Patients' preferences for the management of non-metastatic prostate cancer: discrete choice experimentBMJ 2004; 328 doi: http://dx.doi.org/10.1136/bmj.37972.497234.44 (Published 12 February 2004) Cite this as: BMJ 2004;328:382
- Mark Sculpher, professor ()1,
- Stirling Bryan, professor2,
- Pat Fry, research fellow3,
- Patricia de Winter, research nurse3,
- Heather Payne, consultant in clinical oncology4,
- Mark Emberton, senior lecturer3
- 1Centre for Health Economics, University of York, Heslington, York YO10 5DD
- 2Health Economics Facility, Health Services Management Centre, University of Birmingham, Birmingham B15 2RT,
- 3Institute of Urology, Royal Free and University College Medical School, London W1P 7PN
- 4Meyerstein Institute of Oncology, Middlesex Hospital, London W1T 3AA
- Correspondence: M Sculpher
- Accepted 25 November 2003
Objective To establish which attributes of conservative treatments for prostate cancer are most important to men.
Design Discrete choice experiment.
Setting Two London hospitals.
Participants 129 men with non-metastatic prostate cancer, mean age 70 years; 69 of 118 (58%) with T stage 1 or 2 cancer at diagnosis.
Main outcome measures Men's preferences for, and trade-offs between, the attributes of diarrhoea, hot flushes, ability to maintain an erection, breast swelling or tenderness, physical energy, sex drive, life expectancy, and out of pocket expenses.
Results The men's responses to changes in attributes were all statistically significant. When asked to assume a starting life expectancy of five years, the men were willing to make trade-offs between life expectancy and side effects. On average, they were most willing to give up life expectancy to avoid limitations in physical energy (mean three months) and least willing to trade life expectancy to avoid hot flushes (mean 0.6 months to move from a moderate to mild level or from mild to none).
Conclusions Men with prostate cancer are willing to participate in a relatively complex exercise that weighs up the advantages and disadvantages of various conservative treatments for their condition. They were willing to trade off some life expectancy to be relieved of the burden of troublesome side effects such as limitations in physical energy.
Several situations exist where patients face trade-offs between the risks and benefits of alternative therapies.1 The conservative management of men with organ confined or locally advanced prostate cancer is such a situation. Men diagnosed at a stage when radical treatments such as prostatectomy or radiotherapy are inappropriate face several treatment options, including watchful waiting or oral steroidal or non-steroidal antiandrogen monotherapy.2 Many patients choose castration, performed medically rather than surgically. Alternatively, there is the option of antiandrogen treatment combined with medical or surgical castration.
To make an informed choice, men need to be able to weigh up the slight differences in effectiveness of treatment against a spectrum of side effects associated with alternative strategies. For example, non-steroidal antiandrogen monotherapy offers potential advantages over castration for impotence, loss of libido, and hot flushes, but these may be achieved at the cost of an increased risk of gynaecomastia and breast pain.3
Individuals' preferences for alternative treatments need to be considered in the light of the attributes of the treatments. Discrete choice experimentation, an approach for elicitation of preferences, is now being used widely in health care.4 5 This approach identifies the key characteristics of alternative treatments, such as hot flushes, and selects a series of levels for each (for example, absent, mild, moderate). Respondents choose from several options, each of which details a series of attributes at different levels. The relative importance of attributes to individuals and the trade-offs made between them, can be assessed by changing the levels of the attributes and asking participants to make their choice again. Findings on the reliability and validity of discrete choice experimentation in healthcare settings are encouraging.6 7 We used discrete choice experimentation to elicit treatment related preferences in a sample of men with non-metastatic prostate cancer.
Before the main study, we conducted a two phase pilot study in men with prostate cancer. Fourteen men were recruited into the first phase and were interviewed by a trained research nurse. On the basis of these interviews eight attributes were selected for the main study: diarrhoea, hot flushes, ability to maintain an erection, breast swelling or tenderness, physical energy, sex drive, life expectancy, and out of pocket expenses. In the second phase, nine men were asked to complete a questionnaire based discrete choice experiment with these attributes; there was also a brief unstructured interview with a research nurse. Several men did not understand the nature of the exercise and were only able to complete exercise with guidance from the research nurse. We therefore decided to use an interview format for the main study.
The attributes and levels used in the exercise are described on bmj.com. We chose mild and moderate levels only: it was explained to patients that therapy would be changed in a severe event. The mild level included symptoms that would not interfere with work, study, housework, family, or leisure activities and the moderate level included symptoms that would.
A trained research fellow conducted the interviews, during which personal data on the patients were collected (data were also taken from medical records). The men were presented with two treatment options, each containing a set of attributes at specific levels. The interviewer read out the pair wise options and used show cards as prompts to help the men choose the options they preferred (table 1).
We divided the exercise into two parts to avoid overburdening patients with too many attributes for an option. The first part included three unique attributes—the ability to maintain an erection, physical energy, and libido. The second part included three unique attributes—diarrhoea, hot flushes, and breast tenderness. Both parts included two common attributes—life expectancy and a one off out of pocket expense.
The men had to assume a life expectancy of five years, estimated as the average for the sample considering the mean age (70 years) and clinical stage of disease. The clinical stage was typically not detected by screening but rather diagnosed on the basis of problems relating to progression of the cancer. The five year average was justified because the sample included men with T1 or T2 disease with an estimated seven year survival probability of 65% and men with T3 or T4 disease with a five year survival probability of 30%.8 9 The two parts of the exercise each contained eight pair wise options. We prepared eight different versions of the questionnaire, each representing a new experimental design (orthogonal main effects). Each version of the questionnaire presented different levels of the cost attribute to allow a larger number of intervals between cost levels across the survey. Study patients were randomly allocated to one of the questionnaires.
Our study sample was patients with non-metastatic prostate cancer who had or had never received antiandrogen therapy; there were no exclusion criteria. Potential participants were identified from records at the Middlesex Hospital, London. They were contacted in writing to obtain written informed consent. Respondents were asked to make an appointment for an interview at the Middlesex Hospital.
We analysed the discrete choice exercise by taking each choice between pair wise options as a specific observation. Hence each respondent provided a maximum of 16 observations. Given the non-independence of the data provided by the same respondent, a random effects probit model was used. Two separate models were specified (one for each group of attributes), with the choice responses as the binary dependent variable and the difference in levels for each of the attributes as the independent variables (see bmj.com). The specification of the experiment precluded the testing for interactions between attributes, but we explored the interactions between attributes and patient characteristics (age, prostate specific antigen level, and T stage of cancer at diagnosis).
Between 24 May and 8 September 2000, we invited 180 men to participate in our study. Of these, 129 were interviewed. Participants were similar to those who declined for mean age and T stage at diagnosis. Table 2 shows the characteristics of the participants.
Discrete choice experiment
Table 3 shows the results of the first part of the exercise. The coefficients for the unique attributes were all statistically significantly different from 0; negative values for libido, maintaining an erection, and physical energy indicate that the more severe the problems, the less likely the patient is to prefer that scenario; negative values for out of pocket expenses indicate that the higher the costs, the less likely the patient is to prefer that scenario. Positive values for life expectancy indicate that the greater the life expectancy the more likely the patient is to prefer that scenario. The only statistically significant interaction was between ability to maintain an erection and age; the positive value indicates that older men were less likely to be influenced by the ability to maintain an erection in choosing their preferred scenario.
Table 4 shows the results of the second part of the exercise, which also considered the unique attributes of diarrhoea, hot flushes, and breast tenderness. The coefficients for the attributes were all statistically significantly different from zero. Negative values for diarrhoea, hot flushes, and breast tenderness indicate that the more severe the problem the less likely the patient is to prefer that scenario. None of the interaction terms were statistically significantly different from zero.
Table 5 shows the marginal rates of substitution between life expectancy and other attributes—that is, how much life expectancy the men were willing to trade off to achieve an improvement by one level in one of the other attributes. For example, men were willing to trade off 1.8 months of life expectancy to change diarrhoea from a moderate to mild level or from mild to absent. Because the levels of severity differed between attributes, marginal rates of substitution between attributes should be compared with caution. The least important marginal rates of substitution were for hot flushes and the most important were for physical energy.
Men with prostate cancer are willing to participate in the relatively complex exercise of discrete choice experimentation to weigh up the benefits and risks of various conservative treatments, irrespective of the stage of cancer or whether they had received such treatment. To our knowledge, our study is the first to elicit preferences from patients with prostate cancer using discrete choice experimentation, and provides further evidence that this approach can be applied successfully in health care. A novel feature of our study was the use of two groups of attributes. This allowed the choices to be kept relatively simple (maximum of six attributes), and the inclusion of a common core of two attributes (cost and life expectancy) ensured trade-offs across all attributes.
What is already known on this topic
Various factors need to be considered in making treatment decisions in prostate cancer
Patients' views on which factors of treatment are important to them and how they trade-off these factors is under-researched
What this study adds
Men are willing to contemplate trading off life expectancy to be relieved of the burden of side effects such as limitations in physical energy
The preferences of older men are not the same as those of younger men
The men were willing to trade off some life expectancy to be relieved of side effects. Men with metastatic cancer were not included in the study, so no patients faced imminent death, with life expectancy ranging from around 2 to 10 years. Given the difficulty in estimating, and potential ethical problems in presenting, life expectancy for each patient, the men were asked to assume a life expectancy of five years (the average in the group) as a starting point. The size of the trade-offs between life expectancy and the other attributes should be treated with caution because men with a longer or shorter life expectancy than five years may have indicated different preferences if their actual life expectancy had been presented to them.
The results are averaged across the sample and so there is inevitable variation between the men. Our results are therefore no substitute for careful assessment of individual patient preferences in a clinical setting but do provide some basis for clinicians to prioritise issues they discuss with patients. The implications on physical energy from conservative treatment, for example, may be an important starting point for decisions on treatment.
Our findings could be used by clinicians to help patients choose between conservative treatments; knowing about the preferences of other men with prostate cancer might help patients to clarify their own thoughts. Our findings could also help in the design of new studies in prostate cancer.
We looked at the application of discrete choice experiments in prostate cancer only. The ability and willingness of men to engage in this exercise is encouraging for future research. Perhaps the most common therapeutic dilemma that patients and clinicians face is the timing of androgen suppression. Should a patient start therapy early, once progression of prostate cancer has been identified? The potential benefits of this approach might include a slowing down of disease progression and perhaps reducing the likelihood of death related to the cancer. Alternatively, the patient could defer treatment for an agreed time. This would avoid the immediate side effects of treatment and possibly reduce the medium to long term adverse effects. This type of trade-off is made by many patients with progression of disease everyday, and discrete choice experimentation could gain some insight into the way patients make this difficult choice.
We thank Rob Sheldon (Accent Marketing and Research) for help with the design and analysis of the study, Wendy Coucill for her work on the pilot study, and the patients.
Contributors MS and ME initiated the research and designed the study with SB, HP, and PF. PF undertook the interviews. SB and MS undertook the analyses, and MS produced the main drafts. MS and SB will act as guarantors for the paper. They accept full responsibility for the conduct of the study, had access to the data, and controlled the decision to publish.
Funding This study was funded by an unrestricted grant from AstraZeneca
Competing interests MS, SB, and ME have been paid as consultants for AstraZeneca.
Ethical approval Approval was obtained from the local research ethics committee.