Association between red blood cell distribution width and mortality in diabetic ketoacidosis.

Introduction

Over the past decade, the prevalence of diabetes mellitus (DM) in human populations has increased significantly due to increased intake of high-fat foods and reduced physical activity.[1] Diabetic ketoacidosis (DKA) is a life-threatening acute complication of DM that is characterized by a pathological imbalance between severe insulin deficiency and excessive glucagon, resulting in clinical changes such as hyperglycemia, ketosis, electrolyte imbalance, and metabolic acidosis.[2,3] More than 100,000 DKA patients are admitted to American hospitals each year.[3] Despite implementing standardized protocols for maintaining electrolyte balance and insulin replacement, DKA remains a serious and inevitable complication and a leading cause of death in DM patients, which leads to significant treatment costs.[3-5] Sometimes DKA patients will enter the ICU because hospital-specific policies prevent the use of intravenous insulin infusions on general wards, which causes resource shortages for critically ill patients.[6] Therefore, evaluating the risk factors of DKA is crucial for clinicians with regard to patient outcomes and resource use.

RDW is a laboratory index that measures the size variation of circulating red blood cells (RBCs).[7] Previously, the clinical use of RDW was limited to the differential diagnosis of anemia.[8] Recently, several clinical studies have demonstrated that RDW is an independent prognostic marker for mortality,[9] especially in patients with cardiovascular disease (e.g., heart failure,[10] coronary artery diseases,[11] or acute coronary syndrome[12]) and over the long-term in patients with community-acquired pneumonia.[13] Other studies have further demonstrated that RDW is a novel inflammatory marker that plays an important role in the development of inflammation.[14-16] As a systemic metabolic disease, DKA has also been reported to be associated with the inflammatory and immune responses of the hyperglycemic state.[16,17] Inflammation is of vital significance in the generation and development of DKA. Based on the above findings, we hypothesized that RDW might have a relationship with DKA. However, to our knowledge, no study has directly explored the association between RDW and DKA outcomes. Therefore, this study was designed to investigate whether RDW could independently predict the prognosis of DKA patients in the ICU.

Materials and method

Study cohort

This was a longitudinal, single-center, retrospective cohort study that used multi-parameter intelligent monitoring III version 1.3 (MIMIC-III v1.3) in the ICU, which is a public and freely available ICU database software.[18] The database includes more than 38,000 ICU patients admitted to Beth Israel Deaconess Medical Center from 2001 to 2012.[18,19] MIMIC-III also includes demographics, laboratory results, records of care progress, intravenous medications, fluid balance, and other clinical variables.[18] To protect privacy, information regarding the included patients were hidden.

Population selection criteria

All adult patients (age: ≥18 years) who were initially diagnosed with DKA in the ICU were searched in the database. Cases were defined as adult critically ill patients diagnosed with DKA.[18,20] Patients were excluded according to the following criteria: (1) no RDW measurement during the ICU stay; (2) hematologic disease (e.g., leukemia or myelodysplastic syndrome); (3) ICU stay <48 hours; and (4) missing >5% of individual data. The primary outcome of the study was 365-day mortality from the date of ICU admission; 90-day mortality after ICU admission was the secondary outcome.

Data sources

The PostgreSQL tool (version 9.6) was used to extract data. All clinical parameters were recorded in the first 24 hours after patient admission.[21] Physiologic information including heart rate, mean blood pressure, systolic blood pressure (SBP), diastolic blood pressure (DBP), temperature, and respiratory rate were measured by bedside monitors. Laboratory values included bicarbonate, platelet, blood urea nitrogen (BUN), glucose, sodium, creatinine, chloride, hematocrit, hemoglobin, mean corpuscular volume (MCV), mean hemoglobin concentration (MHC), mean corpuscular hemoglobin concentration (MCHC), and white blood cell (WBC). Comorbidities included congestive heart failure (CHF), atrial fibrillation (AF), chronic renal disease (CKD), chronic liver disease, coronary artery disease (CAD), acute respiratory distress syndrome (ARDS), and malignancy. We also calculated Elixhauser Comorbidity Index, acute physiology score III (APSIII), and systemic inflammatory response syndrome (SIRS).

Statistical analysis

All data were analyzed with R software version 3.42. All statistical analyses were two-sided and p < 0.05 was interpreted as statistically significant. Continuous variables were first examined for normality. Normal data are expressed as mean ± standard deviation (SD), while categorical data are summarized as number or percentage. To compare the groups, we used the chi-square test for categorical variables and analysis of variance, and the Kruskal–Wallis test for continuous variables.[22] Factors related to RDW values were determined by multivariate linear regression. We then used Cox proportional hazards models to determine whether RDW was independently associated with all-cause mortality, from which the hazard ratio (HR) and 95% confidence interval (CI) were calculated.

We ran two models for each endpoint. Covariates were adjusted for age, gender, and ethnicity in model I, while in model II, covariates were further adjusted for APSIII, SIRS, CHF, stroke, ARDS, SBP, MCV, MHC, and MCHC. All selected variables were based on associations with the outcomes or a change in effect estimate of more than 10%. Potential multicollinearity between covariates was quantified by calculating the variance inflation factor (VIF),[23] which provided an index of how much the variance of an estimated regression coefficient was increased due to collinearity. A VIF value >5 was considered evidence of multicollinearity.

Stratification analyses were performed to investigate whether the effect of RDW differed across various subgroups, including CHF, AF, CKD, CAD, vasoactive drugs, and RRT.

Results

Subject characteristics

Among the 38,597 reviewed patients, 495 were eligible for this analysis. Patient characteristics are summarized in Table 1, and patients were divided into survival and mortality groups. In total, 440 patients had a survival outcome, while 55 had a fatal outcome. Among the cohort, 239 patients (48.3%) were men, 256 (51.7%) were white, the mean age was 47.5 ± 17.8 years, and the average RDW was 13.9% ± 1.6%. Patients with high mortality were generally older and had a history of CHF, AF, CAD, and malignancy. Additionally, this group had higher creatinine, glucose, sodium, platelets, and BUN.

Table 1.

Characteristics of the study participants according to their 365-day survival outcomes.

Clinical variable	365-mortality			P value
	Overall	Survival	Mortality
N	495	440	55
Age, years	47.5 ± 17.8	45.4 ± 16.9	64.0 ± 16.3	<0.001
Gender, n (%)				<0.001
Female	256 (51.7)	231 (52.5)	25 (45.5)
Male	239 (48.3)	209 (47.5)	30 (54.5)
Ethnicity, n (%)				0.378
White	306 (61.8)	271 (61.6)	35 (63.6)
Black	103 (20.8)	95 (21.6)	8 (14.5)
Other	86 (17.4)	74 (16.8)	12 (21.8)
RDW, %	13.9 ± 1.6	13.8 ± 1.3	15.0 ± 2.7	<0.001
SBP, mmHg	122.6 ± 17.2	122.4 ± 16.6	124.1 ± 21.6	0.700
DBP, mmHg	63.2 ± 10.4	63.5 ± 10.2	60.5 ± 12.0	0.095
Respiratory rate, beats/minute	19.1 ± 4.0	18.9 ± 3.8	20.6 ± 5.0	0.015
Temperature, °C	36.9 ± 0.5	36.9 ± 0.5	36.8 ± 0.9	0.309
Comorbidities, n (%)
Congestive heart failure	32 (6.5)	24 (5.5)	8 (14.5)	0.017
Atrial fibrillation	21 (4.2)	14 (3.2)	7 (12.7)	0.005
Coronary artery disease	68 (13.7)	55 (12.5)	13 (23.6)	0.035
Chronic kidney disease	56 (11.3)	47 (10.7)	9 (16.4)	0.255
Chronic liver disease	23 (4.6)	20 (4.5)	3 (5.5)	0.733
Malignancy	27 (5.5)	15 (3.4)	12 (21.8)	<0.001
Elixhauser Comorbidity Index	8.9 ± 9.8	8.7 ± 9.7	14.2 ± 11.1	0.042
Laboratory parameters
Bicarbonate, mmol/L	14.1 ± 5.8	14.0 ± 5.8	15.9 ± 6.8	0.234
Creatinine, mEq/L	1.3 ± 1.4	1.3 ± 1.4	2.1 ± 1.7	<0.001
Chloride, mmol/L	97.2 ± 9.1	97.2 ± 9.1	100.4 ± 8.8	0.188
Glucose, mg/dL	122.2 ± 53.4	121.0 ± 52.5	158.1 ± 67.4	0.017
Hematocrit, %	32.4 ± 5.7	32.5 ± 5.6	30.5 ± 7.9	0.441
Hemoglobin, g/dL 12–16	11.1 ± 2.0	11.1 ± 1.9	10.3 ± 2.7	0.282
MCV, fL 80–100	85.2 ± 12.2	85.5 ± 12.1	83.8 ± 18.6	0.256
MHC, pg 27–31	29.4 ± 2.6	29.6 ± 2.4	29.5 ± 3.1	0.721
MCHC, g/L 320–360	346.6 ± 22.4	348.6 ± 21.4	350.2 ± 25.2	0.291
Platelet, 109/L	245.7 ± 100.1	247.9 ± 99.3	180.2 ± 104.8	0.012
BUN, mg/dL	21.6 ± 19.7	21.0 ± 19.4	39.6 ± 21.2	<0.001
WBC, 109/L	11.0 ± 5.0	11.1 ± 5.0	10.9 ± 6.2	0.706
RRT, n (%)	25 (5.1)	16 (3.6)	9 (16.4)	<0.001

SBP: systolic blood pressure; DBP: diastolic blood pressure; MCV: Mean corpuscular volume; MHC: Mean hemoglobin concentration; MCHC: Mean corpuscular hemoglobin concentration; WBC: white blood cell; BUN: blood urea nitrogen; RRT: Renal replacement therapy.

Significance level p < 0.05.

Association between RDW and mortality

In the univariable Cox regression model for 365-day all-cause mortality, RDW was a predictor of all-cause mortality in DKA patients in the ICU (HR: 1.30, 95% CI: 1.19–1.43) (Table 2). For 365-day all-cause mortality adjusted for age gender, and ethnicity (model I), RDW was a particularly strong predictor of all-cause mortality in DKA patients in the ICU (HR: 1.27, 95% CI: 1.14–1.42, P < 0.001). When adjusted for age, gender, ethnicity, APSIII, SIRS, CHF, stroke, ARDS, SBP, MCV, MHC, and MCHC (model II), the predictive role of RDW was still statistically significant (HR: 1.23, 95% CI: 1.05–1.45, P < 0.05). For 90-day all-cause mortality adjusted for the same confounders (model II), the same relationship was observed (HR: 1.29, 95% CI: 1.02–1.65, P < 0.05).

Table 2.

Hazard ratios and 95% confidence intervals for mortality across groups.

Outcome	Non-adjusted		Model Ia		Model IIb
	HR (95% CIs)	Pvalue	HR (95% CIs)	Pvalue	HR (95% CIs)	Pvalue
90-day all-cause mortality	1.25 (1.08–1.44)	0.0027	1.19 (1.01–1.41)	0.0498	1.29 (1.02–1.65)	0.0372
365-day all-cause mortality	1.30 (1.19–1.43)	<0.0001	1.27 (1.14–1.42)	<0.0001	1.23 (1.05–1.45)	0.0089

HR: hazard ratio; CI: confidence interval.

aModel I covariates were adjusted for age, gender and ethnicity.

bModel II covariates were adjusted for age; gender; ethnicity; APSIII; SIRS, CHF; stroke; ARDS; SBP; MCV; MHC and MCHC.

Significance level p < 0.05.

Subgroup analyses

We conducted subgroup analyses to determine the consistency of associations between RDW and risk of hospital mortality in critically ill DKA patients (Table 3). This analysis showed that CKD patients with higher RDW had a significantly higher risk of hospital mortality (HR: 1.35, 1.05–1.74, P < 0.05).

Table 3.

Subgroup analysis of associations between RDW and hospital mortality.

	No. of patients	HR (95% CI)	P value
Congestive heart failure
No	463	1.11 (0.98–1.25)	0.1066
Yes	32	1.11 (0.72–1.71)	0.6286
Atrial fibrillation
No	474	1.05 (0.90–1.23)	0.5128
Yes	21	1.10 (0.83–1.45)	0.5013
Chronic kidney disease
No	439	1.07 (0.93–1.24)	0.3552
Yes	56	1.35 (1.05–1.74)	0.0198
Chronic liver disease
No	472	1.12 (1.00–1.26)	0.0504
Yes	23	0.43 (0.07–2.46)	0.3412
Malignancy
No	468	1.25 (1.06–1.47)	0.0075
Yes	27	1.17 (0.95–1.43)	0.1363
Coronary artery disease
No	427	1.14 (1.02–1.28)	0.0228
Yes	68	1.06 (0.71–1.58)	0.7700
Vasoactive drug
No	436	1.16 (1.01–1.33)	0.0371
Yes	59	1.06 (0.89–1.27)	0.4982
RRT
No	470	1.09 (0.94–1.26)	0.2404
Yes	25	1.01 (0.80–1.28)	0.9214

SBP: systolic blood pressure; DBP: diastolic blood pressure; WBC: white blood cell; BUN: blood urea nitrogen; RRT: Renal replacement therapy.

Significance level p < 0.05.

Discussion

The main finding of this study was that RDW was independently associated with 90-day and 365-day all-cause mortality in DKA patients post ICU admission. The association remained significant after adjusting for age, gender, ethnicity, APSIII, SIRS, CHF, stroke, ARDS, SBP, MCV, MHC, and MCHC. While previous studies have shown that RDW is independently associated with several adverse outcomes,[10,11,24] our study indicated that RDW was an independent predictor of mortality in critically ill DKA patients.

To our knowledge, this is the first study to assess RDW as a long-term prognostic marker of DKA patients admitted to the ICU. The precise mechanism underlying the relationship between RDW and all-cause mortality in DKA is unclear. However, some hypotheses have been proposed, among which inflammatory responses[14,15] and oxidative stress[25-27] are the most popular.

Several observational studies have described an association between elevated RDW and changes in inflammatory biomarkers.[14] The proliferation and maturation of erythrocytes were inhibited by proinflammatory cytokines.[14,15] Increased RDW may reflect the extent of inflammation, which negatively impacts patient survival.[9,28,29] There has also been a reported association between higher RDW and increased levels of the inflammatory marker C-reactive protein. It has also been suggested[9,30,31] that RDW reflects the variability in the size of circulating RBCs. The increase in RDW results from the release of reticulocytes into the circulation. The increased number of smaller erythrocytes may reflect an increase in vesicle shedding by RBCs.[32,33] Additionally, RDW is associated with endothelial function, is primarily assessed by flow-mediated dilation and can be used as a measure of potential metabolic disorders.[34]

The relationship between RDW and mortality in this study may also be related to oxidative stress. High oxidative stress may result in increased RDW, which may be due to reduced RBC survival and increased release of large and small premature erythrocytes into peripheral circulation (anisocytosis).[27,35] This is due to the release of heterogeneous RBCs with poor oxygen carrying capacity into the peripheral circulation, which can damage the microcirculation of local tissue via hypoxia.[25,26] The increased RDW may promote the formation of thrombus, which may be released into the circulation through immature reticulocytes, but the role of immature reticulocytes in thrombosis is unclear. The neurohumoral response to arterial underfilling may also contribute to this association.[36-38]

The main advantage of this study is that it is the first comprehensive assessment of the relationship between RDW and mortality in DKA patients in the ICU. RDW appears to be a long-term prognostic indicator for patients admitted to the ICU for DKA. Additionally, RDW is widely available and adds no additional cost to the patient. Moreover, the number of patients participating in our study was large.

We acknowledge that this study also had some limitations. First, our study was a retrospective observational study with inherent biases. For this reason, selection bias cannot be ignored. Second, the study did not investigate potential changes in RDW over time, which may provide additional prognostic information. Third, the groups differed in the number of patients with malignancy, but our study did not assess whether malignancy affected survival differences in the two groups. Finally, the nutritional status of the patient was unknown, and nutritional deficiency, which can lead to elevated RDW, may be a confounding variable.

Conclusions

This study showed that RDW is a long-term prognostic marker for patients admitted to the ICU for DKA, and that higher RDW is associated with an increased risk of mortality in these patients. However, to confirm the relationship between RDW and poor prognosis in DKA, further studies, especially large-scale prospective studies, are needed.

Ethics and consent statements

Because this was a retrospective analysis of a large clinical database, ethics and consent statements are inapplicable for our study. All data in our study were extracted from a freely accessible database, the Medical Information Mart for Intensive Care Database III (MIMIC-III). The setting and use of this database were approved by the institutional review boards of the Massachusetts Institute of Technology (Boston, MA) and Beth Israel Deaconess Medical Center (Cambridge, MA). All personal information included in the database have been de-identified to safeguard privacy. Data are available from MIMIC-III for researchers who have completed the National Institutes of Health’s online course for Protecting Human Research Participants.