Sage Journals: Discover world-class research

Abstract

Objective

This study aimed to develop and validate a predictive risk nomogram for sepsis-associated severe anemia.

Methods

A prediction model was built using data from 252 sepsis patients in a single institution (January 2022 to December 2023). Severe anemia was defined as a hemoglobin level <60 g/L. Least absolute shrinkage and selection operator regression was used to identify key predictors, and multivariable logistic regression was used to construct the nomogram. Model performance was assessed via the receiver operating characteristic curve (C-index), calibration plots, and decision curve analysis. Internal validation was performed using bootstrapping.

Results

Predictors included age, length of intensive care unit stay, nutritional method, and Acute Physiology and Chronic Health Evaluation II score. The model demonstrated good discrimination (C-index: 0.8848) and calibration, with high internal validation performance. Decision curve analysis indicated optimal clinical utility at risk thresholds between 5% and 75%.

Conclusions

The constructed nomogram, incorporating age, length of intensive care unit stay, nutritional method, and Acute Physiology and Chronic Health Evaluation II score, provides a practical tool for early individualized care in sepsis patients.

Keywords

Sepsis severe anemia nomogram prediction validation

Introduction

Sepsis is a life-threatening medical condition caused by organ dysfunction related to the host’s response to infection. Sepsis has a high mortality rate of 30%–50% even in industrialized countries,¹ and its progression involves various pathophysiologic characteristics that serve as therapeutic targets. In recent years, the use of early goal-directed therapy (EGDT) has ushered in a new era of symptomatic supportive treatments.² Sepsis induces a wide range of effects in erythrocytes, including increased reactive oxygen species production and oxidative stress/damage, decreased erythrocyte deformability and redistribution of membrane phospholipids, and increased erythrocyte death (i.e. eryptosis),^3,4 all of which lead to a condition known as sepsis-associated anemia. Moreover, the systemic inflammation that occurs during sepsis induces a massive release of hemoglobin from broken erythrocytes. Elevated levels of circulating cell-free hemoglobin (CFH) further contribute to progressive anemia via its prosthetic heme group.⁵

Sepsis-associated severe anemia is mainly associated with inflammatory anemia–related parameters, such as erythropoietin (EPO), hepcidin, and ferritin, markers that have been shown to be associated with 28-day mortality in sepsis patients.⁶ Several cytokines and chemokines have been shown to block intestinal iron absorption, resulting in iron-restricted erythropoiesis.⁷ Normal hemoglobin values are >130 g/L for males and >120 g/L for females. The severity of anemia is divided into the following four degrees based on these hemoglobin values: mild anemia (hemoglobin level: 90–120 g/L), moderate anemia (hemoglobin level: 60–90 g/L), severe anemia (hemoglobin level: <60 g/L), and most severe anemia (hemoglobin level: <30 g/L).⁸ Red blood cell (RBC) transfusions are frequently necessary to correct the most severe forms of anemia. Blood transfusions are costly and carry substantial risks.⁹ Although previous studies have identified many variables linked to sepsis-associated anemia, the degree to which these variables can predict clinical outcomes in these patients remains unknown. Nomograms are a proven and useful clinical tool for predicting adverse event risks and overall survival in several diseases and medical conditions. They provide evidence-based and highly accurate risk estimates through visual and practical methods.¹⁰ Based on routine clinical and demographic characteristics, a predictive nomogram could be valuable in identifying sepsis patients who might later present with sepsis-associated severe anemia. Nevertheless, to the best of our knowledge, no such nomogram exists.

The purpose of this study was to develop a simple nomogram to predict sepsis-associated severe anemia and internally validate its clinical applicability in a small, single-site study. This nomogram could not only provide intensive care physicians with a tool that can predict the appearance of severe anemia in patients upon admission, even before blood tests are completed, but may also effectively predict anemia in patients with normal hemoglobin levels throughout the duration of hospitalization.

Patients and methods

Patients

Research approval was obtained from Tongji Hospital affiliated to Tongji Medical College of Huazhong University of Science and Technology. Medical records were retrospectively analyzed from patients admitted to the institution from all over China between 1 April 2022 and 31 March 2023. Participants were included if they met the Sepsis 3.0 (2016) diagnostic criteria. Organ dysfunction was identified as a change in the Sequential (sepsis-related) Organ Failure Assessment score of >2 points. Patient demographic, mortality, and related treatments were extracted from our extensive electronic medical record system. The clinical indicators selected in our research focused on anemia-related factors, including age, sex, presence of infections, use of antibiotics, continuous renal replacement treatment (CRRT), and mechanical ventilation, all of which may influence the human hematopoietic system to varying degrees. Informed consent was obtained from all participants, personal information was kept confidential, and all patient details were deidentified so that they could not be identified in any way. This retrospective study was conducted in accordance with the Helsinki Declaration of 1975, as revised in 2024, and the reporting of this study conforms to the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines.¹¹

Outcome measurements

Severe anemia in our cohort was defined according to the international anemia risk degree (i.e. if hemoglobin level <60 g/L).¹² Based on hemoglobin level, participants were divided into two groups: severe anemia and nonsevere anemia. EGDT was applied accordingly during our standard of care for sepsis, including maintenance of central venous pressure (CVP), mean arterial pressure (MAP), urine output, and central venous oxygen saturation (ScvO₂). Specifically, for the first 6 h, a CVP of 1.07–1.60 kPa was maintained through adequate fluid resuscitation. Vasopressors were used to target a MAP goal of ≥8.67 kPa. Urine output was dynamically monitored to achieve a goal of ≥0.5 mL/kg/h. A ScvO₂ goal of ≥70% was targeted, and packed RBC transfusions were used if necessary.² Mechanical ventilation was applied for those with respiratory failure (PaO₂ (oxygen partial pressure) <8.00 kPa or PaCO₂ (carbon dioxide partial pressure) >6.67 kPa), and CRRT was applied for those with acute kidney injury (e.g. oliguria, anuria, blood pH <7.15, serum bicarbonate ion level <15 mmol/L, blood urea nitrogen level >17.8 mmol/L, or serum creatinine level ≥442 μmol/L).

Clinical variables associated with sepsis included age, sex, 30-day mortality, and length of intensive care unit (ICU) stay. Among these variables, MAP and blood oxygen saturation (SaO₂) were collected on admission. Glasgow Coma Scale score and Acute Physiology and Chronic Health Evaluation II (APACHE II) score were also collected on admission.

Statistical analysis

Demographic characteristics were categorized as continuous or categorical variables. Continuous variables were presented as mean ± standard deviation, whereas categorical variables were expressed as percentages (%). Comparisons between the severe and nonsevere anemia groups were performed using Student’s unpaired t-test for continuous variables and chi-squared test for categorical variables.

The least absolute shrinkage and selection operator (LASSO) method was used to select the optimal predictive features for severe anemia risk. Variables with nonzero coefficients in the LASSO regression model were selected.¹³

The LASSO regression model aims to minimize a loss function incorporating L1 regularization, as defined by the following objective function formula:

Objective function = \min_{β}

(\frac{1}{n} \sum_{i = 1}^{n} {(y_{i} - β_{0} - \sum_{j = 1}^{p} β_{j} x_{i j})}^{2} + λ \sum_{j = 1}^{p} | β_{j} |)

x_ij represents the mean squared error, which quantifies the difference between predicted and actual values. y_i is the L1 penalty, where λ ≥ 0 is the regularization coefficient controlling the strength of the penalty, and ∑∣β_j∣ is the L1 norm of the feature coefficients.

β₁, …, β_j−1, β_j+1, …, β_p denote the intercept term, β_p is the model coefficient for features, j = (1,2,…,p), n is the sample size, and p is the total number of features.

Regarding optimization algorithm, we used the coordinate descent method to solve the LASSO regression. This iterative approach involves fixing all coefficients except one and updating that single coefficient in each step. The process is repeated in a loop until convergence is reached. For instance, holding β₁,…,β_j−1,β_j+1,…., β_p as constant, we update β_j analytically by leveraging the sparsity-inducing property of the L1 penalty.

Then, multivariable logistic regression analysis was applied to construct a predictive model by incorporating the features selected in the LASSO regression model. Risk was expressed as an odds ratio with 95% confidence interval (CI) and p-values. Independent variables with p < 0.05 were considered statistically significant and included in the model.

Calibration curves were plotted to assess the accuracy of this nomogram. A significant statistical test implied that the model did not calibrate perfectly.¹⁴ To quantify the discrimination performance of this nomogram, the receiver operating characteristic (ROC) curve with Harrell’s C-index was measured. The nomogram was further subjected to bootstrapping validation to calculate a relatively corrected C-index. Decision curve analysis was then used to determine the clinical usefulness of the nomogram by quantifying the net benefit (NB) at different threshold probabilities in the sepsis cohort. The NB was calculated as previously described in the literature,¹⁵ using the following equation:

\begin{array}{l} NB = true positive (%) \\ \times benefit value (intervention) \\ - false positive (%) \\ \times loss value (intervention) \\ (i . e . NB = TB - FL) \end{array}

All statistical analyses were performed using R software (version 4.1.0) and IBM SPSS 22.

Results

Demographic characteristics

In accordance with the prevalence of sepsis-associated severe anemia in our hospital over the past 3 years, our model required at least 200 cases to ensure robustness of the nomogram. A total of 252 patients were enrolled from 1 April 2022 to 31 March 2023 in this study at our hospital, of which 106 patients were categorized into the severe anemia group and 146 into the nonsevere anemia group. Clinical and demographic characteristics of our study participants are presented in Table 1.

Table 1.

Demographic and clinical characteristics between severe and nonsevere anemia groups.

Demographic characteristics	Severe anemia	Nonsevere anemia	Total patients	p value
Demographic characteristics	(n = 106)	(n = 146)	(n = 252)	p value
Age (years)	58.83 ± 1.975	50.68 ± 1.945	54.11 ± 1.957	0.007
Sex				0.822
Male	52 (49.06%)	70 (47.95%)	122 (48.41%)
Female	54 (50.94%)	76 (52.05%)	130 (51.59%)
Mortality (30 days)	78 (73.58%)	74 (50.68%)	152 (60.32%)	0.013
Outcome (better)	32 (30.19%)	72 (49.32%)	104 (41.27%)	0.039
Length of ICU stay	10.09 ± 0.9184	6.584 ± 0.5381	8.063 ± 0.7283	0.003
Multiple infection sets	14 (13.21%)	10 (6.85%)	24 (9.52%)	0.287
Underlying diseases	72 (67.92%)	90 (61.64%)	162 (64.29%)	0.538
Antibody				0.503
Β-lactams	74 (69.81%)	92 (63.01%)	166 (65.87%)
Quinolones	32 (30.19%)	54 (36.99%)	86 (34.13%)
Mechanical ventilation (h)	73.72 ± 17.20	34.84 ± 4.844	51.19 ± 9.022	0.047
Duration of CRRT (h)	55.07 ± 7.956	33.12 ± 3.905	42.35 ± 6.030	0.027
Use of vasopressors	86 (81.13%)	120 (82.19%)	206 (81.75%)	0.985
CVPC	94 (88.68%)	136 (93.15%)	230 (91.27%)	0.449
APC	98 (92.45%)	118 (80.82%)	216 (85.71%)	0.150
ECMO	2 (1.89%)	10 (6.85%)	12 (4.76%)	0.205
PICCO	14 (13.21%)	14 (9.59%)	28 (11.11%)	0.612
Nutritional method (liquid food)	30 (28.30%)	86 (58.90%)	116 (46.03%)	0.001
Bacteriological culture (+)	60 (56.60%)	52 (35.62%)	112 (44.44%)	0.007
MAP	83.77 ± 2.510	81.89 ± 1.669	82.68 ± 2.090	0.497
SaO₂	96.21 ± 1.168	98.16 ± 0.414	95.79 ± 0.821	0.376
Oxygen route (Mechanical ventilation)	42 (39.62%)	58 (39.73%)	100 (39.68%)	0.762
Oxygen concentration (%)	53.42 ± 3.033	52.62 ± 2.506	52.95 ± 2.675	0.916
GCS score	9.623 ± 0.568	10.89 ± 0.493	10.36 ± 0.554	0.164
APACHE II score	25.95 ± 0.964	15.53 ± 0.809	19.91 ± 1.773	<0.001

APACHE II: Acute Physiology and Chronic Health Evaluation II; CRRT: continuous renal replacement treatment; GCS: Glasgow Coma Scale score; ICU: intensive care unit; MAP: mean arterial pressure; SaO₂: blood oxygen saturation; CVPC: central venous puncture catheterization; APC: arterial puncture catheterization; ECMO: extracorporeal membrane oxygenation; PICCO: pulse index continuous cardiac output.

The overall average age of the cohort was 54 years, ranging from 11 to 91 years. The mean age of the severe anemia group was significantly higher than that of the nonsevere anemia group (58.8 ± 2.0 vs. 50.7 ± 1.9 years; p = 0.007). The sex ratio was similar between the two groups. Mortality at 30 days was 60% for the overall cohort, and mortality was significantly higher in the severe anemia group (73.6% vs. 50.7%, p = 0.013). After standardized treatments, 41% of all patients in the cohort responded with a positive outcome, and the nonsevere anemia group responded positively to a greater extent than the severe anemia group (49.3% vs. 30.2%, p = 0.039). The average length of ICU stay was 8 days, ranging from 1 to 29 days. Participants in the severe anemia group had a significantly longer ICU stay than those in the nonsevere anemia group (10.1 ± 0.9 vs. 6.6 ± 0.5 days, p = 0.003).

The average mechanical ventilation time for all patients was 51 h, ranging from none to 718 h. The duration of mechanical ventilation in the severe anemia group was significantly longer than that in the nonsevere anemia group (73.7 ± 17.2 vs. 34.8 ± 4.8 h, p = 0.047), despite comparable mechanical ventilation use in the overall cohort. Additionally, the duration of CRRT was significantly longer in the severe anemia group than in the nonsevere anemia group (55.1 ± 8.0 vs. 33.1 ± 3.9 h, p = 0.027). The overall average CRRT duration for all participants was 42 h, ranging from 0 to 295 h. Moreover, the use of vasopressors was comparable between the two groups (p = 0.985). The two nutritional methods provided for all patients included liquid food and parenteral or enteral nutrition. There was significantly greater use of liquid food in the nonsevere anemia group than in the severe anemia group (58.9% vs. 28.3%, p = 0.001). The detection of positive bacteriological cultures of the blood, sputum, and secretions was significantly higher in the severe anemia group than in the nonsevere anemia group (56.6% vs. 35.6%, p = 0.007). Finally, the APACHE II score on admission, which ranged from 4 to 40, was significantly higher in the severe anemia group than in the nonsevere anemia group (26.0 ± 0.9 vs. 15.5 ± 0.8, p < 0.001).

Model feature selection

Among all clinical and demographic characteristics, 25 potential predictor features were reduced to 9 based on univariable analysis. The features included age, 30-day mortality, outcomes, length of ICU stay, mechanical ventilation, duration of CRRT, nutritional method, bacteriological culture, and APACHE II score. These nine features were then included in the multiple regression analysis.

All 25 features were included in the LASSO regression model (∼5:1 ratio, Figure 1(a) and (b)) with nonzero coefficients.

Figure 1.

LASSO binary logistic regression model for demographic and clinical feature selection. (a) Optimal parameter (lambda) selection using 5-fold cross-validation by minimum criteria. The binomial deviance curve was plotted versus log(lambda). Dotted vertical lines were drawn at the optimal values using the minimum criteria and (b) LASSO coefficient profiles of the 23 features. A coefficient plot was constructed against the log(lambda) sequence. A vertical line was drawn at the value selected through 5-fold cross-validation. LASSO: least absolute shrinkage and selection operator.

Construction of the individualized predictive nomogram

The results of the multivariable regression analyses are shown in Table 2. Age, length of ICU stay, nutritional method, and APACHE II score were determined to be independent predictors for sepsis-associated severe anemia. With aging, the risk of severe anemia increased (p = 0.016). Similarly, a longer length of ICU stay (p = 0.036) and a higher APACHE II score on admission (p < 0.001) were strong predictors of severe anemia. Compared with parenteral or enteral nutrition, liquid food was associated with reduced occurrence of severe anemia (p = 0.043). A model that incorporated the above-mentioned independent variables was then developed as the predictive nomogram (Figure 2).

Table 2.

Multiple logistical analysis for prediction factors of sepsis-associated severe anemia.

		Prediction model
Intercept and variable	β	Odds ratio(95% CI)	p value
Intercept	−3.8321	1.193 (0.941–1.607)	<0.001
Age	0.0130	1.044 (1.008–1.081)	0.016
Mortality (30 days)	−0.1114	1.224 (0.197–7.613)	0.829
Outcome	−0.4825	0.609 (0.102–3.653)	0.588
Length of ICU stay	0.1681	1.126 (1.008–1.258)	0.036
Mechanical ventilation (h)	−0.0011	0.996 (0.988–1.004)	0.319
Duration of CRRT (h)	0.0049	1.003 (0.989–1.018)	0.640
Nutritional method	−1.9354	0.322 (0.107–0.963)	0.043
Bacteriological culture (+)	0.5763	1.608 (0.521–4.961)	0.409
APACHE II score	0.8556	2.073 (1.394–3.082)	<0.001

APACHE II: Acute Physiology and Chronic Health Evaluation II; CI: confidence interval; CRRT: continuous renal replacement treatment; ICU: intensive care unit.

Figure 2.

Nomogram for sepsis-associated severe anemia. The sepsis-associated severe anemia nomogram was developed in the cohort, with age, length of ICU stay, nutritional method, APACHE II score incorporated. ICU: intensive care unit; APACHE II: Acute Physiology and Chronic Health Evaluation II.

Performance and clinical applicability of the predictive nomogram

The calibration curve of the risk nomogram demonstrated a good agreement for predicting severe anemia in this cohort of sepsis patients, regardless of the absolute ideal or bias-corrected model (Figure 3). The ROC curve presented a reasonable prediction (Figure 4), with a C-index of 0.885 (95% CI: 0.826–0.943). The C-index was conformed to 0.872 through bootstrapping validation, which suggested good discrimination.

Figure 3.

Calibration curves of the nomogram prediction for severe anemia in the cohort. The X-axis represents the predicted sepsis-associated severe anemia. The Y-axis represents the actual diagnosed severe anemia (B = 1000 repetitions, boot). The red line represents a perfect prediction by an ideal model. The green line presents the performance of the nomogram. The green line shows a closer fit to the red line, which represents a better prediction.

Figure 4.

ROC curve of the nomogram prediction model for severe anemia. The X-axis represents the “1-Specificity” of the nomogram. The Y-axis represents the “Sensitivity” of the nomogram. The blue line represents the nomogram prediction model, and the area under the curve represents the C-index. ROC: receiver operating characteristic.

Decision curve analysis of the nomogram is shown in Figure 5. This analysis showed that if the threshold probability of a patient’s risk of severe anemia was >5% and <75%, using this new nomogram to predict sepsis-associated severe anemia provided greater accuracy across the whole threshold range. Within this range, the NB was comparable on the basis of the risk nomogram.

Figure 5.

Decision curve analysis for sepsis-associated severe anemia. The Y-axis represents the net benefit. The red line represents the sepsis-associated severe anemia risk nomogram. The gray line represents the assumption that all severe anemia patients have been diagnosed. The black line represents the assumption that there are no patients with sepsis-associated severe anemia. The decision curve showed that if the threshold probability of a patient’s risk of severe anemia was >5% and <75%, using this new nomogram to predict sepsis-associated severe anemia added more benefit than the diagnosed-all-patients scheme or the diagnosed-none scheme.

Discussion

Nomograms have previously been demonstrated to be useful diagnostic and prognostic tools for predicting patient outcomes in various diseases, and they have been widely applied in many hospitals and clinics. Nomograms are associated with rapid computation through user-friendly digital interfaces, increased accuracy, and more intuitive presentation compared with conventional staging, which promote a better clinical decision-making environment for providers.¹⁶ Furthermore, nomograms have the ability to calculate severe variables as a numerical probability of the clinical outcomes, which can be customized to individual patients with a certain disease.¹⁷ Considering these superiorities, we established a nomogram to predict the risk of severe anemia in sepsis patients, with variables selected by LASSO and logistic regression analysis. To the best of our knowledge, this is the first predictive model of sepsis-associated anemia. In our model, essential parameters were included that can be objectively measured, are readily available, and are routinely monitored in each patient.¹⁸ This nomogram provided a relatively accurate prediction tool for the incidence of severe anemia among sepsis patients at our hospital. After stepwise selection, four variables, including age, length of ICU stay, nutritional method, and APACHE II score, were incorporated into an easy-to-use nomogram, which facilitated individualized prediction. Internal validation in the cohort demonstrated good discrimination and calibration. In particular, the high C-index in the internal validation confirmed that this nomogram could be widely applied in a large sample size.¹⁹

Sepsis induces a wide range of effects in RBCs, including a massive release of hemoglobin and changes in cell volume, metabolism, morphology, antioxidant status, and membrane protein composition.^3,5 Systemic inflammation during sepsis accelerates fragmentation of RBCs, leading to elevated concentrations of circulating CFH. CFH itself has been shown to independently increase endothelial injury and vascular permeability and contribute to organ dysfunction and death.²⁰ These effects on RBCs during sepsis induce anemia, and patients with sepsis-associated anemia have substantially worse clinical outcomes and increased mortalities compared with those without.²⁰ When the level of hemoglobin further decreases to <60 g/L, RBC transfusions are typically necessary to correct this severe anemia rapidly. In addition to more severe clinical outcomes, this complication is known to cause a significant consumption of financial and material resources. Therefore, a risk nomogram for predicting the incidence of severe anemia during sepsis is needed. We developed a valid risk prediction tool that can assist clinicians with early identification of patients at high risk of this serious complication.

Of the 252 sepsis patients included in the study, 42% developed severe anemia. Globally, sepsis continues to pose a major challenge, with substantial mortality despite advances in critical care. Historically, mortality rates have ranged from 30% to 50% or higher. Although improvements in early diagnosis and timely interventions have contributed to a decline in mortality rates, our study revealed a 30-day mortality rate of approximately 60%, which significantly exceeds the global averages.^1,21 Several factors may account for this elevated mortality. First, the study cohort predominantly comprised patients with severe sepsis syndromes, many of whom presented with hemodynamic instability requiring vasopressor support. Second, a high incidence of severe anemia likely contributed to worsened clinical outcomes. Third, as the leading tertiary care center in our region, our hospital frequently treats patients with infections caused by highly virulent and resistant pathogens, such as Klebsiella pneumoniae, multidrug-resistant Acinetobacter baumannii, and methicillin-resistant Staphylococcus aureus, which are associated with increased invasiveness and fatality. Despite appropriate fluid resuscitation and the administration of empiric antimicrobial therapy within 1 h of admission following blood culture collection, the overall mortality remained high, reflecting the severity and complexity of cases managed at our center.

All patients underwent routine blood tests daily; some patients already had severe anemia upon admission, while others progressively developed this complication during hospitalization. In the risk factor analysis, age, length of ICU stay, nutritional method, and APACHE II score were independently associated with severe anemia. The nomogram suggests that these four factors are keys to determining severe anemia risk in sepsis patients.

It is well known that age plays an important role in sepsis-associated anemia. In older adults, anemia can be attributed to nutritional deficiency, bleeding, or unexplained causes. Unexplained anemia might be explained by various factors, including reduced EPO activity and deteriorating hematopoietic function of bone marrow.²² The systemic proinflammatory state during sepsis further produces adverse effects on cells throughout the body, particularly in the hematological system. These effects include hypoferremia, progressive EPO resistance, elevated phagocytosis of erythrocytes by macrophages, and enhanced eryptosis due to oxidative stress, especially in senescent cells (which abundantly increase with aging). For these and other reasons, sepsis-associated anemia is more common in older patients even when preventive treatments are comprehensively applied. The length of ICU stay is another factor previously shown to be associated with a higher incidence of anemia,²³ which was confirmed in our study. It is undeniable that longer length of ICU stay during sepsis is a result of more serious underlying etiology; hence, it is not surprising that this variable is associated with anemia. Although it is likely that targeted anemia treatments (e.g. EPO, iron supplements, and blood transfusions) were applied whenever possible, the complication of severe anemia can greatly lengthen the entire hospital stay.

Nutritional method also seems to play a significant role in the development of anemia during sepsis. All participants in our study were divided into two groups based on nutritional support type: liquid food or fasting and water deprivation. The latter group was supported with parenteral or enteral nutrition to guarantee their daily dietary requirements. A majority of patients in the severe anemia group required absolute fasting and water deprivation and were supported by parenteral nutrition, and this proportion was significantly higher than that in the nonsevere anemia group. Although parenteral or enteral nutrition provides most of the essential daily energy and protein needs, these preparations cannot fully supply the nutrients required for hematopoiesis compared with liquid food. For example, iron, as an essential nutrient to promote the synthesis of RBCs, was previously shown as insufficient for those with total nutritional support.²⁴ Conversely, liquid food can provide some microelement carriers of iron, such as lactoferrin, which may help mitigate anemia, as previously reported.²⁵

Previous studies have shown that nonsurvivors present with significantly more severe anemia and higher APACHE II scores in sepsis. Sepsis patients with APACHE II scores above 14 were previously shown to be associated with higher risk of death.²⁶ In our study, APACHE II accounted for the largest proportion in the nomogram prediction model compared with age, length of ICU stay, and nutritional method. Sepsis patients with higher APACHE II scores invariably have more severe disease and higher levels of inflammatory factors, any of which might result in adverse effects on the hematopoietic system. Inflammatory reactions are known to trigger hemolysis and an increase in CFH levels, thus promoting increased endothelial permeability and cell apoptosis and accelerating the development of anemia.²⁰

Sepsis patients with severe anemia had worse outcomes than those without, demonstrating that an individualized tool to predict anemia risk might improve patient outcomes through more personalized interventions.²⁷ Additionally, it may serve as a guide for early interventions that might prevent sepsis-associated anemia.²⁸ For example, iron sucrose therapy could be applied for those with iron deficiency anemia; erythropoietin could be supplied for those with renal anemia; and folic acid and vitamin B12 could be provided for those with megaloblastic anemia. Use of a recognized prediction tool to rapidly identify severe anemia risk in sepsis patients could help stimulate implementation of low-cost prevention strategies, which should be encouraged in high-risk patients.²⁹

Having a reliable and accurate prognostic tool will assist physicians in determining the risk of severe anemia in sepsis, allowing more time for appropriate therapeutic strategies in each patient, precluding testing in low-risk situations, and promoting additional supportive treatments when there is a high probability of a favorable NB. In our cohort, a small subset of sepsis patients presented with severe anemia upon admission, while others developed severe anemia progressively during hospitalization.³⁰ Although routine blood tests measure hemoglobin levels, results often require >2 h, thus delaying treatments for patients who may benefit from earlier intervention. Furthermore, a significant number of patients did not present with anemia on admission, and conventional detection methods could not have effectively predicted the occurrence of severe anemia in these patients.³¹ Our nomogram provides a method of rapid prediction for severe anemia so that appropriate therapeutic strategies can be promptly implemented for prevention. Given that predicting the risk of severe anemia can be often difficult, suitable evaluations and multifaceted detection methods might be the most effective approach.

Our study has several limitations. First, as a retrospective study, it may be less reliable for drawing definitive conclusions. Additionally, the relatively small sample size (252 patients) limits the generalizability of our findings, which may only represent a subset of sepsis patients. Second, although univariable analysis identified nine clinical and demographic factors related to sepsis-associated severe anemia, only four remained independently predictive in the multivariable analysis after adjusting for confounders. Whether any of the remaining variables (e.g. duration of mechanical ventilation, CRRT) are significantly associated with anemia remains unclear and requires further investigation in larger patient cohorts. Finally, although the robustness of our nomogram was extensively evaluated through internal validation using bootstrapping, external validation was not performed. Therefore, the generalizability of these findings to the broader global sepsis population remains uncertain.

Conclusion

This study developed and tested a nomogram to assist clinicians in predicting the risk of severe anemia in sepsis patients at the time of admission to the hospital and at the onset of sepsis therapy. The nomogram offers a straightforward presentation, incorporating four readily available clinical indicators—age, length of ICU stay, nutritional method, and APACHE II score—which were identified as independent predictors of this complication. By having an estimate of individual risk for this complication in each patient, clinicians can take a more personalized approach in high-risk patients, either through more intensive monitoring or prophylactic therapies, to prevent severe anemia and thereby minimize the need for blood transfusions. External validation is needed to assess the sensitivity and specificity of this nomogram in future studies.

Footnotes

Acknowledgements

We thank the doctors, nurses, statistician, and all staff in the Department of Intensive Care Unit of Tongji Hospital Affiliated to Tongji Medical College of Huazhong University of Science and Technology who participated in this study.

Author contributions

LX and XR conceived and designed the study; LX performed the literature review and drafted the manuscript; LX and XR statistically analyzed the data. SL contributed to the critical revision of the manuscript. All authors read and approved the final manuscript.

Data availability statement

The data underlying this article will be shared on reasonable request to the corresponding author.

Declaration for conflicting interests

The authors declare no conflicting interests.

Funding

No funding support was received for this study.

ORCID iD

Shusheng Li

References

Chousterman

Swirski

Weber

GF.

Cytokine storm and sepsis disease pathogenesis. Semin Immunopathol 2017; 39: 517–528.

Weisberg

Park

Cherry-Bukowiec

JR.

Early goal-directed therapy: the history and ongoing impact on management of severe sepsis and septic shock. Surg Infect (Larchmt) 2018; 19: 142–146.

Bateman

Sharpe

Singer

, et al. The effect of sepsis on the erythrocyte. Int J Mol Sci 2017; 18: 1932.

Chan

Cheng

Shui

, et al. Erythrocyte degradation, metabolism, secretion, and communication with immune cells in the blood during sepsis: a review. Tzu Chi Med J 2022; 34: 125–133.

Effenberger-Neidnicht

Hartmann

Mechanisms of hemolysis during sepsis. Inflammation 2018; 41: 1569–1581.

Jiang

Kong

, 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: 67.

Weiss

Ganz

Goodnough

LT.

Anemia of inflammation. Blood 2019; 133: 40–50.

Holst

Haase

Wetterslev

; Scandinavian Critical Care Trials Groupet al. Lower versus higher hemoglobin threshold for transfusion in septic shock. N Engl J Med 2014; 371: 1381–1391.

Voigt

Hundeshagen

Malagaris

, et al. Effects of a restrictive blood transfusion protocol on acute pediatric burn care: transfusion threshold in pediatric burns. J Trauma Acute Care Surg 2018; 85: 1048–1054.

10.

Guo

Shen

Zhang

, et al. Prediction of postoperative pancreatic fistula using a nomogram based on the updated definition. Ann Surg Treat Res 2020; 98: 72–81.

11.

Von Elm

Altman

Egger

; STROBE Initiativeet al. The Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement: guidelines for reporting observational studies. Ann Intern Med 2007; 147: 573–577.

12.

Eber

Andres

Anemia. Differential diagnostics and fundamental therapy forms of moderate to severe anemia. Monatsschrift Kinderheilkunde 2016; 164: 59–72.

13.

Zhang

Yin

Zhou

, et al. Identification of ipsilateral supraclavicular lymph node metastasis in breast cancer based on LASSO regression with a high penalty factor. Front Oncol 2024; 14: 1349315.

14.

Yang

Lin

Yang

, et al. Advancing high-throughput MS-based protein quantification: a case study on quantifying 10 major food allergens by LC-MS/MS using a one-sample multipoint external calibration curve. J Agric Food Chem 2024; 72: 6625–6637.

15.

Zhao

Leng

, et al. Understanding decision curve analysis in clinical prediction model research. Postgrad Med J 2024; 100: 512–515.

16.

Balachandran

Gonen

Smith

, et al. Nomograms in oncology: more than meets the eye. Lancet Oncol 2015; 16: e173–e180.

17.

Wang

Chen

, et al. Development and external validation of a nomogram for predicting acute kidney injury in cardiogenic shock patients in intensive care unit. Int J Gen Med 2022; 15: 3965–3975.

18.

Wang

Yuan

Ding

, et al. Development and external validation of a prognostic model for survival of people living with HIV/AIDS initiating antiretroviral therapy. Lancet Reg Health West Pac 2021; 16: 100269.

19.

Wei

Champman

, et al. Beliefs about medicines and non-adherence in patients with stroke, diabetes mellitus and rheumatoid arthritis: a cross-sectional study in China. BMJ Open 2017; 7: e017293.

20.

Meegan

Shaver

Putz

, et al. Cell-free hemoglobin increases inflammation, lung apoptosis, and microvascular permeability in murine polymicrobial sepsis. Plos ONE 2020; 15: e0228727.

21.

Remick

DG.

Pathophysiology of sepsis. Am J Pathol 2007; 170: 1435–1444.

22.

Wacka

Nicikowski

Jarmuzek

, et al. Anemia and Its Connections to Inflammation in Older Adults: a Review. J Clin Med 2024; 13: 2049.

23.

Sinha

Hakendorf

Sobieraj-Teague

, et al. Anaemia in elderly patients at discharge from intensive care and hospital. VOX Sang 2021; 116: 1102–1105.

24.

Cai

Fan

Tao

, et al. 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: 300060520952275.

25.

Marques

Bourbon

Rodrigues

, et al. Lactoferrin as a carrier of iron: development and physicochemical characterization. FOOD Hydrocoll 2023; 142: 108772.

26.

Mohan

Shrestha

Guleria

, et al. Development of a mortality prediction formula due to sepsis/severe sepsis in a medical intensive care unit. Lung India 2015; 32: 313–319.

27.

Jansma

de Lange

Kingma

, et al. ‘Sepsis-related anemia’ is absent at hospital presentation; a retrospective cohort analysis. BMC Anesthesiol 2015; 15: 55.

28.

Nemeth

Ganz

Anemia of inflammation. Hematol Oncol Clin North Am 2014; 28: 671–681, vi.

29.

Maiden

Finnis

Peake

, et al. Haemoglobin concentration and volume of intravenous fluids in septic shock in the ARISE trial. Crit Care 2018; 22: 118.

30.

Lasocki

Pène

Ait-Oufella

, et al. Management and prevention of anemia (acute bleeding excluded) in adult critical care patients. Ann Intensive Care 2020; 10: 97.

31.

Carson

Grossman

Kleinman

, et al. Red blood cell transfusion: a clinical practice guideline from the AABB. Ann Intern Med 2012; 157: 49–58.

Development and validation of a predictive nomogram for sepsis-associated severe anemia

Abstract

Objective

Methods

Results

Conclusions

Keywords

Introduction

Patients and methods

Patients

Outcome measurements

Statistical analysis

Results

Demographic characteristics

Model feature selection

Construction of the individualized predictive nomogram

Performance and clinical applicability of the predictive nomogram

Discussion

Conclusion

Footnotes

Acknowledgements

Author contributions

Data availability statement

Declaration for conflicting interests

Funding

ORCID iD

References