A Discrete Choice Study of Patient Preferences for Dialysis Modalities

DCEs

We used a DCE to quantify patient preferences for characteristics (or attributes) of dialysis modality. DCEs are on the basis of the concept that health care interventions (such as dialysis modality) can be described by a set of attributes or characteristics and that the value an individual places on these attributes when weighing up competing treatment choices depends on the levels of these treatment-specific attributes (16). DCEs are commonly used in health services research to quantify competing health care attributes that might influence patient preferences for treatment to inform health care policy and program development (17,18).

In a DCE, participants are presented with a series of questions describing treatment alternatives and asked to choose their preferred option in each question. Treatment alternatives are described by a set of attributes (for example, out of pocket costs) with varying levels (for example, $0, $50, $100, etc.) (Table 1). Using experimental design principles, the levels of attributes that are presented in each question are varied systematically. Respondents choose the treatment alternative (dialysis option) that they prefer, which is assumed to have the highest value to them. On the basis of the choices made, the relative importance of attributes and their levels can be calculated (19).

DCE methodology is underpinned by random utility theory, consumer theory, and experimental design theory (20). On the basis of the choices within the DCEs, a mathematical function is estimated to numerically describe the value that respondents attach to different attributes (19). Sociodemographic data can also enter into the value functions as explanatory variables. DCEs provide rich data used to quantify the attributes that drive patient preferences, the tradeoffs between attributes that people are willing to accept, and how changes in the levels of these attributes can lead to potential changes in patient preferences (17,18,21).

Participant Selection and Recruitment

The study was conducted in two centers in New Zealand. Patients were considered for participation if they were ages 18 years old or older, were anticipated to require RRT within a year (predialysis), and had received formal education about options for RRT. Patients were eligible to participate in the study if they were deemed clinically suitable for either facility or home dialysis in the opinion of their treating clinical team and able to read English. Patients were invited to participate by a nephrologist, CKD coordinator, or predialysis nurse specialist from a list of all eligible patients. The study was approved by relevant hospital ethics committees (Counties Manukau Institutional Review Board [reference no. 1771] and Hawke’s Bay Institutional Review Board [reference no. 14/06/160]). All respondents provided written informed consent.

Survey Design

The survey was presented to participants as follows: (1) introduction and explanation of the study, (2) description and explanation of attributes, and (3) an example choice set. Respondents were shown 12 choice scenarios and asked to choose the treatment that they most preferred in each scenario (Figure 1). Sociodemographic data were also collected, including self-identified ethnicity, number of people living in household, educational attainment, and employment.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1.

Choice set example presented to patients. Example of choice set presented to each patient before dialysis in the discrete choice survey. HD, hemodialysis; PD, peritoneal dialysis.

We piloted the initial survey design in face to face consultation with 13 patients to ensure an adequate understanding and interpretation of the attributes and levels. We also asked for feedback on the length and number of scenarios in the survey and the demographics questionnaire. Respondent feedback was used to inform the final efficient design, which was generated in NGENE software (www.choice-metrics.com). The final design had a d error of 0.12 and consisted of 12 choice sets of three alternative dialysis modalities (home hemodialysis [HD], peritoneal dialysis [PD], and facility HD) (22).

Data Collection

Patient recruitment and data collection were performed between November of 2015 and July of 2016. After verbal consent was obtained, participants were given the survey by either the investigator (R.C.W.) or the predialysis nurse, who explained how to complete survey and showed this with the example scenario. The respondents then either completed the survey at that time or took it home to complete and returned the completed survey via mail. There were no missing data in the completed surveys.

Attributes

We presented patients with three dialysis modalities in each question: home HD, PD, and in-center HD (Figure 1). The attributes and levels used to describe each dialysis alternative were on the basis of our prior research, including qualitative interviews with patients (13,23) and thematic syntheses of patient and caregiver values, beliefs, and experiences when considering home dialysis (8,9). Levels for patient survival were obtained from the Australia New Zealand Dialysis and Transplant Registry (24) to include a range of plausible values. Dialysis alternatives were described using the following attributes: flexibility of dialysis schedule (easy to change, sometimes possible to change, and hard to change), out of pocket costs to patient, availability of transport services (or reimbursement of travel costs), life expectancy, dialysis training time, levels of nursing support, wellbeing on dialysis, and dialysis schedule (time of day, frequency, and duration) (Table 1).

Data Analyses

We used NLOGIT, version 5.0 (Econometric Software, Castle Hill, NSW, Australia; www.limdep.com/products/nlogit/) to analyze the data. Internal validity (that is, the extent to which results were consistent with the researchers’ prior expectations) was assessed by examining the signs and significance of parameter estimates.

We hypothesized a priori that respondents would prefer home dialysis when out of pocket costs were lower, transportation was not reimbursed, patients felt better on dialysis, and the dialysis schedule had greater flexibility. We built more complex models to increase the goodness of fit and enable greater exploration of observed heterogeneity. Different attribute-level specifications were examined, including using different distributions and coding as continuous or categorical (effects coded) variables and as random and nonrandom parameters. We also examined the influence of sociodemographic characteristics as interactions with both attributes and the constant.

The final model was a mixed multinomial logit model within a panel specification (also termed a random parameters logit model). The final model specification was determined on the basis of the P value of specified parameters, log likelihood tests, McFadden pseudo-R², and Akaike Information Criteria. A panel specification was used to account for correlated choices within an individual, because each respondent completed 12 choice tasks. In the final model, cost, life expectancy, flexibility, wellness on dialysis, and time of treatment were treated as continuous linear variables; all other categorical attributes were effects coded. We included an interaction between distance from the dialysis unit (<100 or >100 km) and availability of transport. The attributes were specified as random with normal distributions, with 5000 Halton draws. Some attributes were specified with generic levels across home HD and PD, whereas others had levels that were specific to either home HD or PD. Interactions between treatment characteristics (attributes) and other sociodemographic variables that did not improve the model fit were not included in the final model. We also examined the influence of sociodemographic characteristics (age, education, ethnicity, home ownership, household size, distance from dialysis unit, preference for access type, income, and employment) as interactions with the alternative specific constants for home HD and PD. Model results were expressed as β-parameters and the odds ratios (ORs; 95% confidence intervals [95% CIs]) of choosing home HD or PD compared with facility HD. Marginal rates of substitution (tradeoffs) between the out of pocket cost attribute and other attributes were also examined.