Clinicians' preferences for managing aneurysmal subarachnoid hemorrhage using endothelin receptor antagonists

Background The endothelin receptor antagonist (ERA) clazosentan is being investigated for the medical prevention of cerebral vasospasm and associated complications, such as delayed cerebral ischemia (DCI), after aneurysmal subarachnoid hemorrhage (aSAH). This study quantified how clinicians weigh the benefits and risks of ERAs for DCI prevention to better understand their treatment needs and expectations. Methods An online choice experiment was conducted to elicit preferences of neurologists, intensivists, and neurosurgeons treating aSAH in the US and UK for the use of ERAs. The design of the choice experiment was informed by a feasibility assessment (N = 100), one-on-one interviews with clinicians (N = 10), a qualitative pilot (N = 13), and a quantitative pilot (N = 50). Selected treatment attributes included in the choice experiment were: one benefit (likelihood of DCI); and three risks (lung complications, hypotension, and anemia). In the choice experiment, clinicians repeatedly chose best and worst treatment options based on a scenario of a patient being treated in the ICU after aneurism repair. A correlated mixed logit model determined the relative attribute importance (RAI) and associated highest density interval (HDI) as well as acceptable benefit-risk trade-offs. Results The final choice experiment was completed by 350 clinicians (116 neurologists, 129 intensivists/intensive care clinicians, and 105 neurosurgeons; mean age, 47.4 years). Reducing the likelihood of DCI (RAI = 56.5% [HDI, 53.6–59.5%]) had the largest impact on clinicians' treatment choices, followed by avoiding the risks of lung complications (RAI = 29.6% [HDI, 27.1–32.3%]), hypotension (RAI = 9.2% [HDI, 7.5–10.8%]), and anemia (RAI = 4.7% [HDI, 3.7–5.8%]). Clinicians expected the likelihood of DCI to decrease by ≥8.1% for a 20% increase in the risk of lung complications, ≥2.4% for a 20% increase in the risk of hypotension, and ≥1.2% for a 20% increase in the risk of anemia. Conclusions Clinicians were willing to accept certain increased risks of adverse events for a reduced risk of DCI after aSAH. The likelihood of DCI occurring after aSAH can therefore be considered a clinically relevant endpoint in aSAH treatment development. Thus, evaluations of ERAs might focus on whether improvements (i.e., reductions) in the likelihood of DCI justify the risks of adverse events.


Qualitative Interviews
Ten one-on-one qualitative interviews with clinicians were conducted in May 2020 to test if the key outcomes from CONCIOUS-1 and CONCIOUS-2 trials presented were clinician-relevant and provided enough information for making treatment decisions. The qualitative interviews also aimed to provide rich qualitative data that could be used to contextualize the quantitative preference findings. Furthermore, clinicians' maximum acceptable risk of delayed cerebral ischemia (DCI), hypotension, lung complications, and anemia were elicited during the qualitative interviews using a simple thresholding technique. During the thresholding, the levels of the key attribute were increased to elicit clinicians' maximum acceptable attribute level. The results were used to determine preference-relevant risk ranges to be covered by attributes in the subsequent preference instruments. 1 The interviews followed a semi-structured interview guide that consisted of three parts: • Part 1 included open-ended questions about clinicians' medical background and experience in treating patients with acute subarachnoid hemorrhage (aSAH).
• Part 2 contained a semi-structured discussion about the risks and benefits of treatments for cerebral vasospasm after an aSAH, and maximum acceptable risks were elicited using thresholding analysis.
• Part 3 focused on discussing a preliminary question in the choice experiment to understand clinicians' perception of relevant benefits and risks, as well as their willingness to make trade-offs within a hypothetical choice context.
At the end of the interview, all participating clinicians completed a short online survey to collect their sociodemographic and clinical practice characteristics. The interview guide was reviewed by two experienced neurologists with expertise in aSAH and was modified during the data collection phase in response to clinicians' feedback.
All audio recordings of the interviews were transcribed verbatim, and the interview transcripts were analyzed using a combination of content and thematic analysis. Coding of the transcripts involved thoroughly reading through and analyzing each sentence in each transcript. An initial coding dictionary was developed based on the interview notes. The coding dictionary consisted of a list of codes with each representing a topic, opinion, or other data point. The coding dictionary was adjusted after a pilot during which two analysts coded the same transcripts and discussed the outcomes in a workshop to resolve disagreement by discussion. The qualitative analysis of the transcripts took a concept elicitation perspective, which means that the coding dictionaries were developed iteratively with regular workshops involving data analysts and senior researchers of the project. Emerging codes were grouped together into themes and subthemes. Two (20%) of the transcripts were double coded for quality assurance purposes, with disagreements being resolved during a discussion with both analysts.
Ten clinicians completed the interviews, of which five were from the UK (50%) and five were from the US (50%). The clinician sample included two (20%) neurologists (one from the US and UK each), three (30%) neurosurgeons (one from the UK and two from the US), and five (50%) intensivists (three from the UK and two from the US). All clinicians had been practicing medicine for at least five years and most of them (n=6; 60%) were occupied with their role for approximately 10 to 20 years. Eighty percent (n=8) had treated more than 20 patients with aSAH during the last 10 years.
When asked about the need of a new pharmacological treatment that would reduce the likelihood of clinical deterioration due to DCI following aSAH using a 1-10 range, 80% (n=8) reported a high need with scoring results of either 9 or 10.
The findings from the qualitative analysis were visually summarized in a conceptual map, highlighting the initial management as well as DCI and cerebral vasospasm as themes emerging from the analysis (below).

Conceptual Map
Abbreviations: CV, cerebral vasospasm; DCI, delayed cerebral ischemia; ICU, intensive care unit; Triple-H, induced hypervolemia, hemodilution, and hypertension therapy Overall, there was a variability in practice in the initial management of aSAH in terms of clinical severity assessment, ICU admission, and medical management after aneurysm repair. However, despite the differences in practice, most clinicians (n=7) reported the use of the Glasgow Coma Scale as an initial measure of consciousness that offers an immediate indication of severity in patients presenting with aSAH. Some clinicians also used other validated scales to determine the initial clinical severity, including the Hunt and Hess (n=6) and Fisher (n=4) scales. 2 Upon diagnosis of aSAH, some clinicians (n=3) reported guaranteed admission of patients to the ICU or high-dependency unit, while others would only admit patients to the ICU or highdependency unit depending on their severity and/or after an aneurysm repair (n=6). Medical management post-aneurysm repair varied across clinicians. Some reported the use of antiepileptics to prevent seizures (n=2) and heparin prophylaxis (n=6), while others emphasized more general monitoring of vital signs, glucose (n=4) and blood pressure (n=5). Across the interviews, all clinicians reported following some guidelines on the management of aSAH. However, they recognized that there was no consensus on a universally accepted guideline. Clinicians reported that the most concerning complications following aSAH were re-bleeding, cerebral vasospasm, and DCI. Other complications such as seizure, hydrocephalus and cardiomyopathy were also reported. In terms of the monitoring of the occurrence of cerebral vasospasm and DCI, clinicians reported a combination of neurological and radiological examinations (e.g., transcranial Doppler or computed tomography scan). In addition to the use of the validated scales (e.g., Glasgow Coma Scale, Hunt and Hess scale, Fisher) during neurological examinations, most clinicians (n=6) also indicated the use of their judgment during clinical examinations to recognize changes in the patient's neurological status. All clinicians reported the routine use of nimodipine for the prevention and management of cerebral vasospasm post-aSAH. However, they acknowledged that, at present, there are limited treatment options available, and there is a need for an effective alternative on top of the current standard of care. For a new treatment to be considered effective, clinicians reported that it must be able to reduce the risk of cerebral vasospasm and DCI. On a longer term, some clinicians would like to see improvements in quality of life, such as the ability to conduct activities of daily living upon recovery. The table below summarizes example quotes on the management of cerebral vasospasm and DCI. Overall, most clinicians (n=7; 70%) were willing to accept up to a 20% chance of DCI occurrence with the use of an endothelin receptor antagonist (ERA) compared to a 30% chance of DCI occurrence without the use of ERA. Acceptable risk levels varied widely between clinicians, such that further adjustment after the quantitative pilot was expected. Clinicians emphasized the importance of having a clear description of the severity of the AEs to clearly judge the importance of these events.

Example Quotes on Management of CV and DCI
Finally, all clinicians were presented with a draft choice task. When making choices between two hypothetical alternatives, they considered all the benefit and risk attributes. Clinicians were willing and able to make trade-offs but re-emphasized the need for clearly describing the severity of adverse events. One clinician suggested making the comparison to the 'no treatment' alternative clearer. In response to this suggestion, the 'no treatment' alternative was moved to the left of the choice task.

Design of the Choice Experiment
The experimental design of a choice experiment describes the combinations of attribute levels that make up the hypothetical profiles from which clinicians choose their preferred treatment. Even a small number of attributes and levels results in far too many possible treatment profiles for clinicians to be presented with. Thus, a subset profiles with desirable mathematical properties was selected to be included in the survey according to an experimental design. [3][4][5] These profiles were selected such that all effects of interest could be estimated independently and to ensure clinicians need to make trade-offs when choosing.
The full combination of the attributes and levels would have resulted in more than 30,000 choice tasks (i.e., 256 × (256 -1 ) /2, where 4 4 = 256 is the number of possible attribute combinations). A D-efficient experimental design was used to identify the subset of choice tasks that ensures all effects of interest can be estimated independently. 6 The employed D-efficient design minimized the standard errors of the preference estimates for a given sample size. Directional priors were used to minimize the risk of dominant choice tasks. The generated design had 28 choice tasks and was split into two blocks of 14 questions to ensure that the same amount of statistical information for the different preference parameters would be collected. The design was generated using the software Ngene ® . 7 Within the choice experiment, every choice task asked clinicians to choose the best and the worst treatment out of three unlabeled hypothetical options: 1) ERA A; 2) ERA B; 3) ERA C. This choice format was chosen to maximize the amount of information collected on clinicians' preference structure. Participants were prevented from choosing the same alternative as best and worst. The experimental design systematically varied the attribute levels to ensure clinicians made trade-offs when choosing their preferred alternative. The order of the experimental choice tasks as well as the presentation order of ERA A, ERA B and ERA C was randomized between clinicians. In addition, clinicians were randomized to one of two attribute presentation orders 8 : • Order 1 Likelihood of DCI presented before randomly arranged risks • Order 2 Likelihood of DCI presented after randomly arranged risks A practice choice task was used to familiarize clinicians with the format of the choice experiment (Task #0). This was followed by the 14 experimental choice tasks (Tasks #1 to #14).

Qualitative Pre-testing
The survey was qualitatively pre-tested in 60-min interviews with a total of 13 clinicians in the UK (n=6) and US (n=7), over two waves. The first wave included 10 participants, and the second wave included three participants. Interviews were conducted using a 'think aloud' approach, in which the clinicians completed the online survey while sharing their screen, while an experienced interviewer observed their responses and listened to their reasoning. Using a semistructured interview guide, interviewers also probed clinicians on the clarity of survey instructions, their understanding of the survey content, questions and choice tasks, completeness of response options and the relevance of each attribute included in the preference elicitation tasks. Clinicians were also asked if they perceived the choice tasks to be complete or whether any relevant information or concepts were missing.
The interviews were conducted in two waves, to iteratively test adjustments: Wave 1 (N=10): Participants in the first wave of interviews completed a standard choice experiment including choice tasks offering two treatment alternatives (ERA A and ERA B) and an opt-out 'No ERA'. Within each choice task, a single-choice option in which participants were asked to choose their preferred treatment was presented to clinicians.
Clinicians were familiar with all attributes and felt that all were at least somewhat important depending on their relative values. Across several interviews, it was observed that for some attributes it was difficult to collect trade-off data, due to their low relative importance (i.e., risks of hypotension and anemia). Further, most clinicians never selected the opt-out (e.g., No ERA), citing the acute nature of aSAH and DCI for this preference, which raised concerns about the opt-out option being dominated and therefore biasing preference estimates. Based on these results, the survey design was updated: The opt-out was removed, and a third treatment option, 'ERA C', was added to the choice tasks to allow for more trade-off information being collected per choice task.
The choice task design was updated from a single choice design to a best-worst scaling-type 3 design in which participants were asked to select both the best and worst treatments in the tasks. This further allowed for the collection of more trade-off information per choice task.
Levels for both the likelihood of DCI and other treatment-related risks were updated; the number of levels for the likelihood of DCI was updated from 5 to 7, with a 3% difference between levels. Levels were also added to all attributes to incorporate values associated with placebo treatment. This change was used to generate more choice situations in which attributes of lower importance had a meaningful impact on clinicians' choices.
To ensure clinicians traded off between all risk attributes, an overlapped design was employed that forced the likelihood of DCI to take the same value for two alternatives. This again facilitated the collection of trade-off data for attributes of lesser importance.
Minor edits were also made to the descriptive language used in the survey introduction and attribute descriptions to add clarity and improve clinical accuracy.
Wave 2 (N=3): Following updates to the survey design after the first wave of interviews, participants in the second wave completed the revised survey. Clinicians showed a good understanding of the updated instrument and appeared to improve the insights on trade-offs collected in the survey. No further edits were made following completion of the second wave of interviews.

Statistical Analysis
Based on random utility maximization framework, clinicians' choices were modeled as a function of the treatment characteristics (i.e., attribute levels included in the discrete choice experiment) and clinicians' sensitivities to changes in treatment characteristics.
The utility (U) of treatment option (j) for clinician (n) in choice tasks (t) was specified as a fully categorical linear and additive function: Where ANEMIA corresponds to the risk of anemia, DCI to the likelihood of delayed cerebral ischemia, LUNG to the risk of lung complication, and HYPOTENSION to the risk of hypotension. The αA and αB parameters are two constant terms capturing potential ordering effects in clinicians' choices (e.g., systematic tendency to choose option A, everything else being equal). The βs are preference parameters measuring the effect of discrete/categorical changes in the attributes on the probability of preferring the treatment option. For example, β1 captures the effect of decreasing the risk of anemia from the reference level (i.e., 40%) to 10%. Similarly, β2 captures the effect of decreasing the risk of anemia from 40% to 20%. The ε component is an error component, which is typically assumed to be identically and independently distributed (as a type I extreme value), leading thus to use of a multinomial logit model for the analysis of clinicians' choices.
The preferences are allowed to vary within the sample by assuming that they are multivariate normally (MVN) distributed.

~( ; )
Where μ is a vector of mean preferences with 14 elements (one for each preference effect), and Ω is the full covariance matrix with 105 elements (14 diagonal and 91 off-diagonal elements) to be estimated.
This model was estimated within the Bayesian framework and is typically referred as hierarchical Bayesian logit model. This model relies on Monte Carlo Markov Chain simulation procedures to estimate the model parameters. The prior distribution for the mean elements was a diffuse MVN distribution with null mean and large variance. The prior distribution for the covariance elements was an inverted Wishart distribution with K degrees of freedom and parameter KI, where I is the identity matrix and K is the number of preference effects.
The Monte Carlo Markov Chain simulation procedure was then used to update these prior distributions with the choice data: 50,000 draws were used before convergence (i.e., burn-in period), 50,000 draws were used after convergence, and every 25 th draw was retained, leaving thus 2000 effective draws to simulate the posterior distribution.

Introduction
Thank you for agreeing to taking part in this online survey. The survey aims to understand the perspectives and treatment preferences of clinicians who are experienced in treating patients with aneurysmal subarachnoid haemorrhage (aSAH).
There are four sections in this survey, which will take about 30 minutes to complete: Section 1 will ask you questions about the treatment characteristics of an endothelin receptor antagonist (ERA).
Section 2 will show you 14 choice questions. Each question will ask you to choose the best and the worst treatment out of three ERA that are available to you. This is an established method that helps us understand what matters to you when you choose a treatment.
Section 3 will explore how specific aspects of ERA treatment effect your choice of treatment.
Section 4 will ask questions about you and your clinical experience in the management of aSAH.
Your responses will be confidential, and your name will never be connected with any of your answers. All data will be fully anonymised and used for medical research purposes.
Thank you for your contribution! About aSAH and ERA • Cerebral vasospasm and the resulting delayed cerebral ischemia (DCI) typically occur 4 to 14 days after an aneurysmal subarachnoid haemorrhage (aSAH) and are responsible for high morbidity and mortality in patients. Signs of cerebral vasospasm include loss of consciousness, focal numbness, weakness, paralysis, confusion, dizziness, problems speaking, worsening headache, mood changes, and blurred or double vision.
• Treatment options for cerebral vasospasm and resulting DCI are typically limited to hemodynamic therapy, and rescue therapy. Thus, prevention of cerebral vasospasm is an important objective in the management of aSAH.
• Currently, only the calcium channel blocker, nimodipine, is approved for the prevention of DCI post-aSAH. Recently, a new drug class, the endothelin receptor antagonist (ERA), has been shown to prevent or reverse cerebral vasospasm following an episode of aSAH in clinical trials.
• ERAs are administered as a continuous intravenous infusion. The most common side effects are hypotension, lung complications (primarily due to pleural effusion, pulmonary oedema, and pneumonia) and anemia.

Section 1: About ERA treatment characteristics
In this survey, we will ask you to choose between different hypothetical treatment profiles that you would prefer to use to manage cerebral vasospasm and the resulting DCI following an aSAH.
Each of these treatment is described by four attributes: 1) Likelihood of delayed cerebral ischemia (DCI) 2) Risk of hypotension

3) Risk of anemia 4) Risk of lung complications
When you choose your preferred treatment, it is important to consider all of these characteristics and weigh the pros and cons.
We will now introduce each of the treatment characteristics over the next screens.

WEB PAGE BREAK
The efficacy of an ERA can be measured by its ability to prevent the occurrence of clinical deterioration due to delayed cerebral ischemia (DCI) in patients with aSAH.
The likelihood of a patient developing DCI will be illustrated like this: 30 out of 100 patients (30%) The figures in red (30 out of 100 = 30%) represent the number of people who will develop DCI. The figures in grey (70 out of 100 = 70%) represent the number of people who do NOT develop DCI.
The three ERAs (A, B and C) below both reduce the likelihood of DCI occurrence in patients, but to a different degree. Which of the two treatments is more effective, assuming that all other aspects are the same? Although ERA have only a small effect on the systemic circulation, some patients may develop hypotension from the use of ERA due to its vasodilation effect. However, the severity of the hypotension is mild to moderate (in the order of 10% reduction in blood pressure) and can be corrected by vasopressor and fluid therapy. Only few patients treated with ERA discontinued treatment due to hypotension.

Risk of anemia
Anemia is a class effect of ERA and is attributed to plasma volume expansion as a results of fluid retention. It is typically reversible after discontinuation of ERA and does not require blood transfusion.

Risk of lung complications
ERAs are associated with lung complications such as pleural effusions, pulmonary oedema and pneumonia. These lung complications are related to fluid retention, which is a known effect of ERAs. Euvolemia can be used for the management of these lung complications within a typical ICU setting.

WEB PAGE BREAK
If you were able to improve one of the four treatment characteristics as described below. Which one of the following treatment characteristics would you choose to improve?