Data analysis
Ngene software (version 1.2.1) was used to construct a fractional factorial efficient design based on the selected attributes and their levels. In the design, one constraint was taken into account in order to avoid implausible combinations (if surgery removes the endometriosis as much as possible without major bowel surgery, then the chance of permanent intestinal symptoms can only be 10%). Thirty choice sets were created with each choice set consisting of two treatment options, that is conservative and surgical treatment. In order to reduce the burden for the patient, the thirty choice sets were blocked into three versions of 10 choice sets.
To assess if the DCE, levels, attributes, explanation text and baseline questions were understood, we performed a pilot test with a group of pre-surgical patients. Small adjustments were made, mainly in the information text of the attributes and levels. After the DCE was online and open for inclusion, an interim analysis was performed to test our expected direction of effect. The results were in line with our expectations; therefore no adjustments were made. For the whole design of this DCE we followed the checklist of the ISPOR Conjoint Analysis Experimental Design Good Research Practices Task Force (25).
For internal consistency, we included one fixed task, with one dominant option (conservative treatment option) which is more favorable (higher chance of pain reduction and little side effects, compared to a surgical option with maximum side effects and little to no beneficial effects). However, it should be noted that this is not automatically and unambiguously always the ‘best’ treatment because some women can have a strong preference for a conservative treatment or surgery regardless of the specific levels of the attributes.
Optimal sample size calculation for estimating for non-linear discrete choice models is complicated as it depends on the true values of the unknown parameters estimated in the DCE (26). Given the lack of a definite method for calculating a sample size, we based our sample size on a literature review (27). Marshall et al. described that most studies published between 2005-2008 had a sample size of 100-300 respondents. We aimed for 300 respondents because we also wanted to study subgroups.