Emergency admission parameters for predicting in-hospital mortality in patients with acute exacerbations of chronic obstructive pulmonary disease with hypercapnic respiratory failure

Chronic obstructive pulmonary disease (COPD) is a common chronic respiratory disease. It has a chronic and progressive course that is often punctuated by "exacerbations", which are clinically defined as exacerbations of respiratory symptoms. Specifically, they include dyspnoea, cough and sputum production, and increased sputum purulence [1]. Acute exacerbation of COPD (AECOPD) is a common presentation at emergency departments (ED) that is associated with a mortality rate of 1.8–8% [2‐4]. The occurrence of hypercapnia in patients with COPD might be indicative of the presence of hypoventilation syndrome, which is associated with a negative prognosis [5, 6]. However, there are significant among-patient differences in the severity, evolution, and outcome of exacerbations. Specifically, some patients make a full recovery within a short time period, while others may die [7]. There is a need for an effective tool for helping emergency care providers promptly determine the mortality risk of patients with AECOPD to inform treatment decisions.

There have been numerous studies on the prognostic risk factors for AECOPD [6‐13], including age, body mass index (BMI), lung function tests, blood gas analysis, inflammation-related indicators, and smoking history. However, there have been inconsistent reports regarding the predictors of poor outcomes in patients with AECOPD due to the heterogeneity across the previous studies with respect to the assessed variables, study population, and outcomes of interest [7]. Moreover, emergency patients, especially those in critical care, lack immediately available data regarding lung function, BMI, and the 6-min walk test. Additionally, individual risk predictors cannot allow systematic risk assessment. Scoring systems for assessing the criticality of patients with AECOPD include the medical research council (MRC) scale [11] and Acute Physiology and Chronic Health Evaluation-II score (APACHE-II) [14]. The MRC dyspnoea scale is a questionnaire for assessing long-term mortality that is comprised of five statements regarding perceived breathlessness. The APACHE-II assesses numerous parameters collected within 24 post-admission hours. However, it is mainly suitable for the intensive care unit (ICU) rather than ED. CURB65, CRB65, BAP65, and NEWS, which mainly involve mental status, respiratory rate, oxygen saturations, pulse, blood pressure, age, BUN level, etc., can be used to predict AECOPD-associated mortality in ED given the simple structure and data availability [2‐4, 15]. However, these scoring systems have been rarely studied; moreover, they have a poor or modest value for predicting mortality (AUC 0.606–0.744) [2‐4, 15‐17]. Additionally, it is difficult for these systems to yield an individual survival assessment.

This study aimed to extensively collect predictors readily available from patients with AECOPD presented at the ED, and subsequently develop and validate a nomogram for individual mortality risk prediction in patients with AECOPD with hypercapnic respiratory failure.

Figure 1 presents the study flowchart. Initially, 1062 patients were recruited; among them, 601 patients remained after applying the exclusion criteria. Table 1 and Additional file 1: Table S1 show the patient characteristics. Among the 601 patients, 213 (35.44%) received non-invasive mechanical ventilation, 137 (22.80%) received invasive mechanical ventilation, 6 (1.00%) died in ED, 172 (28.62%) were admitted to the ICU, and 19 (3.16%) died during hospitalization. Univariate analysis revealed significant differences in temperature, respiratory rate, level of consciousness, PH, PCO₂, BE, lactic acid, lactic dehydrogenase, potassium, magnesium, albumin, albumin/globulin ratio, blood amylase, BUN, C-reactive protein, white blood cell count, neutrophils count, red blood cell count, haematocrit, haemoglobin, platelet distribution width (PDW), international normalized ratio, prothrombin time, and D-dimer levels between the death and survival. Furthermore, we collected the results of lung function tests. However, among the 601 patients, only eight lung function tests were performed within 6 months before admission. Therefore, these data were not included in the statistical analysis.

Among 80 variables, 7 non-zero coefficients were screened through LASSO logistic regression analysis using the minimum criteria, including respiratory rate, lactic acid, PCO₂, BUN, haemoglobin, platelet count, and PDW. We constructed a nomogram incorporating these seven predictors and the the variable of concomitant pneumonia (Fig. 2). An AUC of 0.940 (95% CI 0.895–0.985) indicated good discrimination by the nomogram for predicting the mortality risk in the patients (Fig. 3a). The AUC for the internal validation model was 0.933 (95% CI 0.870–0.975) (Fig. 3b).

Figure 4 presents the DCA for the nomogram. DCA showed that when the model for assessing the in-hospital mortality risk ranged from 0.04 to 0.55, the nomogram added more net benefits than the “treat all” or “treat none” strategies.

The AUCs for in-hospital mortality of BAP65, CRB65, CURB65, NEWS, and the nomogram were 0.732 (95% CI 0.693–0.768), 0.734 (95% CI 0.695–0.770), 0.787 (95% CI 0.751–0.820), 0.771 (95% CI 0.734–0.850) and 0.940 (95% CI 0.895–0.985) respectively (Fig. 5). For ICU admission and invasive mechanical ventilation, the AUCs of the nomogram were 0.7822 (95% CI 0.7408–0.8237) and 0.8044 (95% CI 0.7615–0.8473), respectively.

We developed and validated a nomogram for the individualized mortality prediction in patients with AECOPD with hypercapnic respiratory failure. This nomogram incorporated eight predictors: respiratory rate, PCO₂, lactic acid, BUN, haemoglobin, PDW, platelet count, and concomitant pneumonia. All eight predictors could be readily accessed at the ED. The model allowed good discrimination with an AUC of 0.940 (95% CI 0.895–0.985) for in-hospital mortality. Additionally, the nomogram showed good performance for invasive mechanical ventilation in the ED and ICU admission. Moreover, it showed potential clinical utility. Compared with other prognostic models, including CRB65, CURB65, BAP65, and NEWS, the nomogram showed better performance in predicting the mortality risk of patients with AECOPD. In the subgroups of patients with different concomitant conditions, the model also showed a good predictive performance. To our knowledge, there has been no study that has developed a prediction model for assessing individual mortality risk in patients with AECOPD with hypercapnic respiratory failure.

Numerous factors are associated with AECOPD prognosis. However, there are often strong between-factor correlations, especially in laboratory tests, including PO₂, oxygenation index, PDW, mean platelet volume, PCO₂, PH, etc. Inconsistent results are yielded by incomplete collection of factors or use of different statistical methods. Based on other studies, we extensively collected factors readily available in ED that may be associated with patient outcomes. These included vital signs, state of consciousness, laboratory tests, and major comorbidities. Specifically, we considered factors that have recently attracted more attention, including D-dimer [20], NT-proBNP, troponin [21], red blood cell distribution width [22], platelet count [8], and mean platelet volume [10]. To construct the predictive model, 12 potential predictors were identified from 80 candidate variables by examining the predictor-outcome association through shrinking the regression coefficients using the LASSO method. In practice, it is often easier to implement models with fewer predictor variables. Most clinicians are unwilling to use large models requiring an extensive collection of information [23]. Therefore, we used the stepwise regression to reduce the number of variables to seven, which made the model more practical. The superior AUC suggested that this did not reduce the predictive performance of the model.

In addition to the number of predictors and the prediction accuracy, the variable availability should be considered in model establishment. Particularly in the ED, data access is limited by equipment, time, and patient conditions, including lung function, 6-min walk test, etc. There was no routine data collection on smoking, long-term oxygen treatment, and BMI, which are used for predicting patient 90-day or long-term survival, as well as re-exacerbation [11, 24‐26]. In the ED, it is crucial to determine how to use the available data, quickly and effectively distinguish the criticality of the patients, and make the right treatment decision. Therefore, we mainly included variables readily available in ED; further, real-time data collection reflected the patient's current condition. Our nomogram involved 8 variables, which are all required for routine evaluation in the ED. Therefore, this model has a strong practical value. Furthermore, we made full use of the data to develop the model and used bootstrap for internal validation to prevent the loss of estimation precision in models developed on a data subset [24]. A nomogram can be used to predict individualized survival probability in patients with AECOPD [25], which provides a powerful tool for patients and health care professionals in making decisions.

The respiratory rate is an effective indicator for determining the lung condition and dyspnoea. An increased respiratory rate often indicates lung inflammation or gas exchange inefficiency; moreover, it predicts increased mortality, which is consistent with the study by Akhter [26, 27].

PCO₂ has received increasing attention as an important prognostic factor for patients with AECOPD. Although there have been inconsistent reports [28‐30], clinicians often consider hypercapnia as an indicator of more severe respiratory disease, worse prognosis, and higher complication risk from oxygen therapy including respiratory depression [7, 11, 13, 20, 28, 31‐33]. Additionally, in clinical practice, patients with carbon dioxide retention have a higher mortality risk. Therefore, we enrolled patients with AECOPD with hypercapnia respiratory failure. In patients with oxygen-dependent COPD, PaCO₂ (air) is an independent prognostic factor with a U-shaped association with mortality; specifically, the lowest mortality is observed at approximately 6.5 kPa with the mortality increasing at < 5.0 kPa and > 7.0 kPa [6]. In patients with AECOPD who had PCO₂ ≥ 50 mmHg, there was a positive correlation of mortality risk with PCO₂ levels, which is consistent with the study by Ahmadi [6].

Infection is among the main AECOPD causes. Among ED patients with suspected infection, there is an association of intermediate lactate elevation with moderate-to-high mortality risk [34]. Further, blood levels of lactic acid are affected by the effectiveness of non-invasive ventilation in the treatment of AECOPD with different severities [35]. In our study, 213 (35.44%) patients underwent non-invasive mechanical ventilation therapy. Therefore, the lactate level is another important prognostic factor. There is significant clinical utility in the dynamic monitoring of lactic acid for detecting changes in a patient’s condition [35].

BUN is another important marker of poor outcome in patients with AECOPD. Acute renal failure and elevated BUN levels at hospital admission were associated with higher hospital mortality [12].

The ability to transport oxygen is directly dependent on haemoglobin levels. Inadequate haemoglobin levels could aggravate tissue hypoxia and negatively affect the prognosis [36, 37]. Anaemia is a strong independent dyspnoea predictor in patients with COPD [36]. Haemoglobin levels are negatively associated with mortality in patients with AECOPD.

Platelet and platelet activity (including mean platelet volume, PDW, etc.) are associated with haemostasis and coagulation function [38‐41]. Furthermore, there are associated with tumour development and inflammation [41‐44]. Platelet count has a U-shaped association with increased risk of 3-year all-cause mortality in patients with COPD [8]. For analysis, we divided patients into those with platelet levels > 300 × 10⁹/L and < 100 × 10⁹/L. We found that patients with normal platelet levels had the lowest mortality risk while it was highest among patients with platelet levels < 100 × 10⁹/L. There have been inconsistent reports regarding the relationship between PDW and death. Şahin [45] reported no significant difference in PDW between patients with stable and acute exacerbation of COPD. Another study reported a protective role of PDW; further, inhaled corticosteroids/long-acting β2-agonists therapy increased PDW values in patients with COPD [46]. We found a negative correlation of PDW with mortality.

Respiratory viruses, airway bacteria are common causes of COPD exacerbations [47], patients with AECOPD often complicated with pneumonia. In the cohort we studied, 45.6% patients had pneumonia. Based on this clinical consideration, the variable of concomitant pneumonia was incorporated into the construction of the nomogram. The result shows that the nomogram can be applied to both AECOPD patients with pneumonia and without pneumonia.

This study has several limitations. First, this was a single-centre retrospective study; therefore, the results may not be generalizable to other settings and populations. Second, given the limitations in the ED, we did not collect other factors associated with AECOPD mortality, including lung function, BMI, smoking, and long-term oxygen treatment. Therefore, we could not determine the correlation between these factors and death. However, we built a model with good performance using parameters easily available in the ED. Moreover, the addition of other variables would not significantly improve the prediction effect of the model. Third, we only studied patients with hypercapnia respiratory failure. Further studies are required to determine whether the results apply to non-hypercapnia patients.

In conclusion, this study presents a nomogram that incorporates respiratory rate, PCO₂, lactic acid, BUN, haemoglobin, platelet distribution width, and platelet count.

It can be conveniently used by ED staff to grade patients with AECOPD faster, and accurately, and therefore manage patients more effectively in the complex environment of the ED, and enhancing the quality of patient care.