Severe Hypoglycemia Attributable to Intensive Glucose-Lowering Therapy Among US Adults With Diabetes: Population-based Modeling Study, 2011-2014

Abstract

Objective:

To estimate the contemporary prevalence of intensive glucose-lowering therapy among U.S. adults with diabetes and model the number of hypoglycemia-related emergency department (ED) visits and hospitalizations that are attributable to such intensive treatment.

Patients:

U.S. adults with diabetes and HbA_1c <7.0% who were included in NHANES between 2011-2014. Participants were categorized as clinically-complex if ≥75 years, with ≥2 activities of daily living limitations, end-stage renal disease, or ≥3 chronic conditions. Intensive treatment was defined as any glucose-lowering medications with HbA_1c ≤5.6% or ≥2 with HbA_1c 5.7-6.4%.

Methods:

First, we quantified the proportion of clinically-complex and intensively-treated individuals in the NHANES population. Then, we modeled the attributable hypoglycemia-related ED visits/hospitalizations over a 2-year period based on published data on event risk.

Results:

Half of U.S. adults with diabetes (10.7 million people) had HbA_1c <7.0%; 32.3% were clinically-complex and 21.6% were intensively-treated, with no difference by clinical complexity. Over a 2-year period, we estimated 31,511 hospitalizations and 30,954 ED visits for hypoglycemia; 4,774 (95% CI, 945-9,714) hospitalizations and 4,804 (95% CI, 862-9,851) ED visits were attributable to intensive treatment.

Conclusions:

Intensive glucose-lowering therapy, particularly among vulnerable clinically-complex adults, is strongly discouraged as it may lead to hypoglycemia. Yet, intensive treatment was equally prevalent among U.S. adults, irrespective of clinical complexity. Over a 2-year period, an estimated 9578 hospitalizations and ED visits for hypoglycemia can be attributed to intensive diabetes treatment, particularly among clinically-complex patients. Patients at risk for hypoglycemia may benefit from treatment deintensification to reduce hypoglycemia risk and treatment burden.

INTRODUCTION

Control of hyperglycemia as a way of reducing acute and chronic diabetes symptoms and complications is the cornerstone of diabetes management. Long-term glycemic control is associated with reduced risk of microvascular and, particularly among patients with type 1 diabetes, macrovascular complications and death.^1-6 As a result, most clinical practice guidelines recommend striving for glycosylated hemoglobin (HbA_1c) levels below 6.5%-7.0% in most non-pregnant adults with diabetes, as long as this can be achieved without hypoglycemia, polypharmacy, and undue burden of treatment and the patient is likely to meaningfully benefit from such glycemic control.^7-9 Though the ideal HbA_1c treatment target for patients with diabetes is still debated, higher HbA_1c levels are consistently recommended for patients with multiple or advanced comorbidities,^7-10 as they are at increased risk for hypoglycemia, treatment burden, and adverse drug reactions. Moreover, patients with limited life expectancy are less likely to derive long-term benefit from intensive glycemic control, which can take up to a decade to be realized.³ Despite these recommendations, several studies have demonstrated that very intensive treatment and potential overtreatment remain common, including among the elderly, those with dementia or chronic kidney disease, and the clinically-complex.^11-13

Prior population-level studies of potential overtreatment focused on older patients with complex or very complex health status who were treated with insulin or sulfonylurea drugs, but did not assess for overtreatment more broadly nor explore the association between overtreatment and hypoglycemia.^11-13 These studies, examining treatment patterns in the general U.S. population¹² and the U.S. Veterans Affairs Administration^11,13 up to 2010, found that up to 50% of relatively healthy patients, and up to 60% of patients with poor health status, may be overtreated. These studies defined potential overtreatment as attaining HbA_1c <7.0% using insulin and/or sulfonylureas, without considering polypharmacy at low very low HbA_1c levels as a potential indicator of overtreatment nor the potentially appropriate use of sulfonylurea/insulin in clinical contexts where no alternatives may be available. Another study conducted among commercially-insured and Medicare Advantage beneficiaries with non-insulin requiring type 2 diabetes who achieved HbA_1c <7.0%, estimated the prevalence of potential overtreatment between 2001-2013 to be 25.5% (26.5% of patients with low clinically complexity, 18.7% of patients with high clinical complexity).¹⁴ This study defined potential overtreatment as use of more glucose-lowering medications of any class than recommended or clinically necessary in order to achieve low HbA_1c. Using this definition, overtreatment increased the risk of severe hypoglycemia among clinically-complex patients by 77%.¹⁴ However, this study quantified potential overtreatment in a subset of U.S. adults (those with commercial and Medicare Advantage health insurance) and included information up to 2013. As such, there is no contemporary data about the full extent of intensive treatment (and potential overtreatment) and its effect on the rates of severe hypoglycemia among U.S. adults with diabetes.

Patients are at risk for hypoglycemia may benefit from timely evaluation, screening for hypoglycemia, and proactive treatment de-intensification. While the Centers for Disease Control and Prevention provide general estimates of hypoglycemia-related emergency department (ED) visits and hospitalizations among U.S. adults with diabetes,¹⁵ how many of these events are due to an immediately modifiable factor such as intensive treatment is unknown. The objective of our study was therefore two-fold. First, we quantify the rates of intensive treatment among U.S. adults using population-level data from 2011-2014 provided by the National Health and Nutrition Examination Survey (NHANES). Then, we model the estimated population-level burden of ED visits and hospitalizations for severe hypoglycemia among patients with HbA_1c <7.0%, and calculate the estimated proportion of events directly attributable to intensive treatment by applying event rates observed in comparable patient populations.¹⁴

METHODS

Study Design

In order to estimate the total number of hypoglycemia-related ED visits and hospitalizations in the U.S. and quantify the number of events attributable to intensive treatment, we applied published hypoglycemia-related ED/hospitalization event rate data¹⁴ to the general U.S. population included in NHANES years 2011-2012 and 2013-2014. The NHANES program is comprised of household interviews, physical examinations, and diagnostic/laboratory studies designed to provide a comprehensive cross-sectional representation of the health and nutritional status of adults and children in the U.S. NHANES does not include longitudinal health information or data regarding diabetes-related outcomes, including hypoglycemia; this information therefore had to be obtained from published literature. NHANES uses stratified, multistage, probability-cluster techniques to ensure that sample populations are representative of U.S. non-institutionalized civilians, thereby complementing studies conducted in narrow patient populations. This study used de-identified data and was therefore exempt from review by the Mayo Clinic Institutional Review Board.

Study Population

We included all adults (age ≥18 years) who reported a diagnosis of diabetes from a health professional and had a measured HbA_1c <7.0%. Blood samples were collected in mobile examination centers (MECs) and measurement of HbA_1c was performed using the Tosoh Automated Glycohemoglobin Analyzer HLC-723G8. Interview responses were used to ascertain patient age, sex, and race/ethnicity.

We also used interview responses to identify chronic conditions. Functional limitations were assessed based on a series of questions designed to measure functional status. Impairment in activities of daily living (ADLs) was deemed present if a patient reported some or much difficulty with an ADL (dressing, feeding, walking from room to room, and/or getting in or out of bed) or was unable to perform the ADL entirely.

Respondents were categorized as clinically-complex if they met at least one of the following criteria: (1) age ≥75 years; (2) end-stage renal disease or ≥1 dialysis session during the preceding 12 months; (3) limitation in ≥2 ADLs; (4) self-report of ≥3 chronic conditions from among coronary heart disease, congestive heart failure, stroke, chronic lung disease, kidney disease, and cancer other than non-melanoma skin cancer with first date of cancer diagnosis within 5 years. This classification of clinical complexity is based on prior studies of diabetes intensive treatment or overtreatment and is consistent with American Diabetes Association (ADA)¹⁰ and American Geriatric Society (AGS) guidance.¹⁶ It parallels the definitions used previously by McCoy et al in a study using administrative claims data (from whence estimates of severe hypoglycemia were derived, as described below)¹⁴ and by Lipska et al in a study of glycemic overtreatment using NHANES data.¹² Importantly, because NHANES does not include information about dementia (which was used to define clinical complexity in the McCoy study), and because the ADA and AGS both specify advanced functional limitations as warranting more relaxed glycemic treatment targets,^10,16 we relied on self-report of ADL limitations in lieu of a dementia diagnosis, as previously published.¹²

Defining Intensive Therapy

Based on guideline-supported thresholds for initiating and intensifying pharmacotherapy^7-9 and consistent with McCoy et al,¹⁴ intensive treatment was defined by use of any glucose-lowering medications with HbA_1c ≤5.6% or ≥2 medications with HbA_1c 5.7-6.4%. All other patients were considered non-intensively treated. Medication use is ascertained in NHANES through participant self-report. Combination medications were counted toward both medication classes, and insulin was considered as either short/rapid acting or intermediate/long acting; patients on both short/rapid and intermediate/long acting insulins were categorized as being on two (or more, if used in combination with non-insulin agents) medications.

Primary Outcome

The population-level estimates of rates of ED visits and hospitalizations for hypoglycemia were modeled by applying hypoglycemia rate estimates for clinically complex and non-clinically complex adults receiving intensive and non-intensive glucose-lowering treatment from the observational study by McCoy et al.¹⁴ to participants meeting these definitions within the NHANES sample (Figure 1). While published data in the McCoy study provided 2-year incidence (first event) rates of a composite hypoglycemia measure (ED visits, hospitalizations, and ambulatory face-to-face encounters), we used that study dataset to de novo identify all (not just first) ED visits and hospitalizations over the two year time frame (Supplemental Table 1). Ambulatory visits for hypoglycemia were not included in the primary outcome.

Figure 1. Study design.

Population-level estimates of rates of emergency department (ED) visits and hospitalizations for hypoglycemia were estimated by applying hypoglycemia rate estimates for clinically complex and non-clinically complex adults receiving intensive and non-intensive glucose-lowering treatment from the observational study by McCoy et al.¹⁴ to participants meeting these definitions within the NHANES sample. The shaded boxes provide data from McCoy et al on the percent of patients with any hospitalization(s) or any ED visit(s) for severe hypoglycemia during a two-year study period and the mean (standard deviation [SD]) number of events.

We calculated the number of ED visits and hospitalizations that may be attributed to intensive treatment as the difference between the total number of observed hypoglycemic events among intensively treated patients minus the number of events expected if non-intensive treatment event rates (Supplemental Table 1) were applied to the intensively-treated cohort.

Sensitivity Analysis

The study by McCoy and colleagues excluded patients with a recent episode of severe hypoglycemia and those receiving insulin therapy, as the objective of that study was to establish the relationship between overtreatment and incident hypoglycemia with minimal confounding by known hypoglycemia risk factors.¹⁴ Insulin-treated patients were included in this study in order to estimate the total population burden of hypoglycemia among all potentially overtreated patients. Nonetheless, recognizing that the true hypoglycemia event rate is likely to be much higher among insulin-treated patients^17-19 (and hence the general diabetes population which includes both insulin-treated and non-insulin-treated adults) than would be modeled in the primary analysis, we performed a sensitivity analysis that restricted the study population to NHANES participants not treated with insulin and report those findings in the Supplemental Material. Respondents with history of hypoglycemia were not excluded (even though they were excluded in the study by McCoy et al¹⁴) as this data is not available in NHANES. This further underestimates hypoglycemia event rate, as prior hypoglycemia is one of the strongest risk factors for future hypoglycemia.^18-24

Statistical Analysis

We calculated the weighted proportions of NHANES participants who were clinically-complex and/or intensively treated. Analyses incorporated a complex survey design using NHANES-recommended methods to produce nationally representative estimates. All data show annualized estimates of the number of U.S. adults with the outcome of interest based on the mean of values across the four study years. Univariate between-group comparisons were conducted using Rao-Scott χ² tests for categorical variables and log-linear Wald χ² tests for continuous variables, with mean values and standard deviations presented where applicable. Analyses were performed using SAS, version 9.4 (SAS Institute Inc., Cary, NC).

National estimates of hypoglycemia-related ED visits and hospitalizations among intensively-treated patients were derived via a decision analytic model developed in Microsoft Excel. To assess uncertainty when estimating attributable events, a probabilistic sensitivity analysis (PSA) was performed for hospitalizations and ED visits directly attributable to intensive treatment. The PSA was performed by simultaneously drawing from appropriate distribution functions for each model parameter according to their means and standard errors. This process of drawing parameters and running the model was repeated 1000 times and the results are presented graphically. In the PSA, NHANES estimates (population size, proportion of high complexity, and proportion treated intensively), probability of a severe hypoglycemic event, and the number of hypoglycemia-related ED visits and hospitalizations were included. Results are presented as the resampling derived 95% CI (e.g., the 2.5% and 97.5%-tiles).

RESULTS

Between 2011 and 2014, we identified 662 non-pregnant adults with diagnosed diabetes and HbA_1c <7.0%, representing an estimated 10.7 million individuals (48.8% of the total U.S. adult diabetic population). Mean patient age was 61.24 years; 20.1% were 75 years or older; Table 1. Overall, 21.55% of patients were treated intensively (corresponding to 2.3 million U.S. adults) and 78.45% were treated non-intensively (corresponding to 8.4 million U.S. adults). Nearly one third (32.3%) of the study population were clinically complex, representing 3.5 million Americans. Non-Hispanic whites comprised 64.3% of the population and 26.8% had a less than a high school education.

TABLE 1. Characteristics of U.S. adults with controlled diabetes, 2011-2014.

Percentages are calculated down each column (i.e. with the denominator set to the total, intensively-treated, and non-intensively-treated patient populations). All percentages were calculated taking into account complex survey design. The estimated numbers of U.S. adults corresponding to each percentage are shown in parentheses, except where otherwise specified. Raw numbers are omitted because they do not directly correspond to the percentages due to weighting.

	Total % (N*)	Intensive Treatment % (Na)	Non-Intensive Treatment % (Na)	P
Number of patients	10,719,057	2,309,556	8,409,501
Clinically-complex	32.34% (3,466,713)	32.39% (748,111)	32.33% (2,718,602)	.99
Age, years, mean (SD)	61.24 (0.56)	61.27 (1.11)	61.23 (0.56)	.30
Age category
18-44 years	11.61% (1,244,165)	8.55% (197,564)	12.45% (1,046,600)
45-64 years	44.23% (4,740,936)	48.07% (1,110,114)	43.18% (3,630,822)
65-74 years	24.07% (2,580,142)	28.92% (667,862)	22.74% (1,912,280)
≥75 years	20.09% (2,153,814)	14.46% (334,016)	21.64% (1,819,798)
Female	52.69% (5,647,870)	42.06% (971,461)	55.61% (3,733,092)	.05
Race/ethnicity				.57
Hispanic	12.26% (1,313,667)	9.38% (216,521)	13.05% (1,097,146)
Non-Hispanic White	64.26% (6,888,284)	67.78% (1,565,440)	63.30% (5,322,844)
Non-Hispanic Black	15.03% (1,611,179)	15.77% (364,153)	14.83% (1,247,026)
Non-Hispanic Asian	5.21% (558,978)	5.02% (115,830)	5.27% (443,148)
Other/Multiracial	3.24% (346,950)	2.06% (47,613)	3.56% (299,336)
Educatione				.22
Less than high school	26.80% (2,867,012)	24.48% (561,588)	27.44% (2,305,424)
High school/GED	23.38% (2,501,062)	15.91% (365,001)	25.42% (2,136,062)
Some college	31.72% (3,392,522)	35.71% (819,192)	30.63% (2,573,330)
College degree	18.10% (1,935,686)	23.90% (548,235)	16.51% (1,387,451)
HbA1c range				--
≤5.6%	14.17% (1,519,233)	32.71% (755,493)	9.08% (763,740)
5.7% - 6.4%	52.55% (5,632,568)	67.29% (1,554,063)	48.50% (4,078,504)
6.5% - 6.9%	33.28% (3,567,257)	--	42.42% (1,983,921)
Insulin use b	12.60% (1,350,792)	20.40% (471,195)	10.46% (879,597)	.01
Non-insulin medication use c				<.001
0 drugs	24.95% (2,674,428)	8.18% (188,947)	37.30% (3,137,084)
1 drug	44.67% (4,788,016)	33.56% (775,178)	49.27% (4,143,611)
2 drugs	20.61% (2,209,155)	44.39% (1,025,238)	10.71% (900,948)
3 drugs	8.25% (884,512)	10.98% (253,673)	2.71% (227,858)
4 drugs	1.52% (162,946)	2.88% (66,521)	0.00% (0)
Primary insurance				.04
None	9.67% (1,036,736)	7.46% (172,270)	10.28% (864,465)
Medicare	37.57% (4,027,552)	35.16% (811,937)	38.24% (3,215,615)
Medicaid	7.02% (752,676)	6.12% (141,230)	7.27% (611,445)
Dual-eligible	5.90% (632,466)	8.02% (183,319)	5.32% (447,148)
Private	30.64% (3,284,501)	25.82% (596,374)	31.97% (2,688,127)
Other public d	9.19% (985,126)	17.42% (402,426)	6.93% (582,701)
Comorbiditiese
≥ 2 ADL limitations	12.74% (1,365,383)	14.93% (344,896)	12.13% (1,020,487)	.62
ESRD or dialysis	2.12% (227,766)	3.01% (69,597)	1.88% (158,169)	.62
CKD (non-end-stage)	7.26% (761,159)	4.15% (92,896)	8.11% (668,263)	.06
Heart failure	10.46% (1,116,906)	11.17% (257,382)	10.26% (859,524)	.82
Coronary heart disease	15.64% (1,669,618)	15.99% (367,296)	15.55% (1,302,321)	.92
Stroke	9.36% (1,002,388)	12.39% (286,099)	8.53% (716,290)	.21
Lung disease	18.73% (1,995,211)	25.96% (594,929)	16.74% (1,400,282)	.06
Cancer (past 5 years)	5.29% (528,140)	3.65% (75,607)	5.72% (452,533)	.42
Diabetes duration, years				.81
< 5 years	39.42% (4,225,486)	37.63% (869,088)	39.91% (3,356,397)
5-10 years	22.69% (2,432,379)	20.25% (467,666)	23.36% (1,964,713)
≥10 years	37.89% (4,061,192)	42.12% (972,801)	36.73% (3,088,391)

Abbreviations: ADL, activities of daily living (dressing/bathing, eating, walking, toileting, and hygiene). CKD, chronic kidney disease. ESRD, end-stage renal disease.

^aNHANES survey design is not a simple random sample, but uses complex, multistage techniques to select participants representative of the non-institutionalized U.S. civilian population. Oversampling of particular subgroups is done to account for differences in response rates and improve statistical precision of under-represented populations. Appropriate survey weights are then applied to obtain nationally representative population estimates from this sample.

^bInsulin use, alone or in combination with non-insulin drugs. Each type of insulin (e.g. rapid acting, long-acting) was considered separately.

^cNumber of non-insulin glucose lowering medications used by the patient, with or without concurrent insulin use.

^dOther public health insurance types included SCHIP, military health care, Indian Health Service, state-sponsored health plan, or other government insurance as reported in NHANES. Patients with Medicare and any other insurance type except for Medicaid were considered as being covered by Medicare only.

^eDue to missing information as NHANES respondents refused to answer or did not know, denominators may be smaller than those reported as the number of patients in each column. Percentages reported are based on the denominator associated with the reported count.

Prevalence of intensive glucose-lowering therapy

The prevalence of intensive treatment was 21.6% among clinically complex patients and 21.5% among non-clinically complex patients. Table 1 compares intensively- and non-intensively-treated patients. Intensively-treated patients were more likely to use insulin (20.4% vs. 10.5%) and take two or more glucose-lowering medications. There was no statistically significant difference in comorbidity burden between intensively-treated and non-intensively-treated patients; however, the point estimates for nearly all conditions were higher among intensively-treated patients except for cancer and chronic kidney disease. There was also no difference in patient age, race/ethnicity, diabetes duration, or educational attainment between intensively-treated and non-intensively-treated patients. Intensively-treated patients were significantly less likely to be uninsured or have private health insurance, but much more likely to have public health insurance.

Estimate of Severe Hypoglycemia-Related Events

Over a 2-year period, we predicted 31,511 hospitalizations and 30,954 ED visits to occur for severe hypoglycemia among U.S. adults with diabetes and HbA_1c <7.0%. The vast majority of these events would be experienced by clinically-complex patients (Table 2). Of these, an estimated 4,774 hospitalizations (95% CI, 945-9,714) and 4,804 ED visits (95% CI, 862-9,851) could be directly attributed to intensive glucose-lowering therapy.

Table 2. Severe hypoglycemia-related events within two years.

The number of severe hypoglycemia hospitalizations and emergency department visits was estimated by applying rates from McCoy et al 2016¹⁴ to non-pregnant NHANES participants with diabetes who met analogous inclusion criteria (≥18 years and HbA_1c <7.0%). Events attributable to intensive treatment were calculated by applying event rates expected with non-intensive treatment to those patients receiving intensive therapy. 95% confidence intervals (CI) were estimated using probabilistic sensitivity analysis (PSA).

	Non-intensive treatment	Intensive treatment	Total	Events Attributable to Intensive Treatment
Hospitalizations
Low complexity	8,127	3,654	11,781	787 (95% CI, 0 - 1,812)
High complexity	10,406	9,324	19,730	3987 (95% CI, 374 - 8,925)
U.S. total	18,533	12,978	31,511	4,774 (95% CI, 945 - 9,714)
Emergency department visits
Low complexity	2,902	2,436	5,338	525 (95% CI, 0 - 1,241)
High complexity	15,610	10,006	25,616	4,279 (95% CI, 431 - 9,286)
U.S. total	18,512	12,442	30,954	4,804 (95% CI, 862 - 9,851)

Sensitivity Analysis

There were 573 participants within NHANES who had HbA_1c <7.0% and were not treated with insulin, corresponding to 87.4% of the base population or 9.4 million people (Supplemental Table 2). They were generally similar to the broader study population. When compared to the general population of U.S. adults with HbA_1c <7.0%, the subset not treated with insulin was slightly younger (60.81 vs. 61.24 years), mostly due to having a higher proportion of patients aged 45-64 years. Patients who were intensively treated without use of insulin were also more likely to HbA_1c ≤5.6% (by definition, they were treated with another glucose-lowering medication at that low HbA_1c level). The prevalence of nearly all chronic conditions was lower in the non-insulin treated cohort, with the exception of cancer, which was equally prevalent in both.

In the non-insulin-treated cohort, we estimated a total of 50,337 severe hypoglycemic events with 25,712 hospitalizations (51.1%) and 24,625 ED visits (48.9%) over a two-year period. This vast majority of these events would also occur among the clinically-complex (Supplemental Table 3). Of these, an estimated 3,428 (95% CI, 878-6701) hospitalizations and 3,409 (95% CI, 768-7022) ED visits could be directly attributed to intensive treatment.

DISCUSSION

Over ten million, or nearly half, of U.S. adults with diabetes had HbA_1c <7.0% between 2011 and 2014, and 2.3 million of these patients were treated much more intensively than recommended by current evidence-based guidelines.^7-9 Such intensive treatment is not harmless. It promotes polypharmacy, with 13.86% of intensively-treated patients taking three or more non-insulin medications, sometimes in addition to insulin, thereby increasing treatment burden, risk of adverse drug reactions, and costs of diabetes care. In addition, we estimated that such intensive glucose-lowering therapy contributed to at least 4,774 hospitalizations and 4,804 ED visits for severe hypoglycemia over two years, primarily among clinically-complex patients. Such hypoglycemia can, in turn, lead to death,^25-29cardiovascular disease,^28-30 cognitive decline,³¹ disability,³² impaired quality of life,³³ and high costs of care.³²

There is little evidence that treating to very low HbA_1c improves patient outcomes, especially in the context of multi-morbidity and advanced age.³⁴ Despite explicit guidance to avoid intensive treatment in clinically-complex and elderly patients,^7-10 this did not translate into clinical practice as the prevalence of intensive treatment was nearly identical irrespective of patient clinical complexity. More relaxed glycemic targets are also recommended for patients with long-standing diabetes, and we found no difference in treatment intensity by diabetes duration. These findings are consistent with historic data demonstrating high prevalence of potential overtreatment in a variety of settings in the U.S.,^11-14 and lack of improvement in recent years despite growing recognition of the harms of overtreatment is concerning. The most recent population-level assessment of intensive glycemic control in the U.S. prior to our study focused on the use of insulin and sulfonylurea drugs to achieve HbA_1c <7.0% among patients ≥65 years between 2001 and 2010.¹² By this definition, 55% of older adults with diabetes were potentially overtreated, with greater prevalence of overtreatment among patients with worsening health status. Similarly, there was no variation in potential overtreatment by clinical complexity or life expectancy in studies conducted among Veterans Affairs patients, where 50% of patients ≥75 years or with dementia or CKD were treated with insulin or sulfonylurea to achieve HbA_1c <7.0%,¹¹ or in the privately-insured population, where the prevalence of potential overtreatment (defined by the use of more glucose-lowering medications than recommended at low HbA_1c) among the clinically-complex was 18.7%.¹⁴

The prevalence of intensive treatment in our study was lower than the aforementioned population-based studies^11-13 due to inclusion of adults of all ages and use of a more stringent definition of intensive treatment, which was predicated on both the HbA_1c level and the number of medications used to achieve it. We chose the latter definition because use of insulin and sulfonylurea drugs to achieve HbA_1c <7.0% does not necessarily translate to overtreatment. These medications can be used safely and may be the only treatment options available for the management of diabetes in specific clinical contexts. In contrast, use of glucose-lowering medications to achieve HbA_1c ≤5.6% or the use of ≥2 medications to achieve HbA_1c 5.7-6.4% is not consistent with clinical practice guidelines and is more likely to denote excessive glucose-lowering therapy. When compared to the earlier study of potential overtreatment among commercially-insured and Medicare Advantage beneficiaries between 2001-2013, the current prevalence of intensive treatment among U.S. adults was higher among high-complexity patients but lower among low-complexity patients. Specifically, our study found that 21.6% of high-complexity adults were treated intensively compared to 18.7% of privately-insured complex patients.¹⁴ Conversely, intensive treatment rates were 21.5% among low-complexity U.S adults in this study vs. 26.5% among corresponding low-complexity privately-insured beneficiaries.¹⁴ These differences likely reflect our study’s inclusion of patients with public or no health insurance as well as temporal changes in diabetes management. For example, patients without health insurance were much less likely to be intensively treated and they are also less likely to be clinically complex (e.g. older patients, patients with ESRD, and patients with multiple comorbidities are more likely to be insured); these patients were included in this study but excluded from the earlier work.

There is no population-level data on the rates of ED visits and hospitalizations for hypoglycemia among adults with HbA_1c <7.0%, which based on HbA_1c parameters alone is considered to reflect optimally controlled diabetes. Although our study could not directly quantify all the ED visits and hospitalizations for hypoglycemia across the U.S., these estimates are the best available data about the number of events attributable to intensive treatment and thus potentially avoidable with guideline-concordant care. Our sensitivity analysis, which restricted the study cohort to patients not treated with insulin to fully mimic the source data for hypoglycemic event rate estimates, suggested that a substantial number of ED visits and hospitalizations for severe hypoglycemia occurred among insulin-treated clinically-complex patients. This reinforces the importance of considering treatment de-intensification, including simplification of insulin regimens, among clinically-complex patients in order to reduce their risk of hypoglycemia.

While there is no standard and universally accepted definition of clinical complexity, the approach used in our study was grounded in the framework proposed by the American Geriatrics Society.¹⁶ It is more broad than earlier studies of potential overtreatment, which primarily included patients with chronic kidney disease or dementia,^11,13 but is similar to the approach taken by Lipska and colleagues.¹² Nonetheless, because the same definitions of clinical complexity and intensive treatment were applied to the NHANES population as were used in the source data, the specifics of these definitions do not affect the study results. Specifically, because all events among all patients were counted in the source data, and all comparable patients were identified among NHANES respondents, changing the definition of clinical complexity and/or treatment intensity would not alter the total number of observed events. It would merely shift the distribution of events among the four subsets of patients, keeping the estimated total number of events in the overall population unchanged. However, the reliance on self-report of comorbidities is a limitation, as different conditions and ADL limitations are prone to under- and over-reporting.

A major limitation of our study is its reliance on hypoglycemia-related hospitalization and ED visit event rates ascertained in a different study population.¹⁴ Source data was obtained from the OptumLabs^® Data Warehouse (OLDW), which includes de-identified claims data for privately insured and Medicare Advantage enrollees in a large, private, U.S. health plan.^35,36 People included in the OLDW represent a diverse mix of ages, ethnicities, and geographic regions across the U.S., but nonetheless is not fully representative of the U.S. population. Furthermore, because rates of hypoglycemia-related ED visits and hospitalizations from a different study cohort were applied to population-level estimates for the distributions of treatment intensities and clinical complexity among U.S. adults, results presented herein are an estimate of the number of hospitalizations and ED visits for hypoglycemia, rather than the true count of events. Confidence intervals around these estimates are wide, reflecting the relatively low rate of hypoglycemia-related ED visits and hospitalizations in this otherwise low risk population. This is most noticeable in Table 2, where the 95% CIs includes a lower bound of zero hospitalizations and ED visits due to the low number of those classified as having severe hypoglycemia among those non-clinically complex receiving intensive treatment. Still, this is the only available population-level estimate of the number of hypoglycemia-related events directly attributable to intensive treatment and therefore potentially avoidable. Importantly, our findings underestimate the true scope of hypoglycemia in the U.S., as rates of hypoglycemia were extrapolated from lower risk patients: not using insulin, without prior hypoglycemia, and commercially insured.¹⁴ True numbers are likely to be much higher. These estimates also reflect the most severe episodes which required ED care or hospitalization, and not the much higher true burden of hypoglycemia in the management of diabetes. The vast majority of hypoglycemic events do not come to medical attention but are treated by the patient or caregivers, or are treated by medical personnel outside of the ED or hospital setting.^37-39 Furthermore, most hypoglycemic events occur in patients with elevated HbA_1c and are not captured in our study. However, the objective of this study was to estimate the number of events directly attributable to intensive treatment and thus potentially avoidable with preemptive identification, treatment de-escalation or simplification, and closer monitoring.^40,41

CONCLUSION

High quality diabetes care is predicated on balancing the benefits and harms of glucose-lowering therapy, including avoiding very intensive treatment particularly among the elderly, the clinically-complex, and those with long duration of diabetes.^7,8,10 Hypoglycemia is the most common serious adverse effect of glucose-lowering therapy, but its rates can be reduced with patient-centered evidence-based care. In this study, we found that intensive treatment with glucose-lowering therapy remains common in the U.S. There is also no apparent individualization of therapy based on patient clinical complexity, life expectancy, or likelihood of benefit from targeting low HbA_1c levels. Such uniformity of intensive glycemic control and treatment are estimated to have contributed to over 9,500 potentially preventable ED visits and hospitalizations for severe hypoglycemia in the span of two years.

ABBREVIATIONS

ADA

American Diabetes Association

ADL

Activities of daily living

AGS

American Geriatric Society

CKD

Chronic kidney disease

ED

Emergency department

HbA_1c

Glycosylated hemoglobin

OLDW

OptumLabs^® Data Warehouse

NHANES

National Health and Nutrition Examination Survey

PSA

Probabilistic sensitivity analysis