Discrete choice experiment

Preferences of patients with PD for potential cell-based therapies to treat PD were assessed by a Discrete Choice Experiment (DCE) in Swedish patients with PD. The DCE is a cross-sectional survey method to investigate individuals preferences and can be used to determine the relative importance of different characteristics of an intervention and predict uptake of different interventions [15]. Respondents of a DCE are faced with a set of hypothetical choice questions with two or more alternatives, characterized by different characteristics (i.e., attributes) with varying levels. The DCE method also allows for the calculation of attribute trade-offs [16].

We performed a scoping literature review to identify attributes of treatments for PD that potentially were of importance for patients with PD when choosing treatment. Qualitative and quantitative papers investigating preferences of patients with PD related to treatment for PD were included. All literature searches were performed in PubMed and the keywords used were Parkinson disease, patient preferences, preferences, treatment, medication, and attributes. We identified 193 papers, including 29 papers that were relevant for this project, of which 20 papers remained after excluding duplicates. After reading the full text papers, 209 potential attributes were identified. Out of the 209 attributes identified in the scoping literature review, 115 attributes were unique. These attributes were condensed down to 45 by merging similar concepts. The identified attributes were discussed in a group consisting of a representative patient of a Parkinson patient organization, neurologists, a research coordinator, a nurse working with patients with PD, and researchers knowledgeable in DCE methodology. Based on the discussions in this group, 11 attributes remained. We let 17 patients with PD rank the 11 attributes from most to least important, for their decision about PD treatment. Based on the mean ranks of the attributes and discussions with clinicians, eight attributes remained. These were re-categorized into the five attributes that were assigned relevant levels to be assessed by the DCE: (i) type of treatment, (ii) aim of treatment, (iii) available knowledge of the different types of treatments, (iv) effect on symptoms, and (v) risk for severe side effects (Table1).

We followed methodological guidelines to estimate the sample size needed to identify preferences of patients with PD and differences within those preferences [17]. We considered the number of attributes in the DCE (Table1) and the number of choice questions for each respondent (n=9). Based on the sample size requirements for a DCE and accounting for subgroup analysis, we aimed for a sample size of 500 respondents.

Patients with PD were recruited from neurology clinics at two university hospitals in Sweden. This study was approved by the Swedish Ethical Review Authority (Dnr 201906539). Information about the study was sent out by mail to all potential respondents fulfilling the inclusion criteria: patients diagnosed with PD, 18 years or older, able to read and understand Swedish. Patients with a known dementia diagnosis were excluded. Information about the study was sent out to 1266 patients. Patients who had not responded within two weeks were sent a reminder by mail. All respondents provided their informed consent before entering the survey. Two patients formally declined participation, and five patients were unable to participate due to technical or health-related restrictions. In total, 498 patients participated in the study (i.e., 39% response rate).

This survey was administered as a web-based survey that included three parts: (i) information about the attributes and levels, (ii) the DCE with hypothetical choice scenarios, and (iii) demographic and attitude questions (see supplementary file for survey). The survey was created for this study and administered using Sawtooth Software (Sawtooth Software Inc.). Each respondent was faced with nine hypothetical choice scenarios that each included three alternatives. The respondents were asked to select the alternative that they most preferred out of the three presented to them. The first two alternatives were experimentally designed to assess preferences for potential treatment alternatives for PD and the third was a fixed profile (i.e., nonexperimental) to represented standard care (drugs) for patients with PD (Fig.1). We used a Bayesian D-efficient design to construct the choice scenarios for the DCE using the NGene program (version1.2.1; ChoiceMetrics 2012). Prior information on the attribute importance was gathered from a pilot test (n=142) in patients with PD. The design used 500 Halton draws and 1000 repetitions. Using the pilot data, a multinomial logit (MNL) model was fitted, and the beta estimates was used as priors for the final experimental DCE design.

Some conditions were posted on the design: if the aim of treatment was to repair damage caused by disease, the treatment could not consist of electric stimulation or drug. If the aim of the treatment was to slow down disease progression, the treatment could not consist of electric stimulation. The final discrete choice survey consisted of 36 unique choice scenarios divided into four blocks; each respondent was randomized into one block and answered nine choice scenarios. The choice questions also included a hover function with further explanations of the attributes and the levels (see Table1 for full description of the attribute levels).

Example of a choice scenario

The demographical and attitude questions included background questions (e.g., age, gender, and education) and disease-related questions (e.g., disease duration, treatment, and side effects). Moreover, the respondents attitudes were gauged with a ranking exercise with eight statements that they were asked to place in the order they found most important. The attitude questions asked respondents about their moral stands on the status of an embryo, and a ranking exercise to prioritize eight statements.

The respondents were asked about their views on how to regard the products left over after IVF procedures, which may be used for hESC isolation, that is, the blastocyst. Whether this material was regarded as a lump of cells or something more was used to dichotomize the answers. Questions to assess respondents health literacy [18] and health numeracy [19] were also included to define the sample.

The statistical analyses, in particular the estimation of the latent class model were performed using R 4.0.2 (R Core Team, 2018), the mlogit (version 1.1-1; Yves Croissant, 2009) and the gmnl (version 1.13.3; Mauricio Sarrias, 2017) [20].

Demographics describing the populations age, gender, country of birth, occupational situation, education, health numeracy, health literacy, drug frequency, disease duration, number of experienced side effects, and experience of advanced treatment were presented in mean, median, and percentages. The overall level of health literacy and numeracy was calculated for each respondent. Individuals who responded strongly disagree or disagree to one of the items were categorized as having low health literacy. Individuals who responded with neither agree nor disagree with one of the items were categorized as having medium health literacy. Individuals responding agree or strongly agree to all the items were categorized as having high health literacy, and likewise for numeracy.

Respondents attitude toward the moral status of a couple of days old human embryo was assessed using this question: The human is perceived to have a special moral position, in the sense of having rights just by being human. What moral position does a human embryo that is only a few days old have? The respondents had four statements from which to select: (1) The embryo is just a lump of cells; it is meaningless to discuss its moral status, (2) The embryo has a moral status that is in between being just a lump of cells and being a human being, (3) The embryo in its moral status is closer to being a human than just a lump of cells, and (4) The embryo has the same moral status as a human being. The variable was dichotomized based on the frequency of the data. Respondents answering The embryo is just a lump of cells; it is meaningless to discuss its moral status formed one group, and the rest another group. One-way analysis of variance and nonparametric measures were used to test the differences between the personal characteristics and the different perceptions of whether an embryo is more than a lump or cells or not.

The most important attitudinal statement was given a 1, the second most important the number 2 and so forth. The ranking exercise was illustrated with a boxplot by the median value of each statement, stratified on the different perceptions of whether an embryo is more than a lump of cells.

The latent class analysis was based on the a priori hypothesis that the authors thought would be associated with the willingness to accept a new treatment. Five variables were tested for class membership: (1) a summary of experience of different treatment, (2) experience of the summary of different side effects, (3) the perception of the moral status of the embryo, (4) experience of advanced treatment, and (5) the importance of religion. A sum of how many treatments each respondent had was calculated, and also how many side effects they had experienced. Advanced treatment was based on treatment experience with one or more of apomorphine subcutaneous injection, apomorphine subcutaneous infusion, deep brain stimulation, levodopa-carbidopa intestinal infusion, and levodopa-entacapone-carbidopa intestinal infusion. The variable the perception of the moral status of the embryo did not influence class membership and was therefore not included in the final class assignment model.

The statistical analyses of the preference data were based on a latent class model. A preference weight (i.e., coefficient) and a corresponding SE were estimated for all but one level of each attribute (i.e., reference attribute level) [21]. Dummy coding of the variables was selected for this analysis (i.e., corresponding to zero as the reference value). Each p-value is a measure of the statistical significance of the difference between the estimated preference weights for each level of the attribute compared to the reference attribute level. All results were considered statistically significant at p<0.05. Confidence intervals (95%) were also provided for each preference weight. The Akaike information criterion (AIC) and the log-likelihood values were considered when selecting the appropriate model.

The latent class model was used to identify hidden (latent) classes of respondents preferences [22]. In latent class analysis, unobserved preference heterogeneity among respondents preferences is modeled as classes with similar preference patterns but with different variances across classes. Once preference patterns have been stratified into classes, the model determines the extent to which demographic characteristics impact the likelihood of belonging to a certain class. The systematic utility component (V) describes the latent construct that participant r belonging to class c reported for alternative A, B or C in choice task t. The final utility functions were as follows:

Vr,t,A&B|c=1 * consist_hESCr,t,A&B|c+2 * consist_iPSr,t,A&B|c+3 * consist_electricr,t,A&B|c+4 * aim_slowr,t,A&B|c+5 * aim_repairer,t,A&B|c+6 * know_500r,t,A&B|c+7 * know_5000r,t,A&B|c+8 * effect_50r,t,A&B|c+9 * effect_80r,t,A&B|c+10 * sideeffects_0.001r,t,A&B|c+11 * sideeffects_0.01r,t,A&B|c+.

Vr,t,C|c=1 * consist_drugr,t,C|c+2 ** aim_reliefr,t,C|c+3 * know_5000r,t,C|c+4 * effect_50r,t,C|c+5 * sideeffects_0.01r,t,C|c+.

A class assignment model was fitted after the specified utility function. The variables: experience in treatment, side effects, advanced treatment therapy and religious beliefs were tested for their potential impact on class membership in the model. The final class assignment function was:

Vn|c=0+1* treatment_sum|c+2 * experience_sideeffects|c+3 * advanced_treatment|c+4 * Religion_dum|c+.

The relative importance of the attributes included in the DCE was calculated by estimating the difference in preference weights of the latent class model between the most preferred level of an attribute and the least preferred level of the same attribute [21]. The highest difference value was normalized to 1, which represents the most important value. The difference values were divided by the highest value to reveal the relative distance between all other attributes.

We calculated the predicted acceptance uptake for a potential treatment scenario using hESCs to treat patients with PD. Predicted acceptability can be understood as the probability that a participant will accept a described scenario. The scenario represents a hypothetical treatment scenario of treatments with hESCs based on the attributes assessment in the DCE. Attribute estimates assessed by the latent class model were used to calculate the predicted acceptability of attribute levels (treatment with hESCs, risk of severe side effects is 1 out of 1000 and 50 patients received treatment) in relevant future scenarios; (A) effect on symptoms is 2 out of 10, (B) effect on symptoms is 5 out of 10, and (C) effect on symptoms is 8 out of 10.

The predicted acceptability is presented as the percentage of 100 who would accept the presented scenario. The utility for the specific scenario was calculated by using the following equation:

VScenario 1=A+B+C.

The predicted acceptability, the probability of accepting a specific scenario, was then calculated by using the following equation:

Predicted acceptance uptake=1/(1+expVScenario 1).

Read more:
Patients accept therapy using embryonic stem cells for Parkinson's ... - BMC Medical Ethics