Summary of Findings
In this study, we piloted an approach to explore whether responses to TTO tasks differ when participants are provided with additional contextual information around their need for a carer when living in the health state being valued, as well as the carer’s health state. While participants did rate the combined exercise as more difficult than the traditional exercise, the combined exercise appeared to be feasible to conduct in practice. The mean utilities had face validity in that health state A was preferred to health state B, which was preferred to health state C (irrespective of the exercise). Mean utilities for each health state in the traditional exercise were higher than those observed in the EQ-5D-5L crosswalk value set for the UK .
Mean differences in utilities between the two exercises were generally larger in magnitude than many of the estimated minimally important differences for EQ-5D-5L [33, 34], suggesting that differences of this magnitude could be meaningful. However, there was inconsistency in the differences in valuations between exercises (different directions), which meant that there was no strong evidence for a consistent carer effect at a group level. Nonetheless, it cannot be concluded that participants were valuing health states very similarly between the two exercises, as illustrated by the heterogeneous individual response trajectories in Fig. 2. There was also a considerable split between the proportion of individuals who considered health states B and C to be better or worse than dead in both exercises, which added to the noise in the data.
It was expected that many participants would provide lower values in the combined exercise due to a concern for the carer, i.e. the quality-of-life altruism effect observed by Krol et al. . However, a longevity altruism effect may also have occurred with some participants, and some participants may also have been comforted by the confirmation that they would have a carer, both of which could work in the opposite direction. It was also found that the well-being of the carer in the combined exercise was a concern for many participants (n = 15; 45%). The carer's health state in the combined exercise is likely to have differed to the actual health state of the individual(s) that participants imagined for the carer role (Table 5). These differences in health status could have exacerbated either of the altruism effects, affecting valuations in either direction. Furthermore, TTO is known to be subject to learning effects, which may have had a greater impact on the results of the combined exercise relative to the traditional exercise, as the former was always conducted after the latter . This could also have affected results in either direction.
This broad mixture of potential impacts may go some way in explaining the variation in our results, and highlights the challenges associated with identifying the impact of providing contextual information on valuations. Thus, while the task appeared feasible to conduct in practice, identifying the specific impact of the provision of additional contextual information using this approach might not be feasible without substantial adjustments being made to the approach (see Sect. 4.3).
This pilot study had several limitations. First, the study sample size was small and the study was underpowered for the effect size observed. With only 33 participants, the study would be powered to pick up mean differences of a magnitude of around 0.2, which is larger than the observed mean differences of around 0.07. Second, convenience samples are susceptible to selection effects (explaining our relatively homogenous sample) and our results are likely to have differed if we had recruited a representative sample of the general population. Third, we aimed to follow good practice when conducting the study [19, 25], with a particular focus on EQ-VT protocol . However, we did not employ the EQ-VT digital aid and instead used physical props, given the flexibility required for the combined exercise. We therefore did not benefit from the interactive features of the computer-based tool and the feedback module that has been shown to improve data quality . However, we sought to maximise data quality by providing all interviewers with the same initial training, which was followed up with one-on-one coaching sessions, and by including a warm-up exercise in the interviews. Fourth, while it may be typical to estimate mean utilities using Tobit models that control for sociodemographic characteristics, our analysis focused on direct comparisons of mean differences using paired t-tests. We felt this was justified by our limited sample size, the lack of variation in sociodemographic characteristics, and the need for a panel set up to incorporate observations from both exercises; however, our analysis may be limited as a result. Fifth, while the carer was defined as an informal carer at the beginning of the combined exercise, five participants stated that they considered professional (formal) carers, suggesting that not all instructions were followed precisely by all participants. Finally, another limitation was that the interviews were not audio recorded and it was therefore not possible to obtain full transcripts. Transcripts would have provided richer data for the qualitative analysis.
While our pilot study has shown that providing contextual information in health state valuation exercises is feasible, it has also illustrated the potential difficulties in understanding the differences in quantitative results. Nonetheless, further research in this area would be beneficial and this pilot study may provide useful insights for such studies.
Future studies of this nature may benefit from taking a simpler approach to the present study, potentially controlling for competing effects such as mentioning the need for a carer but not specifying their health state (or simplifying the health state provided). Furthermore, while this pilot study did directly refer to carers as opposed to loved ones more generally, it may have been beneficial to have explicitly instructed participants to consider the information in their valuations as this may have been more impactful than simply providing the information, as the results of the study by Krol et al.  suggest.
Many participants (n = 22; 67%) stated that their responses would have differed had the carer been in full health, suggesting that this information was relevant to the valuations being made. However, informal feedback from interviewers suggested that some participants struggled to take on board multiple EQ-5D-5L health states at the same time, suggesting that the inclusion of this information increased the difficulty of the exercise. An alternative approach that would enable the impact of the carer health states to be explored further would be to have two combined exercises, where only the carer’s existence is described in the first exercise and the carer’s health state is introduced in the second exercise. Furthermore, the carer's health state could be described using a carer-specific instrument such as CarerQol to reduce the focus on EQ-5D-5L and provide a more relevant description .
Future studies may also benefit from collecting more data and recording interviews for transcripts. Directly asking participants to reflect on whether the information about needing a carer is comforting or concerning overall would be useful. It would also be helpful to identify participants who may be unwilling to trade life-years due to longevity altruism, as this has an opposite effect to quality-of-life altruism, but occurs as a result of a similar thought process (i.e. considering the impact on others) .
Additionally, future studies may also benefit from directly asking participants whether they believe that the implications of their responses are consistent with their beliefs. In some interviews, interviewers noticed contradictions in the responses provided by participants. Using alternative methods that have a more deliberative focus, or implementing direct validity tests into the study design, could be more informative and result in better-quality preference data [38, 39]. It may also be worth considering that a fully qualitative approach could be superior to a quantitative or mixed-methods approach in this area, given the wide range of factors that may influence valuations and the lower sample size requirements that these require.