Racial Differences in Trends of Serious Hypoglycemia Among Higher Risk Older Adults in US Veterans Health Administration, 2004–2015: Relationship to Comorbid Conditions, Insulin Use, and Hemoglobin A1c Level

Abstract

Aims:

To evaluate temporal trends in racial/ethnic groups in rates of serious hypoglycemia among higher risk patients dually enrolled in Veterans Health Administration and Medicare fee-for-service and assess the relationship(s) between hypoglycemia rates, insulin/secretagogues and comorbid conditions.

Methods:

Retrospective observational serial cross-sectional design. Patients were ≥ 65 years receiving insulin and/or secretagogues. The primary outcome was the annual (period prevalence) rates (2004–2015), per 1,000 patient years, of serious hypoglycemic events, defined as hypoglycemic-related emergency department visits or hospitalizations.

Results:

Subjects were 77–83% White, 7–10% Black, 4–5% Hispanic, <2% women; 38–58% were ≥75 years old; 72–75% had ≥1 comorbidity. In 2004–2015, rates declined from 63.2 to 33.6(−46.9%) in Blacks; 29.7 to 20.3 (−31.6%) in Whites; and 41.8 to 29.6 (−29.3%) in Hispanics. The Black-White rate differences narrowed regardless of insulin use, hemoglobin A1c level, and frequency and various combinations of comorbid conditions. Among insulin users, The Black-White contrast decreased from 34.7 (98.5 vs. 63.8) in 2004 to 13.2 (43.6 vs. 30.4) in 2015; in non-insulin users, the contrast was 25.7 (44.1 vs. 18.4) in 2004 and 10.1 (18.9 vs. 8.8) in 2015.

Conclusion:

Marked declines in serious hypoglycemia events occurred across race, medications, and comorbidities, suggesting significant changes in clinical practice.

1. Introduction

The US Department of Health and Human Services (DHHS) National Action Plan for Prevention of Adverse Drug Events identified serious hypoglycemia (defined as emergency department visits or hospitalizations) as one of three major drug related adverse events, and developed an inter-agency plan to reduce event rates.¹ Key recommendations included evaluation of rates by specific medications (e.g., insulin and secretagogues), age, comorbid conditions and race/ethnicity. Blacks are known to have higher rates of serious hypoglycemic events, but results conflict about whether racial differences in rates are decreasing in populations enrolled in Medicare² and Healthcare Maintenance Organizations.³

Following the recommendations by the DHHS, this study focused on patients at higher risk for hypoglycemia by limiting the study cohort to patients receiving insulin and or secretagogues. The study objectives were to: (1) evaluate temporal trends in racial/ethnic groups in rates of serious hypoglycemia; and (2) assess the relationship(s) between hypoglycemia rates and comorbid conditions, insulin/secretagogues use, and hemoglobin A1c (HbA1c) levels.

2. Subjects

We analyzed data from patients dually enrolled in Veterans Health Administration (VHA) and Medicare fee-for-service from 2004–2015. Patient-level electronic medical records were utilized in the study and it was approved by the Institutional Review Board of the New Jersey Health Care System under the US Department of Veterans Affairs for exemption of informed consent and HIPAA Authorization.

3. Materials and methods

3.1. Study population, data sources, and study design

This retrospective observational serial cross-sectional design utilized VHA and Medicare patient-level data. For each calendar year between 2004 and 2015, we identified VHA veteran patients dually enrolled in fee-for service Medicare (Part A and Part B), aged ≥ 65 years with diabetes, receiving insulin or secretagogues (sulfonylurea and meglitinides) within 120 days prior to the beginning of each year, and who were alive as of the beginning of the year. Each year in 2004–2015 was a study year, or outcome year, for it was used for outcome evaluation. Baseline year was defined as the one-year period prior to each outcome year.

For each cross-sectional cohort, we removed patients having end stage renal disease (ESRD) or in VHA or Medicare hospice programs, long-term or extended care or rehabilitative facilities in its baseline or outcome years. We determined the diagnosis of diabetes using a previously validated approach based on either diabetes medication prescription in the baseline year or two or more visits from inpatient stays or outpatient clinician visits with diabetes-specific International Classification of Diseases – Clinical Modification, 9^th edition (ICD-9-CM) codes (250.xx, 357.2, 362.0, 366.41, 648.0, E932.3) within two years prior to the outcome year.⁴ ESRD was determined using ICD-9-CM diagnosis and procedure codes for dialysis, VHA clinic stop codes, and Medicare enrollment eligibility status codes.⁵ Patients with unknown race/ethnicity (7 to 161 per year) were removed.

3.2. Outcome measures

The primary outcome was the annual (period prevalence) rates (2004–2015) of serious hypoglycemic events, defined as emergency department visits or hospitalizations for hypoglycemia. We calculated the number of persons having ≥1 event in each year over the total person years of the eligible study population in the same year. Hypoglycemia was determined using a previously validated approach⁶ using ICD-9-CM codes. ICD-9-CM codes were replaced with ICD-10-CM codes since October 2015 in the US, so ICD-10-CM codes were used to determine hypoglycemia for the last three months in 2015. We used the General Equivalence Mappings and guidance provided by Medicare for mapping ICD-9 to ICD-10.⁷ A hospitalization for hypoglycemia was determined by presence of a hypoglycemia-specific principal diagnosis code. To determine emergency department (including urgent care) visits, in VHA files we used primary and secondary clinic stop codes 102, 101 before March 1, 2007 and 130, 131 afterwards⁸; in Medicare data we used place of service and Healthcare Common Procedure Coding System codes, revenue codes, and emergency room charge amounts (value greater than zero).⁹

3.3. Independent variables

Independent variables included age, sex, race/ethnicity, comorbidities, insulin/secretagogue use, and HbA1c values. Age (at baseline) was categorized into 65–74, 75–84, and ≥85 years. Race/ethnicity groups were non-Hispanic Whites (Whites), non-Hispanic Blacks (Blacks), Hispanics, and others/mixed (including individuals of Asians/Pacific Islander, Native American, and mixed race). We reported comparisons only for the first three groups because they are less heterogeneous than the others/mixed group. Race/ethnicity was determined based on VA files with missing data filled in with information from Medicare.

We first selected comorbidities that could increase the risk of serious hypoglycemic events based on systematic review¹⁰ and post-hoc analyses of clinical trials^11,12, and then created composite comorbidity categories based upon our published taxonomy.¹³ These included four comorbidity categories/indicators (yes/no): (1) diminished life expectancy (cancer, end stage hepatic disease, chronic obstructive pulmonary disease (COPD) with exacerbation), (2) advanced diabetes complications (lower extremity amputations, advanced retinopathy, gastroparesis, autonomic neuropathy, and diabetic poly neuropathy, chronic kidney disease), (3) cardiovascular conditions (myocardial infarction, chronic heart failure, ischemic vascular disease, transient ischemic attack), and (4) dementia/cognitive impairment, major neurological disorders (Parkinson disease, hemiplegia, epilepsy, aphasia, dysphasia, apraxia, multiple sclerosis), and major depression, substance/alcohol abuse.

We created a variable representing the number of comorbidity categories (range:0–4). We also derived a variable containing sixteen mutually exclusive comorbidity groups based on various combinations of the above four comorbidity categories. Comorbid conditions for each outcome year were derived from the baseline period.

Prescription of insulin and secretagogues within the last 120 days of the baseline year was a condition of cohort entry. Categories included insulin without secretagogues, insulin with secretagogues, and secretagogues without insulin. To assess the date of the most proximal (recent) medication information in relation to the first hypoglycemic event (the index event) in an outcome year, we evaluated insulin/secretagogue use within 120 days prior to the date of the index event, or within the 120 days prior to the date of death or the end of an outcome year for those without an event.

Because HbA1c values were available only in VHA files, individuals without a value were assigned to the category of “no value.” In each cross-sectional cohort, a patient’s baseline HbA1c information was taken from the last HbA1c value in the baseline year. To assess proximal HbA1c values, we obtained the values of the last HbA1c in the period of 120 and 365 days prior to the date of the index event. Findings from using 120 and 365 days were similar. We chose the latter approach to include more subjects in the analysis. HbA1c values were reported as mean values and by interval: <7%, 7–7.9%, 8–8.9%, and >=9%.

3.4. Statistical techniques

Annual rates of serious hypoglycemic events from 2004 to 2015 were calculated as the number of patients having ≥1 event per 1,000 patient years. Age-and-sex standardized rates using the 2014 cohort as the standard population were reported (Supplementary (Suppl.) Table 1). To describe the overall rate changes in the period of 2004–2015, we calculated differences in years 2004 and 2015 using the absolute rate differences and the percentage change relative to 2004. Graphs of age-and-sex adjusted rates using three-year moving averages were presented to show the rate trends over time; for each year, the moving average was calculated as the mean of the rates from the current and one year before and one year after, except for 2004 and 2015, each with only two annual rates available.

We evaluated temporal trends for the overall study population and in subpopulations by race/ethnicity, comorbidity, insulin use, and HbA1c levels. Furthermore, we evaluated trends in racial/ethnic differences within subgroups of insulin/secretagogue use, comorbidity, and HbA1c level. Log-linear Poisson regression analysis was used to evaluate statistical significance for temporal trends of the annual rates. Person-years were treated as an off-set variable to account for unequal observation time. “Year” was treated as a continuous measure and modeled by either a linear or polynomial function. Observations are inherently correlated because many subjects appeared in more than one serial cross-sectional study cohorts; we therefore used the generalized estimating equation (GEE) approach to account for intra-subject correlation to provide valid statistical inference.¹⁴

As an additional analysis, we used the subset of patients who received insulin/secretagogues and had HbA1c test results available based on proximal information from the outcome years to analyze the inter-relationships among insulin use, HbA1c values, and race/ethnicity.

4. Results

4.1. Demographics and clinical characteristics of the study population

Table 1 shows the characteristics of the study cohorts, overall and by race/ethnicity, for the selected years of 2004, 2009, and 2015; summary statistics for additional independent variables are in the Suppl. Table 2.

Table 1.

Population characteristics, overall and by race/ethnicity in selected study years. Notes: Data presented are column percentages or mean ± standard deviation. Because of rounding, the sum of percentages may not be equal to 100. SD: standard deviation. Baseline medications were derived from the last 120 days prior to the beginning of each evaluation year. Baseline A1c level information used the last A1c value from the one-year interval prior to each year of outcome evaluation.

Year	2004				2009				2015
	Total	Hispanics (H)	Non-H Whites	Non-H Blacks	Total	Hispanics (H)	Non-H Whites	Non-H Blacks	Total	Hispanics (H)	Non-H Whites	Non-H Blacks
Number of persons	214,617 %	10,871 %	177,257 %	15,875 %	200,183 %	8912 %	163,285 %	14,519 %	245,827 %	12,234 %	190,433 %	23,684 %
Age (years)
65–74	49.4	55.7	47.9	55.5	41.9	46.5	39.8	52.7	62.5	68.8	60.0	72.0
75–84	45.7	40.3	47.1	39.9	46.0	43.4	47.3	38.8	27.3	22.9	28.8	21.8
85 and over	4.9	4.1	5.1	4.6	12.2	10.1	12.9	8.5	10.2	8.3	11.2	6.2
Mean ± SD	75 ± 6	74 ± 6	75 ± 6	74 ± 6	76 ± 7	76 ± 7	77 ± 7	75 ± 6	74 ± 7	73 ± 7	74 ± 7	72 ± 6
Sex
Men	98.9	99.4	98.9	99.0	98.8	99.1	98.8	98.9	98.75	98.7	98.86	98.18
Women	1.1	0.6	1.1	1.0	1.2	0.9	1.2	1.2	1.25	1.3	1.14	1.82
Baseline diabetes medications
Insulin, no	22.0	25.7	20.9	31.0	26.3	29.0	25.4	32.5	40.1	42.9	39.1	42.9
secretagogues (Sc)
Insulin, with Sc	6.5	7.1	6.4	7.5	10.2	10.7	10.1	10.3	12.8	12.1	12.9	11.6
No Insulin, with Sc	71.5	67.2	72.7	61.5	63.5	60.3	64.6	57.2	47.1	45.0	48.1	45.5
Comorbidities (number of four comorbidity indicators)
0	27.0	34.3	25.4	36.6	25.0	29.8	23.7	34.2	27.8	30.1	27.3	35.5
1	43.5	41.0	44.2	38.3	42.4	39.3	43.0	38.2	39.6	38.6	40.2	38.0
2	23.7	20.0	24.4	20.1	25.4	23.9	25.9	21.5	24.5	23.6	24.6	20.4
3	5.3	4.4	5.5	4.6	6.6	6.6	6.7	5.5	7.2	6.8	7.1	5.5
4	0.4	0.4	0.5	0.3	0.7	0.4	0.7	0.6	0.9	0.9	0.8	0.6
Baseline HbA1c value
No values	21.1	12.4	22.4	15.6	21.8	17.6	23.2	15.5	19.8	10.7	21.8	15.3
With values	78.9	87.6	77.6	84.4	78.2	82.4	76.8	84.5	80.2	89.3	78.2	84.7
Distribution for those with values
b7%	48.8	38.1	50.3	43.1	48.0	41.7	49.0	43.5	33.2	29.2	33.6	32.6
7.0–7.9%	30.2	30.7	30.3	28.1	32.2	31.6	32.3	31.1	34.6	32.4	35.3	31.8
8.0–8.9%	13.1	16.8	12.6	15.0	12.4	15.1	11.9	14.2	18.8	20.6	18.7	19.0
>9%	8.0	14.4	6.8	13.8	7.5	11.6	6.7	11.2	13.4	17.8	12.4	16.7
Mean ± SD (%)	7.2 ± 1.2	7.5 ± 1.4	7.1 ± 1.2	7.5 ± 1.5	7.2 ± 1.2	7.4 ± 1.4	7.2 ± 1.1	7.4 ± 1.4	7.6 ± 1.3	7.8 ± 1.5	7.6 ± 1.3	7.7 ± 1.5

The number of patients ranged from 214,617 to 245,827 per year; 77.2–83.2% were Whites, 6.8–9.6% Blacks, and 4.2–5.0% Hispanics; 1.1–1.3% were women. Our primary analyses were limited to individuals who self-identified as Black, Hispanic or White. The percentage of patients ≥75 years old ranged 37.5%−58.1%, increasing in later years.

Overall, 25.0–28.0% patients had none of the studied comorbidity categories. Most patients had only one comorbidity category (39.6–43.5%); about a quarter had two comorbidity categories. Patients who had 3 or 4 comorbidity categories increased from 5.7% in 2004 to 8.1% in 2015. The comorbidity category of cardiovascular conditions was most prevalent (52.8%−60.5%), followed by advanced diabetes complications (25.9–30.4%) (Suppl. Table 2). The patients with advanced diabetes complications increased from 25.9% in 2004 to 30.4% in 2015 (Supplementary Table 2).

Percentage of patients who received secretagogues (mostly sulfonylurea) without insulin declined from 71.5% in 2004 to 47.1% in 2015 (Table 1). However, insulin use gradually increased: from 22.0% to 40.1% for insulin without secretagogues, and from 6.5% to 12.8% for insulin with secretagogues. Blacks consistently had higher rates in insulin use than Whites and Hispanics, but the difference markedly declined over the study period.

Mean HbA1c values increased slightly from 2004 to 2015 in all racial/ethnic groups, with the percentages of patients with HbA1c <7% markedly decreased: from 43.1 to 32.6% in Blacks, 50.3 to 33.6% in Whites, and 38.1 to 29.2% in Hispanics (Table 1). Mean HbA1c values in Blacks and Hispanics were similar, but higher than those in Whites. However, the difference between non-Whites and Whites narrowed over time; e.g., for Blacks vs. Whites, the difference went from 0.4% (7.5% for Blacks vs. 7.1% for Whites) in 2004 to 0.1% (7.7% vs. 7.6%) in 2015.

4.2. Rates of serious hypoglycemia events and racial/ethnic differences

The study population was comprised of 689,541 unique patients identified in 2004 to 2015; they incurred a total of 78,057 serious hypoglycemic events. On a yearly basis, 4,918 to 6,457 patients who met study criteria in the beginning of each calendar year incurred 5,538 to 7,467 serious hypoglycemic events (data not shown).

Rates of serious hypoglycemia were higher in individuals ≥75 years than 64–75 years of age. The overall age-and-sex adjusted rates declined during the 12 years (Figure 1A), from 34.1 (per 1,000 person-years) in 2004 to 23.9 in 2015, a relative rate reduction of 29.9% and an absolute rate decrease of 10.2 (Table 2). The overall downward trend was statistically significant (p <0.001). All three racial/ethnic groups experienced downward trends (p<0.001). Blacks consistently had higher hypoglycemia rates than Hispanics and Whites. However, Blacks also showed larger rate declines (relative rate reduction: 46.9%; absolute rate decrease: 29.6/1000 person-years) than Hispanics (29.3%; 12.2/1000 person-years) and Whites (31.6%; 9.4/1000 person-years) (Table 2). The differential downward trends among racial/ethnic groups were statistically significant (p<0.001).

Figure 1. Rates* (per 1,000 person years) of serious hypoglycemic events.

*Age -and-sex standardized rates to the 2014 study cohort, three years moving averages (for 2004, 2015 – two years moving averages). White: Non-Hispanic Whites; Black: Non-Hispanic Blacks. Comorbidities in category (indicators): A: Advanced diabetes complications; C: Cardiovascular conditions; L: Diminished life expectancy; D: Dementia/Cognitive impairment, major neurological disorders, major depression, substance/alcohol use. The number of comorbidities was derived from the sum of the above four comorbidity category indicators, ranging from zero to four. Mutually exclusive groups are based on various combinations of the above four comorbidity category indicators.

Table 2.

Annual age- and sex-standardized rates (per 1000 patient years) of serious hypoglycemic events, overall and by race/ethnicity, within subpopulations. Notes: Proximal medication information was from within the 120 days prior to the index hypoglycemic event, or death or end of the outcome year for those without such an event. Proximal HbA1c level used the last A1c value within the 365 days prior to the index hypoglycemic event, or death or end of the outcome year for those without such an event. Insulin users included patients with/without secretagogues.

Year	2004	2005	2006	2007	2008	2009	2010	2011	2012	2013	2014	2015	Relative rate change (2015 vs. 2004)	Absolute rate change (2015 vs. 2004)
Overall	34.1	34.2	34.5	33.8	32.9	31.9	30.6	29.2	27.6	26.0	24.6	23.9	‐29.9%	‐10.2
Hispanics	41.8	44.0	42.8	34.9	43.9	39.2	31.1	35.1	29.2	34.4	33.2	29.6	−29.3%	−12.2
Whites	29.7	32.0	31.6	31.4	30.1	28.5	28.4	26.2	25.0	22.5	21.3	20.3	−31.6%	−9.4
Blacks	63.2	63.7	57.6	59.2	58.4	55.0	49.8	47.9	40.4	40.0	34.3	33.6	−46.9%	−29.6
Proximal medications
Insulin users	68.8	68.2	66.4	64.8	60.5	58.6	51.2	48.9	44.1	39.4	36.6	34.9	‐49.3%	‐34.0
Hispanics	70.5	71.8	71.7	43.6	58.3	69.1	44.1	49.0	50.3	43.5	48.9	42.6	−39.5%	−27.9
Whites	63.8	63.1	62.3	61.6	55.0	53.6	47.2	43.7	40.9	34.1	33.0	30.4	−52.4%	−33.4
Blacks	98.5	105.1	88.9	101.6	103.9	90.2	83.6	80.3	52.6	60.9	42.2	43.6	−55.8%	−54.9
Insulin non‐users	20.5	21.7	21.9	19.1	16.8	15.9	15.7	14.8	12.4	12.6	11.1	10.8	‐47.4%	‐9.7
Hispanics	27.8	34.2	31.5	19.1	20.5	21.4	19.8	21.0	14.5	21.1	18.6	10.0	−64.1%	−17.9
Whites	18.4	19.4	20.0	17.2	15.3	13.9	13.7	12.9	10.6	9.6	8.7	8.8	−52.2%	−9.6
Blacks	44.1	39.2	36.3	39.0	31.3	28.2	25.1	28.8	22.0	25.6	17.0	18.9	−57.1%	−25.2
Proximal A1c value
<7%	34.5	37.0	38.6	34.7	30.5	32.5	30.8	31.9	29.1	26.7	25.0	24.5	‐29.0%	‐10.0
Hispanics	46.9	43.3	52.8	27.4	32.1	48.9	24.1	41.2	38.7	33.5	41.6	29.1	−37.9%	−17.8
Whites	31.6	34.2	35.1	31.9	27.7	28.8	28.3	28.1	25.7	22.0	21.9	20.7	−34.5%	−10.9
Blacks	65.0	64.9	65.7	69.9	56.6	53.8	51.5	60.3	37.3	47.8	26.7	36.4	−43.9%	−28.5
7.0‐7.9%	30.8	32.3	31.2	30.2	30.4	30.4	28.8	27.3	25.4	23.3	22.1	22.2	‐27.9%	‐8.6
Hispanics	28.9	51.4	39.5	33.0	34.0	32.5	28.6	31.9	30.9	33.7	31.1	22.7	−21.3%	−6.1
Whites	27.9	29.1	28.5	27.5	27.2	27.0	26.0	24.0	23.7	19.0	19.3	19.4	−30.3%	−8.4
Blacks	57.3	56.6	54.3	52.9	62.0	56.5	49.6	47.6	35.7	39.8	29.9	31.0	−45.9%	−26.3
8.0‐8.9%	39.4	37.8	34.8	39.1	38.7	35.1	31.5	31.1	30.4	28.7	26.0	24.8	‐37.2%	‐14.7
Hispanics	48.3	40.1	38.7	19.3	46.9	31.3	33.0	30.7	27.8	34.0	37.3	32.8	−32.0%	−15.4
Whites	33.8	33.1	31.5	36.1	34.3	33.9	29.3	28.1	28.0	25.3	21.9	21.0	−37.6%	−12.7
Blacks	71.9	70.4	52.2	69.1	67.5	49.3	39.6	39.2	42.0	38.3	38.6	27.9	−61.1%	−44.0
>=9%	51.9	47.0	49.2	48.0	48.4	50.5	50.6	43.5	38.1	38.3	33.8	34.0	‐34.6%	‐18.0
Hispanics	57.0	58.8	52.7	36.3	35.2	70.3	63.5	38.3	42.8	34.4	35.8	38.5	−32.5%	−18.5
Whites	45.8	38.6	46.7	45.7	43.1	43.5	41.6	37.9	34.0	34.2	30.8	29.6	−35.5%	−16.3
Blacks	82.6	87.9	55.3	82.4	84.3	83.0	99.2	78.4	45.7	52.2	33.0	41.9	−49.3%	−40.7

4.3. Stratified analyses of rates of serious hypoglycemia events by subpopulation: by comorbidity category

Patients with more comorbidity categories (Figure 1B, Suppl. Table 3) had higher rates of hypoglycemic events. The rate decreased across years regardless of the number of comorbidity categories (p <0.001). Similarly, patients with cardiovascular conditions only, with advanced diabetes complications only, and with both conditions (Figure 1C, Suppl. Table 3) also experienced downward trends (p<0.001).

Similar to the findings in the overall population, Blacks had higher rates and steeper declines in rates within subpopulations defined by the number of comorbidity categories (Figures D1–D4, Suppl. Table 3). The differential downward trends across racial/ethnic groups were statistically significant (p<0.05) for patients with 0, 1, and 2 comorbidity categories. We also observed similar patterns of racial differences in patients with cardiovascular conditions only, with advanced diabetes complications only, and with both conditions (Suppl. Table 3).

4.4. Stratified analyses of rates of serious hypoglycemia events by subpopulation: by glycemic control (insulin use and HbA1c level)

Tables 2 lists rates of serious hypoglycemia events by insulin use and HbA1c level based on proximal information. Patients were grouped into insulin users and non-users; the latter were composed of those who received secretagogues without insulin. Rates were predictably higher for insulin users than non-users; both groups showed marked declines across years (p<0.001), although there were greater declines in insulin users than non-users (p<0.001). The absolute rate change was −34.0 (from 68.8 in 2004 to 34.9 in 2015) among insulin users and −9.7 (20.5 to 10.8) among non-users.

Non-Whites had greater declines in rates than Whites in insulin users (p<0.001) as well as in non-users (p<0.01). For example, among insulin users the difference in rates for Blacks vs. Whites decreased from 34.7 (98.5 vs. 63.8) in 2004 to 13.2 (43.6 vs. 30.4) in 2015. Among insulin non-users, the Blacks-Whites contrasts in rates were 25.7 (44.1 vs. 18.4) in 2004 and 10.1 (18.9 vs. 8.8) in 2015. We observed similar patterns in analysis using the baseline medication (Suppl. Table 3).

The event rates also declined in 2004–2015 in all HbA1c groups (p<0.001). The decline trends differed by HbA1c group (p<0.001); e.g., patients with HbA1c range >=9% had the highest event rates and a larger decline across time. As in other subpopulations, non-Whites had steeper declines in rates than Whites in all HbA1c groups (p-values ranged from <0.01 to 0.04). There were similar patterns for the baseline HbA1c. (Suppl. Table 3).

4.5. Additional analysis: inter-relationships among insulin use, HbA1c values, and race/ethnicity

We then used the subset of patients who received insulin/secretagogues and had HbA1c test results available based on proximal information from the outcome years to analyze the inter-relationships among insulin use, HbA1c values, and occurrence of serious hypoglycemic events (Figure 2). We found that mean HbA1c levels were higher in insulin users than non-users in all racial/ethnic groups and across years (p<0.001). HbA1c levels of Whites were lower than Blacks and Hispanics among both insulin users and non-users (p<0.001). Regardless of race/ethnicity, mean baseline HbA1c values increased from 2004 to 2015 overall and for each racial group, but the main differences occurred between 2009 and 2015. We also found that HbA1c values were comparable (differences generally within 0.2%, in HbA1c NGSP unit) between patients who incurred serious hypoglycemia and those who did not, regardless of race/ethnicity, and whether on insulin or not.

Figure 2. Mean proximal HbA1c levels by medication use and presence of serious hypoglycemic events, overall and by race.

Results were based on 153,748 patients (49,319 insulin users and 104,429 non-users) who satisfied the following criteria: (1) on insulin/secretagogues 120 days prior to the hypoglycemic event, or death or end of the outcome year for those without an event, and (2) having at least one A1c value 365 days prior to the hypoglycemic event, or death or end of the outcome year for those without an event. The last HbA1c value was used (i.e., proximal HbA1c). There were no statistical differences in mean HbA1c levels between patient with and without any hypoglycemic event, regardless of race/ethnicity, in insulin users and non-users and in all three years (p>0.05). There were statistical differences in mean HbA1c levels between insulin user and non-users, regardless of race/ethnicity, and in all three years, based on two-sample t-tests (p<0.001). There were statistical differences in mean HbA1c levels among racial/ethnic groups, regardless of insulin use, and in all three years, based on ANOVA (p<0.001).

5. Discussion

Our study found a marked decrease in serious hypoglycemic event rates among Blacks, Whites, and Hispanics over a 12-year period (2004–2015). Although rates remained higher in Blacks than Hispanics and Whites over the study period, the absolute differences in rates narrowed markedly. The decrease in rates occurred across multiple comorbid conditions and in patients receiving insulin or secretagogues (greater in the insulin group). Additionally, we observed a small absolute increase in mean HbA1c levels, and a large decrease in proportion of patients with HbA1c <7%. Although these findings suggest that while relaxation of glycemic control was one mediator of the decline in hypoglycemia rates, the relationship of HbA1c levels with serious hypoglycemia is more complex. We found that the HbA1c values most proximal to a serious hypoglycemic event in Whites were similar or only slightly lower than those in Blacks and Hispanics (within 0.1%−0.3% difference in HbA1c NGSP unit) and were similar among patients who incurred serious hypoglycemia and those who did not. These observations suggest the importance of other unmeasured risk factors for serious hypoglycemia, including medications dosage changes, changes in medical status, and social determinants of health.

Our results of decreased overall rates of serious hypoglycemia starting in the late 2000 support prior US findings in Medicare² and managed care¹⁵, as well as finding from the UK .^16,17 In contrast to these studies, we report that the overall rates declined to below the initial rates. However, since our study had a later initial study year and included emergency department visits, results are not directly comparable to prior UK or Medicare studies.

With respect to racial differences, our findings of significant declines in both Blacks and Whites are similar to those reported in the US Medicare population in 1999–2011.² However, we found a marked narrowing of differences between Blacks and Whites, whereas a four-fold difference persisted over time in Medicare.² Our findings may more accurately reflect rates in the ambulatory care population since both the numerator and denominator are derived from the same individual-level electronic health records, rather than relying on a population-based estimate of the denominator.

Our results contrast with findings from a prior study that, based on an US managed health care ambulatory population aged 18 years or older, reported a decreased trend of hypoglycemic events for Whites but an increase for Blacks, and lower rates in Hispanics than Whites.³ The two studies have comparable methodologies, including similar definitions of study populations (receiving insulin or secretagogues) and serious hypoglycemia; both reported greater rates of hypoglycemia in patients with HbA1c >9% (than those in other HbA1c groups). However, our study only focused on patients aged 65 years or older and included an additional four years of data. There might also be potential differences in intensification of medication treatment. Furthermore, the prior study did not ascertain comorbid conditions and, hence, could not evaluate rates within comorbid conditions. In another study of the same population by the same group¹⁵, downward trends in the incidence of hypoglycemia were reported in subpopulations defined by presence of selected comorbidities but no discernable trends in the overall cohort. This underscores the importance of evaluating hypoglycemia rates stratified by comorbidity. We observed higher hypoglycemia rates in patients with greater illness burden (more comorbidities), similar to finding from claims studies of older patients in Medicare Advantage Plans.^18,19

Our study found marked declines in serious hypoglycemia events across race, medications and comorbidities; we believe the finding suggests significant changes in clinical practice as a result of modifications in guidelines and performance measures across the years. A strong supporting evidence is that the 2008/2009 onset of declines in rates of hypoglycemia in the current and prior studies^2,3,15–19 corresponds temporally to the termination of the ACCORD trial in early 2008²⁰ and the subsequent, almost immediate withdrawal of the National Committee for Quality Assurance <7% measure.²¹ This measure, first approved in 2006, had been applied to all individuals with diabetes 18–75 years old regardless of comorbid conditions or medications. In early 2009 the American Diabetes Association revised their guidelines to state that less stringent HbA1c values >7% may be appropriate for older patients, those with comorbid conditions, diabetes complications and hypoglycemia.²²

Over the course of this study, the FDA has approved a few medications that do not result in hypoglycemia. Therefore, it is likely that patients on insulin would be those with longer duration of diabetes and greater overall sickness, as evidenced by increased percentages of patients having 3 or 4 comorbidity categories and advanced diabetes complications over the years in this study. Although sicker patients may presumably bias the results towards higher rates, in this study we still detected marked declines as well as narrowed racial differences in rates of serious hypoglycemia in the overall population and within various subpopulations defined by insulin use, HbA1c level, and comorbidity. The robust findings in this higher risk older population add to the literature of serious hypoglycemia for observational population studies.

One plausible, but not provable hypothesis, for the decline in rates of serious hypoglycemia in Blacks is that on a population basis, they have higher HbA1c levels than Whites at any given level of average blood glucose.^23,24 This biological fact has been postulated to an increased risk of serious hypoglycemia if medication therapy, especially insulin, is intensified to meet a guideline or performance measurement goal for HbA1c.²³

This study has significant strengths, including a national scope, large population, and individual-level electronic health records with information on medications, laboratory values, comorbid conditions, and patient outcomes over a 12year period. In addition, it evaluated racial differences in various subpopulations defined by an extensive list of medical and mental conditions not considered in prior studies. Lastly, it was based on the recommendation by the DHHS to evaluate serious adverse drug events by specific medications, age, comorbid conditions and race/ethnicity; therefore, the findings may inform the policy on targeted surveillance and prevention efforts in reducing rates of serious hypoglycemic events.

Limitations of the study include the inability to ascertain Type 1 versus Type 2 diabetes, medication changes (especially for insulin), and HbA1c results performed in non-VA health care settings. However, the percentage of Veterans with type 1 diabetes would likely be less than the general population because disease onset earlier than 18 years old would disqualify individuals from military service. It is also not possible to ascertain hypoglycemic events that are treated only by paramedics; there could also be potential changes in coding practices over time.²⁵ However, these are limitations shared by similar studies evaluating rates of hypoglycemia in an ambulatory population.^3,15. We also acknowledge that we cannot account for possible changes over time in unmeasured social determinants of health, including food insecurity, social isolation, alcohol or drug use in the study population.

In conclusion, our study is the first to demonstrate reduction in rates of racial differences in serious hypoglycemic events in a large healthcare system using linked ambulatory care, pharmacy, laboratory, and claims data. Our findings of narrowing of racial differences were robust. They occurred in patients regardless of insulin or oral hypoglycemic agent only use, categories of comorbid conditions, age, and HbA1c control levels. We speculate that decrease in rates of hypoglycemia could be attributable to changes in national performance measures and professional society guidelines that relaxed recommendations for tight glycemic control. These changes led to adoption of individualized target levels over time.

Our findings provides empiric support for population based recommendations to consider system-level implementation strategies that focus on screening patients who are determined to be at higher risk for serious hypoglycemia. Such patients should not only have their glycemic management assessed, but be evaluated for social determinants of health, and comorbid conditions.^1,26 Several approaches are underway in the private²⁷ and Federal Sectors.²⁸ Because of the concerns of high health care costs associated with hypoglycemia²⁹, as well as of the associated adverse outcomes, including but not limited to falls and factures^30,31, it is important for healthcare systems and payors to consider adapting or adopting approaches for patient screening. Perhaps most importantly, it should be recognized that appropriate screening strategies may improve the quality of life of patients who incur hypoglycemic events, whether they are considered serious events or not.³²