Risk factor analysis and nomogram development for predicting 28-day mortality in elderly ICU patients with sepsis and type 2 diabetes mellitus

Abstract

Background: Type 2 diabetes mellitus (T2DM) significantly contributes to sepsis, with patients suffering from both conditions exhibiting greater severity and higher mortality rates compared to those with sepsis alone. Elderly individuals in the intensive care unit (ICU) are particularly prone to these comorbidities. A nomogram prediction model was developed to accurately assess prognosis and guide treatment for elderly patients with sepsis and T2DM. Methods: Data from 1489 patients with sepsis and T2DM in the Medical Information Mart for Intensive Care IV (MIMIC-IV) database were analyzed and categorized into 28-days survival (n = 1156) and 28-days death groups (n = 333). The dataset’s clinical characteristics were employed to create a nomogram predicting 28-days mortality in elderly ICU patients with sepsis and T2DM. The least absolute shrinkage and selection operator (LASSO) regression identified candidate predictors, followed by a multivariate logistic regression analysis incorporating variables with p < .05 into the final model. A nomogram was then constructed using these significant risk predictors. The model’s discriminatory power was evaluated through a receiver operating curve (ROC) and the area under the curve (AUC). Additionally, model performance was assessed using a calibration plot and the Hosmer-Lemeshow goodness-of-fit test (HL test), and clinical utility was examined via decision curve analysis (DCA). Results: Risk factors incorporated into the nomogram included age, ICU length of stay, mean blood pressure (MBP), metastatic solid tumor, Sequential Organ Failure Assessment (SOFA) score, Logistic Organ Dysfunction System (LODS) score, blood urea nitrogen (BUN), and vasopressor use. The predictive model demonstrated robust discrimination, with an AUC of 0.802 (95% CI 0.768–0.835) in the training dataset and 0.753 (95% CI 0.697–0.809) in the validation set. Calibration was confirmed with the HL test (p > .05), and DCA indicated clinical usefulness. Conclusion: This new nomogram serves as a practical tool for predicting 28-days mortality among elderly ICU patients with sepsis and T2DM. Optimizing treatment strategies based on this model could enhance 28-days survival rates for these patients.

Keywords

sepsis nomogram type 2 diabetes mellitus elderly

Introduction

Sepsis is a critical condition characterized by organ dysfunction arising from an unregulated host response, posing a life-threatening risk.¹ It frequently serves as the proximate cause of death for critically ill patients in the intensive care unit (ICU), remaining a leading contributor to global morbidity and mortality.² Annually, sepsis affects over three million patients worldwide, with mortality rates reaching 30%–50%, thus imposing significant burdens on both individuals and society.³ Sepsis typically presents with rapid onset and swift progression, affecting multiple organs and proving difficult to reverse, necessitating immediate treatment upon diagnosis.

The etiology of sepsis is complex and perilous, originating from various factors and impacting multiple organ functions. Diabetes, a globally escalating metabolic disorder, mirrors this trend. Its prevalence approaches pandemic levels due to lifestyle shifts and the widespread adoption of Western diets rich in processed foods.⁴ Patients with diabetes form a significant group at increased risk for post-sepsis complications and elevated mortality.⁵ Numerous indicators suggest that patients with diabetes^6–9 have a 2 to 6 times higher susceptibility to sepsis than their non-diabetic counterparts.¹⁰ Furthermore, diabetic individuals face higher sepsis-related morbidity and mortality compared to non-diabetic individuals.¹¹ Preclinical studies reveal that diabetes alters elements of the innate immune system and suppresses the adaptive immune system.¹² Particularly in type 2 diabetes, chronic inflammation, immune response suppression, and significant infection-related morbidity occur. In diabetes, inflammatory pathways are activated via toll-like receptors such as TLR2 and TLR4, along with indirect activation through TLR signaling.^13,14

Elderly patients often present to the ICU with complex and severe conditions, frequently leading to poor prognosis. There is controversy regarding the ICU admission of elderly patients, as studies indicate that over half of these admissions involve patients nearing the end of life.¹⁵ The incidence of sepsis increases with age, especially in the extremely elderly population (age ≥80 years), showing a strong association with significantly higher mortality in this demographic.¹⁶

Currently, no effective scale exists to assess the prognosis and condition of elderly patients with sepsis and T2DM, leading to treatment delays and additional challenges. The Sequential Organ Failure Assessment (SOFA) score and the Logistic Organ Dysfunction System (LODS) provide some utility in assessing patient conditions by focusing on organ physiological functions. However, these metrics lack specificity and sensitivity in evaluating clinical practices and prognoses for elderly patients with sepsis and T2DM. Consequently, these scoring systems are inadequate for guiding treatment in this patient population. To address the absence of an effective evaluation scale for predicting 28-days outcomes in elderly patients with sepsis and T2DM, this study analyzed risk factors for 28-days mortality using the Medical Information Mart for Intensive Care (MIMIC-IV) database. A nomogram prediction model was developed and compared with the SOFA score and LODS to thoroughly evaluate patient conditions, predict prognostic outcomes, and guide treatment for elderly patients with sepsis and T2DM.

Materials and methods

Study design

Research data were obtained from the MIMIC-IV database, which includes ICU patients from Beth Israel Deaconess Medical Center.^1,17 Structured Query Language (SQL) combined with Navicat Premium facilitated data retrieval, while R software was used for subsequent data processing. The MIMIC-IV database (version 2.2, https://mimic.physionet.org/) is publicly accessible and offers comprehensive clinical research information, including demographic characteristics, examination results, disease diagnoses, and treatment modalities. It also includes in-hospital and post-discharge patient outcomes. Following authorization from the institutional review boards of both Beth Israel Deaconess Medical Center (Boston, MA, USA) and the Massachusetts Institute of Technology (Cambridge, MA, USA), and approval to access the MIMIC-IV database, data extraction was conducted for our study. The database ensures absolute confidentiality by anonymizing patient identities and private information, thereby maintaining adherence to stringent privacy standards. The data analysis for this study was conducted from January 2024 to February 2024.

Patient admission and data extraction

The inclusion criteria were as follows¹: initial ICU admission,² diagnosis of both sepsis and T2DM, and³ age over 65 years. The exclusion criteria included¹: Sequential Organ Failure Assessment (SOFA) score <2 and² ICU stay duration of less than 24 h.

The Sepsis 3.0 Definition¹ was utilized, identifying patients with life-threatening organ dysfunction due to an uncontrolled host response to infection, with a SOFA score of ≥2. T2DM diagnosis was confirmed using the ninth edition of the International Classification of Diseases (ICD-9) codes applied to patient data in the MIMIC-IV database. Data extraction from the MIMIC-IV database was conducted using SQL with Navicat Premium. The sampling procedure is illustrated in Figure 1. Patients diagnosed with sepsis and T2DM based on ICD-9 codes were identified, and relevant data, including patient ID, SOFA score, and basic characteristics, were extracted. Ineligible patients (age <65 years, SOFA score <2, ICU stay <24 h) were excluded. Demographic data, vital signs, laboratory tests, comorbidities, scoring systems, and interventions were extracted using patient HADM_ID and ICUSTAY_ID. Demographic information included age, sex, and length of ICU stay. Vital signs encompassed heart rate (HR), systolic blood pressure (SBP), diastolic blood pressure (DBP), mean blood pressure (MBP), and oxygen saturation (SpO2). Laboratory tests included hematocrit (HCT), hemoglobin (HB), platelet count (PLT), white blood cell count (WBC), anion gap (AG), standard bicarbonate (SB), chloride, blood urea nitrogen (BUN), creatinine, sodium, and potassium. Comorbidities included congestive heart failure, myocardial infarction, peripheral vascular disease, cerebrovascular disease, chronic pulmonary disease, rheumatic disease, peptic ulcer disease, mild liver disease, renal disease, malignant cancer, severe liver disease, metastatic solid tumor, acquired immune deficiency syndrome (AIDS), and acute kidney injury (AKI). Scoring systems included the simplified acute physiology score II (SAPS II), SOFA score, logistic organ dysfunction system (LODS), and Charlson comorbidity index. Interventions included active dialysis, ventilation, vasopressin, epinephrine, norepinephrine, and neuroblock. Vital signs and laboratory data from blood examinations, basic blood biochemical indexes, and arterial blood gas were collected during the initial 24 h of ICU stay. If a variable was recorded multiple times, the most severe value was used.

Figure 1.

Flow chart of patient selection (MIMIC-IV, Medical Information Mart for Intensive Care I).

Statistical analysis

Continuous variables were summarized using the median and interquartile range, with the Wilcoxon rank-sum test applied for between-group comparisons. Categorical variables were presented as frequencies and proportions, and intergroup comparisons utilized Chi-square tests or Fisher’s exact test, as appropriate. Internal validation followed the bootstrap method, adhering to the Transparent Reporting of a Multivariable Prediction Model for Individual Prognosis or Diagnosis (TRIPOD) guidelines.¹⁸ Data preprocessing involved handling missing data with the Multiple Imputation by Chained Equations (MICE) R package¹⁹ using the neighbor interpolation method. We discarded data with missing values >20% and performed multiple imputation by chained equations (MICE) for data with missing values ≤20%. Sensitivity analyses were performed to ensure that the imputation did not significantly affect our results. Post-imputation, the least absolute shrinkage and selection operator (LASSO) method identified predictors of 28-days mortality .²⁰ Final candidate variables were determined based on cross-validation results, selecting lambda = min.²¹ A multivariate logisticregression model was then constructed using these variables, with multicollinearity assessed through variable inflation factors (VIF).

Model performance was evaluated using the area under the receiver operating curve (AUC), sensitivity, and specificity, with the optimal cutoff point determined by the Youden index. A nomogram was generated using the R package “regplot.” Consistency between observed and predicted outcomes was assessed using the calibration C index (based on 1000 bootstrap resamplings),²² calibration curve (illustrating the relationship between observed and predicted probabilities), Hosmer-Lemeshow goodness-of-fit test (HL test), and Brier score. Decision curve analysis (DCA) evaluated the net clinical benefit.²³ All statistical analyses were conducted using R (version 4.2.1), with a significance level of p < .05.

Results

Patient characteristics

A total of 1489 patients were included in the study, with an average age of 76.13 years. Females comprised 43% (n = 637) of the cohort, with a mortality rate of 22.4%, while males accounted for 57% (n = 852), with a mortality rate of 22.41%. Table 1 contrasts the characteristics of the 28-days death group with those of the survival group. Patients in the 28-days death group were older and had longer ICU stays compared to those in the survival group. Additionally, higher levels of WBC, AG, BUN, creatinine, potassium, and HR were observed in the death group. The Charlson Comorbidity Index, LODS, SOFA, and SAPS II scores were significantly higher, with a higher prevalence of AKI, myocardial infarction, congestive heart failure, renal disease, severe liver disease, and metastatic solid tumors. The usage of dialysis, ventilation, vasopressin, epinephrine, norepinephrine, and neuroblock was significantly higher, indicating poor hemodynamics.

Table 1.

Baseline characteristics of 28-days survival group and death groups.

Variables	Total (n = 1489)	28-days survival group (n = 1156)	28-days death group (n = 333)	p-value
Age, years	76.13 [70.73, 82.51]	75.59 [70.55, 81.85]	78.2 [72.23, 85.34]	<0.001
Female, n (%)	637 (43.78%)	495 (42.82%)	142 (42.64%)	1.000
Los icu, days	3.04 [1.69, 6.30]	2.83 [1.61, 5.98]	4.07 [1.98, 7.46]	<0.001
HCT, %	32.00 [27.00, 36.90]	32.00 [27.10, 36.60]	32.10 [27.00, 37.60]	0.558
HB, g/L	10.10 [8.50, 11.90]	10.10 [8.60, 11.90]	10.00 [8.40, 11.90]	0.523
PLT, K/μL	183.00 [132.00, 249.00]	183.00 [134.00, 248.00]	184.00 [129.00, 253.00]	0.830
WBC, K/μL	11.80 [8.50, 16.40]	11.50 [8.40, 15.90]	12.40 [8.70, 17.30]	0.026
AG, mmol/L	16.00 [14.00, 20.00]	16.00 [13.00, 19.00]	18.00 [14.00, 22.00]	<0.001
SB, mmol/L	21.00 [19.00, 24.00]	22.00 [19.00, 24.00]	20.00 [17.00, 23.00]	<0.001
BUN, mg/dl	28.00 [18.00, 46.00]	26.00 [17.00, 41.00]	39.00 [25.00, 58.00]	<0.001
Chloride, mEq/L	102.00 [98.00, 106.00]	102.00 [98.00, 107.00]	101.00 [96.00, 105.00]	0.002
Creatinin, mg/dL	1.30 [1.00, 2.10]	1.3.00 [0.90, 1.90]	1.70 [1.20, 2.70]	<0.001
Sodium, mEq/L	138.00 [135.00, 141.00]	138.00 [135.00, 141.00]	138.00 [134.00, 141.00]	0.193
Potassium, mEq/L	4.40 [4.00, 5.00]	4.40 [3.90, 4.90]	4.50 [4.10, 5.30]	<0.001
HR, beats/min	86.00 [76.00, 100.00]	85.00 [75.75, 100.00]	89.00 [77.00, 102.00]	0.036
SBP, mmHg	121.00 [105.00, 139.00]	121.00 [105.75, 139.00]	120.00 [103.00, 139.00]	0.419
DBP, mmHg	64.00 [53.00, 76.00]	63.00 [53.00, 76.00]	65.00 [53.00, 78.00]	0.433
MBP, mmHg	80.00 [69.00, 92.00]	80.00 [69.00, 91.25]	80.00 [69.00, 95.00]	0.662
SPO₂, mmHg	98.00 [95.00, 100.00]	98 [95.00, 100.00]	97.00 [94.00, 100.00]	0.051
AKI, n (%)	1166 (78.31%)	864 (74.74%)	302 (90.69%)	<0.001
Myocardial infarct, n (%)	448 (30.09%)	321 (27.77%)	127 (38.14%)	<0.001
Congestive heart failure, n (%)	713 (47.88%)	524 (45.33%)	189 (56.76%)	<0.001
Peripheral vascular disease, n (%)	184 (12.36%)	141 (12.20%)	43 (12.91%)	0.799
Cerebrovascular disease, n (%)	277 (18.60%)	204 (17.65%)	73 (21.92%)	0.092
Chronic pulmonary disease, n (%)	422 (28.34%)	333 (28.81%)	89 (26.73%)	0.501
Rheumatic disease, n (%)	51 (3.43%)	42 (3.63%)	9 (2.70%)	0.515
Peptic ulcer disease, n (%)	56 (3.76%)	39 (3.37%)	17 (5.11%)	0.194
Mild liver disease, n (%)	136 (9.13%)	99 (8.56%)	37 (11.11%)	0.189
Renal disease, n (%)	662 (44.46%)	492 (42.56%)	170 (51.05%)	0.007
Malignant cancer, n (%)	206 (13.83%)	150 (12.98%)	56 (16.82%)	0.089
Severe liver disease, n (%)	60 (4.03%)	39 (3.37%)	21 (6.31%)	0.025
Metastatic solid tumor, n (%)	90 (6.04%)	51 (4.41%)	39 (11.71%)	<0.001
Dialysis active, n (%)	156 (10.48%)	92 (7.96%)	64 (19.22%)	<0.001
Ventilation, n (%)	678 (45.53%)	483 (41.78%)	195 (58.56%)	<0.001
Vasopressin, n (%)	216 (14.51%)	99 (8.56%)	117 (35.14%)	<0.001
Epinephrine, n (%)	150 (10.07%)	95 (8.22%)	55 (16.52%)	<0.001
Norepinephrine, n (%)	599 (40.23%)	396 (34.26%)	203 (60.96%)	<0.001
Neuroblock, n (%)	35 (2.35%)	16 (1.38%)	19 (5.71%)	<0.001
Charlson comorbidity index	8.00 [6.00, 10.00]	8.00 [6.00, 9.00]	9.00 [7.00, 10.00]	<0.001
SAPSII	42.00 [35.00, 51.00]	41.00 [34.00, 49.00]	50.00 [40.00, 58.00]	<0.001
LODS	5.00 [3.00, 7.00]	5.00 [3.00, 7.00]	7.00 [5.00, 9.00]	<0.001
SOFA	6.00 [4.00, 9.00]	6.00 [4.00, 8.00]	9.00 [6.00, 12.00]	<0.001

HR, heart rate; SBP, systolic blood pressure; DBP, diastolic blood pressure; MBP, mean blood pressure; RR, respiratory rate; BUN, blood urea nitrogen; SAPSII, simplified acute physiology score II; SOFA, sequential organ failure; LODS, logistic organ dysfunction system; HCT, hematocrit; HB, hemoglobin; PLT, Platelet; WBC, white blood cell (WBC); AG, aniongap; SB, standard bicarbonate; AKI, acute kidney injury.

Characteristics selection and development of a nomogram

Out of 43 variables, 10 were retained in the LASSO logistic regression model based on the binomial deviance minimum criteria (ratio 5:1) (Figure 2). The multivariable logistic analysis of these 10 variables is presented in Table 2. Eight predictors were included in the final multivariable logistic model: Age (OR: 1.055; 95% CI 1.035–1.076), ICU length of stay (OR: 0.962; 95% CI 0.940-0.984), metastatic solid tumor (OR: 2.902; 95% CI 1.594–5.283), SOFA score (OR: 1.185; 95% CI 1.121–1.253), vasopressin usage (OR: 2.974; 95% CI 1.998–4.427), LODS (OR: 1.075; 95% CI 1.012–1.142), mean blood pressure (OR: 1.013; 95% CI 1.006–1.021), and BUN (OR: 1.007; 95% CI 1.002-1.012). Based on this model, a nomogram was constructed to predict 28-days mortality in elderly ICU patients with sepsis and T2DM (Figure 3).

Figure 2.

Demographic and clinical feature selection using the LASSO logistic regression model. (A) LASSO coefficient profiles of the radiomic features. A coefficient profile plot was plotted versus the log (λ). Each colored line represents the coefficient of each feature. (B) Tuning parameter (λ) selection using LASSO penalized logistic regression with 10-fold cross-validation.

Table 2.

Multivariate logistic regression model of elderly patients with sepsis and T2DM in the training set.

Variables	Multivariable logistics model
Variables	OR	95%CI	p Value
Age	1.055	1.035-1.076	<0.001
Los icu	0.962	0.940-0.984	<0.001
Metastatic solid tumor	2.902	1.594-5.283	<0.001
SOFA	1.185	1.121-1.253	<0.001
Vasopressin	2.974	1.998-4.427	<0.001
BUN	1.007	1.002-1.012	0.010
Charlson comorbidity index	1.058	0.986-1.135	0.117
LODS	1.075	1.012-1.142	0.020
MBP	1.013	1.006-1.021	<0.001

SOFA, sequential organ failure; LODS, logistic organ dysfunction system; MBP, mean blood pressure; BUN, blood urea nitrogen.

Figure 3.

Nomogram to predicting 28-days mortality in ICU elderly patients with sepsis and Type 2 Diabetes Mellitus. SOFA, sequential organ failure; LODS, logistic organ dysfunction system; MBP, mean blood pressure; BUN, blood urea nitrogen; ** means p < .01, ***means p < .001.

Apparent performance of the nomogram

The AUC for the model was 0.802 (95% CI 0.768–0.835) for the training set and 0.753 (95% CI 0.697–0.809) for the validation set (Figure 4(a) and 4(b)). The optimal cutoff values for the nomogram-predicted probability were set at 25.8% in the training set and 26.7% in the validation set, based on the maximum Youden index. Sensitivity, specificity, PPV, and NPV were 64.9%, 80.0%, 48.2%, and 89.3% in the training set, and 63.0%, 79.6%, 49.6%, and 87.1% in the validation set. The ROC values for the SOFA and LODS scoring systems incorporated into the regression were 0.712 and 0.709 in the training set, and 0.736 and 0.648 in the validation set, respectively, indicating that the model demonstrated superior predictive performance compared to commonly used clinical scoring systems. The calibration plot showed a good fit for the nomogram in both cohorts (Figure 5(a) and (b)). Additionally, the HL test of multivariable analysis demonstrated perfect consistency between predicted and observed values (training set, χ2 = 7.713, p = .462; validation set, χ2 = 5.986, p = .469).

Figure 4.

ROC curve and AUROC of SOFA, nomogram and LODS in training set (A) and validation set (B). The AUROC of nomogram is bigger than it of SOFA and LODS in both training set and validation set.

Figure 5.

Calibration curves of the predicted nomogram in the training set (A) and validation set (B). In both sets, the apparent curve and bias-corrected curve slightly deviated from reference line, but a good conformity between observation and prediction is observed.

Clinical practice

DCA for the nomogram was performed on both the training and validation sets (Figure 6(a) and (b)). In the figure, the gray curve represents the scenario where all patients receive intervention, the straight black line indicates no patients receive intervention, and the red curve illustrates the clinical benefit provided by our model. Predicted probability thresholds were set at 10%–89% for the development cohort and 13%–85% for the validation cohort, with net benefits ranging from 0% to 24% and 0%–19%, respectively. Notably, as the threshold decreases, the net benefit also decreases. Our results indicate that the model provides superior net clinical benefit compared to the SOFA and LODS scoring systems.

Figure 6.

The DCA curve of medical intervention in patients with the nomogram, SOFA, and LODS in the training et (A) and validation set (B).

Discussion

Since the clinical definition of sepsis was established, physicians have faced significant challenges in diagnosing and treating this condition, largely due to its high mortality rates, reduced quality of life, and substantial economic burdens on patients.²⁴ Sepsis primarily induces systemic organ damage through an inflammatory response, with capillary endothelial damage and loss of peripheral circulation regulation being critical processes that can lead to septic shock. Physiopathological factors such as leukocyte extravasation, coagulation dysfunction, and inflammation-induced capillary dilation are major contributors to widespread tissue edema.²⁵ A nationwide observational study of over 500,000 individuals with type 2 diabetes revealed that type 2 diabetes is an independent risk factor for mortality in sepsis.²⁶ Given its increased prevalence²⁷ and association with impaired host defense against infections,⁸ diabetes is a particularly significant risk factor for sepsis. The global incidence of diabetes is rising, with projections indicating it will affect over 500 million individuals by 2030, most of whom will develop type 2 diabetes.²⁸ Elderly patients, characterized by reduced immunity,²⁹ decreased organ reserve function, and a higher prevalence of comorbidities such as diabetes and coronary heart disease compared to younger individuals,³⁰ often present atypical clinical symptoms post-infection. This makes the diagnosis prone to oversight or misdiagnosis, leading to the rapid progression of sepsis to multiple organ failure³¹ and subsequently high clinical mortality rates. Additionally, findings from a prospective cohort study indicate that older survivors of sepsis endure a greater burden of persistent disability and a higher 12-months mortality rate compared to younger patients.³²

Despite the scarcity of studies examining the risk factors influencing the prognosis of elderly patients with sepsis and T2DM, developing a nomogram to assist clinicians in identifying high-risk patients and understanding the impact of various risk factors on the 28-days prognosis in this population is essential. The predictive nomogram includes eight predictors: age, length of stay in the intensive care unit (LOS ICU), metastatic solid tumor, Sequential Organ Failure Assessment (SOFA), vasopressin, blood urea nitrogen (BUN), Logistic Organ Dysfunction System (LODS), and mean blood pressure (MBP). The model demonstrated robust performance, with an area under the curve (AUC) of 0.802 in the training set and 0.753 in the validation set. Conversely, the receiver operating characteristic (ROC) values for the SOFA and LODS scoring systems included in the regression were 0.712 and 0.709 in the training set and 0.736 and 0.648 in the validation set, respectively. These results indicate that the model exhibits superior predictive performance compared to commonly used clinical scoring systems. Decision curve analysis (DCA) further validates and endorses the use of this model.^33,34 DCA evaluates prediction models by calculating the clinical net benefit. Findings suggest that interventions guided by the nomogram yield more favorable prognoses in both the training and validation sets compared to those guided by the SOFA score and LODS. The calibration of the prediction model was assessed using the Hosmer–Lemeshow test and calibration plots,³⁵ both of which demonstrated satisfactory calibration ability. Elderly patients with sepsis and T2DM exhibit higher 28-days mortality, and their clinical treatments pose greater challenges. Current sepsis guidelines do not recommend specific treatment methods and evaluation programs for patients with different infection sources and complications. Several systems are available for clinically evaluating patients with sepsis, such as the SOFA score, qSOFA, SAPSII, LODS, APACHE III, and APACHE IV. The SOFA score and qSOFA are primarily used for early and rapid disease assessments of patients with sepsis.³⁶ The Logistic Organ Dysfunction System (LODS) provides an objective tool for assessing the severity levels of organ dysfunction in the ICU,³⁷ with each variable in the LODS score screened and weighted by logistic regression. Research indicates that the LODS score exhibits higher accuracy than the qSOFA score and SOFA score.³⁸ However, the prediction efficacy for elderly patients with sepsis and T2DM remains uncertain.

Among the critical variables for predicting sepsis in elderly patients with both sepsis and T2DM, the most influential factors are the administration of vasopressors, mean blood pressure (MBP), blood urea nitrogen (BUN), Logistic Organ Dysfunction System (LODS), and Sequential Organ Failure Assessment (SOFA) scores following admission to the intensive care unit (ICU). These parameters are essential indicators for assessing circulatory disorders, respiratory complications, and other organ dysfunctions in elderly individuals with sepsis and T2DM.³⁹ The integration of vasopressors and MBP into scoring systems such as SOFA, LODS, SAPSII, and APSIII underscores their significance in disease prediction.⁴⁰ For instance, existing research has shown that BUN is associated with the 28-days mortality risk in sepsis.³³ Additionally, the inclusion of metastatic solid tumor as a new variable is significant. Intensive cytotoxic chemotherapy reduces the absolute neutrophil count in malignancies or solid tumors, leading to decreased patient survival rates.⁴¹ The presence of metastatic solid tumors is closely linked to the immune status of each patient. Numerous studies have emphasized that immunocompromised individuals exhibit significantly higher incidence and mortality rates of sepsis compared to their immunocompetent counterparts.^42,43 Cancer, by employing the immune checkpoint pathway and releasing immunosuppressive factors, can effectively dampen both innate and adaptive immune responses in the host.⁴⁴ Furthermore, research has established a correlation between Type 2 diabetes and an increased susceptibility to various forms of cancer.^44–46 The immune system dysfunction shared by tumors and sepsis is associated with decreased survival rates in elderly patients with sepsis.⁴⁷

The qSOFA score was not included in our study. While qSOFA is extensively utilized in emergency and general wards due to its simplicity and convenience, the SOFA score offers multiple advantages. Primarily, the SOFA score provides superior multi-organ function assessment, evaluating six major organ systems—respiratory, cardiovascular, hepatic, coagulation, renal, and neurological—thereby delivering a comprehensive overview of a patient’s condition. This detailed scoring of each system’s functional impairment results in a more thorough evaluation of the patient’s health status. Conversely, the qSOFA score assesses only three parameters: respiratory rate, altered mental status, and systolic blood pressure, which, though straightforward, do not fully capture the status of multiple organ functions. Additionally, in ICU and dynamic monitoring contexts, the SOFA score proves more effective.^48–50 Initially designed for ICU environments to monitor multi-organ failure, the SOFA score is both more suitable and clinically valuable for managing ICU patients.⁵¹ The qSOFA score, primarily a rapid screening tool for non-ICU settings, shows limited effectiveness in the ICU. Given that this study focuses on ICU patient evaluation, the SOFA score holds greater relevance. Furthermore, the SOFA score enables dynamic monitoring of organ function changes during treatment, aiding physicians in adjusting treatment plans and enhancing outcomes. The qSOFA score, typically used for initial screening, lacks the capability for continuous monitoring.

Several limitations of this study must be acknowledged. Firstly, the diverse diagnostic criteria for sepsis, such as the Martin criteria, necessitate further validation of our model’s effectiveness using these different benchmarks. Additionally, the study’s retrospective observational design inherently limits the ability to establish causal relationships between the identified predictors and patient outcomes. The use of the MIMIC-IV database, with its specific inclusion criteria, resulted in the exclusion of certain populations, potentially limiting the generalizability of our model. While the MIMIC-IV database provides a wealth of high-quality clinical data, it also presents potential selection biases and demographic limitations, as the data primarily originate from a single large academic medical center in Boston, USA, which may restrict the applicability of our findings to other regions or healthcare settings. Consequently, our results may not be fully generalizable to broader populations. Future studies should utilize multicenter data to validate the generalizability and robustness of our model. Furthermore, data limitations prevented the inclusion of certain conventional factors known to influence sepsis outcomes, such as ventilation status, neurological symptoms, cardiac injury biomarkers, procalcitonin levels, neutropenia, and inappropriate empirical antibiotic therapy. The absence of these variables may affect the comprehensiveness and precision of our nomogram. Future research should aim to integrate these additional factors to enhance the model’s accuracy and comprehensiveness. LASSO regression and logistic regression models were selected for prediction analysis due to their high interpretability in clinical settings. Although more advanced machine learning techniques might offer improved predictive performance, they often lack the transparency and interpretability crucial for clinical decision-making. Therefore, the choice of models balances predictive performance with clinical interpretability. The study demonstrated the model’s robustness through internal validation. However, to comprehensively evaluate the robustness and effectiveness of the nomogram, future investigations should include external validation with independent datasets from various medical institutions and regions. This approach will ensure the model’s effectiveness in a wider range of clinical settings. Lastly, our nomogram prediction model is a valuable tool for assessing the 28-days mortality risk among ICU elderly patients with sepsis and T2DM. Clinicians can utilize this model for more accurate patient prognosis evaluations and personalized treatment plans, including initial risk assessment upon ICU admission, identifying high-risk patients for timely intervention, and dynamically adjusting treatment strategies to improve outcomes. Additionally, this model can aid in the efficient allocation of medical resources and optimize ICU management. By acknowledging these limitations and areas for future research, caution is emphasized when interpreting our results, and further studies are recommended to address these issues and validate our model in diverse clinical settings.

Conclusion

The newly developed nomogram, which includes age, length of stay in the intensive care unit (LOS ICU), metastatic solid tumor, Sequential Organ Failure Assessment (SOFA), blood urea nitrogen (BUN), Logistic Organ Dysfunction System (LODS), mean blood pressure (MBP), and vasopressor usage, effectively predicts 28-days mortality in elderly ICU patients with sepsis and Type 2 Diabetes Mellitus. Tailoring treatment strategies to address the identified factors within the model has the potential to improve the 28-days survival rates for this specific cohort of ICU patients.

Footnotes

Acknowledgement

We thank Bullet Edits Limited for the linguistic editing and proofreading of the manuscript. This research was supported by National Natural Science Foundation of China (81672526, 81802560, 82403182), and Special Program for Clinical Research of Shanghai Municipal Health Commission (2018470263, 20184Y0105) and Key Discipline of Tongji Hospital in Shanghai (ZD24-MN) and A nucleotide targeting lncRNA ZNF518A for AR-V7-mediated castration-resistant prostate cancer (ZLZH2202). This data used in this study are available at the database:

Author contributions

H.L.; Y.Z.: Designed the protocol, write the manuscript. X.W.; Q.W.; T.Z.; X.Q.; Y.Z.; W.J.; C.X.: Collected and analyzed the data. X.C.; G.W.: reviewed and edit the manuscript.

Declaration of conflicting interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This research was supported by National Natural Science Foundation of China (81672526, 81802560, 82403182), and Special Program for Clinical Research of Shanghai Municipal Health Commission (2018470263, 20184Y0105) and Key Discipline of Tongji Hospital in Shanghai (ZD24-MN) and A nucleotide targeting lncRNA ZNF518A for AR-V7-mediated castration-resistant prostate cancer (ZLZH2202).

ORCID iD

Haopeng Li

References

Evans

Rhodes

Alhazzani

, et al. Surviving sepsis campaign: international guidelines for management of sepsis and septic shock 2021. Intensive Care Med 2021; 47(11): 1181–1247.

Srzić

Nesek Adam

Tunjić Pejak

. Sepsis definition: what's new in the treatment guidelines. Acta Clin Croat 2022; 61(Suppl 1): 67–72.

Zampieri

Bagshaw

Semler

. Fluid therapy for critically ill adults with sepsis: a review. JAMA 2023; 329(22): 1967–1980.

Worldwide trends in diabetes since 1980: a pooled analysis of 751 population-based studies with 4.4 million participants. Lancet. 2016;387(10027):1513-1530.

Frydrych

Fattahi

, et al. Diabetes and sepsis: risk, recurrence, and ruination. Front Endocrinol 2017; 8: 271.

Jiang

Cheng

. Impact of diabetes mellitus on outcomes of patients with sepsis: an updated systematic review and meta-analysis. Diabetol Metab Syndr 2022; 14(1): 39.

Koh

Peacock

van der Poll

, et al. The impact of diabetes on the pathogenesis of sepsis. Eur J Clin Microbiol Infect Dis 2012; 31(4): 379–388.

Schuetz

Castro

Shapiro

. Diabetes and sepsis: preclinical findings and clinical relevance. Diabetes Care 2011; 34(3): 771–778.

Wang

Ren

Wang

, et al. Association between diabetes mellitus and outcomes of patients with sepsis: a meta-analysis. Med Sci Monit 2017; 23: 3546–3555.

10.

Laupland

Gregson

Zygun

, et al. Severe bloodstream infections: a population-based assessment. Crit Care Med 2004; 32(4): 992–997.

11.

Frydrych

Bian

O'Lone

, et al. Obesity and type 2 diabetes mellitus drive immune dysfunction, infection development, and sepsis mortality. J Leukoc Biol 2018; 104(3): 525–534.

12.

Carlos

Spiller

Souto

, et al. Histamine h2 receptor signaling in the pathogenesis of sepsis: studies in a murine diabetes model. J Immunol 2013; 191(3): 1373–1382.

13.

Jialal

Kaur

. The role of toll-like receptors in diabetes-induced inflammation: implications for vascular complications. Curr Diab Rep 2012.

14.

Sepehri

Kiani

Nasiri

, et al. Toll-like receptor 2 and type 2 diabetes. Cell Mol Biol Lett 2016; 21: 2.

15.

Ritzema

Young

. Admitting elderly patients to ICU: is it just about survival? N Z Med J 2017; 130(1453): 8–10.

16.

Boonmee

Ruangsomboon

Limsuwat

, et al. Predictors of mortality in elderly and very elderly emergency patients with sepsis: a retrospective study. West J Emerg Med 2020; 21(6): 210–218.

17.

Yang

Liu

, et al. Brief introduction of medical database and data mining technology in big data era. J Evid Based Med 2020; 13(1): 57–69.

18.

Collins

Reitsma

Altman

, et al. Transparent reporting of a multivariable prediction model for individual prognosis or diagnosis (TRIPOD): the TRIPOD statement. Bmj 2015; 350: g7594.

19.

Beretta

Santaniello

. Nearest neighbor imputation algorithms: a critical evaluation. BMC Med Inform Decis Mak 2016; 16 Suppl 3(Suppl 3): 74.

20.

Friedman

Hastie

Tibshirani

. Regularization paths for generalized linear models via coordinate descent. J Stat Softw 2010; 33(1): 1–22.

21.

Zhou

Liu

, et al. Margin-based pareto ensemble pruning: an ensemble pruning algorithm that learns to search optimized ensembles. Comput Intell Neurosci 2019; 2019: 7560872.

22.

Harrell

Jr. Califf

Pryor

, et al. Evaluating the yield of medical tests. JAMA 1982; 247(18): 2543–2546.

23.

Vickers

Cronin

Elkin

, et al. Extensions to decision curve analysis, a novel method for evaluating diagnostic tests, prediction models and molecular markers. BMC Med Inform Decis Mak 2008; 8: 53.

24.

Prescott

Langa

Liu

, et al. Increased 1-year healthcare use in survivors of severe sepsis. Am J Respir Crit Care Med 2014; 190(1): 62–69.

25.

Liu

Huang

Sun

, et al. Sepsis-induced immunosuppression: mechanisms, diagnosis and current treatment options. Mil Med Res 2022; 9(1): 56.

26.

Balintescu

Lind

Franko

, et al. Glycemic control and risk of sepsis and subsequent mortality in type 2 diabetes. Diabetes Care 2022; 45(1): 127–133.

27.

Cho

Shaw

Karuranga

, et al. IDF Diabetes Atlas: global estimates of diabetes prevalence for 2017 and projections for 2045. Diabetes Res Clin Pract 2018; 138: 271–281.

28.

Saeedi

Petersohn

Salpea

, et al. Global and regional diabetes prevalence estimates for 2019 and projections for 2030 and 2045: results from the International Diabetes Federation Diabetes Atlas, 9(th) edition. Diabetes Res Clin Pract 2019; 157: 107843.

29.

Müller

Di Benedetto

Pawelec

. The immune system and its dysregulation with aging. Subcell Biochem 2019; 91: 21–43.

30.

Zhang

Huang

, et al. Prediction of prognosis in elderly patients with sepsis based on machine learning (random survival forest). BMC Emerg Med 2022; 22(1): 26.

31.

Pool

Gomez

Kellum

. Mechanisms of organ dysfunction in sepsis. Crit Care Clin 2018; 34(1): 63–80.

32.

Mankowski

Anton

Ghita

, et al. Older sepsis survivors suffer persistent disability burden and poor long-term survival. J Am Geriatr Soc 2020; 68(9): 1962–1969.

33.

Ren

Zhang

, et al. Risk factor analysis and nomogram for predicting in-hospital mortality in ICU patients with sepsis and lung infection. BMC Pulm Med 2022; 22(1): 17.

34.

Rousson

Zumbrunn

. Decision curve analysis revisited: overall net benefit, relationships to ROC curve analysis, and application to case-control studies. BMC Med Inform Decis Mak 2011; 11: 45.

35.

Steyerberg

Vickers

Cook

, et al. Assessing the performance of prediction models: a framework for traditional and novel measures. Epidemiology 2010; 21(1): 128–138.

36.

Seymour

Liu

Iwashyna

, et al. Assessment of clinical criteria for sepsis: for the third international consensus definitions for sepsis and septic shock (Sepsis-3). JAMA 2016; 315(8): 762–774.

37.

Le Gall

Klar

Lemeshow

, et al. The Logistic Organ Dysfunction system. A new way to assess organ dysfunction in the intensive care unit. ICU Scoring Group. JAMA 1996; 276(10): 802–810.

38.

Yan

Gan

, et al. Prognostic values of SOFA score, qSOFA score, and LODS score for patients with sepsis. Ann Palliat Med 2020; 9(3): 1037–1044.

39.

Kobayashi

Nakagawa

Kudo

, et al. Arterial blood pressure correlates with 90-day mortality in sepsis patients: a retrospective multicenter derivation and validation study using high-frequency continuous data. Blood Press Monit 2019; 24(5): 225–233.

40.

Capuzzo

Scaramuzza

Vaccarini

, et al. Validation of SAPS 3 admission score and comparison with SAPS II. Acta Anaesthesiol Scand 2009; 53(5): 589–594.

41.

Kochanek

Schalk

von Bergwelt-Baildon

, et al. Management of sepsis in neutropenic cancer patients: 2018 guidelines from the infectious diseases working party (AGIHO) and intensive care working party (iCHOP) of the German society of hematology and medical oncology (DGHO). Ann Hematol 2019; 98(5): 1051–1069.

42.

Poutsiaka

Davidson

Kahn

, et al. Risk factors for death after sepsis in patients immunosuppressed before the onset of sepsis. Scand J Infect Dis 2009; 41(6-7): 469–499.

43.

Tolsma

Schwebel

Azoulay

, et al. Sepsis severe or septic shock: outcome according to immune status and immunodeficiency profile. Chest 2014; 146(5): 1205–1213.

44.

Kim

Scherer

. Obesity, diabetes, and increased cancer progression. Diabetes Metab J 2021; 45(6): 799–812.

45.

Pearson-Stuttard

Papadimitriou

Markozannes

, et al. Type 2 diabetes and cancer: an umbrella review of observational and mendelian randomization studies. Cancer Epidemiol Biomarkers Prev 2021; 30(6): 1218–1228.

46.

Roy

Sahoo

Kamalanathan

, et al. Diabetes and pancreatic cancer: exploring the two-way traffic. World J Gastroenterol 2021; 27(30): 4939–4962.

47.

Mirouse

Vigneron

Llitjos

, et al. Sepsis and cancer: an interplay of friends and foes. Am J Respir Crit Care Med 2020; 202(12): 1625–1635.

48.

Raith

Udy

Bailey

, et al. Prognostic accuracy of the SOFA score, SIRS criteria, and qSOFA score for in-hospital mortality among adults with suspected infection admitted to the intensive care unit. JAMA 2017; 317(3): 290–300.

49.

Liu

Meng

, et al. Prognostic accuracy of the serum lactate level, the SOFA score and the qSOFA score for mortality among adults with Sepsis. Scand J Trauma Resusc Emerg Med 2019; 27(1): 51.

50.

Khwannimit

Bhurayanontachai

Vattanavanit

. Comparison of the performance of SOFA, qSOFA and SIRS for predicting mortality and organ failure among sepsis patients admitted to the intensive care unit in a middle-income country. J Crit Care 2018; 44: 156–1560.

51.

Qiu

Lei

Zhou

. SIRS, SOFA, qSOFA, and NEWS in the diagnosis of sepsis and prediction of adverse outcomes: a systematic review and meta-analysis. Expert Rev Anti Infect Ther 2023; 21(8): 891–900.