In this study of primary care subjects with type 2 diabetes, we observed a high risk of death in people with a prior admission to hospital for a severe hypoglycaemic episode: in a 60-year-old individual, the 5 year risk of death was around 7% for CVD causes, 1% for cancer causes, and 13% for non-CVD and non-cancer causes, resulting in an overall risk of around 21%. The corresponding risk in a person without severe hypoglycaemia was 2%, 3% and 2%, resulting in an absolute mortality risk difference associated with severe hypoglycaemia of 14%. Accounting for missing data and for a large set of confounders strongly related to the risk of death, the estimate was more conservative and indicated an absolute risk difference of 11%.
Hypoglycaemia in individuals with diabetes has long been recognised as an acute, frequent and, in most cases, reversible complication of intensive glucose control. The scientific interest around hypoglycaemia reignited a decade ago, after the publication of randomised controlled trials demonstrating a neutral or higher risk of death in individuals with type 2 diabetes randomised to an intensive glucose control. Hypoglycaemia was deemed a potential reason, with severe, repeated hypoglycaemic episodes not only thought to pose a risk for acute symptoms and complications but also to predispose an individual to long-term CVD events through direct (i.e. ECG abnormalities) and indirect (i.e. platelet abnormalities) mechanisms . The clinical trade-off would therefore be between long-term complications of hyperglycaemia and short- and long-term complications of hypoglycaemia. As a result, there has been increasing emphasis on inappropriate intensive glucose control in individuals at a higher risk of hypoglycaemia (overtreatment) .
In light of the rising prevalence of severe hypoglycaemia in multimorbid people with type 2 diabetes [12,13,14], clarifying whether hypoglycaemia is causally related to an increased risk of CVD remains important from both a patient and a public health perspective. Most of the available evidence from epidemiological studies, in different geographical regions and from heterogeneous groups of patients with type 2 diabetes, has shown a positive association between severe or non-severe hypoglycaemia and risk of CVD events or death [2, 3, 5, 6, 15,16,17,18,19,20,21,22,23,24,25,26,27]; these observations, however, have been inconsistent [25, 28]. In the attempt to quantify causality, previous studies adjusted for several potential confounders and explored associations with less-plausible outcomes (i.e. negative control outcomes). Virtually all, however, have estimated the risk of complications in terms of hazard ratio, which is difficult to interpret and lacks information on the absolute disease burden. In fact, in a population at a low absolute risk of an event, an intervention has smaller public health implications than in a population at a high absolute risk.
To our knowledge, few studies have specifically quantified the absolute risk of outcomes associated with severe hypoglycaemia in individuals with type 2 diabetes. Zhao et al designed a matched cohort study of 1522 participants with type 2 diabetes (mean age, 63 years) without cardiovascular and microvascular diseases using the Veterans Health Administration electronic health records. In 761 participants reporting a previous episode of any hypoglycaemia, the 3 year risks of fatal and non-fatal CVD events and mortality were 34% and 9%, respectively, while the 5 year risk of CVD events was about 43%; corresponding estimates at 3 years in the 761 participants without hypoglycaemia were 22% for CVD events and 7% for mortality . By applying the same study design, Jensen et al analysed data from 10,130 individuals with type 2 diabetes (mean age, 74 years): the 5 year risk of death was 62% in those admitted to hospital for severe hypoglycaemia (5605 individuals) and 37% in the matched participants without severe hypoglycaemia . In 1209 participants with type 2 diabetes (mean age, 64 years) enrolled in the Atherosclerosis Risk in Communities cohort study, Lee et al investigated the risks of CVD events and all-cause death associated with hypoglycaemia, defined as an episode resulting in an ambulance call, emergency department visit or hospitalisation. In the 195 participants with hypoglycaemia, the unadjusted risk 3 years after the hypoglycaemic episode was about 11% for coronary heart disease and 28% for mortality . Of note, in this study hypoglycaemia was associated with an increased relative risk of CVD mortality and cancer mortality but not non-CVD and non-cancer mortality. Similar to this study, other authors have reported the cumulative risk of CVD events or death without accounting for the heterogeneous clinical characteristics between patients with and without hypoglycaemia, potentially resulting in an overestimation of the difference [19, 21, 30]. Comparisons between these studies and our results are difficult because of the different definitions of hypoglycaemia, outcomes, population, design and analytical approach. In particular, in estimating the absolute risk, previous studies did not account for the competing nature of the events and the majority did not adjust for key confounders (i.e. age).
Our results are, however, in line with previous studies showing an increased risk for outcomes that may be considered negative controls [2, 6, 29]. In both patients with and patients without hypoglycaemia, the most common cause of death in our study was not a CVD event, as the combination of cancer and other causes accounted for around 71% of all deaths. Furthermore, the 5 year risk difference between hypoglycaemia and non-hypoglycaemia, adjusted for the heterogeneous clinical characteristics between the two groups, was higher for other causes of death (11%) than for CVD causes (5%). Among the other causes of death, the combination of ‘unspecified’, ‘vascular’ and ‘Alzheimer’s’ dementia accounted for 19% of all other causes of death in subjects with severe hypoglycaemia. We recognise that dementia may not be considered a negative control outcome as several vascular and nonvascular mechanisms could link hypoglycaemia to dementia ; yet, the same combination of the three causes accounted for a very similar proportion also in subjects without hypoglycaemia (i.e. 20%). As these two estimates are unadjusted and the epidemiological evidence in this area is contrasting [32,33,34,35], further studies should be conducted, ideally with a detailed phenotypical assessment of dementia to minimise coding misspecification.
In contrast with dementia, to our knowledge, no established mechanisms link hypoglycaemia to bronchopulmonary diseases, although they accounted for significant and very similar proportions of other causes of death in both subjects with and subjects without severe hypoglycaemia. The possibility of non-causality is further supported by the reduction in the estimated differences upon the inclusion of further potential confounders, suggesting the presence of residual confounding. This is in line with previous observational studies indicating that hypoglycaemia and CVD share several risk factors , as well as with the bidirectional nature of the association between hypoglycaemia and CVD [24, 36]. Together with the available evidence, our results would therefore strongly point towards a non-causal association between hypoglycaemia and long-term CVD complications in type 2 diabetes. However, regardless of the nature of the association, from a prognostic perspective our study identifies a group of patients with type 2 diabetes at a very high risk of death.
Our study extends previous research on the relationship between hypoglycaemia and risk of death. We defined a large and contemporary cohort of primary care subjects with type 2 diabetes, thus enhancing precision and generalisability of the results. We also focused on the absolute risk and accounted for the competing causes of death. As age is the strongest risk factor of death, we estimated the absolute risks and their differences for different ages; this enables age-specific prognostic comparisons with other risk factors or diseases (i.e. cancer).
This study also has some limitations. We defined hypoglycaemia as an episode of hospitalisation reporting hypoglycaemia as the underlying cause. From a clinical perspective, hypoglycaemia itself may not be the sole reason for admission; the decision to admit could be related to a patient’s clinical complexity, which is associated with an increased risk of death. Our results confirmed this hypothesis; we adjusted for several potential confounders to minimise confounding bias, yet our results would suggest that a residual confounding is still possible. We considered only admission to hospital because both severe hypoglycaemic episodes without hospitalisation and non-severe episodes, available only from CPRD, represent a more heterogeneous exposure. The definition used in this study and the necessity to link detailed clinical information to hospital data also resulted in a relatively small number of subjects with severe hypoglycaemia. Although the use of glucose-lowering medications associated with a lower risk or no risk of hypoglycaemia is increasing [37,38,39], the available evidence shows that the number of patients admitted to hospital for severe hypoglycaemia as the underlying cause was around 79,000 in England during the decade 2005–2014 ; this figure further highlights the public health relevance of our findings. Moreover, using this definition for the exposure may have resulted in an overestimation of the absolute risk of mortality in people admitted to hospital for a severe hypoglycaemic episode and, in turn, in a larger difference in the mortality risk compared with those without severe hypoglycaemia. In fact, severe hypoglycaemic episodes not requiring hospitalisation likely occur in individuals with a lower risk of complications and death. Whether the possible overestimation similarly applies to different causes of death, however, is difficult to envisage.
We used a prevalent cohort design and estimated the age-specific absolute risk of different causes of death in order to investigate the biological plausibility of the association between hypoglycaemia and cause-specific death; of note, when time since the last hypoglycaemic episode was included in the analysis, estimates did not materially change. An incident cohort design, whereby the occurrence of hypoglycaemia is the starting time of follow-up, is more relevant when the goal is to examine temporality, which is another criterion to judge causality. The temporal relationship of hypoglycaemia with CVD events and all-cause mortality has been variably investigated and reported, with some studies showing a constant, and others a reduced, relative risk over time [24, 27]. In this respect, it is also worth mentioning that we explored the long-term risk of mortality associated with hypoglycaemia; however, in acute or sub-acute clinical settings, hypoglycaemia may be the cause (i.e. injuries) or may contribute to a worse prognosis (i.e. infection) for ‘other’ causes of death. Further epidemiological studies should be specifically designed to explore temporality, accounting for the competing nature of the outcomes and the possible varying magnitudes of the associations between hypoglycaemia and each outcome over time. Lastly, similar to other large, observational studies using electronic health records, the risk of miscoding for both the exposure and the outcome cannot be excluded.
Severe hypoglycaemia resulting in hospitalisation is an event that carries a high long-term risk of death. As most of the causes of death in our study were not related to CVD, for which several pathophysiological mechanisms have been proposed, and a large set of potential confounders were accounted for, our results are highly suggestive for a non-causal relationship between severe hypoglycaemia and long-term CVD complications. Rather, severe hypoglycaemia is very likely a marker of frailty, which is causally associated with a higher risk of death. Along with the clinical principle of reducing glucose without causing hypoglycaemia to avoid short-term complications and preserve the quality of life in patients with diabetes , we underline that hospitalisation for hypoglycaemia identifies clinically complex phenotypes of type 2 diabetes: in these patients, further research should be conducted to identify the optimal strategies to reduce the risk of death.