Predictors of Adverse Outcome in Patients Hospitalised for Exacerbation of Chronic Obstructive Pulmonary Disease

Exacerbations of chronic obstructive pulmonary disease (COPD) are unfavourable events in the course of disease for most COPD patients. Published evidence indicates a significant impact of exacerbations, especially if frequent, on patients’ health-related quality of life (HRQL) [1,2], disease progression [3], mortality [4,5,6,7,8,9,10,11,12,13], health care utilisation and costs [14,15]. However, the severity, evolution and outcome of an exacerbation may differ significantly between patients – some patients will recover completely in a short period of time while others may die. The identification of risk factors for an adverse outcome could help in distinguishing patients who require more intense management in order to prevent failures, achieve satisfactory recovery and reduce the negative clinical and socioeconomic impact of exacerbations.

Formerly identified predictors of poor outcome of COPD exacerbations are various and inconsistent due to the heterogeneity of previous studies in terms of assessed variables, study population [outpatients, inpatients or intensive care unit (ICU) patients] and outcomes of interest (e.g. need for hospitalisation, ICU admission, length of hospital stay, changes in lung function and HRQL, readmission or mortality) [4,5,6,7,8,9,10,11,12,13,16,17,18,19,20,21]. Regarding exacerbations requiring hospital treatment, some of the most frequently recognised risk factors for mortality and readmission are age, low body mass index (BMI), low forced expiratory volume in 1 s (FEV₁), hypercapnia, need for long-term oxygen therapy (LTOT), comorbidity and previous exacerbations [4,5,6,7,8,9,10,11,12,13,18,19,20,21]. However, the usefulness of the newly developed multidimensional prognostic indices of COPD for the prediction of outcomes in exacerbated patients has seldom been assessed.

By analysing data recorded in the health care database of our institution, we aimed to investigate the evolution of hospitalised COPD patients and to identify predictors of adverse outcome in routine clinical practice. In contrast to previous studies of risk factors, we used a composite adverse outcome, taking into account different negative events in the course of the exacerbation, during both the index hospitalisation as well as the 1-month post-hospital period.

This was an observational investigation of prospectively collected data from patients hospitalised for a COPD exacerbation in a tertiary acute-care hospital (Hospital Clínic, Barcelona, Spain) between May 2009 and December 2010. The health care database contained the data obtained as part of routine clinical procedures. The patients included had to fulfill the COPD criteria set by the Global Initiative for Chronic Obstructive Lung Disease (GOLD), including demonstration of a post-bronchodilator FEV₁/forced vital capacity <0.7 in the clinical records, performed in the stable phase prior to admission [22]. The diagnosis of COPD exacerbation, decision to hospitalise, time of discharge and choice of in-hospital and post-hospital therapy were made by the treating physician. All patients discharged from our institution after an exacerbation of COPD are scheduled for a follow-up visit at 1 month. Data for each subject were included in the analysis only once, despite the possibility of more than one admission during the observation period. The ethics committee of the Hospital Clínic Barcelona evaluated the study and determined that approval is not required for observational studies of routinely collected clinical data.

An adverse outcome of COPD exacerbation was a composite outcome, as in Rothberg et al. [23], defined by the presence of at least one of the following: (1) death from a respiratory cause during hospitalisation or within 1 month of follow-up; (2) ICU admission; (3) invasive or non-invasive mechanical ventilation; (4) prolonged hospitalisation, defined by the upper quartile of the distribution of hospital stays, and (5) COPD-related emergency room visit or readmission during follow-up. A good outcome was considered to exist in the absence of all the above-mentioned issues.

Demographic and clinical data, including vital signs and laboratory findings at admission, were obtained from the health care database. Besides basic blood laboratory tests (leucocytes, hematocrit, C-reactive protein, glucose, creatinine), arterial blood analysis was recorded with pH, arterial carbon dioxide pressure (PaC O₂) and the ratio of arterial oxygen pressure to the fraction of inspired oxygen (PaO₂/FiO₂).

Apart from spirometry and GOLD categorisation [22], for the additional determination of COPD severity, several multidimensional prognostic indices were calculated based on the information included in the database. The COPD severity score (COPDSS) [24] consists of a series of questions about respiratory symptoms, COPD treatment and previous exacerbations (maximum score 35). It was proven useful in ambulatory patients for the assessment of COPD severity and prediction of health care utilisation and outcomes of exacerbations [24,25,26] and was therefore adopted into routine clinical practice in our institution. The COPDSS questionnaire was previously translated and validated in Spanish [27]. The DOSE index [28] is composed of dyspnoea degree (D), severity of airflow obstruction (O), current smoking status (S) and annual number of exacerbations (E) and was designed to predict unfavourable events in the course of COPD such as future exacerbations, hospitalisations and respiratory failure. The BODEx index [29] is composed of BMI (B), airflow obstruction (O), dyspnoea (D) and number of exacerbations in the previous year (Ex) and was proven to be a good prediction tool for mortality in COPD. The ADO index [30] was also developed to predict mortality in COPD patients and is composed of age (A), dyspnoea (D) and airflow obstruction (O). The dyspnoea domain of the COPDSS (ranging from 0 to 3) was used to build the DOSE, BODEx and ADO indices; therefore, the maximum final scores were 7, 8 and 9 points, respectively. The BODE index [31] could not be calculated because the 6-min walking test is not usually performed in patients admitted to our hospital.

Comorbid medical conditions were evaluated according to the Charlson index [32], which takes into account the number and severity of common chronic diseases. Since there is a high prevalence of cardiovascular diseases in the COPD population [33], we wanted to explore their impact on outcome in more detail. Therefore, besides the cardiovascular conditions that are already in the Charlson index (myocardial infarction, congestive heart failure, peripheral vascular disease), we also evaluated arterial hypertension, arrhythmias, non-ischaemic cardiomyopathy and different forms of coronary artery disease (angina pectoris, ischaemic cardiomyopathy, prior percutaneous coronary artery intervention) which were recorded in the clinical records.

The number of COPD exacerbations in the year preceding the index hospitalisation was assessed based on recorded data and a search of the database of all hospital contacts. Only exacerbations requiring hospital management (emergency room visit or admission) were considered.

Index hospitalisation data included the length of stay, weekend admission (admission on Friday, Saturday or Sunday), admission to the ICU, invasive or non-invasive mechanical ventilation and death from a respiratory cause. During the 1-month follow-up period, data on all COPD-related emergency visits and readmissions were obtained, as well as deaths from a respiratory cause.

Descriptive statistics were used to describe the study population. Categorical variables are presented as absolute numbers and percentages and continuous variables as means ± standard deviations (SD) or medians with the 1st and 3rd quartile in the case of non-normal distribution. Categorical variables were compared with the χ² test or Fisher’s exact test, as appropriate. For the comparison of continuous variables, Student’s t test was used if normality was demonstrated; otherwise, the nonparametric Mann-Whitney U test was performed.

The associations of the demographic and clinical variables, as well as prognostic indices, with the outcome were assessed using univariate and multivariate logistic regression analysis. Multidimensional prognostic indices significantly associated with an adverse outcome on univariate analysis (COPDSS, BODEx and DOSE indices) were used to construct three corresponding multivariate logistic regression backward stepwise models, each including one of the indices together with the remaining variables with a significant univariate association (p < 0.1). The ADO index was not included in the multivariate analysis because it was not associated with adverse outcome in the univariate analysis. Highly correlated variables, as well as those already a part of a prognostic index, were excluded from the multivariate analysis. Receiver operating characteristic (ROC) curves were constructed to determine the best cut-off points for continuous variables. The Hosmer-Lemeshow goodness-of-fit test was performed to assess the overall fit of the models [34]. Comparisons between the areas under ROC curves were assessed according to the method of Hanley and McNeil [35]. All tests were two-tailed, and significance was set at 5%. All analyses were performed with SPSS version 16.0 for Windows (SPSS Inc., Chicago, Ill., USA).

Of the 160 patients admitted during the observation period with the diagnosis of exacerbation of COPD, an alternative diagnosis was recorded in 5 patients (3%) at discharge, and they were excluded from the analysis. The population therefore comprised a total of 155 patients, with a predominance of elderly men [130 men (84%); mean ± SD age 70.0 ± 9.5 years] with high smoking exposure (63.0 ± 32.6 pack-years; table 1). On average, patients had severe airway obstruction with a mean FEV₁ of 42.4 ± 15.5% predicted, while 113 patients (76%) were in GOLD stage III or IV. Previous COPD exacerbations were frequent, with 51% of patients having had at least one severe exacerbation in the year preceding index hospitalisation (mean 1.1 ± 1.5, maximum 8 exacerbations), while 37% were also admitted to hospital.

The study population had a significant comorbid burden, with a median (1st quartile; 3rd quartile) Charlson index of 2.0 (1.0; 3.0); 135 patients (87%) had at least one comorbid disease [median 2.0 (1.0; 4.0), maximum 8 comorbidities], the most frequent being cardiovascular diseases in 114 patients (74%), diabetes mellitus in 40 (26%) and malignant diseases in 26 (17%). Among the cardiovascular diseases, the most prevalent was arterial hypertension in 91 patients (59% of the study population), followed by congestive heart failure in 31 patients (20%), arrhythmia with predominance of atrial fibrillation in 31 (20%), coronary artery disease in 24 (16%), non-ischaemic cardiomyopathy in 23 (15%), 5 of whom had cor pulmonale, and peripheral vascular disease in 15 (10%). The scores of the four multidimensional prognostic indices are presented in table 1 together with demographic and clinical data.

At admission, the average patient was tachypnoeic and afebrile with the presence of mild hypoxaemia, hypercapnia, hyperglycaemia, leucocytosis and elevation of C-reactive protein (table 2).

The mean duration of hospitalisation was 8.8 ± 5.5 days, while 11 days was established as the cut-off to define prolonged hospitalisation (upper quartile). Altogether, 69 patients (45%) had an adverse outcome of exacerbation (fig. 1). Death due to a respiratory cause occurred in 4 patients (3%); 1 patient (1%) died during the index hospitalisation and 3 (2%) during the 1-month follow-up period. Admission to the ICU was required by 16 patients (10%), and 33 patients (21%) were mechanically ventilated with either invasive or non-invasive ventilation or both. Prolonged hospitalisation was recorded in 32 patients (21%). In the first month after discharge, 24 patients (16%) requested hospital assistance for COPD-related symptoms, 7 of whom (5%) had an emergency visit without admission and 17 of whom (11%) were readmitted.

On average, patients with adverse outcomes had a lower FEV₁ (p = 0.004) and higher GOLD stage (p = 0.049), more frequently used oxygen at home (p = 0.042) and had more frequent exacerbations with or without admission within the previous year (p = 0.039 and 0.011, respectively). They also achieved significantly worse scores on the COPDSS, DOSE and BODEx indices (p = 0.016, 0.009 and 0.005, respectively) but not on the ADO index (p = 0.237; table 1). Signs of respiratory failure at admission in terms of lower PaO₂/FiO₂, higher PaC O₂ and lower pH were also associated with adverse outcome (p = 0.006, <0.001 and <0.001, respectively; table 2).

No significant differences in terms of outcome were found for age, sex, BMI, smoking history, common comorbidities, vital signs and basic haematological and biochemistry findings at admission. Since the prevalence of isolated arterial hypertension (without other cardiovascular conditions) and congestive heart failure in the study population was high (30 and 20%, respectively), we explored if there was a difference in the prevalence of these conditions among patients with good and adverse outcome separately but did not find a significant difference.

Fifty-eight patients (37%) were admitted at the weekend but were not found to have a higher risk for adverse outcome compared with patients admitted on weekdays.

Statistically significant variables from univariate analysis and three multivariate logistic regression models (A, B and C) are presented in table 3. Hypercapnia and hypoxaemia at admission were recognised as independent predictors of adverse outcome by all three multivariate models. Hypercapnia was the most relevant predictor, and according to model A, patients with hypercapnia had a 9-fold higher risk of adverse outcome than those with normocapnia [odds ratio 9.4, 95% confidence interval (CI) 3.7–23.6; p < 0.001]. The remaining independent predictors were exacerbation in the previous year, BODEx index score ≥4 and DOSE index score ≥4, identified in models A, B and C, respectively. All three models were well calibrated, with p values of 0.340, 0.511 and 0.329, respectively, with the Hosmer-Lemeshow test.

Significant variables derived from different models were compared for their capacity to predict adverse outcome using ROC curve analysis (fig. 2). The area under the ROC curve was 0.843 (95% CI 0.770–0.916) for model A, 0.839 (95% CI 0.765–0.912) for model B and 0.835 (95% CI 0.759–0.911) for model C, with comparisons among them showing no statistically significant differences (p = 0.84, 0.73 and 0.86 for the comparison of models A and B, A and C, and B and C, respectively).

Predictors from model A were used to calculate the probability of adverse outcome given by the following formula: Exp(β)/(1 + Exp(β)), where β = –2.095 + 1.365 (in the case of COPD exacerbation in the previous year) + 2.237 (if PaC O₂ ≥48.0 mm Hg) + 1.455 (if PaO₂/FiO₂ <226.5 mm Hg). All the previous cut-offs used for continuous variables were identified by means of ROC curves. The risk of adverse outcome increased progressively with the number of predictors, being 11% for the patients without any of these characteristics, 54% for those with hypercapnia and 95% for those having all three characteristics. The probabilities of adverse outcome for the different combinations of risk factors are shown in table 4.

Our results indicate that arterial blood gas analysis together with information about previous exacerbations have a high predictive value for 1-month outcomes in patients hospitalised for COPD exacerbation. Patients with at least one exacerbation requiring hospital management in the previous year and presenting with hypoxaemia and hypercapnia in the current exacerbation had the highest risk for unfavourable evolution.

Although risk factors for poor outcome in COPD exacerbations have already been investigated by other authors, we applied a different approach, analysing patients in routine clinical practice and using a composite adverse outcome described only in intervention trials [23], with several unfavourable in-hospital and 1-month post-hospital events including death, ICU admission, mechanical ventilation, prolonged hospitalisation, repeated emergency visits and readmissions. One advantage of this approach is that despite the low mortality rate in our cohort (3%), the composite adverse outcome allowed the identification of a particular population requiring intensive treatment and follow-up.

According to the published literature, the in-hospital mortality of patients hospitalised for COPD exacerbation is 2–8% (up to 15% for ICU patients), with a 1-year mortality of 22–43% [4,5,6,7,8,9,10,11,12,13]. Some of the most frequently recognised predictors of mortality are higher age, lower BMI, lower FEV₁, hypercapnia, hypoxaemia, hypoalbuminaemia, comorbidity, cor pulmonale, previous exacerbations or admissions, LTOT and severity of dyspnoea [4,5,6,7,8,9,10,11,12,13]. Furthermore, factors associated with prolonged hospitalisations include higher respiratory rate, dyspnoea, comorbidity, cor pulmonale and weekend admission [12,16,17]. Readmissions after hospitalisation for an exacerbation are frequent events, ranging from 14 to 16% in the first month after discharge and 25–58% in the first year [5,19]. According to previous studies, risk factors for readmission are higher age, lower FEV₁, hypercapnia, comorbidity, LTOT, poor HRQL and previous exacerbations or admissions [18,19,20,21].

Some of our results concur with previous reports, demonstrating the association between indicators of COPD severity (FEV_1, LTOT, previous exacerbations, higher PaC O₂, lower PaO₂/FiO₂ and pH) and composite adverse outcome. Additionally, we identified exacerbation in the previous year and hypercapnia and hypoxaemia at admission as independent adverse prognostic factors.

To explore the predictive capacity of different prognostic indices for the outcome of exacerbation, we included four multidimensional scores in our analysis, the COPDSS and the DOSE, BODEx and ADO indices. The COPDSS was adopted into routine clinical practice after previous validation studies [26,27]. Unlike the BODE index [31], the other indices do not require the performance of exercise testing and could be calculated from the data recorded during the routine health care process. It has recently been demonstrated that these indices may be useful in the assessment of COPD severity as well as in the prediction of different negative events in ambulatory COPD patients (e.g. mortality, health care utilisation, future hospitalisations, failures in COPD exacerbations) [24,25,26,27,28,29,30], but to our knowledge, no studies have investigated their usefulness in hospitalised patients. We found that the COPDSS, BODEx and DOSE but not the ADO index were associated with adverse outcome on univariate analysis, but only the BODEx and DOSE indices appeared as independent adverse prognostic factors in the multivariate models B and C, respectively. Since calculation of indices requires additional time and data, it was of interest to determine whether any of these indices provided some additional predictive power. Comparison of significant variables derived from the three models (hypercapnia, hypoxaemia and previous exacerbation from model A; hypercapnia, hypoxaemia and BODEx index ≥4 from model B, and hypercapnia, hypoxaemia and DOSE index ≥4 from model C) did not show any statistically significant difference in their capacity to predict adverse outcome. This observation is particularly important because arterial blood gas analysis and anamnestic information about previous exacerbations are easy to obtain in the hospital setting and sometimes even in ambulatory settings, and our results suggest that this basic set of data provides a high predictive capacity that was not significantly improved by multidimensional prognostic indices. These results indicate that, although multidimensional indices are of great value as prognostic indicators in stable COPD patients, other variables are more strongly related to the clinical outcomes in severely exacerbated individuals.

Surprisingly and contrary to some previous studies [4,5,6,7,8,9,17,18], age and comorbidities did not affect outcome, nor even did cardiovascular diseases, which were highly prevalent in our cohort. This is consistent with Anzueto [36] reporting that cardiac comorbidity is a risk factor for poor outcome in moderate COPD patients, but impairment of respiratory function prevails over cardiac disease in more severe COPD (our patients were hospitalised for an exacerbation and 76% were in GOLD stage III or IV). Along these lines, the possible improvement in outcomes with statins in patients with COPD is still controversial [37]. Concerning comorbidities, it is important to mention that although diabetes did not appear as a risk factor for adverse outcome, hyperglycaemia showed statistical significance on univariate analysis. This finding is relevant because hyperglycaemia with or without confirmed diabetes has already been related to negative outcomes in different acute diseases such as myocardial infarction, stroke, pneumonia and, more recently, ambulatory COPD exacerbations [38,39,40].

Furthermore, no association was found between BMI and outcome, contrary to some previous studies suggesting low BMI to be a risk factor for poor outcome [10,11,12]. The absence of this relationship could be explained by the fact that most of our patients were of normal weight or obese. Unfortunately, we could not evaluate ECG or echocardiographic criteria for cor pulmonale in our population. Another limitation of our study was the lack of data regarding the glomerular filtration rate, which could provide information about mild kidney damage. Similarly, neither were factors related to long-term clinical outcomes such as socioeconomic status [41] included in our clinical database.

In summary, our results indicate that despite a low mortality during or 1 month after admission for exacerbated COPD, probably due to the improvements in the management of COPD exacerbations, other adverse outcomes are still frequent. Few data are necessary for the risk stratification of patients at admission. Patients with hypercapnia, hypoxaemia and previous exacerbations requiring hospital management are especially at risk for poor outcome and require more intense therapy and careful supervision.

Zinka Matkovic from the University Hospital Dubrava, Zagreb, Croatia, is the recipient of a joint European Respiratory Society/Spanish Society of Pneumology and Thoracic Surgery LTRF fellowship 47-2010.

Marc Miravitlles has participated as a speaker in scientific meetings or courses organised and financed by various pharmaceutical companies, including Boehringer Ingelheim, Bayer Schering Pharma, Pfizer, Novartis and AstraZeneca, and has been a consultant for Boehringer Ingelheim, Bayer Schering Pharma, Pfizer, AstraZeneca and Novartis. Antoni Torres has received speaking fees from Astellas, MSD and Pfizer and research grants from Pfizer and Covidien, and he is on the advisory boards of Astellas, MSD and Sanofi Pasteur. Zinka Matkovic, Arturo Huerta, Nestor Soler and Rebeca Domingo have no competing interests to declare.