Hypoglycaemia is defined as a deficiency of glucose in the bloodstream causing neuroglycopenic and autonomic symptoms, and is a well-recognized side effect in the insulin-based management of both type 1 and type 2 diabetes, with a reported overall incidence of 42.9 events per patient-year for type 1 diabetes and 16.4 events per patient-year for type 2 diabetes [1]. Biochemical definitions for low plasma glucose range between 3.0 mmol/L (European Medicines Agency) and 3.9–4.0 mmol/L (American Diabetes Association and Canadian Diabetes Association, respectively) [2–4]. Clinically, events can be subdivided into non-severe events, for which the individual is able to take remedial action, and severe events, for which third-party assistance is required. Episodes of hypoglycaemia may occur during the day (diurnally) or nocturnally, with the latter being particularly unpredictable. Hypoglycaemia symptoms often include, but are not limited to, trembling, hunger, sweating, difficulty concentrating, confusion, unconsciousness, or coma, and, in rare cases, death [5]. The debilitating effects of severe hypoglycaemia are widely acknowledged, but reports have also highlighted the negative impact of nocturnal events on patient wellbeing [6, 7]. Hypoglycaemia can impact patient health-related quality of life (HRQoL) in two ways: firstly, through the direct physiological effects of repeated episodes; and secondly, through fear of future hypoglycaemia, leading to the adoption of avoidant, precautionary or compensatory actions (e.g., restricting holiday choices, avoiding physical exertion, defensive eating, insufficient insulin dosing) [7–10]. Moreover, repeated events can lead to impaired hypoglycaemia awareness, a condition affecting 20–25% of patients with type 1 diabetes and up to 10% of patients with type 2 diabetes, which raises the risk of severe hypoglycaemia and associated morbidity by up to six-fold [11].

To determine the HRQoL impact of hypoglycaemia, the impact per event must be quantified, which can be achieved by determining a ‘health utility’. Utilities enable HRQoL to be placed on a scale where a value of 1 corresponds to perfect health and 0 to death [12, 13]. Although studies describing utility values for diabetes-related hypoglycaemic events exist, participant numbers have been generally small and the methods employed have suffered from various limitations; for example, a failure to estimate utility per event or to determine the impact of onset time or severity [14–18]. It is clear, therefore, that a significant knowledge gap remains regarding the relative utilities associated with a non-severe compared with a severe hypoglycaemic event, and how the timing of onset may further affect HRQoL.

The primary purpose of this study was to elicit a set of utility values for hypoglycaemic events using time trade-off (TTO) methods that can be used to assess the overall patient benefits of using appropriate diabetes medication. These utility values could be used directly as an input to diabetes decision models, and under certain assumptions can be used to estimate quality-adjusted life-years (QALYs) [19, 20]. For example, in a guide to technology appraisal methods, the UK National Institute for Health and Clinical Excellence (NICE) advocates direct health-state valuations determined using the TTO method in a representative population sample in situations where EQ-5D™ data are unavailable [21].

The secondary objective was to examine the relative effects of diurnal and nocturnal hypoglycaemia on HRQoL, based on the hypothesis that nocturnal hypoglycaemic events, irrespective of severity, may have more impact than diurnal events due to their unpredictability.

The TTO method is a standard tool in the health economics evaluation arsenal, advocated for direct health-state valuations by health-technology assessment bodies. This study used a web-based TTO survey to determine the relative effects of severe, non-severe, daytime and nocturnal hypoglycaemic events on HRQoL and, as such, is the first analysis to provide a quantitative disutility value for each event class, and to demonstrate a clear increase in disutility for nocturnal compared with diurnal hypoglycaemic events.

As would be predicted based on clinical and patient-reported experiences, increasing event severity is associated with greater disutility. There were some variations between countries, with significantly lower disutilities reported for Germany and Sweden, and a significantly higher value reported for Canada for non-severe hypoglycaemia compared with the overall population. However, it is perhaps most interesting to note the high overall similarity between populations. The response consistency observed both across populations and between countries supports the credibility of the results and suggests that hypoglycaemia-related disutility is comparable and independent of healthcare system differences.

This study sampled a large number of respondents from multiple countries, 78–81% of whom were willing to trade-off health improvements against projected life expectancy, indicating understanding and acceptance of the TTO concept. The respondents represented three distinct groups: the general population, people with type 1 diabetes and people with type 2 diabetes, providing a unique opportunity to identify any significant differences. Indeed, a significantly higher disutility for severe nocturnal hypoglycaemia was reported by the type 2 diabetes population (p = 0.008).

The web-based approach, whilst facilitating respondent participation, does mean that help was not available if respondents had queries. Additionally, although a time delay was built into the survey, the lack of supervision may have led to some respondents not spending enough time considering their answers. Collectively, these design attributes may have led to inconsistencies within the responses. It should be noted, however, that the low dropout rate (10%) indicates the questionnaire was clear and manageable for most respondents, and potential skewing due to respondent fatigue was minimized by the methods employed.

Although participation bias must be considered, the incentives to participate would be expected to mitigate any initial disinclination. A recent report suggests that the application of discounting to correct for time preferences has an influential effect on outcomes [40]; we did not apply this in the present study. The use of an internet-based survey may also pose a selection bias, since only literate respondents with computer access could participate. However, the literacy rates and proportion of internet users in all sampled countries are high (e.g., the UK has a 99% literacy rate and 51.1 million computer users out of 63 million inhabitants) [41].

A further strength of this study is that the findings are of a similar magnitude compared with previous research. The observed baseline diabetes utility value of 0.844 found in the current study is in line with that reported previously [18]. A recent review found every non-severe event (day and night) may be associated with a utility loss of between 0.0033–0.0052 over 1 year [42]. However, previous studies describing diabetes-related utility values for hypoglycaemic events have various limitations. Cross-sectional studies using generic health instruments either suffer from potential unobserved confounding or do not estimate utility per event [14–16]. Matza and colleagues studied 129 people with type 2 diabetes, using the standard gamble technique, and found a significant disutility (p < 0.001) associated with the fear of hypoglycaemia overall, but did not distinguish between onset time or severity, or calculate disutility per event [17]. In contrast, Levy and colleagues surveyed 51 people with diabetes and 154 respondents without diabetes, using a TTO approach to estimate utility values per hypoglycaemic event, but did not report values for nocturnal or severe events [18]. In addition, these previous studies were limited by small sample sizes.

Moreover, the use of age-dependent, life-expectancy-based adaptation of the TTO questions for each respondent, applied successfully by other groups [32, 33], may avoid some of the disadvantages of the artificial fixed 10- or 30-year horizon used elsewhere [18, 22] by increasing the relevance of the trade-offs to the respondents and so providing more reliable utility value estimates.

Non-severe hypoglycaemic events have been shown to have a measurable detrimental impact on patient wellbeing, reflected by increased healthcare professional visits (25% of participants) and higher testing-strip consumption (5.6 on average), and a quarter of respondents reduced their insulin dose in the days immediately following an event [7]. After a non-severe nocturnal event, 23% of respondents reported arriving late or missing a day of work, and 32% missed a meeting or deadline [7]. Although an association between nocturnal hypoglycaemia and reduced HRQoL was demonstrated, the current investigation is the first analysis to provide quantitative disutility values and to demonstrate a clear increase in disutility for nocturnal compared with diurnal hypoglycaemic events.

Within clinical practice there is recognition of the phenomenon of ‘first being worst’; that is, the effect of each hypoglycaemic event on HRQoL diminishes as frequency increases and the patient adapts. In health economic terms, this is referred to as diminishing marginal disutility. Interestingly, the degree of disutility associated with increasing frequency of non-severe hypoglycaemic events consistently increased in this study, irrespective of onset time. The diminishing marginal disutility may reflect a coping mechanism, a maximum trade-off limit, or study design limitations, where some respondents might pay more attention to the health-state descriptions than the actual frequencies.

Identifying a minimally important difference (MID), described as the smallest change in the patient-reported outcome of interest that is either perceived as beneficial or that would elicit a change in behaviour [43, 44], underpins the clinical relevance interpretation of any HRQoL study. However, no universally accepted method for MID estimation exists, with both anchor-based or distribution-based methods being employed [45–47]. An MID has been reported for some generic health instruments; Drummond reported an MID in utilities of 0.03 for the 15D instrument and the Health Utilities Index (HUI®), with the elaboration that utilities of 0.01 may be meaningful in some contexts [48]. Luo and colleagues reported mean MID estimates of 0.040 for the EQ-5D™ (US algorithm), 0.082 for the EQ-5D™ (UK algorithm), 0.045 for the HUI-2, 0.032 for the HUI-3 and 0.027 for the SF-6D [45]. Generic instruments generally lack sensitivity, which is why direct elicitation using an approach such as TTO becomes relevant. When patients trade a portion of their life expectancy to improve quality of life, they implicitly express the importance of the health state. In this study, the utility differences derived are per event; therefore, whilst they appear small initially, when the predicted annual event frequency is considered, the differences would be quite substantial.

In summary, this analysis provides a unique evaluation of severe compared with non-severe hypoglycaemic events, and of the distinction between daytime and nocturnal hypoglycaemia, providing a comprehensive breakdown of the respective contributions made by each of these distinct event subgroups to patient HRQoL.