Analysis of cluster randomised stepped wedge trials with repeated cross-sectional samples

Table 1 Estimates of treatment effect from the Hussey and Hughes model; and model extensions

Model	Unexposed to intervention	Exposed to intervention	Odds ratio (95% CI)	ICC (95% CI)
Model	N = 1420	N = 1367	Odds ratio (95% CI)	ICC (95% CI)
Number swept	629 (44.3%)	634 (46.4%)
Unadjusted for time models
Naïve model			1.11 (0.95, 1.30)	0.069 (0.028, 0.161)
Time-adjusted models
Basic Hussey and Hughes model			0.78 (0.55, 1.12)	0.073 (0.030, 0.168)
Model extensions
A: Time by strata interaction (FE)			0.80 (0.55, 1.17)	0.075 (0.030, 0.176)
B: Time by cluster interaction (RE)^a			0.79 (0.55, 1.14)	0.073 (0.030, 0.168)
B: Time by cluster interaction (RE)^a			0.79 (0.55, 1.14)	0.078 (0.032, 0.177)
C: Treatment by strata interaction (FE)^b			0.85 (0.58, 1.23)	0.066 (0.026, 0.156)
C: Treatment by strata interaction (FE)^b			0.80 (0.57, 1.13)	0.066 (0.026, 0.156)
D: Treatment by cluster interaction (RE)^c			0.76 (0.52, 1.12)	0.016; 0.045; 0.027
E: Treatment by time interaction (FE)^d			0.86 (0.21, 3.49)	0.075 (0.030, 0.171)

CI confidence interval, ICC intracluster correlation, FE fixed-effect interaction, RE random-effect interaction. ^aICCs presented are within same cluster same period; and same cluster different period; ^bTreatment effects for two strata (hospital A and hospital B); ^cICCs presented are within same cluster both treated; and same cluster both untreated; and same cluster different treatment; ^dTreatment effect given is at mid study week 20 – others are depicted in Fig. 4. Note the ICC is reported on the logistic scale and so is not to be used for planning purposes. All models adjust for clustering
Note the summaries of number and proportion swept in first two columns are unadjusted for time and so should not be interpreted as representative of the treatment effect. Estimates from model D are using xtmelogit as melogit failed to converge

ISSN: 1745-6215