Development and Internal Validation of a Nutrition-Related Indicator-Based Prediction Model for in-Hospital Mortality in Hospitalized Patients with AECOPD

Abstract

Purpose

To investigate the predictive value of routinely available nutrition-related laboratory indicators for in-hospital mortality in hospitalized patients with acute exacerbation of chronic obstructive pulmonary disease (AECOPD) and to develop and internally validate a clinically applicable prediction model.

Patients and Methods

A retrospective cohort study included 2167 patients hospitalized for AECOPD at the Second People’s Hospital of Wuhu between January 1, 2016 and October 21, 2025. Candidate predictors were prespecified based on clinical relevance and previous evidence. Univariable and multivariable logistic regression analyses were performed to identify independent predictors of in-hospital mortality. A prediction model was developed and internally validated using receiver operating characteristic (ROC) analysis, calibration curves with bootstrap resampling, the Hosmer–Lemeshow goodness-of-fit test, and decision curve analysis (DCA).

Results

Among the 2167 patients, 109 (5.0%) died during hospitalization. Multivariable analysis identified five independent predictors of in-hospital mortality: older age (OR = 1.234, 95% CI: 1.077–1.422), lower hemoglobin (OR = 1.095, 95% CI: 1.036–1.152), lower albumin (OR = 1.135, 95% CI: 1.083–1.192), respiratory failure at admission (OR = 6.002, 95% CI: 3.936–9.234), and a history of lung cancer (OR = 4.430, 95% CI: 1.815–9.855). The prediction model achieved good discrimination, with an area under the receiver operating characteristic curve (AUC) of 0.811. The calibration curve demonstrated good agreement between predicted and observed outcomes, and the Hosmer–Lemeshow test indicated good model fit (P = 0.196). Decision curve analysis showed that the model provided greater net benefit than both the treat-all and treat-none strategies across a threshold probability range of approximately 5%–50%.

Conclusion

We developed and internally validated a simple prediction model incorporating routinely available nutrition-related laboratory indicators and key clinical characteristics for predicting in-hospital mortality in hospitalized patients with AECOPD. The model demonstrated good discrimination and calibration and may serve as a practical bedside tool for early risk stratification and individualized clinical decision-making. External validation in independent multicenter cohorts is warranted before widespread clinical implementation.

Introduction

Chronic obstructive pulmonary disease (COPD) is a highly prevalent chronic respiratory disorder and the third leading cause of death worldwide.¹ According to the Global Burden of Disease study, approximately 212 million individuals were affected by COPD globally in 2019, resulting in around 3.3 million deaths.² In China, the burden of COPD is also substantial. A nationwide epidemiological survey reported that the prevalence of COPD among individuals aged ≥40 years was approximately 13.6%.³

During the course of the disease, patients with COPD frequently experience acute exacerbations (AECOPD), which are associated with a significantly increased risk of mortality. Previous studies have reported that the in-hospital mortality rate of patients with AECOPD ranges from 2% to 8%,^4–6 while in critically ill patients admitted to the intensive care unit, the mortality rate may exceed 20%.⁷ Therefore, accurate early assessment of mortality risk at hospital admission is of considerable clinical importance for guiding timely and appropriate interventions.

Several prognostic models for predicting in-hospital mortality in patients with AECOPD have been proposed,^8–10 indicating that multiple clinical factors may influence patient outcomes. However, many existing prediction models incorporate numerous clinical, physiological, and laboratory variables, which may limit their routine clinical application. Among the reported predictors, malnutrition and anemia are common conditions in patients with COPD and may have a significant impact on prognosis.^11–13 Malnutrition has been consistently associated with adverse outcomes in COPD. Previous studies have demonstrated that reduced fat-free mass (FFM) and low body mass index (BMI) are independent predictors of increased mortality in COPD patients.^11,14 In addition, anemia has also been identified as an independent risk factor for poor prognosis in COPD.¹⁵

These findings suggest that nutritional status—reflected by laboratory indicators such as hemoglobin and albumin—may play an important role in determining clinical outcomes in COPD patients. However, although several prognostic models have incorporated laboratory parameters, the prognostic value of routinely available nutrition-related laboratory indicators has not been sufficiently emphasized in concise and clinically applicable prediction models. Therefore, the present study aimed to develop a clinically applicable prediction model for in-hospital mortality in hospitalized patients with AECOPD by integrating key clinical characteristics with nutrition-related laboratory indicators, with particular emphasis on the predictive value of hemoglobin and albumin. A nomogram was further constructed to facilitate individualized risk assessment and assist clinicians in identifying high-risk patients and optimizing management strategies.

Methods

Study Population

This retrospective cohort study included patients hospitalized for acute exacerbation of COPD at the Second People’s Hospital of Wuhu between January 1, 2016 and October 21, 2025. COPD was diagnosed according to the Global Initiative for Chronic Obstructive Lung Disease (GOLD) criteria. Inclusion criteria were: age ≥40 years and a primary diagnosis of AECOPD. Exclusion criteria included: coexisting severe pulmonary diseases such as tuberculosis or interstitial lung disease, active malignancy (a history of malignancy was recorded only as a comorbidity), and patients who died or were discharged within 24 hours after admission.

All patients received standard treatment, including oxygen therapy, bronchodilators, corticosteroids, and anti-infective therapy as appropriate. This study was approved by the institutional ethics committee (Approval No. 2023-KY-026). Given the retrospective and anonymized nature of the study, the requirement for informed consent was waived.

Data Collection and Preprocessing

Baseline demographic characteristics, clinical characteristics, and laboratory parameters were collected from the electronic medical records at hospital admission. Candidate predictors were selected a priori based on clinical relevance, previous evidence, and the objective of developing a clinically practical nutrition-related prediction model. The candidate variables evaluated for model development included age, sex, respiratory failure at admission, history of lung cancer, hemoglobin, albumin, blood urea nitrogen, creatinine, and IL-6. Respiratory failure was defined as respiratory failure present at the time of hospital admission, based on the requirement for noninvasive or invasive ventilatory support and documented in the admission medical records.

For variables with a low proportion of missing data (≤5%), complete-case analysis was performed. For variables with missing rates between 5% and 30%, missing values were imputed using multiple imputation. To improve the stability and accuracy of imputation, the MissForest algorithm was applied for handling mixed-type data.¹⁶

Variable Selection and Model Construction

Candidate predictors were prespecified based on previous literature, clinical relevance, and the objective of developing a clinically practical nutrition-related prediction model. Univariable analyses were performed using the chi-square test, Mann–Whitney U-test, or univariable logistic regression, as appropriate for the variable type. Variables included in the final multivariable logistic regression model were selected by comprehensively considering the results of the univariable analyses together with their clinical importance.

A multivariable logistic regression model was subsequently constructed to identify independent predictors of in-hospital mortality. Regression coefficients were used to estimate the strength and direction of the associations between candidate predictors and in-hospital mortality. Finally, a nomogram was developed based on the final multivariable logistic regression model using R software to facilitate individualized risk prediction.

Model Evaluation and Validation

The discriminative ability of the model was assessed using receiver operating characteristic (ROC) curves and the corresponding area under the curve (AUC). Model calibration was evaluated using the Hosmer–Lemeshow goodness-of-fit test and calibration curves. Calibration curves were generated using bootstrap resampling (B = 1000) to assess the agreement between predicted and observed outcomes.

Decision curve analysis (DCA) was performed to evaluate the clinical net benefit of the model across a range of threshold probabilities. All statistical analyses were conducted using R software (version 4.5.2). All tests were two-sided, and a P value <0.05 was considered statistically significant.

Results

Baseline Characteristics

A total of 2167 patients hospitalized with AECOPD were included in this study, among whom 109 (5.0%) died during hospitalization and 2058 (95.0%) survived. The baseline demographic characteristics, comorbidities, and laboratory parameters of the two groups are summarized in Table 1. Compared with survivors, non-survivors were generally older and had lower hemoglobin and serum albumin levels. They also had a higher prevalence of respiratory failure at admission and a history of lung cancer.

Table 1.

Baseline Characteristics of Hospitalized Patients with AECOPD Stratified by in-Hospital Mortality

Variable	Survivors (n = 2058)a	Non-Survivors (n = 109)a	p-valueb
Sex			0.2
Male	1593 (77%)	90 (83%)
Female	465 (23%)	19 (17%)
Age (years)	76 (69, 81)	82 (74, 86)	<0.001
Respiratory failure at admission			<0.001
No	1617 (79%)	46 (42%)
Yes	441 (21%)	63 (58%)
History of lung cancer			0.003
No	2007 (98%)	100 (92%)
Yes	51 (2.5%)	9 (8.3%)
Hemoglobin (g/L)	128 (117, 140)	119 (102, 132)	<0.001
Albumin (g/L)	36.2 (33.5, 38.9)	31.9 (28.7, 35.2)	<0.001
Urea (mmol/L)	6.04 (4.80, 7.83)	8.50 (5.90, 12.80)	<0.001
Creatinine (μmol/L)	69 (57, 86)	72 (55, 100)	0.5
IL-6 (pg/mL)	8 (2, 25)	24 (6, 79)	<0.001

Notes: ^an (%); Median (Q1, Q3). ^bPearson’s Chi-squared test; Wilcoxon rank sum test; Fisher’s exact test.

Univariate Analysis

The results of the univariable logistic regression analysis are summarized in Table 2. Every 5-year increase in age was associated with a 39% increase in the risk of in-hospital death (OR = 1.39, 95% CI: 1.23–1.57, P < 0.001). Each 5 g/L decrease in hemoglobin was associated with a 13.5% increase in the risk of in-hospital mortality (OR = 1.135, 95% CI: 1.084–1.183, P < 0.001), and each 1 g/L decrease in serum albumin was associated with a 20.2% increase in mortality risk (OR = 1.202, 95% CI: 1.152–1.255, P < 0.001). In addition, higher blood urea nitrogen (OR = 1.009, 95% CI: 1.000–1.017, P = 0.018), higher serum creatinine (OR = 1.004, 95% CI: 1.001–1.007, P = 0.003), and higher IL-6 levels (OR = 1.001, 95% CI: 1.000–1.003, P = 0.024) were significantly associated with an increased risk of in-hospital death. Patients with respiratory failure at admission (OR = 5.455, 95% CI: 3.684–8.136, P < 0.001) or a history of lung cancer (OR = 4.320, 95% CI: 1.925–8.733, P < 0.001) also had a significantly higher mortality risk.

Table 2.

Univariable Logistic Regression Analysis of Candidate Predictors

Variable	OR	95% CI	P value
Age (per 5-year increase)	1.389	1.228–1.574	<0.001
Hemoglobin (per 5 g/L decrease)	1.135	1.084–1.183	<0.001
Albumin (per 1 g/L decrease)	1.202	1.152–1.255	<0.001
BUN	1.009	1.000–1.017	0.018
Creatinine	1.004	1.001–1.007	0.003
IL-6	1.001	1.000–1.003	0.024
Respiratory failure at admission	5.455	3.684–8.136	<0.001
History of lung cancer	4.320	1.925–8.733	<0.001

Abbreviations: OR, odds ratio; CI, confidence interval.

Multivariable Model

Variables with statistical significance in the univariate analysis and clinical relevance were entered into the multivariable logistic regression model (Table 3). The results showed that age (OR = 1.234, 95% CI: 1.077–1.422, P = 0.003), lower hemoglobin (OR = 1.095, 95% CI: 1.036–1.152, P = 0.001), lower albumin (OR = 1.135, 95% CI: 1.083–1.192, P < 0.001), respiratory failure (OR = 6.002, 95% CI: 3.936–9.234, P < 0.001), and a history of lung cancer (OR = 4.43, 95% CI: 1.815–9.855, P < 0.001) were independent risk factors for in-hospital death. In contrast, IL-6, blood urea nitrogen, and serum creatinine were no longer statistically significant in the multivariable model (P > 0.05). Overall, older age, lower hemoglobin, lower albumin, a history of lung cancer, and respiratory failure were associated with a significantly increased risk of in-hospital mortality (Table 3).

Table 3.

Multivariable Logistic Regression Analysis of Independent Predictors of in-Hospital Mortality

Variable	OR	95% CI	P value
Age (per 5-year increase)	1.234	1.077–1.422	0.003
Hemoglobin (per 5 g/L decrease)	1.095	1.036–1.152	0.001
Albumin (per 1 g/L decrease)	1.135	1.083–1.192	<0.001
Respiratory failure at admission	6.002	3.936–9.234	<0.001
History of lung cancer	4.430	1.815–9.855	<0.001

Abbreviations: OR, odds ratio; CI, confidence interval.

To further explore the effects of the major continuous variables, stratified logistic regression analyses were performed by categorizing these variables and entering them into the multivariable model (Table 4). After adjustment for other covariates, each 10-year increase in age was associated with a 64% increase in the risk of in-hospital death (OR = 1.64, 95% CI: 1.23–2.20, P = 0.001), indicating a stable positive association between age and mortality.

Table 4.

Stratified Multivariable Logistic Regression Analysis

Variable	OR	95% CI	P value
Age (per 10-year increase)	1.635	1.231–2.197	0.001
Hemoglobin (g/L)
≥130	1 (reference)	—	—
120–129	1.345	0.724–2.468	0.354
110–119	1.964	1.012–3.758	0.047
100–109	1.160	0.453–2.692	0.730
<100	4.030	2.052–7.881	<0.001
Albumin (g/L)
≥40	1 (reference)	—	—
35–39	0.814	0.346–2.139	0.670
30–34	1.763	0.792–4.496	0.170
<30	4.126	1.749–10.948	0.001
Respiratory failure at admission	6.439	4.122–10.174	<0.001
History of lung cancer	4.638	1.893–10.408	0.001

Note: Reference category is indicated as 1.

Abbreviations: OR, odds ratio; CI, confidence interval.

In the hemoglobin-stratified analysis, using Hb ≥130 g/L as the reference category, patients with Hb <100 g/L had a significantly increased risk of in-hospital death (OR = 4.03, 95% CI: 2.05–7.88, P < 0.001), while the Hb 110–119 g/L group also showed an elevated risk (OR = 1.96, 95% CI: 1.01–3.76, P = 0.047). Although the risk estimates across hemoglobin categories did not show a completely monotonic pattern, the overall trend suggested that lower hemoglobin levels were associated with a higher risk of death.

In the albumin-stratified analysis, using Alb ≥40 g/L as the reference category, patients with Alb <30 g/L had a significantly increased risk of in-hospital death (OR = 4.13, 95% CI: 1.75–10.95, P = 0.001). Although no statistically significant differences were observed in the remaining albumin strata, the overall trend still indicated that lower albumin levels were associated with increased mortality risk. In addition, respiratory failure (OR = 6.44, 95% CI: 4.12–10.17, P < 0.001) and a history of lung cancer (OR = 4.64, 95% CI: 1.89–10.41, P = 0.001) remained significant in the stratified analysis, consistent with the findings of the main model.

Model Performance

The area under the receiver operating characteristic curve (AUC) of the model was 0.811, indicating good discriminatory ability (Figure 1). The optimal cut-off value, determined according to the maximum Youden index, was 0.076. At this threshold, the sensitivity and specificity of the model were 65.1% and 84.2%, respectively. The Hosmer–Lemeshow goodness-of-fit test yielded a χ² value of 11.085 (P = 0.197), indicating no significant deviation between predicted and observed outcomes and suggesting satisfactory model calibration. The calibration curve also showed good agreement between predicted mortality probabilities and the observed mortality rates (Figure 2).

Figure 1.

Receiver operating characteristic (ROC) curve of the model for predicting in-hospital mortality in patients with AECOPD.

Figure 2.

Calibration curve of the model for predicting in-hospital mortality in patients with AECOPD.

Internal validation using bootstrap resampling (1000 repetitions) demonstrated that the bias-corrected calibration curve closely overlapped with the apparent calibration curve, indicating good model stability and predictive accuracy, with a mean absolute error of 0.005. Decision curve analysis further showed that within a threshold probability range of 5% to 50%, the model consistently provided a greater net benefit than either the “treat-all” or “treat-none” strategy, with a relatively stable curve pattern (Figure 3). These findings suggest that the model has potential clinical utility across a broad range of threshold probabilities by reducing unnecessary interventions while minimizing missed identification of high-risk patients.

Figure 3.

Decision curve analysis (DCA) of the model for predicting in-hospital mortality in patients with AECOPD.

Nomogram Construction

Based on the multivariable logistic regression results, a nomogram was constructed to predict the risk of in-hospital death (Figure 4). The nomogram included five predictors. Among them, respiratory failure at admission contributed the most to the total score, whereas hemoglobin and albumin influenced mortality risk in the opposite direction, with higher values corresponding to lower total scores. This nomogram may help clinicians visually and conveniently estimate the individual risk of in-hospital death.

Figure 4.

Nomogram for predicting in-hospital mortality in patients with AECOPD based on age, hemoglobin, albumin, respiratory failure at admission, and history of lung cancer.

Discussion

In the present study, we developed an internally validated prediction model for in-hospital mortality in hospitalized patients with AECOPD based on routinely available nutrition-related laboratory indicators together with key clinical characteristics. The final model identified older age, lower hemoglobin, lower serum albumin, respiratory failure at admission, and a history of lung cancer as independent predictors of in-hospital mortality. The model demonstrated good discrimination and calibration, suggesting that it may serve as a simple and clinically applicable tool for early risk stratification in hospitalized patients with AECOPD. Compared with previous prediction models, the present study focused on a concise set of routinely available laboratory indicators, which may facilitate bedside application while maintaining acceptable predictive performance.

In this study, age was analyzed as a continuous variable, and each 10-year increase was associated with a significant increase in the risk of in-hospital death, suggesting a stable positive relationship between age and mortality. This finding further supports the view that age is an important determinant of prognosis in patients with COPD.¹⁰ Compared with categorized analyses, modeling age as a continuous variable preserves more information and improves statistical efficiency, thereby providing a more accurate reflection of the true association between age and clinical outcome.

Notably, after adjustment for age and major comorbidities, both hemoglobin and albumin remained statistically significant in the multivariable model, indicating that they are independent predictors of in-hospital mortality. This finding is broadly consistent with previous reports.^17,18 From the perspective of routinely available laboratory biomarkers, the present study further quantified the association between the nutrition–inflammation axis and mortality risk. In particular, each 1 g/L decrease in serum albumin was associated with a significant increase in in-hospital mortality, suggesting that assessment based on simple laboratory parameters is clinically feasible and may provide valuable prognostic information in routine practice.

Regarding hemoglobin, the stratified analysis showed that patients with moderate-to-severe anemia (Hb <100 g/L) had a significantly higher risk of in-hospital death, and the Hb 110–119 g/L group also showed an elevated risk. Although the trend across all hemoglobin categories was not completely monotonic, the overall pattern still suggested that lower hemoglobin levels were associated with a higher mortality risk. This nonlinearity may be related to information loss after categorization, uneven sample distribution across strata, and redistribution of effects after multivariable adjustment. Previous studies have also shown that mortality is significantly higher in patients with COPD complicated by anemia than in those without anemia.^12,13,19 Martinez-Rivera et al further reported that anemia was an important predictor of post-discharge mortality in patients hospitalized for AECOPD.¹⁵ In addition to reflecting nutritional status, hemoglobin may also reflect chronic inflammation, anemia of chronic disease, iron metabolism, and the overall systemic burden associated with COPD. Therefore, early identification and appropriate management of anemia may be important for improving patient outcomes.

Serum albumin, as an important marker reflecting both nutritional status and systemic inflammation, demonstrated a stable and significant predictive effect in the present study. Patients with albumin levels <30 g/L had a markedly increased risk of in-hospital death, suggesting that hypoalbuminemia may be a key determinant of poor prognosis in COPD. Low albumin not only reflects malnutrition, but is also closely associated with chronic inflammation, impaired immune function, and reduced tissue repair capacity, all of which may contribute to adverse clinical outcomes. Furthermore, hypoalbuminemia may also reflect disease severity, frailty, increased capillary permeability, and the catabolic state associated with acute illness rather than nutritional deficiency alone. Therefore, the present findings should not be interpreted as evidence that nutritional intervention alone would necessarily reduce mortality, and prospective interventional studies are still required to determine whether improving nutritional status can improve clinical outcomes.

Other variables included in the model also have important clinical implications. A history of lung cancer was identified as an independent risk factor, which is in line with previous studies.^4,20 Antitumor treatment may have long-term adverse effects on lung function, while the risk of tumor recurrence may further increase disease complexity.^21,22 Respiratory failure at admission was one of the strongest predictors in the model, indicating that patients presenting with respiratory failure are at substantially increased risk of in-hospital death. Such patients should therefore receive closer monitoring and earlier intervention during hospitalization.

Compared with previous risk prediction models for mortality in AECOPD, the present model was relatively simplified while retaining key predictors. For example, Sun et al developed a nomogram incorporating a larger number of clinical variables and demonstrated good predictive performance. In contrast, our model focused on a limited number of routinely available nutrition-related laboratory indicators together with key clinical characteristics. Although the predictive performance was comparable to that reported in previous studies, the simplified variable set may facilitate bedside application and improve clinical practicality. Rather than incorporating a large number of variables, selecting a concise set of representative and clinically accessible predictors may be more meaningful in routine care.^23,24

Several limitations of this study should be acknowledged. First, this was a single-center retrospective study, and the data source was relatively limited. Although internal validation was performed, the model has not yet been validated in an external independent cohort, and its generalizability still requires further assessment in multicenter populations. Second, as a retrospective analysis, the available variables were restricted by the completeness of medical records. Some potentially important clinical information was not systematically collected or quantitatively incorporated, and residual confounding cannot be excluded. Radiological variables and pulmonary function parameters, including chest CT findings, GOLD stage, and FEV1, were not consistently available in this retrospective cohort and therefore could not be incorporated into the present model. Future studies integrating imaging findings and pulmonary function measurements may further improve predictive performance. Third, nutritional status was assessed only by two indirect laboratory indicators, namely hemoglobin and albumin, without more comprehensive measures such as body mass index or muscle mass. Although hemoglobin and albumin are readily available and have established prognostic relevance, a more comprehensive nutritional evaluation might provide additional value. Fourth, some acute-phase physiological parameters, including arterial blood gas results and certain inflammatory markers, were not included in the final model because of substantial missing data or a lack of independent predictive value during variable selection. In addition, because the study covered a relatively long inclusion period, changes in clinical practice over time may also have influenced patient outcomes. Future prospective studies with more systematic data collection may further improve model performance. Finally, it should be emphasized that this model was derived from correlational analyses. Although the findings suggest that poor nutritional status is associated with adverse outcomes, they do not imply that correction of malnutrition alone would necessarily reduce mortality. Whether nutritional intervention can improve outcomes in patients with COPD still requires confirmation in prospective studies.

Conclusion

In conclusion, we developed and internally validated a simple and practical prediction model for in-hospital mortality in hospitalized patients with AECOPD. The model incorporates five readily available variables: age, hemoglobin, albumin, history of lung cancer, and respiratory failure at admission, and demonstrated good discrimination and calibration. This model may serve as a practical bedside tool to assist clinicians in early risk stratification and individualized clinical decision-making. However, external validation in independent multicenter cohorts is warranted before widespread clinical implementation.

Funding Statement

This study was supported by the Wuhu Municipal Health Commission Project (No. 2020rkx4-3).

Declarations

This study is in accordance with the Helsinki Declaration.

Ethical Approval

This study was approved by the Ethics Committee of the Second People’s Hospital of Wuhu (Approval No. 2023-KY-026). Due to the retrospective and anonymized nature of the study, the requirement for informed consent was waived.

Author Contributions

All authors made a significant contribution to the work reported, whether that is in the conception, study design, execution, acquisition of data, analysis and interpretation, or in all these areas; took part in drafting, revising or critically reviewing the article; gave final approval of the version to be published; have agreed on the journal to which the article has been submitted; and agree to be accountable for all aspects of the work.

Disclosure

The authors declare no conflicts of interest in this work.