Supplementary statistical methods
In study II, we validated the results with data from the LAFA-trial database. The sample size of the LAFA trial was calculated based on hospital stay as the primary efficacy parameter. Using a 5% significance level, a total sample size of 400 had a power of > 95% to detect a minimum reduction in total hospital stay of 1 day between laparoscopic and open surgery, 1 day reduction in total hospital stay between fast track and standard care, and a power of 80% to detect the same difference between the combination of fast track with laparoscopic surgery and open surgery with standard care.[14] The Kruskal-Wallis test followed by Dunn’s Multiple Comparison test was used to compare the 4 groups of study II.
Postoperative hospital stay was calculated from the day of surgery until the day of discharge.
As postoperative hospital stay was not normally distributed, these data were log-transformed.
The covariates intake of solid food day 3, absence nausea on day 3, flatus before day 4 and defecation before day 4, and 7 baseline characteristics (sex, age (≤65 years), American Society of Anesthesiologists (ASA) grade (I/II), body mass index (BMI), type of perioperative care program (standard, fast track care), type of surgery (open, laparoscopic), and type of resection (right-sided, left-sided)) were tested by univariate linear regression analysis. All variables with P<0.100 were then entered in a sequential multiple linear regression analysis. The variables were: intake solid food day 3, absence nausea on day 3, flatus on/before day 3, passing stool on/before day 3, sex, age, ASA grade, type of perioperative care program, and type of surgery. Forward sequential elimination was used to create a final multivariate model retaining only variables with P<0.05, as this was considered to be statistically significant. A new variable was created for those patients that had achieved tolerance of solid food (i.e. intake solid food without nausea) and passed stool before day 4.
This new variable, named ‘passing stool and tolerance solid food on day 3’ and the predictive baseline characteristics (age, type of perioperative care surgery and type of surgery) were
entered in a new hierarchical multivariable linear regression model. B-values of significant predictive parameters were converted into percentages difference in postoperative hospital stay they would result in if present, with their 95% confidence intervals (CI).