28-Day Mortality and Prognostic Factors in COPD Patients with Sepsis: A Retrospective Study Using eICU-CRD Database Authors: Ailing Zhu, Wei Sun, Dehai Ge, Xiujian Zhang, Jinfeng Zhang, Manman Hu, Qiangjian Zhong, Yangchun Wang, Junjun Liu Published: 2026 ABSTRACT BackgroundChronic obstructive pulmonary disease (COPD) is the third leading global cause of death, and concurrent sepsis markedly increases short-term mortality, making early identification of high-risk patients clinically essential. This retrospective study, based on the multi-center eICU-CRD database, sought to identify independent predictors of 28-day all-cause mortality and develop a combined prognostic model for intensive care unit (ICU) patients complicated with both COPD and sepsis.MethodsA total of 1814 eligible patients (49.4% male, mean age 70.03 ± 11.20 years) from 208 US ICUs (2014–2015) were included. Baseline demographic, vital, laboratory, blood gas and organ function data within 24 h of admission were collected. Univariate and multivariate Cox regression screened independent prognostic factors, and ROC curves evaluated predictive performance of single markers and the integrated model.Results28-day mortality among sepsis-COPD patients was 9.2% (166/1814). After adjusting for confounding variables, the sequential organ failure assessment (SOFA) score was proven to independently affect mortality (HR = 1.19, 95% CI 1.08–1.30, P < 0.001). In contrast, pH (HR = 0.08, 95% CI 0.01–0.96) and body mass index (BMI) (HR = 0.96, 95% CI 0.932–0.998) showed protective effects. All included predictors had area under the curve (AUC) values between 0.57 and 0.67, and the SOFA score achieved the maximum predictive performance. A combined risk model integrating these variables achieved an AUC of 0.70 (95% CI 0.65–0.75, P < 0.001), showing satisfactory discriminatory power. At a threshold value of 0.166 for predictive probability, the model obtained 0.514 sensitivity and 0.791 specificity.ConclusionHigher SOFA, lower pH and lower BMI independently predict worse 28-day survival in ICU COPD-sepsis patients. This integrated model exhibits moderate discriminative ability and can serve as an auxiliary tool for rapid risk screening among ICU patients. Introduction Chronic obstructive pulmonary disease (COPD), a major global contributor to mortality, is a progressive chronic respiratory disorder characterized by persistent airflow limitation and a substantial socioeconomic burden.1,2 It is associated with chronic systemic inflammation, immune dysfunction, and reduced cardiopulmonary reserve. These factors collectively establish a susceptible host microenvironment, increasing the risk of severe infection.1,3 COPD is a common comorbidity in patients with sepsis, with a prevalence of approximately 15.3% among critically ill patients with sepsis, and has been confirmed to be an independent predictor of 28-day all-cause death in the population.4 Sepsis, a life-threatening complication, triggers a superimposed inflammatory cascade that exacerbates endothelial injury, mitochondrial dysfunction, and respiratory muscle fatigue, leading to a “second-hit” injury.5 A growing body of evidence confirms that patients with both COPD and sepsis face an increased risk of negative clinical events, including acute respiratory failure, multiple organ dysfunction, and elevated 28‑day mortality.6 Therefore, urgent efforts are required to identify prognostic factors in this high‑risk population as early as possible to guide clinical management and improve patient survival. During the last few decades, a growing body of research has attempted to explore the prognostic factors of COPD and sepsis separately, and a series of reliable prognostic indicators have been identified, providing important guidance for clinical diagnosis and treatment.7–13 However, despite extensive research on single diseases, studies focusing on the specific population of patients with sepsis complicated by COPD remain relatively scarce. More importantly, existing studies on this comorbid population mostly focus on single indicators or a single category of indicators, and there is a lack of comprehensive prognostic analysis that integrates three key aspects closely related to the disease progression of this population: arterial blood gas analysis, nutritional status, and organ function. Arterial blood gas analysis directly reflects the respiratory compensation ability and acid–base balance of patients with COPD and sepsis, which is closely related to the severity of respiratory dysfunction.14 Nutritional status (such as body mass index and serum albumin) is an important indicator of the body’s reserve capacity and immune function, and its impairment is often associated with poor prognosis in critically ill patients.15,16 Organ function scores (eg, SOFA score) can comprehensively evaluate the degree of multi-organ damage caused by sepsis, which is a classic prognostic tool for patients with sepsis.17 The lack of an integrated analysis of these three aspects limits the comprehensiveness and accuracy of prognostic evaluation in this specific population. Yu et al and Long et al have conducted relevant studies based on the Medical Information Mart for Intensive Care (MIMIC) database to explore 28-day mortality prognostic factors among COPD patients with sepsis.18,19 Nevertheless, those analyses only assessed isolated clinical and laboratory markers, failing to integrate three core dimensions (arterial blood gas, nutritional status, and organ function scores) for comprehensive prognostic evaluation. Each of the three dimensions carries distinct clinical significance for this specific population: arterial blood gas directly reflects impaired respiratory compensation and acid–base imbalance inherent to COPD exacerbation; nutritional indices such as BMI mirror metabolic reserve and anti-inflammatory capacity; organ function scoring like SOFA quantifies the degree of systemic inflammatory injury induced by superimposed sepsis. Furthermore, few comparable studies have been performed using the eICU Collaborative Research Database (eICU-CRD). Encompassing 208 intensive care units built on a telemedicine framework, eICU-CRD differs substantially from the single-center MIMIC database in admission patterns and patient profiles. MIMIC cohorts feature higher mechanical ventilation rates, greater burdens of advanced malignancy and chronic liver disease, more severe baseline illness, and elevated in-hospital mortality. By contrast, eICU-CRD enrolls heterogeneous patients from both community and academic ICUs with milder average organ dysfunction, better mirroring routine real-world ICU populations. Such fundamental cohort disparities suggest the combined predictive value of blood gas, nutritional, and organ function metrics may diverge between the two databases, further underscoring the necessity of the present multicenter analysis. Given the scarcity of studies on patients with sepsis and COPD in the eICU-CRD and the absence of comprehensive prognostic analysis integrating arterial blood gas parameters, nutritional status and organ function scores, this study sought to investigate the 28-day mortality risk factors in this specific population. We further explored the combined prognostic value of the three above dimensions, hoping to fill this research gap and provide evidence for the clinical risk stratification and individualized management of critically ill patients with sepsis and COPD. Methods Study Data Resources This study was designed as a retrospective observational study. Data were obtained from the eICU-CRD database (v2.0) available on PhysioNet.20 The authors completed the CITI Program training and obtained ethical certification, submitted a formal data access application, and received approval from PhysioNet. Raw data were downloaded and managed using a local PostgreSQL database. This database contains relevant clinical data from 208 hospitals in the United States, spanning 2014 to 2015. As a telehealth data repository dedicated exclusively to adult ICU patients, it comprises vital signs, laboratory test results, medication records, detailed diagnoses and treatment plans, admission diagnoses, and other treatment-related information.20 The database used in this study was released in strict compliance with the safe harbor provisions of the Health Insurance Portability and Accountability Act (HIPAA). Given the retrospective nature of this study, which involved no direct intervention on patients, and following review by the Privacert Certification Body in Cambridge, Massachusetts, USA, the data de-identification protocol was confirmed to meet safe harbor standards for minimizing re-identification risks. Given the above reason, this research did not require formal ethical review and authorization from the Massachusetts Institute of Technology Institutional Review Board (IRB No. 40859994), and informed consent from individual patients was not mandatory. The whole research was carried out in full accordance with the principles of the Declaration of Helsinki, and all experimental procedures strictly followed standard academic and clinical research guidelines. This retrospective study was granted ethical exemption by the Ethics Committee of Nanjing Meishan Hospital. Study Population This retrospective cohort included adult patients admitted to ICUs with concurrent COPD and sepsis at enrollment. COPD cases were identified solely via ICD-9 diagnostic codes recorded in electronic medical records; spirometry confirmation (the gold standard for COPD diagnosis) was unavailable in the database. Sepsis was defined according to the Sepsis-3 international consensus criteria, operationalized as an acute ≥2‑point increase in the SOFA score accompanied by confirmed or suspected infection.21,22 For database screening, suspected infection was captured using ICD‑9‑CM codes, including systemic bacteremia (038 series), localized severe bacterial infections, and explicit sepsis labels (995.91, 995.92, 785.52). Patients were included only if both infection‑related ICD‑9 codes and SOFA score elevation were documented within the first 24 hours of ICU admission. Individuals with any of the following conditions were excluded from the analysis: (1) repeated ICU admission rather than the first hospitalization; (2) ICU hospitalization duration shorter than 48 hours; (3) age under 18 years old; and (4) more than 30% of variable data missing. The detailed screening and enrollment flow of the research population is presented in Figure 1.Figure 1Flow chart of study population selection. A total of 1814 sepsis patients complicated with COPD were finally enrolled from 23,136 initial ICU admissions through sequential exclusion criteria.Flowchart of sepsis patient selection for final analysis from ICU admissions.Flowchart detailing the selection process of sepsis patients for final analysis from initial ICU admissions. Starting with 23,136 patients diagnosed with sepsis on ICU admission, the first step includes only the first ICU admission of each patient, resulting in 15,926 patients, with 7,210 excluded. Next, patients with ICU stays longer than 48 hours are selected, totaling 10,216, with 5,710 excluded. Patients older than 18 years are then included, resulting in 10,211, with 15 excluded. The next criterion is patients complicated with COPD, totaling 1,882, with 8,329 excluded. Finally, patients with more than 30 percent missing data are excluded, leaving 1,814 patients for the final analysis. Flow chart of study population selection. A total of 1814 sepsis patients complicated with COPD were finally enrolled from 23,136 initial ICU admissions through sequential exclusion criteria. Data Collection Data were extracted from version 2.0 of the publicly available eICU-CRD database using Structured Query Language (SQL) scripts. All target variables were retrieved within the first 24 hours after ICU admission. For continuous variables, the initial recorded value at ICU admission was extracted; for categorical variables (including diagnoses and treatment records), all documented entries within the 24‑hour window were reviewed for eligibility. SQL queries were standardized to align with official eICU-CRD table schemas, with strict filters applied to exclude duplicate records and extreme outliers. The extracted variables were categorized into four groups: Demographic characteristics and vital signs: sex, age, heart rate, mean arterial pressure, core temperature, and SOFA score.Arterial blood gas parameters: arterial carbon dioxide tension (PaCO2), arterial oxygen tension (PaO2), and bicarbonate level.Nutritional markers: including BMI and serum albumin.Laboratory test results included: white blood cell count (WBC), red cell distribution width (RDW), blood glucose, hemoglobin, blood urea nitrogen (BUN), creatinine, sodium, potassium, calcium, alanine aminotransferase (ALT), and aspartate aminotransferase (AST). Demographic characteristics and vital signs: sex, age, heart rate, mean arterial pressure, core temperature, and SOFA score. Demographic characteristics and vital signs: sex, age, heart rate, mean arterial pressure, core temperature, and SOFA score. Arterial blood gas parameters: arterial carbon dioxide tension (PaCO2), arterial oxygen tension (PaO2), and bicarbonate level. Arterial blood gas parameters: arterial carbon dioxide tension (PaCO2), arterial oxygen tension (PaO2), and bicarbonate level. Nutritional markers: including BMI and serum albumin. Nutritional markers: including BMI and serum albumin. Laboratory test results included: white blood cell count (WBC), red cell distribution width (RDW), blood glucose, hemoglobin, blood urea nitrogen (BUN), creatinine, sodium, potassium, calcium, alanine aminotransferase (ALT), and aspartate aminotransferase (AST). Laboratory test results included: white blood cell count (WBC), red cell distribution width (RDW), blood glucose, hemoglobin, blood urea nitrogen (BUN), creatinine, sodium, potassium, calcium, alanine aminotransferase (ALT), and aspartate aminotransferase (AST). Outcome The primary outcome was 28-day all-cause mortality from the time of ICU admission. Patients were followed from ICU admission until the occurrence of death, censoring at 28 days, or loss to follow‑up, whichever occurred first. Statistical Analysis Data analyses were performed using IBM SPSS Statistics version 26.0 software. The Kolmogorov–Smirnov test was used to examine the normal distribution of continuous indicators. Normally distributed data are described as mean ± standard deviation, whereas non-normally distributed variables are reported as median with interquartile range. Categorical variables were summarized as counts and constituent ratios. Missing values for variables with missingness less than 30% were imputed using the random forest algorithm via the R “missForest” package. Five-fold cross-validation was adopted to optimize imputation performance and mitigate bias.23 Variables with missingness exceeding this threshold were excluded from the analysis. The imputation model incorporated all predictors and the outcome variable to reduce estimation bias.23 For prognostic variable screening, a standardized three-step workflow was conducted: First, the time‑to‑event analysis was performed as described in the Outcomes section (2.4), with follow‑up starting at ICU admission and censoring at 28 days. Cox proportional hazards regression was then applied to estimate hazard ratios (HR) and 95% confidence intervals (CI) for each predictor. The included variables included age, gender, body mass index, body temperature, respiration, blood pressure, SOFA score, liver and kidney function, electrolytes, blood gas analysis, and blood routine indicators. HR together with CI were estimated, and the Wald test was applied to assess statistical significance. Variables yielding a P value below 0.1 in univariate regression were further entered into the multivariate model. Second, to address multicollinearity, variables with a variance inflation factor (VIF) >5 were excluded from the multivariate Cox regression model. Accordingly, bicarbonate (HCO3−), which had a VIF of 6.42, was not retained in the final model. Since the SOFA score, Acute Physiology Score III, and APACHE IV score reflect overlapping severity of critical illness and exhibit strong multicollinearity (all are comprehensive scores for evaluating the severity of critically ill patients, with overlapping indicators including blood pressure, oxygenation, renal function, coagulation, consciousness, and acid–base status), only the SOFA score (the most widely recognized scoring system for sepsis in ICU settings) was involved in the final multivariate Cox regression model to avoid statistical bias and overfitting. In addition, ALT and AST were excluded from the final model, as their hazard ratio was close to 1 (HR = 1.001) with an extremely narrow 95% confidence interval (1.000–1.001), indicating a negligible effect size with no clinically meaningful impact on 28-day mortality. The model was then used to compute adjusted HR and 95% CI associated with mortality risk. The proportional hazards assumption of the Cox model was assessed using Schoenfeld residual testing, and no significant violations were detected for any of the included predictors (global P > 0.05). With 166 observed deaths (9.2% mortality rate among 1814 patients), the study had >80% power to detect a hazard ratio of ≥1.3 for the primary predictors at a two‑sided significance level of α = 0.05, assuming an event proportion of 10% in the reference group. Meanwhile, we used ROC curves to evaluate the prognostic performance of these markers for mortality among COPD patients with sepsis. After identifying significant prognostic factors using Cox regression analysis, we constructed ROC curves based on the predicted probabilities from the Cox model to compare the discriminative performance of individual independent predictors and the combined three-variable model. For variables negatively associated with the outcome or with an area under the ROC curve (AUC) < 0.5, reciprocal transformation (1/x) was used to unify the direction of prognostic effects and improve discriminative performance. AUC between 0.7 and 0.8 was considered to have certain practical discriminative value, and the optimal cut-off values of continuous variables were determined by maximizing the Youden index (Youden index = sensitivity + specificity - 1). Notably, this study neither conducted internal cross-validation nor external cohort validation for the combined risk model, which constitutes an important limitation of the predictive analysis. A two-tailed test was adopted in the study, and statistical significance was set at P < 0.05. Result Baseline Demographics and Clinical Features of Subjects The study included 1814 patients with sepsis and concomitant COPD who were enrolled. The cohort included 49.4% male and 50.6% female individuals, with a mean age of 70.03 ± 11.20 years. Baseline demographic and clinical characteristics are summarized in Table 1. Mean BMI was 28.88 ± 9.57 kg/m2, and mean SOFA score was 3.63 ± 2.82. Blood gas indicators revealed the following values: pH 7.33 ± 0.11, PaO2 108.37 ± 75.57 mmHg, and PaCO2 50.03 ± 18.43 mmHg. Regarding biochemical parameters, AST was 30.0 U/L (interquartile range 19–61), ALT was 26.0 U/L (17–46.75), mean albumin was 2.6 ± 0.61 g/dL, mean creatinine was 1.65 ± 1.50 mg/dL, and mean BUN was 32.38 ± 22.50 mg/dL. WBC was 15.04 ± 9.52 × 109/L, RDW was 16.0 ± 2.45%, and mean blood glucose was 156.16 ± 75.14 mg/dL. Serum electrolyte levels were as follows: sodium 138.31 ± 5.38 mmol/L, potassium 4.20 ± 0.75 mmol/L, and calcium 8.12 ± 0.82 mg/dL. Overall, 17.4% of patients underwent endotracheal intubation.Table 1Participants Characteristics at Baseline (N=1814)VariablesMean±SD OR N (%)Age (years)70.03±11.20BMI (kg/m2)28.88±9.57Body temperature (°C)36.56±0.94Mean BP (mmHg)77.64±42.09Respiration (bpm)29.63±13.78SOFA score3.63±2.82APS III50.36±23.42APACHE IV score65.14±24.07PH7.33±0.11PaO2 (mmHg)108.37±75.57PaCO2 (mmHg)50.03±18.43HCO3 (mmol/L)25.49±8.09Creatinine (mg/dl)1.65±1.50BUN (mg/dl)32.38±22.50AST (U/L)30.0(19–61)ALT (U/L)26.0(17–46.75)Albumin (g/dL)2.6±0.61WBC (*109/L)15.04±9.52RDW (%)16.0±2.45Glucose (mg/dl)156.16±75.14Sodium (mmol/L)138.31±5.38Potassium (mmol/L)4.20±0.75Calcium (mg/dl)8.12±0.82Sex Female918(50.6) Male895(49.4)Intubated No1442(82.6) Yes303(17.4)Note: Mean and SD are shown for the continuous variables. N and % are shown for the categorical variables.Abbreviations: BMI, body mass index; BP, blood pressure; SOFA, sequential organ failure assessment; APS III, Acute Physiology Score III; PaO2, partial pressure of arterial oxygen; PaCO2, partial pressure of arterial carbon dioxide; HCO3, Bicarbonate; AST, Aspartate Aminotransferase; ALT, Alanine Aminotransferase; BUN, blood urea nitrogen; WBC, white blood cell; RDW, red cell distribution width. Table 1Participants Characteristics at Baseline (N=1814)VariablesMean±SD OR N (%)Age (years)70.03±11.20BMI (kg/m2)28.88±9.57Body temperature (°C)36.56±0.94Mean BP (mmHg)77.64±42.09Respiration (bpm)29.63±13.78SOFA score3.63±2.82APS III50.36±23.42APACHE IV score65.14±24.07PH7.33±0.11PaO2 (mmHg)108.37±75.57PaCO2 (mmHg)50.03±18.43HCO3 (mmol/L)25.49±8.09Creatinine (mg/dl)1.65±1.50BUN (mg/dl)32.38±22.50AST (U/L)30.0(19–61)ALT (U/L)26.0(17–46.75)Albumin (g/dL)2.6±0.61WBC (*109/L)15.04±9.52RDW (%)16.0±2.45Glucose (mg/dl)156.16±75.14Sodium (mmol/L)138.31±5.38Potassium (mmol/L)4.20±0.75Calcium (mg/dl)8.12±0.82Sex Female918(50.6) Male895(49.4)Intubated No1442(82.6) Yes303(17.4)Note: Mean and SD are shown for the continuous variables. N and % are shown for the categorical variables.Abbreviations: BMI, body mass index; BP, blood pressure; SOFA, sequential organ failure assessment; APS III, Acute Physiology Score III; PaO2, partial pressure of arterial oxygen; PaCO2, partial pressure of arterial carbon dioxide; HCO3, Bicarbonate; AST, Aspartate Aminotransferase; ALT, Alanine Aminotransferase; BUN, blood urea nitrogen; WBC, white blood cell; RDW, red cell distribution width. Participants Characteristics at Baseline (N=1814) Note: Mean and SD are shown for the continuous variables. N and % are shown for the categorical variables. Abbreviations: BMI, body mass index; BP, blood pressure; SOFA, sequential organ failure assessment; APS III, Acute Physiology Score III; PaO2, partial pressure of arterial oxygen; PaCO2, partial pressure of arterial carbon dioxide; HCO3, Bicarbonate; AST, Aspartate Aminotransferase; ALT, Alanine Aminotransferase; BUN, blood urea nitrogen; WBC, white blood cell; RDW, red cell distribution width. Univariate Analysis of Mortality-Related Factors During the 28-day follow-up period after ICU admission, 166 participants died, corresponding to an overall mortality rate of 9.2%. With 28-day mortality defined as the dependent variable and baseline characteristics as independent variables, survival analysis was performed using the Cox regression analysis in SPSS. Univariate Cox regression indicated that BMI (HR=0.97, 95% CI: 0.95–0.99), body temperature (HR=0.73, 95% CI: 0.62–0.86) and pH value (HR=0.03, 95% CI: 0.01–0.13) were inversely correlated with 28-day mortality risk. Conversely, age (HR=1.03, 95% CI: 1.02–1.05), SOFA score (HR=1.17, 95% CI: 1.12–1.23), APS III (HR=1.02, 95% CI: 1.02–1.03), APACHE IV score (HR=1.03, 95% CI: 1.02–1.03), WBC (HR=1.02, 95% CI: 1.01–1.03), RDW (HR=1.09, 95% CI: 1.04–1.16) and BUN (HR=1.01, 95% CI: 1.01–1.02) could independently increase the risk of 28-day death and served as positive prognostic indicators. Gender (HR=0.98, 95% CI: 0.72–1.33), endotracheal intubation status (HR=0.75, 95% CI: 0.54–1.05), PaO2 (HR=1.000, 95% CI: 0.997–1.003) albumin (HR=0.73, 95% CI: 0.53–1.02) and serum electrolyte levels all showed no significant correlation with mortality, with the differences lacking statistical significance (Table 2).Table 2Result of Univariate and Multivariate Analysis for the Mortality Risk Factors in Spesis-COPD PatientsVariablesUnivariate AnalysisMultivariate AnalysisHR95% CIP valueHR95% CIP valueAge (years)1.031.02~1.05<0.0011.020.99~1.040.293BMI (kg/m2)0.970.95~0.990.0010.960.932~0.9980.039Temperature (°C)0.730.62~0.86<0.0010.820.63~1.070.148Mean BP (mmHg)0.9980.995~1.0020.343–––Respiration (bpm)1.011.00~1.020.264–––SOFA1.171.12~1.23<0.0011.191.08~1.30<0.001APS III1.021.02~1.03<0.001–––APACHE IV score1.031.02~1.03<0.001–––PH0.030.01~0.13<0.0010.080.01~0.960.047PaO2 (mmHg)1.0021.001~1.0040.008–––PaCO2 (mmHg)1.0010.992~1.0110.815–––HCO3 (mmol/L)0.970.95~0.990.013–––AST (U/L)1.0001.000~1.0010.000–––ALT (U/L)1.0011.000~1.0010.000–––Albumin (g/dL)0.730.53~1.020.0620.780.48~1.250.298WBC (*109/L)1.021.01~1.030.0021.020.99~1.040.177RDW (%)1.091.04~1.160.0011.020.93~1.120.712Creatinine (mg/dl)1.070.99~1.150.0980.810.62~1.050.114BUN (mg/dl)1.011.01~1.02<0.0011.011.00~1.020.374Glucose (mg/dl)1.0010.999~1.0030.171–––Sodium (mmol/L)0.9960.968~1.0250.800–––Potassium (mmol/L)1.241.03~1.490.0241.200.87~1.660.259Calcium (mg/dl)0.940.77~1.140.496–––Sex Female1.00(Reference)1.00(Reference) Male0.980.72~1.330.8890.830.47~1.460.521Intubated Yes1.00(Reference)– No0.750.54~1.050.090–––Abbreviations: BMI, body mass index; BP, blood pressure; SOFA, sequential organ failure assessment; APS III, Acute Physiology Score III; PaO2, partial pressure of arterial oxygen; PaCO2, partial pressure of arterial carbon dioxide; HCO3, Bicarbonate; AST, Aspartate Aminotransferase; ALT, Alanine Aminotransferase; BUN, blood urea nitrogen; WBC, white blood cell; RDW, red cell distribution width. Table 2Result of Univariate and Multivariate Analysis for the Mortality Risk Factors in Spesis-COPD PatientsVariablesUnivariate AnalysisMultivariate AnalysisHR95% CIP valueHR95% CIP valueAge (years)1.031.02~1.05<0.0011.020.99~1.040.293BMI (kg/m2)0.970.95~0.990.0010.960.932~0.9980.039Temperature (°C)0.730.62~0.86<0.0010.820.63~1.070.148Mean BP (mmHg)0.9980.995~1.0020.343–––Respiration (bpm)1.011.00~1.020.264–––SOFA1.171.12~1.23<0.0011.191.08~1.30<0.001APS III1.021.02~1.03<0.001–––APACHE IV score1.031.02~1.03<0.001–––PH0.030.01~0.13<0.0010.080.01~0.960.047PaO2 (mmHg)1.0021.001~1.0040.008–––PaCO2 (mmHg)1.0010.992~1.0110.815–––HCO3 (mmol/L)0.970.95~0.990.013–––AST (U/L)1.0001.000~1.0010.000–––ALT (U/L)1.0011.000~1.0010.000–––Albumin (g/dL)0.730.53~1.020.0620.780.48~1.250.298WBC (*109/L)1.021.01~1.030.0021.020.99~1.040.177RDW (%)1.091.04~1.160.0011.020.93~1.120.712Creatinine (mg/dl)1.070.99~1.150.0980.810.62~1.050.114BUN (mg/dl)1.011.01~1.02<0.0011.011.00~1.020.374Glucose (mg/dl)1.0010.999~1.0030.171–––Sodium (mmol/L)0.9960.968~1.0250.800–––Potassium (mmol/L)1.241.03~1.490.0241.200.87~1.660.259Calcium (mg/dl)0.940.77~1.140.496–––Sex Female1.00(Reference)1.00(Reference) Male0.980.72~1.330.8890.830.47~1.460.521Intubated Yes1.00(Reference)– No0.750.54~1.050.090–––Abbreviations: BMI, body mass index; BP, blood pressure; SOFA, sequential organ failure assessment; APS III, Acute Physiology Score III; PaO2, partial pressure of arterial oxygen; PaCO2, partial pressure of arterial carbon dioxide; HCO3, Bicarbonate; AST, Aspartate Aminotransferase; ALT, Alanine Aminotransferase; BUN, blood urea nitrogen; WBC, white blood cell; RDW, red cell distribution width. Result of Univariate and Multivariate Analysis for the Mortality Risk Factors in Spesis-COPD Patients Abbreviations: BMI, body mass index; BP, blood pressure; SOFA, sequential organ failure assessment; APS III, Acute Physiology Score III; PaO2, partial pressure of arterial oxygen; PaCO2, partial pressure of arterial carbon dioxide; HCO3, Bicarbonate; AST, Aspartate Aminotransferase; ALT, Alanine Aminotransferase; BUN, blood urea nitrogen; WBC, white blood cell; RDW, red cell distribution width. Multivariate Analysis of Mortality Risk Factors in Sepsis-COPD Patients Variables with a univariate P < 0.1, including age, BMI, body temperature, endotracheal intubation, SOFA score, pH, bicarbonate, serum creatinine, BUN, serum potassium, albumin, WBC, and RDW, were incorporated into the multivariate Cox regression model for prognosticating 28-day ICU death in specific groups of patients. Given its VIF greater than 5, HCO3 was omitted from the final multivariate Cox model. The findings of the regression analysis are presented in Table 2. After controlling for confounding variables, the SOFA score (HR = 1.19, 95% CI: 1.08–1.30, P < 0.001) was identified as an independent risk factor for mortality in the training cohort. In contrast, BMI (HR = 0.96, 95% CI: 0.932–0.998, P = 0.039) and blood pH (HR = 0.08, 95% CI: 0.01–0.96, P = 0.047) served as independent protective factors against the risk of mortality (Table 2). ROC Curve Analysis to Assess Predictive Values ROC analysis elucidated the individual discriminative performances of each prognostic indicator. The respective AUC values were 0.67 for the SOFA score, 0.57 for the inverse BMI, and 0.61 for the inverse pH. Integrating these three variables into a composite prediction model moderately improved discriminatory performance, yielding an elevated AUC of 0.70 (P < 0.001, 95% CI: 0.65–0.75) for the prediction of 28-day mortality among patients with COPD complicated by sepsis (Table 3 and Figure 2). With the optimal Youden index determined, the optimal threshold for the combined predictive probability was established at 0.166, with corresponding diagnostic parameters of 51.4% sensitivity and 79.1% specificity.Table 3Area Under the Curve of SOFA Score, 1/BMI,1/PH, CombinationVariablesAUC (95% CI)P valueCutoff PointSensitivitySpecificitySOFA score0.67(0.62~0.73)<0.0013.50.7930.4661/BMI0.57(0.52~0.63)0.0150.030.7750.3551/PH0.61(0.55~0.67)<0.0010.140.4590.758Combination0.70(0.65~0.75)<0.0010.170.5140.791Abbreviations: BMI, body mass index; BP, blood pressure; SOFA, sequential organ failure assessment; APS III, Acute Physiology Score III; PaO2, partial pressure of arterial oxygen; PaCO2, partial pressure of arterial carbon dioxide; HCO3, Bicarbonate; AST, Aspartate Aminotransferase; ALT, Alanine Aminotransferase; BUN, blood urea nitrogen; WBC, white blood cell; RDW, red cell distribution width; AUC, area under the curve. Figure 2ROC curve analysis. Comparison of predictive performance among SOFA score, 1/BMI, 1/pH, and the combination model.ROC curve multi-line graph comparing SOFA Score, 1 over BMI, 1 over pH, prediction probability and reference line. Table 3Area Under the Curve of SOFA Score, 1/BMI,1/PH, CombinationVariablesAUC (95% CI)P valueCutoff PointSensitivitySpecificitySOFA score0.67(0.62~0.73)<0.0013.50.7930.4661/BMI0.57(0.52~0.63)0.0150.030.7750.3551/PH0.61(0.55~0.67)<0.0010.140.4590.758Combination0.70(0.65~0.75)<0.0010.170.5140.791Abbreviations: BMI, body mass index; BP, blood pressure; SOFA, sequential organ failure assessment; APS III, Acute Physiology Score III; PaO2, partial pressure of arterial oxygen; PaCO2, partial pressure of arterial carbon dioxide; HCO3, Bicarbonate; AST, Aspartate Aminotransferase; ALT, Alanine Aminotransferase; BUN, blood urea nitrogen; WBC, white blood cell; RDW, red cell distribution width; AUC, area under the curve. Area Under the Curve of SOFA Score, 1/BMI,1/PH, Combination Abbreviations: BMI, body mass index; BP, blood pressure; SOFA, sequential organ failure assessment; APS III, Acute Physiology Score III; PaO2, partial pressure of arterial oxygen; PaCO2, partial pressure of arterial carbon dioxide; HCO3, Bicarbonate; AST, Aspartate Aminotransferase; ALT, Alanine Aminotransferase; BUN, blood urea nitrogen; WBC, white blood cell; RDW, red cell distribution width; AUC, area under the curve. ROC curve analysis. Comparison of predictive performance among SOFA score, 1/BMI, 1/pH, and the combination model. Discussion In this investigation, the 28‑day mortality rate among ICU patients with sepsis and concomitant COPD was 9.2%, which was notably lower than the rates reported in previous studies: 21.45% by Yu et al and 23.6% by Chen et al.4,18 One possible reason for the decreased mortality rate in our cohort might be the higher proportion of critically ill patients within the MIMIC-III dataset adopted by Chen et al.4 In contrast, the mean SOFA score in our study was considerably lower, indicating a milder baseline illness severity. Univariate Cox regression revealed multiple factors correlated with 28-day mortality, including older age, higher SOFA score, and elevated BUN as risk factors, while higher BMI, higher body temperature, and higher arterial pH correlated with improved survival outcomes. In contrast, sex, endotracheal intubation status, and serum electrolyte levels showed no significant univariate association with mortality, implying these variables do not independently correlate with prognosis in this specific cohort. After controlling for potential confounders via multivariate Cox analysis, SOFA score, BMI and pH remained independent correlations of 28-day mortality. Our study demonstrated that the SOFA score is independently linked to 28-day mortality among patients with COPD complicated by sepsis, which is consistent with previous findings.24–26 An elevated SOFA score indicates progressive multiple organ dysfunction induced by uncontrolled systemic inflammation, including impaired circulation, respiration, renal function, coagulation, and neurological status, indicating a severely diminished physiological reserve and increased mortality risk.25 Prior research has indicated that among various scoring models, the SOFA score exhibits the highest sensitivity and AUC for predicting prognosis in patients with COPD and multiple organ dysfunction syndrome.24 However, interpreting the SOFA score in a COPD-specific population requires consideration of unique pathophysiological characteristics. First, patients with COPD often experience chronic hypoxemia and type II respiratory failure, and consequently, the PaO2/FiO2 component of the respiratory SOFA score is frequently impaired at baseline. An equivalent rise in respiratory SOFA may therefore represent poorer baseline pulmonary reserve rather than de novo acute lung injury, leaving these patients with minimal compensatory capacity.27 Second, auto PEEP resulting from airflow limitation in COPD can increase cardiac afterload and impede venous return, meaning that the shock state reflected by the cardiovascular SOFA score may be partially attributable to worsening pulmonary dynamics rather than solely to myocardial depression.28,29 This exacerbation of cardiopulmonary interaction makes patients with COPD particularly sensitive to dynamic elevations in the SOFA score. Therefore, for this patient population, dynamic monitoring of the SOFA score combined with other scoring models30,31 may provide superior prognostic value compared with using a single cutoff value of the SOFA score. This study demonstrated that reduced pH (acidemia) independently predicts 28-day mortality among individuals with sepsis and COPD showing outstanding prognostic value in this physiologically vulnerable cohort. Acidemia weakens diaphragmatic function, impairs cardiac performance, decreases tissue perfusion, and aggravates systemic inflammatory responses.32 COPD patients possess poor physiological reserve; therefore, even mild acidosis may rapidly cause circulatory and respiratory decompensation.14 Long-term impaired alveolar ventilation results in chronic hypercapnia and baseline respiratory acidosis, greatly reducing the tolerance to acid–base disturbances.33 Combined with sepsis-induced metabolic acidosis, the body’s buffering capacity is rapidly exhausted. Airflow limitation and respiratory muscle fatigue prevent effective hyperventilation compensation, worsening persistent acidosis.34,35 In addition, acidosis worsens organ damage via COPD-related mechanisms. It exacerbates diaphragmatic dysfunction in patients with pre-existing respiratory muscle impairment36 and induces pulmonary vasoconstriction to increase right ventricular load in patients with pulmonary hypertension.37 Therefore, a mild-to-moderate pH decline serves as a sensitive early prognostic indicator in this high-risk group. Clinically, our results emphasize the need for close acid–base monitoring and active interventions, including anti-infection treatment, proper fluid management and early noninvasive ventilation.38 A higher BMI correlates with lower 28-day mortality in patients with COPD and sepsis, supporting the “obesity paradox”—overweight/obese patients may have survival benefits in critical illness despite obesity’s traditional risks.39 Three mechanisms explain this: 1) Greater metabolic reserve resists sepsis-induced hypercatabolism and malnutrition;40 2) adipose tissue-derived anti-inflammatory adipokines (eg, adiponectin) and abundant lipoproteins neutralize endotoxins, reducing inflammation;41,42 3) Low BMI in COPD often indicates advanced disease (eg, cachexia), reflecting disease severity rather than adiposity.43,44 Clinically, moderate nutritional reserve confers protective benefits in COPD patients with acute sepsis.45 From a clinical management perspective, patients at the extremes of BMI warrant different considerations. For underweight patients (BMI < 18.5 kg/m2), our findings and prior evidence consistently indicate higher mortality risk;43,44 this subgroup may benefit from early nutritional support and close monitoring for cachexia‑related complications. Conversely, for patients with obesity (BMI ≥ 28 kg/m2), while the obesity paradox confers a survival advantage, clinicians should remain vigilant for obesity‑related mechanical complications, including ventilation difficulties and increased risk of pressure ulcers, which may counteract the protective metabolic effects. Zhou et al reported that a BMI cutoff of 20.8 kg/m2 was associated with distinct ICU mortality risks among critically ill patients with COPD,46 while Song et al47 identified a low‑BMI threshold of 18.5 kg/m2 as a marker of worse pulmonary function and higher mortality in stable COPD. Collectively, these studies suggest that a BMI range of approximately 18.5–28.0 kg/m2 may represent a reasonable target for optimizing outcomes in this population, with underweight patients deserving particular clinical attention. Further prospective validation is required to confirm these conclusions and establish more precise BMI targets. ROC curve analysis confirmed the moderate discriminative ability of the candidate predictors for 28-day all-cause mortality in patients with sepsis complicated by COPD. When evaluated individually, the SOFA score exhibited the strongest predictive performance (AUC = 0.67), followed by reverse-coded arterial pH (AUC = 0.61) and reverse-coded BMI (AUC = 0.57). The three-variable combined model produced a slightly higher AUC of 0.70, a marginal statistical gain that translated to only minimal incremental discrimination compared with any single predictor alone. At the optimal Youden index-derived cutoff value of 0.166, the composite model achieved moderate sensitivity (51.4%) and relatively high specificity (79.1%). Although the combined model yielded statistically superior discrimination versus individual markers, its clinical applicability warrants cautious interpretation. The modest sensitivity of 51.4% indicates that nearly half of patients who ultimately died would be falsely classified as low-risk, which prohibits its use as an independent prognostic tool; this shortcoming is partially compounded by the low 28-day mortality event rate (9.2%) within our cohort. Conversely, the model’s high specificity (79.1%) enables reliable positive triage to identify patients requiring intensified monitoring and early intervention. Compared with recently developed nomograms for COPD‑sepsis patients—Yu et al’s eight‑variable model (AUC 0.784/0.689 in MIMIC‑IV)18 and Yin et al’s seven‑variable model (AUC 0.765/0.788)48—our AUC of 0.70 is lower. Nevertheless, our tool relies solely on routinely collected bedside metrics with no additional specialized laboratory testing, and the multicenter eICU-CRD cohort enrolled patients with markedly milder baseline illness severity relative to single-center MIMIC-III populations; this floor effect inherently suppresses the maximum achievable AUC. Collectively, this parsimonious model strikes a pragmatic balance between predictive accuracy and clinical practicality for resource-constrained ICUs, though external validation across independent cohorts is still mandatory before widespread clinical adoption. This study has several limitations. First, COPD was identified using ICD codes rather than confirmatory spirometry, which may have introduced selection bias, and detailed data on specific causes of death were unavailable. Second, unmeasured confounders (eg, socioeconomic status, insurance coverage) could not be fully adjusted. Third, the retrospective design and missing data represent inherent limitations, despite the use of multiple imputation. Fourth, data were derived from a single‑country database (eICU), requiring external validation for generalizability. Beyond these methodological constraints, we acknowledge the inherent limitations regarding novelty and performance. All three predictors have been previously examined in MIMIC‑based studies, and the obesity paradox is well recognized. Our three‑variable model achieved an AUC of 0.70, representing only a modest incremental gain over SOFA alone (0.67), with moderate sensitivity (51.4%) precluding standalone use. The eICU‑CRD database lacks inflammatory biomarkers and body composition data, which prevented deeper mechanistic exploration. Nevertheless, these limitations point to concrete future directions: (1) prospective cohorts with spirometry‑confirmed COPD diagnoses; (2) dynamic trajectory modeling of SOFA, pH, and BMI during the first 72 hours rather than single baseline values; (3) incorporation of inflammatory markers (eg, IL‑6, Procalcitonin, CRP) and body composition metrics to elucidate the obesity paradox; and (4) external validation across diverse populations before clinical translation. Conclusion In this multicenter retrospective cohort study, elevated SOFA score, low arterial pH, and lower BMI were independently associated with 28-day mortality in critically ill COPD patients with sepsis. A parsimonious model integrating these three routinely available bedside variables demonstrated moderate discriminative performance for early mortality risk stratification. All predictors are obtainable at ICU admission without additional cost or complex computation, facilitating immediate risk assessment even in busy, resource-constrained clinical environments. Nevertheless, given its modest sensitivity, this model should be interpreted as an auxiliary screening tool rather than a definitive guide for individualized therapeutic decisions. Its primary utility lies in rapid “rule-out” triage—identifying patients at relatively lower risk who may be triaged to step-down units, thereby reserving intensive resources for those with overt high-risk profiles. External validation across diverse geographical and healthcare settings is essential before clinical implementation. Future model refinement should examine how these three predictors interact and evolve over time during ICU treatment, rather than simply incorporating additional biomarkers that may offer limited incremental value while reducing clinical practicality.