Since our themes were generated from grounded theory, their titles may be interpreted in a variety of ways. We therefore created clear descriptions for each attribute in relation to remote care (see Table 1). For each attribute, we chose two levels, positive and negative, corresponding to the level of attainment of any given attribute, with neutral included as a negative level [8,9].

Each question forced the participant to choose between two hypothetical remote care technologies with opposing attribute levels, that is, a positive level in an attribute in one choice means that the alternative choice will have a negative level of that same attribute. The forced choice design reduced the complexity of adding an opt-out alternative to each question, which minimized questionnaire fatigue [10].

The choice sets (the combination of levels of each attribute that were grouped together per question) were assigned based on a predetermined, orthogonal design algorithm [11]. For a discrete choice questionnaire containing 5 attributes each with 2 levels, this resulted in 32 profiles split across 16 questions. The order of the questions was randomized to mask the pattern of the choice sets. The attributes were listed in alphabetical order in each question [12,13].

We used an established method for determining the minimum sample size for conjoint analyses [14]:

Where N is the minimum sample size; c is the number of levels; t is the number of questions; and a is the number of alternative answers.

For a 16-question survey with 2 choices, the recommended minimum response size is 32 participants. We took this as a minimum and left the online survey open until the end of the study window to capture as many responses as possible.

Patients who were aged 18 years or over and had a diagnosis of chronic heart failure were included in the study. Exclusion criteria included (a) diagnosis of acute heart failure without any chronic component and (b) non-English speaking patients (the questionnaire was only available in English).

A patient participation group consisting of five patients with heart failure and related conditions was formed to aid the outputs of the research. These patients were recruited via a free access public engagement event held at the University of Liverpool on October 23, 2017. This event involved talks from cardiology and technology experts to inform on upcoming heart failure technology research and generate interest in public participation. After the formation of the group, several informal discussions and feedback sessions were conducted between February and March 2018, where the group piloted the questionnaire and had input into the patient information leaflet. Design changes were made due to this feedback, including shortening the questions, formatting for better readability, as well as expanding on the patient information leaflet to provide more context for the study (Multimedia Appendix 1 and Multimedia Appendix 2). Furthermore, the patient group members helped to suggest places where the survey could be distributed online to heart failure care communities.

As per HRA guidance [15], responses to online surveys imply consent as long as participants are provided with sufficient information to reach an informed decision. We worked with our patient group to develop substantially descriptive participant information for them to make an informed choice. This study was approved by the Research Ethics Committee at the University of Liverpool (ref: 3314). The survey was exported online using a secure digital platform (SurveyMonkey), which complies with EU Privacy Laws and General Data Protection Regulations, and is registered under the Data Protection Act. This online platform was used to create a web link, which was the primary means of distributing the survey to participants. In accordance with the principles of data minimization and purpose limitation under General Data Protection Regulations, no personal or demographic data were collected by the research team; therefore, participants were not identifiable, nor was there any direct contact between the research staff and participants. No monetary compensation was offered to any participant for completing the questionnaire. Raw and processed data were stored securely on encrypted university intranet servers.

The link to the survey was distributed to national and international heart failure patient groups, which were accessed via social media and communications through heart failure charities. A list of organizations approached for distribution can be found in the supplementary information (Multimedia Appendix 3). It is important to note that while the study was conducted in the United Kingdom, the survey was distributed worldwide, and so the respondents were not limited by geographic location.

Responses were analyzed using limited dependent-variable models to determine preference weights of each attribute [16]. From this, we can infer which attributes participants are willing to trade in favor of others. Our DCE is a forced-choice, five-attribute, two-level, two-alternative questionnaire (Multimedia Appendix 4). As both the choices and the levels were binary, binary logit [16] was used to determine the likelihood of the outcome. The logarithmic function ensures the likelihood values are constrained between 0 and 1 [17].

The logit definition is as follows: [18]:

Logit(P)=log(odds)=log(P/(1−P))

As part of the regression, we assign logit(P) as a linear function of any given attribute Xi, so that:

Where P=probability (of choosing this option); α=reference value or constant; β=coefficient of attribute X; i=attribute number; U=utility

The logit value is proportional to the odds of an attribute, affecting the probability of choosing an alternative. Thus, these values can be compared directly as preference weights for each variable. The preference value for each attribute is known as utility, which is the measure of importance of each attribute or combination of attributes. In order to standardize for participant heterogeneity, random effects were added to create a mixed binary logit model [18,19].

The utility value of each combination of attribute level was obtained by adding the constant coefficient of attribute X from the logit model, with the coefficients of each positive attribute present. The odds were obtained by exponentiating the utility. To convert this to percentage uptake probability, that is, the likelihood of choosing this remote care device as opposed to the alternative, we divided the Odds by 1+Odds [20]. The dataset was analyzed using RStudio version 1.0.136. These calculations were also corroborated using STATA/MP 13.0.

The analysis ranked the remote care attributes in the following order of importance for patients with heart failure: (1) clinical care; (2) education; (3) convenience; (4) ease of use; and (5) communication. Based on the coefficients of the logit fit, clinical care was almost twice as important as the lowest scoring variable, communication. Remote care technology design should therefore prioritize clinical care improvements first and foremost. The attributes of education and convenience had similar preference values, which were around 20% greater than communication. Ease of use was 11% more important than communication. This pattern of preference shows a disproportionately high preponderance toward clinical care, with the second, third, and fourth ranked attributes plateauing at a similar level. Therefore, if a trade-off is required, any other attribute may be sacrificed for the sake of preserving clinical care, while still incentivizing patient engagement.

A number of DCE and conjoint analyses have been published regarding patient preferences for telecare since the COVID pandemic [23-25]. However, there have been no other DCEs evaluating engagement of remote care technologies in this patient cohort of chronic heart failure. Therefore, the study provides a valuable insight into the factors of remote care devices that encourage engagement. In a post-COVID era, remote care technologies have gained greater importance in health care. Patients with heart failure are a vulnerable cohort and so are more likely to be offered remote consultation. Therefore, these preference rankings are all the more vital at this time to help remote care become better established in medical practice for those that need it most.

Improving this and similar surveys may require shifts in methodologies to make it more generalizable, albeit with additional feedback. In the first instance, while patient participation was a key determinant in the design of the methodology, additional value may have been obtained by reaching out to technology designers and start-ups that create devices within the space. Obtaining this kind of feedback would enable tailoring of the outputs in such a way as to provide the most deliverable benefit in the context of future design by, for instance, presenting realistic alternatives grounded in existing technologies as opposed to theoretical ones.

To mitigate some of the limitations further, it may be also useful to obtain demographic information and location of participants so as to correctly contextualize the responses based on patient profile, recognizing that different subpopulations may have differences in preference.

Finally, in order to address noncompletion rates, the questionnaire could be shortened in order to be less mentally taxing, while also ensuring a process of gathering feedback from participants as to reasons for noncompletion.

Our questionnaire used a DCE method to elicit preferences for remote care technology from patients with heart failure from around the world. Results of the analysis indicate that clinical care was substantially more valued as a factor for engagement with remote technology than the four other themes of education, convenience, ease of use, and communication. Our findings also allow approximations of increase in engagement by sequentially adding in these individual factors to an existing remote care device based on their preference values. This hierarchy could provide useful insights for technology designers to check the effectiveness of an intervention’s features in engaging the end user and help develop a plan of improvement for devices based on their missing attributes. Incorporating these attributes appropriately will ultimately bring remote care technology to these patients in a more effective and engaging manner, to reduce the burden of morbidity from chronic heart failure.