Background
Chronic obstructive pulmonary disease (COPD) is a progressive lung condition and a leading cause of adult morbidity and mortality worldwide [
1‐
3]. Exacerbation of chronic obstructive pulmonary disease (ECOPD) is an event characterized by a sustained worsening of the respiratory symptoms of a patient (including cough, phlegm production, and dyspnea), beyond normal day-to-day variations, which often necessitates additional therapies [
4,
5]. These episodes requiring hospitalization are associated with increased morbidity, mortality, and put enormous burden upon healthcare systems [
6,
7]. Inflammation is a key component in the pathogenesis of COPD [
8]. It has previously been observed that COPD is not only associated with abnormal inflammatory response of the lung, but also with systemic inflammation, including systemic oxidative stress, activation of circulated immune cells and inflammatory cells, and the increased circulating levels of inflammatory cytokines [
9]. It is generally considered that ECOPD reflects a flare-up of these underlying inflammatory processes [
10], and is linked to a neutrophilic signature response [
11].
A recent systematic literature review concluded that ECOPDs are extremely dangerous events. There is an urgent need to identify tolerable treatment guidelines and manage acute exacerbations in hospitalized ECOPD patients [
12]. The Global Initiative for Chronic Obstructive Pulmonary Disease (GOLD) uses the ratio of forced expiratory volume in one second to forced vital capacity (denoted as FEV
1/FVC) as the diagnostic criteria for airflow obstruction, whose nominal value shall be smaller than 0.70; and classifies the airflow obstruction severity based on the value of forced expiratory volume in one second as percentage of predicted value (denoted as FEV
1%pred) as shown in Table
1. GOLD stages 1–4 respectively represent mild, moderate, severe and very severe. Inadequate diagnosis of COPD and the lack of spirometric assessment can lead to inadequate treatment strategies, with health costs and risks for patients, leading to delays in diagnosing and treatment of the true cause of the symptoms [
13‐
15]. Patients hospitalized with COPD exacerbations due to poor health status, are unable to complete the pulmonary function test reliably. As one longitudinal study indicated, 50% of pulmonary function test results are unacceptable [
16].
Table 1
Classification of airflow obstruction severity according to GOLD
GOLD 1 | Mild | FEV1%pred ≥ 80 |
GOLD 2 | Moderate | 50 ≤ FEV1%pred < 80 |
GOLD 3 | Severe | 30 ≤ FEV1%pred < 50 |
GOLD 4 | Very severe | FEV1%pred < 30 |
Many studies on evaluating the disease severity in COPD patients have been focused on the stable stage instead of exacerbating stage. 78% of the ECOPD patients have clear evidence of viral or bacterial infection [
17]. The percentage of neutrophils (NEU%) is commonly used as clinical bacterial infection indicators. Based on the needs of unified assessment criteria that can accurately reflect the pulmonary function of hospitalized ECOPD patients, we explored the predictive capability of the percentage of blood NEU% and demographic parameters in GOLD stage and created a prediction model based on support vector regression (SVR) for predicting GOLD stage in Hospitalized ECOPD Patients [
18].
Discussion
ECOPD is a kind of acute attack process, where the patients' respiratory symptoms continue to worsen over their daily status. The frequent episodes of ECOPD resulted in an accelerated decline in FEV
1. Meanwhile, the rapid decline of FEV
1 performs as an independent hazard factor for ECOPD. The vicious circle between the decline of FEV
1 and the frequent attack of ECOPD affect the prognosis and mortality of the patients [
23]. In this analysis, we focused on the discrimination value of blood NEU% as a biomarker for a severe episode of ECOPD, and the GOLD stage prediction in hospitalized ECOPD patients. We attempted to create an easy-to-use measure to estimate the value of FEV
1%Pred and to identify the GOLD stage that could assist clinicians in choosing appropriate measures of medical care to decrease future hospitalization rates and mortality in hospitalized ECOPD patients.
In line with previous studies, the outcome of pulmonary function test relied on the cooperation of ECOPD patients, most likely due to the limitation by force–velocity characteristics of expiratory muscles [
16,
24]. Biomarkers were required for effective risk stratification and making individualized treatment decision.
The pathophysiological mechanism of most cases of ECOPD is an acute burst of local or systemic inflammatory mediators following respiratory bacterial or virus infection. Usually, high levels of non-specific inflammatory biomarkers are expected [
25]. Neutrophils are the most abundant inflammatory cells in blood and sputum. As neutrophil proteases can generalize many of the characteristics of ECOPD including emphysema and mucus hypersecretion [
26], ECOPD is characterized as a neutrophil inflammatory disorder in most cases. A study on peripheral blood neutrophils from ECOPD patients conducted by Milara et. al. showed that compared with healthy control group, the release of the neutrophil activation marker neutrophil elastase (NE) and reactive oxygen species (ROS) increased by 2 times and 30% respectively [
27]. Jones et al. observed that compared with the healthy controls, bacteria stimulated neutrophil degranulation was greater in the ECOPD group [
28]. Corhay et al. focused on exacerbation whichever its trigger, and found that neutrophil inflammatory markers declined after treatment [
29]. We designated a statistically significant difference in the NEU% between ECOPD patients with different GOLD stages to extend these findings. ECOPD patients with higher blood NEU% had a higher tendency of severe episode of ECOPD, whose GOLD stage risk stratification could thus be higher. The differences between ECOPD patients with different GOLD stages are consistent with the results of Perera et al. They found that there were significant differences in systemic markers of inflammation between patients with GOLD stages 3 and 4 vs. controls without COPD; while there was no significant difference between GOLD 2 patients and controls [
30].
We sought for factors that would discriminate a severe episode of ECOPD in clinical cases. Although the multivariable demographic parameters or NEU% values reflected the relative risk of a severe episode of ECOPD, considering the moderate values of areas under the ROC curves, the overall prediction performance is still quite limited. No matter which cut-off level is chosen, the false positive rate is still very high, so the specificity for acceptable value of sensitivity is low. With increase in blood NEU%, the risk of a severe episode of ECOPD increased. The overall discrimination value of multivariable factors including demographic parameters and blood NEU% was encouraging with the area under the ROC curve of 0.84.
To further study the FEV1%Pred prediction and the GOLD stage categorization capability of the blood NEU% and demographic parameters, we randomly divided the data collected from the ECOPD patients into a training data set to develop a prediction model and a testing data set to validate the predictive performance. The selected demographic parameters included sex, age, weight and BMI, which had demonstrated their relevance to the target values. We used supervised learning algorithm to evaluate the predictive capability of the risk factors, and classified the subjects to 4 different GOLD stages. Searching for the right subjects was one of the major difficulties of our study.
On the other hand, support vector machine (SVM) is a learning method based on the principle of structural risk minimization of statistical learning theory. It shows many unique advantages in solving the problem of small sample and nonlinearity [
18]. SVR is a model dealing with the SVM regression problems, which showed acceptable regression capacity in estimating the value of FEV
1%Pred and identifying the GOLD stage.
To our knowledge, this is the first study in ECOPD patients to predict the value of FEV1%Pred and identify the GOLD stage based on demographic parameters and blood NEU%. In the absence of a clear biomarker to categorize the GOLD stage of ECOPD patients, our research provides an auxiliary guidance value for the clinicians to diagnose GOLD stage and establish appropriate clinical care, since the demographic parameters and blood NEU% are easy to be obtained.
Limitations of our current study should also be noted. First, the relatively small number of subjects enrolled in this study could limit the predictive performance of the model, especially when comparing to the previous work of Cristóbal et al. [
31] and Godtfredsen and coworkers [
32]. The predictive performance of the prediction model was limited in the ECOPD patients with optimistic degree of airflow obstruction, which could also be resulted from the lower influence of inflammatory factors when the symptoms were moderate. To find proper ECOPD patients and guide them to complete the pulmonary function test turned out to be one of the biggest difficulties during our research. To overcome this limitation, we used the most widely accepted learning method SVM to establish the prediction model. The grouping strategy of the training set and testing set was able to tackle the problem of multiple covariates larger than the samples (patients) or “p > n problem”. Importantly, the overall ECOPD GOLD stage prediction accuracy of the establish prediction model was 90.24%. Besides, Sørheim and coworkers showed that pulmonary function injury may differ between sexes. There was a sexual imbalance in our study, as the ECOPD patients included were mostly male (135/155). The model’s predictive performance on female patients could be limited. Considering the low population of the study, comorbidity and different treatments during hospitalization that are not reported herein, could influence the result of this work. Therefore, our future work is to balance the sex composition and extend the observation time to carry out larger scale research to verify our findings. As an additional limitation of the study, the patient's general condition, comprehension and cooperative degree could also influence the accuracy of pulmonary function test results. Nevertheless, every enrolled patient was trained and guided by the same professional physician to minimize the impact of external factors on the measurement.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.