Sas statistical software package 9

Demographic characteristics in baseline and follow-up study samples in the intervention and control groups were summarised by percentages, means and standard deviations. Outcomes were compared between intervention and control groups using intention to treat analysis. Outcomes were compared at each time point and changes in each group compared between time points. These comparisons were made first without adjustment, followed by repeated-measures linear or logistic regression, as appropriate, for continuous or dichotomous measures, respectively, with random intercepts for supervisors and CHWs within the experimental group. The regression models used change between time points as the dependent variable. To account for within-subject correlation of measurements over time, we used the SAS MIXED procedure for numeric outcomes [21 ] and SAS GENMOD procedure to implement the generalised estimating equation (GEE) approach for binary outcomes [22 (link), 23 ]. Results were expressed as odds ratios (OR) with 95% CI. The p values were two-sided and statistical significance was defined at the 5% level. All analyses were conducted with type I error set at 0.05 for each pairing of dependent and independent variables. The SAS statistical software package 9.2 was used for all computations.