Patients’ priorities in selecting chronic lymphocytic leukemia treatments

In the United States, 130 000 people live with chronic lymphocytic leukemia (CLL).¹  The American Cancer Society estimates that, in 2017, there will be 20 110 new cases of CLL in the United States, which represents approximately one-third of all leukemia diagnoses.²  Treatment options for patients have increased, with 4 new treatments approved since 2013 (obinutuzumab, ibrutinib, idelalisib, and venetoclax) and more under development.^3,4  Because many patients relapse after treatment, patients may receive a number of different treatments.

Treatments available for CLL vary in efficacy, safety, routes of administration, and cost. With many available choices, patients’ preferences in treatment selection are becoming increasingly important. With the advent of value-based treatment decision-making like the American Society of Clinical Oncology value framework, which scores efficacy, safety and costs, the patients’ perspectives and preferences are often missing in the scoring algorithm.⁵  Understanding these perspectives and preferences is critical if we want to truly consider what patients value in the delivery of cancer care. Nonetheless, there is little systematic research in this area.

This study helps to fill the knowledge gap on patient preferences by conducting a relatively large-scale study of treatment preferences among patients with CLL in the United States using a discrete-choice experience (DCE) survey. The study includes a novel way to explore the impact of out-of-pocket cost on treatment choice. The study also investigates patient interest in more sensitive tests for minimal residual disease (MRD) or small numbers of cancer cells remaining after treatment.

We designed a patient preference survey that included a set of DCE questions that offer a choice between pairs of hypothetical treatments for CLL. DCE questions reflect the assumption that goods can be described by a set of attributes with varying levels and that individuals have preferences for certain combinations of attribute levels.⁶  We conducted 2 patient focus groups in February and June 2015 to learn about treatment attributes of interest to patients. Candidate attributes were developed based on the patient input, clinical input from experts in CLL, and data on existing CLL treatments. The final attributes and levels were selected at a meeting including representatives from the Leukemia & Lymphoma Society (LLS) and the Lymphoma Research Foundation.

Table 1 summarizes the final set of 5 attributes, which include progression-free survival (PFS), mode of administration and schedule, chance of diarrhea, chance of serious infection, and chance of organ damage (tumor lysis syndrome). The attribute levels were designed to cover most existing treatments as of the fall of 2015. The 2 modes of administration were assigned different durations: the IV mode was fixed at 6 months, and the pills were taken every day until the medicine stopped working (the length of PFS for that hypothetical treatment). Respondents were told they would experience the side effects and elevated risks (diarrhea, infection, and organ damage) as long as they were taking the medicine. Figure 1A presents a sample choice question from the survey.

The online survey instrument started with screening questions and informed consent. After providing consent, the respondents were asked background questions about their disease and treatment histories. Each DCE attribute was described individually using the text in Table 1, followed by questions about the attribute. The risk-based attributes (chance of serious infection and chance of organ damage) were presented both numerically and graphically. Risk levels were presented graphically using risk grids, each including 100 human figures, in which each figure represents 1 person who takes the medicine and figures in color represent patients for whom the event occurred.⁷  Respondents were asked a comprehension question about the risk grid and presented with the correct answer to reinforce the information. The survey text was pretested in semistructured, face-to-face interviews with 15 patients prior to survey administration.

Each respondent answered 8 DCE questions, choosing between pairs of experimentally designed hypothetical CLL treatments, each a combination of the 5 attributes shown in Table 1 with varying levels. Following good research practices in conjoint analysis and DCE,⁸  we created a set of 48 DCE questions, known as the experimental design, using the SAS implementation of a common D-optimal design.^9,10  The 48 DCE questions were divided into 6 blocks of 8 questions. Respondents were randomly assigned to one of these blocks, and the order of the DCE questions within a block was randomized to control for question order effects.

There are significant cost differences in treatments for CLL, which can translate to variation in out-of-pocket costs depending on a patient’s insurance coverage. To explore the impact of cost on preferences, we added a question after the DCE questions in which respondents were again asked to choose between medicine A and medicine B, only this time with out-of-pocket cost included. We created 2 versions of the cost question, and respondents were randomly assigned to 1 of the 2 versions. The noncost attributes for medicine A and medicine B were fixed at the attribute levels presented in Figure 1A. In the first version, medicine A was $200 per month and medicine B was $600 per month. In the second version, medicine A was $25 per month and medicine B was $100 per month. The 2 versions are presented in Figure 1B-C.

Detection of MRD can help inform patient and physician decisions about frequency of monitoring and potential length of remission. After the cost questions, we assessed respondent interest in more sensitive blood and bone marrow tests of MRD using a 5-point Likert scale (see Figure 5 for the questions and response categories). Before the questions on interest in blood or bone marrow tests of MRD, we provided a brief description of MRD and MRD testing (see the supplemental Appendix for the text used to describe the tests). For the bone marrow test, we explained that physicians are interested in whether there are cancer cells in both the blood and the bone marrow, because bone marrow disease is a major concern in patients with CLL. We described the process of getting a bone marrow aspiration and noted that the test can be painful.

A convenience sample was recruited through the LLS during March and April 2016. The LLS sent e-mails to potential respondents in their database of patients. The e-mails contained an invitation to the survey and a unique link to the survey. Respondents were offered a $20 incentive to complete the survey. Adults aged ≥18 years with a self-reported physician diagnosis of CLL and who were able to provide informed consent were eligible to participate in the online survey.

The study was approved by an institutional review board at RTI International.

The DCE questions generate panel data, which were estimated using a main effects random-parameters logit (RPL) model with NLOGIT Software version 5.0 (Econometric Software, Inc., Plainview, NY).¹¹  RPL accounts for differences in preferences across respondents that can bias results from conventional conditional logit models.¹²  Parameter estimates from the model can be interpreted as relative preference weights that indicate the average relative preference for one attribute level compared with other attribute levels. The mean preference weights were used to calculate the relative importance of each attribute and to calculate the minimum acceptable benefit in terms of the months of PFS respondents required in order to accept a worsening of the other attributes.

In order to assess whether patients’ treatment status influenced their response, we performed subgroup analysis, dividing the sample into 3 categories: first-line patients, relapsed/refractory patients, and treatment-naive patients. The first-line category included respondents who were receiving or had completed a first-line oral or IV treatment at the time of the survey. The relapse category included respondents who reported having been treated more than once with oral or IV treatment and who reported that their condition worsened after experiencing a complete or partial remission. The treatment-naive category included respondents who had not received an oral or IV treatment of CLL.

The percentage of respondents who selected medicine A and medicine B in the 2 versions of the cost question was compared with an estimate of the percentage of respondents who would have selected each medicine based on a prediction using the coefficients from the RPL model, where cost was not included. Posterior preferences for each individual in the sample were computed, conditional on the pattern of observed choices.^12-14 

The percentage of respondents selecting each point on the 5-point scale was tabulated. The 2 questions regarding interest in a blood or bone marrow MRD test were analyzed using an ordered logistic regression model.^15,16  The dependent variable was a categorical variable ranging from 1 to 5 indicating the respondent’s level of interest in the blood test. A 5 indicated high interest, and a 1 indicated no interest. The independent variables were age, gender, education, geographic residence, employment, race, type of health insurance, treatment line, years since diagnosis, and whether a respondent has had a bone marrow test in the past (for the bone marrow question).

Of 4420 people in the LLS patient database who were sent an invitation e-mail by LLS, 610 accessed the survey, 432 met the eligibility criteria, and 384 provided consent and completed the survey. Table 2 presents select demographic and treatment experience summary statistics. The age, race, and gender of the sample are similar to what is known about the population of patients with CLL.¹⁷  The average age of the sample was 65 years, approximately half were male, and the majority were white.

Sixty-nine percent of respondents had a form of public insurance (Medicare, Medicaid, and/or Veterans Health Insurance), and approximately half of the respondents were retired. Fifty-three percent of respondents had received financial aid from patient support programs through their doctor, hospital, or drug maker or from a patient organization for their out-of-pocket treatment costs. Forty percent of patients who had to pay something out of pocket for their medicines reported difficulty in paying out-of-pocket costs. Many respondents (26%) stated they were either participating in a clinical trial for CLL or had participated in one in the past.

Respondents placed the most importance on a change in PFS from 10 months to 60 months, followed by a change in the chance of infection from a 30% risk to no risk, a change in the chance of organ damage from an 8% risk to no risk, a change in diarrhea from severe to none, and a change in mode of administration from IV to pill. Figure 2 presents the normalized preference weights for each attribute level. The preference weights indicate the relative strength of preference for each attribute, where larger positive numbers indicate greater preference and smaller, negative numbers indicate less preference. All the levels within each attribute are statistically different from each other at the 5% level. The results followed the expected ordering, with stronger preferences for better levels of the attributes. The vertical distance between the most-preferred and least-preferred level within each attribute represents the degree of importance of that attribute within the ranges presented for the survey.

In the subgroup analysis of differences among first-line patients, relapsed/refractory patients, and treatment-naive patients, there were no statistically significant differences between the preferences of any subgroup and the full sample, or across subgroups.

Figure 3 provides the results from the minimum acceptable benefit calculation. The graph shows the additional months of PFS necessary for the average respondent to accept a worsening in the other attributes. Moving from the lowest risk of serious infection to the highest generated the largest minimum acceptable benefit. On average, 36 additional months of PFS would compensate respondents for an increase in the risk of serious infection from 0% to 30%.

Figure 4 presents the results from the cost sensitivity assessment. Using results from the DCE in absence of cost, it was forecasted that 91% of respondents would choose the medicine with the longest PFS, medicine B. However, when respondents were presented with a choice between the 2 medicines and their out-of-pocket cost, 75% of respondents who saw Figure 1B ($400 difference in out-of-pocket cost per month) chose the lower-cost medicine, and 50% of the respondents who saw Figure 1C ($75 difference in out-of-pocket cost per month) chose the lower-cost medicine. The difference between respondents’ forecasted choice based on the DCE results and their actual choice was most pronounced for respondents who were assigned to the version with the largest difference in cost between the 2 medicines (Figure 1B).

Approximately 82% and 58% of respondents were interested or very interested in the blood test or bone marrow test for MRD, respectively (Figure 5). The results of the ordered logistic regression for the blood test show that interest in a blood test for MRD decreased with age, was lower for women, and was higher for respondents with public insurance. Interest in the bone marrow test for MRD was also lower for women but higher for respondents living in rural areas, those who were employed part-time, those who were black, and those who had previously had a bone marrow test (results in supplemental Tables 1 and 2; supplemental Appendix).

This study investigated patient preferences for different, hypothetical CLL treatments; the relationship between treatment preferences and patient out-of-pocket cost; and patient interest in more sensitive tests for MRD. The study included a relatively large sample of patients with CLL, including first-line patients, relapsed/refractory patients, and treatment-naive patients. The respondents valued efficacy, but they also valued safety. Any increase in adverse events or the risk of adverse events would require a potentially large improvement in efficacy to offset the disutility to respondents associated with the adverse events. For simplicity, the chance of organ damage (tumor lysis syndrome) is presented as constant for the duration of treatment. This overstates the actual risk, which is highest at the start of treatment.^18,19  Cost had a big impact on treatment choice.

Evidence suggests that patients prefer oral over IV mode of administration in oncology.²⁰  In the context of the treatment attributes included in this survey, respondents were generally unwilling to trade off more convenient dosing for lower efficacy and risk attributes.

We did not find any statistically significant differences among the preferences of first-line, relapsed/refractory, and never-treated respondents. This suggests that patient preferences may be stable over the stages of the disease; however, it is possible that sample size was not large enough to estimate the difference in preferences across the groups.

The impact of out-of-pocket costs for care to patients has become an important issue in cancer care, and studies have found that higher out-of-pocket costs may be associated with noncompliance.^21-23  In our study, more than half the respondents had received financial aid from patient support programs, and almost half of patients who had to pay something out of pocket for their treatments reported difficulty in paying out-of-pocket costs.

Including costs in a DCE when the cost levels are considered too high for some fraction of respondents to pay presents a challenge to studies examining the impact of cost on treatment choice. If too many respondents select a hypothetical treatment based only on cost in the DCE questions, nothing is learned about relative preferences for the other attributes. When respondents ignore the cost attribute in the DCE because they cannot afford either option, it reduces the precision of all the estimates. We used a novel approach to collecting information on the impact of cost on preferences, adding a fixed follow-up question that contained the same attributes as the DCE and added cost. The addition of cost to the treatment profiles resulted in a large swing in the predicted choices toward the treatments with lower out-of-pocket costs. These results give a strong indication of cost sensitivity in patients with CLL.

Detection of MRD can help inform patient and physician decisions about the frequency of monitoring and potential length of remission. The value of MRD as a prognostic factor in CLL has been demonstrated by several studies, including randomized trials,^24-26  but very little is known about patients’ interest in knowing their MRD status. In this study, a significant number of patients indicated interest in a test of their MRD status, especially if the sample was taken from the peripheral blood rather than by bone marrow aspiration. If further studies establish the usefulness of MRD testing for treatment decisions, our study suggests that patients may be interested.

The only other DCE study to investigate treatment preferences for CLL using a DCE was that conducted by Landfeldt et al.²⁷  The study included physicians, the general population, and a small sample of relapsed/refractory patients in Germany and Sweden. Among 6 attributes (overall survival, PFS, fatigue, nausea, risk of serious infections, and treatment administration), overall survival was the most important attribute for all 3 groups, and treatment administration was more important to patients than it was to physicians. We selected PFS as our measure of efficacy because patients often receive a series of treatments followed by remission and relapse and there is more consistent clinical data on PFS across existing treatments. Although in the larger picture, patients may care more about overall survival, patients with CLL often take a series of treatments, and it can be hard to tie one particular treatment to a change in overall survival. During the focus groups, patients talked more about remission than overall survival, and in the pretests, no patients asked about overall survival, so the use of PFS as an efficacy end point appeared to be acceptable to patients.

The study results should be viewed in light of several limitations. The respondents were recruited through a patient advocacy organization, and this population may represent a subgroup of the overall CLL patient. The DCE method relies on hypothetical scenarios, and only a limited number of treatment attributes can be included. Although the attributes and choices reflected in the survey are similar to choices that patients face, the answers do not carry the weight of real choices. In addition, the results are conditional on the set of attributes included and the range of levels. If different attributes or ranges had been considered, the relative preferences for any individual attribute might shift.

The study demonstrates the importance of efficacy to respondents and a willingness to trade off efficacy to avoid serious risks. Cost considerations drove a large shift in preferences, and patients expressed interest in MRD testing. Combined with the responses to other questions indicating difficulty paying for current treatments, cost is an important factor in patient decisions. Patients with CLL have a number of options for treating their disease. These results could help patients play a more active role in discussions about treatments and help physicians work with their patients to find treatments that best fit patients’ preferences and circumstances.

Kimberly Moon (RTI Health Solutions) was responsible for overall project management for this study. Alicia Patten (LLS) provided assistance with patient recruitment. Editorial assistance during preparation of this manuscript was provided by Joyce Hicks (RTI Health Solutions) and funded by Genentech, Inc.

This study was supported by research funding from Genentech, Inc., to RTI Health Solutions.

Contribution: C.M., A.M., E.W., M.G., J.W., J.L., and C.R. contributed to the conception and design of the work; C.M., J.S., and M.B. conducted the analysis; and all authors were involved in the acquisition of data, interpretation of results, drafting the work, and revising it critically for important intellectual content.

Conflict-of-interest disclosure: C.M., M.B., and J.S. are full-time employees of RTI Health Solutions and were paid contractors of Genentech, Inc., in the development of the survey, the analysis and interpretation of the data, and preparation of the study report. A.M., J.L., and C.R. are employees of Genentech, Inc. The remaining authors declare no competing financial interests.

Correspondence: Carol Mansfield, RTI Health Solutions, 200 Park Offices Dr, PO Box 12194, Research Triangle Park, NC 27709-2194; e-mail: carolm@rti.org.