Patient preferences for epilepsy treatment: a systematic review of discrete choice experimental studies

We included primary studies using discrete-choice experiments to elicit preferences for various attributes of pharmacological and surgical interventions in patients diagnosed with epilepsy or the parents/carers of children with epilepsy.

The following types of studies were excluded:

We searched the PubMed, Web of Science and Scopus databases from inception until the date of the search (February 2022 or April 2022). The reference lists of the included articles were reviewed for potential studies.

Search terms for stated preferences derived from previous systematic reviews on DCEs and recommended searching strategies [19, 20] for preference studies were used. This included words such as “conjoint analysis”, “conjoint choice experiment”, and “stated preference”. These terms were combined with subject headings and free text terms for epilepsy. The specific terms used to search each database and search date are available in Appendix 1.

Three reviewers screened the titles and abstracts. First, two reviewers (RA, BA) independently screened the titles and abstracts. A third unblinded reviewer (SA) screened all titles and abstracts, made decisions and identified disagreements between the reviewers. Disagreements regarding study inclusion or exclusion were resolved by discussion. The Ryaan tool was used for title and abstract screening and selection.

Data were extracted by one reviewer (RA, BA) and reviewed for accuracy and completeness by a second reviewer (SA). Excel was used to collect data.

We collected data on study characteristics (e.g., first author, year of publication, study objective, country, sample size, main results), attributes and levels of information (e.g., attribute and level identification, selection and labeling, the mode of survey administration, relative importance of each attribute category).

The risk of bias assessment was assessed using the PREFS (Purpose, Respondents, Explanation, Findings, Significance) checklist [21] and ISPOR Conjoint Analysis Applications in Health Checklist [10], a 10-item checklist. Two independent reviewers carried out risk of bias assessment, and disagreements were resolved by discussion.

A descriptive analysis of the data in a narrative format, accompanied by tables to convey relevant study components, was conducted.

The search identified 961 records. After removing duplicate records (n = 312), we screened the titles and abstracts and excluded 633 articles. At this stage, the main reasons for exclusion were not DCE (n = 617), DCE but not treatment (n = 11), and no stated preference (n = 5). The full text of 16 articles was assessed for eligibility; 7 were included [22‐28], and 9 were excluded. Three titles were without abstracts, and after obtaining the full text were excluded because they were not DCEs. One study was not a DCE [29]. One study was a DCE of self-management programs [30]. One discrete choice study that compared society and patients’ preferences for health priorities and epilepsy patients was 2% of their sample [31]. One study was excluded because it was a focus group of physicians and caregivers to identify attributes for future discrete choice experiments [32]. Two abstracts were excluded. We contacted the authors of these abstracts to enquire if they were published in full. One abstract was not published [33], the other abstract was published, and the full text was excluded for the abovementioned reason [32]. Figure 1 shows the flow diagram of the included studies and the reasons for exclusion.

The majority of the studies were published after 2017 (n = 5). Five studies explored patients’ preferences, and two compared the preferences of patients with those of physicians [22, 27]. Six studies compared two medications, and one compared two surgical options to continuing medication options [28]. The online survey was the most common mode of administration (n = 6). The questionnaire was mainly self-administered by participants (n = 6). Three studies were from the US, two from the UK, one from China, and one with participants from different European countries. The sample size ranged from 148 to 518 patients (Table 1). The studies were published in clinical journals, with only one published in a specialized pharmacoeconomic journal [25]. Three studies [22, 25, 27] were funded by pharmaceutical companies.

Using the PREFS checklist for assessing quality (Appendix 2), we identified one major shortcoming: the lack of reporting on the number of candidates approached and those who declined to participate and the differences between responders and nonresponders, if any. One study used a postal survey and reported the response rate to be 28% [25], and another study reported 33 patients who consented to participate in the study but did not complete the survey [23]. Both studies did not report any details of nonrespondents’ characteristics.

Appendix 2 presents the ISPOR risk of bias checklist for the included studies. The following section will discuss the risk of bias in the included studies in reference to the ISPOR checklist.

This section summarizes which attributes are most important for people with epilepsy as reported in identified studies. Further details on the study aims and level of attributes are available in Table 1.

In Lloyd et al. [25], a pair of hypothetical AED profiles were described using five attributes related to adverse effects plus seizure control and cost (to estimate willingness to pay). The study presented the WTP for a unit improvement in each attribute level; for example, participants are willing to pay £2.67 per month per 1% reduction in the chance of hair loss. The study also presented the absolute WTP for different health states; for instance, respondents were willing to pay £709 per month for being seizure free with no adverse effects (the currency year is 2002). The study found that participants were willing to give up improvement in seizure control to avoid adverse effects. Segmented models showed that participants would be willing to pay 34%–42% of their income for being seizure free and experiencing no adverse effects on their health state and that women were willing to pay twice as much as men to avoid weight gain.

The study aim of Manjunath et al. [26] was to quantify the relative importance of 8 attributes of add-on AEDs. Seizure reduction was the most preferred outcome, with a mean relative importance of 10.0, followed by limitations due to long-term confusion or memory problems (8.6) and weight change (8.2). A treatment that provides freedom from seizures with no side effects and a single dosing was valued by respondents at $126.36 more than their current treatment (the currency year not reported is assumed to be 2012).

Ettinger et al. [22] compared patient and neurologist preferences regarding two hypothetical AEDs characterized by six attributes. Both patients and neurologists ranked seizure control as the most important attribute; however, seizure control had a significantly greater weighting in neurologists' decision-making than among patients (45% vs. 32%, p < 0.005). Patients placed more importance than neurologists on psychiatric adverse effects (19% vs. 15%, p < 0.05), diminished coordination and balance (16% vs. 10%, p < 0.05), and fatigue or diminished energy (13% vs. 11% m p < 0.05).

Holmes et al. [23] compared patient benefit-risk preferences between recently diagnosed patients and women of childbearing age. The results show that respondents had stronger preferences for reductions in the risk of AEs than improvements in 12-month seizure remission. In the recently diagnosed group, the maximum acceptable incremental risk of adverse effects for an AED that increased the 12-month seizure remission by 10% was 3.1% for depression, 3.0% for memory problems, and 2.5% for aggression. Women of childbearing age would accept an incremental risk of 5.6% for depression, 3.4% for memory problems and 2.0% for fetal abnormality for an AED that increases the 12-month seizure remission by 10%. The authors also applied the preference weights derived from discrete choice experiments to clinical trial data to estimate the expected utility of five alternative drugs.

Hua et al. [24] examined Chinese patients’ risk–benefit preferences and WTP for AED treatment. Seizure control was the most preferred treatment attribute (10.0; 95% CI 8.9–11.1), followed by the effects of AEDs on the fetus (8.9; 95% CI 7.7–10.1), the duration of side effects in the neuropsychic system (4.9; 95% CI 3.7–6.0), and adverse reactions of the digestive system (3.2; 95% CI 1.5–4.2). The participants were willing to spend ¥1,246 (95% CI, U 632–U 1,861) per month to ensure 100% seizure control and ¥ 1,112 (95% CI, U 586–U 1,658) to reduce the risk of the drug affecting the fetus by 3%. The currency year was not reported and was assumed to be 2020. Furthermore, patients who have the intention of getting pregnant have a high preference for lowering the effects on the fetus than patients who are not and value the effect of AEDs on the fetus more than the efficacy of AEDs.

Sinha et al. [28] determined patients acceptability of benefit-risk trade-offs in selecting treatment options for drug-resistant epilepsy, including open brain surgery, laser ablation or continued medications. The study findings show that respondents who were willing to undergo a procedure were willing to accept a markedly lower chance of freedom from seizures (23%) if they were able to undergo laser treatment as their first surgical option as opposed to 70% seizure freedom for open brain surgery. For 30-day mortality, the minimum acceptable benefit was 52% (95% CI 39–68%) for risk reduction from 1 to 0%. For the risk of long-term problems, the minimum acceptable benefit was 39% (95% CI 28–51%) for a risk reduction from 10 to 0%.

Rosenow et al. [27] explored patients' trade-offs between treatment efficacy and the risk of side effects of AEDs before and up to 14 days after treatment consultation and to elicit physicians’ preferences in selecting treatment for a specific patient after consultation with that patient. The most important attribute for patients before consultation was the chance of becoming seizure free with a mean conditional relative importance value of 10.0 (95% CI 7.8–12.2), followed by trouble thinking clearly (8.2, 95% CI 6.6–9.9) and personality changes (6.9, 95% CI 5.4–8.3). Patient preferences before and after treatment consultation were generally qualitatively similar. The mean conditional relative importance calculations showed that the most important attribute for physicians was the chance of becoming seizure free (10.0, 95% CI 8.6–11.4), followed by personality changes (7.5, 95% CI 6.2–8.9) and trouble thinking clearly (6.2, 95% CI 4.9–7.5).

The number of identified studies is limited. There is a need for more research in this area. Future research should build on the findings of this review with regard to important elements of study design that have received little attention in the published literature. The first area is the identification and selection of attributes to describe alternative choices. We recommend more details on the purpose of the qualitative process, e.g., to identify or validate attributes and levels, order attributes, classify level ranges or test the complexity of the survey; the qualitative approaches used, e.g., structured or semistructured interviews, focus groups, think aloud exercises, or observations of patients’ decision-making processes; and approaches used to analyze qualitative data. Helter et al. [41] suggested four stages for attribute development: raw data collection; data reduction; removing inappropriate attributes; and wording using different qualitative and alternative methods for each stage. This framework can be used to describe attribute development in future research. Secondary areas that require attention include justification for the choice of experimental design used to create tasks and the model estimations approach to analyze data.

The focus of available evidence was on efficacy and side effects, and we recommend further qualitative research to explore other attributes that might also be important for patients with epilepsy. People with epilepsy have many concerns about living with epilepsy in addition to seizure control and medication side effects [42] that may influence their preferences for interventions.

New studies should also examine questions not explored by the published research, which could be achieved, for instance, by investigating underrepresented patient groups such as children, adolescents or elderly individuals, the preference for discontinuation of treatment and the impact of patient characteristics on preferences. Indeed, understanding aspects that may explain preference heterogeneity, such as patient characteristics, is one of the methodological priorities for patient preference research in general [43].

The main strength of this review is the systematic search for evidence and quality appraisal of the included studies. There is a chance that we missed some studies published in journals not indexed in the databases we searched. Another limitation of our method is that data were extracted from published manuscripts without attempts to obtain or confirm data from the investigators. In our review, we excluded studies measuring stated preference using direct elicitation of monetary values of an intervention, such as willingness-to-pay methods, and preference-based studies, such as time-trade-offs or standard gambles, referred to as preference-based but not considered stated preferences. The reason for their exclusion was to review studies with homogenous methodologies, as excluded studies differ in their methodologies, results use and implications. Future reviews may examine such types of studies.