Chronic Obstructive Pulmonary Disease

Exploration of preferences among people with COPD to inform resource allocation: a discrete choice experiment study

Abstract

Introduction Treatment options for chronic obstructive pulmonary disease (COPD) are numerous but adherence remains a key challenge. We performed a discrete choice experiment (DCE) of patients’ preferences in accessing care for the management of COPD. The aim of this study was to understand patients’ preferences for modes of accessing care for the management of COPD. This piece of work was then used to inform resource allocation decisions in five integrated care systems (ICSs) in England.

Methods People with diagnosed COPD in five ICSs were invited to complete an online survey from August to September 2022. An experimental design built on the principles of minimal overlap, level balance and orthogonality was used to create 20 sets of 11 scenarios for participants to assess. Participants were presented with three hypothetical options and asked to select their most preferred or state that none was preferred. Data were analysed using a hierarchal Bayes algorithm.

Results Of 82 639 patients with COPD in the study area, 520 completed the survey. The mean health-related quality of life score derived using EuroQol 5-Dimensions 5-Level was 0.57 (0.29). The attributes assigned greatest importance were treatment outcomes, treatment delivery and the type of staff who deliver treatment. Mean utility level scores were substantially higher for little relief (22.75 (SD 78.80)) or some relief from symptoms (20.67 (46.77)) than for complete relief (‒43.42 (83.03)). Of the treatment delivery options, in-person individual appointments were preferred (mean utility score 48.34 (SD 48.14)), and care being provided by healthcare professionals was viewed as very important (77.50 (64.39)).

Conclusions The DCE approach can help resource allocation decisions by indicating attributes most important to patients and trade-offs they are willing to make in treatment access and delivery.

What is already known on this topic

  • Treatment options for chronic obstructive pulmonary disease are numerous throughout the disease pathway but adherence remains a key challenge. Regulators have called for more emphasis to be placed on patients’ views to inform resource allocation decision-making. Discrete choice experiments can provide insights into patients’ preferences through their choices of hypothetical scenarios with a range of attributes and possible levels of outcome.

What this study adds

  • A surprising preference was for some or little relief rather than complete relief, suggesting that participants are aware of high-risk intervention trade-offs involved in achieving the latter (eg, invasive surgery or quitting smoking). The findings also provided empirical evidence for healthcare delivery.

How this study might affect research, practice or policy

  • The data yielded from a large study sample were suitable to enable mesolevel resource allocation decisions specific to integrated care systems. Assessing patients’ preferences in this way could be especially useful to increase uptake of, and adherence to, therapies, such as pulmonary rehabilitation. A future direction of research would be to engage people not already interacting with healthcare services to improve reach and engagement for subgroups at risk of missed diagnosis.

Introduction

Chronic obstructive pulmonary disease (COPD) is the fifth-leading cause of mortality worldwide and is a substantial burden on healthcare resources.1 In England, the estimated prevalence of COPD in 2011 was 1.79%, leading to a healthcare cost of roughly £1.5 billion. These values are expected to rise to 2.19% and £2.32 billion, respectively, by 2030.2 Such information is of particular concern in COPD, where the treatment options are numerous throughout the disease pathway and adherence to medical and other interventions remains a key challenge.3 4 For example, despite substantial evidence indicating that pulmonary rehabilitation improves exercise capacity and quality of life, uptake and completion rates remain low.5 6 Non-health factors are also important to the process of receiving and completing care, such as treatment costs, treatment risks and travel times.1 7 8 In the example of pulmonary rehabilitation, perceived barriers to accessing care were lack of transport and work commitments.6

Understanding how patients like to receive care can be vitally important in priority setting and resource allocation decisions. Indeed, regulators have called for more emphasis to be placed on patients’ views to inform decision-making.9–15 One method to obtain relevant evidence is discrete choice experiments (DCEs). They are being increasingly used by health technology assessment agencies and as a part of clinical decision-making in areas like oncology and end-of-life care. However, their use to support decision-making for medium-sized populations (mesolevel) has been limited.16 17

DCEs are surveys in which respondents are presented with a series of hypothetical scenarios. Each scenario is described as a set of attributes and possible levels of outcome. Scenarios presented can consist of attributes which address a range of relevant factors that affect patients’ decisions (eg, factors that affect access to care, such as travel time, travel costs, and geography, and receipt of care, such as who delivers treatment, and types of tests) and drivers of demand for services. They are developed under the assumption that individuals derive utility from them, and respondents are required to select one of the three hypothetical options presented to them in each scenario. The choice made elicits preferences and reveals the relative importance placed on different attributes.18 19 There is the potential for DCEs to contribute to resource allocation decisions and service design by enabling health-service planners to quantify and understand the relative importance of treatment attributes and acceptance levels of potential trade-offs in different scenarios.20

DCEs have been conducted to assess care for COPD, but so far have mainly focused on benefits and convenience of inhaler treatments.8 With few data available on risk tolerance, the data cannot be used by health-service planners to inform resource allocation across the wide variety of services and interventions that are offered for COPD.

The aim of this study was to understand patients’ preferences for modes of accessing care for chronic disease management of COPD. It formed part of a pilot scheme aimed at supporting integrated care systems ((ICSs) partnerships of National Health Service (NHS) and local authority organisations responsible for planning health services21) to make decisions about resource allocation for COPD care. The data were used as inputs in a sociotechnical allocation of resources (STAR) analysis,22 to inform resource allocation; this further analysis is beyond the scope of this article and will be reported elsewhere.

Methods

Study design and participants

This was a prospective DCE study. Eligible respondents were patients older than 16 years with a pre-existing diagnosis of COPD. Potential participants were contacted via five ICSs (Gloucestershire, Birmingham and Solihull, Northamptonshire, Coventry and Warwickshire, and Nottingham and Nottinghamshire) participating in the STAR process. The research team provided COPD teams in each of the ICSs with promotional materials, including leaflets, bulletins, template emails, text messages and social media posts to help them disseminate the survey. As the aim of this piece of work was to inform their decision-making, the research team gave as much ownership to the local teams as possible. The ICSs then disseminated the open link for the survey using their existing communication channels. This included sending text messages to registered COPD patients, bulletins in general practices and ICS practices and social media platforms, information via COPD community teams and respiratory teams. The survey could be completed by patients face to face in COPD clinics or treatment groups, online, and/or via the telephone. The aim was to sample as many of the patients with COPD in the ICSs as possible. Data on the UK Government Public Health Profiles website (https://fingertips.phe.org.uk/) indicated that there were 82 639 patients with COPD in the ICSs of interest at the time of the study. No incentives were offered to patients for completing the survey.

On clicking on the survey link, potential participants were presented with written information about the study and provided with a specific email address that could be used in case of any questions before proceeding. They were asked to agree that they had read the information, accepted the collection and use of personal and health-related data, and voluntarily consented to participate. They could only proceed to the screening questions by clicking ‘yes’; the survey automatically ended if they clicked ‘no’. The survey similarly automatically ended following the screening stage if the individual did not meet the inclusion criteria. All data were collected in an anonymised format and were stored and processed under strict compliance with the UK Data Protection Act 2018. The ICSs could also undertake their own data protection impact assessment if they thought it necessary.

Survey

The survey was available for 3 weeks in August–September 2022 and included the DCE alongside specific questions about sociodemographic factors, health and COPD. We used the modified Medical Research Council dyspnoea scale (MMRC) to identify the impact that COPD has on respondents’ everyday lives23 and the EuroQol 5-Dimensions-5-Level (EQ-5D-5L) questionnaire to measure respondents’ health status and predict severity of COPD.24 We further computed utility scores via the three-level EQ-5D (EQ-5D-3L), using utility weights as recommended by National Institute for Health and Care Excellence,25 26 to split the sample into health quartiles. The first quartile indicated the worst COPD severity and the fourth quartile indicated the least COPD severity. These quartiles were used for subsample analysis of the DCE findings by COPD severity. The aim was to reach as many people with COPD in the five ICSs as possible within the limited survey period.

Attributes and levels

We developed the DCE attributes and levels in three predefined steps. First, we identified 10 attributes and their respective levels from a systematic review reported by Collacott et al.8 Second, these were shared with an expert group of clinicians, researchers and healthcare managers from the participating ICSs for review of relevance and appropriateness to the patient population. The panel could add or edit attributes and levels and was asked to rank the attributes’ importance. Third, based on the ranking, we selected the six most important attributes for the DCE (figure 1).

Figure 1
Figure 1

DCE key attributes and levels. COPD, chronic obstructive pulmonary disease.

Discrete choice experiment

An experimental design built on the principles of minimal overlap, level balance and orthogonality was used to create 20 sets of 11 scenarios for participants to assess. Each scenario included the six key attributes, which were accompanied by different level options that were varied to allow identification of trade-offs that participants used in their choices. An example of how the scenarios were displayed to participants is included in figure 2. Participants were presented with a series of three hypothetical options and asked to select their most preferred option or to select that none of the options were preferred. In each task, participants could select the set of care options that they preferred or select to choose none of them.

Figure 2
Figure 2

Example task and choice frame. Participants could choose one or none of the options. COPD, chronic obstructive pulmonary disease.

Data quality checks

There were no missing data, as all respondents completed all 11 scenarios. However, before estimation, we excluded responses from participants who had selected a concept in the same position in at least 9 of 11 tasks. This is based on the assumption that, due to the randomness of the experimental design, it is statistically improbable that a participant making reasoned choices would choose a concept in the same position this frequently.

After estimation, the consistency across each participant’s choices was calculated using a root likelihood statistic (RLH), which measures the goodness of fit of the model to the observed data. The RLH shows the geometric mean of estimated probabilities associated with the alternatives chosen by participants and is computed by taking the nth root of the likelihood, where n is the number of scenarios. In null model in which the utilities are the same for all alternatives (ie, a model that randomly chooses alternatives with equal probability) has an RLH of 1/k, where k is the number of alternatives. Participants with an RLH lower than 0.25 were removed from the analysis.

Data analysis

The analysis of the choice data was conducted using a hierarchal Bayes algorithm run in the conjoint-based simulation/hierarchical bayes system from Sawtooth Software.27 This technique fits a multinomial logit model to each participant using an iterative approach that maximises the posterior likelihood. That is, it calculates the optimum set of parameters, known as part worth utilities, given the observed participant choice data and knowledge about the population (other participants in the study). It is called ‘hierarchical’ because it has two levels: at the higher level, it assumes that participant utilities are described by a multivariate normal distribution characterised by a vector of means and a matrix of covariances; at the lower level, it assumes that, given a participant’s utilities, their probability of choosing particular alternatives is governed by a multinomial logit model.

The additional information from the population strengthens the estimation by allowing the estimation of how different an individual participant’s data are from the population data. This approach enables robust estimates for each participant even when individual-level data are sparse.

The Bayesian model outputs are part-worth utility estimates for each level tested in the experimental design per participant. These scores were the basis of all summary metrics and simulations. In modelling the preferences of each participant, the utilities predict what participants would select when faced with different service options. To convert utilities to preference shares, the following calculation was used:

Display Formula

where Ui represents the product utility for the concept of interest; and Uj + … are the product utilities for each of the competing concepts in the simulation. The total utility for each concept is calculated by summing the utility scores for each level within a concept.

Preference shares

Using a dynamic Excel-based simulation tool, we estimated the preferences of respondents for three total treatment delivery options: option 1 was the status quo of treatment delivery in the current health system providing in-person individual appointments by healthcare professionals but with longer waiting and travelling times and associated higher out-of-pocket expenditures for patients to access their treatment; option 2 was in-person group appointments delivered in the community by healthcare professionals, which would reduce waiting and travelling times and out-of-pocket expenditure for patients to access care; and option 3 was individual online appointments delivered by healthcare professionals, which would have the lowest waiting time, no travelling time, and no out-of-pocket expenditure. The outputs were preference shares, which enable understanding of overall support (including no support) for a complete treatment delivery approach compared with an alternative option. They also show the relative impact of each attribute level on the degree of support.

Informing resource allocation

The results of the DCE were presented to each ICS’s COPD team at decision conferences as part of the STAR process.28 The method combines extensive stakeholder engagement with analysis of value for money.28 Each decision conference was attended by patients, caregivers, respiratory consultants, nurses, physiotherapists, public health professionals, ICS management teams and members of local respiratory charities. The DCE results were presented by the research team alongside the results of a systematic literature review of non-pharmacological interventions for COPD (the results of which will be published elsewhere). The local ICS team also presented any data of interest locally (such as rates of emergency admissions for COPD patients). Using this information, the ICS teams developed recommendations for how to allocate resources in future budget years (which will be reported elsewhere).

Patient and public involvement

Patients and/or the public were not involved in the development of this study.

Results

Characteristics of respondents

Of 82 639 patients with COPD in the study area, 520 agreed to participate, met the inclusion criteria and completed the survey. The characteristics of the participants are presented in table 1. Respondents were well balanced in terms of sex and level of education. Most respondents (98%) were white, lived in urban areas (70%) and resided in Gloucestershire (45%) or Coventry & Warwickshire (30%). COPD diagnosis had generally been received within the previous 1‒5 years. Most respondents (90%) reported using inhalers and many had experienced exacerbations in the previous 12 months (70%). About one-third of respondents reported moderate problems with mobility, usual activities and pain or discomfort. About half of the respondents reported no problem with self-care and slight to moderate problems with anxiety and/or depression. The mean health-related quality of life scores derived using EQ-5D-3L was 0.57 (SD 0.29) and 60% of patients had an MMRC score of 3–5.

Table 1
|
Baseline characteristics of survey respondents

Importance of attributes and utilities

The attributes assigned greatest importance by respondents were treatment outcomes, method of treatment delivery and the type of staff who deliver treatment (table 2). Less importance was placed on travel time to the appointment, waiting times from referral to treatment and out-of-pocket expenditures. When assessed in subgroups for sex, educational, region, place of residence and health utility quartiles, we observed similar rankings for attributes (online supplemental appendix table A1).

Table 2
|
Average attribute importance levels

Utility-level scores for treatment outcomes showed that respondents found little or some relief important, whereas complete relief was given low importance (table 3). Of the treatment delivery options, respondents selected in-person individual appointments, and care being provided by healthcare professionals as very important. For the less important attributes, little difference was seen between utility levels, but patients placed marginally greater importance on travelling for a maximum of 30 min, spending £1‒£5 on attendance, and waiting a maximum 3 months for an appointment. All findings were consistent in our subgroup analysis except for treatment outcomes in patients with the most severe disease who placed similar importance on some and complete relief from symptoms (online supplemental appendix table A2).

Table 3
|
Average attribute utility levels

Preference shares

When considering total treatment delivery options, respondents assigned similar preference shares to options 1 and 2 and a marginally lower preference for option 3, but most frequently the responses indicated no preference. The preference shares for the whole cohort were 18.4% for each of options 1 and 2, 16.6% for option 3 and 46.6% for none of these.

Discussion

In this DCE, we were able to assess attribute importance and utility levels in a large sample of people accessing care for COPD from multiple regions in the UK. Furthermore, the study was designed to provide data suitable to help stakeholders make decisions about ICS-specific resource allocation, which we believe is a novel use of the DCE method. There were some key ways in which the results of the DCEs informed the discussions in the STAR decision conferences.

Because ICS teams disseminated the survey through existing care channels, only people actively engaging with the health system could have responded. When presented with the characteristics of patients included in the survey, ICS staff discussed how to get healthcare services to reach communities under-represented in this survey.

Results showed that patients selected treatment which led to some or little relief from their COPD compared with complete relief. This was taken in discussions to indicate that participants may have considered that complete relief was not possible for them.

The survey provided empirical evidence of a strong preference for treatment delivery by healthcare professionals and for in-person individual appointments. However, there was little difference in the average utility level between online individual appointments (–0.56 (SD 37.70)) or in-person group appointments (–3.15 (SD 36.52)). One ICS decision conference used this evidence as a basis for discussion on how existing channels, such as the voluntary group sessions run by Asthma and Lung UK Breathe Easy programme and the myCOPDapp (mHealth, Bournemouth, UK), could be leveraged to get healthcare professionals in front of a wider population.

It suggests that strategies should be developed that would increase flexibility for patients. Barradell et al 29 found that a choice of pulmonary rehabilitation programmes was available for patients but not routinely offered at referral. They suggested building menus of treatment delivery options and educating patients to engage with them rather than taking only to the typical hospital-based offerings. Collacott et al’s system review of DCEs of asthma and COPD8 also indicated that patients are willing to make trade-off between attributes such as convenience to improve treatment benefit and reduce risk. Even though convenience was the most studied attribute, risk (eg, adverse event occurrence, likelihood and types and severity of side effects) was the most important attribute to patients in 39% of studies. This finding suggests that risk should be given more consideration in future research to maximise the scope of preference data in decision-making.

Only a small number of DCEs have included more than 300 patients with COPD as respondents.30–33 These studies assessed patients’ preferences about specific treatments aimed at improving adherence rather than attempting to inform resource allocations, and all found that patients place great importance on medication attributes (eg, ease of use and speed of onset of action) as well as safety and efficacy. In two of the studies, patients indicated that they would be prepared to trade off a slight increase in the frequency of exacerbations to maximise treatment attributes.30 33 In another study,34 402 members of the public, valued different elements of healthcare and found that the public valued patient experience and length of life over quality of life improvements.

A systematic review of 28 DCEs shows that few such studies have been performed since 2010. However, regulators are moving towards increased inclusion of patients’ perspectives during the development and assessment of treatments.35 36 DCEs could be a useful tool for not only gathering information on drugs but also understanding how patients see entire treatment pathways. This information can be used by healthcare planners at the meso and local levels to plan services in ways which work best for patients. Improving decision-making in this way was a key aim of the reforms brought about by the Health and Social Care Act 2022 in England.37

This suggests that DCE findings could be used to develop strategies to allocate existing resources in ways that might increase treatment uptake and adherence and could be applicable to a range of diseases and public health issues. Healthcare decision-makers can use the preference data that DCEs generate to design services that meet the needs of patients in a structured and rigorous process.

Addressing poor uptake and adherence in COPD is particularly important, as modelling has suggested that there is a potentially large undiagnosed population.38 While diagnosis has improved over time, notable proportions of patients remain undiagnosed, and in certain subgroups risk of missed COPD diagnosis is increased, including Black, Asian and other ethnic groups, people with no recorded smoking status, and those who are overweight or obese.39 It was a topic of discussion in the decision conferences that the very high proportion (98%) of white patients in the sample population is reflective of the people currently interacting with community services.

Limitations

A limitation of this study was that it only reflected the preferences of people who were already interacting with healthcare services, as the survey was distributed by service providers. This approach meant that population was not fully representative of all COPD patients. For example, this study seemed to have more severe COPD than the general population. In another study that looked at the characteristics of 322 991 patients with COPD in the Clinical Research Data Link, 25% in COPD Gold and 52% in COPD Aurum had a MMRC scores of 3–5 and 28% and 22%, respectively, had had exacerbations in the past 12 months.40 Furthermore, 82% of the total population of England and Wales is white compared with 98% in this study.41 These factors along with wide SDs make some of the results difficult to interpret and, therefore, the findings may not be generalisable to harder-to-reach communities and populations who are not engaged with services. A further limitation is that the survey did not capture information using a respiratory-specific health-related quality-of-life measure like the COPD assessment tool or the St George’s respiratory questionnaire. However, the EQ-5D is a responsive measure of health status in people with COPD24 and it provides a general understanding of health-related quality of life rather than focusing only on COPD outcomes.

We did not perform pilot testing of the DCE, as is recommended in the The Professional Society for Health Economics and Outcomes Research (ISPOR) guidelines,42 due to the limited time and budget required for the study. Instead, a systematic literature review and a clinical reference group were used to inform the survey design. Additionally, data gathered in the first week of the survey were checked and no inconsistencies were observed.

A result of implementing research in an operational setting is that there was variation in the dissemination of the survey, which was due partly to differences in resources and time available across ICSs and agreements that were in place with providers. For example, in Gloucestershire and Coventry, agreements in place with general practices meant the survey link could be texted to patients with COPD directly, whereas other ICSs relied on the support and goodwill of general practice and hospital staff to distribute the survey by hand.

The patient did not participate in the development of this study. As this work had to be conducted in time for ICS budget allocations, there was a trade-off between the timeliness of results and the time needed for study design. This limitation to the study design was offset using a systematic literature review of similar DCEs and the use of expert clinicians to identify attributes reported as being important to patients. The interpretation of the results of this DCE was informed by patients in the STAR decision conferences.

Further development of DCE use by ICSs

DCEs have been used in the decision-making criteria of other diseases, for example, to inform the weighting in multiple-criteria decision models or to inform economic evaluations through estimating willingness to pay.43 Furthermore, DCEs have been successfully used to elicit respondents’ preferences over the design of goods or services.44 Therefore, using preference shares as a criterion within the STAR or another framework could help to ensure that patients’ preferences are explicitly considered by decision-makers and that decisions better reflect them. As the Excel-based simulation tool allows assessment of different hypothetical scenarios, it could be useful in approaches such as decision conferences to design pathways of care based on empirical evidence of patient’s preferences. Time should be incorporated in the service design stage to ensure that stakeholders understand this or a similar tool to produce meaningful insights. Dehmel et al note ‘(using approaches in) combination can bring deeply contextualised, user-centred, operational and experimentally verified ideas for development interventions prior to their implementation’ We found that the hypothetical options we assessed were useful for designing and implementing COPD treatment in the context of local service delivery while considering patients preferences. Wider system impacts of scenario analyses might be related to cost, resources and health outcomes. For example, given that our participants equally supported in-person group appointments and online individual appointments, the latter is likely to be less resource-intensive, thus potentially saving providers money and providing quicker access to treatments and relief of symptoms for patients.

Conclusions

Our DCE provided insight into the preferences of 520 people with COPD. It also produced empirical findings to apply to resource allocation decision-making. This dual approach could be especially useful in areas where uptake and adherence to therapies are low, such as pulmonary rehabilitation. Furthermore, it enables strategies to be considered at the mesolevel, which will be important as ICSs develop throughout England. A future direction of research would be to engage people not already interacting with healthcare services to help design strategies that will improve reach and engagement, particularly in subgroups at risk of missed diagnosis.