Development of general healthcare preference items

A literature search identified five relevant questionnaires which covered autonomy, shared decision making and beliefs about medications, but no suitable existing measure of general healthcare preferences. [11–15] We modified concepts from these scales and from Mulley [16], informed by similar established ELSA questions on attitudes to general and financial risk [17], to develop six questions about general healthcare preferences for use in ELSA. The questions covered risk aversion, future orientation, quality or length of life, body function compared to looks, openness to experimental treatments, and preference to leave treatment decisions to a doctor (Table 2).

These new questions were asked for the first time to all participants in wave 8 of ELSA (2016/ 2017), and the questionnaire is freely available. [17] The questions had not been asked in previous surveys. The questions were asked in the paper-based self-completion questionnaire (question numbers 56–61), which was either returned by the interviewer or posted back by the respondent in a postage-paid envelope provided by the interviewer. [18] The questions followed on from the similar established questions about attitudes to general and financial risk. Each item was assessed on a scale from 0 to 10, with 10 indicating a preference for low risk. The order of responses for the first question (avoid risks) was reversed for all analyses in this paper, so as to be consistent with 10 indicating a preference for lower risk. This question was originally asked with 0 being ‘Avoid taking risks’ and 10 being ‘Fully prepared to take risks’. The remaining 5 questions were analysed with the response order unchanged from that in the questionnaire.

Data

Demographic characteristics were recorded from the main questionnaire, with the exception of ‘frequency of alcohol consumption’ which was asked about in the self-completion questionnaire. Age was categorised into three groups (50–64, 65–74 and over 75 years). Sex was categorised as male or female. Education was categorized into high (College/University and above), intermediate (Advanced-level) and low (Ordinary-level or lower). Region of residence was categorised into 9 English regions. Coronary heart disease (CHD), diabetes, respiratory disease, stroke and cancer were assessed using self-reported doctor diagnosis. Smoking status was classified as never smoked, ex-smoker or current smoker. Respondents asked whether they ‘take part in sports or activities that are vigorous / moderately energetic / mildly energetic’ in three separate questions. A binary variable was created for ‘active or ‘inactive’ by classifying participants who reported doing vigorous or moderate physical activity at least once per week as ‘active’, and those who reported doing both vigorous and moderate activity less often than once per week as ‘inactive’. The frequency of drinking alcoholic drinks over the last 12 months was classified into ‘daily’ (almost every day) or ‘less than daily’ (five or six days a week or less, including no alcoholic drinks in the last 12 months).

ELSA was conducted in accordance with the Declaration of Helsinki. ELSA Wave 8 received ethical approval from the South Central – Berkshire Research Ethics Committee on 23rd September 2015 (15/SC/0526). All ELSA participants provided written consent to take part in the study. Data are made available through the UK Data Service.

Statistical analysis

Descriptive statistics (means, percentages and 95% CIs) and statistical analyses were estimated using the wave 8 self-completion weight (w8scwt), accounting for clustering (idahhw8) and stratification by government office region (gor). The ELSA wave 8 self-completion weight was a product of a non-response weight designed to minimise any bias arising from differential non-response to the self-completion questionnaire, and the wave 8 cross-sectional weight (scaled so that the average weight was equal to 1). Linear regression models were used to estimate the fraction of variance explained by different clusters of covariates, and the association between each health preference item and the full set of included covariates. Principal component analysis was used to explore whether the health preference items could be represented by fewer underlying core components. ‘Don’t know’ values were recoded as missing. All analyses were run in Stata SE15 and R version 4.5.2.

Sensitivity analyses

The impact of missing data on the results was tested with a sensitivity analysis. We tested whether the prevalence of variance explained by each predictor was similar when we restricted the analyses to the 4,727 respondents with valid answers on all items and no missing data on predictors. A non-negligible proportion of the sample selected the middle score of 5 on almost all items, perhaps capturing people who did not know, rather than people who wanted to report a central value. Thus we also re-ran all analyses excluding responses reporting 5, to check whether results were driven by this score. Lastly, we calculated odds ratios from logistic regressions for the likelihood of responding 5 vs not 5, and for the likelihood of strong positives and strong negatives for each healthcare preference item.

Sensitivity analysis

The percentage of variance explained by each predictor was similar when we restricted the analyses to participants with valid answers on all items and no missing data on predictors (n = 4,727). Associations between participant characteristics and healthcare preferences were also similar.

We re-ran the linear regression analyses after excluding responses reporting 5, to check whether results were driven by this score. Some clustering of responses around the midpoint in attitudinal surveys may reflect genuine ambivalence in health preferences rather than measurement error. When responses with a score of 5 were removed, the direction and overall pattern of associations remained similar with the main analysis. This indicates that the main analysis results are not driven by midpoint responding.

The odds ratios from logistic regressions for the likelihood of responding 5 vs not 5, and for the likelihood of strong positives and strong negatives for each healthcare preference item showed selective associations across certain covariates. Older adults were less likely to select 5 (vs not 5) for ‘Body function’ (odds ration 0.581, p < 0.001), however overall patterns did not indicate systematic covariate-specific inflation of mid-point responses.

Women were more likely to provide strong positive responses for ‘avoid risks’ and ‘avoid experimental treatment’. Lower educational attainment was also associated with strong positive responses, for 4 out of 6 healthcare preference items. Older age, female sex and lower educational attainment were each associated with higher odds of selecting strong negative scores for some items.

Implications for research and practice

Further research is needed to test these general healthcare preferences in other populations, and with real health care decisions. Healthcare preferences are not static but can evolve over time as patients’ conditions and life circumstances change.(2) Further longitudinal research should explore how healthcare preferences evolve over time, particularly with respect to changes in health status or life circumstances, and how often they should be reassessed to ensure that medical decisions remain aligned with patient preferences. Including more domains of healthcare preferences, such as familial considerations, could provide a more holistic understanding of patient-centred care and support tailored approaches that respect patient preferences.

These findings have important implications for clinical practice. Whilst healthcare preferences vary between individuals and cannot be assumed to hold across demographic groups, there are some general preference traits by age and sex and education that may be useful to clinicians seeking to elicit patients’ preferences within the constraints of a short consultation. Shared decision making tools such as decision aids can facilitate patient-provider conversations that reveal individual preferences, particularly where patients may have distinct perspectives on treatment risks, benefits, and autonomy.

Conclusions

The new information presented here about associations between healthcare preferences and patient characteristics offers potential for the quicker elicitation of individual preferences in medical practice. These associations cannot replace assessment of preferences with individual patients, and further research and testing is needed, but cautious use of these results to facilitate the elicitation of health preferences is likely to lead to better health outcomes and greater patient satisfaction with care.