The development of the DCE questionnaire was undertaken in sequential steps (Fig.1) based on best practice guidance (19,22–24). The methods and approach for analysis for each stage are summarized in the following sections.
Stage 1: Attribute Development
Attribute development consisted of two distinct steps: (1) attribute identification; (2) attribute selection and framing.
The process of attribute identification started with a review of literature followed by in-depth interviews with pregnant women and decision-makers.
A systematic review of literature conducted by two researchers (HMN, BS) on the use of DCE in the field of prenatal screening for chromosomal anomalies was performed. The search strategy was validated by a librarian. Data extracted on the identification of attributes and levels for preferences of prenatal screening are presented in additional file 1 (full-unpublished report is available on request). The review led to identifying potential attributes that have been found to influence preferences to undertake a screening test for a new condition in a NIPS-based prenatal screening program.
A qualitative study was then undertaken to test the attributes suggested from the literature and identify others that would be regarded by both groups, the group of pregnant women and the group of decision-makers, as important to consider in the decision-making process regarding the content of a public prenatal screening program.
Semi-structured interviews, based on an interview guide (see additional file 2), were conducted with pregnant women (n=12) and decision-makers (n=4). This guide had been beforehand pretested with two pregnant women and one decision maker. These three persons were not included into the sample for this project. They were only asked to comment on the understanding of the interview guide.
Inclusion criteria for pregnant women were to be primigravida, aged 18 years or above, and being consulting at the obstetric department of the CHUL hospital in Quebec City (Canada) for a first prenatal echography. Inclusion criteria for decision-makers were to be either member of a permanent scientific committees of the Quebec's HTA Agency, INESSS (Institut National d'Excellence en Santé et Services Sociaux), or a public servant at the Ministry of Health and Social Service (Québec), and to be involved in decision-making processes on services to be provided to mothers and children.
All interviews were digitally recorded. The verbatim were transcribed by using NVivo Transcription (QRS International 2020). The transcriptions were independently checked by two researchers (HMN, CL) while relistening to the recordings.
The analysis was independently performed by these two researchers (HMN, CL). It aimed at identifying key attributes and their levels. The initial framework used to guide the identification of the attributes was based on the interview guide (deductive approach) (25,26). Additional codes were generated where required (inductive approach) (25,26).
The analysis started with five pre-established dimensions that were expected to come out from the interviews: 1) monthly family expenses associated with a child’s disability; 2) prevalence of a new disease added to the list of diseases searched for by NIPS; 3) performance of the test for this new disease, i.e., probability of identifying a child with a disability; 4) probability that a child tested positive has a severe phenotype; 5) out-of-pocket cost associated with being tested. New dimensions suggested by the interviews were added. Interviews were built in such a way that, at the beginning of the interview, respondents were incited to say whatever they thought was an important characteristic to consider when deciding which condition to screen for, without being interrupted or influenced.
The analysis was first based on triangulation (i.e., categories and themes have to be derived from several sources of information) (27). Moreover, all interviews were coded independently by the two researchers. In case of disagreement, discussions were held until a consensus on the final coding was reached.
Attributes identified by the two groups of respondents were merged. All possible levels arising from the interview were retained. Attributes defined by only one group were also included in the list of attributes/levels for the following selection procedure.
- Attribute selection and framing
Attribute selection followed an iterative process aiming at finding a consensus between both groups of participants, on a set of attributes to be included in the DCE questionnaire. The iterative approach was based on consultations and a focus group discussion. Discussions were held between research team members after each step to refine the list of attributes/levels.
Firstly, the consultation process was undertaken with the same participants who had participated in the attribute identification step and had accepted to be contacted for future study. A list of potential attributes and levels retrieved from the previous codification procedure was presented to the participants. This consultation process aimed to refine the first list of potential attributes and levels. This list was sent to each participant via e-mail, to explore their opinion regarding the dimensions' meaning and relevance. Participants were also asked to provide justification in case they considered one of the attributes to be irrelevant. Attributes that were considered relevant by the participants were thus retained and modified if needed, while those considered irrelevant were excluded.
At the second step, the list of retained attributes was refined based on a focus group discussion conducted with three pregnant women and one decision-maker solicited from the same hospital (focus group discussion with decision-makers alone were not held due to the limited pool of potential participants). They were asked to give their opinion on the relevance of attributes and levels.
Information gained from the attribute-selection process was synthesized by one researcher (MHN). The content of attributes and levels was then revised by the research team members to ensure their relevance and the comprehension of the wording.
Eight key attributes and their levels were identified.
Stage 2: Experimental design and construction of tasks
DCE design and construction of tasks followed the 10-points checklist best practice guidance for conjoint experiment design proposed by the International Society for Pharmacoeconomics and Outcomes Research (ISPOR) (22).
- Construction of choice tasks
An experimental design was performed in order to generate unforced choice questionnaires (28). The questionnaires consist of a series of two options that differ by their level at each of the questionnaire's dimensions (i.e., attributes). The attributes represent characteristics of a hypothetical additional test that could be added to the list of tests already included in a public prenatal screening program. Respondents must choose the preferred option or declare that they cannot decide which of the two options is the preferred one.
The number of possible combinations (so-called alternatives) is determined by the number of attributes and levels. This experiment setup involved eight attributes (five attributes with two levels; three attributes with three levels), for a total of 25 x 33 = 864 screening combinations in the full factorial design (29). Since this number was too high, we built a fractional factorial design using efficient design (instead of orthogonal design due to its limitations of interaction effect estimations), to reduce the sample of scenarios to a manageable level, while still being able to evaluate the effect of each attribute and their 2-way interactions (28,30). The design allowed defining the scenario based on orthogonality (no option dominates another), level balance (all levels of each attribute have an equal frequency), and minimum overlap (there is no overlap in attribute levels) (22,28,30).
- Design of the questionnaire
Generic labels were used to identify the options, called test A and test B across all the choice sets, because they do not reflect exact screening options and there was a possibility that labels might encourage the use of a heuristic approach (31). An opt-out option (no preference for option A or B) was added although there is risk that an opt-out option could lead to high levels of non-response where a trade-off is judged to be difficult (32). Test A and B were constructed in such a way that trade-offs were expected.
Questionnaires are built in such a way that pregnant women are asked which prenatal test they would prefer to have, whereas decision-makers are asked which prenatal test they would prefer to offer to pregnant women.
Stage 3: DCE survey design
Survey components include an introduction to the survey and explanation on its purpose, consent form, an explanation of attributes and levels, an example of DCE choice task followed by the eight/seven DCE choice tasks (for pregnant women and decision-makers, respectively), and questions on demographic data (age, sex (for decision-makers only), income) and on experience of having or knowing a child with disability.
The DCE questionnaires were transferred on Université Laval’s (Canada) LimeSurvey platform (version 3.23.6+200929; https://www.questionnaires.cstip.ulaval.ca/).
Stage 4: Pilot testing the DCE Survey
A pilot project was undertaken to explore the feasibility of the survey that will be administered to a large sample of respondents (i.e., explore the understanding of the tasks, the complexity of the choices, the time needed to fill the questionnaire), as well as the statistical relevance of the dimensions and levels of the first version of the final questionnaire. For practical reasons (a limited pool of potential respondents in the group of decision-makers), we could only test the questionnaires through this pilot project with pregnant women.
The pilot project was also used to elicit a dominant choice that will be added in the full-scale study with pregnant women and decision-makers.
A D-optimal design (33–35) was constructed by judiciously selecting 22 screening combinations, and thus conducted to a pairwise-DCE pilot project (2 by 2 scenarios per choice task) of C(22,2)=231 pairs. This project was undertaken using 33 individuals who were asked to give their opinion on 7 different choice tasks (33*7=231). Detail on sample’s calculation of this pilot project can be found in the additional file 3.
Participants had to be at a gestation age between 28 and 30 weeks. They were recruited among participants to a clinical trial (PEGASUS-2, ClinicalTrials.gov Identifier: NCT03831256) who had accepted to be solicited to participate in an additional study. In this stage, participants were asked to choose the screening option that either suits them the best in each of the tasks or state that they could not express a preference.
- Data collection and analysis
Choices in the DCE choice task were collected automatically by the LimeSurvey platform. Eligible participants received an invitation message containing information regarding the nature of the study, and a link that led them to participate in the study. Once the informed consent form was given by clicking on the “accept” function, participants had access to the DCE survey.
Participants were given two weeks to answer the survey. After two weeks, the link to the non-answered questionnaire was inactivated and sent to new solicited participants until the sample size had been reached.
The preference data were codified on Excel for analysis using a conditional logit model (SAS the LOGISTIC procedure, release 9.4) to estimate the effects of each attribute on the preference of the participants. The significance level was fixed at 0.25% (36).