Introduction

Community-acquired pneumonia (CAP) is a significant cause of morbidity and mortality [1, 2]. Severe CAP is conceptually pneumonia requiring admission to the intensive care unit (ICU) or carrying a high risk of death [2, 3].

The cornerstone of the treatment of pneumonia is antibiotic therapy, and ventilatory support in patients with severe respiratory failure [4]. Invasive mechanical ventilation presents multiple complications [5, 6]. For this reason non-invasive ventilation (NIV) has been used for acute or acute-on-chronic respiratory failure of diverse aetiologies [79]. The most important rationale for using NIV is to overcome an episode of severe acute respiratory failure (ARF) without the need for endotracheal intubation. However, few studies have assessed specifically the usefulness of NIV in patients with pneumonia [10], and it is even considered controversial due to a major variability in failure rates [1013], which are generally higher than those observed in chronic obstructive pulmonary disease (COPD) [14] or cardiogenic pulmonary oedema [15]. Moreover, predictive factors for NIV failure have been studied in different aetiologies [13, 14, 16, 17]. A recent study found that lack of improvement in arterial oxygenation was the main predictor of NIV failure in patients with CAP [18]. However, this study assessed a limited number of patients. Therefore, studies assessing factors predicting outcome for NIV in larger populations of CAP with severe ARF are still needed.

In spite of the controversy surrounding the use of NIV out of COPD exacerbations or cardiogenic pulmonary oedema, several observational reports illustrate that hypoxaemic ARF is among the most frequent indications for NIV in real practice [19, 20]. The ICU of our institution has been using NIV for the management of patients with severe ARF secondary to pneumonia who do not need emergency intubation for more than 15 years.

Since the available evidence on the efficacy of NIV in pneumonia is not compelling, appropriate patient selection should be a key point to improve the success of this technique. To determine the clinical results and evolution of patients with CAP and severe ARF treated with NIV at our institution, we assessed the characteristics and outcomes of these patients and determined factors that predict failure of the technique and mortality.

Methods

Patients

We prospectively followed all consecutive patients with severe ARF due to CAP receiving NIV treatment in an 18-bed ICU of Hospital Morales Meseguer, a general university hospital in Murcia, Spain, from January 1997 to December 2008. The study was approved by the Ethics Committee of the institution.

Pneumonia was defined as a new pulmonary infiltrate on the admission chest radiograph and symptoms and signs of lower respiratory tract infection. Exclusion criteria were any degree of immunosuppression and other criteria previously published [21]. All patients received timely empiric antimicrobial therapy according to the local protocols.

All patients received stress ulcer and deep venous thrombosis prophylaxis. Patients with COPD, other chronic respiratory diseases, or in the presence of wheezing received nebulised bronchodilators and systemic steroids. Patients with hypotension received initially intravenous fluid therapy with crystalloids; if haemodynamic instability persisted, norepinephrin was initiated. Patients who responded to vasoactive drugs remained on NIV unless they required intubation for other reasons.

Non-invasive ventilation protocol

The indication for NIV was done by the attending physicians. Patients were continuously monitored with electrocardiogram, pulse oximetry, invasive and/or non-invasive blood pressure, and respiratory rate. The criteria used for implementing NIV were: (1) moderate to severe dyspnoea accompanied by respiratory rate ≥30 breaths/min or signs of increased work of breathing such as active contraction of the accessory respiratory muscles, and (2) arterial oxygen tension to inspired oxygen fraction (PaO2/FiO2) ratio <250. Patients in need for emergency intubation at admission (respiratory or cardiac arrest, respiratory pauses with loss of alertness or gasping for air, major agitation inadequately controlled by sedation, signs of exhaustion, massive aspiration, inability to manage respiratory secretions appropriately, and haemodynamic instability without response to fluids and vasoactive agents) [13] were not treated with NIV.

Non-invasive ventilation was applied with specific NIV ventilators (BiPAP ST-D and VISION Ventilator; Respironics, Inc., Murrysville, PA, USA). As previously described [13, 21], patients were ventilated using the bi-level positive airway pressure mode (BiPAP). Face mask was used as first choice, but nasal mask was optionally used if patients did not tolerate face mask. The initial inspiratory positive airway pressure (IPAP) was set at 12 cmH2O, and levels were raised by 2–3 cmH2O as tolerated but did not exceed 25 cmH2O. The initial expiratory positive airway pressure (EPAP) was set at 5 cmH2O, and the levels were raised by 1–2 cmH2O if needed to improve hypoxaemia. The FiO2 was set to achieve SpO2 >92% or PaO2 >65 mmHg. Arterial blood gas samples were obtained from each patient before the connection to the ventilator and after 1 h of NIV. Subsequently, samples were obtained as clinically indicated.

Effectiveness of the technique

Non-invasive ventilation was considered successful if a patient avoided endotracheal intubation, was discharged from the ICU and remained alive and conscious for at least 24 h after being transferred to the ward. Failure of NIV therapy occurred when a patient experienced worsening of gas exchange or respiratory distress leading to intubation or death. If NIV failed, patients who had not declined intubation underwent such procedure according to the criteria stated above [13, 21].

Data collection

At the start of NIV therapy demographic, clinical and laboratory variables were collected. The variables specifically related to pneumonia were the presence of unilateral or bilateral involvement, involvement of one or more pulmonary lobes, and pleural effusion on chest radiography, and the presence of aetiologic pathogens in diagnostic samples. Worsening pulmonary infiltrate was considered when a chest radiography done in the first 24 h of ventilation therapy showed an increase in pulmonary involvement for at least one lobe compared with the previous radiography. Patients’ degree of severity and organ failure were estimated with the Simplified Acute Physiology Score (SAPS)-II [22], the Acute Physiology and Chronic Health Evaluation (APACHE)-II [23] and Confusion, Elevated Blood Urea Nitrogen, Respiratory Rate and Blood Pressure plus Age ≥65 years (CURB65) score [24] calculated at admission, and the Sepsis-Related Organ Failure Assessment (SOFA) [25] score calculated daily. Septic shock [26] at admission was also noted.

Acute respiratory failure was considered “de novo” in patients without previous cardiac or respiratory disease [19].

Statistical analysis

Qualitative or categorical variables, expressed as number and percentages, were compared with the χ 2 or Fisher’s exact tests. Continuous variables were expressed as mean ± standard deviation (SD), and compared between groups using the Student t test for independent data, and with Pearson’s correlation between two continuous variables. All the analyses were two-tailed, and p values <0.05 were considered significant.

The variables identified as predictors of NIV failure and hospital mortality on univariate analysis were included in a multivariate logistic regression with a conditional stepwise forward model (p in < 0.10, p out < 0.05) to correct for co-linearity. Adjusted odds ratios and 95% confidence intervals were computed for variables independently associated with NIV failure or hospital mortality. The predictive capacity for NIV failure or hospital mortality of quantitative variables were assessed with receiver-operating characteristic (ROC) curves; the area under the curve (AUC), optimal cut-off values, sensitivity, specificity and positive and negative likelihood ratio [27] were calculated. Data were processed with SPSS 16.0.

Results

Patients

During the study period 250 patients were admitted to the ICU with primary diagnosis of CAP; 50 (20%) received oxygen therapy, 16 (6%) were intubated at ICU admission, and 184 (74%) received NIV. Patients received NIV for 44 ± 33 h (mean ± SD) along 2.8 ± 1.9 days. During NIV treatment the maximal inspiratory and expiratory positive airway pressure were 19 ± 3 and 9 ± 1 cmH2O, respectively.

Non-invasive ventilation was successful in 116 patients (63%). Among patients with NIV failure, 59 were intubated (32%), and the remaining 9 patients died without intubation after a do-not-intubate order. The main reason for intubation was worsening of respiratory insufficiency in 39 patients, uncontrolled shock in 17, and intolerance to NIV in 3 cases.

Patients with “de novo” ARF and previous cardiac or respiratory disease (Table 1)

Acute respiratory failure was considered “de novo” in 102 (55%) patients, whereas the remaining had history of cardiac or respiratory disease. Patients with previous cardiac or respiratory disease failed NIV less frequently than patients with “de novo” ARF. They were older and more frequently males. At admission they had higher PaCO2 and bicarbonate, lower arterial pH and respiratory rate, and more frequently decreased consciousness.

Table 1 Clinical and ventilatory characteristics of patients with “de novo” acute respiratory failure and with previous cardiac or respiratory disease

Complications related with non-invasive ventilation

Sixty-nine (38%) patients had at least one complication related with NIV. The most common complications were skin lesions, from erythema to skin ulceration (57, 31%), eye irritation (24, 13%), claustrophobia (15, 8%), and gastric distension (6, 3%). The duration of NIV was longer in those who presented skin lesions (69 ± 38 versus 32 ± 23 h, respectively, p < 0.001), without differences between the two groups. Serious complications were three cases of total intolerance, one of nosocomial pneumonia during NIV, one pneumothorax, one patient with vomiting and bronchial aspiration and two cases with mucus plugs detected on intubation.

Prediction of non-invasive ventilation failure (Table 2)

In both groups, patients with successful NIV treatment were less severe, assessed by less need for vasoactive drugs at onset of NIV, severity scores and acidosis, had less extensive radiologic findings, less acute respiratory distress syndrome (ARDS) criteria, and lower respiratory and heart rate and better oxygenation after 1 h of NIV. In patients with “de novo” ARF, successful NIV was also associated with lower respiratory rate and better oxygenation at onset of NIV. In patients with previous cardiac or respiratory disease, successful NIV was also associated with lower heart rate at onset of NIV and higher arterial pH and better level of consciousness after 1 h of NIV. Age, sex and PaCO2 were not related with NIV outcome in both groups.

Table 2 Variables associated with successful and unsuccessful treatment with non-invasive ventilation in patients with “de novo” ARF and previous cardiac or respiratory disease

The ICU stay was shorter, and mortality lower in patients with successful treatment with NIV.

Since the variables significantly associated with NIV outcome were essentially the same for patients with “de novo” ARF and previous cardiac or respiratory disease (Table 2), the multivariate analysis of predictors of NIV outcome was performed in the overall population. The variables independently associated with NIV failure were worsening of radiological infiltrate 24 h after admission, maximum SOFA score during NIV, and higher heart rate, lower PaO2/FiO2 ratio and lower bicarbonate after 1 h of NIV (Table 3). Except for heart rate, these variables were highly accurate in predicting NIV failure, as shown in this table.

Table 3 Multivariate analysis of variables independently associated with non-invasive ventilation failure in the overall population

Analyses of survival (Table 4)

Hospital survivors from both groups were younger, less severe, assessed by severity scores and acidosis, had failed NIV less frequently, and had lower respiratory rate after 1 h of NIV. In patients with “de novo” ARF, survival was also associated with less worsening of pulmonary infiltrate 24 h after the onset of NIV and better oxygenation. In patients with previous cardiac or respiratory disease, survival was also associated with bilateral radiologic involvement and ARDS criteria, and need for vasoactive drugs at onset of NIV. As for NIV outcome, sex and PaCO2 were not related with survival in both groups.

Table 4 Variables associated with hospital mortality in patients with “de novo” ARF and previous cardiac or respiratory disease

In the overall population, the multivariate analysis identified maximum SOFA score during the ICU stay, NIV failure and older age as independent predictors of hospital mortality (Table 5). Maximum SOFA score and NIV failure were the most accurate variables in predicting NIV failure, as shown in this table.

Table 5 Multivariate analysis of variables independently associated with hospital mortality in the overall population

Delayed intubation and hospital mortality

Among intubated patients with “de novo” ARF, the duration of NIV in hospital survivors was shorter than that of non-survivors (32 ± 24 versus 78 ± 65 h, p = 0.014, Fig. 1 left panel), with an AUC of 0.70 for predicting hospital mortality. The duration of NIV before intubation that best predicted mortality was ≥53 h, with sensitivity of 69% and specificity of 83%. The SAPS-II, CURB65 and SOFA scores at admission did not correlate with the duration of NIV in these patients, indicating that patients with longer time on NIV before intubation were not more severely ill at admission. Moreover, the need for vasoactive drugs at onset of NIV was associated with shorter duration of NIV before intubation (31 ± 40 versus 65 ± 52 h, respectively, p = 0.030), indicating that patients treated with vasoactive drugs who need intubation fail earlier than those without these drugs. After adjusting for these variables, increased duration of NIV before intubation in hours was significantly associated with decreased hospital survival (adjusted odds ratio 0.978, 95% confidence interval 0.962–0.995, p = 0.012).

Fig. 1
figure 1

Duration of non-invasive ventilation in patients who needed intubation and survived or died in the hospital. Left panel patients with “de novo” acute respiratory failure. Right panel patients with previous cardiac or respiratory disease. Horizontal bars represent mean values of survivors and non-survivors for intubated patients from each group

By contrast, no relationship was found between duration of NIV before intubation and mortality in patients with previous cardiac or respiratory disease (Fig. 1, right panel).

Discussion

Patients with CAP and previous cardiac or respiratory disease responded better to NIV than those with “de novo” ARF, as expected from previous studies [10, 19]. Successful NIV treatment strongly predicted survival. We have also found a consistent association between delayed intubation and decreased survival in patients with “de novo” ARF.

The only randomised clinical trial that assessed the efficacy of NIV in patients with CAP and severe ARF showed that the benefits of NIV in terms of less need for intubation and better survival were restricted to patients with COPD and hypercapnic respiratory failure only [10]. More recently, NIV was effective in decreasing the rates of intubation and mortality in severely hypoxaemic patients with pneumonia, compared with high-concentration oxygen therapy, with faster improvement of hypoxaemia and tachypnoea [13]. Although those results were promising, routine use of NIV in patients with CAP and without COPD is not clearly recommended by recent guidelines [1, 28].

Patients with CAP are among those with the highest rate of NIV failure, either in COPD patients [29] or those with hypoxaemic ARF [11, 12, 17]. The rate of NIV failure in patients with pneumonia is widely variable, being lower in controlled clinical trials (21–26%) [10, 13, 30] than observational studies (33–66%) [11, 12, 17, 18, 31], possibly due to selection criteria of patients. We found similar rates of NIV failure in our population.

The risk factors for NIV failure have been studied in mixed patients with hypoxaemic ARF [17, 32, 33]. In specific populations with severe CAP treated with helmet NIV, a 56% failure rate was reported [18]. Higher severity of patients, worse oxygenation, lower pH, and less improvement of blood gases after ventilatory support were related with NIV failure on univariate analysis. This study, however, was restricted to patients with “de novo” ARF, and did not consider variables other than general severity scores and ventilatory and respiratory parameters.

Studies on NIV often include pneumonia as a cause of “de novo” or hypoxaemic ARF only. However, the great proportion of previous cardiac or respiratory disease in our patients is not surprising, since large published series of hospitalised patients with CAP report high rates of chronic respiratory or cardiac co-morbidities [34, 35]. Hence, the outcome of NIV in patients with CAP from studies that have excluded COPD or hypercapnic patients [11, 12, 17, 18] should not be extrapolated to general CAP populations treated with NIV.

Similarly to previous reports [1719, 21, 33], severity scores at admission and worse respiratory and cardiovascular function were associated with NIV failure. Variables related with initial severity and worse clinical evolution in pneumonia, such as the need for vasoactive drugs, organ system dysfunction [28, 36], and extension and worsening of radiological infiltrates, were also associated with failure. These variables had not been assessed in previous reports.

Higher organ system failure during NIV treatment, including metabolic acidosis, and variables related with worse initial response to the empiric treatment and NIV support were independently associated with NIV failure on multivariate analysis. These results indicate that the optimal management of these patients seems not restricted to appropriate support with NIV but should also include general support measures and effective empiric treatment.

Concerns have been raised due to the high mortality rate of patients who fail NIV treatment, particularly in “de novo” ARF, and the possibility that unnecessary delay of intubation results in excess mortality [19, 32]. Particularly, an actual mortality of patients intubated after NIV failure higher than mortality predicted by severity scores has been reported [32]. However, these comparisons may be misleading, since severity scores often underestimate hospital mortality in ICU patients [37, 38]. To our knowledge, this is the first time a consistent association between delayed intubation and increased mortality in patients with CAP and “de novo” ARF has been shown. Longer duration of NIV before intubation was not related with severity of patients at admission. Moreover, patients with vasoactive drugs who needed intubation failed NIV earlier than those without. Therefore, this excess of mortality should be attributed to delayed intubation rather than a more severely ill selected population.

Several variables independently associated with NIV failure, such as higher SOFA score, worsening of radiologic infiltrate and low oxygenation after 1 h of NIV, were highly predictive of this event (Table 3). In patients with “de novo” ARF, we strongly advise careful management in order to avoid potential harm related to excessive delay in intubation when the response to NIV is not positive, particularly when they require vasoactive drugs, a condition particularly associated with NIV failure in patients with acute lung injury [32], as well as in those patients who do not achieve those PaO2/FiO2 cut-offs after 1 h of NIV treatment, as previously described [17, 33]. By contrast, this recommendation cannot apply for patients with previous cardiac or respiratory disease.

In the overall population, older age, NIV failure and higher organ system failure were independently associated with hospital mortality. Thus, further improvement of survival could be potentially achieved by combining adequate support against organ system dysfunction and optimal NIV management of these patients.

The present study has several limitations. First, the effectiveness of any treatment must be determined by a randomised controlled trial. In the absence of a control group, definitive conclusions cannot be drawn. Second, the study was conducted in a centre with major experience in the use of NIV, and therefore these results cannot be extrapolated to less trained and equipped hospitals.

In conclusion, successful treatment with NIV, which is strongly related with less organ system failure of patients and a good initial response to the antimicrobial treatment and NIV support, was strongly associated with better survival. In presence of predictors for NIV failure, avoiding delayed intubation of patients with “de novo” ARF is strongly advised in order to potentially minimise mortality.