V(conservative treatment/surgery treatment) =β0+ β1*chance of pain
reduction_med + β2* chance of fatigue + β3* chance of pregnancy +
β4*presence of endometriosis_med+ β5 *chance on depressed mood_med +
β6 * chance on osteoporosis_med+ β7*chance of pain reduction_surg +
β8* presence of endometriosis_surg + β9* chance on temporary
stoma_surg + β10* permanent intestinal symptoms_surg + β2* chance of
fatigue + β3* chance of pregnancy + ε
V represents the relative utility that a respondent derives from
choosing conservative treatment or surgery. β0 is the
alternative specific constant, reflecting a preference for the label
irrespective of the levels of the attributes
β1- β10 are the alternative specific coefficients of each attribute with
the exception of β2 chance of fatigue and β3 chance of pregnancy. The
levels of these attributes were generic across the treatments.
A priori, we expected that patients prefer a higher chance of pain
reduction (positive coefficient), lower levels of fatigue (positive
coefficient for decreasing fatigue, negative for increasing fatigue),
higher pregnancy chance, reduction of endometriosis nodules/spots, low
chance of getting a depressed mood, osteoporosis, temporary stoma and
permanent intestinal symptoms.
ε is an unobserved component of the utility function or error term. Pain
reduction, chance on depressed mood, chance on osteoporosis, chance on
temporary stoma and permanent intestinal symptoms were included as
continuous variables while for fatigue, chance on pregnancy and presence
of endometriosis dummy coding was used. In addition, we performed
subgroup analysis with women with or without a future child wish.
Relative importance was calculated by multiplying the coefficient of an
attribute with the range used for the attribute levels or using the
difference in coefficients between the best and worst level of the same
attribute (in case of dummy coding). Subsequently, the resulting
part-worth utility of each attribute was divided by the sum of all
part-worth utilities which gives the relative importance per attribute
(28). A significance level of 5% was chosen to determine statistically
significant coefficients.