Study design
This study was designed in two phases: Phase I was a qualitative study to identify important aspects of SAA treatment and to develop the survey questionnaire and Phase II involved the conduct of a one-time, non-interventional anonymous online survey study among adult patients with SAA to elicit patient preferences for attributes associated with treatment of SAA, including transfusion independence. All study materials were approved by the Ethical and Independent Review Services (E&I) Institutional Review Board (E&I study number 15128–01). No compensation was offered and informed consent was obtained from all patients who participated in the survey study.
Phase II included an interactive discrete choice experiment (DCE). The DCE was selected because it is a choice-based survey research methodology that captures patient’s value for specific characteristics of treatment and patient’s willingness to accept tradeoffs between characteristics [12, 23]. It is an established methodology for eliciting patient preferences in regards to treatment characteristics [17, 25, 34]. A patient survey using discrete choice experiment design presents patients with a series of hypothetical choice scenarios between two treatments, characterized by the same set of attributes but where one or more of the attributes have different levels, and requests patients to make a choice between the two options. By presenting patients with a DCE, the study is able to elicit patient preferences without directly asking the patient about their preference, which is indicative of patients’ actual choice regarding potential treatment [9, 23].
The survey questionnaire was developed based on findings from Phase I of this study, which included seven one-time, one-hour long, phone-based focus group discussions (Phase I focus groups). A literature review was performed to assist with the development of the moderator’s guide for the Phase I focus group discussions. Two separate moderator’s guides were developed – one for the clinicians and one for the patients. A list of potential factors that may influence a patient’s value of transfusion independence was summarized from the literature review. Participants were recruited through the physician referrals in the United States and France. The Phase I focus groups aimed to identify and refine a list of attributes that were important to the management and treatment of SAA. Focus groups in the United States and France were conducted separately with physicians (N = 3), a nurse (N = 1), and patients (N = 9) to discuss the illness experience and the treatment and management of SAA, followed by discussion of the value of being transfusion-independent from the patient perspective. Providers and patients gave feedback on the comprehensiveness of the attributes. The most meaningful treatment characteristics to patients were discussed, including frequency of transfusions, symptoms, risks, convenience, and costs. The attributes were rated in terms of importance on a scale from 1 (least important) to 10 (most important) from the perspective of a patient with severe aplastic anemia who is considering treatment/management options. They were asked to focus on the importance of each attribute prior to transfusion independence. The self-rating of each attribute was transformed into a rank based on relative importance and the ratings were reviewed with each respondent. Respondents were given an opportunity to reconsider their ranking and then confirm the ranking of each attribute. Providers were asked to rank and then rate the importance of each attribute from a patient perspective, and provided input on the appropriate wording for the levels for each attribute that would constitute meaningful categories.
In Phase II, the online survey consisted of three parts: 1) study qualification questions and questions collecting general patient characteristics and medical condition, 2) questions aiming to assess potential impacts of achieving transfusion independence on different aspects of patients’ well-being relative to being transfusion dependent, and improvement in quality of life with a new therapy using 5-point Likert-scale rating questions, adapted from the “MDS Patient Quality of Life Survey” that was available online and has been published [29], and 3) an interactive discrete choice experiment (DCE) designed to capture patients’ willingness to accept tradeoffs between hypothetical treatment characteristics. The online survey was designed to be completed in English or French. The survey was designed in English and translated into French based on the efforts of three professional linguists, each chosen based on their knowledge of the language, the subject area, and the target audience. Once the translation was completed, an editor carefully reviewed and refined the document to ensure that it read as if it were originally written in French. A proofreader then provided quality control, checking the translation against the original to verify that it appropriately represented the source.
The key attributes and levels for the DCE experiment, selected based on the findings from the Phase I focus groups, included the following: risk of serious bleeding (low: platelet count > 50,000, moderate: platelet count 10,000–50,000, high: platelet count < 10,000), risk of infection (low: ANC > 1000 cells/mm3, moderate: ANC 500–1000 cells/mm3, high: ANC < 500 cells/mm3), fatigue (none, moderate, severe), and frequency of transfusion (none, 4 per month, 8 per month). These attributes were recommended based on the independence of the attributes relevant to the management of SAA (relative to other attributes initially considered such as convenience, which converged as a concept with frequency of transfusion), importance as rated by clinicians and patients and for their consistent relevance in focus groups with clinicians and patients from the US and France.
In the questionnaire, patients were asked to choose between pairs of hypothetical treatments characterized by a common set of attributes in order to estimate values for the attributes. To reduce the burden on the patient in responding to multiple hypothetical treatment comparisons, each patient reviewed 12 comparison cards. An opt-out option was not included, as receiving no treatment does not mirror the real-life experience of patients with SAA. The DCE design was generated using Sawtooth Software’s Balanced Overlap method [26]. This method minimizes the frequency respondents will have to compare the same levels throughout the tasks, but allows that in a given task both treatment options can compare the same level for an attribute, which results in more plausible choice options. All presented treatment profiles with different levels of attributes are plausible (e.g., low level of transfusions and low levels of symptoms/risks, high levels of transfusions and low levels of symptoms/risks, etc.). Versions of the design that included a dominant comparison, where one treatment was better in all attributes than the alternative, were removed. The final DCE design consisted of 20 versions, each with 12 comparison cards that compared two treatments. An example of the comparison card is shown in Fig. 1.
Data source
Patients who participated in this study were recruited through announcements in the newsletter of the Aplastic Anemia and Myelodysplastic Syndromes International Foundation and referrals from clinical sites in the US and France. Thirty patients qualified for the study and completed the online survey anonymously between February 2, 2016 and April 15, 2016. Patients did not receive a monetary incentive for completion of the survey.
Study population
Patients who completed the survey were screened to meet the following criteria to qualify for this study: age 18 years or older, diagnosed with SAA by a doctor or healthcare provider, prescribed IST with antithymocyte globulin but had an insufficient response, required blood transfusions at least monthly for a period of at least three months over the previous two years, and not currently pregnant.
Outcomes and statistical analysis
Demographic and SAA-related characteristics questions (country, age, gender, race [asked in the USA only], education level, employment status, health insurance coverage, history of treating SAA with drugs, transfusions, and iron chelation therapy) consisting of categorical variables were analyzed using descriptive statistics, and the number and proportion of patients were reported. Questions in which the patient rated their current condition and characteristics related to being transfusion independent were also analyzed using descriptive statistics, and the number and proportion of patients with top-two box (most favorable) Likert-scale ratings were reported.
For the DCE, a conditional logit model was used to estimate part-worth utilities for different attribute levels and assess the relative importance of each attribute [12]. Part-worth utility is a computed preference weight for each attribute and its respective levels. A negative part-worth utility indicates less desired levels of the attributes, whereas a positive part-worth utility indicates more desired levels of the attributes. In our model, attribute levels were effect-coded, whereby omitted categories were coded as − 1, so the part-worth utility estimate of the omitted category for each attribute is the negative sum of the part-worth utility estimates for the included categories. For example, if the treatment alternative had the level of the omitted attribute level (e.g., high risk of serious bleeding) then the dummy variables of the attribute levels included in the model (e.g., low risk of serious bleeding and moderate risk of serious bleeding) had a value of − 1. With effects coding, part-worth utility estimates represent the deviation of the mean of each attribute level from the overall mean/the model intercept. The model assumes that an individual’s utility function, or choice, can be defined by the level’s part-worths, or coefficient estimates [4, 8]. An individual’s probability of choosing a given treatment when considering two treatment options can be expressed as shown in eq. 1 [22].
$$ \mathrm{Probability}\ \mathrm{of}\ {\mathrm{treatment}\ \mathrm{choice}}_{Treat ment\ X}=\frac{e^{\left({\mathrm{Total}\ \mathrm{utility}}_{Treat ment\ X}\right)}}{e^{\left({\mathrm{Total}\ \mathrm{utility}}_{Treat ment\ X}\right)}+{e}^{\left({\mathrm{Total}\ \mathrm{utility}}_{Treat\mathrm{m} ent\ Y}\right)}} $$
(1)
Attribute relative importance and level ranking are made based on comparisons of part-worths [21]. An attribute with a larger difference between its lowest level part-worth and its highest level part-worth is considered to have a greater relative importance in the patient’s choice [4, 21]. Within a given attribute, the relative differences between the levels correspond to the relative respondent preferences. The relative contribution of each treatment attribute to patient treatment preferences was calculated based on the coefficient range for each attribute as a proportion of the summed range of coefficients across all attributes as shown in eq. 2.
$$ {\mathrm{Relative}\ \mathrm{contribution}}_{Attribute}\%=\frac{100\times \left(\mathrm{Maximum}\ \mathrm{part}-{\mathrm{worth}}_{Attribute}-\mathrm{Maximum}\ \mathrm{part}-{\mathrm{worth}}_{Attribute}\right)}{\sum \left(\mathrm{Maximum}\ \mathrm{part}-{\mathrm{worth}}_{Attribute}-\mathrm{Maximum}\ \mathrm{part}-{\mathrm{worth}}_{Attribute}\right)} $$
(2)
Predicted utility scores associated with different levels of transfusion frequency were also estimated to understand the effect of transfusion frequency on choice. The predicted utility scores with different levels of transfusion frequency were estimated using the parameter estimates from the conditional logit model and out of sample prediction of utility scores assuming scenarios where all treatment alternatives required 0, 4, and 8 transfusions per month, respectively.
All analyses were performed using SAS 9.4. DCE analyses were validated using Stata 14.1.