Prediction of one-year post-discharge mortality in patients with acute exacerbation of chronic obstructive pulmonary disease based on clinical characteristics, blood gas indicators, and duration of non-invasive ventilation Authors: Li-Na Huang, Xiao-Hong Zhu, Jin-Mei He Published: 2026 ABSTRACT BackgroundPatients with acute exacerbation of chronic obstructive pulmonary disease (AECOPD) receiving non-invasive ventilation (NIV) represent the most severe group with the poorest prognosis. This study aimed to develop a predictive model for the early identification of patients at high risk of post-discharge mortality, in order to improve patient outcomes.MethodsThis study retrospectively included 148 AECOPD patients who underwent NIV and divided them into survival and death groups according to survival status within 1 year after discharge. Univariate analysis was conducted to compare baseline characteristics, blood gas indicators, and inflammatory markers between the two groups. Least absolute shrinkage and selection operator (LASSO) regression and Cox regression were further used to screen factors significantly associated with mortality and construct a prediction model. Finally, interaction analysis and survival analysis of the prediction model were performed.ResultsCompared with the survival group, patients in the death group had more severe respiratory dysfunction at admission, a higher comorbidity burden, more severe disease, more severe respiratory impairment and inflammatory response at both admission and discharge, and a longer duration of mechanical ventilation. Cox regression analysis showed that chronic kidney disease (CKD), higher modified Medical Research Council (mMRC) grade, higher Global Initiative for Chronic Obstructive Lung Disease (GOLD) stage, and longer duration of mechanical ventilation were independent risk factors for mortality, while an increase in ΔPaO2 [partial pressure of arterial oxygen (PaO2) at discharge − PaO2 at admission] was a protective factor. The mortality prediction model constructed based on these factors demonstrated excellent performance. Interaction analysis revealed a significant synergistic effect between the prediction model and the Acute Physiology and Chronic Health Evaluation II (APACHE II) score. Survival analysis showed that patients with higher model scores had lower survival rates.ConclusionsCKD, mMRC, GOLD, ΔPaO2, and duration of mechanical ventilation were significant predictors of one-year post-discharge mortality in AECOPD patients undergoing NIV. The prediction model demonstrated good predictive performance and showed a significant synergistic effect with the APACHE II score. This study provides a new reference for early identification of high-risk patients and improvement of prognosis in clinical practice. Highlight box Key findings • This study developed a predictive model for early identification of acute exacerbation of chronic obstructive pulmonary disease (AECOPD) patients receiving non-invasive mechanical ventilation who are at high risk of death within one year after discharge, and demonstrated that the model has good predictive performance. What is known and what is new? • Patients with AECOPD have a high post-discharge mortality rate, and early risk assessment is of great importance for improving prognosis. • This study established a predictive model based on clinical and blood gas indicators and found a significant synergistic effect with the APACHE II score. What is the implication, and what should change now? • This model can be used for early clinical risk stratification and may provide a reference for improving the prognosis of AECOPD patients. Introduction Chronic obstructive pulmonary disease (COPD) is a common chronic respiratory disease characterized by persistent airflow limitation, with chronic inflammation, airway remodeling, and destruction of lung parenchyma as its main pathological basis (1,2). As the disease progresses, patients often experience acute exacerbations [acute exacerbation of chronic obstructive pulmonary disease (AECOPD)], which in severe cases may lead to respiratory failure or even death, typically manifesting as worsening cough, increased sputum volume, changes in sputum characteristics, or aggravated dyspnea (3,4). AECOPD can be triggered by various factors, including respiratory tract infections (viral or bacterial), air pollution, climate changes, and other comorbidities such as heart failure or pulmonary embolism (5). Among them, patients requiring non-invasive ventilation (NIV) represent the most severe group with the poorest prognosis (6). Although some patients may be successfully weaned off and discharged after active treatment, their long-term survival after discharge remains concerning, with high rates of rehospitalization and mortality within 1 year (7). Although several studies have explored the factors influencing mortality in AECOPD patients (8-10), systematic evaluation of the risk of death within 1 year after discharge in patients receiving NIV remains relatively insufficient. Early identification of high-risk patients is of great clinical value for optimizing treatment strategies, prolonging survival, and improving quality of life. Current commonly used assessment indicators include age, inflammatory markers, and blood gas parameters at admission, but few studies have incorporated dynamic change indicators to comprehensively reflect the course of disease and treatment response. This study integrates baseline characteristics, disease features, and dynamic change indicators to systematically evaluate their impact on the risk of death within 1 year after discharge, aiming to provide more reliable evidence for early clinical intervention and optimization of treatment strategies. We present this article in accordance with the TRIPOD reporting checklist (available at https://jtd.amegroups.com/article/view/10.21037/jtd-2026-1-0290/rc). Methods Study population This retrospective study included 148 patients with AECOPD who were admitted to Longquan Branch of Sichuan Provincial People’s Hospital between January 2021 and June 2024. Inclusion criteria were: (I) diagnosis of AECOPD according to the Chinese guidelines for the diagnosis and treatment of chronic obstructive pulmonary disease (11); (II) following the Global Initiative for Chronic Obstructive Lung Disease (GOLD) guidelines, patients received NIV along with standard treatment (12); (III) age ≥18 years. Exclusion criteria were: (I) severe systemic diseases (e.g., advanced malignant tumors); (II) severe cardiac, hepatic, or renal dysfunction; (III) coexisting severe pulmonary diseases (e.g., lung cancer, severe pneumonia); (IV) severe psychiatric disorders (bipolar disorder, schizophrenia) or poor compliance; (V) missing mechanical ventilation data or survival status within 1 year after discharge. An independent dataset from July 2024 to January 2025 was additionally selected for external validation. This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. This study was approved by the Ethics Committee of Longquan Branch of Sichuan Provincial People’s Hospital (No. 202642). The research content involved in this research meets the requirements of medical ethics and academic morality of our hospital, and the research content is reasonable, the risks are controllable, and there are no violations. The relevant research carried out is in line with the safe, standardized, and true scientific research guiding principles, and in line with the requirements of the clinical research ethics code. Data collection Data were collected from the electronic medical record system and included age, sex, body mass index (BMI), smoking history, hypertension, diabetes, chronic kidney disease (CKD), coronary artery disease, disease duration, modified medical research council (mMRC) grade, GOLD stage, history of exacerbations in the past year, home oxygen therapy, previous mechanical ventilation, Charlson comorbidity index (CCI) on admission (13), Acute Physiology and Chronic Health Evaluation II (APACHE II) score, Glasgow coma scale (GCS) score (14), blood gas parameters on admission and discharge [pH, partial pressure of oxygen (PaO2), partial pressure of carbon dioxide (PaCO2), PaO2/fraction of inspired oxygen (PaO2/FiO2)], respiratory rate, white blood cell (WBC) count and neutrophil percentage, and duration of NIV (days). Least absolute shrinkage and selection operator (LASSO) regression and Cox regression analysis Significant variables from the univariate analysis and the differences in clinical and laboratory indicators at admission and discharge (denoted as Δ) were used as independent variables, and patient survival status was defined as the dependent variable. Data were randomly divided into a training set (75%) and a testing set (25%). Model construction and 10-fold cross-validation were performed in the training set. Variables with nonzero coefficients were extracted as the final predictors. The model was then evaluated in the testing set by calculating the confusion matrix and key performance metrics, including accuracy, sensitivity, specificity, and precision. The discriminative ability of the model was assessed by plotting receiver operating characteristic (ROC) curves and calculating the area under the curve (AUC). The final predictors were further entered into a Cox proportional hazards regression model to estimate hazard ratios (HRs) and 95% confidence intervals (CIs). Construction of mortality risk prediction model The mortality risk prediction model was constructed based on the significant factors and their coefficients identified from the Cox proportional hazards model. The discriminative ability and predictive performance of the model were evaluated using ROC curves. In addition, interaction analysis and survival analysis were performed. For the interaction analysis, patients were divided into high and low groups based on the median value of the overall predictive model score. Other continuous variables were also categorized according to their respective median values. Statistical analysis All analyses and visualizations were performed using R software version 4.4.1. Continuous variables were expressed as median (minimum–maximum) and compared using the independent-samples t-test or Mann-Whitney U test. Categorical variables were expressed as frequency (percentage) and compared using Fisher’s exact test or the Chi-squared test. A P value <0.05 was considered statistically significant. Results Differences in baseline characteristics between patients surviving and dying within 1 year after discharge The results showed that patients in the death group had a significantly higher proportion of CKD (29.09% vs. 7.53%, P=0.001), previous mechanical ventilation (16.36% vs. 2.15%, P=0.004), and longer disease duration (9 vs. 7 years, P=0.02). The proportion of patients with mMRC grade 4 (94.55% vs. 56.99%, P<0.001) and GOLD stage 4 (98.18% vs. 70.97%, P<0.001) was significantly higher in the death group, and the frequency of acute exacerbations in the past year was also higher (≥2 times or ≥1 hospitalization: 90.91% vs. 72.04%, P=0.01). In addition, the CCI (6 vs. 4, P=0.01) and APACHE II score (25 vs. 21, P=0.005) were significantly higher in the death group, whereas the GCS score was significantly lower (9 vs. 12, P=0.003). Other characteristics such as age, sex, and BMI showed no significant differences (Table 1). Table 1Differences in baseline characteristics between patients who survived and those who died within 1 year after dischargeVariableAll patients (n=148)Alive (n=93)Dead (n=55)P valueAge (years)71 [64–80]71 [64–80]74 [64–80]0.21Gender0.58 Male97 (65.54)63 (67.74)34 (61.82) Female51 (34.46)30 (32.26)21 (38.18)BMI (kg/m2)27.3 [20.7–32.6]27.1 [20.9–32.6]27.4 [20.7–32.6]0.64Smoking0.39 Yes94 (63.51)62 (66.67)32 (58.18) No54 (36.49)31 (33.33)23 (41.82)Hypertension0.10 Yes91 (61.49)52 (55.91)39 (70.91) No57 (38.51)41 (44.09)16 (29.09)Diabetes mellitus0.24 Yes39 (26.35)21 (22.58)18 (32.73) No109 (73.65)72 (77.42)37 (67.27)Chronic kidney disease0.001 Yes23 (15.54)7 (7.53)16 (29.09) No125 (84.46)86 (92.47)39 (70.91)Coronary artery disease0.59 Yes38 (25.68)22 (23.66)16 (29.09) No110 (74.32)71 (76.34)39 (70.91)Disease course (years)8 [1–16]7 [1–15]9 [2–16]0.02Modified Medical Research Council grade<0.001 343 (29.05)40 (43.01)3 (5.45) 4105 (70.95)53 (56.99)52 (94.55)GOLD grade<0.001 GOLD 328 (18.92)27 (29.03)1 (1.82) GOLD 4120 (81.08)66 (70.97)54 (98.18)Exacerbation history (past year)0.01 0–1, no admission31 (20.95)26 (27.96)5 (9.09) ≥2 or ≥1 admission117 (79.05)67 (72.04)50 (90.91)Home oxygen therapy0.24 Yes92 (62.16)54 (58.06)38 (69.09) No56 (37.84)39 (41.94)17 (30.91)Prior history of mechanical ventilation0.004 Yes11 (7.43)2 (2.15)9 (16.36) No137 (92.57)91 (97.85)46 (83.64)Charlson comorbidity index5 [2–8]4 [2–8]6 [2–8]0.01APACHE II22 [17–28]21 [17–28]25 [17–28]0.005Glasgow coma scale score11 [7–14]12 [8–14]9 [7–13]0.003Data are presented as median [minimum–maximum] or n (%). APACHE II, Acute Physiology and Chronic Health Evaluation II; BMI, body mass index; GOLD, Global Initiative for Chronic Obstructive Lung Disease. Differences in baseline characteristics between patients who survived and those who died within 1 year after discharge Data are presented as median [minimum–maximum] or n (%). APACHE II, Acute Physiology and Chronic Health Evaluation II; BMI, body mass index; GOLD, Global Initiative for Chronic Obstructive Lung Disease. Differences in clinical and laboratory indicators between patients surviving and dying within 1 year after discharge At admission, patients in the death group had lower pH (7.26 vs. 7.31, P<0.001), higher PaCO2 (83 vs. 80 mmHg, P=0.02), lower PaO2 (54 vs. 57 mmHg, P=0.02), and lower oxygenation index (PaO2/FiO2: 180 vs. 183, P=0.02), as well as higher respiratory rate (38 vs. 35 breaths/min, P<0.001), WBC count (15.1×109/L vs. 12.9×109/L, P=0.01), and neutrophil percentage (87.9% vs. 85.6%, P=0.048). At discharge, indicators improved in both groups, but pH and oxygenation index remained significantly lower in the death group, while PaCO2 and respiratory rate remained significantly higher. In addition, the WBC count, neutrophil percentage, and duration of NIV were significantly higher in the death group compared with the survival group (Table 2). Table 2Differences in clinical and laboratory parameters between patients who survived and those who died within 1 year after dischargeVariableAll patients (n=148)Alive (n=93)Dead (n=55)P valuepH On admission7.28 (7.21–7.36)7.31 (7.23–7.36)7.26 (7.21–7.35)0.001 At discharge7.38 (7.32–7.45)7.40 (7.33–7.45)7.37 (7.32–7.43)0.001PaCO2 (mmHg) On admission81 (73–91)80 (73–91)83 (73–91)0.02 At discharge54 (48–60)53 (48–60)55 (49–60)0.042PaO2 (mmHg) On admission56 (47–65)57 (49–65)54 (47–61)0.02 At discharge71 (63–79)71 (63–79)71 (63–79)0.69PaO2/FiO2 On admission181 (145–217)183 (145–217)180 (148–216)0.02 At discharge382 (343–427)387 (348–427)374 (343–415)0.02Respiratory rate (breaths/min) On admission37 (32–41)35 (32–39)38 (34–41)<0.001 At discharge19 (16–22)19 (16–22)20 (16–22)0.048White blood cell count (×109/L) On admission13.4 (7.2–18.4)12.9 (7.7–18.4)15.1 (7.2–18.3)0.01 At discharge8.0 (4.9–11.2)7.7 (4.9–11.2)8.8 (5.0–11.2)0.03Neutrophil percentage (%) On admission86.6 (81.0–92.0)85.6 (81.0–91.6)87.9 (81.2–92.0)0.048 At discharge69.3 (63.1–74.9)68.3 (63.1–74.9)70.2 (63.1–74.8)0.03Duration of non-invasive ventilation (days)5 (2–7)4 (2–7)5 (2–7)0.003Data are presented as median [minimum–maximum] or n (%). FiO2, fraction of inspired oxygen; PaCO2, partial pressure of arterial carbon dioxide; PaO2, partial pressure of arterial oxygen. Differences in clinical and laboratory parameters between patients who survived and those who died within 1 year after discharge Data are presented as median [minimum–maximum] or n (%). FiO2, fraction of inspired oxygen; PaCO2, partial pressure of arterial carbon dioxide; PaO2, partial pressure of arterial oxygen. Selection of significant factors associated with mortality by LASSO and Cox regression Because there were numerous significant factors (P<0.05) in “Differences in baseline characteristics between patients surviving and dying within 1 year after discharge” and “Differences in clinical and laboratory indicators between patients surviving and dying within 1 year after discharge” sections, variables with P<0.01 were selected as independent variables for the subsequent analysis. LASSO regression with cross-validation showed that as the regularization parameter λ [log(λ)] increased, model complexity decreased and the number of selected variables dropped (from 18 to 3) (Figure 1A). Considering that there were 55 death events in this study, and following the “events per variable (EPV) ≥10” rule to avoid overfitting, the maximum number of independent variables included in the multivariable Cox regression model was set to six, which corresponded to λ =0.142. In the testing set, the LASSO model yielded an AUC of 0.776 (95% CI: 0.622–0.930), an accuracy of 69.4%, a recall (sensitivity) of 61.5%, a specificity of 73.9%, and a precision of 57.1%, indicating good model performance (Figure 1B). The final variables selected were CKD, mMRC grade, GOLD stage, ΔPaO2, pH at discharge, and duration of NIV (Table 3). Construction and validation of the LASSO regression model. (A) Cross-validation curve of the LASSO regression model. (B) ROC curve of the LASSO model in the testing set. AUC, area under the curve; CI, confidence interval; LASSO, least absolute shrinkage and selection operator; ROC, receiver operating characteristic. Table 3Variable selection via LASSO regressionVariableS1CKD0.112mMRC grade 0.577GOLD grade0.364Prior history of mechanical ventilation –APACHE II–GCS score–ΔpH –ΔPaCO2 –ΔPaO2 −0.010ΔOI–ΔRR –ΔWBC –ΔNeutrophil –pH 0 –pH 1 −0.070RR 0 –Duration of NIV 0.016S1: non-zero coefficients variables. pH 0 represents the pH value on admission, pH 1 represents the pH value at discharge, and RR 0 represents the respiratory rate on admission. Δ represents the difference between the value at discharge and the value on admission. APACHE II, Acute Physiology and Chronic Health Evaluation II; CKD, chronic kidney disease; GCS, Glasgow coma scale; GOLD, Global Initiative for Chronic Obstructive Lung Disease; LASSO, least absolute shrinkage and selection operator; mMRC, modified Medical Research Council; NIV, non-invasive ventilation; OI, oxygenation index (PaO2/FiO2). PaCO2, partial pressure of arterial carbon dioxide; PaO2, partial pressure of arterial oxygen; RR, respiratory rate; WBC, white blood cell. Variable selection via LASSO regression S1: non-zero coefficients variables. pH 0 represents the pH value on admission, pH 1 represents the pH value at discharge, and RR 0 represents the respiratory rate on admission. Δ represents the difference between the value at discharge and the value on admission. APACHE II, Acute Physiology and Chronic Health Evaluation II; CKD, chronic kidney disease; GCS, Glasgow coma scale; GOLD, Global Initiative for Chronic Obstructive Lung Disease; LASSO, least absolute shrinkage and selection operator; mMRC, modified Medical Research Council; NIV, non-invasive ventilation; OI, oxygenation index (PaO2/FiO2). PaCO2, partial pressure of arterial carbon dioxide; PaO2, partial pressure of arterial oxygen; RR, respiratory rate; WBC, white blood cell. When these six variables were included in the Cox regression model, CKD (HR =8.492; 95% CI: 2.245–32.121; P=0.002), higher mMRC grade (HR =10.139; 95% CI: 2.467–41.661; P=0.001), and higher GOLD stage (HR =11.377; 95% CI: 1.354–95.562; P=0.02) were significantly positively associated with mortality risk. Additionally, each additional day of NIV increased the risk of death by 25.8% (HR =1.258; 95% CI: 1.092–1.448; P=0.001). In contrast, ΔPaO2 was a protective factor, with each unit increase associated with a 15.6% reduction in mortality risk (HR =0.844; 95% CI: 0.739–0.964; P=0.01). pH at discharge was not statistically significant (Table 4). Table 4Cox regression analysis of significant factors influencing mortalityFactorβStandard errorStatisticP valueHRCI-lowerCI-upperCKD2.1390.6793.1510.0028.4922.24532.121mMRC2.3160.7213.2130.00110.1392.46741.661GOLD2.4321.0862.2390.0211.3771.35495.562ΔPaO2−0.1700.068−2.5010.010.8440.7390.964pH 1−9.5706.518−1.4680.14<0.001<0.00124.663Duration of NIV0.2290.0723.1900.0011.2581.0921.448pH 1 represents the pH value at discharge. Δ represents the difference between the value at discharge and the value on admission. CI, confidence interval; CKD, chronic kidney disease; GOLD, Global Initiative for Chronic Obstructive Lung Disease; HR, hazard ratio; mMRC, modified Medical Research Council; NIV, non-invasive ventilation; PaO2, partial pressure of arterial oxygen. Cox regression analysis of significant factors influencing mortality pH 1 represents the pH value at discharge. Δ represents the difference between the value at discharge and the value on admission. CI, confidence interval; CKD, chronic kidney disease; GOLD, Global Initiative for Chronic Obstructive Lung Disease; HR, hazard ratio; mMRC, modified Medical Research Council; NIV, non-invasive ventilation; PaO2, partial pressure of arterial oxygen. Construction of the prediction model A mortality prediction model was constructed using the significant factors and their coefficients identified from the Cox regression model. The formula was as follows: prediction model = CKD × 2.139 + mMRC × 2.316 + GOLD × 2.432 − ΔPaO2 × 0.170 + duration of NIV × 0.229. The ROC curve of this model showed an AUC of 0.89 (95% CI: 0.839–0.94), indicating excellent predictive performance (Figure 2). Comparison with commonly used prognostic tools showed that the AUC of the proposed model was significantly higher than that of the CCI (AUC =0.618) and the APACHE II score (AUC =0.634) (Figure 3A,3B). DeLong test results indicated statistically significant differences in pairwise comparisons (all P<0.001), suggesting that the proposed model outperforms existing commonly used scoring systems in predictive ability. ROC curve of the prediction model. AUC, area under the curve; CI, confidence interval; ROC, receiver operating characteristic. Commonly used prognostic tools for mortality prediction. (A) ROC curve of CCI for predicting mortality rate. (B) ROC curve of APACHE II for predicting mortality rate. APACHE II, Acute Physiology and Chronic Health Evaluation II; AUC, area under the curve; CCI, Charlson comorbidity index; CI, confidence interval; ROC, receiver operating characteristic. In the external validation cohort, the proposed model also demonstrated good discriminative ability. ROC curve analysis showed that the AUC of the model in the external validation dataset was 0.828 (95% CI: 0.719–0.937) (Figure 4). Performance of the prediction model in the external validation cohort. AUC, area under the curve; CI, confidence interval. Interaction analysis of the prediction model Interaction analysis with other significant factors showed that only the interaction term with the APACHE II score was statistically significant (HR =13.125; 95% CI: 1.164–147.943; P=0.03). This indicates that the effect of a one-point increase in the prediction model score on mortality risk differs between patients with high and low APACHE II scores. The large HR suggests that the prediction model has a greater impact on mortality risk in patients with higher APACHE II scores (Table 5). Table 5Interaction analysis in the prediction modelInteraction termβStandard errorStatisticP valueHRCI-lowerCI-upper(Intercept)−2.0790.474−4.384<0.0010.1250.0490.317Group2.0100.6033.3360.0017.4672.29224.325APACHE II group−1.2531.123−1.1160.260.2860.0322.580Group*APACHE II group2.5751.2362.0830.0313.1251.164147.943(Intercept)−1.8330.539−3.4030.0010.1600.0560.460Group2.9610.6414.623<0.00119.3185.50467.799GCS group−1.2350.902−1.3700.170.2910.0501.702Group*GCS group0.1761.0350.1700.861.1920.1579.069(Intercept)−2.0790.612−3.3960.0010.1250.0380.415Group2.9370.6904.254<0.00118.8574.87272.980pH 0 group−0.6060.855−0.7090.470.5450.1022.915Group*pH 0 group0.1230.9930.1240.901.1310.1627.923(Intercept)−1.7490.542−3.2290.0010.1740.0600.503Group2.7110.6324.287<0.00115.0384.35551.932pH 1 group−1.3640.903−1.5110.130.2560.0441.500Group*pH 1 group0.6261.0350.6050.541.8700.24614.228(Intercept)−3.1140.723−4.309<0.0010.0440.0110.183Group3.4880.8224.244<0.00132.7276.536163.866RR 0 group1.3640.9031.5110.133.9130.66722.973Group*RR 0 group−0.8821.035−0.8520.390.4140.0543.147pH 0 represents the pH value on admission, pH 1 represents the pH value at discharge, and RR 0 represents the respiratory rate on admission. (Intercept) represents the intercept of the logistic regression model, and Group indicates high- and low-risk groups stratified according to the median value of the predicted model score. APACHE II, Acute Physiology and Chronic Health Evaluation II; CI, confidence interval; GCS, Glasgow coma scale; HR, hazard ratio; RR, respiratory rate. Interaction analysis in the prediction model pH 0 represents the pH value on admission, pH 1 represents the pH value at discharge, and RR 0 represents the respiratory rate on admission. (Intercept) represents the intercept of the logistic regression model, and Group indicates high- and low-risk groups stratified according to the median value of the predicted model score. APACHE II, Acute Physiology and Chronic Health Evaluation II; CI, confidence interval; GCS, Glasgow coma scale; HR, hazard ratio; RR, respiratory rate. Survival analysis Survival analysis showed that patients in the high prediction model score group had significantly lower survival rates compared with those in the low score group (Figure 5). Kaplan-Meier survival curves stratified by high and low prediction model scores. Discussion This study ultimately identified CKD, mMRC dyspnea scale, GOLD stage, ΔPaO2, and duration of NIV as significant factors associated with 1-year post-discharge mortality in AECOPD patients requiring NIV. Patients with CKD had a significantly higher risk of death after discharge, which may be explained by the fact that CKD is often accompanied by a certain degree of immune dysfunction and chronic systemic inflammation (15,16). Once infection occurs, pathogens are more difficult to eliminate, leading to more frequent and severe exacerbations and ultimately increased mortality. AECOPD itself represents an acute pulmonary inflammatory event, and when combined with the chronic systemic inflammation caused by CKD, a “superimposed” or “synergistic” effect may occur, resulting in more extensive lung tissue injury and multi-organ dysfunction, thereby elevating mortality risk. Furthermore, AECOPD patients undergoing NIV experience substantial cardiac load; concomitant CKD worsens cardiac function, reducing tolerance to hemodynamic fluctuations induced by mechanical ventilation. Patients with higher mMRC grades at admission also had significantly increased post-discharge mortality risk. Higher mMRC grades indicate longer COPD duration and more severe airway obstruction, which can cause pronounced dynamic pulmonary hyperinflation (DPH) and airflow limitation (17), typically irreversible. These patients experience dyspnea even during mild activity and thus reduce physical activity, leading to disuse muscle atrophy and further functional decline, creating a vicious cycle of “dyspnea-inactivity-declining fitness-worsening dyspnea” (18). Dyspnea also increases energy expenditure and may contribute to sarcopenia, complicating weaning and adversely affecting prognosis and survival after discharge. Similarly, higher GOLD stages at admission reflect very severe airflow limitation and pronounced DPH, essentially representing exhaustion of respiratory physiological reserve. Such patients are on the verge of respiratory decompensation, where even minor infections can have severe consequences, markedly increasing post-discharge mortality. ΔPaO2 was negatively associated with post-discharge mortality in this study. As a dynamic indicator, ΔPaO2 reflects, to some extent, the overall trend of improvement in oxygenation during hospitalization. An increase in ΔPaO2 may suggest improved oxygenation compared with admission, which to a certain extent reflects the patient’s response to treatment and alleviation of disease status, including factors such as infection control, reduction of airway inflammation, and alveolar recruitment. However, it should be noted that PaO2 and its changes are influenced not only by the disease itself but also by treatment-related factors such as oxygen therapy, FiO2, and adjustments in NIV parameters. Therefore, ΔPaO2 is essentially a composite indicator reflecting both disease severity and treatment response, and its clinical interpretation is relatively complex. In this context, the findings of this study only indicate a statistical association between ΔPaO2 and mortality risk, rather than supporting a direct causal inference. Nevertheless, ΔPaO2 may still serve as an auxiliary indicator, providing additional information for prognostic assessment in patients with AECOPD. Longer duration of mechanical ventilation was associated with increased post-discharge mortality. Patients requiring prolonged ventilation likely experienced more severe pulmonary infections, poorer lung function, and were more susceptible to ventilator-associated pneumonia (VAP) (19,20), further prolonging ventilation and exacerbating existing lung injury. Prolonged ventilation also induces diaphragmatic dysfunction, complicating weaning. Continuous sedation during intubation is often required, and long-term sedation is closely linked to intensive care unit (ICU) delirium (21), a cognitive sequela that negatively impacts post-discharge survival. In general, patients with more severe disease and poorer lung function require longer durations of NIV support. NIV duration may, to some extent, reflect the severity of respiratory failure and the recovery rate, and is closely associated with prognosis; therefore, it may serve as a surrogate marker of disease severity. However, it should be noted that the association between NIV duration and outcomes mainly reflects an indirect manifestation of disease severity rather than an independent causal factor. Causal interpretation of this variable should therefore be avoided. The prediction model constructed in this study carries important clinical significance. Clinicians can apply the model at discharge like a “calculator”, integrating these indicators to rapidly and objectively identify patients at extremely high risk of post-discharge death. These high-risk patients could then receive more intensive management strategies, such as more frequent follow-up visits, closer monitoring, and more detailed self-management education for patients and families, thereby reducing rehospitalization risk. A key strength of this model is the inclusion of ΔPaO2. Unlike previous studies relying on single time-point blood gas indices (e.g., admission or discharge), this dynamic parameter captures patients’ response to treatment during hospitalization, essentially representing the cumulative therapeutic effect, and provides a novel and valuable perspective for future research. Interaction analysis showed that the APACHE II score is a strong effect modifier. Our prediction model had a greater impact on mortality risk in patients with higher APACHE II scores, likely because these patients are already in a state of severe physiological decompensation with significantly impaired multi-organ function. In this context, multiple factors in the prediction model (such as CKD) may act synergistically with this instability, so that any additional risk burden can lead to a sharp increase in mortality. This study also has certain limitations. As a retrospective study, all data were derived from historical medical records, which may introduce a degree of selection bias. The study was single-center with a relatively small sample size, and all patients were from the same medical center, limiting the generalizability of the findings. Unmeasured or unknown confounding factors, such as patients’ socioeconomic status, may still exist. Although home oxygen therapy did not show statistical significance in the univariate analysis, this does not exclude its potential association in other studies or in larger sample sizes. In addition, some oxygen therapy-related variables (such as oxygen flow rate and home NIV use) were not fully collected, which may to some extent underestimate the potential impact of oxygen therapy-related factors on prognosis. Some chronic comorbidities such as hypertension, diabetes mellitus, and cardiovascular disease may not have shown statistical significance in this study due to limitations such as sample size or the retrospective study design; however, this does not imply that these factors are not associated with the prognosis of patients with AECOPD. Therefore, future multicenter prospective studies incorporating more potential influencing factors are warranted to validate and extend the conclusions of this study. Conclusions This study constructed a prediction model for 1-year post-discharge mortality in AECOPD patients requiring NIV, based on CKD, mMRC, GOLD stage, ΔPaO2, and duration of NIV. The model demonstrated a significant synergistic amplification effect with APACHE II score and provides an important tool for identifying high-risk patients. Supplementary The article’s supplementary files as Available at https://jtd.amegroups.com/article/view/10.21037/jtd-2026-1-0290/prf