster.
χ Design effect was calculated using 1+(average cluster size-1)*ρ and 1-ρ for expertise-based trial and stratified design respectively.

ISSN: 1745-6215