Objective Transition towards value-based healthcare requires insight into what makes value to the individual. The aim was to elicit individual preferences for cardiovascular disease screening with respect to the difficult balancing of good and harm as well as mode of delivery.
Methods A discrete choice experiment was conducted as a cross-sectional survey among 1231 male screening participants at three Danish hospitals between June and December 2017. Participants chose between hypothetical screening programmes characterised by varying levels of mortality risk reduction, avoidance of overtreatment, avoidance of regretting participation, screening duration and location. A multinomial mixed logit model was used to model the preferences and the willingness to trade mortality risk reduction for improvements on other characteristics.
Results Respondents expressed preferences for improvements on all programme characteristics. They were willing to give up 0.09 (95% CI 0.08 to 0.09) lives saved per 1000 screened to avoid one individual being over treated. Similarly, respondents were willing to give up 1.22 (95% CI 0.90 to 1.55) or 5.21 (95% CI 4.78 to 5.67) lives saved per 1000 screened to upgrade the location from general practice to a hospital or to a high-tech hospital, respectively. Subgroup analysis revealed important preference heterogeneity with respect to smoking status, level of health literacy and self-perceived risk of cardiovascular disease.
Conclusions Individuals are able to express clear preferences about what makes value to them. Not only health benefit but also time with health professionals and access to specialised facilities were important. This information could guide the optimal programme design in search of value-based healthcare.
- quality and outcomes of care
- cardiac risk factors and prevention
Statistics from Altmetric.com
Moving to value-based healthcare has become a strategic priority across healthcare systems. Within this framework, the overarching goal is to achieve high value for patients and/or users of healthcare services. Information on individual preferences for process and outcome is thus required in order to prioritise alignment of evidence-based medicine with individual preferences.1 Population-based screening programmes are typically characterised by clear benefits from early detection and prevention but also unavoidable harms such as psychological distress, over diagnosis and overtreatment.2 The balancing of good over harm is an ever-challenging dilemma to evidence-based medicine, decision-makers and practitioners, but also to the individual, who must decide whether or not to participate if invited. Individuals are thus a direct source of information, which is increasingly relevant to take advantage of as methods for preference elicitation advances.
Within the screening literature individual preferences for cancer screening have been studied. These studies confirm that the expected health benefit is of top priority to screening participants but also that there is an unclear impact of mode of delivery.3 Primary and secondary screening programmes for cardiovascular disease (CVD) have been implemented in the forms of simple health checks4 and screening for abdominal aortic aneurysm (AAA).5 More recently, advanced multifaceted screening programmes targeted also peripheral artery disease, hypertension and ischaemic heart disease among others have been proposed in men.6 7 However, contrary to the field of cancer, preferences for CVD screening seem to be uninformed.
There is a fair knowledge on the factors that affect the participation decision in screening. These include sociodemographic and socioeconomic status, lifestyle, lack of readiness to face the outcome and consequences, fear of overtreatment, the views of significant others, information about screening, location, accessibility, time and costs.8 9 Such results provide no guidance in relation to the difficult trade-offs that are required when deciding about whether or not to offer screening and how to design the optimal screening programme to maximise value to users. By eliciting individuals’ preferences and quantifying their trade-offs between good and harm, valuable guidance is offered to healthcare management with a strategic priority of transition towards value-based healthcare.
In this study, we used a discrete choice experiment (DCE), which is established in other areas of health science3 10 but less so within cardiovascular research.11 The basic concept is that an individual is repeatedly asked to choose between hypothetical alternative healthcare offers, for example, screening programmes. For each choice, selected characteristics of the programme are varied such that the importance and relative weight of these characteristics can be statistically modelled in order to describe what makes value to the individual.12 The objective of this study was to elicit individual preferences for CVD screening with respect to the difficult balancing of good and harm as well as mode of delivery.
A DCE was conducted as a cross-sectional survey in accordance with state-of-the-art recommendations.13 The study complies with the Declaration of Helsinki.
Setting and participants
Participants were recruited consecutively from Danish general population males aged 60–74 years, who had accepted to participate in screening at three Danish hospitals between June and December 2017.7
Prior to the screening appointment, screening invitees received an invitation letter describing the purpose, benefits and risks of screening along with information on venue and duration, instructions on how to book a time slot via web, email or telephone, and illustrations of the screening tests. The tests included: (1) low-dose non-contrast CT scan to detect coronary artery calcification and aortic/iliac aneurysms, (2) brachial and ankle blood pressure index to detect peripheral arterial disease and hypertension, (3) a telemetric assessment of the heart rhythm and (4) a measurement of the cholesterol and plasma glucose levels. Invitees were further informed that in case of positive findings they would be offered a consultation and initiation of evidence-based therapy.7
Discrete choice experiment
The DCE was administered at the screening site following the screening appointment but before any knowledge of the test results. Respondents had the option of completing and returning the DCE on site, to complete it later electronically or to return it by mail. The DCE included six choice sets per respondent. An example is illustrated in figure 1. Respondents were asked to choose between unlabelled alternatives (screening programme A and screening programme B) each characterised by five attributes, that is, characteristics (mortality risk reduction, avoidance of unnecessary medical treatment, avoidance of regretting participation, screening duration and location). For each attribute, three levels were defined based on balancing of plausibility and ensuring a relevant contrast between the levels. Table 1 shows the attributes, levels and the expected direction of the preferences. The DCE was generated from two steps:development of the DCE and construction of the experimental design based on a pilot DCE.
Development of the DCE questionnaire
A first version of the DCE including 12 choice sets with the five attributes was constructed for qualitative evaluation. The attributes were selected based on literature reviews3 8 9 and discussions among researchers and experts on DCE methods or CVD screening and considered of particular importance for decision-making and programme valuation. Structured interviews and think-aloud choices of the DCE were undertaken with four respondents to discuss the overall design (layout, length, complexity and understandability), the selected attributes (number, terminology, relevance, conciseness, dominance and interconnectedness) and the attribute levels (formulation and range). This version proved to include too many choice sets per respondent. Also, the terminology and framing could be optimised. In order to avoid that one of the selected attributes mortality risk reduction dominated the choices, the upper attribute level was reduced from 20 to 10 lives saved per 1000 screened and the row of the attributes were altered.
Constructing the experimental design
A revised second version of the DCE based on input from the qualitative evaluation was generated for a pilot test among 60 respondents based on an orthogonal fractional factorial design using Ngene version 220.127.116.11 This attribute level balance design is a subset of all possible attribute level combinations in which there are no correlations between attributes (ie, each pair of attributes appears an equal number of times within the design).15 16 To minimise respondent burden, this second version included six choice sets (18 choice sets generated from the design were blocked into three different versions with respondents randomised to one of the versions). The pilot data were analysed in a multinomial logit model.16 The analysis confirmed the feasibility and face validity of the DCE and provided prior values (best guesses) of the parameters that were used to replace the orthogonal design with a design that more efficiently captures information from respondents. This final Bayesian design was evaluated through the expected D-efficiency measure.15
The DCE questionnaire was appended questions on health literacy, health status, family history of CVD, self-perceived risk of CVD, previous screening history and decision-making. Health literacy was measured using the 5 items Health Literacy Questionnaire scale ‘Understand health information well enough to know what to do’.17 Health status was measured using the EuroQoL EQ-5D-3L instrument.18 Information on body mass index, smoking status, comorbidity and use of preventive CVD medication was obtained from screening data.6
A mixed multinomial logit model was used to analyse the choice data. This model accounts for the panel nature of the data (here because multiple observations are obtained from the same respondent) and allows for preference heterogeneity across respondents.16 19 20 Different model specifications were tested to determine the best model fit. In the final model, screening location was specified as a dummy variable and all other variables as linear. Mortality risk reduction was specified as fixed and the remaining parameters as random (which allows respondents to have heterogeneous preferences) with a normal distribution. Missing responses were excluded from the analysis. The final model used 1000 quasi-random Halton draws from the normal distributions to ensure stable estimation results.
The respondents’ trade-offs between screening characteristics were estimated by marginal rates of substitution (MRS) where the model parameters are converted to a single scale, sometimes referred to as a payment attribute for which extra levels of another attributes can be bought. Mortality risk reduction was used as payment attribute with the interpretation of how much risk reduction (number of lives saved) a respondent is willing to give up or require in order to get more or less attractive levels of other attributes. The Krinsky and Robb parametric bootstrapping method was used to compute the CIs of the MRS.21 To inform the role of respondent characteristics on preferences selected characteristics (smoking status (never, previous, current), level of health literacy (dichotomous variable of being in the worst off quintile of all respondents (no/yes)), self-perceived risk of CVD (high, medium, low) and existing CVD (either stroke, myocardial infarction, peripheral arterial disease, AAA, arrhythmia, previous cardiac surgery or percutaneous coronary intervention) (no/yes) were tested in the model specified as interaction terms. These characteristics were chosen since they are either central with regards to current screening practice8 22 or shown to impact decision-making and thus valuation of screenings.23 24
One choice set included a dominant alternative to assess the respondent’s understanding of the attribute levels. Following the DCE, respondents were further asked to report their level of certainty about their choice. Sensitivity analyses were used to assess the quality of the responses and the methodological robustness of the analysis. This included excluding observations where: (1) respondents stated feeling very insecure about their choices, (2) respondents were insensitive to level changes (chose the same alternative consistently across choice sets), (3) respondents failed to choose the dominant alternative and (4) a model specifying the attributes avoidance of overtreatment and of regretting participation as factor variables.
All analyses were performed using Stata14.
Of the 1231 participants in the survey, 830 (68%) responded. Of the 830 respondents, 264 (31%) responded electronically and the remaining by postal mail. The characteristics of respondents and non-respondents are shown in table 2. Minor differences between the groups were observed with regards to age, body mass index and health status. Of the 830 respondents, 699 (84%) responded to all choice sets of the DCE, 112 (13%) responded to some of the choice sets, while 19 (3%) did not respond to any of the choice sets. The resulting data for the DCE analysis was 9016 observations.
Individual preferences for programme characteristics
Table 3 shows the results of statistical modelling of the individual preferences. As expected, respondents expressed clear preferences for higher mortality risk reduction, avoidance of overtreatment and avoidance of regretting participation. They also expressed clear preferences for a screening programme of longer duration and for a high-tech hospital as location. That the direction of the preferences are consistent with our ex-ante hypotheses provides some theoretical validity.
Willingness to give up mortality risk reduction
Despite mortality risk reduction being a top priority, respondents were willing to give up some risk reduction to gain improvements in other programme characteristics. Table 4 shows the number of lives saved that individuals are willing to give up in order to upgrade one level on other programme characteristics. Respondents were willing to give up 0.09 (95% CI 0.08 to 0.09) life saved per 1000 screened in order to avoid one individual being overtreated. Similarly, respondents were willing to give up 1.22 (95% CI 0.90 to 1.55) lives saved per 1000 screened to upgrade the location from general practice to a hospital and 5.21 (95% CI 4.78 to 5.67) lives saved per 1000 screened if upgrading was to a high-tech hospital.
The statistical significance of the SDs of table 3 indicates that different subgroups hold different preferences. Subgroup differences with respect to smoking status, level of health literacy and self-perceived risk of CVD were observed to explain some of the preference heterogeneity (table 5). As compared with never smokers, previous and current smokers had stronger preferences for a hospital based versus a general practice-based programme. For instance, while never smokers were willing to accept 0.69 (95% CI 0.24 to 1.16) fewer lives saved in order to have screening at a hospital, current smokers were willing to give up a statistically significant additional 1.00 (95% CI 0.08 to 1.91) lives. Similarly, current smokers had stronger preferences for the highest level upgrading, to a hospital setting with CT, than never smokers. Individuals with low health literacy also reported stronger preferences for the hospital-based programme than individuals with higher health literacy. Finally, individuals who perceived their risk of CVD to be low were less willing to give up lives saved to avoid overtreatment than their counterparts, who perceived their risk of CVD to be high.
The certainty of individual responses appeared to be good with only 46 (6%) feeling very uncertain about their choices. The sensitivity analyses (table 6) did not alter the overall findings of the base case analysis, indicating that the results are robust to unsecure or irrational answers as well as econometric specification.
In this study, we found that individuals were able to express clear preferences on complex constructs on the balancing of good over harm in a population screening programme for CVD. Health benefit and also time with health professionals and access to more high-tech facilities were important. The study represents one of the first examples on choice-based preference elicitation for cardiovascular therapy. Trade-offs between good and harm of screening have been examined in the field of cancer screening, and our results are overall in consensus. For example, men have expressed willingness to trade risk reduction for death caused by prostate cancer in order to decrease their risk of unnecessary treatment at a ratio around 1:5.25 The ratio in our study was less than 1:10, which is intuitive given the distinctly different characteristics of cancer and CVD overtreatment (chemo and/or radiation vs medication).
Alignment of evidence-based medicine with individual preferences
In times of many healthcare services focusing on value-based healthcare, what makes value to the individual is important for clinical practice guideline panels, clinicians and policy-makers concerning the difficult screening ethics and also more general within the design of cardiovascular therapies. The main reason for incorporating values and preferences in guideline development is that recommendations aligned with patient values and preferences may be more easily accepted, implemented and adhered to since different stakeholders, for example, healthcare professionals and patients may choose different treatment options although presented with the similar evidence. Increasingly, groups such as the Canadian Task Force on Preventive Health Care use findings from systematic reviews to improve their understanding of patient preferences when they weigh the benefits and harms of specific screening interventions as part of guideline development.26 Other strategies to inform the impact of values and preferences on the strength of recommendations employed by, for example, the National Institute for Health and Clinical Excellence include asking patient representatives to reveal their experiences and to include reviews of qualitative research and cross-sectional surveys.27 Information on individual preferences and trade-offs are thus highly requested but still limited with regards to CVD prevention.11
Importance of mode of delivery
The results of our study are also useful to inform the practical design and organisation of cardiovascular therapies. Screening programmes in the form of primary and secondary prevention general health checks have been investigated28 and found to have limited effects on hard endpoints. The current results show that a high-tech hospital-based programme is strongly preferred by individuals even when the mortality risk reduction was held even between different locations. Although that is not necessarily realistic, it suggests that specialised equipment and personnel provide higher value to individuals than the security of facing a known general practitioner.
Targeting of vulnerable subgroups
The subgroup analysis focused at selected dimensions of relevance to targeting high-risk individuals. If different groups have distinct preferences, it might be possible to favour their participation in the programme design. For instance, with regards to screening for AAA, the US Preventive Services Task Force conclude that screening is most relevant for ever-smokers.22 Our results show that it is particularly important for current and ever-smokers that screening is hospital based. In addition, if current smokers are the main target group, a high-tech hospital with CT scan holds the highest value. Similarly with regards to lower health literacy, which in increasingly recognised as an aspect that require specific attention,29 our results identify location as important dimensions for choice.
Strengths and limitations
The main strengths of this study include a large sample of individuals with real-life experience in deciding about screening. Further, the study was designed and analysed according to current guidelines,13 including allowance for preference heterogeneity and sensitivity analysis of the methodological robustness. Still, our study suffers the usual limitations when eliciting preferences on complex services. In particular, the intangible character of, for example, avoidance of regretting screening participation might have been difficult to understand, and the fact that some characteristics were to some extent overlapping, for example, screening location and duration of screening might have impacted the estimates. The main limitations include the fact that we surveyed only men since women were not invited and could not be surveyed as ‘experienced’ individuals, who had reflected on their decision. Surveying such individuals would incur a risk of hypothetically bias. Another main limitation is that the timing of surveying (after decision to participate in screening) might have affected the results if respondents ‘defended’ their actual choice logic when completing the choice task. This means that generalisability to non-participants should be made with great caution. The reason for the timing was a priority about ethics and not impacting participation because screening participants were enrolled in a clinical trial.7
Individuals are able to express clear preferences about what makes value to them. In this case, not only health benefit but also access to specialised facilities and time with health professionals were important. This information could guide the optimal programme design in search of value-based healthcare.
What is already known on this subject?
There is increasing interest in patient preferences in order to support patient-experienced value of clinical practice, to inform guidelines development and more generally to guide priority setting in the direction of value-based healthcare. However, there is a lack of studies reporting on patient preferences for cardiovascular disease screening.
What might this study add?
This study serves as one of the first examples of systematic preference elicitation based on a discrete choice experiment within preventive cardiology.
How might this impact on clinical practice?
The study might hold importance for the development of practice guidelines and for the design of cardiovascular disease screening programmes.
We thank the survey respondents and the staff for the distribution. We also thank statistician Gunnar Hellmund Laier for assistance on the generation of the DCE design and data analysis.
Contributors TBH and RS conceived the idea for the study. All the authors contributed to the design and planning of the study. TBH, MCJB and RS performed the statistical analyses. TBH wrote the first draft of the manuscript. All revised the manuscript critically. All have given their final approval of the version to be published. TBH and RS are responsible for the overall content.
Funding The randomised trial from which the participants in the current study was recruited has received financial support from the South Region of Denmark, the Danish Research Council, the Danish Heart Foundation, the Helse Foundation, Odense University Hospital and the Elitary Research Centre of Individualized Medicine in Arterial Disease (CIMA). For the DCE study, financial support has been provided by the Region Zealand (RSSF2017000614).
Competing interests None declared.
Patient consent Not required.
Ethics approval The study was approved by the Danish Data Protection Agency (REG-003-2017).
Provenance and peer review Not commissioned; externally peer reviewed.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.