Case-control Investigation of Previously Undiagnosed Diabetes in the Critically Ill

Abstract

Context

The outcome of patients requiring intensive care can be influenced by the presence of previously undiagnosed diabetes (undiagDM).

Objective

This work aimed to define the clinical characteristics, glucose control metrics, and outcomes of patients admitted to the intensive care unit (ICU) with undiagDM, and compare these to patients with known DM (DM).

Methods

This case-control investigation compared undiagDM (glycated hemoglobin A_1c [HbA_1c] ≥ 6.5%, no history of diabetes) to patients with DM. Glycemic ratio (GR) was calculated as the quotient of mean ICU blood glucose (BG) and estimated preadmission glycemia, based on HbA_1c ([28.7 × HbA_1c] – 46.7 mg/dL). GR was analyzed by bands: less than 0.7, 0.7 to less than or equal to 0.9, 0.9 to less than 1.1, and greater than or equal to 1.1. Risk-adjusted mortality was represented by the Observed:Expected mortality ratio (OEMR), calculated as the quotient of observed mortality and mortality predicted by the severity of illness (APACHE IV prediction of mortality).

Results

Of 5567 patients 294 (5.3%) were undiagDM. UndiagDM had lower ICU mean BG (P < .0001) and coefficient of variation (P < .0001) but similar rates of hypoglycemia (P = .08). Mortality and risk-adjusted mortality were similar in patients with GR less than 1.1 comparing undiagDM and DM. However, for patients with GR greater than or equal to 1.1, mortality (38.5% vs 10.3% [P = .0072]) and risk-adjusted mortality (OEMR 1.18 vs 0.52 [P < .0001]) were higher in undiagDM than in DM.

Conclusion

These data suggest that DM patients may develop tolerance to hyperglycemia that occurs during critical illness, a protective mechanism not observed in undiagDM, for whom hyperglycemia remains strongly associated with higher risk of mortality. These results may shed light on the natural history of diabetes.

The prevalence of diabetes (DM) is increasing worldwide. Recent data suggest that the age-standardized prevalence has increased in the United States from 9.8% of the population in 1999 to 2000 to 14.3% in 2017 to 2018 [1]. The percentage of hospitalized patients with DM is considerably higher, owing to the high burden of comorbidities associated with this disease [2]. Hyperglycemia can also occur in hospitalized patients not previously diagnosed with DM, especially among those admitted with critical illness. For these patients, the association between hyperglycemia and hospital outcome is different from that for patients previously known to have DM [3–8]. Among patients without DM, higher mean blood glucose (BG) during intensive care unit (ICU) admission has been consistently demonstrated to be associated with higher mortality [3–8]. In contrast, the relationship between mean ICU BG and mortality for DM patients is not consistent [3–8]. Recent literature has demonstrated the importance of preadmission glycemia, reflected by glycated hemoglobin A_1c (HbA_1c) levels. For patients with low HbA_1c, high mean ICU BG is strongly associated with high mortality, but the opposite is seen for patients with high HbA_1c at admission, for whom high mean ICU BG is associated with lower mortality [9–18].

Many patients admitted to the ICU without a previous diagnosis of DM who, based on measurement of HbA_1c at the time of ICU admission, are diagnosed as having DM after hospital admission, referred to herein as previously undiagnosed DM (undiagDM) [19–24]. Previous studies have not provided detailed information about comorbidities, or glucose metrics and their relationships to outcome for this subset.

The purpose of this investigation is to delineate clinical characteristics, glucose metrics, and mortality of patients admitted to a mixed medical-surgical ICU with undiagDM, based on HbA_1c greater than or equal to 6.5% obtained at or near ICU admission. We used case-control methodology to compare this group of patients to those with previously diagnosed DM. We hypothesized that (1) glucose metrics during ICU admission would differ between the 2 groups and (2) that the relationship of ICU glucose metrics to risk-adjusted mortality would be different for undiagDM and patients with known DM.

Materials and Methods

Patients and Study Center

This investigation includes adult patients admitted consecutively to the 20-bed Stamford Hospital ICU with at least 4 BG levels obtained during ICU admission and HbA_1c level obtained at admission, or within 3 months of admission. The study excludes patients younger than 18 years, patients admitted with diabetic ketoacidosis or hyperglycemic hyperosmolar state, and patients admitted following cardiovascular surgery (because of their very low mortality). Fig. 1 details the flow of included and excluded patients.

Figure 1.

Derivation of the patient cohorts, including dates of admission, inclusions, and exclusions.

The center is a teaching hospital associated with Columbia University College of Physicians and Surgeons. The ICU cares for a broad array of medical and surgery patients; organ transplantation is not performed. Care is delivered by multidisciplinary teams led by medical and surgical intensivists; nursing-to-patient ratio is 1:2 or 1:1, based on acuity and severity of illness.

Glucose Control Strategy

Details of the ICU's glucose control protocol have been published previously [25]. In brief, before October 14, 2014, all patients had a BG target range of 90 to 125 mg/dL. After this date “tight” (80-140 mg/dL) or “loose” (110-160 mg/dL) BG target ranges were chosen based on HbA_1c less than 7.0% or greater than or equal to 7.0%, respectively. Continuous intravenous (IV) regular insulin infusion was initiated when BG exceeded 180 mg/dL on 2 successive readings. Mild hyperglycemia in the 140 to 180 mg/dL range was treated with short-acting subcutaneous analogue insulin injections. The protocol mandated monitoring frequency as every 3 hours for all patients and every hour if the patient was treated with IV insulin. While arterial or venous blood was used preferentially, capillary blood was also tested. Measurement technology included the GEM 4000 arterial blood gas analyzer (Instrument Laboratories) or Accu-Chek Inform II glucometer (Roche Diagnostics).

Case-Control Methodology

Patients with a history of diabetes were prospectively identified at the time of ICU admission (DM). Those with newly diagnosed diabetes (undiagDM) were defined as having an HbA_1c greater than or equal to 6.5% but no prior history of diabetes.

A total of 294 patients were identified as undiagDM. Two investigators (Y.W. and J.B.), unaware of other patient comorbidities, glucose control metrics, or clinical outcomes, performed 1:1 matches of undiagDM with appropriate DM matches using 3 criteria:

• HbA_1c 6.5% to 7.9% or HbA_1c greater than or equal to 8.0%;

• admission to the ICU with a medical or surgical diagnosis; or

• admission to the ICU before or after the change in BG treatment protocols.

Data Abstraction and Statistical Plan

The ICU's comprehensive database including demographics, admitting diagnoses, severity of illness scores and clinical outcomes, was maintained manually by one of the authors (J.K.) through December 31, 2018. After this date these data were abstracted automatically by the Phoenix data management system (Medical Decisions Network). BG values, including all point-of-care data, were abstracted from the hospital's electronic medical record. Comorbidities were abstracted by review of the electronic medical record by 2 of the investigators (Y.W. and J.B.).

Continual data are presented as median (interquartile range) and compared using the Mann-Whitney test for nonparametric values. Categorical data are presented as numbers and percentages and compared using the chi-square test.

Risk-adjusted mortality was represented by the observed:expected mortality ratio (OEMR), calculated using the Acute Physiology and Chronic Health Evaluation IV predictions of mortality (APIV PM) [26]; OEMR equals the quotient of observed mortality percentage and the mean APIV PM, with ratios less than 1 indicating that the patient cohort had a lower mortality rate than the model predicted and ratios greater than 1 indicating that the patient cohort had a higher mortality rate than the model predicted. We compared these ratios using the Z test for independent proportions for between-group comparisons [27].

The APACHE IV model calculates an individual patient's prediction of mortality before hospital discharge based on a large number of different parameters: age; 15 vital sign and laboratory values derived from the first 24 hours of ICU admission; the Glasgow Coma Scale, for assessment of neurologic status; a discrete group of comorbidities (AIDS, cirrhosis, hepatic failure, immunosuppression, lymphoma, leukemia or myeloma, metastatic tumor); ICU admission diagnosis (n = 116); ICU admission source (general ward, emergency room, operating/recovery room, stepdown unit, direct admission, other ICU, other hospital); the length of stay in the hospital before ICU admission; and mechanical ventilation. The model does not include diabetes among the comorbidities as it did not contribute to the model's precision; maximum BG greater than or equal to 300 during the first 24 hours contributes a very small amount to the score.

We calculated glycemic ratio (GR) [18], a metric that describes the extent of divergence of acute glycemia from preadmission glycemia, as the quotient of mean ICU BG and estimated preadmission BG, using the Nathan formula based on HbA_1c ([28.7 × HbA1c] – 46.7 mg/d) [28]. We stratified the relationship between GR and mortality as well as GR and risk-adjusted mortality, represented by OEMR, by the bands of GR less than 0.7, 0.7 to 0.9, 0.9 to 1.1 and 1.1 or greater.

A P value of less than .05 was considered statistically significant.

We used MedCalc Statistical Software (version 18.11.6) for statistical analysis (MedCalc Software bvba; https://www.medcalc.org; 2019).

Results

Of 5567 patients who met inclusion criteria during the period of the investigation, 3721 (66.8%) had no prior history of diabetes and an HbA_1c less than 6.5%, 1552 (27.9%) had a history of diabetes (DM), and 294 (5.3%) were undiagDM (see Fig. 1). Table 1 displays clinical characteristics of undiagDM and their DM matches. Patients with DM were older and less likely to be male than were undiagDM. They were more likely to have hypertension, hyperlipidemia, coronary artery disease, and peripheral vascular disease and their renal function was worse than were undiagDM.

Table 1.

Clinical characteristics and outcomes of known diabetes mellitus patients and previously undiagnosed diabetes mellitus patients

	DM	undiagDM	P
No.	294	294
Age, y	72 (60-82)	68 (59-79)	.0117
Male, %	53.7	62.9	.0238
BMI	28.9 (25.0-34.1)	28.1 (24.5-34.0)	.23
eGFR, % of patients, mL/min/1.73 m2
> 60	46.6	60.2	.0010
40-59	21.1	19.0	.53
20-39	16.3	13.6	.36
< 20	16.0	7.1	.0007
Comorbidities, %
Current smoker	8.2	11.2	0.22
Past smoker	39.8	34.0	.15
Hypertension	83.3	68.0	< .0001
Hyperlipidemia	78.2	60.5	< .0001
Coronary artery disease	54.1	33.0	< .0001
Cerebrovascular disease	25.2	24.8	.91
Peripheral vascular disease	29.3	12.6	< .0001
Preadmission meds, %
Statins	68.4	52.0	< .0001
ARB/ARB	50.7	29.9	< .0001
Insulin	48.3	N/A
Metformin	32.7	N/A
Sulfonylureas	21.5	N/A
SGLT2	1.7	N/A
GLP1	1.4	N/A
DPP-4	17.0	N/A
Diagnostic categories
Medical	77.9	77.9
Cardiac	28.2	22.8	.13
Respiratory	20.7	19.0	.61
Neurologic	10.6	10.2	.87
Septic shock	9.9	10.9	.69
Gastrointestinal	4.1	7.5	.0781
Other	4.4	7.5	.11
Surgical/trauma, %	22.1	22.1
Vascular	6.5	5.1	0.47
Gastrointestinal	6.1	4.8	.49
Neurologic	2.4	3.4	.47
Respiratory	1.0	2.7	.13
Other	2.7	1.7	.41
Trauma	3.4	4.4	.53
ICU LOS, d	1.9 (1.1-3.9)	2.0 (1.0-4.2)	.91
APACHE IV predicted mortality, %	22.5 (1.4)	23.2 (1.5)	.75
Mortality, %	13.6	16.7	.29
Observed:expected mortality ratio	0.60	0.72	.0021

Abbreviations: ARB/ARB, angiotensin-converting enzyme/angiotensin receptor blockade; APACHE, Acute Physiology and Chronic Health Evaluation; BMI, body mass index; DM, known diabetes patients; DPP-4, dipeptidyl peptidase 4; eGFR, estimated glomerular filtration rate; LOS, length of stay; GLP-1, glucagon-like peptide 1; ICU, intensive care unit; SGLT2, sodium-glucose cotransporter 2; undiagDM, previously undiagnosed diabetes patients.

Although hospital mortality was statistically similar between the 2 groups, undiagDM had higher risk-adjusted mortality (represented by OEMR) than did DM.

Table 2 details glucose metrics for the 2 groups. Patients with DM had higher mean BG and higher coefficient of variation (CV) than did undiagDM and a higher percentage received insulin in the ICU.

Table 2.

Glucose metrics of known diabetes patients and previously undiagnosed diabetes mellitus patients

	DM	undiagDM	P
Mean blood glucose, mg/dL	151 (136-171)	138 (123-158)	< .0001
Coefficient of variation, %	27.4 (20.7-34.8)	21.7 (16.3-29.5)	< .0001
Hypoglycemia < 70 mg/dLa	23.8	18.0	.0840
IV insulin received in ICUa	52.0	28.6	< .0001
SC insulin received in ICUa	89.5	81.6	.0065
No. of blood glucose tests in ICU	22 (11-57)	16 (8-48)	.0045
HbA1c, %	7.2 (6.7-7.9)	6.9 (6.6-7.8)	.0007
Glycemic ratio	0.91 (0.79-1.01)	0.86 (0.76-0.97)	.0081

Abbreviations: DM, known diabetes patients; HbA_1c, glycated hemoglobin A_1c; IV, intravenous; ICU, intensive care unit; SC, subcutaneous; undiagDM, previously undiagnosed diabetes patients.

^a Percentage of patients.

Table 3 compares the 227 undiagDM with HbA_1c 6.5% to 7.9% to the 67 undiagDM with an HbA_1c greater than or equal to 8.0%. The latter group had higher mean ICU BG and CV and a higher percentage received insulin in the ICU.

Table 3.

Comparison of previously undiagnosed diabetes patients with glycated hemoglobin A_1c 6.5% to 7.9% and 8.0% or greater

	HbA1c 6.5%-7.9%	HbA1c ≥ 8.0%	P
No.	227	67
Mean blood glucose, mg/dL	134 (121-153)	153 (137-181)	< .0001
Coefficient of variation, %	20.3 (15.4-26.2)	29.5 (22.0-36.9)	< .0001
Hypoglycemia < 70 mg/dLa	17.2	20.9	.49
IV insulin received in ICUa	21.1	53.7	< .0001
SC insulin received in ICUa	79.3	89.6	.0561
No. of blood glucose tests in ICU	15 (7-46)	20 (9-54)	.24
HbA1c, %	6.7 (6.6-7.1)	9.1 (8.3-10.5)	< .0001
Glycemic ratio	0.90 (0.81-0.99)	0.72 (0.60-0.79)	< .0001
Male, %	61.7	67.1	.42
APACHE IV predicted mortality, %	29.5 (7.6)	18.0 (22.4)	< .0001
ICU LOS, d	1.9 (1.0-5.1)	2.0 (1.0-3.9)	.77
Mechanical ventilation, %	44.9	32.8	.0785
Mortality	18.1	11.9	.23
Observed:Expected mortality ratio	0.61	0.66	.46

Abbreviations: APACHE, Acute Physiology and Chronic Health Evaluation; DM, diabetes patients; HbA_1c, glycated hemoglobin A_1c, ICU, intensive care unit; IV, intravenous; LOS, length of stay; SC, subcutaneous; undiagDM, previously undiagnosed diabetes patients.

^a Percentage of patients.

Table 4 describes the relationship of hypoglycemia to mortality for undiagDM and DM. Mortality was significantly higher among patients with hypoglycemia in both cohorts. Risk-adjusted mortality was higher in both cohorts with hypoglycemia less than 55 mg/dL than with hypoglycemia 55 to 69 mg/dL.

Table 4.

Mortality and severity-associated mortality stratified by severity of hypoglycemia

	No.	Mortality %	APACHE IV predicted mortality, %	Observed:Expected mortality ratio	P Mortalitya	P Observed:Expected mortality ratioa
DM
No hypoglycemia	224	10.7	18.9	0.57
Hypoglycemia: Minimum BG, mg/dL
< 70	70	22.9	34.1	0.67	.0080	.14
55-69	48	22.9	37.4	0.61	.0222	.61
< 55	22	22.7	27.0	0.84	.0963	.0139
40-54	19	21.1	27.2	0.78
< 40	3	33.3	25.5	1.31
undiagDM
No hypoglycemia	241	12.4	19.5	0.64
Hypoglycemia: Minimum BG, mg/dL
< 70	53	35.9	40.2	0.89	< .0001	.0004
55-69	33	21.2	33.2	0.64	.17	≥ .999
< 55	20	60.0	51.8	1.16	< .0001	< .0001
40-54	15	60.0	45.8	1.31
< 40	5	60.0	70.0	0.86

Abbreviations: APACHE, Acute Physiology and Chronic Health Evaluation; BG, blood glucose; DM, diabetes patients; undiagDM, previously undiagnosed diabetes patients.

^a Hypoglycemia vs no hypoglycemia.

Finally, Figs. 2A and 2B illustrate the association of GR, the quotient of mean ICU BG and estimated preadmission BG, with mortality, and the risk-adjusted mortality (represented by OEMR). Absolute as well as risk-adjusted mortality were similar for the 2 groups in the broad range of GR 0.7 to less than or equal to 1.1. However, for patients with GR greater than or equal to 1.1, mortality (38.5% vs 10.3% [P = .0072]) and risk-adjusted mortality (OEMR 1.18 vs 0.52 [P < .0001]) were considerably higher in undiagDM than in DM, though there were no statistically significant differences in mean BG, CV, or hypoglycemia between the 2 groups (Table 5).

Figure 2.

A, Comparison of mortality of DM (known diabetes patients) and undiagDM (previously undiagnosed diabetes patients) stratified by glycemic ratio, the quotient of mean intensive care unit (ICU) blood glucose (BG) and estimated preadmission BG. B, Comparison of the observed:expected mortality ratio, the quotient of mortality, and predicted mortality using the mean APACHE IV prediction of mortality for each patient, of DM and undiagDM stratified by glycemic ratio, the quotient of mean ICU BG, and estimated preadmission BG.

Table 5.

Comparison of known diabetes patients and previously undiagnosed diabetes patients stratified by bands of glycemic ratio

GR
	GR<0.7
	DM	undiagDM	P
No.	33	44
Age, y	62 (54-78)	64 (55-74)	.67
Male, %	60.6	63.6	.79
BMI	28.4 (23.7-33.5)	27.6 (23.7-34.4)	.84
eGFR, % of patients
> 60	48.5	61.4	.26
40-59	9.1	20.5	.18
20-39	21.2	13.6	.38
< 20	21.2	4.5	.0248
Comorbidities, % of patients
Current smoker	3.0	18.2	.0412
Past smoker	42.2	25.0	.11
Hypertension	84.8	61.4	.0254
Hyperlipidemia	81.8	45.5	.0013
CAD	39.4	31.8	.49
Cerebrovascular disease	33.3	20.5	.21
Peripheral vascular disease	24.2	13.6	.24
Preadmission meds, %
Statins	66.7	52.3	.21
ARB/ARB	66.7	22.7	.0001
Glucose metrics
Mean BG, mg/dL	141 (111-163)	133 (115-147)	.23
Coefficient of variation, %	31.4 (23.1-37.2)	27.7 (20.9-35.6)	.34
Hypoglycemia < 70 mg/dLa	33.3	40.9	.50
SC insulin received in ICUa	72.2	72.7	≥ .999
IV insulin received in ICUa	57.6	45.4	.29
HbA1c, %	10.5 (7.7-12.1)	8.9 (7.9-10.9)	.14
GR	0.58 (0.54-0.64)	0.61 (0.54-0.66)	.29
Clinical
Mechanical ventilation, %	42.4	38.6	.74
ICU LOS, d	2.6 (1.2-5.8)	1.6 (0.9-4.0)	.18
APACHE IV predicted mortality, %	19.0 (21.9)	22.9 (28.0)	.51
Mortality, %	15.2	18.2	.73
Observed:Expected mortality ratio	0.80	0.79	.91
	GR 0.7-<0.9
	DM	undiagDM	P
No.	108	127
Age, y	71 (61-82)	69 (57-80)	.12
Male, %	48.1	59.1	.0924
BMI	29.9 (25.3-34.4)	28.4 (25.0-34.6)	.39
eGFR, % of patients
> 60	48.1	59.4	.0838
40-59	19.4	16.4	.55
20-39	15.7	17.2	.76
< 20	16.7	7.0	.0203
Comorbidities (% of patients)
Current smoker	11.1	7.1	.29
Past smoker	31.5	37.8	.31
Hypertension	83.3	71.7	.0355
Hyperlipidemia	83.3	73.2	.0637
CAD	64.8	30.7	< .0001
Cerebrovascular disease	25.9	23.6	.68
Peripheral vascular disease	29.6	11.8	.0007
Preadmission meds, %
Statins	72.2	49.6	.0004
ARB/ARB	49.1	30.7	.0041
Glucose metrics
Mean BG, mg/dL	138 (123-154)	126 (118-143)	.0042
Coefficient of variation, %	25.7 (20.0-31.9)	20.3 (15.5-28.9)	.0002
Hypoglycemia < 70 mg/dLa	30.6	16.5	.0106
SC insulin received in ICUa	84.3	76.4	.13
IV insulin received in ICUa	43.5	21.3	.0003
HbA1c, %	7.2 (6.8-8.1)	6.9 (6.6-7.8)	.0026
GR	0.81 (0.76-0.86)	0.81 (0.76-0.85)	.73
Clinical
Mechanical ventilation, %	48.2	49.5	.79
ICU LOS, d	1.8 (1.1-3.9)	2.1 (1.3-4.1)	.40
APACHE IV predicted mortality, %	21.9 (24.3)	22.6 (23.6)	.84
Mortality, %	13.9	14.2	.95
Observed:Expected mortality ratio	0.63	0.63	≥ .999
	GR 0.9-<1.1
	DM	undiagDM	P
No.	114	97
Age, y	75 (62-82)	68 (58-79)	.16
Male, %	56.1	66.0	.14
BMI	29.4 (24.5-34.4)	28.9 (24.9-32.4)	.76
eGFR, % of patients
> 60	45.2	60.2	.0301
40-59	24.3	19.4	.39
20-39	14.8	12.2	.58
< 20	15.7	8.8	.13
Comorbidities, % of patients
Current smoker	7.9	14.4	.13
Past smoker	41.2	33.0	.22
Hypertension	83.3	63.9	.0013
Hyperlipidemia	74.6	49.4	.0012
CAD	50.9	34.0	.0137
Cerebrovascular disease	25.4	28.9	.57
Peripheral vascular disease	29.8	14.4	.0079
Preadmission meds, %
Statins	64.0	53.6	.13
ARB/ARB	50.0	30.9	.0051
Glucose metrics
Mean BG, mg/dL	155 (144-169)	148 (136-158)	.0005
Coefficient of variation, %	27.5 (20.8-35.0)	21.1 (16.1-26.9)	< .0001
Hypoglycemia < 70 mg/dLa	21.1	12.4	.0952
SC insulin received in ICUa	99.1	91.8	.50
IV insulin received in ICUa	56.1	25.8	< .0001
HbA1c, %	7.0 (6.7-7.5)	6.7 (6.5-7.2)	.0003
GR	0.98 (0.93-1.02)	0.97 (0.94-1.01)	.61
Clinical
Mechanical ventilation, %	42.1	47.4	.44
ICU LOS, d	2.5 (1.2-4.3)	1.9 (1.0-5.6)	.60
APACHE IV predicted mortality, %	25.0 (25.1)	21.7 (24.1)	.32
Mortality, %	14.0	13.4	.90
Observed:Expected mortality ratio	0.56	0.62	.38
	GR ≥ 1.1
	DM	undiagDM	P
No.	39	26
Age, y	75 (65-82)	71 (61-85)	.76
Male, %	56.4	69.2	.30
BMI	28.(26.2-32.4)	25.4 (24.4-35.2)	.23
eGFR, % of patients
> 60	46.2	57.7	.37
40-59	25.6	26.9	.91
20-39	7.7	0.0	.15
< 20	20.5	15.4	.61
Comorbidities, % of patients
Current smoker	5.1	7.7	.67
Past smoker	56.4	34.6	.0872
Hypertension	82.1	76.9	.61
Hyperlipidemia	72.0	65.4	.57
CAD	48.2	42.3	.64
Cerebrovascular disease	15.4	23.1	.44
Peripheral vascular disease	30.8	7.7	.0277
Preadmission meds, %
Statins	72.0	57.7	.24
ARB/ARB	43.6	34.6	.47
Glucose metrics
Mean BG, mg/dL	181 (170-211)	170 (160-187)	.11
Coefficient of variation, %	28.2 (21.7-38.6)	24.7 (18.1-34.1)	.14
Hypoglycemia < 70 mg/dLa	5.1	7.7	.67
SC insulin received in ICUa	89.7	84.6	.54
IV insulin received in ICUa	59.0	46.2	.31
HbA1c, %	7.1 (6.6-7.3)	6.7 (6.6-6.9)	.13
GR	1.21 (1.14-1.33)	1.19 (1.12-1.27)	.49
Clinical
Mechanical ventilation, %	19.6	22.2
ICU LOS, d	1.2 (0.8 = 2.5)	1.5 (1.1-2.8)	.39
APACHE IV predicted mortality, %	19.9 (23.5)	32.5 (35.2)	.0867
Mortality, %	10.3	38.5	.0072
Observed:Expected mortality ratio	0.52	1.18	< .0001

Abbreviations: ACE/ARB, angiotensin-converting enzyme/angiotensin receptor blockade; APACHE, Acute Physiology and Chronic Health Evaluation; BG, blood glucose; BMI, body mass index; CAD, coronary artery disease; DM, diabetes patients; eGFR, estimated glomerular filtration rate; GR, glycemic ration; HbA_1c, glycated hemoglobin A_1c; ICU, intensive care unit; IV, intravenous; LOS, length of stay; SC, subcutaneous; undiagDM, previously undiagnosed diabetes patients.

^a Percentage of patients.

Discussion

To our knowledge, this is the first investigation that explores glucose metrics and their relationship to mortality among a cohort of critically ill patients with previously undiagnosed DM (undiagDM). The salient findings of this case-control study include the following:

• Risk-adjusted mortality was higher in undiagDM than in DM, attributable to a markedly higher mortality among patients with GR (the quotient of mean ICU BG and estimated preadmission BG, using the Nathan formula [28]) greater than or equal to 1.1 in this group.

• UndiagDM had a significantly lower burden of comorbidities, including hypertension, hyperlipidemia, coronary artery disease, peripheral vascular disease, and renal insufficiency, at the time of admission than did DM.

• UndiagDM had lower mean BG, lower CV, and a trend toward lower percentage of patients with hypoglycemia less than 70 mg/dL than did DM.

• Hypoglycemia less than 70 mg/dL was associated with higher risk-adjusted mortality in undiagDM but not in DM. Hypoglycemia less than 55 mg/dL was associated with higher risk-adjusted mortality in both cohorts.

Relationship to Prior Literature

Prevalence of previously undiagnosed diabetes

Several prior studies have identified undiagDM in hospitalized patients using HbA_1c obtained at admission, or follow-up glucose tolerance tests [19–24], with percentages ranging from 9.3% [20] to 24% [21]. In contrast, the present investigation identified 5.3% with undiag DM. It is possible that the lower prevalence of undiag DM in our cohort is due to a high intensity of outpatient care in the community.

Glucose metrics and mortality of previously undiagnosed diabetes

In a single-center cohort of 1886 non-ICU patients, Umpierrez and colleagues [29] identified 26% with known DM and 12% with no prior diagnosis of DM but with a fasting BG greater than 126 mg/dL or 2 random values greater than 200 mg/dL; this group of patients sustained markedly higher mortality (16.0%) than did the groups with no diabetes (1.7%). A multicenter ICU study found that undiagDM had higher mean ICU BG, CV, hypoglycemia, and mortality than did patients without DM [21]. In addition, this study stratified the known DM cohort based on HbA_1c less than or greater than or equal to 6.5%. The undiagDM cohort had values of mean ICU BG, CV, and hypoglycemia intermediate between DM patients with HbA_1c less than 6.5% and those with HbA_1c greater than or equal to 6.5% in this study.

Biologic Plausibility

In this case-control study, higher mean ICU BG in the undiagDM cohort was associated with higher mortality, whereas the opposite was seen with DM patients. Here, it is possible that this finding reflects the fact that patients previously diagnosed with diabetes were exposed to higher circulating insulin levels before admission, owing to treatment either with insulin itself (48.3%) or with drugs such as sulfonylureas (21.5%) that increase insulin secretion. At issue is whether such treatments induce stable physiological adaptations that are protective against hyperglycemia in the ICU.

The Diabetes Control and Complications Trial (DCCT) (1982-1993) compared intensive insulin therapy (INT) aimed at achieving levels of glycemia close to the nondiabetic range with conventional insulin therapy (CON) [30]. Following these findings, the Epidemiology of Diabetes Interventions and Complications (EDIC) study was initiated to examine the long-term effects of the original DCCT interventions on diabetic complications [31]. During EDIC, both INT and CON groups had identical HbA_1c levels of approximately 8% at follow-up. Even though the HbA_1c level was now comparably elevated between groups, the beneficial effects of previous lower HbA_1c in the INT group persisted as if there had been no deterioration in their HbA_1c. This outcome was shown to reflect stably altered DNA methylation patterns that occurred during the DCCT [32].

The markedly higher mortality risk seen in undiagDM patients with high mean ICU BG could also involve other factors that raise sensitivity to ICU hyperglycemia. For example, the lower preadmission use of statins in undiagDM group could have contributed, since this has been associated with reduced ICU mortality [33]. It is also possible that patients in the undiagDM group were less likely to have received regular medical attention or were burdened with unrecognized disease contributing to impaired response to physiologic stress.

Strengths and Limitations

Strengths of this investigation include its rich data set, including comorbidities, severity of illness, and detailed glucose metrics. The database was maintained and updated prospectively, and the accuracy of the designation of DM and undiagDM was validated by detailed review of individual patient records. In addition, the case-control methodology, with 100% matching of insulin treatment protocol eras, medical vs surgical admission to the ICU, and stratification of HbA_1c (6.5%-7.9% and ≥ 8.0%), allowed a more rigorous comparison of the 2 groups. We acknowledge that the use of HbA_1c in determining DM status is subject to the known limitations of this metric [34]. Furthermore, while the ICU admits patients with a broad array of medical and surgical diagnoses, the generalizability of the investigation is limited by its single-center status. Finally, as is true of all observational data, our conclusions must be considered hypothesis generating, rather than as proof of causality.

Clinical Implications

These data shed light on the natural history of DM, glucose metrics during critical illness in these patients, and the relationship of ICU glycemia to mortality.

The undiagDM cohort may have sustained a shorter duration of DM before ICU admission compared to the matched DM group, as demonstrated by their significantly lower burden of important comorbidities. Therefore, this group of patients can be placed on the spectrum of impaired glucose regulation between patients without DM and those with established disease.

This inference sheds light on the observed differences in glucose control. While both groups were treated with the same ICU BG management protocols, undiagDM had lower mean BG, lower CV, and less frequent hypoglycemia than did DM patients. Previous literature has documented similar differences in glucose metrics when comparing cohorts of ICU patients with and without DM, as well as the relationship of higher HbA_1c with greater derangements in ICU glucose control [9–18].

Finally, our data demonstrate important differences in the relationship of ICU glycemia to mortality for the 2 groups. We used the GR, the quotient of mean ICU BG and estimated preadmission glycemia [18], to reflect the degree of divergence of ICU glycemia from baseline glycemia. The undiagDM cohort had higher risk-adjusted mortality than did the DM cohort. This is attributable to the much higher mortality associated with GR greater than or equal to 1.1 in undiagDM than in DM; for GR values less than 1.1, both absolute and risk-adjusted mortality were similar (see Fig. 2A and 2B, respectively) comparing DM and undiagDM.

Elevated GR reflects hyperglycemia relative to preadmission glycemia. The undiagDM, therefore, demonstrated the same association of elevated mean ICU BG with mortality that has been documented in numerous observational studies of patients without DM [3–8]. In contrast, both absolute and risk-adjusted mortality were numerically lower in the DM cohort with relative hyperglycemia (GR ≥ 1.1). These data suggest that the tolerance to hyperglycemia among patients with established DM—the attenuation of an association of hyperglycemia with mortality in the critically ill—as described in prior observational studies [3–8], had not yet developed in this cohort of undiagDM.

In summary, glucose control in the critically ill is abnormal in undiagDM but not as abnormal as in DM and, furthermore, undiagDM retain sensitivity to the deleterious effect of hyperglycemia, an association not seen in DM. These data place the undiagDM cohort on a spectrum between patients without DM and those with established DM. These data strengthen the association of hyperglycemia with mortality in patients without DM and the relative tolerance to hyperglycemia among those with established DM, especially those with high HbA_1c. Finally, these data provide impetus to the need to perform well-designed randomized trials to test the paradigm of individualized glucose control in the critically ill.

Abbreviations

APIV PM

Acute Physiology and Chronic Health Evaluation IV predicted mortality

BG

blood glucose

CON

conventional insulin therapy

CV

coefficient of variation

DCCT

Diabetes Control and Complications Trial

DM

diabetes mellitus

EDIC

Epidemiology of Diabetes Interventions and Complications

HbA_1c

glycated hemoglobin A_1c

GR

glycemic ratio

ICU

intensive care unit

INS

insulin

INT

intensive insulin therapy

IQR

interquartile range

IV

intravenous

OEMR

observed:expected mortality ratio

undiagDM

previously undiagnosed diabetes mellitus

Financial Support

The authors received no financial support for the research, authorship, and/or publication of this article.

Author Contributions

J.K. wrote the first draft of the manuscript and was responsible for the designation of the methodology. All authors were responsible for the critical revision of the article for important intellectual content. J.K., Y.W., and J.B. were responsible for the collection and assembly of data. J.K. was responsible for data management. J.K. completed the main part of the data analyses, and all authors discussed the analysis plan and results and provided input to the manuscript. All authors had access to the final study results and were responsible for the final approval of the manuscript.

Disclosures

J.K. is a consultant for Dexcom. G.R. has no relevant disclosures. J.C.P. has no relevant disclosures. M.B. has no relevant disclosures. M.S. has received research support from Novo Nordisk. P.R. has no relevant disclosures. Y.W. has no relevant disclosures. J.B. has no relevant disclosures. G.U. is partly supported by research grants from the National Institutes of Health (NIH/NATS UL 3UL1TR002378-05S2) from the Clinical and Translational Science Award program, and from the National Institutes of Health and National Center for Research Resources (NIH/NIDDK 2P30DK111024-06). He has also received research support (to Emory University) from Dexcom, Bayer, Abbott, and Astra Zeneca. I.H. has research grants or contracts with Dexcom, Beta Bionics, Omnipod, and Medtronic Diabetes, and receives consulting fees from Abbott Diabetes Care, Roche, Lifescan, and GWave.

Data Availability

The data that support the findings of this study are not publicly available. However, the corresponding author will on request detail the restrictions and any conditions under which access to some data may be provided.

Clinical Trials Information

ClinicalTrials.gov registration number NCT05256043 (registered February 25, 2022).