This was an observational study of participants with MS or epilepsy. The participants were asked to choose between alternative mobile technologies that were described according to a set of characteristics in a DCE survey. The survey was administered online and was given ethical approval by the National Research Ethics Service Committee London—Social Care (19/IEC08/0013).

Formal sample size calculations are challenging for DCEs, but their level of precision increases rapidly up to 150 participants, [33] so we adopted this as our sample size guide for each group (ie, MS and epilepsy). Analysis was undertaken using a mixed multinomial logit model which allows for unobserved heterogeneity of preferences across respondents. This modeling approach also accommodates multiple observations per respondent and relaxes the assumption of independence of irrelevant alternatives required in the commonly applied multinomial logit model. Responses of “I don’t know” were treated as missing data rather than rejection of the device. It was hypothesized that higher levels of privacy, greater levels of support, increased accuracy of detection, and clearer benefit to the user would influence respondents’ decisions. Coefficient signs and significance were explored to investigate whether the data supported these hypotheses. The attribute accuracy of detection was specified as a categorical variable with 3 levels to allow for a nonlinear relationship with utility. All categorical variables were effects coded to aid interpretation [34]. Marginal rates of substitution were calculated to express the degree to which respondents would trade off attribute levels and accuracy. It was assumed that the change in accuracy was linearly related to the change in utility for changes in accuracy within the levels modeled.

The effects of 7 prespecified patient characteristics on response data were investigated. The characteristics were health condition (ie, MS or epilepsy), age, gender, education (categorized as low, medium, or high), depression within the past 2 years (categorized as yes or no), current user of wearable technology (categorized as yes or no), and score for acceptance of technology. Each patient characteristic was investigated separately by inclusion of interaction effects. Improvement in the Akaike information criterion was used as a criterion for a significant difference in preferences by subgroup. Interaction terms for all subgroups identified as having a significant impact on preferences were included in a single model. Backwards elimination was undertaken to assess whether the identified characteristics were capturing different underlying distributions of preferences across respondents. Characteristics were retained when inclusion of their interaction terms minimized the Akaike information criterion. The impact of these characteristics on preferences was quantified by recalculating the marginal rates of substitution by patient subgroup, utilizing a mixed logit model including interaction terms for that patient characteristic.

A total of 318 respondents completed the survey. Of these, 141 (44%) of the respondents had MS, 175 (55%) had epilepsy, and 2 (0.6%) had both MS and epilepsy. All respondents answered at least one question. Table 1 reports the demographic and other characteristics of these respondents.

The coefficients from the mixed logit model excluding interactions with patient characteristics are shown in Table 2. Improved accuracy, higher privacy, increased level of benefit to the user, and the availability of clinical support were all associated with an increased likelihood of selecting a mobile technology device. Accuracy was the most important attribute, with a nonlinear effect; a move from low to moderate accuracy was valued higher than a move from moderate to high accuracy. The next most important attribute was privacy. Again, there was a modest nonlinear effect, with a stronger preference for moving from low to moderate privacy. Clinical support was the least valued of the attributes. The SDs reflect the impact of unobserved heterogeneity of preferences. There was evidence of unobserved heterogeneity in preferences for low versus moderate accuracy and for clinical support.

Table 3 reports the marginal rates of substitution for each attribute compared to accuracy. This is the percentage of accuracy that respondents were prepared to trade to achieve an improvement in the remaining attributes. Respondents were prepared to trade a modest amount of accuracy to achieve an improvement in privacy. Clinical support, in contrast, was not valued; respondents were only prepared to accept small reductions in accuracy in exchange for a high level of clinical support.

The coefficients from the mixed logit model including interactions with patient characteristics can be found in Table 4 and Table 5. Regression analyses indicated that health condition (ie, MS or epilepsy), age, gender, and depression had no significant effect on preferences. Current use of wearable technology had a marginal impact on preferences. Technology acceptance did have an impact, as did education level. After a multivariate analysis that included wearable technology use, education level, and technology acceptance score, wearable technology use was no longer significant. Preferences varied significantly for patients in the low education group but not between those in the average or high education groups. Table 6 reports the marginal rates of substitution for each attribute compared with accuracy for respondents with education beyond age 18 compared to those without, and for patients with a high technology acceptance score (in the 75th percentile) compared to those with a low score (in the 25th percentile).

Respondents with lower technology acceptance scores were prepared to trade far more accuracy for an improvement in privacy. The impact of technology acceptance on preferences for clinical benefit and clinical support was more modest. Respondents with high technology acceptance were prepared to trade slightly more in terms of loss of accuracy to improve clinical benefit or clinical support. These data suggest privacy is a far greater concern for respondents with a low technology acceptance score. The impact of education was more modest. Respondents with low education appeared less inclined to trade accuracy for improvements in clinical benefit and privacy but were more prepared to trade accuracy for improvements in clinical support.

Including missing data as an active decision to reject both technologies for the respective question (ie, an “opt-out”) had a minimal impact on the magnitude of the model coefficients, and their significance and direction were unchanged. These results are available on request.

Our analyses revealed that people with epilepsy or MS value higher levels of privacy, greater levels of support, increased accuracy, and a clearer benefit to the user when selecting mHealth technology devices. This is a key finding that is in line with our hypotheses and validates feedback received from people with epilepsy and MS in other studies [1,5,6]. Of all these factors, people preferred a higher degree of accuracy, regardless of their diagnosis. When asked to make compromises between levels of privacy, clinical support, accuracy, and benefit to the user, people were willing to trade some accuracy for greater privacy but were less influenced by the other factors. This analysis reveals a hierarchy in the importance of factors influencing engagement with mobile technologies (Figure 2).

We wanted to explore individual variability in how people traded off these attributes, because this should lead to a more tailored or personalized approach to RMT development. We found no evidence that age, gender, or experience of depression affected preferences. This is in contrast to other studies, such as one that found that age influences technology use in a different patient population [11]. We did, however, discover the following: (1) the people who were less accepting of technology placed greater value on privacy and were willing to give up some degree of accuracy for privacy, and (2) people with no qualifications beyond those that might have been obtained before the age of 18 were willing to compromise some degree of accuracy to receive greater clinical support. These moderating factors are shown in Figure 2.

This model will help those designing mobile technologies to prioritize features for development that can maximize engagement. Accuracy is the key feature; individuals were willing to make compromises on accuracy, but the reduction in accuracy they were prepared to accept for improvements in other characteristics was relatively small (<20%). We did not find that people with MS and people with epilepsy had different preferences; however, accuracy may have a different meaning for devices that detect seizures in epilepsy and devices that measure symptom recurrence or deterioration in MS. Issues to do with privacy and the willingness to share information with a clinician were more transdiagnostic, with subgroup differences relating more to trust and familiarity with technology and educational level. This indicates that some level of personalization may be required in the design of devices.

This study is the first of its kind to try to understand the relative value and influences of factors affecting engagement with mobile technologies for people with neurological conditions. It used a quantitative approach, which is novel in this area of research, adopted from the field of health economics. The study size was large enough to explore some subgroups, selected for their hypothesized relationship with technology use, but future researchers may wish to focus on a greater number of health-related variables, such as illness severity, which we were unable to discuss in this paper. Basing our study on the completion of an online survey, albeit one that was supervised for participants who were recruited from clinics, may have led to a sampling bias toward respondents who were more familiar with technology. However, there were few differences in the patterns of the data and in the makeup of the samples. An additional limitation is that we did not design the DCE in a way to allow validity checks on choice data, such as by including repeated tasks or dominant alternatives.

We have shown that people with epilepsy and MS are influenced by factors such as the accuracy, privacy, benefit of the technology, and the amount of clinical support received, but that in some instances they are willing to make compromises. These preferences should be factored into the design of mHealth technologies, alongside the views of other stakeholders, in the future.