Derivation and Validation of a New Score for Predicting In-Hospital Mortality in Exacerbations of Chronic Obstructive Pulmonary Disease

Abstract

Background and Objective

Since exacerbation of chronic obstructive pulmonary disease (ECOPD) is both common and often fatal, accurate prognostication of patients hospitalized for ECOPD is critical. We aimed to develop and validate a new score for predicting hospital mortality.

Methods

Independent predictors of in-hospital mortality were identified from a large prospective cohort of ECOPD from 10 medical centers in China between September 2017 and July 2021, and incorporated into a clinical prediction score. Application of this score was then prospectively evaluated in an external validation cohort. The prognostic value of the scores was assessed by the area under the receiver operating characteristic curve (AUC).

Results

A total of 14,007 patients were included in the derivation cohort, and 201 patients (1.43%) died in the hospital. The 11 strongest independent predictors were combined to form the Chronic Heart failure, Age ≥ 75, altered Mental status, Pneumonia, Interstitial lung disease, diastolic blOod pressure ≤ 70 mmHg, Neutrophil ratio > 85 (%), Anemia, Long-term bed rest, hEart Rate > 100 (beats/minute), and blood urea niTrogen > 7.3 mmol/L, CHAMPION-ALERT Score, which enabled patients to be stratified according to increasing risk of in-hospital mortality: score 0–5, 0.3%; score 6–9, 4.4%; score 10–16, 25.9%. The score displayed excellent predictive value for mortality with an AUC of 0.89 (95% CI 0.87–0.92), which was reproduced in the validation cohort of 3048 patients. The discrimination of the new score was superior to or comparable to that of existing prognostic scores (DECAF, BAP-65, and CURB-65).

Conclusion

A new scoring system composes of 11 routinely available clinical variables on admission can accurately predict in‐hospital mortality in ECOPD and shows a trend toward better performance compared to previous prognostic scores.

Supplementary Information

The online version contains supplementary material available at 10.1007/s11606-025-09660-x.

INTRODUCTION

Chronic obstructive pulmonary disease (COPD) is a serious lung disease that affects health and quality of life, with high rates of illness and death¹. COPD worsening is an acute respiratory crisis requiring additional therapy². ECOPD accounts for the greatest proportion of the total COPD burden on the healthcare system³.Patients with frequent acute exacerbations have worse quality of life and a higher risk of dying than those with stable COPD^4–6. Because of the aging of the population, the burden of ECOPD is expected to rise.

A clinical prediction tool can help clinicians evaluate the prognosis of COPD patients, including high-risk patients who are transferred to higher levels of care when appropriate and low-risk patients who are discharged early⁷. Some scores have been proposed to predict short-term mortality in COPD patients, including BAP-65 and DECAF^7–10. The CURB-65 score can predict short-term mortality in patients hospitalized with COPD^11,12. However, none of the scores have been examined in a large Chinese population. Furthermore, some of the prognostic scores were derived in highly selected rather than unselected patients.

The purpose of this study was to develop and validate a simple prediction score capable of accurately stratifying unselected patients admitted for ECOPD based on the risk of in-hospital mortality, and to compare with several existing prognostic scores (DECAF, BAP-65, and CURB-65), through a large, multicenter derivation cohort and a prospective multicenter external validation cohort from China.

METHODS

Patients and Study Design

The derivation cohort was derived from the MAGNET ECOPD Registry study. This study included patients hospitalised for COPD in 10 large Chinese hospitals between September 2017 and July 2021. It examined management strategies and adverse outcomes¹³. It also identified independent risk factors for in-hospital mortality.

The external validation cohort comprised consecutive patients admitted between September 2021 and December 2023 at 5 tertiary hospitals not involved in the MAGNET study. To minimize selection bias, enrollment was conducted consecutively, with all medical admissions screened prospectively. In both the derivation and validation cohorts, ECOPD diagnosis was based on the following: (1) a history of COPD, including (i) exposure to risk factors (e.g., tobacco smoking, environmental exposures); (ii) chronic symptoms such as dyspnea, cough, or sputum production; (iii) post-bronchodilator spirometry (performed during stable disease) demonstrating an FEV1/FVC ratio < 70%; and (2) acute worsening of respiratory symptoms necessitating additional therapy. The exclusion criteria were as follows: (1) primary cause of hospitalization other than ECOPD; (2) age < 40 years; (3) refusal to sign the informed consent; and (4) incomplete clinical data. Patient admission and treatment were determined by attending physicians, with no additional interventions imposed by the study. The study was approved by each participating center’s ethics committee, with reference to the coordinating centre (West China Hospital of Sichuan University, 2019–1056). The study had adhered to the “Declarations of Helsinki.” Written informed consent was obtained from all the participants.

Data Collection

The MAGNET ECOPD registry collected data on patients admitted with ECOPD, including demographics, clinical characteristics, lab results, chest x-ray findings, management strategies, and outcomes. Most clinical variables were obtained within 24 h of admission. A case report form was completed for each patient. In the derivation cohort, the scores of the prediction tools were evaluated based on the information available at admission. The majority of variables in these prediction tools were already present in the case report form, except for the eMRCD score, which is required for the DECAF score. The MRC dyspnea score was used for inpatients with ECOPD in the MAGNET study to assess stable-state dyspnea. Patients with MRC dyspnea grade 5 and bedridden for 14 + days were classified as eMRCD grade 5b, while the remainder were grade 5a. The DECAF score used in the derivation cohort was modified with the eMRCD score (the “modified DECAF”). All investigators were well trained (included detailed instructions on how to interpret clinical criteria and apply the definitions uniformly across all study sites) and blinded to patients’ clinical outcomes.

In the validation cohort, the scores of the developed clinical prediction tool and the scores of DECAF, BAP-65, and CURB-65 were prospectively calculated by study operators who received in-depth training to ensure data reliability and consistency within 24 h of admission. The prediction probability does not refit the model in the validation cohort, but directly applies the model parameters of the derivation cohort (CHAMPION-ALERT) or the original study (DECAF, etc.) to ensure the independence of the external validation. The risk factors and points assigned are presented in Table S1.

Study Outcomes

The primary outcome was all cause in-hospital mortality. Secondary outcomes involved invasive mechanical ventilation, ICU admission, and length of stay (LOS).

Statistical Methods

The data were analyzed using R software (Version 4.2.0) or SPSS Version 26.0. Proportions, means with SD, or medians with interquartile ranges were used to characterize the patient sample. Bivariate comparisons were performed using Student’s t test (parametric data), the Mann–Whitney U test (nonparametric data), and the chi-square test (categorical data). Multiple imputations (MI) were conducted using the Mice package in R software (created 10 imputed datasets and performed the analysis separately in each dataset and combined the results using Rubin’s rules) to impute missing data (annotated in Table 1) if the missing values were less than 20%, and variables with a missing rate of more than 20% were excluded.

Table 1.

Population Description of the MAGNET (Derivation) Cohort and Comparison Between Survivors and Non-survivors

Variable	Total population, n = 14,007	Survivors, n = 13,806	Non-survivors, n = 201	P value*
Socio-demographic details
Age(years), mean (SD)	72.32(10.39)	72.20(10.34)	80.90(10.07)	< 0.001
Female, %	2987(21.3)	2942(21.3)	45(22.4)	0.711
Current smoker, %	2833(20.2)	2813(20.4)	20(10.0)	< 0.001
Body mass index(kg/m2), mean (SD)	21.58(4.07)	21.58(4.05)	20.94(5.40)	0.258
Markers of disease severity
Frequency of hospitalization due to COPD in past year (times ≥ 2), %	1726(12.3)	1699(12.3)	27(13.4)	0.630
MRCD 5, %	1616(11.5)	1567(11.4)	49(24.4)	< 0.001
Long-term oxygen therapy, %	528(3.8)	516(3.7)	12(6.0)	0.099
Long-term bed rest, %	1828(13.1)	1753(12.7)	75(37.3)	< 0.001
Long-term prednisolone, %	233(1.7)	231(1.7)	2(1.0)	0.639
Cor pulmonale, %	2905(20.7)	2828(20.5)	77(38.3)	< 0.001
Comorbidity
Hypertension, %	4679(33.4)	4585(33.2)	94(46.8)	< 0.001
Coronary heart disease, %	1546(11.0)	1508(10.9)	38(18.9)	< 0.001
Chronic heart failure, %	1609(11.5)	1534(11.1)	75(37.3)	< 0.001
Arrhythmia, %	1187(8.5)	1132(8.2)	55(27.4)	< 0.001
Atrial fibrillation, %	808(5.8)	770(5.6)	38(18.9)	< 0.001
Interstitial lung disease, %	403(2.9)	389(2.8)	14(7.0)	< 0.001
Bronchiectasis, %	1198(8.6)	1188(8.6)	10(5.0)	0.068
Pneumonia, %	3495(25.0)	3364(24.4)	131(65.2)	< 0.001
Diabetes, %	1804(12.9)	1751(12.7)	53(26.4)	< 0.001
Osteoporosis, %	347(2.5)	337(2.4)	10(5.0)	0.039
Stroke, %	822(5.9)	785(5.7)	37(18.4)	< 0.001
Lung Cancer, %	477(3.4)	468(3.4)	9(4.5)	0.399
Other malignancies, %	316(2.3)	310(2.2)	6(3.0)	0.644
Chronic renal insufficiency, %	514(3.7)	479(3.5)	35(17.4)	< 0.001
Anxiety and depression, %	106(0.8)	99(0.7)	7(3.5)	< 0.001
Sleep disorders, %	143(1.0)	137(1.0)	6(3.0)	0.015
Sepsis, %	58(0.4)	47(0.3)	11(5.5)	< 0.001
Admission clinical data +
Altered mental status, %	465(3.3)	392(2.8)	73(36.3)	< 0.001
Respiratory rate/min, mean (SD)	20.86(2.29)	20.86(2.26)	21.15(3.70)	0.260
Systolic blood pressure, mean (SD)	132.06(19.44)	132.12(19.34)	127.76(24.88)	0.014
Diastolic blood pressure, mean (SD)	78.83(12.58)	78.97(12.48)	69.35(15.38)	< 0.001
Heart rate/min, mean (SD)	89.13(16.63)	89.06(16.56)	93.52(20.80)	0.003
Laboratory investigations +
Red blood cell count (10^12/L), mean (SD)	4.28(0.79)	4.29(0.78)	3.77(0.92)	< 0.001
Hemoglobin(g/L), mean (SD)	127.28(22.53)	127.53(22.36)	109.53(26.51)	< 0.001
Anemia, %	3692(28.3)	3847(27.9)	115(57.2)	< 0.001
White blood cell count (> 10 or < 4) (10^9/L),%	4217(30.1)	4126(29.9)	91(45.3)	< 0.001
Neutrophil ratio, mean (SD)	74.34(12.76)	74.22(12.74)	82.67(11.43)	< 0.001
Eosinophil ratio, median (IQR)	1.00(0.10,2.60)	1.00(0.10,2.60)	0.20(0,1.30)	< 0.001
Eosinophil ratio < 2%, %	9220(65.8)	9061(65.6)	159(79.1)	< 0.001
Eosinopenia < 0.05 × 10^9/L, %	5390(42.3)	5804(42.0)	126(62.7)	< 0.001
Platelet count (10^9/L), mean (SD)	205.56(94.80)	205.96(94.82)	177.52(89.48)	< 0.001
Albumin(g/L), mean (SD)	36.73(5.55)	36.78(5.53)	33.17(5.71)	< 0.001
Blood urea nitrogen(mmol/L), median (IQR)	5.70(4.30,7.61)	5.64(4.30,7.60)	9.50(6.83,15.01)	< 0.001
Creatinine(umol/L), mean (SD)	98.53(88.43)	98.14(88.13)	126.02(104.62)	< 0.001
NT-proBNP(pg/ml), median (IQR)	320(103,1309.50)	311(101,1247)	2124(742.75,4808.50)	< 0.001
Troponin T(ng/L), median (IQR)	18.00(10.80,33.20)	17.50(10.60,32.20)	43.20(23.90,80.85)	< 0.001
D-dimer(mg/L), median (IQR)	0.73(0.40,1.53)	0.72(0.40,1.49)	2.26(1.12,5.18)	< 0.001
Procalcitonin(ng/ml), median (IQR)	0.10(0.05,0.26)	0.09(0.05,0.24)	0.34(0.14,1.53)	< 0.001
CRP(mg/L), median (IQR)	12.40(3.88,47.40)	12.00(3.84,45.50)	57.80(21.10,110.00)	< 0.001
Sodium(mmol/L), mean (SD)	138.56(5.15)	138.57(5.11)	137.80(7.33)	0.146
Potassium(mmol/L), mean (SD)	3.91(0.54)	3.91(0.54)	4.11(0.77)	< 0.001
Chlorine(mmol/L), mean (SD)	100.53(6.12)	100.56(6.06)	98.58(9.07)	0.003
pH, median (IQR)	7.41(7.38,7.45)	7.41(7.38,7.45)	7.41(7.36,7.45)	0.071
PaCO2(mmHg), mean (SD)	45.29(14.21)	45.22(14.11)	48.73(18.13)	0.015
PaCO2 > 50 mmHg	2380(17.0)	2315(16.8)	65(32.3)	< 0.001
PaO2(mmHg), mean (SD)	90.34(33.23)	90.29(33.12)	94.09(39.56)	0.206
Chest radiograph +
Pleural effusion, %	2619(18.7)	2535(18.4)	84(41.8)	< 0.001
Consolidation, %	868(6.2)	837(6.1)	31(15.4)	< 0.001
In-hospital outcomes
LOS, median (IQR)	9(6,13)	9(6,13)	15(8,28)	< 0.001
Invasive mechanical ventilation, %	462(3. 3)	390(2. 8)	72(35. 8)	< 0.001
ICU admission, %	1035(7. 4)	957(6. 9)	78(38. 8)	< 0.001

There were missing data for body mass index, MRCD, and variables in the admission clinical data, laboratory investigations, and chest radiograph

SD standard deviation, IQR inter-quartile range, ECOPD exacerbation of chronic obstructive pulmonary disease, MRCD Medical Research Council (MRC) dyspnea, NT-proBNP N-terminal pro-brain natriuretic peptide, CRP C-reactive protein, PaCO2 partial arterial carbon dioxide pressure, PaO2 partial arterial oxygen pressure, LOS length of stay, ICU intensive care unit

Neutrophil ratio, percentage of neutrophils in the white blood cells. Eosinophil ratio, percentage of eosinophils in the white blood cells

Long-term bed rest means bed rest ≥ 14 d before admission for illness

Altered mental status was defined as a Glasgow Coma Scale score lower than 14 or a designation of disoriented, stupor, or coma by a physician

Anemia with hemoglobin less than 110 g/L in females and less than 120 g/L in males

Chronic heart failure is heart failure caused by reasons other than chronic pulmonary heart disease, such as coronary heart disease, hypertension, cardiomyopathy, and various forms of arrhythmia

Arrhythmias include common heart rhythm abnormalities such as atrial fibrillation, atrial flutter, and supraventricular tachycardia

Other malignancies refer to non-hematologic malignancies except lung cancer

All of the above diagnoses were made by attending physicians and confirmed by trained study investigators

^*Comparison of survivors and those who died in hospital

⁺At the time of hospital admission

To explore the independent prognostic factors of in-hospital mortality in the derivation cohort, 41 factors based on clinical relevance and prior literature on ECOPD outcomes were involved in LASSO regression analysis (optimal regularization was determined through tenfold cross-validation). LASSO was used to filter variables and minimise risk of overfitting. The variables (with non-zero coefficients at λ_min) identified by LASSO analysis were further included in a logistics regression (Enter Method) model to identify in-hospital mortality factors. Odds ratios were calculated with 95% confidence intervals. To develop a predictive tool, continuous variables were converted to binary. Cut-off points identified using the area under the curve (AUC) and the Youden index. Alternatively, a clinically relevant or median split was used.

For pragmatic reasons, the 11 independent predictors with the regression coefficients greater than 0.4 derived from multivariate logistic analysis were included in the final clinical prediction tool. Scores were assigned according to the regression coefficient, with minor adjustments made¹⁴. A model was constructed using the strongest categorical predictors of mortality. The prognostic tools’ performance was evaluated by comparing the AUCs between our clinical prediction score and the published BAP-65, DECAF, and CURB-65 models¹⁵. Furthermore, calibration curve (Bootstrap Procedure, to address potential overfitting and improve the reliability of calibration curves) and decision curve analyses were conducted to assess the goodness of fit, accuracy, and clinical applicability of the developed predictive score and the published models in the derivation and/or validation cohorts. Clinically relevant risk thresholds for mortality prediction in ECOPD typically range from 1% (identifying low-risk patients for early discharge) to 60% (guarding against undertreatment of high-risk patients requiring intensive care). To align with these practical decision-making scenarios, the DCA x-axis was restricted to 1–60%, while the y-axis (net benefit) was truncated at 40% to enhance visual clarity. Excluding patients with pneumonia, sensitivity analyses were also conducted. A two-tailed p-value of less than 0.05 was deemed statistically significant for each analysis.

RESULTS

Patient Characteristics

A total of 14,007 patients were enrolled in the MAGNET ECOPD registry study. The mean age of the participants was 72.3 years (SD 10.4) and 21.3% were female. More than 20% of the patients were current smokers. A total of 201 (1.43%) patients died within the hospital, 462 patients (3.3%) received invasive mechanical ventilation, and 1035 (7.4%) were admitted into the ICU during hospitalization. The median LOS was 9 days. Table 1 describes the characteristics of all patients in the MAGNET (derivation) cohort and demonstrates the differences between the survivors and non‐survivors.

The validation cohort composed of 3048 consecutive patients hospitalized for ECOPD from 5 large tertiary hospital not participating the MAGNET study (the derivation cohort) between September 2021 and December 2023. In the validation cohort (n = 3048), 55 patients (1.8%) died during hospitalization, compared to 1.4% in the derivation cohort. The characteristics of the validation cohort are compared with those of the derivation cohort in Table S2.

Factors Associated with In-Hospital Mortality

Cross-validation of LASSO regression was performed for variables based on clinical relevance and prior literature on ECOPD outcomes in Table 1. After LASSO regression selection (Figure S1), 22 variables remained significant predictors of in-hospital mortality, including age ≥ 75, chronic heart failure, interstitial lung disease, chronic pulmonary heart disease, pneumonia, diabetes, long-term bed rest, altered mental status, diastolic blood pressure ≤ 70 mmHg, heart rate > 100 (beats/minute), anemia, neutrophil ratio > 85 (%), blood urea nitrogen > 7.3 mmol/L, arrhythmia, stroke, chronic renal insufficiency, sepsis, white blood cell count > 10 or < 4 (10^9/L), platelet count < 100 (10^9/L), serum albumin < 35 (g/L), D-dimer > 0.5 (mg/L), and potassium > 5.5 (mmol/L). Inclusion of these 22 variables in a logistic regression model resulted in 13 variables that were independently statistically significant predictors for in-hospital mortality in these unselected inpatients with ECOPD (Table 2). These variables included age ≥ 75 (OR, 2.33; 95% CI, 1.62–3.34; P < 0.001), chronic heart failure (OR, 1.76; 95% CI, 1.25–2.49; P = 0.001), interstitial lung disease (OR, 1.93; 95% CI, 1.04–3.58; P = 0.038), chronic pulmonary heart disease (OR, 1.44; 95% CI, 1.05–1.99; P = 0.025), pneumonia (OR, 2.70; 95% CI, 1.96–3.72; P < 0.001), diabetes (OR, 1.49; 95% CI, 1.04–2.12; P = 0.028), long-term bed rest (OR, 1.66; 95% CI, 1.18–2.33; P = 0.003), altered mental status (OR, 5.01; 95% CI, 3.48–7.23; P < 0.001), diastolic blood pressure ≤ 70 mmHg (OR, 2.02; 95% CI, 1.48–2.75; P < 0.001), heart rate > 100 (beats/minute) (OR, 1.66; 95% CI, 1.19–2.32; P = 0.003), anemia (OR, 1.61; 95% CI, 1.16–2.22; P = 0.004), neutrophil ratio > 85 (%)(OR, 1.97; 95% CI, 1.41–2.74; P < 0.001), and blood urea nitrogen > 7.3 mmol/L (OR, 2.20; 95% CI, 1.57–3.08; P < 0.001). Figure S1 Feature selection using the least absolute shrinkage and selection operator (LASSO) binary logistic regression model. (A) LASSO coefficient profiles of the 41 baseline features. (B) Tuning parameter (λ) selection in the LASSO model used tenfold cross-validation via minimum criteria. Figure S1 Feature selection using the least absolute shrinkage and selection operator (LASSO) binary logistic regression model. (A) LASSO coefficient profiles of the 41 baseline features. (B) Tuning parameter (λ) selection in the LASSO model used tenfold cross-validation via minimum criteria. Figure S1 Feature selection using the least absolute shrinkage and selection operator (LASSO) binary logistic regression model. (A) LASSO coefficient profiles of the 41 baseline features. (B) Tuning parameter (λ) selection in the LASSO model used tenfold cross-validation via minimum criteria.

Table 2.

Multivariable Logistic Regression for Predictors of In-Hospital Mortality

Variable	B	Odds ratio (95% CI)	Significance
Age ≥ 75	0.846	2.330(1.624,3.344)	< 0.001
Chronic heart failure	0.565	1.760(1.246,2.487)	0.001
Interstitial lung disease	0.656	1.927(1.036,3.581)	0.038
Chronic pulmonary heart disease	0.367	1.444(1.047,1.991)	0.025
Pneumonia	0.992	2.698(1.956,3.721)	< 0.001
Diabetes	0.397	1.487(1.043,2.120)	0.028
Long-term bed rest	0.507	1.659(1.183,2.327)	0.003
Altered mental status	1.612	5.013(3.476,7.230)	< 0.001
Diastolic blood pressure ≤ 70 mmHg	0.702	2.017(1.479,2.750)	< 0.001
Heart rate > 100 (beats/minute)	0.508	1.663(1.194,2.315)	0.003
Anemia	0.473	1.605(1.159,2.222)	0.004
Neutrophil ratio > 85 (%)	0.677	1.968(1.414,2.739)	< 0.001
Blood urea nitrogen > 7.3 mmol/L	0.789	2.201(1.571,3.084)	< 0.001
Arrhythmia	0.214	1.238(0.843,1.817)	0.276
Stroke	0.394	1.483(0.966,2.276)	0.072
Chronic renal insufficiency	0.191	1.211(0.770,1.904)	0.408
Sepsis	0.700	2.014(0.855,4.743)	0.109
White blood cell count > 10 or < 4 (10^9/L)	− 0.236	0.790(0.566,1.102)	0.165
Platelet count < 100 (10^9/L)	− 0.251	0.778(0.495,1.224)	0.278
Serum albumin < 35 (g/L)	0.264	1.302(0.939,1.805)	0.114
D-dimer > 0.5 (mg/L)	0.308	1.360(0.902,2.052)	0.143
Potassium > 5.5 (mmol/L)	0.620	1.859(0.822,4.203)	0.136
Intercept	− 7.401

95% CI 95% confidence interval. B: regression coefficient

Development of the Prediction Score

To simplify the model and enhance clinical utility, we selected 11 variables with regression coefficients greater than 0.4 to develop the final prediction tool. Weights were assigned based on regression coefficients. Table 3 illustrates the CHAMPION-ALERT score, based on Chronic Heart failure, Age ≥ 75, altered Mental status, Pneumonia, Interstitial lung disease, diastolic blOod pressure ≤ 70 mmHg, Neutrophil ratio > 85 (%), Anemia, Long-term bed rest, hEart Rate > 100 (beats/minute), and blood urea niTrogen > 7.3 mmol/L. Altered mental status were assigned 3 points because of their high risk of in-hospital mortality; age ≥ 75, pneumonia, and blood urea nitrogen > 7.3 mmol/L were assigned 2 points; and other variables were assigned 1 point each. A 16-point score was devised to enable patients to be stratified according to mortality risk, with rates of 0.3% (low risk), 4.4% (intermediate risk), and 25.9% (high risk) (Fig. 1). Table 4 shows the secondary outcomes for patients with different CHAMPION-ALERT scores and risk levels. Those with higher scores and levels were more likely to need invasive ventilation, ICU admission, and longer stays.

Table 3.

The CHAMPION-ALERT Score

Variable	B	Score
Chronic heart failure	0.674	1
Age ≥ 75	0.920	2
altered Mental status	1.764	3
Pneumonia	1.090	2
Interstitial lung disease	0.731	1
diastolic blOod pressure ≤ 70 mmHg	0.766	1
Neutrophil ratio > 85 (%)	0.642	1
Anemia	0.581	1
Long-term bed rest	0.578	1
hEart Rate > 100 (beats/minute)	0.543	1
blood urea niTrogen > 7.3 mmol/L	0.869	2
Total CHAMPION-ALERT Score		16

CHAMPION-ALERT, Chronic Heart failure, Age ≥ 75, altered Mental status, Pneumonia, Interstitial lung disease, diastolic blOod pressure ≤ 70 mmHg, Neutrophil ratio > 85 (%), Anemia,

Long-term bed rest, hEart Rate > 100 (beats/minute), blood urea niTrogen > 7.3 mmol/L

Figure 1.

In-hospital mortality rate by CHAMPION-ALERT risk group.

Table 4.

In-Hospital Mortality and other Adverse Outcomes at Each Scoring Level of CHAMPION-ALERT

Score	Population, %	In-hospital mortality, %	Risk groups	Invasive mechanical ventilation, %	ICU admission, %	Median length of stay, median (IQR)	Mortality by risk group, %
0	1973(14.1)	1(0.1)	Low risk	131(1.1)	544(4.7)	9(6,12)	39(0.3)
1	1811(12.9)	0(0)
2	2332(16.6)	3(0.1)
3	2295(16.4)	10(0.4)
4	1817(13.0)	14(0.8)
5	1382(9.9)	11(0.8)
6	958(6.8)	23(2.4)	Intermediate risk	214(10.0)	365(17.1)	12(7,18)	94(4.4)
7	595(4.2)	30(5.0)
8	368(2.6)	26(7.1)
9	213(1.5)	15(7.0)
10	126(0.9)	23(18.3)	High risk	117(44.5)	126(47.9)	16(9,25)	68(25.9)
11	63(0.4)	14(22.2)
12	35(0.2)	12(34.3)
13	29(0.2)	13(44.8)
14 +	10(0.1)	6(60.0)

Risk groups: low = CHAMPION-ALERT score 0–5; intermediate = CHAMPION-ALERT score 6–9; high = CHAMPION-ALERT score 10–16

ICU intensive care unit, IQR inter-quartile range

The area under the CHAMPION-ALERT receiver operating characteristic curve for predicting in-hospital mortality was 0.89 (95% CI 0.87 to 0.92). The calibration curve showed good predictive consistency between the predicted probability of in-hospital death by the score and the actual probability (Figure S2, left). Decision curve analysis demonstrated that the application of the score yielded superior overall net benefits across a wide range of threshold probabilities (Figure S2, right).

Risk groups: low risk = CHAMPION-ALERT score 0–5; intermediate risk = CHAMPION-ALERT score 6–9; high risk = CHAMPION-ALERT score 10–16. Figure S2 Calibration curve analysis and decision curve analysis for the CHAMPION-ALERT score in the derivation cohortFigure S2 Calibration curve analysis and decision curve analysis for the CHAMPION-ALERT score in the derivation cohort. Figure S2 Calibration curve analysis and decision curve analysis for the CHAMPION-ALERT score in the derivation cohort

External Validation and Comparison of the Prediction Score

When tested prospectively in the temporally and spatially distinct external validation cohort, the new scoring system (CHAMPION-ALERT) reproduced an excellent performance as in the derivation cohort, with an AUC of 0.89 (95% CI 0.84–0.93), superior discrimination to DECAF (AUC = 0.89 vs. 0.74, P < 0.001) and BAP-65 (AUC = 0.89 vs. 0.80, P = 0.012), and comparable performance to CURB-65 (AUC = 0.89 vs. 0.85, P = 0.132) in the validation cohort (Table S3 left, Fig. 2 left). The calibration curve (Figure S3) and decision curve (Figure S4) analysis also demonstrated the accuracy and applicability of the new predictive score in the validation cohort. Calibration curve analysis and decision curve analysis of previous scores (DECAF, BAP-65, CURB-65) in the validation cohort are presented in Figure S3 and Figure S4.

Figure 2.

Receiver operator characteristic (ROC) curves of prognostic scores for in-hospital mortality among the entire population (left) and patients excluding pneumonia (right) in the validation cohort.

Since pneumonia as a comorbidity of ECOPD is controversial, we conducted a sensitivity analysis excluding patients with pneumonia diagnosis. We found the CHAMPION-ALERT was still a significantly stronger predictor of in-hospital mortality with an AUC of 0.85 (95% CI 0.75–0.94), outperforming DECAF (AUC = 0.85 vs. 0.65, P = 0.003), and was non-inferior to BAP-65 (AUC = 0.85 vs. 0.79, P = 0.278) and CURB-65 (AUC = 0.85 vs. 0.83, P = 0.724) among patients hospitalized for ECOPD without pneumonia in the validation cohort (Table S3 right, Fig. 2 right).

DISCUSSION

Through a multicenter derivation cohort and a prospective multicenter validation cohort of patients hospitalized with exacerbations of COPD (ECOPD), we developed and validated the CHAMPION-ALERT score, a bedside tool incorporating 11 routinely available admission variables, chronic heart failure, age ≥ 75, altered mental status, pneumonia, interstitial lung disease, diastolic blood pressure ≤ 70 mmHg, neutrophil ratio > 85 (%), anemia, long-term bed rest, heart rate > 100 (beats/minute), and blood urea nitrogen > 7.3 mmol/L. The score reproduced excellent discrimination for in-hospital mortality, with AUCs of 0.89 (95% CI 0.84–0.93) and 0.85 (95% CI 0.75–0.94) in the entire validation cohort and pneumonia-excluded subgroup, respectively. Risk stratification using CHAMPION-ALERT identified distinct mortality tiers: low (score 0–5; mortality = 0.3%), intermediate (score 6–9; mortality = 4.4%), and high risk (score 10–16; mortality = 25.9%).

Some independent predictors of in-hospital mortality among inpatients with ECOPD identified by our study, such as interstitial lung disease (ILD), anemia, and diastolic hypotension, are rarely studied as a risk factor for short-term mortality in prior ECOPD studies. The lack of published evidence about the association of ILD with adverse outcomes of ECOPD may be attributed to the strict exclusion of such patients in previous randomized controlled trial studies and even in some cohort studies. The mechanisms of the development of anemia in COPD are complex, and systemic inflammation causing erythropoietin resistance and disordered iron metabolism may play a role in the process^16–19, Thus, anemia could be accepted as a surrogate of severe systemic inflammation and might be helpful in identifying sicker patients during the acute exacerbation of COPD. Notably, diastolic hypotension (≤ 70 mmHg) outperformed systolic measures, aligning with findings from the DECAF⁹ and Spannella et al²⁰ cohort, where low blood pressure signaled hemodynamic compromise¹³. Existing scores for predicting in-hospital mortality risk in patients with AECOPD face limitations: CURB-65 was derived for pneumonia, BAP-65 excluded critically ill patients, and DECAF relies on stable-state dyspnea assessments—often impractical during acute exacerbations. CHAMPION-ALERT score developed by this study circumvents these issues by using objective, admission-based variables, optimizing cutoffs (e.g., age ≥ 75 vs. ≥ 65, urea > 7.3 mmol/L vs. > 7 mmol/L in CURB-65) for elderly ECOPD populations. The CHAMPION-ALERT score outperformed established prognostic tool, DECAF (AUC = 0.74 vs. 0.89, P < 0.001), and BAP-65(AUC = 0.80 vs. 0.89, P = 0.012), and performed comparably to CURB-65 (AUC = 0.85 vs. 0.89, P = 0.132) in the validation cohort.

The CHAMPION-ALERT score advances precision medicine in acute exacerbations of chronic obstructive pulmonary disease (ECOPD) by translating multivariate risk profiles into actionable clinical prognostic stratification. Its stratification of patients into low-risk (score 0–5; 82.9% of the cohort; mortality 0.3%) and high-risk (score ≥ 10; 1.9% of the cohort; mortality = 25.9%) subgroups addresses a critical need for tailored resource allocation. For low-risk patients, early supported discharge protocols could reduce median length of stay (LOS) from 9 to < 7 days, mitigating hospital-acquired complications and optimizing bed utilization—an imperative in resource-constrained settings. Conversely, high-risk patients demand timely escalation to intensive monitoring, advanced therapies, or palliative care, aligning with value-based healthcare principles. The proportion of ECOPD patients with coexisting pneumonia is high, and the trend is increasing yearly²¹. Coexisting pneumonia amplifies systemic inflammation, accelerates respiratory failure, and complicates therapeutic responses²². Pneumonic ECOPD is not simply treated as pneumonia, but requires specific management of the ECOPD, including controlled oxygen therapy, corticosteroids, nebulized bronchodilators, and, if respiratory acidemia is present, noninvasive ventilation²³. The inclusion of pneumonia reflects real-world clinical overlap between ECOPD and respiratory infections^21,22,24,25. While debate persists about whether pneumonia represents a distinct entity^26–29, our data reinforce its prognostic relevance in ECOPD. Importantly, the score retained accuracy in pneumonia-excluded subgroups (AUC = 0.85), supporting its applicability across heterogeneous ECOPD phenotypes. The CHAMPION-ALERT score incorporates pneumonia as a key variable, acknowledging its real-world prevalence and mortality impact without sacrificing model performance in pneumonia-negative cases. Furthermore, the score’s inclusion of objective parameters like advanced age (≥ 75), an elevated percentage of neutrophils, increased blood urea nitrogen levels, anemia, diastolic hypotension (≤ 70 mmHg), and increased heart rate (> 100 beats/min) facilitates rapid, data-driven decisions at admission. Future studies should explore dynamic recalibration of the CHAMPION-ALERT score during hospitalization, particularly in patients with evolving infections, to refine mortality predictions. Additionally, the high prevalence of chronic comorbidities (e.g., heart failure, 11.5% in derivation cohort) underscores the necessity for multidisciplinary care models that address both pulmonary and systemic contributors to mortality.

The strength of this study lies in the large sample size, the prospective and systematic collection of relevant clinical information, the use of optimal statistical methods for selecting associated risk factors (thus avoiding spurious associations), and the inclusion of unselected patients admitted for ECOPD. Besides, external validation with cohort from the spatially and temporally separate hospital was performed. The strong and consistent performance of the CHAMPION-ALERT emphasizes the external validity and generalizability of the score. However, our study does have some shortcomings. First, information on certain COPD-specific covariates, particularly FEV1/FVC and FEV1%Pred, were missing in many patients. Nevertheless, spirometry data often are not available to physicians in the real-life management of ECOPD, and our goal was to create a score that could be used easily and broadly using routinely available data. More importantly, it remains unclear whether and how FEV1/FVC and FEV1%Pred relates to outcomes in ECOPD. Second, the lack of follow-up data prevented us from further evaluating the association of CHAMPION-ALERT with long-term outcomes in patients with ECOPD after discharge. Additionally, external validation in non-Chinese populations is warranted to address global demographic and practice variations influencing ECOPD outcomes, and comparison with newer models (e.g., machine learning approaches), though not explored in this study, could enhance prognostic precision in future work.

The CHAMPION-ALERT score is a simple, validated tool for predicting in-hospital mortality in ECOPD. Its integration of objective, readily available variables facilitates timely clinical decisions. Future studies should validate its utility in diverse populations other than Chinese and evaluate its impact on patient outcomes.

Supplementary Information

Below is the link to the electronic supplementary material. Dichotomization criteria for continuous variables in the clinical prediction model and definition of predictors used in developing or validating the multivariable prediction model.

Author Contribution:

Lige Peng: writing—original draft; analysis and interpretation of the data.

Qun Yi: conceptualization funding acquisition; project administration; supervision; writing—original draft; writing—review and editing.

Yuanming Luo: investigation, software, writing—original draft.

Hailong Wei: investigation, resources, writing—original draft.

Huiqing Ge: investigation, resources, writing—original draft.

Huiguo Liu: investigation, resources, writing—original draft.

Jianchu Zhang: investigation, resources, writing—original draft.

Xianhua Li: investigation, resources, writing—original draft.

Xiufang Xie: investigation, writing—original draft.

Pinhua Pan: investigation, resources, writing—original draft.

Mengqiu Yi: investigation, writing—original draft.

Lina Cheng: investigation, writing—original draft.

Hui Zhou: investigation, writing—original draft.

Liang Liu: investigation, writing—original draft.

Chen Zhou: investigation, writing—original draft.

Haixia Zhou: conceptualization funding acquisition, project administration, conception and design, revising it critically for intellectual content.

All authors agree to be accountable for all aspect of the work.

Funding:

This study was supported by the Suzhou Collaborative Medical Health Foundation (Y117) and National Key Research Program of China (2016YFC1304202).

Data Availability:

Due to the inclusion of patient-identifiable information, the raw clinical data are subject to ethical restrictions and cannot be shared publicly. Qualified researchers who have a legitimate scientific need for the data may request access to anonymized data by contacting the corresponding author, Haixia Zhou (zhouhaixia@wchscu.cn). Access will be granted after signing a data use agreement and receiving approval from the relevant ethics committee.

Declarations:

Human/Animal Ethics Approval:

The Ethics Committee of each participating academic medical center approved the study based on the ethical approval obtained from the Ethics Committee of the coordinating center, West China Hospital of Sichuan University (No: 2019–1056).

Conflict of Interest:

The authors declare no competing interests.