Sample size was calculated using methods described by Jones et al to demonstrate equivalence (to within 15%) in the primary outcome. An exacerbation rate of 10% was assumed; 168 patients were needed to give a power of 80% with significance level of 0.05. Note that since this is an equivalence trial, the emphasis of the analysis should be on confidence intervals (CIs) rather than p values because the failure to detect a difference via significance testing does not necessarily imply equivalence. However, the CI provides a range of values for the possible magnitude of difference between the two groups, and we can be reasonably confident that the “true” difference lies within this range. Thus, if all points in the CI correspond to a difference that is not clinically important (for example < 15% for the exacerbation rates), we can conclude that the groups are equivalent in this respect. The number of patients with exacerbations and readmissions were tabulated against group, and the odds ratio for the nurse compared to the doctor was calculated, with 95% CIs. fully
Exacerbation rates are expressed as the number of exacerbations per patients for descriptive purposes. The number of exacerbations per patient was tabulated against study group, and Pearson’s x2 statistic was used to assess the association. Poisson regression with a log-link function was used to model the number of exacerbations. The nurse/doctor group was included in the model, and the exponential of the coefficient for this term was used to estimate the relative exacerbation rate for nurse-led follow-up compared with doctor-led follow-up. Significance levels for these terms are taken from the likelihood ratio test, comparing the model without this term to the model including it.
Similar models were fitted to assess the effect of nurse-led follow-up and clinic appointments made. The number of clinics attended was tabulated against group and compared using a x2 test. The number of clinics attended was analyzed using Poisson regression. In these models, terms for clinic outcome, group, and outcome/group interaction were included. The interaction term was of primary interest to establish if, for each outcome, different counts could be expected from the two groups. A likelihood ratio test was used to compare the full model to the model incorporating only differences between outcomes. Two-sample t tests were used to compare the groups for changes from baseline in the AQ20 and in the dimensions of the SGRQ. The Mann-Whitney U test was used to compare the groups for change in peak flow.