Severe Hypoglycemia Identifies Vulnerable Patients With Type 2 Diabetes at Risk for Premature Death and All-Site Cancer: The Hong Kong Diabetes Registry

Hypoglycemia is an important but often neglected aspect in diabetes management. The increased risk of hypoglycemia during intensive blood glucose lowering in the ADVANCE (Action in Diabetes and Vascular Disease: Preterax and Diamicron Modified Release Controlled Evaluation) (1,2) and ACCORD (Action to Control Cardiovascular Risk in Diabetes) (3) studies have rekindled the subject of a risk–benefit ratio of intensive glycemic control in type 2 diabetes. In both the ADVANCE (2) and the ACCORD (4) studies, intensive blood glucose lowering was associated with an increased risk of severe hypoglycemia. Although in both studies researchers reported high rates of multiple adverse events, including hospitalizations and all-cause death in the 12 months after the incident event of hypoglycemia, detailed analysis did not reveal increased risk of cardiovascular disease (CVD) (2). Furthermore, although intensive treatment was associated with an increased risk of hypoglycemia, in both ACCORD (4) and ADVANCE (2), patients treated intensively tended to have a lower risk of death than those treated conventionally (4). These findings suggest that intensive monitoring in the former group might have prompted corrective actions to reduce adverse clinical outcomes.

On the other hand, hypoglycemia is known to increase the risk of CVD and mortality in diabetic patients (5) due to arrhythmia (5), abnormal hemostasis, and neurohormonal dysregulation, among other conditions. Other host factors such as age, comorbidities, and, notably, chronic kidney disease (CKD) might also influence clinical outcomes (6). In a hospital-based survey, age, CKD, presence of other complications, and institutionalizations were independent predictors for hospitalizations caused by hypoglycemia (7). Against this background, professional bodies and experts have recommended comprehensive assessments and individualizing treatment goals based on patient risk profiles, comorbidities, coping skills, cognitive states, and social support to maximize benefits and minimize harm (8,9).

Although the nature of the association between hypoglycemia and increased risk of death remain elusive, the 1.3- to 2.3-fold increased risk of all-site cancer and related death in type 2 diabetes has gained increasing attention (10). Our group first reported the linear association of glycated hemoglobin (HbA_1c) and nonlinear risk associations of lipids and BMI with all-site cancer in type 2 diabetes (11). Through the Hong Kong Diabetes Registry, a prospective cohort with detailed documentation of risk factors and clinical outcomes, we asked the question of whether type 2 diabetic patients with severe hypoglycemia exhibited particular phenotypes that might predict future events, including premature death and all-site cancer.

The methodology of the Hong Kong Diabetes Registry, established as a quality improvement program, has been previously described (11,12). Briefly, the registry was established in 1995 at the Prince of Wales Hospital, which serves a population of >1.2 million. Patients with diabetes attending medical clinics at the Prince of Wales Hospital can be referred to the Diabetes Mellitus and Endocrine Centre for comprehensive assessment based on the European DIABCARE protocol (13). Hong Kong has a heavily subsidized health-care system. The majority of patients with diabetes and chronic illnesses are managed in public hospitals governed by the Hospital Authority, which provides 95% of the total hospital bed-days in Hong Kong (14). After participants are enrolled in the registry, their outcomes, including hospital admissions, are monitored until their time of death (11,14). Up to 31 December 2007, 10,129 patients were enrolled in the cohort. After excluding 417 patients with type 1 diabetes (including missing data on classification of diabetes type) and 945 with missing variables used in the analysis, 8,767 patients were included in the present study.

At enrollment and regularly thereafter, type 2 diabetic patients underwent a comprehensive 4-h assessment for diabetes-related complications and risk factors (13). The assessment included an interview by diabetes nurses, anthropometric measurements, biochemical evaluations, fundus examination, and podiatry assessment. After fasting for 8 h overnight, blood was sampled for assays of fasting lipids, glucose, HbA_1c, and renal and liver function tests. Treatment goals were defined as HbA_1c <7%, blood pressure (BP) <130/80 mmHg, and LDL cholesterol (LDL-C) <2.6 mmol/L (12). We used spot urinary albumin-to-creatinine ratio (ACR) to define albuminuria (ACR ≥2.5 mg/mmol in men and ≥3.5 mg/mmol in women). The abbreviated Modification of Diet in Renal Disease equation recalibrated for Chinese (15) was used to define CKD as an estimated glomerular filtration rate (eGFR) <60 mL/min/1.73 m². All laboratory analyses were performed with standard methods in the Department of Chemical Pathology of the Prince of Wales Hospital. The laboratory is accredited by the Australian National Association of Testing Authorities. Informed written consent was obtained from all participants, and the study was approved by The Chinese University of Hong Kong Clinical Research Ethics Committee.

Severe hypoglycemia was defined as one or more hospitalizations for hypoglycemia in the 12 months before enrollment (at enrollment) or during the follow-up period from enrollment to death or 31 January 2009 to ensure that the last enrolled patient in 2007 had at least 2 years of observation. We ascertained all clinical outcomes through the Hospital Authority Central Computer Management System, which records diagnoses of all hospital discharges, including mortality, based on ICD-9 codes. The mortality data were cross-checked with the Hong Kong Death Registry, and the cause of death was defined by the principal discharge diagnosis.

The statistical analysis was performed with SAS release 9.30 software (SAS Institute Inc., Cary, NC), unless otherwise specified. Cox proportional hazards analysis was used to obtain hazard ratios (HRs) and 95% CIs. Data analysis was divided into two parts: 1) identification of risk factors for severe hypoglycemia and 2) identification of cancer-associated subphenotypes for severe hypoglycemia.

Follow-up time was calculated as the period from enrollment to the date of the first admission for severe hypoglycemia, death, or 31 January 2009, whichever came first. We also ascertained the temporal relationship among severe hypoglycemia, all-site cancer, and death. A two-step scheme was used to identify risk factors for severe hypoglycemia. First, we examined the demographic, lifestyle, and clinically relevant factors measured at enrollment, including age, sex, BMI, high BP (systolic BP ≥130 mmHg or diastolic BP ≥80 mmHg), suboptimal glycemic control (HbA_1c ≥7.0%), high LDL-C (≥2.60 mmol/L), low HDL cholesterol (HDL-C) (<1.03 mmol/L in men and <1.29 mmol/L in women), high triglyceride (TG) (≥1.70 mmol/L), albuminuria, and CKD. Second, we further adjusted for drug use at enrollment, including insulin, oral glucose-lowering drugs, lipid-lowering drugs, and renin-angiotensin system (RAS) inhibitors.

Because we detected an increased occurrence of all-site cancer in patients who developed severe hypoglycemia in the first stage of analysis, we further examined risk associations between severe hypoglycemia and cancer phenotypes previously reported by our group. In brief, we identified five phenotypes in the Hong Kong Diabetes Registry predictive of incident cancers in type 2 diabetes in addition to the linear association between HbA_1c and cancer (16). These cancer phenotypes included 1) high LDL-C (≥3.80 mmol/L) (11), 2) copresence of low LDL-C (<2.60 mmol/L) and albuminuria (17), 3) copresence of low LDL-C (<2.60 mmol/L) and low TG (<1.70 mmol/L) (18), 4) low HDL-C (<1.0 mmol/L) (19), and 5) high and low BMI (≥27.6 and <24.0 kg/m²) (20). By examining their associations with severe hypoglycemia, we aimed to ascertain whether the high incidences of cancer in patients who developed severe hypoglycemia could be predicted by these lipid phenotypes.

To assess whether the type of oral antidiabetic agents used in this study population, specifically sulfonylureas, compared with other agents with less risk of hypoglycemia, such as metformin, would have an impact on the analyses, we performed a sensitivity analysis among patients enrolled between December 1996 and January 2005, and followed up to July 2005 (because we censored data related to drug use during this period). We used this analysis to check whether the results were valid after further adjustment for drug use, similar to what we did and reported previously (19).

The plots of LOG [−LOG (survival function)] versus LOG (follow-up time in years) were used to check the proportional hazards assumption for categorical variables, whereas the Supremum test was used to check the assumption for continuous variables (21). P < 0.05 was considered statistically significant.

The mean age of the cohort was 57.4 (SD 13.2) years, with a median duration of diabetes of 5 (interquartile range 1–11) years. During a total of 59,375 years (median 6.71 [interquartile range 3.47–10.38] years) of follow-up, 235 patients were hospitalized for hypoglycemia, with an incidence of 3.96 (95% CI 3.45–4.46) per 1,000 patient-years. Compared with patients without severe hypoglycemia, patients with severe hypoglycemia were older, were more likely to be ex-smokers and ex-drinkers, had longer disease duration, and had lower BMI. They also had higher HbA_1c and BP but lower TG and LDL-C and worse renal function and were more likely to have a history of CVD and insulin treatment at enrollment (Table1).

After adjustment for confounders, old age, low BMI, high HbA_1c, high BP, low LDL-C, low TG, albuminuria, and CKD were independent risk factors for severe hypoglycemia. After further adjustment for drug use at enrollment, a 10-year increase in age was associated with an 86% increased risk, whereas 1 kg/m² BMI increase was associated with 5% reduced risk of severe hypoglycemia. Patients with suboptimal glycemic control (HbA_1c ≥7%) were at a 1.78-fold risk of experiencing severe hypoglycemia compared with those with HbA_1c <7%. Nonattainment of LDL-C (≥2.60 mmol/L) and TG (≥1.7 mmol/L) targets was associated with reduced risks of severe hypoglycemia (HR 0.71 [95% CI 0.53–0.96] and 0.70 [0.51–0.95], respectively). Apart from age, low BMI, high HbA_1c, albuminuria, and CKD, low TG and low LDL-C remained highly predictive of severe hypoglycemia after adjustment for drug use (Table 2).

The consistent associations of low HDL-C (<1.0 mmol/L), low TG (<1.7 mmol/L), and low LDL-C (<2.6 mmol/L) with severe hypoglycemia in the two multivariable analyses prompted us to investigate whether these lipid phenotypes, reported to be associated with all-site cancer by our group (11,19,22,23), might also be associated with hypoglycemia. These relationships were nonlinear, with both high and low TG, HDL-C, and LDL-C values being associated with cancer risk, even in patients not using lipid-regulating drugs (11,19,22,23). After adjusting for all covariables, the HR of HDL-C <1.0 mmol/L versus HDL-C of 1.10–1.29 mmol/L (reference) was 1.88 (1.37–2.58) and that of low LDL-C plus low TG was 1.69 (1.15–2.47) (Table 3). Other cancer phenotypes were not found to be predictive of severe hypoglycemia.

The significant results in multivariable analysis remained statistically significant, except for albuminuria, which became insignificant (P = 0.1025) after adjustment for drug use in the 2.5 years before enrollment and during follow-up (Table 4).

In this large prospective cohort of type 2 diabetes, the rate of severe hypoglycemia requiring hospitalization was 3 per 1,000 patient-years. Similar to the ACCORD and ADVANCE studies, this cohort had high rates of death, including all-site cancer. After adjusting for confounders, apart from old age, high HbA_1c, and CKD, previously reported cancer-associated phenotypes, including low BMI, low HDL-C, low LDL-C, and low TG, were also predictors of severe hypoglycemia. These results concurred with those reported in the ADVANCE study (2), where old age, long disease duration, high serum creatinine levels, low BMI, low cognitive function, use of two or more oral blood glucose–lowering drugs, history of smoking or microvascular disease, and assignment to intensive glucose control were all independent risk factors for severe hypoglycemia.

The nonlinear relationship between BMI and premature mortality, including cancer deaths, has been widely reported (24). The associations between cancer and CKD is now emerging, which may be due to sharing of common risk factors and improved survival from CVD (25). Low BMI may denote a state of catabolism and malnutrition, which are often found in patients with occult malignancy and renal impairment. Besides, CKD can be both a cause and a consequence of hypoglycemia. Aside from reduced glucose production (about one-fifth of the plasma glucose originated from renal gluconeogenesis), diabetic patients with CKD often have altered drug metabolism, autonomic neuropathy, and hypoglycemic unawareness, which culminate and place them at high risk of hypoglycemia (6).

Based on 24-h monitoring of blood glucose and electrocardiogram, silent hypoglycemia and cardiac ischemia were not uncommon in type 2 diabetic patients who had long disease duration, prior CVD, and insulin treatment (26). Thus, when the ACCORD study was stopped prematurely because of an increased death rate in the intensively treated group, hypoglycemia-associated cardiovascular death was considered a possible culprit (27). Indeed, there were many plausible mechanisms for hypoglycemia to increase the risk of CVD, including sympathetic activation, prolongation of QT interval, increased thrombogenesis, inflammation, and vasoconstriction (28). However, subsequent analysis of ACCORD (4) failed to confirm increased CVD-related death in patients treated intensively. Indeed, in patients without prior CVD and HbA_1c <8%, intensive treatment was associated with a trend of reduced CVD (4). Further analysis of the ADVANCE databases revealed that diabetic patients with severe hypoglycemia were more likely to die of non-CVD causes, including cancer, compared with those without severe hypoglycemia (2). In support of these findings, patients with severe hypoglycemia in the present observational cohort had a higher mortality rate mainly attributable to cancer, with rates of CVD similar to patients without hypoglycemia.

The present findings and those of others (2) thus raise the possibility that severe hypoglycemia might actually reflect the consequences of coexisting conditions and denote a subgroup of diabetic patients who are particularly fragile and deserve more clinical attention. A novel observation from this study is that cancers occurred more frequently among diabetic patients who had developed severe hypoglycemia during the follow-up period. This adverse event could be predicted by our previously reported cancer subphenotypes, notably low BMI, low TG, and low LDL-C. Large-scale randomized studies have reported an annual cancer rate of 1% in type 2 diabetes (ORIGIN [Outcome Reduction With Initial Glargine Intervention]), although few studies have reported an association between hypoglycemia and cancer except as a terminal event in patients with metastatic disease (29).

Our group was the first to report the high rates of cancer in type 2 diabetic patients, which accounted for 20% of all-cause death (30). Using the same registry, apart from the linear association of cancer with HbA_1c levels (16), we also reported nonlinear associations between cancer and lipid subphenotypes (11). From Hong Kong Diabetes Registry data, we reported an association between low TG level (<1.7 mmol/L) and increased cancer risk that was attenuated by statin use (23), implicating that the activation of the hydroxymethylglutaryl-CoA reductase with increased production of Ras signals, which are known to be oncogenes, may be involved (23). In this connection, sterol regulatory element-binding proteins (SREBPs), regulated by both insulin and insulin-like growth factor 1 (IGF1), are involved in lipid synthesis and may play a plausible biological link for cancer development in type 2 diabetic patients (23). Based on these observations and experimental studies by others, we argued that dysregulation of lipid and glucose metabolism with activation of the SREBP-IGF1-hydroxymethylglutaryl-CoA reductase pathway might be a possible mechanism (23,31,32).

Apart from highlighting the high-risk nature of type 2 diabetic patients with severe hypoglycemia, the present findings might have implications for clinical management. In this analysis, patients with severe hypoglycemia were less likely to be at HbA_1c or BP goals and had renal dysfunction despite being more likely to be taking insulin and RAS inhibitors. These difficult-to-treat patients might have other unmet needs, such as treatment noncompliance and psychological distress. Of note, several studies have now reported high rates of depression associated with hypoglycemia in type 2 diabetes (33). These epidemiological data once again highlight the complex and high-risk nature of type 2 diabetic patients with hypoglycemia who would require comprehensive assessment for individualized care.

On the other hand, patients with hypoglycemia were more likely not to reach LDL-C and TG treatment goals despite having similar usage rates of lipid-lowering drugs as those without hypoglycemia. Given the predictive values of low TG and low LDL-C levels on severe hypoglycemia, physicians might need to be more cautious when interpreting these treatment goals, especially in patients with long disease duration, low BMI, and CKD. The nature of the association between low HDL-C level and hypoglycemia was not clear, although we have reported that reduced renal function might predict low HDL-C (23). Taken together, understanding these complex interrelationships among renal impairment, dyslipidemia (low HDL-C, low LDL-C, low TG), severe hypoglycemia, and cancer may shed light on the nature of comorbidities in type 2 diabetes (10).

The strengths of this study are its large sample size, long follow-up duration, and detailed documentation of risk factors, complications, drug use, and outcome within a heavily subsidized health-care system with a high retention rate. These real-world data confirm findings collected in clinical trial settings and provide alternative perspectives regarding the clinical significance of severe hypoglycemia with possible therapeutic implications. On the other hand, we acknowledge the retrospective and hypothesis-generating nature of epidemiological analysis and its inherent limitations. The ascertainment of hypoglycemia through ICD-9 codes might contain errors due to misclassification and/or undercoding. However, there is a general consensus that administrative data for clinical outcomes are generally satisfactory, with an accuracy rate as high as 95% for acute myocardial infarction and stroke (34,35). Furthermore, the impact of the use of oral antidiabetic agents on severe hypoglycemia shown in the present sensitivity analysis should be interpreted with caution because the use of observational studies to address drug effects might introduce potential bias, requiring studies designed specifically to confirm the results (36). Finally, we only included severe hypoglycemic episodes requiring hospitalizations, which were specific but not sensitive indicators of these adverse events. Because this database did not systematically capture minor hypoglycemia events not requiring admission, it is likely that we underestimated the true rates of hypoglycemia. The analysis did not confirm the risk association between severe hypoglycemia and CVD-related deaths, but we cannot exclude the possibility that repeat minor hypoglycemic episodes could have cumulative adverse effects on cardiovascular and cerebrovascular functions with direct or indirect impacts on mortality. In addition, there are different definitions of severe hypoglycemia adopted by various research groups, and the commonly used one would be a hypoglycemic event requiring assistance of another person to actively administer carbohydrates, glucagons, or other resuscitative actions. The definition we used in this study might not be useful in a comparison with other cohort data using different definitions.

In conclusion, in this real-world pragmatic study, we identified risk factors predictive of severe hypoglycemia in type 2 diabetes, including old age, low BMI, high HbA_1c, low TG, low LDL-C, albuminuria, and CKD. Patients with severe hypoglycemia were also more likely to die and to develop all-site cancer (2) than those without severe hypoglycemia. Given the complexity and multiple comorbidities of type 2 diabetes, experts have emphasized the importance of periodic phenotyping, individualizing treatment goals, and comanagement of specialist and primary care, especially in difficult-to-treat patients such as those with hypoglycemia, poor glycemic control, and comorbidities. To this end, the present analysis has provided additional evidence to support these expert recommendations, highlighting the vulnerability of patients with severe hypoglycemia who are at high risk for all-cause death and all-site cancer.

Acknowledgments. The authors thank all medical, nursing, and research staff at the Diabetes Mellitus and Endocrine Centre, Prince of Wales Hospital, for conducting this study.

Funding. This study was supported by the Li Ka Shing Institute of Health Science and Hong Kong Foundation for Research and Development in Diabetes under the auspices of The Chinese University of Hong Kong.

Duality of Interest. A.P.S.K. has received honoraria for consultancy or giving lectures from Novo Nordisk, Eli Lilly, Merck Serono, and Nestlé. R.C.W.M. has received research grants and/or honoraria for consultancy or giving lectures from AstraZeneca, Boehringer Ingelheim, Danone, Eli Lilly, Nestlé, Pfizer, and Sanofi. J.C.N.C. has received a research grant and/or honoraria for consultancy or giving lectures from Bayer, Boehringer Ingelheim, Daiichi-Sankyo, Eli Lilly, GlaxoSmithKline, Merck Sharp & Dohme, Merck Serono, Pfizer, AstraZeneca, Sanofi, Novo Nordisk, and Bristol-Myers Squibb. The proceeds have been donated to The Chinese University of Hong Kong, American Diabetes Association, and other charity organizations to support diabetes research and education. No other potential conflicts of interest relevant to this article were reported.

Ethical Approval. The Joint Chinese University of Hong Kong and New Territories East Cluster Clinical Research Ethics Committee.

Author Contributions. A.P.S.K. collected data and wrote the manuscript. X.Y. analyzed data and wrote the manuscript. A.L., R.C.W.M., W.Y.S., R.O., R.T., K.C., and C.C.C. collected data. C.S.H. and M.H.M.C. supervised the quality control and advised on the laboratory data analysis. J.C.N.C. collected data and reviewed and edited the manuscript. A.P.S.K. and X.Y. are the guarantors of this work and, as such, had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.