Development and internal validation of a nomogram for predicting in-hospital mortality in patients with acute exacerbation of chronic obstructive pulmonary disease

Abstract

Background

Acute exacerbation of chronic obstructive pulmonary disease (AECOPD)-related in-hospital mortality poses a significant clinical challenge. Existing prognostic tools—including the DECAF (Dyspnoea, Eosinopenia, Consolidation, Acidaemia, and atrial Fibrillation) score, which requires eosinophil count and chest radiography, and the BAP-65 (Blood urea nitrogen, Altered mental status, Pulse rate, and age ≥65 years) score, which incorporates blood urea nitrogen measurement—depend on haematological or radiological investigations that may not be immediately available at admission. This study aims to develop a nomogram based on clinical admission variables and arterial blood gas analysis for early risk stratization in patients with AECOPD.

Methods

This single-centre retrospective cohort study enrolled consecutive patients hospitalised for AECOPD between December 2016 and December 2018. Eligible patients had spirometry-confirmed COPD [post-bronchodilator forced expiratory volume in one second/forced vital capacity (FEV₁/FVC) <0.70] admitted for an acute exacerbation, with a hospital stay of at least 24 hours; those with severe concurrent conditions or substantial missing data were excluded. Univariate and multivariable logistic regression identified independent admission-level predictors of in-hospital mortality. A nomogram was constructed and internally validated using 1,000-iteration bootstrap resampling.

Results

Among 1,719 patients [75.6% male; median age 77.0 years, interquartile range (IQR): 69.0–84.0], 188 (10.9%) died during hospitalisation. Multivariable logistic regression (n=1,570) identified six independent predictors: older age [odds ratio (OR) =1.03, 95% confidence interval (CI): 1.02–1.05], higher modified Medical Research Council dyspnoea scale (mMRC) grade (OR =1.28, 95% CI: 1.05–1.56), lower peripheral oxygen saturation (SpO₂) (OR =0.96, 95% CI: 0.94–0.99), malnutrition (OR =1.81, 95% CI: 1.06–2.97), elevated arterial partial pressure of carbon dioxide (PaCO₂) (OR =1.01, 95% CI: 1.00–1.02), and hypertension (OR =0.54, 95% CI: 0.35–0.81, inversely associated). The nomogram demonstrated moderate discrimination [apparent area under the curve (AUC) =0.678, 95% CI: 0.633–0.722; bootstrap-corrected C-statistic =0.665, optimism =0.012], good calibration (Hosmer-Lemeshow P>0.99), and net clinical benefit across threshold probabilities of 2–49%.

Conclusions

This internally validated nomogram integrates five clinical variables assessable without laboratory testing (age, mMRC grade, SpO₂, nutritional status, and hypertension history) with admission PaCO₂—a routine investigation in AECOPD—to provide immediate, individualised prediction of in-hospital mortality. The model demonstrated moderate discrimination (bootstrap-corrected C-statistic: 0.665), good calibration, and net clinical benefit across a broad range of threshold probabilities, requiring fewer investigations than existing tools such as DECAF and BAP-65. Prospective external validation in independent multicentre cohorts is warranted before broader clinical implementation.

Highlight box.

Key findings

• A three-factor nomogram incorporating hypertension, bronchoscopy use, and work impact grade showed good predictive performance for in-hospital mortality in acute exacerbations of chronic obstructive pulmonary disease (AECOPD) (AUC =0.798).

• Hypertension (OR =2.290) and higher work impact grade (OR =1.540 per grade) were independent risk factors, while bronchoscopy was protective (OR =0.287).

What is known and what is new?

• In-hospital mortality in patients hospitalised for AECOPD is substantial, and several prognostic tools—including the DECAF (Dyspnoea, Eosinopenia, Consolidation, Acidaemia, and atrial Fibrillation) and BAP-65 scores—have been developed for risk stratification. However, these tools require laboratory or specialist investigations not consistently available at the bedside, and nutritional status is not routinely incorporated.

• This study developed and internally validated a six-variable nomogram—incorporating age, modified Medical Research Council (mMRC) dyspnoea grade, SpO₂, malnutrition, PaCO₂, and hypertension—for predicting in-hospital mortality in hospitalised AECOPD patients. All predictors are ascertainable at the bedside at the time of admission, without additional laboratory investigations. Bootstrap internal validation confirmed minimal overfitting (optimism =0.012).

What is the implication, and what should change now?

• This bedside-available nomogram provides a practical tool for immediate, individualised risk stratification of AECOPD patients at hospital admission, potentially supporting early escalation of care and clinical decision-making. Given its derivation from a single Chinese tertiary centre, external validation in independent multicentre cohorts is required before broader implementation.

Introduction

Chronic obstructive pulmonary disease (COPD) is a common progressive respiratory disease characterised by persistent airflow limitation driven by chronic airway inflammation, oxidative stress, and parenchymal destruction secondary to prolonged exposure to noxious stimuli, most notably tobacco smoke (1,2). According to the Global Burden of Disease Study 2019, COPD was the third leading cause of death worldwide, responsible for approximately 3.23 million deaths annually (3). Global prevalence estimates indicate that approximately 391.9 million individuals aged 30–79 years were affected by COPD in 2019, with the majority residing in low- and middle-income countries, underscoring the disproportionate burden borne by resource-limited health systems (4).

Acute exacerbations of COPD (AECOPD) represent critical inflection points in the disease trajectory, associated with accelerated lung function decline, impaired health-related quality of life, and substantially increased mortality risk (5). A recent meta-analysis of 65,945 hospitalised patients across multiple countries reported a pooled in-hospital mortality rate of 6.2% [95% confidence interval (CI): 6.0–6.4] following AECOPD-related admission, with post-discharge mortality reaching 10.9% at one year (6). Accurate risk stratification at the point of admission is therefore essential to guide escalation of care, prioritise clinical resources, and enable timely, individualised intervention (1,7).

Several prognostic tools have been developed for AECOPD risk stratification. The DECAF score—incorporating stable-state dyspnoea (modified Medical Research Council dyspnoea scale, eMRCD), eosinopenia, consolidation, acidaemia, and atrial fibrillation—demonstrated strong discriminative performance in both internal and external multicentre validation cohorts [area under the receiver operating characteristic curve (AUC) 0.83 and 0.82, respectively], outperforming CURB-65, CAPS, APACHE II, and BAP-65 for in-hospital mortality prediction (8). The BAP-65 score, integrating blood urea nitrogen (BUN), altered mental status, pulse rate, and age ≥65 years, achieved an AUC of 0.79 (95% CI: 0.62–0.94) in an independent validation cohort (9). A single-centre Chinese nomogram combining eight clinical and laboratory variables similarly demonstrated good discrimination, with a C-index of 0.858 (95% CI: 0.819–0.897) and a bootstrap-corrected estimate of 0.851 in 1,224 hospitalised AECOPD patients (10). Collectively, these tools highlight the growing interest in developing prediction models tailored to specific clinical settings and patient populations.

While these tools represent meaningful advances in AECOPD risk stratification, their applicability across diverse clinical settings and patient populations remains constrained by several factors. Both DECAF and BAP-65 were originally derived and validated in UK or North American populations (8,9), and their performance in Chinese hospitalised cohorts—characterised by distinct comorbidity profiles, disease management practices, and patient demographics—warrants further examination. Furthermore, the DECAF score requires eosinophil count and chest radiography in addition to clinical assessment, meaning that complete scoring depends on the availability of laboratory and imaging results and cannot be performed instantaneously at the bedside; existing tools also do not incorporate nutritional status, a clinically identifiable and potentially modifiable condition affecting a substantial proportion of patients with COPD and independently associated with increased mortality risk in this population (11).

In this context, the present study aimed to systematically identify independent predictors of in-hospital mortality among 1,719 patients hospitalised with AECOPD at a Chinese tertiary centre and to develop a nomogram based on variables routinely available at the point of admission—requiring no imaging and no haematological testing beyond standard arterial blood gas analysis—thereby enabling immediate, individualised risk stratification without the need for chest radiography or specialised laboratory investigations. We present this article in accordance with the TRIPOD reporting checklist (available at https://jtd.amegroups.com/article/view/10.21037/jtd-2025-1-2780/rc).

Methods

Study design and participants

This was a single-centre retrospective cohort study. All consecutive patients admitted with a primary diagnosis of AECOPD to the Clinical Medical College and Affiliated Hospital of Chengdu University between December 2016 and December 2018 were screened for eligibility.

Patients were eligible for inclusion if they met both of the following criteria: (I) a confirmed diagnosis of COPD based on post-bronchodilator (BD) spirometry demonstrating a forced expiratory volume in 1 second to forced vital capacity ratio (FEV₁/FVC) of less than 0.70, with the current admission representing an acute exacerbation, in accordance with GOLD guidelines (3); and (II) a hospital length of stay of at least 24 hours.

Patients were excluded if they had any of the following: (I) significant congenital heart disease; (II) acute myocardial infarction during the index admission; (III) severe hepatic or renal failure; (IV) advanced malignancy with an estimated life expectancy of less than three months; or (V) substantial missing data for key variables required for analysis.

The primary outcome was in-hospital mortality, defined as death from any cause occurring at any point during the index hospitalisation. Patients who survived to discharge were classified as the survival group. Given that in-hospital mortality is an objectively recorded administrative outcome derived directly from the electronic medical record (EMR), no outcome assessment blinding was required.

Of 1,719 eligible patients, 188 (10.94%) died during hospitalisation (in-hospital death group) and 1,531 (89.06%) survived to discharge (survival group).

Data source and ethical considerations

Clinical data were retrospectively extracted from the EMR system of the Clinical Medical College and Affiliated Hospital of Chengdu University. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the Medical Ethics Committee of the Clinical Medical College and Affiliated Hospital & Chengdu University (approval No.: PJ2021-045-01) and individual consent for this retrospective analysis was waived. All patient data were anonymised and de-identified prior to analysis. As all predictor variables were objectively extracted from the EMR, blinding of predictor assessment was not applicable.

Candidate predictor variables

A comprehensive set of variables routinely available at hospital admission was collected for each participant across the following domains:

• ❖ Demographics and clinical characteristics: age, sex, duration of COPD (years), and smoking history (smoking status, daily cigarette consumption, pack-year history, and cessation history).

• ❖ Vital signs at admission: body temperature, heart rate, respiratory rate, systolic and diastolic blood pressure, and peripheral oxygen saturation (SpO₂).

• ❖ Symptom burden and functional impact: disease-specific health status was assessed using the COPD Assessment Test (CAT) (12), a validated eight-item instrument in which each item is scored on a 0–5 scale (total score range, 0–40). All eight individual item scores were recorded: Items 1–4 assess symptom severity (cough, phlegm, chest tightness, and breathlessness, respectively), and Items 5–8 assess functional impact (limited activities, confidence leaving home, sleeplessness, and energy, respectively). The modified Medical Research Council (mMRC) dyspnoea scale (grade 0–4) was also administered. Permission to use the CAT was obtained from the copyright holder (GlaxoSmithKline) prior to the study.

• ❖ Arterial blood gas parameters: pH, partial pressure of carbon dioxide (PaCO₂), partial pressure of oxygen (PaO₂), bicarbonate (HCO₃⁻), and lactate.

• ❖ Pulmonary function: spirometric indices including forced vital capacity (FVC), FVC% predicted, forced expiratory volume in one second (FEV₁), FEV₁% predicted, FEV₁/FVC ratio, peak expiratory flow (PEF), PEF% predicted, and maximal mid-expiratory flow (MMEF). Pre- and post-BD values for FEV₁ and FVC, the percentage change in these values, and post-BD FEV₁/FVC ratio and FEV₁% predicted were also recorded.

• ❖ Comorbidities: hypertension, malignancy, malnutrition, asthma, liver disease, cardiovascular disease, neuropsychiatric disorders, renal disease, diabetes, and bronchiectasis.

Treatment-related variables—including mechanical ventilation, bronchoscopy, antibiotic regimens (class, combination, and timing), N-acetylcysteine (NAC), bromhexine, Ambroxol, and Tanreqing Injection—were documented for descriptive purposes but excluded from prediction model development, as these are determined dynamically during hospitalisation and their inclusion as static admission-level predictors would introduce immortal time bias.

Sample size

The study enrolled all consecutive eligible patients over the two-year study period (n=1,719; 188 in-hospital deaths). As a formal a priori power calculation is not applicable to retrospective prediction model studies where sample size is determined by data availability, adequacy was evaluated post hoc: the final six-predictor model yielded an events-per-variable (EPV) ratio of 31.3, substantially exceeding the commonly recommended minimum threshold of 10 (13).

Statistical analysis

All statistical analyses and visualisations were performed in R (version 4.3.1; R Foundation for Statistical Computing, Vienna, Austria). Continuous variables are reported as mean ± standard deviation (SD) or median (interquartile range, IQR) according to distributional normality assessed by the Shapiro-Wilk test, and compared between groups using the independent-samples t-test or Mann-Whitney U test, respectively. Categorical variables are reported as frequency (percentage) and compared using the Pearson chi-square or Fisher’s exact test as appropriate.

For prediction model development, variables significantly associated with in-hospital mortality on univariate logistic regression (P<0.05) were entered into multivariable logistic regression with bidirectional stepwise selection based on the Akaike Information Criterion (AIC). To address missing data in pulmonary function indices, two parallel models were constructed: Model A included pulmonary function variables restricted to complete cases (n=1,119); Model B was fitted on the full available sample without pulmonary function indices (n=1,570). Model B was designated the primary model based on its larger effective sample size and superior discriminative performance. Multicollinearity was assessed using the variance inflation factor (VIF).

A nomogram was constructed based on the independent predictors in the final model. Discriminative performance was quantified by the C-statistic (area under the receiver operating characteristic curve, AUC) with 95% CIs. Calibration was assessed graphically with a calibration plot and formally with the Hosmer-Lemeshow goodness-of-fit test. The full dataset was used for model development without a separate training/validation split. Internal validation was performed using 1,000-iteration bootstrap resampling, repeating the full modelling procedure in each replicate to estimate optimism; the bias-corrected C-statistic was obtained by subtracting the mean optimism from the apparent C-statistic. Decision curve analysis (DCA) was conducted to evaluate net clinical benefit across clinically relevant threshold probabilities. A two-tailed P<0.05 was considered statistically significant.

Results

Comparison of general information between the two groups

There were statistically significant differences between the survivors and non-survivors in age, heart rate, respiratory rate, SpO₂, and hypertension (P<0.05). However, no statistically significant differences were observed in disease duration, length of hospital stay, daily smoking amount, smoking duration, years since quitting smoking, body temperature, systolic blood pressure, diastolic blood pressure, gender, smoking status, or quit status (P>0.05) (Table 1).

Table 1. Baseline demographic and clinical characteristics of study participants.

Variables	Survivors (n=1,531)	Non-survivors (n=188)	Z/χ2	P
Age (years)	76.00 (69.00, 83.00)	79.00 (73.75, 86.00)	−4.16	<0.001
Gender			0.33	0.57
Male	1,161 (75.83)	139 (73.94)
Female	370 (24.17)	49 (26.06)
Disease duration (years)	10.00 (7.00, 20.00)	10.00 (10.00, 20.00)	−0.42	0.67
Temperature (℃)	36.50 (36.50, 36.70)	36.50 (36.50, 36.70)	−0.29	0.77
Heart rate (beats/min)	86.00 (80.00, 98.00)	90.00 (80.00, 101.25)	−2.6	0.009
Respiratory rate (breaths/min)	21.00 (20.00, 24.00)	22.00 (20.00, 25.00)	−3.29	0.001
Systolic blood pressure (mmHg)	126.00 (120.00, 138.00)	128.00 (120.00, 137.25)	−0.58	0.56
Diastolic blood pressure (mmHg)	79.00 (72.00, 82.00)	80.00 (72.00, 82.00)	−0.92	0.36
Oxygen saturation (%)	96.00 (92.00, 96.00)	95.00 (90.00, 96.00)	−4.2	<0.001
Hypertension			11.96	<0.001
No	1,078 (70.41)	155 (82.45)
Yes	453 (29.59)	33 (17.55)
Smoking			0.2	0.66
No	502 (33.33)	64 (34.97)
Yes	1,004 (66.67)	119 (65.03)
Smoking cessation			0.44	0.51
No	381 (38.92)	40 (35.71)
Yes	598 (61.08)	72 (64.29)
Daily smoking amount (cigarettes/day)	20.00 (10.00, 20.00)	20.00 (20.00, 20.00)	−1.17	0.24
Smoking history (years)	30.00 (30.00, 40.00)	30.00 (30.00, 40.00)	−0.76	0.45
Years since quitting smoking	10.00 (4.00, 17.75)	10.00 (5.00, 20.00)	−1.71	0.09
Hospital stay (days)	10.00 (8.00, 14.00)	11.00 (7.00, 19.25)	−0.76	0.45

Continuous variables are expressed as median (Q1, Q3) and compared with the Mann-Whitney U test; categorical variables are expressed as n (%) and compared with the χ² test. Missing data were <5% for all variables.

Compared with the survivors, the non-survivors were older [79.00 (73.75, 86.00) vs. 76.00 (69.00, 83.00) years, P<0.001], had a faster heart rate [90.00 (80.00, 101.25) vs. 86.00 (80.00, 98.00) beats/min, P=0.009], a higher respiratory rate [22.00 (20.00, 25.00) vs. 21.00 (20.00, 24.00) breaths/min, P=0.001], a lower SpO₂ [95.00% (90.00%, 96.00%) vs. 96.00% (92.00%, 96.00%), P<0.001], and a lower prevalence of hypertension (17.55% vs. 29.59%, χ²=11.96, P<0.001) (Table 1).

Comparison of symptom burden, arterial blood gas, pulmonary function, and treatment parameters between the two groups

As presented in Table 2, there were statistically significant differences between the two groups in CAT total score, mMRC grade, CAT Items 3–8 (chest tightness, breathlessness, limited activities, confidence leaving home, sleeplessness, and energy), partial pressure of carbon dioxide (PaCO₂), HCO₃⁻, FVC, FVC%, FEV₁, FEV₁%, PEF, PEF%, MMEF, pre-bronchodilation FEV₁, and pre-bronchodilation FVC (P<0.05). In contrast, no statistically significant differences were found in CAT Item 1 (cough), CAT Item 2 (phlegm), blood gas pH, PO₂, lactate, VC, VC%, FEV₁/FVC pre-BD, MMEF%, FEV₁ change post-BD (%), FEV₁ post-BD, FVC change post-BD (%), FVC post-BD, FEV₁/FVC post-BD (ratio), or FEV₁% post-BD (P>0.05).

Table 2. Comparison of symptom burden, arterial blood gas analysis, and pulmonary function parameters.

Variables	Survivors (n=1,531)	Non-survivors (n=188)	Z/χ2	P
Symptom scores
CAT total score (0–40)	20.00 (16.00, 24.00)	22.00 (18.00, 27.00)	−4.03	<0.001
mMRC grade			31.95	<0.001
0	17 (1.11)	2 (1.06)
1	160 (10.45)	14 (7.45)
2	592 (38.67)	41 (21.81)
3	508 (33.18)	76 (40.43)
4	254 (16.59)	55 (29.26)
CAT Item 1: cough (0–5)	3.00 (2.00, 4.00)	3.00 (2.00, 4.00)	−0.59	0.55
CAT Item 2: phlegm (0–5)	2.00 (2.00, 3.00)	2.50 (2.00, 3.00)	−1.86	0.06
CAT Item 3: chest tightness (0–5)	2.00 (2.00, 3.00)	2.00 (2.00, 3.00)	−3.80	<0.001
CAT Item 4: breathlessness (0–5)	2.00 (2.00, 3.00)	2.00 (2.00, 4.00)	−3.31	<0.001
CAT Item 5: limited activities (0–5)	3.00 (2.00, 3.00)	3.00 (2.00, 4.00)	−3.80	<0.001
CAT Item 6: confidence leaving home (0–5)	2.00 (2.00, 3.00)	3.00 (2.00, 3.00)	−4.67	<0.001
CAT Item 7: sleeplessness (0–5)	3.00 (2.00, 4.00)	3.00 (2.00, 4.00)	−3.25	0.001
CAT Item 8: energy (0–5)	3.00 (2.00, 3.00)	3.00 (2.00, 4.00)	−4.01	<0.001
Arterial blood gas
pH	7.40 (7.37, 7.43)	7.41 (7.37, 7.44)	−0.41	0.68
PaCO2 (mmHg)	41.00 (36.32, 49.50)	43.10 (37.68, 62.25)	−3.42	<0.001
PaO2 (mmHg)	76.90 (64.00, 99.00)	78.00 (63.55, 111.50)	−1.22	0.22
HCO3− (mmol/L)	25.80 (23.33, 29.98)	27.60 (24.45, 33.40)	−3.81	<0.001
Lactate (mmol/L)	1.22 (0.88, 1.79)	1.26 (0.80, 1.90)	−0.19	0.85
Pulmonary function
FVC, pre-BD (L)	1.79 (1.22, 2.50)	1.45 (0.93, 2.10)	−3.33	<0.001
FVC%, pre-BD (% predicted)	63.00 (46.20, 83.00)	58.00 (39.10, 77.00)	−2.80	0.005
VC (L)	1.93 (1.33, 2.60)	1.67 (1.27, 2.40)	−1.79	0.07
VC (% predicted)	67.00 (51.00, 86.00)	65.00 (48.75, 81.25)	−1.11	0.27
FEV1, pre-BD (L)	0.86 (0.56, 1.30)	0.71 (0.42, 1.12)	−3.05	0.002
FEV1%, pre-BD (% predicted)	44.00 (29.00, 65.00)	40.00 (22.80, 55.40)	−2.17	0.03
FEV1/FVC, pre-BD (%)	49.36 (39.41, 61.00)	49.52 (39.92, 60.38)	−0.42	0.67
PEF, pre-BD (L/s)	2.32 (1.44, 3.51)	1.77 (1.07, 3.03)	−3.10	0.002
PEF%, pre-BD (% predicted)	38.00 (23.65, 56.12)	27.00 (17.10, 46.00)	−3.43	<0.001
MMEF, pre-BD (L/s)	0.48 (0.27, 0.93)	0.40 (0.22, 0.80)	−2.20	0.03
MMEF%, pre-BD (% predicted)	23.00 (12.90, 44.15)	21.00 (10.80, 34.00)	−1.70	0.09
FEV1, post-BD (L)	1.03 (0.70, 1.46)	1.05 (0.72, 1.32)	−0.38	0.71
FVC, post-BD (L)	2.22 (1.62, 2.90)	2.17 (1.54, 2.61)	−0.81	0.42
FEV1/FVC, post-BD (ratio)	0.49 (0.40, 0.59)	0.49 (0.40, 0.59)	−0.45	0.66
FEV1 (%), post-BD (% predicted)	53.00 (37.00, 71.00)	47.00 (42.50, 62.00)	−0.70	0.48
FEV1 change, post-BD (%)	8.33 (3.41, 15.62)	7.41 (4.90,14.29)	−0.09	0.93
FVC change, post-BD (%)	8.02 (2.29, 14.95)	9.09 (3.94, 14.60)	−0.77	0.44
TFAD (min)	77.45 (53.57, 152.54)	100.30 (55.68, 286.65)	−2.89	0.004
Treatment
Mechanical ventilation			46.07	<0.001
No	1,211 (79.10)	107 (56.91)
Yes	320 (20.90)	81 (43.09)
Bronchoscopy			30.60	<0.001
No	750 (48.99)	52 (27.66)
Yes	781 (51.01)	136 (72.34)
Aminoglycosides			6.44	0.01
No	1,358 (98.76)	163 (95.88)
Yes	17 (1.24)	7 (4.12)
Fluoroquinolones			5.27	0.02
No	1,224 (89.02)	161 (94.71)
Yes	151 (10.98)	9 (5.29)
NAC			0.01	0.96
No	893 (58.48)	109 (58.29)
Yes	634 (41.52)	78 (41.71)
Bromhexine			1.71	0.19
No	396 (25.87)	57 (30.32)
Yes	1,135 (74.13)	131 (69.68)
Ambroxol			0.01	0.91
No	1,126 (73.55)	139 (73.94)
Yes	405 (26.45)	49 (26.06)
Tanreqing			7.45	0.006
No	994 (64.92)	103 (54.79)
Yes	537 (35.08)	85 (45.21)
Comorbidities
Tumor			3.93	0.048
No	1,478 (96.54)	176 (93.62)
Yes	53 (3.46)	12 (6.38)
Liver disease			0.93	0.34
No	1,393 (90.99)	167 (88.83)
Yes	138 (9.01)	21 (11.17)
Cardiovascular disease			0.01	0.95
No	591 (38.60)	73 (38.83)
Yes	940 (61.40)	115 (61.17)
Neurological/psychiatric disease			0.17	0.68
No	1,367 (89.29)	166 (88.30)
Yes	164 (10.71)	22 (11.70)
Renal disease			2.88	0.09
No	1,454 (94.97)	173 (92.02)
Yes	77 (5.03)	15 (7.98)
Malnutrition			17.66	<0.001
No	1,443 (94.25)	162 (86.17)
Yes	88 (5.75)	26 (13.83)
Asthma			7.50	0.006
No	1,472 (96.15)	188 (100.00)
Yes	59 (3.85)	0 (0.00)
Diabetes			0.14	0.71
No	1,220 (79.69)	152 (80.85)
Yes	311 (20.31)	36 (19.15)
Bronchiectasis			0.21	0.65
No	1,461 (95.43)	178 (94.68)
Yes	70 (4.57)	10 (5.32)

Continuous variables are expressed as median (Q1, Q3) and compared with the Mann-Whitney U test; categorical variables are expressed as n (%) and compared with the χ² test. CAT items 1–8 were scored on a 0–5 scale and the total CAT score on a 0–40 scale, administered by trained respiratory nurses at admission. Arterial blood gas parameters were measured within 2 hours of admission. Pulmonary function indices are reported as absolute values and percentages of predicted; pre- and post-bronchodilator values were obtained per ATS/ERS criteria using 400 μg salbutamol. ATA/ERS, American Thoracic Society/European Respiratory Society; TFAD calculated from electronic medical record timestamps. BD, bronchodilator; CAT, Chronic Obstructive Pulmonary Disease Assessment Test; FEV₁, forced expiratory volume in 1 second; FVC, forced vital capacity; HCO₃⁻, bicarbonate; MMEF, maximal mid-expiratory flow; mMRC, Modified Medical Research Council dyspnoea scale; NAC, N-acetylcysteine; PaCO₂, partial pressure of arterial carbon dioxide; PEF, peak expiratory flow; PaO₂, partial pressure of arterial oxygen; TFAD, time to first antibiotic dose; VC, vital capacity.

Compared with the survivors, the non-survivors had a higher CAT total score [22.00 (18.00, 27.00) points vs. 20.00 (16.00, 24.00) points, Z=−4.03, P<0.001], a more severe mMRC grade (χ²=31.95, P<0.001)—with the proportion of patients with mMRC grade 4 (29.26%) being significantly higher than that in the survival at discharge group (16.59%)—significantly higher scores in CAT Items 3–8 (chest tightness, breathlessness, limited activities, confidence leaving home, sleeplessness, and energy; all P<0.05), higher levels of PaCO₂ [43.10 (37.68, 62.25) vs. 41.00 (36.32, 49.50) mmHg, P<0.001] and HCO₃⁻ [27.60 (24.45, 33.40) vs. 25.80 (23.33, 29.98) mmol/L, P<0.001], and poorer pulmonary function indicators. For example, FVC [1.45 (0.93, 2.10) vs. 1.79 (1.22, 2.50) L, P<0.001], FEV₁ [0.71 (0.42, 1.12) vs. 0.86 (0.56, 1.30) L, P=0.002], and PEF [1.77 (1.07, 3.03) vs. 2.32 (1.44, 3.51) L/s, P=0.002] in the non-survivors were significantly lower than those in the survivors.

Comparison of treatment-related indicators and comorbidities between the two groups

There were statistically significant differences between the two groups in time to first antibiotic dose (TFAD), bronchoscopy, use of aminoglycosides, use of fluoroquinolones, use of Tanreqing Injection, mechanical ventilation, tumour, malnutrition, and asthma (P<0.05). No statistically significant differences were observed in the use of NAC, bromhexine, ambroxol, liver disease, cardiovascular disease, neurological/psychiatric disease, renal disease, diabetes, or bronchiectasis (P>0.05) (Table 2).

Compared with the survivors, the non-survivors had a longer TFAD [100.30 (55.68, 286.65) vs. 77.45 (53.57, 152.54) min, P=0.004], a higher bronchoscopy rate (72.34% vs. 51.01%, P<0.001), a higher rate of aminoglycoside use (4.12% vs. 1.24%, P=0.01), a lower rate of fluoroquinolone use (5.29% vs. 10.98%, P=0.02), a higher rate of Tanreqing Injection use (45.21% vs. 35.08%, P=0.006), a higher rate of mechanical ventilation (43.09% vs. 20.90%, P<0.001), a higher prevalence of tumour (6.38% vs. 3.46%, P=0.048), a higher prevalence of malnutrition (13.83% vs. 5.75%, P<0.001), and no patients with comorbid asthma (0.00% vs. 3.85%, P=0.006).

Univariate logistic regression analysis of in-hospital mortality

To identify admission-level predictors of in-hospital mortality, univariate logistic regression was performed on all candidate variables except treatment-related variables.

A total of 20 variables were significantly associated with in-hospital mortality on univariate analysis (P<0.05; Figure 1). Among demographic variables, older age was associated with higher mortality risk [odds ratio (OR) =1.03, 95% CI: 1.02–1.05, P<0.001]. Higher mMRC grade (OR =1.53, 95% CI: 1.29–1.82, P<0.001) and greater CAT total score (OR =1.06, 95% CI: 1.03–1.08, P<0.001) were both significantly associated with mortality. Among individual CAT items, confidence leaving home (OR =1.42, 95% CI: 1.23–1.63), energy (OR =1.40, 95% CI: 1.21–1.62), chest tightness (OR =1.36, 95% CI: 1.19–1.56), limited activities (OR =1.34, 95% CI: 1.16–1.55), sleeplessness (OR =1.33, 95% CI: 1.13–1.58), breathlessness (OR =1.32, 95% CI: 1.14–1.53), and phlegm (OR =1.16, 95% CI: 1.00–1.35) were all positively associated with mortality risk (all P<0.05). Among vital signs, lower SpO₂ was associated with increased mortality risk (OR =0.94, 95% CI: 0.92–0.97, P<0.001), while elevated heart rate (OR =1.01, P=0.01) and respiratory rate (OR =1.01, P=0.02) showed modest positive associations. Among comorbidities, malnutrition was the strongest predictor (OR =2.63, 95% CI: 1.62–4.14, P<0.001), whereas hypertension was inversely associated with mortality (OR =0.51, 95% CI: 0.34–0.74, P<0.001). Among arterial blood gas parameters, elevated PaCO₂ (OR =1.02, 95% CI: 1.01–1.03, P<0.001) and HCO₃⁻ (OR =1.05, 95% CI: 1.03–1.07, P<0.001) were positively associated with mortality, while PaO₂ showed a marginal positive association (OR =1.00, P=0.02). Pulmonary function indices, including FVC% predicted (OR =0.99, P=0.002) and FEV₁% predicted (OR =0.99, P=0.03), were inversely associated with mortality, though the magnitude of effect was modest.

Figure 1.

Univariate logistic regression analysis of in-hospital mortality predictors in hospitalized AECOPD patients. Variables are ordered by OR magnitude. Red indicates P<0.05; grey indicates P≥0.05. AECOPD, acute exacerbation of chronic obstructive pulmonary disease; BP, blood pressure; CAT, Chronic Obstructive Pulmonary Disease Assessment Test; CI, confidence interval; DLCO, diffusing capacity for carbon monoxide; FEV₁, forced expiratory volume in 1 second; FVC, forced vital capacity; HCO₃⁻, bicarbonate; MMEF, maximal mid-expiratory flow; mMRC, modified Medical Research Council dyspnoea scale; OR, odds ratio; PaCO₂, partial pressure of arterial carbon dioxide; PaO₂, partial pressure of arterial oxygen; RV/TLC, residual volume/total lung capacity; SpO₂, peripheral oxygen saturation.

Multivariate logistic regression analysis of in-hospital mortality

Of the 20 variables that reached statistical significance in univariate analysis, pulmonary function indices (FEV₁% predicted, FVC% predicted) were excluded from multivariate modelling due to a substantial proportion of missing values. To evaluate the impact of this exclusion, two parallel stepwise models were constructed using bidirectional stepwise selection based on the AIC: Model A retained pulmonary function variables and was restricted to complete cases (n=1,119; AIC =599.5; AUC =0.654), whereas Model B was fitted on the full available sample without pulmonary function indices (n=1,570; AIC =1,008.6; AUC =0.678). Model B was designated as the final model given its larger sample size and superior discriminative performance.

The final multivariable logistic regression model identified six independent predictors of in-hospital mortality (Table 3). Older age (OR =1.03, 95% CI: 1.02–1.05, P<0.001), higher mMRC dyspnoea grade (OR =1.28, 95% CI: 1.05–1.56, P=0.01), malnutrition (OR =1.81, 95% CI: 1.06–2.97, P=0.02), and elevated PaCO₂ (OR =1.01, 95% CI: 1.00–1.02, P=0.03) were each independently associated with increased mortality risk. Lower SpO₂ was also an independent predictor of mortality (OR =0.96, 95% CI: 0.94–0.99, P=0.008). Hypertension was inversely and independently associated with in-hospital death (OR =0.54, 95% CI: 0.35–0.81, P=0.004). VIF analysis confirmed the absence of multicollinearity among all predictors (VIF range, 1.03–1.21).

Table 3. Multivariable logistic regression analysis of independent predictors of in-hospital mortality in hospitalized AECOPD patients (final model, n=1,570).

Variables	OR (95% CI)	P value
Age (years)	1.03 (1.02–1.05)	<0.001
mMRC grade	1.28 (1.05–1.56)	0.01
SpO2 (%)	0.96 (0.94–0.99)	0.008
Malnutrition (yes vs. no)	1.81 (1.06–2.97)	0.02
Hypertension (yes vs. no)	0.54 (0.35–0.81)	0.004
PaCO2 (mmHg)	1.01 (1.00–1.02)	0.03

AECOPD, acute exacerbation of chronic obstructive pulmonary disease; CI, confidence interval; mMRC, modified Medical Research Council dyspnoea scale; OR, odds ratio; PaCO₂, partial pressure of arterial carbon dioxide; SpO₂, peripheral oxygen saturation.

Nomogram development and model validation

Based on the six independent predictors retained in the final multivariable logistic regression model—age, mMRC dyspnoea grade, SpO₂, malnutrition, hypertension, and PaCO₂—a nomogram was developed to enable individualised estimation of in-hospital mortality risk (Figure 2). Each predictor is assigned a point score proportional to its regression coefficient; the total points map directly to a predicted probability of in-hospital death on the bottom scale.

Figure 2.

Nomogram for predicting in-hospital mortality in hospitalised patients with AECOPD. The model was developed using 1,570 hospitalised AECOPD patients with complete data (survival at discharge: n=1,404; in-hospital death: n=166; in-hospital mortality: 10.6%), derived from a total cohort of 1,719 patients (149 excluded due to missing predictor values). AECOPD, acute exacerbation of chronic obstructive pulmonary disease; mMRC, modified Medical Research Council dyspnoea scale; PaCO₂, partial pressure of arterial carbon dioxide; SpO₂, peripheral oxygen saturation.

The model demonstrated moderate discriminative ability, with an apparent AUC of 0.678 (95% CI: 0.633–0.722) (Figure 3A). The optimal cut-off probability determined by the Youden index was 0.094, corresponding to a sensitivity of 70.5% and a specificity of 57.3%; this relatively low threshold is consistent with the overall in-hospital mortality rate of 10.9% observed in this cohort. Internal validation via 1,000-iteration bootstrap resampling—applied to the full dataset without a separate training/validation split—yielded an optimism estimate of 0.012 and a bias-corrected C-statistic of 0.665, indicating minimal overfitting (Table S1).

Figure 3.

Model performance of the nomogram. (A) ROC curve. The AUC was 0.678 (95% CI: 0.633–0.722). The dot indicates the optimal cut-off (0.094) by the Youden index (sensitivity 70.5%, specificity 57.3%). The diagonal dashed line represents chance discrimination. (B) Calibration curve. Each point represents one decile of predicted probability; vertical bars indicate 95% confidence intervals for observed proportions. The red curve (LOESS with 95% confidence band) reflects the overall calibration trend. The diagonal dashed line indicates perfect calibration. Hosmer-Lemeshow test: χ²=1.272, df=8, P>0.99. (C) Decision curve analysis. Net benefit is plotted against threshold probability for the nomogram model (red solid line), treat-all strategy (blue dashed line), and treat-none strategy (grey dotted line). The nomogram outperformed both reference strategies across threshold probabilities of approximately 2–49%. AUC, area under the curve; CI, confidence interval; LOESS, locally estimated scatterplot smoothing; ROC, receiver operating characteristic.

The calibration curve demonstrated good agreement between predicted and observed probabilities across the full risk spectrum (Figure 3B), confirmed by the Hosmer-Lemeshow goodness-of-fit test (χ²=1.272, df=8, P>0.99). DCA showed that the nomogram consistently provided greater net benefit than both the “treat all” and “treat none” strategies across threshold probabilities of approximately 2–49% (Figure 3C), supporting its clinical utility for risk-stratified management of hospitalised AECOPD patients.

Discussion

This study identified six independent predictors of in-hospital mortality among 1,719 hospitalised patients with AECOPD: older age, higher mMRC dyspnoea grade, lower SpO₂, malnutrition, elevated PaCO₂, and hypertension (inversely associated). These admission-level variables were integrated into a nomogram with moderate discriminative ability (AUC =0.678), excellent calibration (Hosmer-Lemeshow P>0.99), and consistent net clinical benefit on DCA across threshold probabilities of 2–49%. While existing AECOPD prognostic tools achieve higher discrimination—the DECAF score (AUC 0.82–0.86) (14) and comparable single-centre nomograms (C-index up to 0.858) (10) rely on laboratory variables such as inflammatory markers and haematological indices that may not be immediately available in primary care or resource-limited settings—our model is based exclusively on variables universally accessible at the bedside upon admission, enabling immediate individualised risk stratification without additional testing.

One of the most clinically counterintuitive findings is the independent inverse association between hypertension and in-hospital mortality (OR =0.54, 95% CI: 0.35–0.81), with hypertension prevalence markedly lower in non-survivors (17.6%) than in survivors (29.6%). Several mechanisms may account for this observation. First, patients with established hypertension are more likely to be receiving long-term antihypertensive therapy, particularly beta-blockers, which have demonstrated mortality-reducing effects in COPD: a meta-analysis of 15 observational cohort studies found that beta-blocker use was associated with significant reductions in overall mortality (RR =0.72, 95% CI: 0.63–0.83) and AECOPD rate (RR =0.63, 95% CI: 0.57–0.71) (15). Second, the absence of hypertension in a hospitalised AECOPD population may partly reflect haemodynamic fragility rather than physiological protection. The SUMMIT trial (n=16,485) demonstrated a U-shaped relationship between blood pressure and all-cause mortality in COPD, with both high [systolic blood pressure (SBP) ≥140 mmHg or diastolic blood pressure (DBP) ≥90 mmHg] and low (SBP <120 mmHg or DBP <80 mmHg) blood pressure independently associated with increased mortality (16). Consistent with this, a large prospective multicentre study from the MAGNET AECOPD Registry (n=13,633) confirmed that low DBP (<70 mmHg) was an independent predictor of in-hospital mortality (HR =2.16, 95% CI: 1.53–3.05) across all AECOPD inpatients, regardless of cardiovascular comorbidity status (17); this finding was corroborated by two further multicentre cohort studies specifically in more critically ill AECOPD patients requiring non-invasive ventilation (NIV) or intensive care unit (ICU) admission, both of which identified DBP <60 mmHg as an independent predictor of in-hospital death (18,19). Third, hypertensive patients typically receive more intensive multidisciplinary follow-up, which may confer survival advantages through earlier comorbidity detection and management. It is important to emphasise that this inverse association should not be interpreted as implying that hypertension is itself beneficial; rather, it reflects the complex interplay between long-term medication use, haemodynamic reserve, and healthcare engagement—all of which are difficult to disentangle in a retrospective cohort without detailed medication records.

Age, mMRC dyspnoea grade, and SpO₂ reflect the accumulated severity of disease at presentation—markers that characterise a patient’s physiological state at admission but are not themselves directly modifiable targets for intervention. The independent contribution of age (OR =1.03 per year) to in-hospital mortality is consistent with the well-established physiological vulnerability of elderly patients with COPD, encompassing progressive respiratory muscle decline, immune senescence, and accumulated comorbidity burden. A prospective observational study of 449 hospitalised AECOPD patients confirmed that age was an independent predictor of 90-day mortality both as a continuous variable (OR =1.05, 95% CI: 1.01–1.10) and categorically, with patients aged 68–76 years (HR =6.6, 95% CI: 1.5–28.8) and ≥77 years (HR =7.2, 95% CI: 1.6–32.6) at markedly higher risk than those aged ≤67 years, even after adjustment for comorbidities, disease staging, and severity (20). The mMRC grade (OR =1.28) captures not only acute dyspnoea but also the cumulative functional reserve that determines a patient’s capacity to compensate during an exacerbation—a construct central to the DECAF score, which incorporated the extended MRC dyspnoea scale as its primary predictor and achieved an AUC of 0.86 in its derivation cohort (n=920) (14). Admission SpO₂ (OR =0.96 per 1%) directly reflects the degree of ventilation-perfusion mismatch and gas exchange failure. A large observational study using the MIMIC-III database (n=996) demonstrated a U-shaped relationship between SpO₂ and all-cause in-hospital mortality in AECOPD, with nadir mortality at SpO₂ ≈89.5% and increasing risk at both extremes (21), underscoring that over-oxygenation—which worsens hypercapnia via the Haldane effect—is as hazardous as hypoxaemia, and supporting titrated oxygen therapy targeting SpO₂ 88–92% in patients at risk of hypercapnic respiratory failure (22).

In contrast to the severity markers discussed above, malnutrition and elevated PaCO₂ represent pathophysiological mechanisms with direct therapeutic implications. To our knowledge, this study is among the first to demonstrate that both clinically recorded malnutrition (OR =1.81) and elevated admission PaCO₂ as a continuous variable (OR =1.01 per mmHg) independently predict in-hospital mortality in hospitalised AECOPD patients. Prior studies in AECOPD have linked nutritional status to length of stay and hospitalisation costs using composite indices such as the Prognostic Nutritional Index (PNI) and Geriatric Nutritional Risk Index (GNRI), but have not established clinically diagnosed malnutrition as an independent predictor of in-hospital mortality in this population (23). This is consistent with evidence from stable COPD, where low body mass index (BMI) and low fat-free mass independently predicted approximately four-fold greater two-year mortality (HR =3.7 and HR =3.6, respectively) after adjustment for age and lung function in a prospective cohort (n=167) (24). For PaCO₂, existing evidence derives principally from stable COPD with long-term follow-up, where PaCO₂ independently predicted overall survival (HR =1.026 per mmHg, 95% CI: 1.011–1.042) after adjustment for FEV₁%, BMI, and comorbidities (25)—a magnitude strikingly consistent with our finding and supporting its extension to the acute inpatient setting. The absence of pH significance on univariate analysis—likely reflecting compensated hypercapnia—further suggests that PaCO₂ captures chronic ventilatory burden beyond acute acid-base disturbance. Together, these findings support integrating nutritional assessment and PaCO₂ monitoring into routine AECOPD admission risk stratification, where both represent potentially modifiable targets for intervention.

Several limitations warrant acknowledgement. First, the retrospective single-centre design and absence of an independent external validation cohort limit generalisability; the bootstrap-corrected C-statistic (0.665, optimism =0.012) provides an internal estimate of discrimination, but prospective multicentre external validation is required before broad clinical deployment. Second, 149 patients (8.7%) were excluded due to missing predictor values—predominantly pulmonary function indices—and as patients too ill to undergo spirometry are disproportionately represented among non-survivors, this may have systematically underestimated the prognostic contribution of lung function and introduced selection bias. Third, the absence of detailed medication records precludes definitive examination of whether the inverse association with hypertension reflects antihypertensive medication effects—particularly those of beta-blockers—or patient-level characteristics such as haemodynamic reserve.

Conclusions

This study identified six independent predictors of in-hospital mortality in hospitalised AECOPD patients—older age, higher mMRC dyspnoea grade, lower SpO₂, malnutrition, elevated PaCO₂, and hypertension (inversely associated)—and integrated them into a nomogram demonstrating good calibration (Hosmer-Lemeshow P>0.99) and consistent net clinical benefit across threshold probabilities of 2–49%. Comprising variables universally accessible at the bedside upon admission, the model enables immediate individualised risk stratification without additional laboratory testing, while introducing clinically recorded malnutrition as a novel predictor absent from existing tools. External validation in independent multicentre cohorts is warranted to confirm generalisability before broader clinical deployment.

Supplementary

The article’s supplementary files as

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the Medical Ethics Committee of the Clinical Medical College and Affiliated Hospital & Chengdu University (approval No.: PJ2021-045-01) and individual consent for this retrospective analysis was waived.

Data Sharing Statement

Available at https://jtd.amegroups.com/article/view/10.21037/jtd-2025-1-2780/dss

jtd-18-04-322-dss.pdf ^{(73.7KB, pdf)}

DOI: 10.21037/jtd-2025-1-2780