1. Introduction

Decisional background

The value drivers of neurorehabilitation are largely unknown, especially from the patient’s perspective. Patients are increasingly exposed to a variety of unknown therapies. Accordingly, there is limited evidence on the characteristics of patient acceptance regarding neurorehabilitation following stroke. Little is known about how patients make decisions for desired outcomes of rehabilitation care, especially after stroke. A multitude of factors influence the decisions: the clinical prognosis, goals for rehabilitation [10], social attributes [11] and desired activities of daily living [12] are important in deciding rehabilitation needs.

Decision-making, especially concerning impaired clinical functions addressed in rehabilitation of stroke patients, is currently guided by the clinical perspective with no specific tools to guide the process. Hence, deciding and assessing rehabilitation for patients with high levels of disabilities is subject to several ethical dilemmas and subjective biases including the value of the clinicians about treating severely disabled patients [4, 13], overestimating or underestimating the impact of stroke on the patient’s health and their expected quality of life [14], fear of litigation and knowledge of the likely outcomes of various interventions [15], which may all contribute to variation in decision-making.

The decision and evaluation of rehabilitation for patients with neuro-disabilities is therefore subject to several manifold dilemmas and subjective biases, including clinicians’ assessments for the best treatments for patients [16, 17], overestimation or underestimation of the impact of stroke on the patient’s health status, quality of life [14] and expected activities, all of which may contribute to differential decision making [16].

The decisions for rehabilitation of severe stroke patients must not be based on known prognostic indicators or rehabilitation requirements alone, but on patient-relevant characteristics [16, 18]. To assess the degree of impairment in clinical functions and activities following stroke, most clinicians use the universal International Classification of Functioning, Disability and Health (ICF) [19]. Within the ICF rehabilitation practitioners can rely for the first on a worldwide accepted model providing a universal language for the description and classification of functioning. The framework is used to describe human functioning and disability. The overarching concept of the ICF is functional health (= functioning), which is mapped in its components of body function and structure, and activity and participation (Fig 1) [19].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Model of physical disabilities according to ICF [19].

https://doi.org/10.1371/journal.pone.0295267.g001

The ICF is divided into different subchapters and subcategories within each component. For every function and activity, the ICF gives a generally valid, very clinical definition. But the question arises of how to assign a value to these functions.

The ICF is a framework on which tools for measuring or ‘assessing’ individual functioning may be based, and to which they can be mapped. The broad framework puts assessment in context and provides the focus for selecting relevant aspects of functioning and disability for assessment. The credo of the ICF is: "From body functions arise activities" [19]. Since the value of a body function is derived from associated activities, this study will conduct an extra choice experiment on the activities to subsequently link this to the functions and measure their value.

The ICF is shown to be an essential tool for identifying and measuring efficacy and effectiveness of rehabilitation services, both through functional profiling and intervention targeting [20]. However, all functions and activities within the classification are equally important.

2. Objectives

It is unclear whether patients consider all ICF factors collectively or prioritize certain factors in deciding rehabilitation treatments. Assessing key clinical factors and related activities from both patients and public perspectives is crucial for guiding therapists and shaping post-stroke rehabilitation programs. The effectiveness of interventions relies on clinical outcomes, patient self-assessment, and perceived effort, yet there is limited empirical evidence on which neurorehabilitation attributes patients consider important [21].

This study aims to understand the trade-offs stroke patients are willing to make for neurorehabilitation and identify pivotal attributes influencing their treatment adherence. It will explore heterogenous preferences among patients and use personalized information for a person-centered approach.

Understanding how patients and citizens value different body functions in rehabilitation decision-making is vital [8]. The research aims to identify critical clinical factors and their significance in decision-making for robot-assisted interventions, incorporating perspectives from both patients and the general population.

3. Descriptive framework

Preference evidence is a series of human behaviors. Different elicitation tasks can be used to observe behavior. Preference elicitation involves setting up tasks that confront subject with assigned objects around which their preferences can be expressed and observed.

In this study, we will elicit patients’ willingness to accept tradeoffs among treatment attributes that were derived from the International classification of functioning, disability and health [19].

For the experiment on body functions the basis is ICFs chapter 7 “Neuromusculoskeletal and movement related functions” of the body function section. The first branching level ("second level classifications") categorized in three sub-chapters is used as attributes and encompasses twelve aspects. The general scale of assessment characteristics with negative scale of the ICF are used as levels. The levels reflect the labeling of the extent or magnitude of the impairment. The characteristics "not specified" and "not applicable" of the ICF scale are not considered. This results in three levels (Table 1).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. List of attributes.

https://doi.org/10.1371/journal.pone.0295267.t001

For the second experiment on neurological and perceptual disorders (neglect) the first chapter on “Mental functions” was used to define attributes. However, unlike the body functions not the whole chapter was used but only those aspects related to a neglect following stroke. On the first branching level this includes “Orientation functions” (b114) and “Visuospatial perception” (b1565). Within these, 6 factors are used to describe impairments due to neglect [19].

The third experiment is based in the activities and participation section of the ICF. All nine chapters and its respective first branching levels ("second level classifications") are used as attributes as all factors may influence the decision-making for rehabilitation and care for people experiencing impairments after stroke.

For all three experiments, the categories/criteria titled "otherwise specified" and "unspecified", respectively, were not considered for the decision model because no patient-relevant functions are categorized under these classifications that cannot also be subsumed under the other criteria. The descriptive framework of attributes and levels was developed with the MECE principles in mind in order to encompass all aspect while identifying the independent and important variables [22, 23].

After ethical approval, semi-structured interviews were conducted with patients and public representatives to further confirm this list of factors and provide feedback as to what extent the list is understandable and was appropriately re-worded from the clinical wording of the ICF to a patient-friendly expression. Interviewees indicated that the list of attributes had minimal overlap but was understandable. There were only a few minor suggestions to improve comprehensibility.

This validation activity led to minor amendments (e.g., wording) resulting into the final list of thirty-four attributes for the activity experiment (BWS case 1), twelve attributes for the arm paresis experiment (BWS case 2), six attributes for the neglect experiment (BWS case 3) (see Table 1 and Appendix A in S1 File for further description).

4. Methods

Stated preferences [24, 25] are an increasingly accepted approach to compare health technologies. The use of such choice experiments is rapidly increasing in medicine[26, 27], and professional societies have developed guidelines for the use of conjoint analysis in outcomes research [28, 29]. Stated preferences are an increasingly accepted method for generating preference data for ranking specific attributes or technologies within a disease class. Best-Worst Scaling (BWS) is a development of the classical Discrete Choice Experiment (DCE), which is becoming increasingly popular to elicit preferences in healthcare [28]. In contrast to DCE, where respondents choose only the best from a set of options, BWS requires respondents to identify the best and worst options in a choice scenario [30]. The main principle of the approach is that respondents define the extremes of a latent, subjective continuum. Essentially, they are asked to choose the pair that maximizes the value difference (on the scale of latent utility) between them [31]. The BWS method distinguishes three cases: the objective case (Case 1), the profile case (Case 2), and the multiprofile case (Case 3). Case 1 is used when the relative values of a set of objects or items (mostly called attributes) are of main interest [32]. The Case 2 scenario shows a single choice alternative, e.g., rehabilitation, defined by different attribute levels. The respondents are asked to choose the best (most important) and the worst (least important) attribute level of the given alternative. Case 3 is comparable to a classical DCE and provides multiple alternatives. However, next to the most preferred alternative respondents also choose the least preferred alternative. Supposing that there are only three choice alternatives in a choice scenario, a full preference ranking of the given alternatives can be determined [25, 33].

In this study, we will conduct three best–worst scaling (BWS) experiments, specifically one BWS experiment using each case to quantitatively assess the relative importance of body functions as well as impaired activities of daily living due to stroke long-term effects. A BWS object case is used for the activities because the respective activities and participation section of the ICF has no attribute and level structure [32]. These questions suit the purpose of this study as they are robust for scale-related biases, simplify ranking tasks for participants, and effectively discriminate between the ratings of different factors involved in complex decisions [34]. On the contrary, a certain attribute and level structure can be assumed for the body functions and the neurological and perceptual disorders. Hence more complex BWS cases can be applied. However, given the cognitive burden on respondents due to the high number of attributes we decided to use a profile case for the body functions while preferences on neglect will be assesses using a multiprofile case.

5. Data analysis

The models account for the panel structure of BWS data with the same respondent providing multiple outcomes for a sequence of different choice scenarios [40]. Upon reaching the target number of study respondents, analysis files will be constructed and data-quality checks as well as consistency tests on the choice data derived from the BWS questions will be performed. Subsequently, statistical analyses will be carried out on the BWS responses to quantify the relative importance of the study’s included attributes. The analysis will be conducted using Stata 15.1 [41].

The complexity of data-analysis tools, such as the specific logit-based model version, will be determined based on the data’s characteristics. Initial analysis will involve count analyses. The detailed analysis of BWS data will encompass evaluating validity issues among respondents and assessing the impact of these issues on average preference estimates. In this model, the dependent (response) variable will be the discrete treatment choice, while the independent (explanatory) variables include the attributes and levels describing the effects of neurorehabilitation treatment alternatives on boy function and activities.

Results from the logit-based models will constitute the primary outcome for this study, providing estimates of preferences for neurorehabilitation attributes. This will include conditional as well as mixed logit models. Second, an exploratory latent class analysis will be conducted to investigate potential variations in preferences based on patient characteristics, including but not limited to sociodemographic factors, attitudes, and experiences. If statistically significant differences in preferences are found among subgroups, subgroup-specific preference information will be reported [37, 42].

6. Expected results

To our knowledge, this is the first study assessing the relative importance of (sub-)chapters of the ICF against the background of decision-making for neurorehabilitation following stroke. When designing and evaluating rehabilitation care, it is important to understand the preferences of patients on how they value various factors, to be able to optimize patient-centered care [8, 16].

Results are expected by the end of May 2022. The expected results can inform value judgements of clinical, policy and regulatory decision makers in approval, reimbursement, pricing, and utilization decisions in terms of robot assisted neurorehabilitation. Findings will inform about the weighting of two subchapters as well as the activity section of the ICF among patients and public representatives in Germany, confronted with the actual decision or a hypothetical judgment. The research will add to the current literature on clinical-decision-making in stroke rehabilitation. The findings will inform future research and consensus activities to derive a more patient-centered classification of impairments and treatment goals to guide decision-makers in stroke units to prioritize the factors for deciding rehabilitation for stroke patients.

7. Discussion

Clinical studies in neurorehabilitation measure clinical parameters and indicators that are an expression of arm mobility or visuospatial perception. These are the result of certain bodily functions. However, it is unclear whether measurable functions or improvement in function have an impact on the assessable (patient) benefit. If the fulfillment or improvement of functions has effects on patients’ activities and assuming these activities have measurable influences in terms of patients’ (health-related) quality of life raises the question how the value of functions can be measured?

After assessing the quality of clinical evidence, clinicians and decision makers often must make complex decisions about access, reimbursement, and use. These decisions require nonclinical value judgments about whether the benefits of robot-assisted therapies justify the associated expense. Such societal value judgments are traditionally made by experts. However, because patients are the ones who ultimately experience the positive and negative outcomes of treatment, most decision makers agree that decisions related to new therapeutic options should be patient-centered and reflect patient values.

The value of an intervention depends on what clinical effects can be achieved, how patients evaluate them for themselves, and what effort patients associate with an intervention. The acceptance of stroke patients is an essential prerequisite for achieving clinical effects. In addition, it is important to find out which characteristics are an incentive for the patients to participate and which ones may be an effort and thus lead to the rejection of an intervention. This is where health preference research comes in, providing information about patient value judgments relevant to decision-making.

Due to the enormous importance of preferences, methods for their measurement are increasingly used in health services research and in health economics. However, in the search for the best possible solution, decision-makers almost always must weigh several attributes of an alternative against each other, which requires so-called multi-attribute evaluation methods.

The BWS method used has several strengths. The main strength of BWS is the better information yield by inducing respondents to make two statements rather than just one. The preference structure of the respondents can thus be determined more precisely or with equal precision, but with a smaller sample size. BWS presents an alternative preference elicitation method that can resolve some of the limitations of the traditional DCE technique and especially manage a high number of attributes [43]. In addition, repeated BWS questions allow to derive complete rather than partial rankings by gradually excluding alternatives previously identified as "best" or "worst." Reduced cognitive load on subjects is cited as another advantage of BWS [34]. It seems to be easier for the subjects to determine the two respective extreme points on their utility scale than to select the most preferred alternative between two or more alternatives with many attributes in complex decision scenarios or even to determine a complete ranking [44]. In addition to advantages, BWS also has weaknesses, which ultimately stem from the simple fact that even in experiments additional information can only be had at increased cost. Thus, for the subjects, the time required for the choice decision increases [45, 46].

8. Limitations

This study has potential limitations. First, there is a certain risk of an interviewer bias within the results of the pretest-interviews. Since the participants were interviewed in person and/or virtually, socially desirable statements cannot be avoided. In addition, members of the research consortium were always present in the room with the patients if the interviews were conducted online by the responsible researcher. Even if these members are not directly involved in these preference studies but only the clinical study, influences on the participants’ answers cannot be eliminated.

Second, as the factors and attributes used in the BWS experiments cover only the chapters of the ICF related to arm paresis, neglect, and activities of daily living. Hence, the findings of this study may not be generalizable to other chapters of the ICF or the ICF in general. Third, there is a potential risk for certain groups of patients and or the public to be under- or over-represented in the survey due to actual non-response. Additionally, the number of participants may not be equally distributed among the three experiments. Hence, the overall ranking of factors may over-represent specific subgroups. However, latent class analyses will outline to what extent there are group differences, and heterogeneities [42].

Fourth, the recruitment of participants uses a vendor with an online panel. It could be argued that this leads to a certain selectin bias as only this with internet access may participate. Moreover panelists might more likely belong to a homogenous educational class and it has to be mentioned that they earn money via panels when completing numerous surveys [47]. Fifth, given the severity of the disease and the potential impairments following a stroke, possible barriers to participation due to physical conditions may apply. However, persons with certain impairments (e.g., visual impairments) are not excluded. The vendor that will recruit patients /public is open to everyone. Hence, it is assumed that participants with impairments have the necessary technical equipment to enable them to participate in the online study (e.g., reading software, magnification) or have the assistance of caregivers in answering the questions. Sixth, participants may not be familiar with the BWS process since it is a relatively new elicitation method, which could lead to non-response response bias due to dropping out of the survey. Seventh, this study aims to understand the preferences of patient and public representatives confronted with decision-making for robot-assisted neurorehabilitation after stroke. However, considering shared decision-making, future studies may analyze preferences from the clinician perspective.

Finally, the survey is being conducted during the Covid 19 pandemic. It is entirely possible that the results would have been different if the survey had been conducted before or after the outbreak and that patient preferences for robotic and/or non-personal therapies are significantly higher during the pandemic than without pandemic. With the coronavirus outbreak and the concomitant increase in digital counseling and care, a differential view and choice of individuals is also conceivable or even assumed also for robotic assistance. Avoiding the risk of infection through contact in waiting rooms, therapists’ offices or public transport could be crucial in this regard. However, healthcare also thrives on the personal caregiver-patient relationship. Moreover, the discussion about data protection could also be perceived negatively and increase the unpleasant feeling of being constantly monitored. Another study under different circumstances could shed light on this. Finally, the pandemic could function as a digital health accelerator for changing perceptions and acceptance of digital or robotic health solutions.

Supporting information