Association between dysglycemia and the Charlson Comorbidity Index among hospitalized patients with diabetes

Abstract

Aim:

Inpatient dysglycemia has been linked to short-term mortality, but longer-term mortality data are lacking. Our aim was to evaluate the association between inpatient dysglycemia and one-year mortality risk.

Methods:

Retrospective chart review of adults with diabetes hospitalized between 2015 and 2019. The Charlson Comorbidity Index (CCI) was used to estimate 1-year mortality risk, stratified into low (CCI ≤ 5) and high risk (CCI ≥6). Simple and multivariable logistic regression was used to evaluate the association between dysglycemic measures and high mortality risk.

Results:

Among 22,639 unique admissions, BG ≥ 180, ≥300, ≤70, <54 and <40 mg/dL were associated with adjusted odds of 1.43 (95 % CI, 1.33, 1.54), 1.58 (95 % CI, 1.48, 1.68), 2.16 (95 % CI, 2.01, 2.32), 2.58 (95 % CI, 2.32, 2.86), and 2.56 (95 % CI, 2.19, 2.99) for high mortality risk, respectively. Older age and Black race were positively associated with hyperglycemia and hypoglycemia. Myocardial infarction, congestive heart failure (CHF), and moderate to severe liver disease were most strongly associated with hyperglycemia, while renal disease, CHF, peripheral vascular disease, and peptic ulcer disease were most strongly associated with hypoglycemia.

Conclusions:

Inpatient hypoglycemia and hyperglycemia were both positively associated with higher one-year mortality risk, with stronger magnitude of association observed for hypoglycemia. The association appears to be mediated mainly by presence of diabetes-related complications.

1. Introduction

Approximately 25 % of all hospital inpatient days are incurred by patients with a diagnosis of diabetes mellitus though they represent about 10 % of the adult population.¹ Dysglycemia, whether hyperglycemia or hypoglycemia, is common among inpatients with diabetes, with more than half experiencing at least one blood glucose (BG) of >250 mg/dL during their stay and up to 12–38 % experiencing a BG of <70 mg/dL.^2,3

Several studies have shown that inpatient hyperglycemia is associated with poor short-term outcomes, including increased rates of infection, higher risk of intensive care unit admission, longer hospital length of stay and increased in-hospital mortality^4–9 Inpatient hypoglycemia is perhaps even more significantly associated with increased length of stay and inpatient mortality.^10–14 By and large, studies have investigated the association between inpatient dysglycemia and short-term outcomes such as 30-day or in-hospital mortality, and evidence on long-term mortality risk is less robust. Turchin et al¹⁰ found an association of dysglycemia and inpatient mortality at one year using a registry of research patient data within a specific healthcare system. Akirov et al¹⁵ found an association between increased inpatient glycemic variability, after adjusting for hypoglycemia, and “end of follow up” mortality with the mean follow up period being 3 years. Glycemic variability was assessed by standard deviation and coefficient of variation of glucose (defined as the ratio of standard deviation to mean glucose values). This study utilized a population health registry to assess longer term mortality risk, but comparable registries are not as readily accessible in the United States. One of the challenges of evaluating the association between inpatient dysglycemia and longer-term mortality is the difficulty of acquiring validated death data for all discharged patients (e.g. using the National Death Index).

The Charlson Comorbidity Index (CCI) was first developed in 1987 to predict one-year mortality based on comorbidity data obtained from the hospital chart.¹⁶ It is the most commonly used mortality risk index and has been updated and validated for use with ICD-10 codes¹⁷ to predict long-term mortality risk of varying durations, even up to 15 years, in different clinical settings.^18,19 Though studies have looked at finding models for long-term mortality predictions in patients with diabetes,^20–22 these models are difficult to make use of in clinical practice given the large number of variables they incorporate. To our knowledge, two studies have utilized the CCI to predict long term mortality specifically in patients with diabetes.^23,24 However, the former included only patients with diabetic nephropathy and the latter was limited to the outpatient setting. Moreover, these studies did not explore the association between glycemic control and CCI, which is the focus of our study.

Based on previous studies, we hypothesized that inpatient dysglycemia is associated with an increased risk of mortality at one year as predicted by CCI and that hypoglycemia would be more strongly associated with mortality risk than hyperglycemia. We further hypothesized that the association between dysglycemia and high mortality risk would be mediated by a higher burden of diabetes-related complications in those with a higher CCI score, with dysglycemia being a marker of diabetes “severity” conceptualized as the presence of associated complications.

2. Subjects, material, and methods

2.1. Study design and participants

This was a retrospective cohort study of adult patients who were hospitalized at Johns Hopkins Hospital between December 1, 2015, and June 31, 2019, or Johns Hopkins Bayview Medical Center between June 1, 2016 and June 31, 2019, two academic medical centers located in Baltimore, Maryland. The start dates of each hospital cohort coincided with the introduction of the Epic electronic medical record (EMR) system in the inpatient setting at each hospital. Patient records were obtained from a bulk EMR query by Epic clarity certified data analysts. This study was approved by the Johns Hopkins School of Medicine Institutional Review Board.

Hospitalized adults aged ≥18 were included if they had a diagnosis of diabetes mellitus as indicated by the presence of an International Statistical Classification of Diseases and Related Health Problems, 10th revision (ICD-10) diagnosis code of E10 (Type 1 diabetes), E11 (Type 2 diabetes), E12 (Malnutrition-related diabetes), E13 (Other specified diabetes), and E14 (Unspecified diabetes) in hospital claims data. Patients were excluded if their BG was not measured during the admission. In cases where a patient had multiple admissions, only the most recent admission was used.

2.2. Exposure variables

All laboratory BG and point-of-care (POC) fingerstick glucose results were collected from the EMR. We classified inpatient hyperglycemia as either having one or more BG or POC glucose results ≥180 mg/dL (hyperglycemia) or ≥300 mg/dL (severe hyperglycemia). We chose these thresholds for the following reasons: ≥180 mg/dL has been used to define inpatient hyperglycemia requiring intervention by the American Diabetes Association (ADA)²⁵ and ≥300 mg/dL has been associated with increased in-hospital mortality.²⁶

We classified inpatient hypoglycemia as either having one or more BG or POC glucose results: ≤70 mg/dL, <54 mg/dL, and <40 mg/dL. The 70 mg/dL and 54 mg/dL thresholds correspond to the Level 1 and Level 2 glucose levels as defined by the International Hypoglycemia Study Group.²⁷ Although no specific glucose threshold defines severe hypoglycemia, in this study we used a level of <40 mg/dL (commonly used for this outcome) since it is difficult to ascertain the presence of cognitive impairment requiring third-party assistance in a retrospective study. Other relevant exposure variables included in the analysis were covariates that were thought to be associated with either CCI score or inpatient dysglycemia. The covariates included were age, sex, race, and body mass index (BMI).

2.3. Outcome: Charlson Comorbidity Index

The primary outcome of interest was the CCI score in relation to dysglycemic measures. The original version of the CCI developed by Mary E. Charlson consisted of 19 weighted items corresponding to different medical comorbidities. Over the years, several adaptations of the CCI for use with different data sources have been developed. Sundararajan et al²⁸ and later Quan et al²⁹ adapted the CCI for use with ICD-10 codes. These weighed the following factors: Myocardial infarction (MI), congestive heart failure (CHF), peripheral vascular disease (PVD), cerebrovascular disease (CEVD), dementia, chronic obstructive pulmonary disorder (COPD), rheumatoid disease, peptic ulcer disease (PUD), liver disease (LD) (uncomplicated designated as mild, or with end-organ damage designated as moderate to severe LD), hemiplegia (HP)/paraplegia (PAPL), diabetes mellitus (complicated or uncomplicated), moderate to severe renal disease (RD), any malignancy (including leukemia and lymphoma), metastatic solid tumor, and AIDS.

We used the STATA code developed by Stagg et al to calculate the CCI with ICD-10 codes derived from hospital claims data.³⁰ All patients in the study had either Type 1 or Type 2 diabetes mellitus, which corresponds to 1 point on the CCI. Therefore, no patients had a CCI score of 0 in this study. In the initial analysis, patients were divided into four levels based on their CCI score, low risk (score of 1 i.e. Diabetes without any other Charlson comorbidity), mild risk (score of 2–3), moderate risk (score of 4–5) and high risk (score of 6 or greater). This resulted in a scale similar to the one used by Charlson et al for their validation. The number of observations allowed us to simplify our classification of patients into one of two CCI score categories for subsequent analysis: low mortality risk (CCI ≤5) and high mortality risk (CCI ≥ 6). These categories were based on arbitrary cut offs similar to other studies using the CCI.¹⁸

2.4. Statistical analyses

Descriptive statistics were used to summarize the study population by stratification of CCI score category. Normality of continuous measures was assessed using the Shapiro-Wilk test. None of the continuous measures were found to be normally distributed, so medians and interquartile ranges (IQR) were reported for each continuous variable. The number and percentage were reported for categorical variables.

Univariable analyses were conducted to determine the differences in patient characteristics by stratification of CCI score category. Kruskal-Wallis tests were conducted for non-normally distributed continuous variables. The association between categorical and binary variables and CCI score categories were compared using Pearson’s χ². Multivariable logistic regression was then used to adjust for age, sex, race, and BMI.

Odds ratios (ORs) were reported with 95 % confidence intervals (CI) and two-sided p-value with significance set at 0.05. Statistical analyses were performed using Stata Statistical Software: Release 17 (College Station, TX: StataCorp LP).

3. Results

A total of 22,639 patients were included in this analysis. Table 1 shows the characteristics of the study population in the full cohort and stratified by mortality risk category. Overall, this cohort consisted of older adults (median age 65 years) with a slight male predominance (52.5 %). Greater than 50 % were Caucasian, with African American being the second most common race included (38.7 %) and the remaining races collectively accounting for 9.1 % of the sample population. Most patients were either overweight or obese with a median BMI of 29.6 kg/m². Most patients had a diagnosis of type 2 diabetes (95.7 %), with 3.8 % having type 1 diabetes and 0.5 % having diabetes otherwise not specified.

Table 1.

Characteristics of the study population for full cohort and stratified by mortality risk.

Factor	Full cohort	Low mortality risk	High mortality risk	p-Valuea
N	22,639	16,311	6328
Age, median (IQR)	65.0 (55.0, 74.0)	63.0 (53.0, 72.0)	68.0 (60.0, 76.0)	<0.001
Female, no. (%)	10,973 (48.5)	8179 (50.1)	3534 (55.8)	<0.001
Race, no. (%)				<0.001
White	11,810 (52.2)	8554 (52.4)	3256 (51.5)
Black or African American	8769 (38.7)	6145 (37.7)	2624 (41.5)
Other	2060 (9.1)	1612 (9.9)	448 (7.1)
BMI, median (IQR)	29.6 (25.1, 35.3)	29.9 (25.4, 35.8)	28.7 (24.3, 34.3)	<0.001
Diabetes type, no. (%)				<0.001
Type 2 diabetes	21,668 (95.7)	15,536 (95.2)	6132 (96.9)
Type 1 diabetes	856 (3.8)	685 (4.2)	171 (2.7)
Other	115 (0.5)	90 (0.6)	25 (0.4)
Charlson Index, median (IQR)	4.0 (2.0, 6.0)	–	–	–
Charlson Index Category, no (%)				–
1	3566 (15.8)	3566 (21.9)	0 (0)
2–3	7407 (32.7)	7407 (45.4)	0 (0)
4–5	5338 (23.6)	5338 (32.7)	0 (0)
6+ (high risk)	6328 (28.0)	0 (0)	6328 (100)
Comorbidities, no (%)
MI	4483 (19.8)	2252 (13.8)	2231 (35.3)	<0.001
CHF	6872 (30.4)	3272 (20.1)	3600 (56.9)	<0.001
PVD	2112 (9.3)	986 (6.0)	1126 (17.8)	<0.001
CEVD	2726 (12.0)	1616 (9.9)	1110 (17.5)	<0.001
Dementia	1470 (6.5)	817 (5.0)	653 (10.3)	<0.001
COPD	5987 (26.4)	3639 (22.3)	2348 (37.1)	<0.001
Rheumatoid disease	653 (2.9)	401 (2.5)	252 (4.0)	<0.001
PUD	336 (1.5)	151 (0.9)	185 (2.9)	<0.001
Mild LD	2240 (9.9)	1176 (7.2)	1064 (16.8)	<0.001
Diabetes with complication(s)	10,120 (44.7)	5190 (31.8)	4930 (77.9)	<0.001
HP/PAPL	689 (3.0)	351 (2.2)	338 (5.3)	<0.001
RD	6672 (29.5)	2297 (14.1)	4375 (69.1)	<0.001
Cancer	3133 (13.8)	1353 (8.3)	1780 (28.1)	<0.001
Moderate/severe LD	536 (2.4)	99 (0.6)	437 (6.9)	<0.001
Metastatic cancer	1300 (5.7)	0 (0.0)	1300 (20.5)	–
AIDS	242 (1.1)	0 (0.0)	242 (3.8)	–
Any hyperglycemic event, no (%)				<0.001
≥180 mg/dL	17,453 (77.2)	12,303 (75.6)	5150 (81.4)
≥300 mg/dL	7294 (32.3)	4883 (30.0)	2411 (38.1)
Any hypoglycemic event, no (%)				<0.001
≤70 mg/dL	4284 (19.0)	2520 (15.5)	1764 (27.9)
<54 mg/dL	1656 (7.3)	870 (5.3)	786 (12.4)
<40 mg/dL	689 (3.0)	353 (2.2)	336 (5.3)

Abbreviations: IQR, interquartile range; BMI, body mass index; MI, myocardial infarction; CHF, congestive heart failure; PVD, peripheral vascular disease; CEVD, cerebrovascular disease; COPD, chronic obstructive pulmonary disease; PUD, peptic ulcer disease; LD, liver disease; HP/PAPL, hemiplegia or paraplegia; RD, renal disease.

^aFor comparison between low and high mortality risk.

Notably, all the patient characteristics and dysglycemic measures differed between the two mortality risk groups. Patients in the high mortality risk group were older (median age of 68 vs. 63 years), more likely to be male (55.8 % vs. 50.1 %), more likely to be Black/African American (41.5 % vs. 37.7 %), have lower BMI (28.7 vs. 29.9 kg/m²), and less likely to have type 1 diabetes (2.7 % vs. 4.2 %). All the CCI comorbidities were more prevalent in the higher mortality risk group, with between-group differences greatest for MI, CHF, PVD, CEVD, dementia, mild LD, PUD, diabetes with complication(s), RD, cancer/metastatic cancer. Of note, renal disease occurred nearly four times more frequently in the high-risk group (69.1 % vs. 14.1 %). All dysglycemic outcomes were more common in the high-risk group. Rates of severe hypoglycemia (BG <40 mg/dL) were nearly double in the high mortality risk group.

Table 2 shows the unadjusted and adjusted association of the dysglycemic measures and high mortality risk. In the unadjusted model, BG ≥180 mg/dL was associated with a 1.41-fold (95 % CI 1.31, 1.52) increased odds of high mortality risk, with similar effect size observed with BG ≥300 mg/dL (OR 1.44, 95 % CI 1.35, 1.53). There were even greater effect sizes observed with the hypoglycemic outcomes, with odds ratios of 2.11 (95 % CI 1.97, 2.26), 2.51 (95 % CI 2.27, 2.78), and 2.53 (95 CI 2.17, 2.95) for BG ≤70 mg/dL, <54 mg/dL, and <40 mg/dL, respectively. In a model adjusting for age, sex, race, and BMI, the magnitude of the effect slightly increased for all dysglycemic measures. For the hyperglycemic measures, the adjusted odds ratios (aORs) for high mortality risk category were 1.43 (95 % CI 1.33, 1.54) and 1.58 (95 % CI 1.48, 1.68) for BG ≥180 mg/dL and ≥300 mg/dL, respectively. For the hypoglycemic measures, the aORs for high mortality risk category were 2.16 (95 % CI 2.01, 2.32), 2.58 (95 % CI 2.32, 2.86), 2.56 (95 % CI 2.19, 2.99) for BG ≤70 mg/dL, <54 mg/dL, and <40 mg/dL, respectively.

Table 2.

Unadjusted and adjusted association of mortality risk with dysglycemic measures.

	Unadjusted model	Adjusted modela
	OR of high mortality risk (95 % CI)	OR of high mortality risk (95 % CI)
Any hyperglycemic event
≥180 mg/dL	1.41 (1.31, 1.52)	1.43 (1.33, 1.54)
≥300 mg/dL	1.44 (1.35, 1.53)	1.58 (1.48, 1.68)
Any hypoglycemic event
≤70 mg/dL	2.11 (1.97, 2.26)	2.16 (2.01, 2.32)
<54 mg/dL	2.51 (2.27, 2.78)	2.58 (2.32, 2.86)
<40 mg/dL	2.53 (2.17, 2.95)	2.56 (2.19, 2.99)

^aAdjusted for age, sex, race, and BMI.

Table 3 shows the association of dysglycemic measures with patient characteristics and each of the Charlson comorbidities in a fully adjusted model (adjusted for age, sex, race, BMI, and each of the other CCI comorbidities). This analysis was done to explore how dysglycemia may be linked to higher CCI scores through associations with specific comorbidities. There was an association between male sex and lower risk of hypoglycemia of ≤70 mg/dL and <54 mg/dL and severe hyperglycemia of ≥300 mg/dL. African American race was positively associated with all degrees of hypoglycemia. The adjusted model revealed a negative association between African American race and mild hyperglycemia with aOR 0.76 (95 % CI 0.71, 0.81). Those with higher BMI were less likely to have a hypoglycemic event.

Table 3.

Multivariable association of patient characteristics and comorbidities with dysglycemic outcomes.

	OR (95 % CI)a of any dysglycemic event
	Any hyperglycemic event		Any hypoglycemic event
	≥180 mg/dL	≥300 mg/dL	≤70 mg/dL	<54 mg/dL	< 40 mg/dL
Age	1.00 (1.00, 1.00)	0.98 (0.97, 0.98)	0.99 (0.98, 0.99)	0.98 (0.98, 0.99)	0.98 (0.98, 0.99)
Female	1.00 (ref)	1.00 (ref)	1.00 (ref)	1.00 (ref)	1.00 (ref)
Male	1.10 (1.03, 1.17)	0.92 (0.87, 0.97)	0.77 (0.72, 0.83)	0.81 (0.73, 0.91)	0.87 (0.74, 1.02)
Race
White	1.00 (ref)	1.00 (ref)	1.00 (ref)	1.00 (ref)	1.00 (ref)
Black	0.76 (0.71, 0.81)	1.00 (0.94, 1.06)	1.26 (1.17, 1.36)	1.17 (1.04, 1.30)	1.43 (1.21, 1.68)
Other	1.27 (112, 1.44)	1.18 (1.07, 1.31)	0.94 (0.83, 1.07)	0.83 (0.68, 1.02)	0.96 (0.71, 1.30)
BMI	1.00 (0.99, 1.00)	0.98 (0.97, 0.98)	0.96 (0.95, 0.96)	0.95 (0.94, 0.96)	0.95 (0.94, 0.96)
MI	1.31 (1.20, 1.43)	1.28 (1.19, 1.38)	1.09 (0.99, 1.19)	1.14 (1.00, 1.29)	1.23 (1.02, 1.48)
CHF	1.10 (1.02, 1.18)	1.28 (1.19, 1.37)	1.31 (1.21, 1.42)	1.40 (1.24, 1.58)	1.37 (114, 1.63)
PVD	0.91 (0.82, 1.02)	0.89 (0.80, 0.99)	1.20 (1.07, 1.34)	1.18 (1.02, 1.39)	1.28 (1.01, 1.62)
CEVD	1.12 (1.01, 1.24)	0.99 (0.90, 1.08)	1.19 (1.07, 1.32)	1.10 (0.95, 1.29)	1.21 (0.97, 1.52)
Dementia	0.86 (0.75, 0.98)	1.17 (1.04, 1.32)	1.32 (115, 1.51)	1.23 (1.01, 1.49)	1.20 (0.89, 1.61)
COPD	0.85 (0.79, 0.91)	1.01 (0.95, 1.08)	1.02 (0.94, 1.11)	1.05 (0.94, 1.18)	1.06 (0.89, 1.27)
Rheumatoid disease	0.74 (0.62, 0.88)	0.95 (0.80, 1.12)	1.15 (0.94, 1.39)	1.23 (0.94, 1.61)	1.47 (1.01, 2.15)
PUD	1.15 (0.87, 1.52)	0.93 (0.74, 1.18)	1.86 (1.47, 2.37)	1.59 (115, 2.21)	1.67 (1.05, 2.64)
Mild LD	1.00 (0.89, 1.12)	0.93 (0.84, 1.03)	1.10 (0.98, 1.24)	1.11 (0.93, 1.32)	1.15 (0.88, 1.48)
Diabetes and complications	1.82 (1.67, 1.98)	1.65 (1.54, 1.78)	1.32 (1.21, 1.44)	1.41 (1.23, 1.61)	1.29 (1.05, 1.58)
HP/PAPL	1.22 (0.99, 1.50)	1.23 (1.04, 1.46)	1.21 (0.99, 1.47)	1.51 (1.15, 1.96)	1.50 (1.02, 2.20)
RD	0.82 (0.75, 0.90)	0.91 (0.84, 0.98)	1.69 (1.54, 1.86)	1.72 (1.50, 1.97)	1.65 (1.34, 2.03)
Cancer	1.06 (0.95, 1.17)	0.95 (0.86, 1.05)	1.04 (0.92, 1.17)	1.07 (0.90, 1.28)	1.19 (0.92, 1.54)
Moderate/severe LD	1.51 (1.18, 1.95)	1.44 (1.19, 1.75)	1.85 (1.50, 2.28)	2.16 (1.64, 2.84)	2.43 (1.66, 3.54)
Metastatic cancer	1.10 (0.94, 1.28)	1.02 (0.88, 1.17)	1.13 (0.96, 1.33)	1.12 (0.87, 1.43)	0.87 (0.59, 1.29)
AIDS	0.59 (0.45, 0.77)	0.80 (0.61, 1.06)	1.02 (0.75, 1.39)	0.69 (0.42, 1.12)	0.73 (0.36, 1.45)

OR, odds ratio; CI, confidence interval; BMI, body mass index; MI, myocardial infarction; CHF, congestive heart failure; PVD, peripheral vascular disease; CEVD, cerebrovascular disease; COPD, chronic obstructive pulmonary disease; PUD, peptic ulcer disease; LD, liver disease; HP/PAPL, hemiplegia or Paraplegia; RD, renal disease; AIDS, acquired immunodeficiency syndrome.

^aBold text indicates statistical significance.

Among the Charlson comorbidities, MI, CHF, and moderate/severe LD were associated with increased risk of hyperglycemia of all degrees, while dementia was linked only to increased risk of severe hyperglycemia. History of dementia, COPD and AIDS appeared to be negatively associated with mild hyperglycemia. MI, CHF, PVD, CEVD, dementia, rheumatoid disease, PUD, hemiparesis, RD and moderate to severe LD were all predisposed to hypoglycemia. MI was associated only with severe hypoglycemia in the adjusted model with aOR of 1.23 (95 % CI 1.02, 1.48). CHF and PVD were associated with all degrees of hypoglycemia. CEVD and dementia were only associated with hypoglycemia of ≤70 mg/dL. PUD, RD and moderate to severe LD were associated with hypoglycemia of all degrees, the latter having the strongest relationship with severe hypoglycemia. The aOR for PUD, RD and moderate to severe LD and having a BG of ≤40 mg/dL were 1.67 (95 % CI 1.05, 2.64), 1.65 (95 % CI 1.34, 2.03) and 2.43 (95 % CI 1.66, 3.54) respectively.

Supplemental Table 1 shows the univariable association of dysglycemic outcomes with each of the comorbidities used to calculate the CCI score. Supplemental Tables 2 and 3 show the univariable and multivariable associations of dysglycemic outcomes with patient characteristics and CCI comorbidities among the high-mortality risk subgroup, respectively. Supplemental Table 4 shows the breakdown of various medications that patients received during admission that could either predispose to hypoglycemia (i.e. insulin and non-insulin antihyperglycemic agents) or hyperglycemia (i.e. systemic glucocorticoids). Notably, approximately 38 % and 74 % of the cohort received basal (long-acting, intermediate-acting, or fixed combination) insulin and rapid-acting insulin, respectively. Per our hospital protocol, very few patients were treated with non-insulin agents during admission. Nearly 20 % of the patients were receiving systemic glucocorticoids.

4. Discussion

In this large retrospective study, we observed an association between dysglycemic measures, both hyperglycemia and hypoglycemia, and high one-year mortality risk among hospitalized patients with diabetes as estimated using the CCI score. Both mild and severe hyperglycemia were associated with high mortality risk; notably, all degrees of hypoglycemia had an even stronger association with increased mortality risk, with BG <54 mg/dL having aOR of 2.58 (95 % CI 2.32, 2.86) and BG <40 mg/dL with aOR of 2.56 (95 % CI 2.19, 2.99). This demonstrates the negative prognostic value of even a single hypoglycemic event during hospitalization as has been seen in previous studies.¹⁰ Our study was unique in its evaluation of mortality risk at several different glucose thresholds for both hyperglycemia and hypoglycemia.

Although our study identified associations between dysglycemia and increased mortality risk, it does not confirm causation. Though studies have shown glycemic variability may independently increase mortality risk after controlling for other variables,³¹ these looked at short term outcomes. It is plausible that patients with multiple comorbidities (i.e. higher CCI score independent of their diabetes status) would experience larger swings in BG related to changes in renal and liver function, exposure to medications for underlying conditions (e.g. steroids) or stress-hyperglycemia in setting of acute illness (e.g. MI). We hypothesized that the association between dysglycemia and mortality risk would be mediated by diabetes-related complications, such as MI, CHF, RD, mild and moderate LD, etc., more so than comorbidities that could not otherwise be clearly linked with a history of diabetes (e.g. COPD, PUD, rheumatoid disease, AIDS).

Our findings confirmed our hypothesis as dysglycemia was indeed associated strongly with conditions that are known sequalae of diabetes. Specifically, hyperglycemia was linked to history of MI and CHF. Multiple studies have shown that hyperglycemia is common following MI and that elevated plasma glucose levels are an independent predictor of in-hospital mortality regardless of diabetes status.³² Glycemic variability has also been found to be associated with increased risk of cardiovascular events in patients with CHF.³³ Acute hyperglycemia has multiple negative effects on cardiovascular health by increasing endothelial dysfunction, platelet hyperreactivity and oxidative stress³⁴ that may impact long-term cardiovascular health and contribute to worse outcomes. Alternatively, these conditions themselves could trigger acute stress-related hyperglycemia, with hyperglycemia being a measure of the degree of severity of illness and, in turn, predicting a poor prognosis.

Moderate to severe LD remained strongly correlated with predisposition to hyperglycemia. Poorly controlled diabetes mellitus could predispose to nonalcoholic fatty liver disease (NAFLD), currently the most common cause of liver disease in the US, the leading cause of hepatocellular carcinoma and indication for liver transplant.³⁵ Dysglycemia could exacerbate these inflammatory processes and, consequently, predispose to more severe LD and worse long-term prognosis. As previously discussed, since our study does not confirm causation and only an association, this finding could be another example of stress hyperglycemia in the setting of severe illness. Though hemiparesis, rheumatoid disease and RD were associated with hyperglycemia in the univariable analysis, this was not seen in the multivariable analysis. COPD was interestingly negatively associated with mild hyperglycemia for unclear reasons.

Hypoglycemia was linked to several patient factors and conditions in this study. African American race was particularly associated with hypoglycemia of all degrees. It has been postulated that African Americans tend to be prescribed more aggressive antihyperglycemic regimens due to provider specific factors. Studies have shown that African Americans tend to be “high glycators,” with hemoglobin A1c (A1C) overestimating the mean BG.³⁶ If falsely elevated A1C is being used to make treatment decisions, African Americans could be at higher risk of hypoglycemia. Higher BMI was associated with lower prevalence of hypoglycemia of all degrees, likely due to insulin resistance in these patients.

All the Charlson comorbidities were positively associated with hypoglycemia except for COPD, mild LD, AIDS, cancer and metastatic cancer, for which there was a nonsignificant difference in the multivariate analysis. Hypoglycemia is known to have multiple effects on cardiovascular health, mostly through activation of the sympathoadrenergic system, potentially leading to endothelial activation, and localized tissue ischemia precipitated by major vascular events such as MI and CEVD.³⁷ Among cardiovascular diseases, CHF was most strongly associated with hypoglycemia. PVD was also associated with all degrees of hypoglycemia, while MI was only associated with severe hypoglycemia. A recent study³⁸ found that CHF development in patients with type 2 diabetes is particularly predictive of shorter lifespan when compared to development of other cardiovascular diagnoses such as MI and CEVD. It is possible that the noted association of hypoglycemia frequency and severity with CHF plays a role in the higher mortality risk compared to other cardiovascular diseases, which seem to have a more modest predisposition to hypoglycemia. The mechanism of hypoglycemia in severe CHF remains unclear but could be related to inhibited gluconeogenesis given the frequent finding of elevated lactic acid levels in this condition.³⁹ A history of PUD was unexpectedly associated with increased risk of hypoglycemia. PUD could be associated with other changes in nutritional intake (e.g. enteral nutrition, NPO status for endoscopic procedures), which increased risk of hypoglycemia, but further studies are needed to elucidate the mechanism behind this association.

RD was strongly associated with hypoglycemia, which is expected given the known role of the kidney in clearance of insulin and gluconeogenesis. One study also showed that CCI has been able to predict worse outcomes in patients with diabetic nephropathy, indicating more than one factor is in play.⁴⁰ Hypoglycemia was also associated with the diagnosis of moderate to severe LD, likely related to impaired gluconeogenesis⁴¹ and thus inadequate ability to respond to a drop in BG.

Several studies report that patients with type 2 diabetes have a poorer prognosis following cardiovascular and renal complications than the general population as noted above. Most studies focus on a single complication only, hence, information on the impact of more than one diabetic complication or the relative magnitude of each complication is lacking prior to our study. An increased awareness of the potential comorbidities in patients with type 2 diabetes that contribute to increased mortality may provide insights into this complicated disease.

Once diabetes-related complications develop, this may be a time to reconsider glycemic targets as our main goal should then be to reduce risk of hypoglycemia in susceptible patients.⁴² In recent years, there has been a shift in practice in which glycemic targets are liberalized in patients with advanced age or multiple comorbidities. The American College of Physicians (ACP) recommends an A1C target of 7–8 % for most patients with diabetes and reserves a target of 6.5 % to 7 % for patients with newly diagnosed diabetes and those without diabetes-related complications.⁴³ Furthermore, the ACP recommended against specific glycemic targets for older adults and those with limited life expectancy, without providing specific guidance on what equates to a limited life expectancy, which can be a barrier for providers in making objective assessments about intensity of glycemic control. Though studies have looked at finding models for long-term mortality predictions in patients with diabetes given the fact that clinicians frequently under or overestimate life expectancy,²⁰ these models are difficult to make use of in clinical practice given the large number of variables they incorporate. We propose that the CCI may be an effective, simpler, and more recognizable tool to identify those patients who have a limited life expectancy and would therefore be candidates for more relaxed glycemic goals.

4.1. Limitations and strengths

This study was not designed to determine a causal relationship between dysglycemia and mortality. Indeed, other studies have found that hypoglycemia in particular could marker of illness rather than direct cause of mortality.¹⁴ Another limitation was that our study used claims data extracted from the EMR to calculate the CCI. This method, though more efficient than manual extraction, is dependent on accurate data entry by providers. We recently found that ICD-10 based diagnoses for diabetes are fairly reliable when compared to gold-standard diagnostic indicators.⁴⁴ Of the types of electronic data we could have analyzed, we chose claims data that tends to agree most with patient chart characteristics in previous studies,⁴⁵ though some comorbidities tend to be under-reported.

There are strengths to this study, including the use of glucose data from a large cohort of hospitalized patients. To our knowledge, this is the first study to explore differing degrees of both hyperglycemia and hypoglycemia in relation to one-year mortality risk using CCI. This study used CCI, a well-validated, non-proprietary, and relatively simple tool that can easily calculate risk using data that are already available in electronic health records.

5. Conclusions

Our study found a positive association between multiple measures of inpatient hyperglycemia and hypoglycemia and one-year mortality as predicted by a validated mortality risk assessment tool, the CCI. Dysglycemic measures were most strongly associated with MI, CHF, PVD, RD and LD, which is most likely driven by higher prevalence of most of these conditions in patients with long-standing diabetes. Further studies could evaluate use of the CCI score to determine individualized A1C targets by providing objective information on life-expectancy, an important component in developing personalized glycemic goals.

Abbreviations

CCI

Charlson Comorbidity Index

ICD-10

International Statistical Classification of Diseases and Related Health Problems, 10th revision

BG

Blood glucose

POC

Point-of-care

BMI

Body mass index

MI

Myocardial infarction

CHF

Congestive heart failure

PVD

Peripheral vascular disease

CEVD

Cerebrovascular disease

COPD

Chronic obstructive pulmonary disorder

PUD

Peptic ulcer disease

LD

Liver disease

HP

Hemiplegia

PAPL

Paraplegia

RD

Renal disease

A1C

Hemoglobin A1C

Appendix A. Supplementary data

Supplementary data to this article can be found online at https://doi.org/10.1016/j.jdiacomp.2022.108305.