Predicting in-hospital mortality in a real-world population of ward-based non-invasive ventilation in acute COPD exacerbations

Abstract

Background:

Accurate tools for patient stratification by likely outcome are needed to support complex decision-making and improve acute non-invasive ventilation (NIV) delivery.

Objectives:

To evaluate the potential of an emerging NIV outcomes (NIVO) score tool to predict in-hospital mortality to aid its validation in a real-world UK hospital population of ward-based NIV for acute exacerbations of chronic obstructive pulmonary disease (AECOPD).

Design:

This was a retrospective observational cohort study of all consecutive patient admissions with AECOPD managed with NIV for acute hypercapnic respiratory failure at a teaching hospital.

Methods:

Clinical parameters were collected as part of an ongoing quality improvement project. Patients were grouped based on their survival status at hospital discharge. First admission of each patient was included in the analysis. NIV failure, defined as NIV withdrawal or intubation requirement due to clinical deterioration on NIV, along with in-hospital mortality, was modelled using logistic regression.

Results:

There were 249 unique patient AECOPD admissions with ward-based NIV. Across first admissions, NIV failure rate was 37.3%, in-hospital mortality 26.5%, and 1-year mortality 47.0%. NIVO score was significantly associated with both NIV failure and in-hospital mortality, with odds ratios (95% Confidence intervals) of 1.33 (1.13–1.58, p < 0.001) and 1.52 (1.26–1.86), p < 0.001, respectively. A progressive increase in in-hospital mortality was observed with increasing NIVO scores (p < 0.0001).

Conclusion:

This study demonstrates that the NIVO score shows promise as a predictive tool for in-hospital mortality in patients with AECOPD receiving ward-based NIV. Furthermore, it suggests that NIVO may be able to support decision-making for enhanced NIV delivery in new clinical pathways to address the growing burden of chronic obstructive pulmonary disease exacerbations.

Plain language summary

Predicting death in patients with chronic obstructive pulmonary disease (COPD) who need breathing support through a mask on hospital wards

People with long-term lung conditions like bronchitis or emphysema - often called chronic obstructive pulmonary disease (COPD) - can sometimes have flare-ups that make it hard to breathe. In these patients, doctors may use a breathing machine to help them breathe more easily using a mask. This breathing machine is called non-invasive ventilation. However, it can be difficult for doctors to know which patients are more likely to benefit from this machine and which are not. In our study, we looked at a new scoring system called the Non-Invasive Ventilation Outcomes (NIVO) score. This score is designed to help doctors predict which patients are more likely to die in hospital or for whom the breathing machine might not work. We aimed to provide information to help this tool be used in the future, for doctors to make better decisions about when to use this treatment. We reviewed the medical records of 249 patients who came to hospital and needed this breathing machine because their COPD had worsened. Some patients came to hospital more than once, so patients came to the hospital 314 times in total. We found that patients with a higher NIVO score were more likely to die and the breathing machine was less likely to help them recover. Also, when we grouped patients by their NIVO scores, those in the higher score groups had a greater risk of dying than those with lower scores. Our findings suggest that the NIVO score could help doctors predict which patients using the breathing machine are most at risk. This could help doctors decide who might benefit from this treatment in the future and improve how this treatment is used. Ultimately, this could lead to better care for patients with COPD.

Keywords

AECOPD COPD NIV NIV treatment outcomes NIVO score

Introduction

Acute exacerbations of chronic obstructive pulmonary disease (AECOPD) are a substantial clinical burden globally and drive considerable morbidity, mortality, and healthcare resource utilisation.^1,2 Acute hypercapnic respiratory failure (AHRF), characterised by elevated PaCO₂ (>45 mmHg) and acidaemia (pH < 7.35), is a life-threatening common complication of AECOPD. A recent meta-analysis of 65,945 patients reported the strongest predictors of mortality and hospital readmissions, with non-invasive ventilation (NIV) emerging as a key intervention that reduces the need for intubation and mortality by addressing ventilatory failure and correcting respiratory acidosis.^3,4

The decision to start NIV is multifaceted and requires complex clinical thinking about patient-specific factors and considerations around tolerability, expected magnitude of treatment effect, likelihood of treatment success and short- and long-term outcomes. Clinical uncertainty and pessimism regarding prognosis can lead to underutilisation and suboptimal delivery of acute NIV, particularly in resource-limited settings.^5
–7

There is a clear need for novel tools that can stratify patients based on their likelihood of treatment success, thereby supporting clinician decision-making and facilitating informed discussions with patients and their families.^6,8 The HACOR score (incorporating heart rate, acidosis, consciousness, oxygenation, and RR), has shown promise in predicting NIV failure in patients with acute hypoxemic respiratory failure and has recently been updated to include parameters such as the presence of pneumonia or immunosuppression.⁹ However, its sensitivity and specificity in AECOPD populations have been limited.¹⁰ Similarly, the Dyspnoea, Eosinopenia, Consolidation, Acidaemia, atrial Fibrillation (DECAF) score has demonstrated good potential to predict hospital mortality in AECOPD, but its performance was less strong in patients requiring ventilation.^11,12

More recently, the Non-Invasive Ventilation Outcomes (NIVO) score has been developed to stratify patients with AHRF due to AECOPD according to their in-hospital mortality risk. The 9-point NIVO score is based on six easily obtainable clinical parameters - Extended Medical Research Council Dyspnoea (eMRCD) score, time from admission to acidaemia (>12 h), severe acidaemia (pH < 7.25), atrial fibrillation, Glasgow Coma Scale (GCS) ⩽14 and consolidation on chest radiograph. NIVO demonstrated a strong predictive accuracy with an area under curve (AUC) of 0.79, which was replicated in a secondary cohort, and in chronic obstructive pulmonary disease (COPD) patients without prior demonstrated airway obstruction.^12,13 Among these, dyspnoea severity, assessed by the eMRCD scale, appears to be the most powerful individual predictor of mortality.¹² Compared to more complex scoring systems, such as APACHE II and COPD and Asthma Physiology Score (CAPS), NIVO offers advantages in simplicity and disease-specific relevance, making it potentially more feasible for bedside use.^12,14
-16

NIVO was developed in a study of 489 patients, and then substantially validated in 844 COPD patients (733 with spirometry-confirmed COPD and 111 with a clinical diagnosis) across 10 UK Trusts in diverse healthcare settings, including intensive care units, high-dependency units, respiratory support units and medical wards.^12,13 However, despite its promise, NIVO has not yet been integrated into routine clinical practice.¹⁷ While subsequent studies have demonstrated that it has shown good utility in predicting mortality in diverse cohorts, including a large Chinese cohort with a comparatively low mortality rate¹⁸ and a French ICU cohort with extended follow-up¹⁹; further real-world validation, particularly in ward-based NIV cohorts, remains essential. We, therefore, aimed to evaluate the performance of the NIVO score in a ward-based NIV cohort of patients with AECOPD, to strengthen the evidence base required for broader implementation of NIVO in guiding NIV decisions and meeting the burden of COPD exacerbations on healthcare systems.

Methods

Study design and data source

In this retrospective observational cohort analysis, we included patients with AECOPD, managed with ward-based NIV for AHRF at an acute teaching hospital, Heartlands Hospital, Birmingham, between June 2021 and December 2024. All consecutive admissions of patients with a primary diagnosis of AECOPD were screened, based on data extracted from the hospital’s NIV database, which was developed by one of the authors (RM), in collaboration with the respiratory physiotherapy department. This database contains demographic and clinical information for all patients initiated on NIV for AHRF. Data were collected by physiotherapists prospectively and routinely in a ward-based NIV service. Patients were included uniquely only once, at their first admission. After identification of the admissions from the database, related data were individually extracted from the medical records. This activity involved retrospective analysis of existing data and was registered as an audit in our National Health Service Institutions’ Clinical Audit Registration & Management System (Registration Number CARMS-19085).

Inclusion criteria were (1) Primary diagnosis of AECOPD, defined by the following medical team as noted on clinical records determined by the consultants of Respiratory Medicine or based on spirometry evidence of a post-bronchodilator FEV₁/FVC ratio of < 0.7, when available; (2) AHRF requiring NIV management (pH < 7.35 or PCO2 > 6.5 kPa after medical management with clinical signs of respiratory failure). Those with AHRF mainly due to other causes (such as obesity hypoventilation syndrome (OHS), chest wall disorders, dysfunctional breathing, etc.) were excluded. NIV was set up by a qualified respiratory physiotherapist and overseen by the acute respiratory or general medical team, in line with British Thoracic Society guidelines.²⁰ NIV settings and management were guided by arterial blood gas (ABG) measurements, patient comfort, and clinical judgement. The reporting of this study conforms to the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) Statement for cohort studies²¹, and the corresponding checklist has been included as a Supplemental File.

Data collection and extraction

Data were extracted from electronic patient records and the acute NIV quality database, including demographic data (age, gender, BMI, smoking status and FEV₁% predicted from the most recent documented spirometry), comorbidities (restrictive thoracic pathologies (including chest wall and neuromuscular disorders), OHS, cardiovascular disease, chronic liver and kidney diseases), clinical descriptors (vitals, ABG findings, DNAR status, ceiling of care for AHRF), and treatment history (domiciliary NIV, CPAP, long-term oxygen therapy (LTOT), acute NIV initiation location, maximum IPAP and EPAP settings within first 24 h). The details of comorbidities were obtained from medical records. Cardiovascular disease includes ischaemic heart disease, valvular heart disease, heart failure, cerebrovascular disease and peripheral vascular disease. When OHS was recorded as a comorbidity, it was defined by the respiratory physician as a contributing rather than primary cause of respiratory failure in patients with a BMI of more than > 30 kg/m². Clinical severity was assessed using the NIVO score,¹² calculated based on chest radiographs, Glasgow Coma Score at NIV initiation, and other clinical parameters (see Table 1). Clinical Frailty Score (CFS) and DECAF scores were computed as described in previous studies.^11,12,14,22

Table 1.

Demographics, clinical parameters and outcomes of included patients stratified according to the presence of in-hospital mortality.

Variable^a	Total (N = 249)	Survivors (N = 183)	Non-survivors (N = 66)	p-Value
Demographics
Age, years	71.4 ± 9.5	69.8 ± 9.2	75.7 ± 9.0	<0.001
Female n (%)	135 (54.2)	104 (56.8)	31(47)	0.168
BMI, kg/m²	25.1 (20.5–31.1)	25.0 (20.5–31)	25.6 (20.4–31.3)	0.964
Current smoker	114 (46.3)	93 (51.7)	21 (31.8)	0.004
FEV1, % predicted	40.9 ± 17.4	40.4 ± 17.2	42.8 ± 18.6	0.714
LTOT	42 (16.9)	32 (17.5)	10 (15.2)	0.664
Domiciliary NIV	19 (7.6)	15 (8.2)	4 (6.1)	0.575
CPAP	10 (4.0)	6 (3.3)	4 (6.1)	0.324
DNR status at admission	178 (71.5)	124 (67.8)	54 (81.8)	0.030
Ceiling of treatment				<0.001
Ward-based NIV	194 (77.9)	132 (72.1)	62 (93.9)
HDU-based NIV	7 (2.8)	5 (2.7)	2 (3.0)
Full escalation	48 (19.3)	46 (25.1)	2 (3.0)
Comorbidities
Obesity hypoventilation	33 (13.3)	24 (13.1)	9 (13.6)	0.915
Restrictive thoracic pathology^b	14 (5.6)	12 (6.6)	2 (3)	0.286
Atrial fibrillation	39 (15.9)	27 (14.9)	12 (18.8)	0.471
Cardiovascular disease	91 (36.8)	63 (34.6)	28 (43.1)	0.225
Diabetes mellitus	60 (24.2)	45 (24.7)	15 (22.7)	0.745
Hypertension	115 (46.4)	82 (45.1)	33 (50)	0.490
Chronic kidney disease	29 (11.8)	19 (10.5)	10 (15.4)	0.295
Chronic liver disease	12 (4.9)	11 (6.1)	1 (1.6)	0.150
Before NIV start
pH	7.24 ± 0.08	7.24 ± 0.08	7.23 ± 0.07	0.384
pCO₂, mmHg	9.8 (8.6–12.2)	9.7 (8.6–12.2)	10.7 (8.9–12.3)	0.057
Heart rate, per min	104.0 ± 20.1	103.6 ± 19.8	104.9 ± 20.9	0.664
Respiratory rate, per min	24 (20–28)	23 (20–28)	25 (21–31)	0.121
Door-to-mask time, h	9.4 (4.8–43.4)	7.6 (4.5–35.0)	18.6 (5.5–72.9)	0.008
Location of NIV initiation	Ward: 169 (67.9)	Ward: 115 (62.8)	Ward: 54 (81.8)	0.017
	A&E: 79 (31.7)	A&E: 67 (36.6)	A&E: 12 (18.2)
	ITU: 1 (0.4)	ITU: 1 (0.5)	ITU: 0
Outcomes & scores
NIV failure	93 (37.3)	29 (15.8)	64 (97.0)	<0.001
Clinical frailty score	4.0 (3.0–5.0)	4.0 (3.0–5.0)	4.0 (3.0–6.0)	0.066
DECAF score	2.0 (1.0–3.0)	2.0 (1.0–3.0)	3.0 (2.0–4.0)	<0.001
NIVO score	3 (1–4)	2 (1–4)	4 (2.8–5)	<0.001
Duration of NIV, days	3 (1–6)	4 (2–6)	1 (0–4)	<0.001

Missing data as follows (n = total/survivor/nonsurvivor): BMI (1/0/1), spirometry (99/62/37), smoking status (3/3/0), comorbidities: atrial fibrillation (4/2/2), cardiovascular disease (2/1/1), diabetes mellitus (1/1/0), hypertension (1/1/0), chronic kidney disease (3/2/1), chronic liver disease (4/2/2), heart rate (12/10/2), respiratory rate (15/12/3), clinical frailty score (4/2/2).

Includes kyphoscoliosis (3), lobectomy (2), polio sequelae (1), prematurity (1), alcohol-/opioid-related hypoventilation (5), pulmonary oedema (3), pleural calcification/surgery (2).

Values are presented as mean ± standard deviation, median (interquartile range) or number (percentage).

BMI, body mass index; CPAP, continuous positive airway pressure; DECAF, dyspnoea, eosinopenia, consolidation, acidaemia, atrial fibrillation; DNR, do-not-resuscitate; FEV₁, forced expiratory volume in 1 s; HDU, high-dependency unit; LTOT, long-term oxygen treatment; NIV, non-invasive ventilation; NIVO, non-invasive ventilation outcomes.

Study outcomes

Mortality data were sourced from hospital records, with in-hospital mortality as the primary outcome, defined as death occurring before discharge, and one-year mortality tracking outcomes up to a year after admission. NIV failure was defined as NIV withdrawal because of intolerance or intubation requirement because of worsening clinical conditions despite NIV therapy.

Statistical analysis

To reduce the risk of bias, statistical analyses were pre-specified. Statistical analysis was conducted using SPSS statistical analysis software, version 12.0 (SPSS Inc, Chicago, IL, USA) and R version 4.2.3, The R Project for Statistical Computing.

All analyses were performed at the patient level, restricted to each patient’s first hospital admission requiring NIV during the study period, to ensure independence of observations. Repeat admissions were excluded from inferential analysis and were described descriptively only. Continuous variables were expressed as mean ± standard deviation (SD) if parametric or as median (interquartile range (IQR)) if non-parametric; groups were compared using independent-samples t tests or Mann-Whitney U test, respectively. The chi-square test was used for categorical data, summarised as frequencies with percentages. Missing data were handled by analysis-specific case exclusion in SPSS, with cases excluded listwise for each statistical test.

Longitudinal changes in pCO2 across serial blood gas measurements were evaluated using the Friedman test, performed separately in survivors and non-survivors. These repeated-measures analyses included only patients with complete pCO2 data across all time points. Missing data were handled using analysis=specific listwise exclusion as implemented in SPSS.

Hospital mortality was modelled using logistic regression using the first admission for unique patients only. An initial full model included age, gender, BMI, current smoking status, COPD, OHS, CVD, OSA, domiciliary NIV, LTOT, and NIVO score. To avoid multicollinearity, variables overlapping with the NIVO score were excluded. Backward stepwise selection was performed, forcing the NIVO score as a core variable. Model assumptions, including linearity in the log-odds, absence of multicollinearity, and independence of observations, were checked and deemed appropriate. A sensitivity analysis including FEV₁% predicted (where available) was conducted to account for disease severity, as this variable had some missing data. Kaplan-Meier survival analysis was also performed using first admissions only and stratified by NIVO score categories (low: 0–2, medium: 3–4, high: 5–6 and very high: 7–9). Differences in survival across strata were assessed using the log-rank test.

Discrimination and calibration of the NIVO score for in-hospital mortality were also evaluated using the first hospital admission for each patient requiring NIV. Logistic regression was performed with in-hospital mortality as the dependent variable and the NIVO score as the predictor. Predicted probabilities were derived from the fitted model. Receiver operating characteristic (ROC) curve analysis was performed using the pROC package in R to assess model discrimination, with the AUC reported alongside the ROC plot. Model calibration was evaluated graphically by grouping patients into deciles of predicted risk and plotting observed versus predicted mortality. Goodness-of-fit was further assessed using the Hosmer–Lemeshow test, with p < 0.05 indicating potential miscalibration.

Results

Patient demographics across the whole cohort

Two-hundred forty-nine patients (with a total of 314 admissions due to AECOPD) were included in the primary analysis, representing each patient’s first ward-based NIV-treatment for AECOPD at Heartlands Hospital, Birmingham, from June 2021 to December 2024 (demographics and clinical parameters in Table 1). The mean (SD) age of patients was 71.4 (9.5), with 54.2% being female, 46.3% were current smokers and 51.1% were ex-smokers. Spirometry data were available for 60.0% of patients, and their mean (SD) FEV₁% predicted was 40.9 (17.4). 106 patients out of 150 (70.7%) had severe or very severe COPD. 15.7% of the patients also had OHS or restrictive thoracic pathology, including chest wall and neuromuscular disorders. Before admission, 16.9% patients were receiving LTOT, 7.6% were on domiciliary NIV, and 4% were on CPAP.

Clinical parameters and differences between patients with in-hospital mortality versus those who survived

Among included patients, the overall NIV failure rate was 37.3% (n = 93), with an in-hospital mortality rate of 26.5% (n = 66) and a one-year mortality rate of 47.0% (n = 117; Table 1). Compared to survivors, patients who died during the index admission were older, less likely to be current smokers, and more frequently had a DNACPR status or ward-based NIV ceiling of care at admission. Non-survivors also had a longer door-to-mask time and different locations for NIV initiation, with initiation occurring more frequently on the ward rather than the emergency department. These patients presented with higher respiratory rates compared to survivors and required similar maximum IPAP and EPAP in the first 24 h (19.4 (3.9) versus 19.6 (3.9) cm H₂O and 4.9 (1.6) versus 5.0 (1.6), survivors versus non-survivors, p = 0.78 and p = 0.68, respectively). On the other hand, survivors demonstrated a more pronounced and sustained reduction in median pCO2 over time (Figure S1). Among survivors, median length of stay was 10 (6–17) days, whereas among non-survivors, median time to death was 5 (2–12.5) days, p < 0.001. Similarly, survivors required a longer duration of NIV support (median of 4 (2–6) days) compared with non-survivors (1 (0–4) days), p < 0.001).

Differences in NIVO score components in patients with in-hospital mortality

When the individual NIVO score parameters were compared (Table 2), there was no significant difference in the proportion of patients with an eMRCD score of 5a between groups. However, a significantly higher proportion of non-survivors had an eMRCD score of 5b. Non-survivors were more likely to have a time from admission to acidaemia exceeding 12 h, with a nonsignificantly higher rate of blood acidaemia at pH < 7.25. Consolidation on chest x-ray was also significantly more common among non-survivors. A non-significant trend was observed for lower GCS (<14) in non-survivors. However, the proportion of patients with atrial fibrillation at presentation did not differ significantly between the groups. It is important to note that this study was not powered to re-validate the NIVO score or assess redundancy between its component variables, as the score was originally derived and validated in over 1200 patients across 10 centres.¹²

Table 2.

NIVO score parameters across survivors versus non-survivors.

Variable^a	Total n = 249	Survivors n = 183	Non-survivors n = 66	p-Value
eMRCD 5a	82 (33.1)	59 (32.4)	23 (34.8)	0.719
eMRCD 5b	44 (17.7)	24 (13.2)	20 (30.3)	0.002
CXR consolidation	70 (28.1)	39 (21.3)	31 (47.0)	<0.001
Glascow coma scale ⩽14	66 (26.7)	44 (24.3)	22 (33.3)	0.156
Atrial fibrillation	41 (16.6)	27 (14.8)	14 (21.5)	0.212
pH < 7.25	97(39.0)	67(36.6)	30 (45.5)	0.207
Time to acidaemia > 12 h	48 (19.3)	28 (15.3)	20 (30.3)	0.008

Values are presented as number (percentage).

Missing data as follows (n = total/survivor/nonsurvivor): eMRCD5a (1/1/0), eMRCD 5b (1/1/0), Glascow Coma Scale < 14 (2/2/0), Atrial fibrillation (2/1/1).

CXR, chest X-ray, eMRCD, extended medical research council dyspnoea scale; NIVO, non-invasive ventilation outcomes.

Overall, in this real-world ward-based NIV cohort, NIVO scores were significantly higher among non-survivors compared to survivors, with a median (IQR) of 4 (2.8–5) versus 2 (1–4), respectively, p < 0.001. DECAF scores were also significantly higher in non-survivors, with median (IQR) values of 3 (2–4) versus 2 (1–3), p < 0.001.

Association of NIVO score with in-hospital mortality

To explore the association between NIVO scores and in-hospital mortality and therefore the potential utility of the NIVO score to predict outcomes in clinical practice, we developed a logistic regression model across the first hospital admission of the 249 patients. The final logistic regression model included current smoking status, LTOT, and NIVO score (Table 3). The NIVO score was significantly associated with in-hospital mortality, with an odds ratio OR of 1.52 (1.26–1.86), p < 0.001, indicating that higher NIVO scores were associated with increased risk of death. Age was independently associated with higher odds of in-hospital mortality (OR 1.07, 95% CI 1.03–1.11; p = 0.001).

Table 3.

Logistic regression model for in-hospital mortality for unique patients on their first admission only.

Predictors	Mortality (hospital)
Predictors	Odds ratios	CI	p
Age	1.07	1.03–1.11	0.001
Male	1.85	0.97–3.57	0.063
NIVO score	1.52	1.26–1.86	<0.001

Logistic regression results for in-hospital mortality, using patients’ first admission only, including key demographic and clinical variables such as age, current smoking status, long-term oxygen therapy (LTOT), and NIVO score. Observations: 240. R² = 0.197.

CI, 95% confidence intervals; NIVO, non-invasive ventilation outcomes; OR, Odds ratios.

A sensitivity analysis, incorporating FEV₁% predicted into the model, demonstrated that the association between NIVO score and in-hospital mortality remained significant, OR (95% CI) of 1.50 (1.16–1.98), p = 0.003, Table S1). The NIVO score demonstrated moderate discrimination for in-hospital mortality with an area under the ROC curve (AUC) of 0.71 (Figure S2). Calibration of the model was broadly acceptable (Figure S3), with predicted and observed mortality showing a generally increasing relationship across deciles of predicted risk, although there was mild underestimation of mortality in the highest-risk groups. The Hosmer–Lemeshow goodness-of-fit test indicated a statistically significant difference between predicted and observed outcomes (p = 0.008), suggesting some lack of perfect calibration, likely reflecting minor deviations at the extremes of risk prediction.

Recognising that not all patients with NIV failure died during hospitalisation, we conducted an additional regression analysis to assess the association between NIVO score and NIV failure. NIVO score remained significantly associated with higher odds of NIV failure, 1.33 (1.13–1.58), p < 0.001 (Table S3).

Finally, we assessed patient outcomes according to NIVO score categories by stratifying patients into low (0–2), medium (3–4), high (5–6) or very high (7–9) risk groups (Table 4); which demonstrated a progressive increase in in-hospital mortality rates across ascending categories, 13.6%, 26.7%, 59.0% and 80.0% in low, medium, high and very high-risk groups, respectively. The Kaplan-Meier survival curves demonstrated a clear gradient in survival across these categories with an increase in in-hospital mortality across ascending NIVO score groups (log-rank full test p < 0.0001; Figure 1).

Table 4.

In-hospital mortality rates according to NIVO score categories.

NIVO score	In-hospital mortality, n (%)
Low (0–2)	16 (13.6)
Moderate (3–4)	23 (26.7)
High (5–6)	23 (59.0)
Very high (7–9)	4 (80.0)

Showing in-hospital mortality in patients split according to NIVO score risk groups, including the following groups low (0–2) moderate (3–4) high (5–6) very high (7–9).

NIVO, non-invasive ventilation outcomes.

Figure 1.

Kaplan–Meier curve showing survival in patients split according to NIVO score. In hospital survival shown over days for patients split according to NIVO score, (Dark blue) Low (0–2), (Red) medium (3–4), (Green) High (5–6), (Light blue) Very high (7–9).

Discussion

In this study, we demonstrate the utility of the NIVO tool in predicting in-hospital mortality and NIV failure among patients with AECOPD treated with ward-based NIV in a real-world UK hospital setting. Our findings confirm a strong, graded association between higher NIVO scores and increased in-hospital mortality, with significant differences in outcomes across NIVO score categories. These findings underscore the potential use of the NIVO score as a practical tool for patient stratification in real-world settings to aid clinical decision-making and improve COPD patient outcomes in acute NIV care.

The original study, which derived and provided initial validation for the NIVO score, was undertaken in diverse care settings. However, while previous follow-on studies have reported the predictive value of the NIVO score in ICU settings and international cohorts,^{12,13,18,19,23} this present study is the first to further validate its use specifically in a UK ward-based NIV population. For example, NIVO has been associated with mortality in a large Chinese cohort with a comparatively low mortality rate.¹⁸ A Turkish study demonstrated its potential for predicting risk of requiring assisted ventilation, while a French ICU study showed its association with NIV failure and long-term outcomes in a younger population of ICU patients.^19,23 Moreover, the NIVO score was recently incorporated into the national Respiratory Support Unit audit, which achieved 100% completion and demonstrated strong performance as a risk-adjustment tool; notably, Respiratory Support Unit-level care was associated with improved survival after adjustment for baseline risk using the NIVO score.²⁴ Our study builds on these findings by demonstrating the NIVO score’s predictive value for in-hospital mortality in an UK cohort, characterised by a comparatively older, multimorbid population, with a high proportion of patients having ward-based ceiling of care (72% had a DNACPR status). This UK study represents an important extension of prior follow-up studies, as the UK healthcare system faces unique challenges with NIV.²² In addition, patients with ward-based ceiling of care often have higher levels of frailty and comorbidity, leading to complex decision-making needs; further reinforcing the demand for validated, pragmatic decision tools to aid this process in this population specifically. Unlike prior ICU studies, NIV in this cohort was delivered in ward settings, which allows early initiation, greater accessibility, efficient allocation of resources, and continuity of patient-centred care.^25,26 Therefore, our results strongly support the integration of NIVO scoring into routine clinical assessment to guide decision-making and escalation planning in patients requiring acute NIV for AECOPD.

Our analysis of model performance showed that the NIVO score demonstrated moderate discrimination for in-hospital mortality and generally good calibration, with only minor deviations at the extremes of predicted risk. These findings indicate that, within this ward-based NIV population, the NIVO score provides a reasonable balance of discrimination and calibration suitable for use in routine clinical assessment. Moreover, although our study primarily focused on in-hospital outcomes, our descriptive analysis of one-year mortality suggests that the prognostic value of the NIVO score may extend beyond the index admission, consistent with recent reports in other cohorts.^19,23

We add further context for the clinical utilisation of the NIVO tool by demonstrating its association with NIV failure, an outcome that does not always result in mortality. In this study, NIV failure was defined as NIV withdrawal or intubation requirement following clinical deterioration despite NIV therapy; notably, none of the patients in this cohort proceeded to invasive mechanical ventilation. The observed association between NIVO score and NIV failure represents an important additional finding and extends the findings of prior work by Yavuz Yıldırım et al. and Alzaabi, who assessed this outcome in patients requiring assisted ventilation and ICU-level care, respectively.^19,23 Alongside total NIVO scores, we identified differences in individual score components between survivors versus non-survivors, although this study was not powered to re-validate the NIVO score or assess redundancy among its component variables. Notably, in non-survivors, time to academia was significantly longer and NIV was more frequently initiated on the ward rather than in the emergency department. Consistent with this, door-to-mask time was significantly longer among non-survivors, and although we could not fully distinguish between delayed recognition/treatment and later development of acidaemia, a small subgroup (n = 20) presented with normal pH values at admission, suggesting both factors may have contributed. These findings underscore the importance of early detection of academia and prompt NIV initiation for appropriately selected patients, in line with the BTS quality standards for NIV delivery.^25,27,28 Conversely, inappropriate hyperoxygenation, the development of infectious complications, and/or multiple organ failure during hospitalisation should be considered potential contributors to delayed acidaemia, underscoring the need for vigilant monitoring throughout the admission.

Our study was a single-centre, retrospective observational study involving 249 patients’ first admissions for AECOPD managed with ward-based NIV. While this represents a substantial cohort, we acknowledge that larger multicentre studies may reveal additional associations not detected here. Moreover, it is possible that missing data or the misclassification or underreporting of data in routine clinical records could have introduced bias or impacted our results. Important treatment-related variables, such as exacerbation history, background medications, and AECOPD cause, were incompletely recorded in our retrospective dataset, therefore not provided for analysis; however, this limitation has also been noted in other large multicentre NIVO cohorts.^12,18 Our descriptive finding of a lower proportion of current smokers in the non-survivor group is consistent with a “healthy smoker” effect, a form of bias whereby patients with more advanced disease are more likely to have stopped smoking before admission.²⁹ Notably, smoking did not associate with outcome in our analysis. A minority of patients were treated based on a clinical diagnosis of AECOPD without spirometric confirmation, reflecting routine acute practice where spirometry is frequently unavailable at the time of admission.³⁰ Moreover, the use of the NIVO score has been shown to be valid and supported in patients with a clinical diagnosis of COPD as well.¹³ This enhances the generalisability of our findings to real-world ward-based NIV populations. Our study provides insights into the use of this tool in an older, multimorbid NIV cohort, many of whom had ward-based ceiling of care. While these findings provide important evidence for this important population, a high rate of DNR orders may have influenced treatment decisions, thereby limiting generalisability to other populations (such as ICU-managed cohorts). Further multicentre validation studies are therefore warranted to confirm the broader applicability of the NIVO score.

Conclusion

Accurate, pragmatic tools are essential to support complex clinical decision-making, facilitate patient stratification and guide discussions with patients and families for improved delivery of NIV. This study demonstrates the prospect of a novel NIVO tool to predict in-hospital mortality in a real-world cohort of patients with AECOPD receiving ward-based NIV. Our findings support its use as a practical aid in guiding treatment decisions, planning escalation of care, and enhancing the efficiency and quality of NIV delivery within new clinical pathways designed to address the growing burden of COPD exacerbations.

Supplemental Material

sj-doc-1-tar-10.1177_17534666261441160 – Supplemental material for Predicting in-hospital mortality in a real-world population of ward-based non-invasive ventilation in acute COPD exacerbations

Supplemental material, sj-doc-1-tar-10.1177_17534666261441160 for Predicting in-hospital mortality in a real-world population of ward-based non-invasive ventilation in acute COPD exacerbations by Aylin Ozsancak Ugurlu, Alastair Watson, Paul Ellis, Ravi K. Bange, Sridivyareddy Yangannagari, Madeeha Ahmed, Amy Oakes, Kirsty-Anne Ebbage and Rahul Mukherjee in Therapeutic Advances in Respiratory Disease

Supplemental Material

sj-docx-2-tar-10.1177_17534666261441160 – Supplemental material for Predicting in-hospital mortality in a real-world population of ward-based non-invasive ventilation in acute COPD exacerbations

Supplemental material, sj-docx-2-tar-10.1177_17534666261441160 for Predicting in-hospital mortality in a real-world population of ward-based non-invasive ventilation in acute COPD exacerbations by Aylin Ozsancak Ugurlu, Alastair Watson, Paul Ellis, Ravi K. Bange, Sridivyareddy Yangannagari, Madeeha Ahmed, Amy Oakes, Kirsty-Anne Ebbage and Rahul Mukherjee in Therapeutic Advances in Respiratory Disease

Footnotes

Acknowledgements

All contributors met ICMJE criteria and are therefore included within the author list. There are no further acknowledgements.

Declarations

ORCID iDs

Alastair Watson

Rahul Mukherjee

Supplemental material

Supplemental material for this article is available online.

References

Chan

Hui

, et al. Acute exacerbation of COPD. Respirology 2016; 21: 1152–1165.

Safiri

Carson-Chahhoud

Noori

, et al. Burden of chronic obstructive pulmonary disease and its attributable risk factors in 204 countries and territories, 1990–2019: results from the Global Burden of Disease Study 2019. BMJ 2022; 378: e069679.

Osadnik

Tee

Carson-Chahhoud

, et al. Non-invasive ventilation for the management of acute hypercapnic respiratory failure due to exacerbation of chronic obstructive pulmonary disease. Cochrane Database Syst Rev 2017; 7: CD004104.

Waeijen-Smit

Crutsen

Keene

, et al. Global mortality and readmission rates following COPD exacerbation-related hospitalisation: a meta-analysis of 65 945 individual patients. ERJ Open Res 2024; 10: 50838–52023.

Davies

Allen

Bentley

, et al. British thoracic society quality standards for acute non-invasive ventilation in adults. BMJ Open Respir Res 2018; 5(1): e000283.

National Confidential Enquiry into Patient Outcome and Death. Inspiring change: a review of the quality of care provided to patients receiving acute non-invasive ventilation [Internet]. London: NCEPOD, https://www.ncepod.org.uk/2017niv.html (2017, accessed 3 November 2025).

Wildman

Sanderson

Groves

, et al. Implications of prognostic pessimism in patients with chronic obstructive pulmonary disease (COPD) or asthma admitted to intensive care in the UK within the COPD and asthma outcome study (CAOS): multicentre observational cohort study. BMJ 2007; 335: 1132.

Northumbria Healthcare NHS Foundation Trust. NIV outcomes study [Internet]. London: Health Research Authority, https://www.hra.nhs.uk/planning-and-improving-research/application-summaries/research-summaries/niv-outcomes-study/ (2024 accessed 2024).

Duan

Chen

Liu

, et al. An updated HACOR score for predicting the failure of noninvasive ventilation: a multicenter prospective observational study. Crit Care 2022; 26: 196.

10.

Ding

Han

Bai

, et al. Impact of HACOR score on noninvasive ventilation failure in non-COPD patients with acute-on-chronic respiratory failure. Can Respir J 2021; 2021: 9960667.

11.

Echevarria

Steer

Heslop-Marshall

, et al. Validation of the DECAF score to predict hospital mortality in acute exacerbations of COPD. Thorax 2016; 71: 133–140.

12.

Hartley

Lane

Steer

, et al. The Noninvasive Ventilation Outcomes (NIVO) score: prediction of in-hospital mortality in exacerbations of COPD requiring assisted ventilation. Eur Respir J 2021; 58: 20210812.

13.

Lane

Hartley

Steer

, et al. The noninvasive ventilation outcomes score in patients requiring NIV for COPD exacerbation without prior evidence of airflow obstruction. ERJ Open Res 2024; 10: 00193–02024.

14.

Knaus

Draper

Wagner

, et al. APACHE II: a severity of disease classification system. Crit Care Med 1985; 13: 818–829.

15.

Wildman

Harrison

Welch

, et al. A new measure of acute physiological derangement for patients with exacerbations of obstructive airways disease: the COPD and Asthma Physiology Score. Respir Med 2007; 101: 1994–2002.

16.

Confalonieri

Garuti

Cattaruzza

, et al. A chart of failure risk for noninvasive ventilation in patients with COPD exacerbation. Eur Respir J 2005; 25: 348–355.

17.

Elshof

Duiverman

Wijkstra

PJ.

The NIVO score: can it help to improve noninvasive ventilation in daily clinical practice?

Eur Respir J 2021; 58: 00336–02021.

18.

Zhang

Zhou

, et al. Risk factors of in-hospital mortality and discriminating capacity of NIVO score in exacerbations of COPD requiring noninvasive ventilation. Chron Respir Dis 2024; 21: 14799731241249474.

19.

Alzaabi

Guerot

Planquette

, et al. Predicting outcomes in patients with exacerbation of COPD requiring mechanical ventilation. Ann Intensive Care 2024; 14: 159.

20.

Davidson

Banham

Elliott

, et al. BTS/ICS guideline for the ventilatory management of acute hypercapnic respiratory failure in adults. Thorax 2016; 71: ii1–ii35.

21.

von Elm

Altman

Egger

, et al. The Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement: guidelines for reporting observational studies. J Clin Epidemiol 2008; 61: 344–349.

22.

NCEPOD. Inspiring change: improving the outcomes for patients receiving acute non-invasive ventilation. London: National Confidential Enquiry into Patient Outcome and Death, 2017.

23.

Yavuz Yıldırım

Akbay

MÖ

Kesen Yurtcanlı

, et al. Non-invasive ventilation outcome (NIVO) score for predicting in-hospital and late mortality of chronic obstructive pulmonary disease (COPD) patients with acute hypercapnia. Eurasian J Pulmonol 2025; 27: 27.

24.

British Thoracic Society. Respiratory Support Unit National Audit: National Patient-Level Report, June 2024 [Internet]. https://www.brit-thoracic.org.Uk/media/q2zdyzkk/bts-respiratory-support-audit-national-patient-level-report-final-june-2024.pdf (2024, accessed 3 November 2025).

25.

British Thoracic Society. BTS quality standards for acute NIV in adults. [Internet]. https://www.brit-thoracic.org.uk/document-library/quality-standards/niv/bts-quality-standards-for-acute-niv-in-adults/ (2018, accessed 4 June 2025).

26.

Mukherjee

Nenna

Turner

Early ward-based acute noninvasive ventilation: a paper that changed practice. Breathe (Sheff) 2018; 14: 153–155.

27.

Watson

Barnard

Shanmugarajah

, et al. Response times for acute non-invasive ventilation set-ups. Ulster Med J 2022; 91: 169.

28.

Watson

Barnard

Antoine-Pitterson

, et al. The impact of COVID-19 on acute non-invasive ventilation services: a case for change. Respirology 2021; 26: 1106.

29.

Becklake

Lalloo

The ‘healthy smoker’: a phenomenon of health selection?

Respiration 1990; 57: 137–144.

30.

Hurst

Amusan

Akinyooye

, et al. National asthma and chronic obstructive pulmonary disease audit programme (NACAP). COPD clinical audit 2019/20: data analysis and methodology report. London: Royal College of Physicians of London, 2021.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.09 MB

0.20 MB