Abstract
Background
The relationship between baseline hemoglobin levels and in-hospital mortality in septic patients remains unclear. This study aimed to clarify this association in critically ill patients with sepsis.
Methods
Patients with sepsis were retrospectively identified from the Medical Information Mart for Intensive Care-IV (MIMIC-IV 2.2) and eICU Collaborative Research Database (eICU-CRD). Multivariate logistic regression analysis and restricted cubic spline regression were used to investigate the association between hemoglobin and the risk of in-hospital mortality. Additionally, a two-part linear regression model was used to determine threshold effects. Stratified analyses were also performed.
Results
A total of 21,946 patients from MIMIC-IV and 15,495 patients from eICU-CRD were included in the study. In-hospital mortality was 14.95% in MIMIC-IV and 17.40% in eICU-CRD. Multivariate logistic regression showed that hemoglobin was significantly and nonlinearly associated with the risk of in-hospital mortality after adjusting for other covariates. Furthermore, we found a nonlinear association between hemoglobin and in-hospital mortality, with mortality plateauing at 10.2 g/dL. The risk of mortality decreased with increasing hemoglobin levels below 10.2 g/dL but increased when hemoglobin levels exceeded 10.2 g/dL. These findings were validated in the eICU-CRD dataset.
Conclusions
A nonlinear correlation between hemoglobin levels and in-hospital mortality was observed in patients with sepsis, with a threshold of 10.2 g/DL. These findings suggested that hemoglobin levels below or above the threshold may be associated with worse outcomes, warranting further investigation in prospective studies.
Supplementary Information
The online version contains supplementary material available at 10.1186/s12879-024-10335-x.
Keywords: Hemoglobin, In-hospital mortality, Sepsis, Intensive care unit, Anemia
Background
Sepsis is a life-threatening, complex clinical syndrome characterized by organ dysfunction caused by a dysregulated host response to infection [1]. Sepsis remains a significant cause of global mortality, estimated to affect approximately 49 million people annually [2] with a 90-day mortality rate of 35.5% [3]. It means that raising awareness of sepsis and further identifying risk factors associated with sepsis outcomes are important for developing early prevention strategies.
Hemoglobin, a critical factor in oxygen delivery and tissue perfusion, plays a central role in the management of critically ill patients, including those with sepsis [4, 5]. There are many reasons for fluctuating hemoglobin levels in septic patients, including nutritional deficiencies, medical anemia, drug reactions, diminished bone marrow response to erythropoietin, inflammatory anemia, defective erythropoiesis, and hemolysis due to disseminated intravascular coagulation [6–11]. Sepsis-induced anemia and altered oxygen consumption may exacerbate microcirculation and increased tissue hypoxia. Hemoglobin also plays a central role in defending against microorganisms and assisting leukocytes. It promotes non-specific immunity by chelating iron, which deprives bacterial nutrition, maintains hemodynamic stability, and assists in the absorption and transit of antibiotics [6]. Given these critical functions, maintaining hemoglobin levels within an appropriate range is essential for critically ill septic patients. However, the relationship between baseline hemoglobin levels and in-hospital mortality in septic patients remains poorly understood. Previous studies have primarily focused on transfusion thresholds rather than exploring the prognostic implications of baseline hemoglobin levels in sepsis. This study specifically investigates the association between baseline hemoglobin levels and the risk of in-hospital mortality in septic patients, focusing on potential threshold effects and nonlinear trends. By analyzing data from two large, multi-center databases, we hypothesize that hemoglobin levels have a nonlinear association with in-hospital mortality, with both low and high levels associated with worse outcomes. This study aims to clarify the prognostic significance of hemoglobin levels in patients with sepsis.
Methods
Study population
Data used in this study were extracted from the Medical Information Mart for Intensive Care-IV (MIMIC-IV version 2.2) and the eICU Collaborative Research Database (eICU-CRD Version 2.0) [12, 13]. The MIMIC-IV database contains patient demographics, clinical measurements, laboratory tests, treatments, pharmacotherapy, medical data, survival data, and diagnoses of patients admitted to the Beth Israel Deaconess Medical Center from 2008 to 2019. The eICU-CRD is a multi-center resource containing deidentified health data from over 200,000 ICU admissions across the United States between 2014 and 2015. Individuals who have completed the Collaborative Training Program exam (author Sheng’s certification number 48693098) can access the databases. Patient information was deidentified in both datasets. The institutional review board at the Beth Israel Deaconess Medical Center approved the study. Consequently, patient informed consent and ethics approval requirements were waived.
Critically ill adult patients who met the Sepsis 3.0 criteria were eligible [1, 14]. The exclusion criteria for this study were as follows: (1) patients with an ICU length of stay less than 24 h, (2) multiple admissions to the ICU, for whom only data from the first ICU admission were extracted, (3) patients with no hemoglobin recorded, (4) patients receiving red blood cell transfusion within 24 h of ICU admission (Fig. 1).
Fig. 1.
Inclusion and exclusion criteria. MIMIC, Medical Information Mart for Intensive Care; eICU-CRD, eICU Collaborative Research Database
Variable extraction and data collection
Data extraction was performed with PostgreSQL tools (version 16.0) through the execution of the Structured Query Language. The following data were extracted from the MIMIC-IV database and the eICU-CRD database on the first day of ICU admission: demographics, laboratory tests, vital signs, comorbidities, disease-related scores, and treatment regimens. If a variable was recorded multiple times during the 24 h of admission to the ICU, we used the value associated with the greatest severity of illness. Variables with missing values greater than 30% were excluded from the analysis. Multiple imputation was performed for variables with missing values less than 30% [15].
Before investigating the association between hemoglobin and in-hospital mortality in patients with sepsis, we first used the Boruta machine learning algorithm for feature selection to determine their importance in the analytical model [16, 17]. By estimating the distribution of the importance values of the variables and selecting predictor variables with significantly higher importance values (Fig. 2).
Fig. 2.
Feature selection for analyzing the relationship between hemoglobin and in-hospital mortality using the Boruta algorithm. The horizontal axis shows the name of each variable, while the vertical axis indicates the importance of each variable. The box plot depicts the importance of each variable in the model calculations, where green boxes represent important variables, yellow boxes represent tentative variables, and red boxes represent unimportant variables. The importance of each variable is compared to shadow variables to determine its classification
Primary outcome and secondary outcomes
The primary outcome of the study was in-hospital mortality, and the secondary endpoint was mortality within 28 days after admission to the ICU.
Statistical analysis
Continuous variables were presented as median and interquartile range (IQR). Categorical variables were presented as numbers (percentages). Comparisons between groups were calculated using the Student’s t-test or Mann–Whitney U test as appropriate for continuous variables and chi-square test or Fisher’s exact test for categorical variables. To assess the association between hemoglobin and the risk of in-hospital mortality and 28-day mortality, multivariate logistic regression analyses were performed to calculate odds ratios (OR) and 95% confidence intervals (CI). The restricted cubic spline (RCS) regression analysis was performed to explore the potential nonlinear relationship between hemoglobin and in-hospital mortality. In addition, we used a recursive algorithm to calculate the inflection point between hemoglobin and in-hospital mortality and investigated the association between hemoglobin and the risk of in-hospital mortality using two logistic regressions on either side of the inflection point. Stratified analyses were conducted based on gender, comorbidities, and treatment approaches. All analyses were carried out with R 4.3.2 (R Foundation), and a two-tailed P < 0.05 was considered statistically significant.
Results
Baseline characteristics
Based on the inclusion and exclusion criteria, a total of 21,946 patients from the MIMIC-IV and 15,495 patients from the eICU-CRD were included in this study. In the MIMIC-IV database, the in-hospital and 28-day mortality rates were 14.95% and 18.67%, respectively, while in the eICU-CRD, these rates were 17.40% and 16.59%. Baseline characteristics of the study population were shown in Table 1. The non-survivor group demonstrated more severe disease severity with higher Sequential Organ Failure Assessment (SOFA), and simplified acute physiology score II (SAPS II) scores compared with the survivor group (36 vs. 48, P < 0.001). Charlson comorbidity index (CCI) scores indicating patient comorbidities were also higher in the non-survivor group. The average value of hemoglobin was 10.10 (8.80, 11.50) in MIMIC-IV and 9.90 (8.50, 11.50) in eICU-CRD. Other demographic data, vital signs, and laboratory results were also compared between the survivor and non-survivor groups in Table 1. When dividing participants into groups based on the quartiles of the hemoglobin, patients in the lower quartiles had significantly higher SAPS II scores, and higher proportion of receiving vasoactive medication than those in the upper quartiles (P < 0.001 for all). In addition, higher in-hospital mortality and 28-day mortality were observed in the lower quartiles. Furthermore, hospital stay time and ICU stay time were much longer in the first quartile of hemoglobin (Table 2 and Supplementary Table 1, Additional File 1).
Table 1.
Baseline characteristics of the survivor and non-survivor groups in MIMIC-IV and eICU-CRD databases
| Variables | MIMIC-IV | eICU-CRD | ||||||
|---|---|---|---|---|---|---|---|---|
| Overall | Survivor | Non-survivor | P-value | Overall | Survivor | Non-survivor | P-value | |
| (n = 21946) | (n = 18664) | (n = 3282) | (n = 15495) | (n = 12799) | (n = 2696) | |||
| Male, [n (%)] | 12,712 (57.92) | 10,877 (58.28) | 1835 (55.91) | 0.012 | 8065 (52.05) | 6674 (52.14) | 1391 (51.59) | 0.618 |
| Age, [years] | 68 (57, 78) | 67 (56, 78) | 72 (61, 82) | < 0.001 | 68 (57, 79) | 67 (56, 78) | 71 (61, 81) | < 0.001 |
| Weight, [kg] | 80 (67.1, 95.9) | 80.7 (67.7, 96.6) | 76.9 (64.5, 92.0) | < 0.001 | 168 (160, 177.8) | 168 (160, 177.8) | 167.6 (160., 177.1) | 0.173 |
| Severity of illness | ||||||||
| SOFA score | 3 (2, 4) | 3 (2, 4) | 4 (2, 5) | < 0.001 | 5 (4, 8) | 5 (3, 7) | 8 (5, 10) | < 0.001 |
| SAPS II | 38 (30, 47) | 36 (29, 45) | 48 (39, 59) | < 0.001 | 41 (32, 52) | 39 (31, 49) | 51 (41, 63) | < 0.001 |
| CCI | 6 (4, 8) | 6 (4, 8) | 7 (5, 9) | < 0.001 | 5 (3, 7) | 5 (3, 6) | 6 (4, 7) | < 0.001 |
| Vital signs | ||||||||
| Heart rate, [beats/min] | 104 (91, 119) | 103 (90, 118) | 110 (95, 127) | < 0.001 | 113 (99, 129) | 112 (98, 128) | 119 (104, 136) | < 0.001 |
| MAP, [mmHg] | 58 (51, 64) | 58 (52, 64) | 55 (47, 62) | < 0.001 | 55 (47, 62) | 55 (48, 63) | 51 (42, 58) | < 0.001 |
| RR, [breaths/min] | 28 (24, 32) | 27 (24, 32) | 30 (25, 34) | < 0.001 | 30 (25, 35) | 29 (25, 35) | 32 (28, 38) | < 0.001 |
| Temperature, [℃] | 37.4 (37.0, 37.9) | 37.4 (37, 37.9) | 37.3 (36.9, 37.9) | < 0.001 | 37.6 (37.1, 38.3) | 37.6 (37.1, 38.4) | 37.4 (36.9, 38.2) | < 0.001 |
| Laboratory measurements | ||||||||
| Hemoglobin, [g/dL] | 10.10 (8.80, 11.50) | 10.10 (8.80, 11.50) | 9.80 (8.50, 11.40) | < 0.001 | 9.90 (8.50, 11.50) | 10.00 (8.60, 11.50) | 9.60 (8.20, 11.20) | < 0.001 |
| Platelet, [10^9/L] | 172 (122, 237) | 173 (124, 236) | 171 (110, 242) | 0.003 | 168 (113, 238) | 171 (117, 241) | 155 (86, 229) | < 0.001 |
| WBC, [10^9/L] | 13.6 (9.7, 18.4) | 13.40 (9.7, 18.1) | 14.9 (10.2, 20.4) | < 0.001 | 15.3 (10.2, 21.8) | 15.1 (10.1, 21.4) | 16.6 (10.6, 23.9) | < 0.001 |
| Anion gap, [mmol/L] | 16 (13, 19) | 16 (13, 19) | 18 (15, 22) | < 0.001 | 12 (9, 16) | 12 (9, 15) | 14 (11, 18) | < 0.001 |
| Bicarbonate, [mmol/L] | 22 (19, 24) | 22 (19, 24) | 20 (16, 24) | < 0.001 | 21 (17, 24) | 21 (18, 25) | 19 (15, 23) | < 0.001 |
| BUN, [mmol/L] | 23 (16, 39) | 22 (15, 36) | 33 (21, 54) | < 0.001 | 33 (20, 52) | 31 (20, 50) | 40 (26, 60) | < 0.001 |
| Creatinine, [mg/dL] | 1.1 (0.8, 1.9) | 1.1 (0.8, 1.7) | 1.5 (1.0, 2.6) | < 0.001 | 1.6 (1.0, 2.8) | 1.6 (1.0, 2.7) | 2.0 (1.2, 3.2) | < 0.001 |
| Chloride, [mmol/L] | 106 (102, 110) | 106 (102, 110) | 105 (100, 110) | < 0.001 | 107 (102, 111) | 107 (102, 111) | 107 (102, 112) | 0.379 |
| Sodium, [mmol/L] | 140 (137, 143) | 140 (137, 142) | 140 (136, 143) | 0.280 | 139 (136, 143) | 139 (136, 142) | 140 (136, 144) | < 0.001 |
| Potassium, [mmol/L] | 4.5 (4.1, 5.0) | 4.4 (4.1, 4.9) | 4.6 (4.2, 5.3) | < 0.001 | 4.3 (3.9, 4.9) | 4.3 (3.9, 4.8) | 4.5 (4.0, 5.1) | < 0.001 |
| INR | 1.4 (1.2, 1.6) | 1.3 (1.2, 1.6) | 1.5 (1.2, 2.1) | < 0.001 | 1.5 (1.3, 1.9) | 1.5 (1.2, 1.8) | 1.8 (1.4, 2.3) | < 0.001 |
| Lactate, [mmol/L] | 2.2 (1.6, 3.0) | 2.2 (1.5, 2.9) | 2.7 (1.8, 4.4) | < 0.001 | 2.4 (1.5, 4.0) | 2.3 (1.4, 3.6) | 3.5 (2.1, 6.1) | < 0.001 |
| Urine output, [mL] | 1569 (950, 2385) | 1655 (1048, 2465) | 1012 (500, 1780) | < 0.001 | 1320 (675, 1968) | 1434 (794, 2060) | 834 (305, 1385) | < 0.001 |
| Comorbidities | ||||||||
| CHF, [n (%)] | 7312 (33.32) | 5993 (32.11) | 1319 (40.19) | < 0.001 | 3546 (22.88) | 2869 (22.42) | 677 (25.11) | 0.003 |
| CVD, [n (%)] | 3401 (15.50) | 2682 (14.37) | 719 (21.91) | < 0.001 | 2199 (14.19) | 1759 (13.74) | 440 (16.32) | 0.001 |
| Diabetes, [n (%)] | 7201 (32.81) | 6136 (32.88) | 1065 (32.45) | 0.646 | 470 (3.03) | 395 (3.09) | 75 (2.78) | 0.438 |
| Renal disease, [n (%)] | 5396 (24.59) | 4378 (23.46) | 1018 (31.02) | < 0.001 | 3685 (23.78) | 2988 (23.35) | 697 (25.85) | 0.006 |
| Liver disease, [n (%)] | 2913 (13.27) | 2219 (11.89) | 694 (21.15) | < 0.001 | 1287 (8.31) | 885 (6.91) | 402 (14.91) | < 0.001 |
| Malignancy, [n (%)] | 3233 (14.73) | 2421 (12.97) | 812 (24.74) | < 0.001 | 2870 (18.52) | 2189 (17.10) | 681 (25.26) | < 0.001 |
| Treatment | ||||||||
| MV, [n (%)] | 10,279 (46.84) | 8323 (44.59) | 1956 (59.60) | < 0.001 | 6813 (43.97) | 5135 (40.12) | 1678 (62.24) | < 0.001 |
| Vasopressor use, [n (%)] | 9373 (42.71) | 7611 (40.78) | 1762 (53.69) | < 0.001 | 5403 (34.87) | 4086 (31.92) | 1317 (48.85) | < 0.001 |
| RRT, [n (%)] | 793 (3.61) | 595 (3.19) | 198 (6.03) | < 0.001 | 954 (6.16) | 745 (5.82) | 209 (7.75) | < 0.001 |
SOFA, Sequential Organ Failure Assessment; SAPS II, Simplified Acute Physiology Score II; CCI, Charlson Comorbidity Index; MAP, mean arterial pressure; RR, respiration rate; WBC, white blood cells; BUN, Blood Urea Nitrogen; INR, international normalized ratio; CHF, Congestive Heart Failure; CVD, cerebrovascular disease; MV, mechanical ventilation; RRT, renal replacement therapy
Table 2.
Characteristics and outcomes of participants from MIMIC-IV categorized by hemoglobin quartiles
| Variables | Q1 (≤ 8.8) | Q2 (8.9–10.1) | Q3 (10.2–11.5) | Q4 (≥ 11.6) | P-value |
|---|---|---|---|---|---|
| (n = 5874) | (n = 5523) | (n = 5308) | (n = 5241) | ||
| Male, [n (%)] | 2937 (50.00) | 3090 (55.95) | 3127 (58.91) | 3558 (67.89) | < 0.001 |
| Age, [years] | 69 (58, 79) | 69 (59, 79) | 68 (57, 79) | 65 (53, 77) | < 0.001 |
| Weight, [kg] | 77.10 (65.00, 91.60) | 79.70 (66.70, 95.20) | 80.60 (67.00, 96.80) | 83.70 (70.90, 100.20) | < 0.001 |
| Severity of illness | |||||
| SOFA score | 3 (2, 5) | 3 (2, 4) | 3 (2, 4) | 3 (2, 4) | < 0.001 |
| SAPS II | 40 (32, 49) | 38 (31, 47) | 37 (29, 46) | 35 (27, 45) | < 0.001 |
| CCI | 7 (5, 9) | 6 (4, 8) | 6 (4, 8) | 5 (3, 7) | < 0.001 |
| Vital signs | |||||
| Heart rate, [beats/min] | 103 (90, 119) | 103 (90, 118) | 104 (91, 119) | 105 (92, 121) | < 0.001 |
| MAP, [mmHg] | 56 (50, 62) | 57 (50, 63) | 58 (51, 64) | 61 (53, 68) | < 0.001 |
| RR, [breaths/min] | 28 (24, 32) | 28 (24, 32) | 28 (24, 32) | 28 (24, 32) | 0.12 |
| Temperature, [℃] | 37.30 (37, 37.80) | 37.40 (37, 37.90) | 37.40 (37, 37.90) | 37.40 (37, 38) | < 0.001 |
| Laboratory measurements | |||||
| Platelet, [10^9/L] | 159 (105, 242) | 170 (120, 237) | 174 (126.75, 237) | 183 (139, 234) | < 0.001 |
| WBC, [10^9/L] | 13.30 (9.20, 18.60) | 13.50 (9.60, 18.40) | 13.70 (10.00, 18.30) | 13.90 (10.20, 18.50) | < 0.001 |
| Anion gap, [mmol/L] | 16 (13, 19) | 16 (13, 19) | 16 (13, 19) | 17 (14, 19) | < 0.001 |
| Bicarbonate, [mmol/L] | 21 (18, 24) | 22 (19, 24) | 22 (19, 24) | 22 (19, 25) | < 0.001 |
| BUN, [mmol/L] | 27 (17, 48) | 23 (16, 40) | 22 (15, 35) | 21 (15, 31) | < 0.001 |
| Chloride, [mmol/L] | 106 (101, 111) | 106 (102, 110) | 106 (102, 110) | 106 (102, 109) | < 0.001 |
| Creatinine, [mg/DL] | 1.30 (0.80, 2.40) | 1.10 (0.80, 1.90) | 1.10 (0.80, 1.70) | 1.10 (0.80, 1.50) | < 0.001 |
| Sodium, [mmol/L] | 139 (137, 142) | 140 (137, 142) | 140 (137, 142) | 140 (138, 143) | < 0.001 |
| Potassium, [mmol/L] | 4.50 (4.10, 5.00) | 4.50 (4.10, 4.90) | 4.40 (4.10, 4.90) | 4.40 (4.00, 5.00) | < 0.001 |
| INR | 1.40 (1.20, 1.70) | 1.40 (1.20, 1.60) | 1.30 (1.20, 1.60) | 1.20 (1.10, 1.60) | < 0.001 |
| Lactate, [mmol/L] | 2.20 (1.50, 3.10) | 2.20 (1.60, 3.00) | 2.20 (1.60, 3.00) | 2.30 (1.70, 3.10) | < 0.001 |
| Urine output, [mL] | 1440 (842, 2230) | 1555 (945, 2360) | 1595 (975, 2391) | 1705 (1068, 2573) | < 0.001 |
| Comorbidities | |||||
| CHF, [n (%)] | 2254 (38.37) | 1859 (33.66) | 1703 (32.08) | 1496 (28.54) | < 0.001 |
| CVD, [n (%)] | 760 (12.94) | 696 (12.60) | 822 (15.49) | 1123 (21.43) | < 0.001 |
| Diabetes, [n (%)] | 2230 (37.96) | 1976 (35.78) | 1630 (30.71) | 1365 (26.04) | < 0.001 |
| Renal disease, [n (%)] | 2040 (34.73) | 1465 (26.53) | 1102 (20.76) | 789 (15.05) | < 0.001 |
| Liver disease, [n (%)] | 958 (16.31) | 713 (12.91) | 639 (12.04) | 603 (11.51) | < 0.001 |
| Malignancy, [n (%)] | 1102 (18.76) | 862 (15.61) | 681 (12.83) | 588 (11.22) | < 0.001 |
| Treatment | |||||
| MV, [n (%)] | 2764 (47.05) | 2589 (46.88) | 2434 (45.86) | 2492 (47.55) | 0.357 |
| Vasopressor use, [n (%)] | 2864 (48.76) | 2612 (47.29) | 2210 (41.64) | 1687 (32.19) | < 0.001 |
| RRT, [n (%)] | 301 (5.12) | 214 (3.87) | 187 (3.52) | 91 (1.74) | < 0.001 |
| Outcomes | |||||
| In-hospital mortality [n (%)] | 1064 (18.11) | 762 (13.80) | 707 (13.32) | 749 (14.29) | < 0.001 |
| In-ICU mortality [n (%)] | 692 (11.78) | 541 (9.80) | 510 (9.61) | 561 (10.70) | < 0.001 |
| 28-day mortality [n (%)] | 1316 (22.40) | 973 (17.62) | 892 (16.80) | 916 (17.48) | < 0.001 |
| ICU-LOS [days] | 3 (2, 6) | 3 (2, 6) | 3 (2, 6) | 4 (2, 8) | < 0.001 |
| Hospital-LOS [days] | 9 (6, 17) | 8 (5, 14) | 8 (5, 14) | 9 (5, 15) | < 0.001 |
SOFA, Sequential Organ Failure Assessment; SAPS II, Simplified Acute Physiology Score II; CCI, Charlson Comorbidity Index; MAP, mean arterial pressure; RR, respiration rate; WBC, white blood cells; BUN, Blood Urea Nitrogen; INR, international normalized ratio; CHF, Congestive Heart Failure; CVD, cerebrovascular disease; MV, mechanical ventilation; RRT, renal replacement therapy; ICU, intensive care unit; LOS, length of stay
Relationship between hemoglobin and mortality
We conducted three logistic regression models to explore the relationship between hemoglobin and mortality. After multivariate adjustment for age, gender, weight, severity score, vital signs, laboratory test results, chronic diseases, and interventions (model 3), the third (Q3) quartile of hemoglobin was associated with lower risks of in-hospital and 28-day mortality compared with the first quartile (Q1) (Table 3 and Supplementary Table 2, Additional File 1). The multivariable-adjusted ORs and 95% CIs for the risk of in-hospital mortality from the first to the fourth quartile of hemoglobin (≤ 8.8 g/DL, 8.9–10.1 g/DL, 10.2–11.5 g/DL, and ≥ 11.6 g/DL) were 1.00 (reference), 0.868 (0.775, 0.972), 0.871 (0.774, 0.980), 1.043 (0.925,1.176); the ORs and 95% CIs for the risk of 28-day mortality were 1.00 (reference), 0.883 (0.795, 0.980), 0.875 (0.785, 0.975), and 1.038 (0.928, 1.161), respectively. The results for the eICU-CRD were shown in Supplementary Table 2, Additional File 1.
Table 3.
Association between hemoglobin and in-hospital and 28-day mortality in patients with sepsis from MIMI-IV
| Exposure | Model 1 | Model 2 | Model 3 | |||
|---|---|---|---|---|---|---|
| OR (95% CI) | P-value | OR (95% CI) | P-value | OR (95% CI) | P-value | |
| In-hospital mortality | ||||||
| Hemoglobin as continuous | 0.956 (0.938, 0.974) | < 0.001 | 0.971 (0.952, 0.990) | 0.003 | 1.018 (0.996,1.041) | 0.105 |
| Q1 | Ref | Ref | Ref | |||
| Q2 | 0.724 (0.654, 0.801) | < 0.001 | 0.725 (0.655, 0.803) | < 0.001 | 0.868 (0.775, 0.972) | 0.014 |
| Q3 | 0.695 (0.627, 0.770) | < 0.001 | 0.706 (0.636, 0.784) | < 0.001 | 0.871 (0.774, 0.980) | 0.021 |
| Q4 | 0.754 (0.681, 0.835) | < 0.001 | 0.818 (0.738, 0.908) | < 0.001 | 1.043 (0.925,1.176) | 0.494 |
| P for trend | < 0.001 | < 0.001 | 0.743 | |||
| 28-day mortality | ||||||
| Hemoglobin as continuous | 0.951(0.934, 0.968) | < 0.001 | 0.972(0.954, 0.990) | 0.002 | 1.017(0.996, 1.038) | 0.116 |
| Q1 | Ref | Ref | Ref | |||
| Q2 | 0.741(0.675, 0.812) | < 0.001 | 0.742(0.675, 0.815) | < 0.001 | 0.883(0.795, 0.980) | 0.019 |
| Q3 | 0.700 (0.636, 0.769) | < 0.001 | 0.713(0.648, 0.785) | < 0.001 | 0.875(0.785, 0.975) | 0.016 |
| Q4 | 0.734(0.668, 0.806) | < 0.001 | 0.822(0.746, 0.905) | < 0.001 | 1.038(0.928, 1.161) | 0.518 |
| P for trend | < 0.001 | < 0.001 | 0.818 | |||
Model 1: Unadjusted
Model 2: Adjusted for age, gender, weight
Model 3: Adjusted for age, gender, weight, SOFA, SAPS II, CCI, heart rate, MAP, RR, temperature, platelets, WBC, anion gap, bicarbonate, BUN, creatinine, chloride, sodium, potassium, INR, lactate, urine output, CHF, CVD, diabetes, renal disease, liver disease, malignancy, MV, vasopressor use, RRT
Nonlinear relationship between hemoglobin and mortality
We found a nonlinear association between hemoglobin and in-hospital mortality by using RCS models with adjustment for other variables (P for nonlinear = 0.003 in MIMIC-IV, P for nonlinear = 0.022 in eICU-CRD) (Fig. 3). Subsequently, a logistic regression model with a two-piecewise logistic regression model was conducted to assess the nonlinear relationship between hemoglobin and in-hospital mortality in patients with sepsis (P for log-likelihood ratio < 0.001) (Table 4). We found an inflection point for in-hospital mortality was 10.2 g/DL. When hemoglobin was less than 10.2 g/DL, a 1-unit decrease in hemoglobin was associated with a 4.6% increase in adjusted OR. When hemoglobin exceeded 10.2 g/DL, the risk of in-hospital death was increased by 7.6% for each 1 g/DL increase in hemoglobin. Similar trends were validated in the eICU-CRD (Supplementary Table 3, Additional File 1).
Fig. 3.
Restricted cubic spline analysis for the nonlinear association between hemoglobin levels and the risk of in-hospital mortality of patients with sepsis from MIMIC-IV (A), and eICU-CRD (B). Adjusted for age, gender, weight, SOFA, SAPS II, CCI, heart rate, MAP, RR, temperature, platelets, WBC, anion gap, bicarbonate, BUN, creatinine, chloride, sodium, potassium, INR, lactate, urine output, CHF, CVD, diabetes, renal disease, liver disease, malignancy, MV, vasopressor use, RRT
Table 4.
Threshold effect analysis of hemoglobin on in-hospital mortality in patients with sepsis
| Adjusted OR (95% CI) | P-value | |
|---|---|---|
| Fitting by the standard linear model | 1.018 (0.996, 1.041) | 0.105 |
| Fitting by the two-piecewise linear model | ||
| Inflection point | 10.2 g/DL | |
| Hemoglobin < 10.2 g/DL | 0.954 (0.914, 0.997) | 0.034 |
Hemoglobin 10.2 g/DL |
1.076 (1.035, 1.118) | < 0.001 |
| P for Log-likelihood ratio | < 0.001 |
Adjusted for age, gender, weight, SOFA, SAPS II, CCI, heart rate, MAP, RR, temperature, platelets, WBC, anion gap, bicarbonate, BUN, creatinine, chloride, sodium, potassium, INR, lactate, urine output, CHF, CVD, diabetes, renal disease, liver disease, malignancy, MV, vasopressor use, RRT
Stratified analyses
The data of Fig. 4 illustrated the association between baseline hemoglobin levels and in-hospital mortality, stratified by gender, age, congestive heart failure (CHF), cerebrovascular disease (CVD), diabetes, renal disease, liver disease, mechanical ventilation (MV), vasopressor use, and renal replacement therapy (RRT). The results demonstrated that higher hemoglobin levels were significantly associated with lower mortality risk among patients younger than 65 years, those receiving MV, and those using vasopressors, with statistically significant interaction P-values. The data of supplementary Fig. 1, Additional File 2 presented the stratified analysis from the eICU-CRD database, where no statistically significant interactions between hemoglobin levels and the stratified variables were observed.
Fig. 4.
Forest plots of stratified analyses of hemoglobin and in-hospital mortality of patients with sepsis from MIMIC-IV. Age, gender, weight, SOFA, SAPS II, CCI, heart rate, MAP, RR, temperature, platelets, WBC, anion gap, bicarbonate, BUN, creatinine, chloride, sodium, potassium, INR, lactate, urine output, CHF, CVD, diabetes, renal disease, liver disease, malignancy, MV, vasopressor use and RRT were all adjusted except the variable itself
Discussion
This large multi-center retrospective cohort study of adults with sepsis revealed that hemoglobin was significantly associated with the risk of in-hospital mortality. We identified a nonlinear association between hemoglobin and in-hospital mortality. This relationship was characterized by an inflection point of 10.2 g/DL for overall sepsis patients. Below this threshold, a 1 g/DL decrease in hemoglobin was associated with a 4.6% increase in the risk of in-hospital mortality, while above this threshold, a 1 g/DL increase was associated with a 7.6% increase in the risk of in-hospital mortality.
The development of acute organ dysfunction in sepsis is caused by the systemic excessive pro-inflammatory and pro-coagulant responses to the infection. Organ dysfunction in patients with sepsis is associated with increased mortality. Most patients with sepsis develop hematologic changes, and hemoglobin changes are one of the most common abnormalities [18, 19]. Several retrospective studies have established that hemoglobin at admission or discharge is an important prognostic biomarker for sepsis [20–22]. Rapid identification and treatment of hematologic dysfunction may improve survival. The etiology of anemia in ICU patients includes decreased erythrocyte viability, blood loss, dysregulated iron metabolism, and bone marrow suppression [4, 23]. Tyler et al. found that persistent inflammation in sepsis patients is associated with persistent anemia [24]. The combination of anemia and oxygen consumption changes caused by sepsis may exacerbate the impairment of tissue oxygenation [25], and maintaining adequate blood hemoglobin levels may serve as one way to reduce sepsis-induced tissue damage. Our research found that the risk of in-hospital mortality increased with decreasing hemoglobin when hemoglobin was below 10.2 g/DL. These results were similar to several previous related studies [26–29]. Chen et al. showed a negative correlation between hemoglobin and the risk of 28-day mortality in patients with sepsis in the range of 41–104 g/L and a positive correlation in the range of 128–207 g/L [26]. A study by Tan et al. noted that patients with sepsis who were moderately anemic (hemoglobin 7–10 g/DL) would have a worse prognosis than those who were not moderately anemic (hemoglobin > 10 g/DL) [18].
Several previous studies pointed out a linear negative correlation between hemoglobin levels and the prognosis of patients with sepsis [25, 30]. It appeared that a higher hemoglobin level was associated with a more favorable prognosis. However, it was of concern that elevated hemoglobin levels did not necessarily improve the prognosis of patients with sepsis [31]. Our results suggested a positive correlation between the risk of in-hospital mortality and hemoglobin changes when hemoglobin was higher than 10.2 g/DL. Although the mechanism underlying this conclusion was not yet clear, we speculated that one significant factor was volume deficiency due to inadequate fluid resuscitation. Moreover, sepsis is characterized by diffuse endothelial injury and capillary hyperpermeability, resulting in greater extravasation of fluid [32]. All of these factors may lead to hemoconcentration and thus elevated hemoglobin levels, ultimately affecting hemodynamics and oxygen delivery. Importantly, the maintenance of microcirculatory oxygen supply depended not only on the oxygen-carrying capacity of erythrocytes but also on adequate blood viscosity. Excessively high hemoglobin levels may increase blood viscosity, which may lead to reduced oxygen delivery [33]. Furthermore, it has also been reported that hemoglobin increased leukocyte-endothelial adhesion in inflammation and mediated tissue damage through TLR4 signaling [34].
Identifying septic patient populations with an elevated risk of in-hospital mortality holds significant clinical relevance. Based on the stratified analysis, we observed the advantage in in-hospital mortality from higher hemoglobin (≥ 10.2 g/DL) in patients with sepsis, particularly in those younger than 65 years old, those receiving MV, and those receiving vasopressor treatment. The significant effect of hemoglobin levels on in-hospital mortality in patients younger than 65 years may be related to their better baseline health status, higher intensity of treatment, and greater physiological reserve capacity. Although the mechanism underlying these findings remains unclear, we speculated it may be linked to reduced oxygen utilization in these patients. Practical clinical attention should be directed towards individuals undergoing mechanical ventilation and vasoactive drug therapy. In conclusion, our study supported the view that maintaining a stable hemoglobin status facilitated the clinical management of septic patients. At the same time, individualized hemoglobin management strategies need to be developed based on patient age, treatment regimen received, and other characteristics.
There are some limitations to our study. Firstly, being an observational study, it elucidated a correlation between hemoglobin levels and sepsis prognosis; however, establishing a causal relationship was not within its scope. Secondly, our examination centered around a single baseline hemoglobin level, potentially overlooking valuable insights into dynamic hemoglobin changes [35]. Finally, our analysis did not incorporate fluid resuscitation volume, particularly positive fluid balance, which could potentially influence hemoglobin levels and oxygen delivery. Our observational research suggested that hemoglobin could serve as a robust marker for evaluating the prognosis of sepsis patients. The present study provided a foundation for future prospective investigations to assess the impact of hemoglobin on sepsis patient outcomes and determine causality.
Conclusions
A nonlinear relationship was shown between hemoglobin and in-hospital mortality in adult patients with sepsis. Both abnormally low and excessively high levels of hemoglobin adversely affected the prognosis of septic patients.
Electronic supplementary material
Below is the link to the electronic supplementary material.
Acknowledgements
Not applicable.
Abbreviations
- MIMIC-IV
Medical Information Mart for Intensive Care-IV
- OR
Odds ratios
- CI
Confidence intervals
- RCS
Restricted cubic spline
- CHF
Congestive heart failure
- CVD
Cerebrovascular disease
- MV
Mechanical ventilation
- SAPS II
Simplified acute physiology score II
- CCI
Charlson comorbidity index
- MAP
Mean arterial pressure
- RRT
Renal replacement therapy
- SOFA
Sequential organ failure assessment
- RR
Respiration rate
- WBC
White blood cells
- BUN
Blood urea nitrogen
- INR
International normalized ratio
Author contributions
SS performed formal analysis, and drafted the original manuscript. AL was responsible for the methodology, software implementation, and drafted the original manuscript. CZ managed data curation and contributed to drafting the original manuscript. XL was involved in software implementation, and reviewed and edited the manuscript. WZ handled visualization, conducted investigation, and reviewed and edited the manuscript. TS contributed to visualization, investigation, and reviewed and edited the manuscript. QM performed data analyses and reviewed and edited the manuscript. SM provided conceptualization, supervision, and reviewed and edited the manuscript. FZ provided conceptualization, supervision, and reviewed and edited the manuscript. All authors read and approved the final manuscript.
Funding
This work was supported by the National Key R&D Program “Stem Cell and Transformation Research” Key Special Project(2019YFA0110601) and the peak supporting clinical discipline of Shanghai health bureau (2023ZDFC0104).
Data availability
The datasets supporting the conclusions of this article are available in the PhysioNet (https://physionet.org/content/mimiciv/2.2/; https://physionet.org/content/eicu-crd/2.0/). To access the data, you must be a credentialed user, complete the required training (CITI Data or Specimens Only Research), and sign the data use agreement for the project.
Declarations
Ethics approval and consent to participate
The institutional review board at the Beth Israel Deaconess Medical Center approved the study and granted a waiver of informed consent. All the patients in the database were deidentified.
Consent for publication
Not applicable.
Competing interests
The authors declare no competing interests.
Footnotes
Publisher’s note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Shuyue Sheng, Andong Li and Changjing Zhang contributed equally to this work.
Contributor Information
Shaolin Ma, Email: mslin@sohu.com.
Feng Zhu, Email: alexzhujunchi@hotmail.com.
References
- 1.Singer M, Deutschman CS, Seymour CW, Shankar-Hari M, Annane D, Bauer M, et al. The Third International Consensus definitions for Sepsis and septic shock (Sepsis-3). JAMA. 2016;315(8):801–10. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Rudd KE, Johnson SC, Agesa KM, Shackelford KA, Tsoi D, Kievlan DR, et al. Global, regional, and national sepsis incidence and mortality, 1990–2017: analysis for the global burden of Disease Study. Lancet. 2020;395(10219):200–11. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Xie J, Wang H, Kang Y, Zhou L, Liu Z, Qin B, et al. The epidemiology of Sepsis in Chinese ICUs: a National Cross-sectional Survey. Crit Care Med. 2020;48(3):e209–18. [DOI] [PubMed] [Google Scholar]
- 4.Hayden SJ, Albert TJ, Watkins TR, Swenson ER. Anemia in critical illness: insights into etiology, consequences, and management. Am J Respir Crit Care Med. 2012;185(10):1049–57. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Warner MA, Hanson AC, Frank RD, Schulte PJ, Go RS, Storlie CB, et al. Prevalence of and Recovery from Anemia following hospitalization for critical illness among adults. JAMA Netw Open. 2020;3(9):e2017843. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Qi D, Peng M. Early Hemoglobin Status as a predictor of long-term mortality for Sepsis patients in Intensive Care Units. Shock. 2021;55(2):215–23. [DOI] [PubMed] [Google Scholar]
- 7.Mustahsin M, Maitra S, Anand RK, Soneja M, Madan K, Darlong V, et al. Transfusion trigger in the critically ill with sepsis or septic shock: a prospective study. Indian J Med Res. 2023;158(3):276–83. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Effenberger-Neidnicht K, Hartmann M. Mechanisms of Hemolysis during Sepsis. Inflammation. 2018;41(5):1569–81. [DOI] [PubMed] [Google Scholar]
- 9.Barie PS. Phlebotomy in the intensive care unit: strategies for blood conservation. Crit Care. 2004;8(Suppl 2):S34–36. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.François T, Sauthier M, Charlier J, Dessureault J, Tucci M, Harrington K, et al. Impact of blood sampling on Anemia in the PICU: a prospective cohort study. Pediatr Crit Care Med. 2022;23(6):435–43. [DOI] [PubMed] [Google Scholar]
- 11.Walsh OM, Davis K, Gatward J. Reducing inappropriate arterial blood gas testing in a level III intensive care unit: a before-and-after observational study. Crit Care Resusc. 2020;22(4):370–7. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Johnson AEW, Bulgarelli L, Shen L, Gayles A, Shammout A, Horng S, et al. MIMIC-IV, a freely accessible electronic health record dataset. Sci Data. 2023;10(1):1. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Pollard TJ, Johnson AEW, Raffa JD, Celi LA, Mark RG, Badawi O. The eICU Collaborative Research Database, a freely available multi-center database for critical care research. Sci Data. 2018;5:180178. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Yang M, Liu C, Wang X, Li Y, Gao H, Liu X, et al. An explainable Artificial Intelligence Predictor for early detection of Sepsis. Crit Care Med. 2020;48(11):e1091–6. [DOI] [PubMed] [Google Scholar]
- 15.van Buuren S, Groothuis-Oudshoorn K. Mice: multivariate imputation by chained equations in R. J Stat Softw. 2011;45(3):1–67. [Google Scholar]
- 16.Degenhardt F, Seifert S, Szymczak S. Evaluation of variable selection methods for random forests and omics data sets. Brief Bioinform. 2019;20(2):492–503. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Zheng R, Qian S, Shi Y, Lou C, Xu H, Pan J. Association between triglyceride-glucose index and in-hospital mortality in critically ill patients with sepsis: analysis of the MIMIC-IV database. Cardiovasc Diabetol. 2023;22(1):307. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Tan SMY, Zhang Y, Chen Y, See KC, Feng M. Association of fluid balance with mortality in sepsis is modified by admission hemoglobin levels: a large database study. PLoS ONE. 2021;16(6):e0252629. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.Aird WC. The hematologic system as a marker of organ dysfunction in sepsis. Mayo Clin Proc. 2003;78(7):869–81. [DOI] [PubMed] [Google Scholar]
- 20.Shankar-Hari M, Rubenfeld GD, Ferrando-Vivas P, Harrison DA, Rowan K. Development, Validation, and clinical Utility Assessment of a Prognostic score for 1-Year unplanned rehospitalization or death of adult Sepsis survivors. JAMA Netw Open. 2020;3(9):e2013580. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Singh A, Bhagat M, George SV, Gorthi R, Chaturvedula C. Factors Associated with 30-day unplanned readmissions of Sepsis patients: a retrospective analysis of patients admitted with Sepsis at a Community Hospital. Volume 11. Cureus; 2019. p. e5118. 7. [DOI] [PMC free article] [PubMed]
- 22.Denstaedt SJ, Cano J, Wang XQ, Donnelly JP, Seelye S, Prescott HC. Blood count derangements after sepsis and association with post-hospital outcomes. Front Immunol. 2023;14:1133351. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23.Rayes HA, Vallabhajosyula S, Barsness GW, Anavekar NS, Go RS, Patnaik MS, et al. Association between anemia and hematological indices with mortality among cardiac intensive care unit patients. Clin Res Cardiol. 2020;109(5):616–27. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24.Loftus TJ, Mira JC, Stortz JA, Ozrazgat-Baslanti T, Ghita GL, Wang Z, et al. Persistent inflammation and anemia among critically ill septic patients. J Trauma Acute Care Surg. 2019;86(2):260–7. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25.Muady GF, Bitterman H, Laor A, Vardi M, Urin V, Ghanem-Zoubi N. Hemoglobin levels and blood transfusion in patients with sepsis in Internal Medicine Departments. BMC Infect Dis. 2016;16(1):569. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26.Chen Y, Chen L, Meng Z, Li Y, Tang J, Liu S, et al. The correlation of hemoglobin and 28-day mortality in septic patients: secondary data mining using the MIMIC-IV database. BMC Infect Dis. 2023;23(1):417. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27.Reade MC, Weissfeld L, Angus DC, Kellum JA, Milbrandt EB. The prevalence of anemia and its association with 90-day mortality in hospitalized community-acquired pneumonia. BMC Pulm Med. 2010;10:15. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28.Oh SM, Skendelas JP, Macdonald E, Bergamini M, Goel S, Choi J, et al. On-admission anemia predicts mortality in COVID-19 patients: a single center, retrospective cohort study. Am J Emerg Med. 2021;48:140–7. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29.Lin IH, Liao PY, Wong LT, Chan MC, Wu CL, Chao WC. Anaemia in the first week may be associated with long-term mortality among critically ill patients: propensity score-based analyses. BMC Emerg Med. 2023;23(1):32. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30.Cai N, Fan W, Tao M, Liao W. A significant decrease in hemoglobin concentrations may predict occurrence of necrotizing enterocolitis in preterm infants with late-onset sepsis. J Int Med Res. 2020;48(9):300060520952275. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 31.Chan YL, Han ST, Li CH, Wu CC, Chen KF. Transfusion of Red Blood Cells to Patients with Sepsis. Int J Mol Sci. 2017; 18(9). [DOI] [PMC free article] [PubMed]
- 32.Quispe-Cornejo AA, Alves da Cunha AL, Njimi H, Mongkolpun W, Valle-Martins AL, Arébalo-López M, et al. Effects of rapid fluid infusion on hemoglobin concentration: a systematic review and meta-analysis. Crit Care. 2022;26(1):324. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 33.Zimmerman R, Tsai AG, Salazar Vázquez BY, Cabrales P, Hofmann A, Meier J, et al. Posttransfusion increase of hematocrit per se does not improve circulatory oxygen delivery due to increased blood viscosity. Anesth Analg. 2017;124(5):1547–54. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 34.Conger AK, Tomasek T, Riedmann KJ, Douglas JS, Berkey LE, Ware LB, et al. Hemoglobin increases leukocyte adhesion and initiates lung microvascular endothelial activation via toll-like receptor 4 signaling. Am J Physiol Cell Physiol. 2023;324(3):C665–73. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 35.Jiang Y, Jiang FQ, Kong F, An MM, Jin BB, Cao D, et al. Inflammatory anemia-associated parameters are related to 28-day mortality in patients with sepsis admitted to the ICU: a preliminary observational study. Ann Intensive Care. 2019;9(1):67. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Data Availability Statement
The datasets supporting the conclusions of this article are available in the PhysioNet (https://physionet.org/content/mimiciv/2.2/; https://physionet.org/content/eicu-crd/2.0/). To access the data, you must be a credentialed user, complete the required training (CITI Data or Specimens Only Research), and sign the data use agreement for the project.