Predictors of Nonsevere and Severe Hypoglycemia During Glucose-Lowering Treatment With Insulin Glargine or Standard Drugs in the ORIGIN Trial

Hypoglycemia is an important adverse effect of treatment with glucose-lowering agents such as insulin and sulfonylureas. Although only a subset of patients to whom these agents are prescribed experience hypoglycemia, these episodes limit the level of glycemic control that can be achieved, affect quality of life, and may lead to loss of consciousness and hospitalization. Moreover, epidemiologic studies have shown that people who experience hypoglycemic events are also at increased risk of a wide range of adverse health consequences including cardiovascular (CV) events, cancers, and death (1–5). Whether nonsevere and severe hypoglycemia occur in similar settings and require similar preventive tactics is unclear. These observations highlight the importance of identifying clinical characteristics that are associated with a higher risk of either nonsevere or severe hypoglycemia, so that susceptible individuals can be identified and provided with strategies to reduce the risks of both hypoglycemic events and their potential consequences.

The Outcome Reduction with an Initial Glargine Intervention (ORIGIN) trial compared treatment with titrated basal insulin glargine targeting normal fasting glucose levels with standard oral agent–based therapy for a median period of 6.2 years, in a population of 12,537 people with dysglycemia and additional CV risk factors (6,7). As previously reported, the glargine intervention had a neutral effect on CV and cancer outcomes (7). Although absolute rates were low in both treatment groups, hypoglycemia was more common with glargine treatment than with standard care (7,8). Because both nonsevere and severe hypoglycemic events were carefully ascertained during long-term treatment, the ORIGIN trial provides a unique opportunity to identify clinically important baseline risk factors for both categories of hypoglycemia, and to assess the independent effects of allocation to glargine and on-treatment HbA_1c levels on such events.

As previously reported, the ORIGIN trial was a large, randomized, 2 × 2 factorial design trial that tested the effects of two pairs of treatments on CV outcomes in people with high CV risk and dysglycemia (6,7). People from 40 countries who were ≥50 years of age and had impaired fasting glucose levels, impaired glucose tolerance, or previously known or recently detected type 2 diabetes were enrolled in the study between 2003 and 2005. Those with a prior diagnosis of diabetes could be taking no more than one oral glucose-lowering agent together with a lifestyle intervention. Participants were randomized to either add insulin glargine to prior treatment, with titration of the dosage seeking a fasting plasma glucose level of ≤5.3 mmol/L (≤95 mg/dL), or to use standard glycemic care beginning with oral therapies according to local guidelines. Prior glucose-lowering therapies were usually continued after the initiation of randomized treatment. Usage of specific glucose-lowering agents at study enrollment and after randomization has been reported previously (9). Participants were also randomized to take either 1 g of n-3 fatty acids daily or a placebo. The main end point of the comparison of glargine with standard care was a composite of nonfatal myocardial infarction, stroke, or CV death.

During a median on-treatment follow-up period of 6.2 years (interquartile range 5.8–7.2 years), participants kept diaries recording symptoms suggesting hypoglycemia and self-measured plasma-referenced glucose values at the time of symptoms. These were transcribed at clinic visits at 0.5, 1, 2, and 4 months, and every 4 months thereafter. Nonsevere confirmed hypoglycemia was a symptomatic event with concurrent glucose measurement of ≤3.0 mmol/L (≤54 mg/dL), but not requiring assistance by another person. Severe hypoglycemia was defined as requiring assistance, with either prompt recovery after oral administration of a carbohydrate, intravenous glucose, or glucagon, or documentation by self-measurement or laboratory measurements of plasma glucose levels of ≤2.0 mmol/L (≤36 mg/dL).

All incidence rates were expressed as the number of incidents per 100 person-years of follow-up. To determine the yearly incidence of hypoglycemia, the number of person-years of exposure for each participant was calculated as the number of years until the first hypoglycemic event. The incidence of at least one episode of hypoglycemia was estimated as the number of people having experienced one or more episodes divided by the total number of person-years of observation. The annualized incidence of all hypoglycemic events (e.g., with an individual counted twice if they had experienced two events, three times if they had experienced three events, etc.) was estimated as the total number of events divided by the number of person-years of follow-up. For each individual who had experienced at least one hypoglycemic event, the annualized incidence of all hypoglycemic events during the study divided by the number of person-years of follow-up contributed by that individual was estimated and the mean (SD) and median (interquartile range) was calculated. Continuous variables were expressed as means and SDs or medians and interquartile ranges, and were compared using t tests or Kruskal-Wallis tests as appropriate. Baseline categorical variables were expressed as numbers and percentages, and were compared using χ² tests.

Cox proportional hazards models were used to estimate the unadjusted hazard of confirmed nonsevere and severe hypoglycemic events for 33 baseline demographic, medical history, examination, medication use, and laboratory characteristics (which have been linked to hypoglycemia in other epidemiologic studies), as well as baseline treatment allocation and the updated HbA_1c level (estimated using the mean of all of the available HbA_1c levels from the time of randomization until the end of the study). All hazard ratios (HRs) were calculated using a frailty model that accounts for the multiplicity of events and the propensity of experiencing a second event in a person who has already experienced one event (10). Variables for which the univariate HR was associated with a P value of <0.1 were then included in multivariable analyses using the frailty model. The effect of a diagnosis of diabetes prior to participation in the ORIGIN trial and of the updated HbA_1c level was assessed by estimating multivariable frailty models with and without these variables. Whether the relationship between updated HbA_1c levels and the risk of hypoglycemia differed in participants allocated to receive insulin glargine versus standard care was assessed by adding an interaction term (updated HbA_1c*allocation) to the model. The relationship between the independent variables and the risk of a hypoglycemic event was expressed as an HR and its 95% CI.

Kaplan-Meier plots illustrating the cumulative incidence of the first nonsevere hypoglycemic event in people allocated to receive glargine versus standard care, and in people with a history of diabetes versus no diabetes were plotted. Plots illustrating the relationship between the incidence of hypoglycemic events and the updated HbA_1c level were estimated from a multivariable model based on just the glargine participants and another model based on just the standard care participants. For each group, the HRs per HbA_1c increment from the model were used to generate an estimated incidence of hypoglycemia at each updated HbA_1c level with reference to the observed rate of hypoglycemia at the median updated HbA_1c level, and expressed as the percentage per year. All statistical analyses were performed using SAS version 9.2 (SAS Institute), and figures were generated using SPlus version 8.1.1 (TIBCO).

As previously reported (7), 5,155 of 12,537 randomized participants (41%) experienced at least one symptomatic hypoglycemic event. As noted in Table 1, at least one hypoglycemic event that was confirmed by glucose measurement but did not require third-party assistance (nonsevere event) was reported for 3,518 participants (28%). Larger proportions of participants randomized to receive glargine than standard care experienced nonsevere events (42% [2,614/6,264] vs. 14% [904/6,273]; mean yearly proportions 9.8% vs. 2.7%). Also, more participants with diabetes at baseline experienced nonsevere events compared with those without diabetes (30% [3,296/11,081] vs. 15% [222/1,456]; mean yearly proportions 6.3% vs. 2.8%). Patterns were similar for severe events. A total of 472 participants (3.8%) had experienced at least one severe hypoglycemic episode, with the proportions of participants affected yearly being 1.00% vs. 0.31% for glargine versus standard care, and 0.68% vs. 0.42% for individuals with or without diabetes at baseline. The time course of the appearance of the first nonsevere and first severe events during treatment is shown in Fig. 1. The frequency of new events was consistently higher with glargine treatment and for people with diabetes versus those without diabetes during the whole period of follow-up.

The 3,518 participants with at least one confirmed nonsevere event had a total of 24,680 events during the study, with a mean (SD) of 1.2 (2.1) events per person-year (Table 1). For participants allocated to treatment with glargine vs. standard care, the annualized frequency of events was 1.3 vs. 0.8 (P < 0.001) yearly, respectively. Similarly calculated yearly rates for diabetes versus no diabetes at baseline were 1.2 vs. 0.8 (P < 0.008). A similar analysis for severe hypoglycemia (Table 1) revealed that the 472 participants affected reported 591 events. While 359 participants allocated to glargine reported at least one severe event, 113 of those receiving standard care did so. Of participants with and without diabetes at baseline, 436 and 36, respectively, had experienced severe events. Among those participants who had experienced at least one severe event, the yearly rates were 0.3 and 0.2 (P = 0.8) for glargine versus standard treatment, and also 0.3 vs. 0.2 (P = 0.3) for diabetes versus no diabetes.

People with both nonsevere and severe events were more likely to be using a sulfonylurea and to have higher HbA_1c levels at baseline. Those who had experienced nonsevere events generally were younger, whereas those who had experienced severe events were older. A complete listing of unadjusted associations with nonsevere and severe events is shown in Supplementary Table 1.

The results of Cox proportional hazards frailty models (which account for multiple events in a given individual) describing independent associations of risk with various clinical factors for each category of hypoglycemia are shown in Table 2. The models included baseline characteristics selected for having univariate associations with events with a P value of <0.1, and also allocation to glargine treatment versus standard care. Both use of a sulfonylurea at study enrollment and allocation to glargine treatment were independently associated with an increased risk of both nonsevere and severe events. Specifically, HRs for nonsevere hypoglycemia were 2.07 (95% CI 1.78–2.40) for sulfonylurea use and 4.53 (4.00–5.13) for glargine allocation, while for severe events they were 1.35 (1.07–1.70) and 3.57 (2.80–4.55), respectively, for sulfonylurea use and glargine allocation. Additional independent predictors of hypoglycemia depended on whether nonsevere or severe events were being analyzed. For nonsevere hypoglycemia, independent predictors included younger age, South Asian ethnicity, depression, lower BMI, higher LDL level, diabetes at study enrollment, metformin use, and higher baseline HbA_1c level. In contrast, independent predictors of severe hypoglycemia included older age, no more than 12 years of education, hypertension, lower triglyceride and higher creatinine levels, and greater impairment of cognitive function. Similar predictors were noted in a supplemental analysis in which all univariate risk factors for either nonsevere or severe hypoglycemia were included in a single multivariable model that was run to identify independent risk factors for each category of hypoglycemia (Supplementary Table 2). In this unified model, nonsevere events themselves were independently associated with severe hypoglycemia (HR 2.39 [95% CI 1.88–3.04], P < 0.001).

A highly significant interaction between updated HbA_1c level and treatment allocation was noted for both nonsevere (P = 0.004) and severe (P < 0.001) hypoglycemia. To further explore the relationship of on-treatment HbA_1c level with hypoglycemia, separate fully adjusted models were generated for each treatment group and used to estimate curves across the range of on-treatment HbA_1c values for both nonsevere and severe events (Fig. 2). Progressively higher on-treatment HbA_1c values were associated with a lower risk of nonsevere events in both treatment groups; a lower risk of severe events in the glargine group; and a higher risk of severe events in the standard care group.

These analyses provide a detailed description of the determinants of hypoglycemia during a median follow-up period of 6.2 years in people who at baseline had moderately elevated glucose levels while taking only 0 or 1 oral glucose-lowering drugs. They show that, regardless of the assigned therapy, the absolute incidence of hypoglycemia of any sort was low. Thus, despite a threefold higher rate of hypoglycemia, the annual incidences of confirmed nonsevere and severe hypoglycemia in people allocated to insulin glargine–mediated fasting normoglycemia were only 10% and 1% respectively. Nonsevere hypoglycemia was more likely to occur in people with diabetes than in those without diabetes, regardless of whether normal fasting glucose levels were sought with basal insulin administration. The fact that the rates of hypoglycemia were even lower among people without diabetes than those with diabetes suggests that, in addition to promoting hyperglycemia, diabetes may impair the ability to avoid hypoglycemia.

Allocation to treatment with glargine and the use of a sulfonylurea at baseline were independent risk factors for both nonsevere and severe hypoglycemia. However, additional risk factors for nonsevere and severe hypoglycemia clearly differed, with no independent predictors, other than glargine assignment and prior use of a sulfonylurea, shared by the two categories. Thus, consistent with other reports (11,12), younger age, lower BMI, depression, a history of diabetes, a higher baseline HbA_1c level, and a lower HbA_1c level achieved during therapy were independently associated with nonsevere hypoglycemia. In contrast, the independent determinants of severe hypoglycemia were older age, limited education, hypertension, renal insufficiency, and mild cognitive impairment. These observations are also consistent with those of other reports (5,13–17). Stated in clinically relevant terms, these analyses suggest that an individual with a BMI of 25 kg/m² would be ∼30% more likely to have nonsevere hypoglycemia than one with a BMI of 35 kg/m², but with no difference in the risk of severe hypoglycemia. Also, an individual who is 70 years old would be predicted to have ∼40% less risk of nonsevere hypoglycemia but 80% higher risk of severe hypoglycemia compared with a person who is 50 years of age.

Of note is the relationship between the HbA_1c level achieved during the trial, and the risk of nonsevere and severe events with the two forms of treatment. In the glargine group, lower achieved HbA_1c levels were associated with a higher risk of both types of hypoglycemia. A similar relationship was noted for the standard care group with respect to nonsevere hypoglycemia. However, the relationship during standard care was reversed for severe hypoglycemia, with higher achieved HbA_1c levels associated with a higher risk of severe hypoglycemia. Such a phenomenon has been noted in other studies (12–14) and may reflect an increased susceptibility to hypoglycemia due to sulfonylurea use or (in <10% of the standard care participants in the ORIGIN trial) use of insulins other than glargine (9) among those people who did not easily maintain good glycemic control.

Finally, nonsevere hypoglycemia itself was an independent predictor of the occurrence of severe hypoglycemia, a finding that is not unexpected. It is unknown whether this association is due more to the effects of unmeasured covariates, such as behavioral factors leading to glycemic variability, or to hypoglycemia-induced unawareness of falling glucose levels. Further study of the relationships between the occurrence of nonsevere hypoglycemia and subsequent severe hypoglycemia or its medical consequences is warranted.

These findings add to a growing body of evidence suggesting that, rather than just representing two parts of a range of below-normal glucose levels, hypoglycemic events requiring or not requiring third-party assistance tend to occur in different situations, have differing implications for clinical management, and should be evaluated separately. Nonsevere events can occur when otherwise healthy people with type 2 diabetes are seeking nearly normal levels of glycemic control, especially with sulfonylurea or insulin treatment. Although they did not independently predict an increased risk of CV events in the ORIGIN trial (8), they may interfere with quality of life and the attainment of glycemic treatment goals. In contrast, because severe hypoglycemia is more likely among older people and individuals with established medical problems who are attempting to maintain excellent glycemic control, more conservative glucose-lowering regimens and targets should be considered for this population. The finding that severe hypoglycemia during standard care was more likely when the HbA_1c level was above the usual target levels, together with the prior observation (8) that severe hypoglycemia in the ORIGIN trial was two to three times more strongly linked to subsequent CV events in people allocated to receive standard care rather than glargine therapy, suggests that severe hypoglycemia identifies a highly vulnerable population at risk for serious medical problems. Similar observations from other large outcomes trials (4,18–20) in people with type 2 diabetes provide further support for this view. People who have experienced even a single severe hypoglycemic episode clearly deserve increased attention, with the aim of mitigating the risk of both later severe hypoglycemia by adjusting glucose-lowering medications and of other outcomes, including both CV and non-CV events, by optimizing cardioprotective and other preventive therapies. These observations further suggest that the current nomenclature of hypoglycemia, which describes a continuum based on glucose levels, might be reconsidered to better distinguish between events that interfere with management but do not predict serious outcomes and those that are strongly linked with adverse outcomes, whether causally or by association.

The strengths of this investigation include the large and geographically diverse population studied, the median follow-up period of >6 years, and (unlike most prior studies of hypoglycemia) the prospectively defined data collection that provided enough events to allow the comparison of nonsevere and severe hypoglycemia. The randomized treatment comparison permitted evaluation of two forms of treatment. Among the limitations of the study are the post hoc design of the analytic plan, the likelihood that some hypoglycemic events associated with a glucose level of ≤3.0 mmol/L (≤54 mg/dL) were unreported by participants, and the lack of generalizability to people with type 1 diabetes or a longer duration of type 2 diabetes. The occurrence of glucose levels <3.9 mmol/L (<70 mg/dL), whether symptomatic or not, was not systematically ascertained in the ORIGIN trial, and so the potential significance of such events is not clarified by these findings.

In summary, individuals with early dysglycemia enrolled in the ORIGIN trial had relatively low absolute rates of hypoglycemia that were nevertheless higher when normal glucose levels were targeted with administration of insulin glargine and sulfonylureas. Moreover, nonsevere and severe hypoglycemic events differed in ways beyond glucose levels and have different implications for individualizing therapy.

Writing Committee for the ORIGIN Trial Investigators. Matthew C. Riddle (Department of Medicine, Oregon Health & Science University, Portland, OR); Hertzel C. Gerstein and Hyejung Jung (Department of Medicine and Population Health Research Institute, McMaster University and Hamilton Health Sciences, Hamilton, Ontario, Canada); Linda G. Mellbin and Lars Rydén (Department of Medicine Solna, Karolinska Institute, Stockholm, Sweden); Julio Rosenstock (Dallas Diabetes and Endocrine Center at Medical City and University of Texas Southwestern Medical Center, Dallas, TX); and Jeffrey Probstfield (Department of Medicine, University of Washington, Seattle, WA).

Acknowledgments. An international steering committee designed and conducted the trial, and all data were collected and analyzed by the ORIGIN Project Office located at the Population Health Research Institute in Hamilton, Ontario, Canada. The authors thank Rose Hastings and Russell Standley Memorial Trusts for their support in the preparation of the manuscript.

Funding. The ORIGIN Trial was funded by Sanofi and Pronova BioPharma, Norway, which provided n-3 fatty acid supplements and placebo.

Duality of Interest. M.C.R. reports receiving research grant support from Sanofi, AstraZeneca, Eli Lilly, and Novo Nordisk; honoraria for speaking from Sanofi; and honoraria for consulting from Sanofi, AstraZeneca, Eli Lilly, Elcelyx, and Valeritas. These dualities of interest have been reviewed and managed by Oregon Health & Science University. H.C.G. reports receiving consulting fees from Sanofi, Novo Nordisk, Lilly, Bristol-Myers Squibb, Roche, AstraZeneca, and GlaxoSmithKline; lecture fees from Sanofi and Bayer; and support for research or continuing education through his institution from Sanofi, Lilly, Takeda, Novo Nordisk, Boehringer Ingelheim, Bristol-Myers Squibb, and AstraZeneca. H.J. declares no potential conflicts of interest relevant to this article. L.G.M. reports receiving research grant support from the Swedish Heart Lung Foundation, the Swedish Diabetes Association, Merck, Bayer AG, and Sanofi; and lecture honoraria from Merck, Sanofi, Novartis, Bayer Schering, AstraZeneca, Roche, and Eli Lilly. L.R. reports receiving research grant support from the Swedish Heart Lung Foundation, the Swedish Diabetes Association, Roche, and Bayer AG; and lecture/consulting honoraria from Sanofi, Bristol-Myers Squibb, AstraZeneca, and Merck. J.R. reports receiving consulting fees from Sanofi, Novo Nordisk, Eli Lilly, GlaxoSmithKline, Takeda, Merck, Daiichi Sankyo, Janssen, Novartis, Boehringer Ingelheim, MannKind, Halozyme, Intarcia, and Lexicon; and research grants from Merck, Pfizer, Sanofi, Novo Nordisk, Roche, Bristol-Myers Squibb, Eli Lilly, GlaxoSmithKline, Takeda, Novartis, AstraZeneca, Amylin, Janssen, Daiichi Sankyo, MannKind, Boehringer Ingelheim, Intarcia, and Lexicon. J.P. reports receiving honoraria for consulting from Sanofi and research grant support from Sanofi, Bristol-Myers Squibb, and AstraZeneca. No other potential conflicts of interest relevant to this article were reported.

These potential conflicts of interest have been reviewed and managed by Oregon Health & Science University. The authors acknowledge the collaborative relationship with Sanofi and the Population Health Research Institute, which made the study and this analysis possible.

Author Contributions. The current analyses were conceived and executed by the Writing Committee. M.C.R. helped to lead the design of the analyses, had access to the results of the analyses, wrote the first draft of the manuscript, contributed to revisions of the manuscript, and helped to make the decision to submit it for publication. H.C.G., L.G.M., and L.R. helped to lead the design of the analyses, had access to the results of the analyses, contributed to revisions of the manuscript, and helped to make the decision to submit it for publication. H.J. performed the analyses, had access to the results of the analyses, contributed to revisions of the manuscript, and helped to make the decision to submit it for publication. J.R. and J.P. had access to the results of the analyses, contributed to revisions of the manuscript, and helped to make the decision to submit it for publication. M.C.R. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Prior Presentation. Parts of this study were presented in abstract form at the 74th Scientific Sessions of the American Diabetes Association, San Francisco, CA, 13–17 June 2014.