To investigate men preferences, we performed a discrete choice experiment (DCE), an econometric method increasingly used in health economics (15).
a. Discrete choice experiment
DCEs allow fictive screening program characteristics to be ranked according to their relative importance in the decision. The method is based on Lancaster’s consumer theory (16), which stipulates that a program or an intervention in healthcare can be described by its main characteristics, called attributes, and their relative levels. In a DCE, the respondent states which alternative he/she prefers among the fictive scenarios. These scenarios are composed of several attributes (e.g. efficacy of the test, out-of-pocket costs, etc.) and differ according to several levels of attribute. Preferences are extracted from the respondent’s stated choices.
b. Identification and selection of attributes and levels
To implement a stated DCE, attributes and levels are selected and fictive scenarios are created. Attributes and levels were chosen to test the effect of the benefit-risk ratio on prostate cancer screening choices. We performed PubMed and Econlit searches in May 2018 to identify attributes and corresponding levels using keywords: “discrete choice” and “cancer screening”. Four DCE were found to explore preferences with regard to prostate cancer screening with discrete choice analysis (17–20). Attributes used in these studies were related to death from prostate cancer (17–20), recommended screening frequency (17), number of biopsies (19) and PSA false positive results (17,18), number or percentage of prostate cancers diagnosed (18–20), risk of overdiagnosis (20), risk of overtreatment (17), treatment side-effects (impotence and incontinence) (18–20) and out-of-pocket costs (17,18,20). False negative results were not introduced into these choice models. We interviewed 5 experts (i.e. epidemiologist, ethicist, health economist and physicians) to select and formulate the attributes and levels. They based their choice on key objective elements (e.g. available care strategies, benefits and risks of each procedure), which should be provided by GPs during a consultation. Finally, six attributes were selected: risk of mortality by prostate cancer, risk of false positive results, risk of false negative results, risk of overdiagnosis, recommended screening test frequency and out-of-pocket costs.
Out-of-pocket costs for the patient only concern medical expenses related to a cancer screening procedure (i.e. GP consultation, and PSA blood assay). In France, since 2017, a routine consultation GP is charged € 23,00 of which € 16,50 are reimbursed by the health insurance system and up to € 5,50 by the patient’s private insurance policy. A PSA assay costs € 10,80 of which 60 % is reimbursed by the health insurance system and up to € 3,32 by private insurance. Levels of out-of-pocket cost attribute vary according to these rates and to the various fictive reimbursement rates applied by the social security system, i.e. from no reimbursement at all to complete reimbursement of medical expenses.
Since there is no national prostate cancer screening program in France, the frequency of the PSA assay and the rectal examination depends on the GP. Levels associated with the recommended frequency attribute were based on the frequency tested during the main surveys and on GPs’ prescription habits.
Levels of the four risk attributes were extracted from the major clinical trials on prostate cancer screening (ERSPC (21), PLCO and CAP (4)) (21–25). Based on recent progress in risk communication (26), the wording of attribute levels based on risks was established with the same indicator (i.e. per 1,000 persons regularly screened). Table 1 gives an overview of the attributes and levels used in this study. Respondents could obtain more details about the attribute definitions (i.e. false positive rate, false negative rate, overdiagnosis, out-of-pocket costs) by clicking on the attribute’s label.
Table 1.
Given the selection of attributes and levels, 972 combinations (41*35) were available for this survey. To reduce the cognitive effort caused by too many tasks per respondent, an experimental design was created to obtain 14 scenarios by using the OPTEX procedure in the SAS software (version 9.4) . In a second time of the procedure, scenarios were paired to obtain 7 choice situations with 2 screening alternatives. This fractional nearly orthogonal design maximized the D-efficient score (90.0788). Alternative scenarios extracted from this design were randomly distributed between two fictive screening options by applying a blocking strategy (28). At the end, a total of 7 tasks per respondent was obtained.
Figure 1
c. Study design and questionnaire
Before completing the questionnaire, half of the respondents had access to a 6-minute video on prostate cancer screening produced by our research team. A simple randomization was performed to determine this access. Several patients, a urologist and GPs watched a previous version of the video and suggested changes to improve its clarity and neutrality and to limit its cognitive demand. The video is available in the supplementary files. The video started with information on prostate anatomy and physiology. Then, key epidemiological data on prostate cancer were illustrated with diagrams. Next, the screening procedure was presented. Its benefit and risks were graphically represented with two icon arrays (consequences for 1.000 men with and without screening) as it is used by The Canadian Task Force on Preventive Health Care (29), for example. This format is recommended for communicating about risks and benefits, especially among men with a low level of numeracy (30). At the end, the official French guidelines were explained to the participants.
Either directly or after the video, each participant received instructions on the stated preference experiment (i.e. context of prostate cancer screening, background to DCE). Respondents had to express their intentions regarding hypothetical and fictive screening programs. In this experiment, the participants chose one of two prostate cancer screening programs. Since it is irrelevant to force respondents to choose between screening programs potentially considered as unimportant (31), choice sets included two fictive prostate cancer screening programs and one opt-out option (i.e. “do not undergo a prostate cancer screening test”). An opt-out option is an alternative whose attribute levels do not change according to the choice situations. Figure 1 is an example of a choice situation proposed to respondents. A within-set dominated-pairs test was added to test the rationality of DCE responses (32,33). In this choice situation, a dominated screening alternative was less interesting for each attribute level. Respondent characteristics likely to influence choice of cancer screening adherence were also collected (e.g. age, prostate cancer screening experience, highest level of education). The last question concerned difficulty in completing the questionnaire (from easy to hard).
A pilot study in 50 respondents tested the relevance of the attributes, the level of comprehension and the feasibility of the full questionnaire. A few changes were made to the introductory section after this phase. The mean survey duration was 17 minutes, including viewing.
a. Study sample
A survey institute oversaw the recruitment process. They performed a sampling approach with the quota method to be representative of the male population aged from 50 to 75 years old and without any prostate cancer diagnosis. For this purpose, criteria used were age, French regions, type of urban agglomeration, and socio-professional categories. In January 2019, the survey institute used e-mail (16,064 emails sent) to contact potential respondents from a French panel. Among the recipients, 2.703 men clicked on the study link. If respondents agreed with the terms, they could complete the online survey. Finally, a total of 1.024 respondents completed the entire questionnaire.
Figure 2
Two tests were included to evaluate choice rationality. The dominant alternative of the within-set dominated pairs test was chosen by 62.21 % of respondents. 170 men failed the rationality test and/or systematically selected the same screening alternative, whatever the screening scenario content. They were excluded from the analysis. Finally, statistical analysis was performed on data from 854 participants. Among them, 427 respondents had to watch the video before completing the questionnaire.
In accordance with French law, ethical committee (CLERS: Comité Local d’Ethique de la Recherche) and CNIL (Commission Nationale de l’Informatique et des Libertés) approval was obtained before the survey began.
b. Statistical analyses
Based on the maximization of utility principle, the relative importance of the choice components could be estimated through alternative utility functions. In these utility functions, utility is explained by a measurable part composed of attributes. All attributes were included in a logistic model in the SAS software (version 9.4) as continuous variables. The main effect model for an individual n and a choice alternative j is presented below:
Unj=β0+ASCopt-out+β1x DRnj + β2 x FPnj + β3 x FNnj + β4 x ODnj + β5 x COnj + β6 x FRnj + Ɛnj
Where β0 is the Alternative Specific Constant (ASC) representing choice parameters unmeasured, ASC opt-out is another alternative specific constant which is equal to 1 if the no-screening option is chosen, 0 otherwise. DR n j, FP nj, FN nj, OD nj, CO nj, FR nj are vectors of the attributes mortality by prostate cancer, false positive result rate, false negative result rate, overdiagnosis rate, out-of-pocket costs and recommended frequency of screening, β1, β2, β3, β4, β5, and β6 their vector of parameters, and Ɛnj represents the random and unobservable part. We assumed that the latter component was independently and identically distributed (i.i.d.).
A ranking of attribute importance in men’s choices is then available with the sign and the magnitude of each coefficient. A priori expectations had a negative impact on alternative utility for all attributes.
Willingness to pay
Marginal Willingness-To-Pay (MWTP) was then calculated from out-of-pocket costs and risk attributes. For example, MWTP represents how much men were willing-to-pay in order for an additional man not to succumb to prostate cancer per 1,000 men screened . Confidence intervals of these estimations were estimated by using the delta method (34), which stipulated that the confidence interval of WTP
[Please see supplementary files section to access the equation.]
Interactions with individual characteristics, anxiety and video
Various specifications of the model were tested by incorporation different interaction components like socio-demographic data. Health anxiety level was broken down into three levels (i.e. low, medium and high) according to terciles. A high level of anxiety was hypothesizedto reinforce the negative estimation of mortality by prostate cancer, false negative, false positive and overdiagnosis attributes. Men with a high level of anxiety were also hypothesized to increase the value of screening.
Video access was also added as an interaction term to test our hypothesis that an informative video could modify choice preferences. The video was hypothesized to improve understanding of the benefits and risks of screening and thus to reinforce the negative effect of mortality by prostate cancer, false positive rate, false negative rate and overdiagnosis. It was also hypothesized to reduce the positive perception of screening by representing the benefit-risk ratio of prostate cancer screening or the statement by the French health authorities.