Wouter Leclercq

Chapter 10 164 Statistical analysis We calculated the sample size needed for our study using the following procedure. Based on a previous study, we expected a hazard ratio (HR) of 1.4 for the difference in return to normal activities between both groups. 25 Using a two-sided log-rank test at a significance level of 5%, we needed to observe 285 events (participants returning to normal activities) to achieve a power of 80%. We aimed for a total of 308 participants (154 in both groups), allowing for an expected proportion of 2.5% of participants not returning to normal activities within the 6-month follow-up period and a dropout rate of 5%. We did analyses in SPSS version 22. We did statistical analyses according to the intention- to-treat (ITT) principle. We also did per-protocol analyses, which included only participants who used the intervention as intended (i.e., those from the intervention group who at least generated a convalescence plan on the website, and all patients in the control group). We used Kaplan-Meier curves to describe differences between study groups in time to return to normal activities after surgery and time to return to work, and we compared these outcomes between groups using Cox regression. For the primary analysis, we included study group as an independent variable and adjusted for covariates used for stratification (i.e., sex, hospital, surgical procedure). Moreover, we did a sensitivity analysis in which baseline factors that might affect the length of the recovery period after surgery (i.e., RI5 score, recovery expectations at baseline, and employment status) were added to the model with study group and stratification factors. 5 We did a final sensitivity analysis in which we defined return to normal activities after surgery as time elapsing between randomisation and return to normal activities. We analysed differences in continuous outcomes between groups over time by mixed models, with time as a categorical independent variable. We included an interaction between group and time in these models, to compare the course of the mean outcome over time between the groups. When the interaction was found not to be significant we removed it from the model, and the main effect for group was considered to establish whether means when averaged over follow-up differed between groups. We did an economic evaluation from a health-care and societal perspective for the primary outcome measure (return to normal activities) and QALYs. 9 Cost differences were—in line with the primary outcome analysis—adjusted for factors for which stratification took place. We used bivariate regression analyses to estimate total cost and effect differences. 26 We used bias-corrected accelerated bootstrapping (5000 replications) to estimate statistical uncertainty. Cost-effectiveness acceptability curves were estimated. A detailed description of methods and results of the cost-effectiveness analysis are reported in the appendix. This trial is registered in the Netherlands National Trial Register, number NTR4699.