Citizen Participation in Patient Prioritization Policy Decisions: An Empirical and Experimental Study on Patients' Characteristics

Adele Diederich; Joffre Swait; Norman Wirsik

doi:10.1371/journal.pone.0036824

. 2012 May 9;7(5):e36824. doi: 10.1371/journal.pone.0036824

Citizen Participation in Patient Prioritization Policy Decisions: An Empirical and Experimental Study on Patients' Characteristics

Adele Diederich ^1,^*, Joffre Swait ², Norman Wirsik ³

Editor: David W Dowdy⁴

PMCID: PMC3348901 PMID: 22590619

Abstract

Health systems worldwide are grappling with the need to control costs to maintain system viability. With the combination of worsening economic conditions, an aging population and reductions in tax revenues, the pressures to make structural changes are expected to continue growing. Common cost control mechanisms, e.g. curtailment of patient access and treatment prioritization, are likely to be adversely viewed by citizens. It seems therefore wise to include them in the decision making processes that lead up to policy changes. In the context of a multilevel iterative mixed-method design a quantitative survey representative of the German population (N = 2031) was conducted to probe the acceptance of priority setting in medicine and to explore the practicability of direct public involvement. Here we focus on preferences for patients' characteristics (medical aspects, lifestyle and socio-economic status) as possible criteria for prioritizing medical services. A questionnaire with closed response options was fielded to gain insight into attitudes toward broad prioritization criteria of patient groups. Furthermore, a discrete choice experiment was used as a rigorous approach to investigate citizens' preferences toward specific criteria level in context of other criteria. Both the questionnaire and the discrete choice experiment were performed with the same sample. The citizens' own health and social situation are included as explanatory variables. Data were evaluated using corresponding analysis, contingency analysis, logistic regression and a multinomial exploded logit model. The results show that some medical criteria are highly accepted for prioritizing patients whereas socio-economic criteria are rejected.

Introduction

The need for reform of healthcare provision in first-world countries is widely recognized, both in public and privately funded systems [1]. Aging populations, shifting demographics and social values, increasing costs and reductions in tax revenues are working together to deeply stress healthcare systems in Australia, Canada, France, Germany, Great Britain and the United States, among others. While specifics differ from country to country, controlling or reducing the cost of health care delivery takes a central role in the debate about reform in all of these places [2], [3].

With this background motivation, this paper addresses empirically the issue of prioritization of patient care, a necessary step to establishing service priorities, designing cost containment policies and imposing user fees. Most of the parties involved in healthcare reform debates – governments, politicians, healthcare professionals, pharmaceutical companies, special interest groups – actively work to make their desires known. Despite their obvious interest in this debate, however, it is the patients who will likely have the greatest difficulty in providing input to these discussions.

A common approach taken to policy formulation in the face of resource constraints is to adopt an utilitarian framework that seeks maximization of societal health benefits through reliance on the cost-effectiveness of health services (see discussion in [4]). However, as argued in [4], and supported by the health service prioritization exercise for the state of Oregon in the early 1990's, cost-effectiveness as a criterion does not seem to generate socially and politically palatable solutions due to the Rule of Rescue. This rule dictates that people cannot remain unresponsive or inactive when a specific, identified person's life is in peril and there exist effective means of “rescue” or aid. Hadorn [4] discusses in some detail one approach and experience to addressing this basic response in humans to the needs of other humans. We seek to introduce citizen participation into health prioritization dialogues as a complementary method that directly incorporates phenomena such as the Rule of Rescue and other psychological, emotional and social responses.

One controversial set of criteria for prioritizing health care concerns patients' personal characteristics. Whereas medical features such as severity of illness are generally supported as valid criteria for priority setting [5]–[7], it is less clear whether personal characteristics such as life-style and self-infliction of disease [8]–[10] or age [9], [11]–[13] are acceptable for the purpose. Research results have been inconsistent and seem to depend partly on the study design.

In this research we employ two different approaches used on the same representative sample of German citizens (N = 2031). First, we elicit the acceptance of possible prioritization criteria (medical, socioeconomic and lifestyle situation of the patient) via questionnaire items with binary responses to probe citizens' acceptance of broad and general criteria in a rather abstract way. Second, we investigate one rigorous mechanism, stated preference experiments [14] – also called discrete choice experiments (DCEs) – for injecting a “voice of the patient” (related to the long-standing “voice of the customer” concept in the marketing literature; see, e.g., [15]) into the healthcare reform debate. Here the criteria are more specific and a tradeoff between several criteria level is required to set a patient at the head of the priority list.

We illustrate the approach through a representative survey of German citizens, eliciting rankings of hypothetical patients (described by illness type and severity, demographics and lifestyle characteristics) in terms of service prioritization – who should be served first, second, etc. On the basis of this data, we relate reported patient service rankings to specified patient descriptions via a multinomial “exploded logit” [16] discrete choice model. This model is the mechanism whereby decision makers can “consult” the German population in terms of this specific issue when dealing with policy issues related to service prioritization. (See [17] for a similar focus with respect to public participation in transportation policy formulation.)

Discrete choice experiments have experienced increased acceptance in the health care research literature. We cite recent work to exemplify the penetration of DCEs in health research. Burge, Netten and Gallo [18], for example, use a DCE to estimate willingness-to-accept (WTA) valuations of a population for a large number of social care outcome domains. Johnson et al. [19] investigate the role of what they term a “recoding heuristic” to characterize how health care costs may be treated by subjects in DCEs due to distortions introduced by insurance coverage. Lancsar et al. [20] employ a DCE to derive distributional weights for quality adjusted life years (QALYs), based on characteristics of beneficiaries (see also [21]). To compare patient and physician perception of patient preferences concerning multiple myeloma therapy, Mühlbacher and Nübling [22] employ separate DCEs applied to the two respondent types; they find that there is broad agreement in preferences between the two groups. Green and Gerard [23] used DCEs for the process of health technology appraisal. As a final example, Scuffham et al. [24] proposed the use of DCEs as an aid to policy makers in designing health care system characteristics (e.g. health, equity, responsiveness and financing). Thus, the health care system literature has DCE applications with a variety of objectives: health system design, patient decision rules, comparative preferences between physicians and patients for a specific treatment, and support of economic valuations.

The DCE reported in this research is based on the same general approach employed in the above literature, but differs from them in that population tastes for preferential treatment decisions are elicited. In this sense, the purpose of this DCE is not to characterize the patient as the receiver of treatment in a health care system, but rather to permit the population's direct representation in the process of redesigning cost-related policies for their health care system.

In the remainder of the paper we discuss the survey structure and sampling methods that were employed, present an overview of the model estimation results, discuss the inferences that arise from the survey and model concerning German citizen preferences about service prioritization, then conclude with a discussion about the usefulness of methods such as stated preference experiments as mechanisms for bringing citizen preferences into public policy discussions.

Materials and Methods

The reported methods and results are part of a more comprehensive study on prioritizing in medicine using a multilevel iterative mixed-method design (for details see, e.g., [25], [26]) for combining a qualitative interview study, a quantitative survey representative of the German public and focus groups. The quantitative survey included a DCE that will be described in due course. Approval for this study was granted by the Ethics Board of the University of Bayreuth (Ethik-Kommission für Forschungsfragen der Universität Bayreuth), 95440 Bayreuth, Germany.

Sampling

The population survey was conducted in Germany by TNS Healthcare between July and September 2009, covering people aged 18 and over living in private households. Data were collected by computer assisted personal interviews (CAPI). The sampling followed a three-stage random route procedure, with a design developed by ADM (Association of German Market and Social Researchers). The first stage comprises electoral wards for national elections, the second the households, and the third the individuals within the target households selected by the Kish-table method (see, e.g., [27], [28], for details). Participants gave a verbal informed consent (i.e., agreed to participate) after they had been informed about the goals and content of the study, as well as about data protection and privacy. Participants' co-operation in this research project was entirely voluntary at all stages.

Material for the Questionnaire

A survey with 34 questions comprising 135 response items was organized around health care and health system related themes [28]. Both the topics addressed in the questionnaire and the attribute dimensions used for the discrete choice scenarios are based on results obtained from an exploratory interview study on prioritizing health care with 45 members of six different stakeholder groups [29]. One of the themes of the questionnaire was concerned with person-specific characteristics as possible criteria for preferential treatment.

The setting for this portion of the qualitative interview was established by the following statement: “We would like to know whether a specific patient or specific patient groups should receive preferential treatment if medical services are not provided for by the public health insurance to the extent they used to be.” If respondents asked for further clarification, the interviewer stated that preferential treatment meant that this specific person would be treated first, but did not mean that other patients were not to be treated at all. That is, patients might receive treatment later or with fewer resources.

A first block of questions described patients in an abstract fashion with only one person characteristic. Responses to the question “Do you think it is justifiable to treat the following patient groups in preference to all others?” were provided for 18 different groups: 1) people who are active in the community (e.g. volunteer workers); 2) patients with a life-threatening disease/illness; 3) senior citizens; 4) patients with psychological illness; 5) people with a healthy lifestyle; 6) people with high income; 7) patients with chronic illness; 8) children; 9) patients with a low quality of life; 10) people with high level of professional responsibility (e.g. people in an executive position supervising several employees); 11) patients with physical handicaps; 12) patients with acute diseases; 13) socially disadvantaged people; 14) people with children; 15) people of working age; 16) patients with mental handicaps; 17) unemployed people; 18) people with social responsibilities (e.g. caring of relatives). The response categories included a “Yes” and “No” response; “Don't know” and “Response refused” options were offered only when the person did not respond with one of the first two categories.

In addition to questionnaire items, two different discrete choice experiments [14] were presented. One experiment described patients with several person characteristics (specified below), whereas the second dealt with other new treatments not considered in this paper. The questionnaire ended with socio-demographic questions and a self-report on the respondent's life-style and on health, the latter measured by the Short-Form Health Survey (SF-8™) [30].

Material for the Discrete Choice Experiment

The discrete choice experiment requires construction of choice alternatives (called “profiles”) which are characterized by several attributes and different levels (or values) on these attributes. In the current context the choice alternatives are hypothetical patients.

The following attributes and their levels were used to create the profiles (i.e. person-condition descriptions) for the discrete choice experiment (DCE).

Health Status (levels: severe disease, light disease)

This attribute summarizes a large number of diseases and injuries. Purposely no concrete examples were given to avoid personal conflict. We relied on a general agreement among all participants that all (acute) life threatening diseases, emergencies and diseases that generally lead to death (e.g. cancer or heart failure) are severe diseases, whereas non-life threatening and most chronic diseases are considered light illnesses.

Quality of life (levels: severely restricted, restricted, no restrictions)

The categories followed the EuroQol classification [31]: a strongly restricted person is unable to perform usual activities and has extreme pain and discomfort; a restricted person has some problems in usual activities and moderate pain and discomfort; and a person with no restrictions is in full health with no restrictions in usual activities.

Unhealthy life style (levels: yes, no)

The attribute is intended to address the person's individual responsibility to care for his/her health. An unhealthy lifestyle includes smoking, unhealthy diet, excessive drinking, no physical exercise or excessive sun bathing.

Age of patient (levels: 25, 43, 68, 87 years)

Each level is representative of an age group and reflects the age structure of adults (over 18 years) in Germany. The level 25 years represents the group of young adults in their early career; the level 43 years represents the middle aged adults, mainly with a settled career; the level 68 years represents retired people and the level 87 years the old-age group.

Family status (levels: single with/out dependents (children, relatives to care for), couple with/out dependents (children, relatives to care for))

For plausibility the description with/out dependents was specified as single/couple with/out children when the age in the profile was 25 or 43 years and single/couple with/out family members to care for when the age in the profile was 68 or 87 years. Couple means living together with a partner and includes all kinds of partnerships.

Occupational status (levels: high, medium, low)

The attribute served as a proxy for socioeconomic status. A high occupational status includes CEOs, physicians, lawyers; a medium status includes clerical workers and craftsmen; low status includes unskilled workers and long-term unemployed people. For profiles with levels 68 or 87 years, the occupational status refers to the time before retirement.

Experimental Design: Generation and presentation of patient profiles

With six attributes, two of them having two levels, two with three levels, and two with four different levels, 576 ( = 4²×3²×2²) different profiles are possible. The number of profiles was reduced to 23 by applying an orthogonal fractional factorial main effects design, utilizing the SAS PROC OPTEX routine which ensures d-optimality. Combining three profiles into one choice set implies there are Inline graphic possible sets; a subset of 25 choice sets was selected using the SAS %choiceff macro which also ensures d-optimality (see [32] for procedural details). Of these 25 choice sets, each respondent saw four different randomly selected choice sets which were embedded in the study's questionnaire. All profiles were presented as full profiles (i.e. all attributes presented); the participant had to indicate which hypothetical patient should be treated first and which patient should be treated last. Table 1 illustrates how the choice set was presented to respondents, including the settings preface.

Table 1. Illustrative Choice Set in Discrete Choice Experiment.

Patient Characteristic	Patient A	Patient B	Patient C
Occupational Status	high	medium	low
Health status	light disease	severe disease	severe disease
Quality of life	severely restricted	restricted	no restrictions
Unhealthy life style	yes	yes	no
Age	25	43	87
Family status	single with child	single with child	single with no relatives to care for
	Which of the patients would you treat first?
	□	□	□
	Which of the patients would you treat last?
	□	□	□

Open in a new tab

Above we introduce three patients with different characteristics. Which of the patients would you prefer be treated first and which last?

Limiting the choice sets to four within one questionnaire was done to minimize cognitive burden. Obviously that limits the amount of data provided by any single respondent: instead of 25·N total choice responses only 4·N are obtained, where N is the number of participants in the survey. However, since the choice task produces the ranking of the three hypothetical patients described, the information content in the preference rank ordered choice sets can be exploited to double the amount of data [16].

Analysis

Socioeconomic status

The socioeconomic status was determined by the “Winkler-Index” [33]. This measure is a three-dimensional, additive, non-weighted social class index using academic/vocational education, monthly net household income and current/last occupation as indicators. Each indicator ranges from one to seven points, where one point represents the lowest and seven the highest social status; hence the Winkler-Index can take values between three and 21 points. Three social status groups with equal ranges can be defined on the basis of this index: lower status (3–8 points), middle status (9–14 points) and higher status (15–21 points) [34].

Health status

The SF-8 Inline graphic Health Survey produces a physical (PCS) and a mental (MCS) component summary measure. Based on the scores and according to the instrument norm of 50 [35] each participant was categorized as average and above (score≥50) and below average (score<50), separately for each component. In addition, the sample medians were taken to categorize the participants in above and below average.

Family status

The family status of the sample was inferred from the combination of three questions in the survey: participants indicated their 1) marital status, 2) partner status (yes/no), and 3) number and age of persons living regularly in the household.

Lifestyle

The lifestyle measure we employ is based on 1) smoking habits (non-smokers vs. smokers); 2) alcohol consumption habits (none/little, moderate, heavy); 3) weight and height of participants, converted to the Body Mass Index (BMI) (underweight (BMI<18.5), normal weight (18.5≤BMI<25), overweight (25≤BMI<30), obese (BMI≥30); 4) body exercise habits (often - three and more times a week, moderate - one to two times per week, and never/seldom). Each category value was assigned one, two and three points, where one point represents behavior that presumably does not affect health negatively, and three points represents behavior that presumably affects health most negatively. A lifestyle index was determined by adding the points. The lifestyle category was defined on the basis of this index: healthy (4–5 points), average (6–7 points), and unhealthy (8–12 points). Note that the categories for non-smoking and smoking received one and three points, respectively.

Survey data

A multiple correspondence analysis [36] was performed on the overall results to discover patterns of criteria of acceptance. To investigate the influence of respondents' age, sex, socio-economic status, health status, family status, and lifestyle on attitudes towards patient characteristics, we conducted a supporting contingency analysis with adjusted residuals as follow-up chi-square tests. An adjusted residual that exceeds about 2 or 3 in absolute value indicates a lack of fit of H₀ in that cell [37]. Here we applied them as posthoc tests with a 5% (adjusted residual = 2) and 1% (adjusted residual = 3) significance level. A binary logistic regression analysis was carried out with age, PCS (physical health status), and MCS (mental health status) as covariates, and socio-economic status and lifestyle as factors.

Utilities, choice probabilities, and relative importance

Data for the discrete choice experiment were analyzed using a random utility model for rank ordered data [16]. The utility Inline graphic of the k^th profile in the l^th choice set is determined by the sum of the partworth utilities of attribute i with level j, perturbed by error , i.e.,

(1)

where Inline graphic indicates whether profile has attribute i with level j () or not (). The error components are identically and independently distributed Gumbel (double exponential) variates with zero mean and variance π²/6. The probability that choice alternative (profile) k of the l^th choice set is chosen is therefore given by

(2)

According to the rank order explosion rule [16], the rank ordered observations can be exploded (decomposed) into statistically independent choice observations under these assumptions. For instance, adding the observed frequency of the rankings of the patients Inline graphic and ( means “is preferred to”) gives us the frequency ranking A as first out of the choice set {A,B,C}. The observed frequency of the ranking , and gives the frequency for ranking A first out of the choice set {A,B}.

The relative importance of attribute i is defined in terms of the range of its partworth utility values relative to the partworth utility ranges of all attributes:

graphic file with name pone.0036824.e018.jpg

(3)

Data of all participants were aggregated to obtain this measure.

The random assignment of four choice subsets out of the 25 possible choice sets to a questionnaire does not necessarily lead to a balanced distribution of choice sets over completed questionnaires. This may lead to heterogeneous precision for the partworth estimates of different attribute levels. To assess this possibility and adjust the estimates we simulated a null model by assigning random answers to the observed (distribution of) choice sets. The procedure was performed on the entire data set and on population subgroups. The confidence intervals (CI) for the attribute levels are directly obtained from the covariance matrix; the 95% (CI) for the relative importance were determined via bootstrap sampling: 1000 random samples with 2031 observations were drawn from the original data set (N = 2031). For each random sample we calculate the relative importance of the attributes. The 1000 estimates of relative importance were ordered by size and the 25th and 97.5th estimates were used as estimators for the 2.5% and 97.5% quantiles, respectively, and hence the lower and upper limit of a 95% confidence interval.

Results

Sample Description

The number of selected addresses was 3729, of which 3% were ineligible (e.g., no private household). Of the remaining 3617 addresses, 22% of the target persons were unavailable, 13% refused to take part and 8.2% were unable to do so for other reasons (e.g., illness), resulting in a response rate of 56.8% (2031 respondents). The sample is representative for the adult population (18 years and above) of Germany. It includes 1131 (55.6%) female and 900 male respondents. Mean, median and standard deviation of their age is 52, 52, and 18 years, respectively. For the analysis, respondents are grouped into three age groups: 18–29 years (14.1% of the sample), 30–59 years (46.7% of the sample) and 60 years and above (39.2% of the sample). The first group represents young adults in their early career; the second the working age group and the last the elderly. The average de facto retirement age is 60 years in Germany.

Socio-economic status

According to the Winkler-index, 47% of the respondents belong to a lower social status, 39.8% to a middle social status, and 13% to a higher social status groups. Three participants (0.2%) did not give any information and therefore could not be classified.