Non invasive ventilation (NIV) is a relatively new ventilatory mode that has shown positive effects in adult patients with different types of respiratory failure [1]. Its most important effects are inspiratory muscle work reduction and gas exchange improvement [2].

Several studies in adult populations have shown that NIV is a safe and effective therapy for patients with hypercapnic acute respiratory failure (ARF) due to chronic obstructive pulmonary disease (COPD) exacerbation [3, 4] and hypoxemic ARF due to cardiogenic pulmonary oedema [5], community-acquired pneumonia [6], and immunocompromised patients with pulmonary infiltrates [7]. In hypoxemic ARF due to acute respiratory distress syndrome (ARDS)/acute lung injury (ALI) and in postextubation ARF, the role of NIV remains unclear and it should be applied in carefully selected patients [812].

NIV diminishes the rate of intubation when compared to conventional medical therapy [13]. NIV diminishes the risk of ventilator-associated pneumonia and other nosocomial infections [14]. Moreover, NIV may improve survival when compared to conventional mechanical ventilation (CMV) [15].

NIV is also a well established therapy in neonates [16]. It has shown to reduce the need of CMV when applied after surfactant treatment [17], to be effective for early stabilization in very low birthweight infants [18], and in primary treatment of newborns with respiratory distress syndrome [19].

In children, most knowledge is derived from retrospective studies [2025], little case series or case reports [26, 27] and preliminary results [28]. Most of reported results do not include a large number of patients [29, 30] or only focus on one disease [31, 32]. As a consequence, evidence is lacking in order to define indications, methods, conditions of use and complications of NIV in children.

Since respiratory pathologies seen in paediatric critical care population widely differ from pathologies seen in the adult or neonatal intensive care unit (ICU), and pathophysiology of ARF in children is different [33], extrapolation from adult or neonatal to paediatric population cannot be done.

It is very important to identify patients at risk of NIV failure. Some studies have been performed in adults to find outcome predictors [8, 9]. Predictive factors for NIV failure have been reported in paediatric patients, but they were based on retrospective data [20, 21] or they only analyzed bronchiolitis [31, 32]. There is only one prospective study which identifies some NIV outcome predictors in critically ill children [34]. Therefore, prospective information on paediatric NIV characteristics, and risk factors for NIV failure in children could be useful to improve therapeutic strategies of respiratory support. The primary objective of this study was to identify predictive factors of NIV failure in paediatric intensive care unit (PICU) patients. The secondary objective was to determine the characteristics of NIV in critically ill children.


Setting and patients

A prospective observational study set in an eight-bed PICU of a university hospital was performed. All patients admitted to our unit from August 2004 to July 2008 and deemed to be candidates to receive continuous positive airway pressure (CPAP) or pressure support ventilation (PSV) were included. CPAP was considered a type of NIV [35, 36]. Criteria to initiate NIV were: ARF or acute-on-chronic respiratory failure without improvement despite medical treatment, severe dyspnea at rest (modified Wood’s Clinical Asthma Score ≥5 in asthma [37] or bronchiolitis [38]), a respiratory rate above 2 standard deviations (SD) for child’s age normal range, or a partial pressure of arterial oxygen to the fraction of inspired oxygen (PaO2:FiO2) ratio under 250 and above 150, or venous PCO2 >55 mmHg or arterial PCO2 >50 mmHg (modified from Antonelli et al [8]). Contraindications to NIV support were cardiorespiratory arrest, hemodynamic instability despite fluid load and vasoactive treatment, Glasgow coma scale score under 10, facial deformity, facial trauma or surgery, vocal cords palsy, bullous pneumopathy, pneumothorax, endotracheal intubation to manage secretions or airway protection, and upper gastrointestinal tract active bleeding.

ARF was defined as failure to sustain a threshold level of alveolar exchange to meet the metabolic demands of cellular respiration and classified in type 1 and 2 using pathophysiological and clinical criteria modified from Teague [33]. ARF with ventilation–perfusion impair, hypoxemia and parenchymal condensations on X-ray was considered as type 1. ARF with hypoventilation, hypercapnia without hypoxemia, and parenchymal condensations absence on X-ray (excluding atelectasis) was considered as type 2. Bronchiolitis was the only ARF cause classified in either type 1 or 2 group.

NIV technique

CPAP or pressure support ventilation (PSV) was delivered using a nasal mask, face mask, full-face mask or nasal prongs. The interface was chosen according to child’s age and size achieving comfortability and avoiding air leaks. Initially it was applied manually onto the patient’s face, and then it was held by a paediatric head cap. Colloid dressings were placed on the major pressure points to minimize skin injury. A heated humidifier (Fisher and Paykel Healthcare, Auckland, New Zealand) was used in all cases.

Ventilation strategy

Based on the protocol published by “Respiratory Group of the Spanish Society of Paediatric Intensive Care” [39], CPAP initial ventilator setting was 4–5 cmH2O; PSV was started at 4–6 cmH2O and expiratory positive airway pressure (EPAP) at 4 cmH2O. PSV was increased if the attending physician considered that inspiratory volume was low according to auscultation and thoracic motion, and if hypercarbia increased or did not decrease. CPAP or EPAP were increased if no improvement of pulse-oximetry O2 saturation or arterial PO2 was achieved. NIV was stopped and patients were intubated when oxygen saturation was below 85% or venous PCO2 above 65 mmHg despite maximal NIV setting, or when any of the exclusion criteria appeared.


Midazolam intravenous boluses (0.1 mg/kg) followed by continuous perfusion (0.05–0.1 mg/kg/h) when needed were used if the child was stressed with consequent patient–ventilator asynchrony.


A nasogastric tube was placed in all patients to avoid gastric distension or vomiting, and afterwards it was used to provide feeding when possible.


All patients were continuously monitored by means of electrocardiography, pulse oximeter and respiratory rate. Blood gas analysis was only performed when considered necessary by the attending physician.

NIV outcome

NIV was deemed successful when conventional mechanical ventilation (CMV) was not necessary. If CMV was needed, the episode was considered as a failure.

Data collection

Patients with multiple admissions were considered individually, since each episode requiring NIV presents new variables potentially affecting outcome. For each episode, the following variables were collected: age, sex, weight, PICU and hospital stay, ARF type, ARF cause, underlying disease, Paediatric Risk of Mortality (PRISM) score, type of interface, NIV duration, NIV outcome, CMV duration, use of sedatives, NIV complications, mortality and causes of death. Clinical data collected were respiratory rate (RR), heart rate (HR) and FiO2 before NIV was started. The same data and CPAP, EPAP and PSV were collected at 1, 6, 12, 24 and 48 h. Complementary explorations collected were blood gas analysis and X-ray.

Statistical analysis

Descriptive analysis

Mean, median, standard deviation and range were used to describe the sample. We compared success and failure groups as well as type 1 and type 2 ARF groups. Quantitative continuous variables were compared between groups using non parametric tests (Mann Whitney’s U test), and qualitative variables using Chi-square test (χ 2).

Predictive factors for NIV failure

A logistic regression analysis was performed in order to identify possible predictors of NIV failure. Multiple logistic regression analysis was performed before NIV start and after 1, 6, 12, 24 and 48 h. Variables included in the multivariate analysis were those which had a p value under 0.2 in univariate analysis between success and failure groups, and also those variables which were considered clinically important in order to control statistical confusion. Variables included in the model were age, weight, mean HR difference to initial HR, mean RR difference to initial RR, FiO2, ARF type and PRISM score. Before NIV was started, RR and HR were also included, instead of HR or RR difference. At 24 and 48 h, EPAP was also included. Venous PCO2 was excluded due to the lack of data. A p value < 0.05 was considered statistically significant.

This research project was approved by the Research Ethics Committee of the Hospital Universitario Central de Asturias. Written informed consent was obtained from patients’ parents or guardians.


Descriptive study

During the study period, there were 1,111 consecutive admissions to the PICU. There were 214 episodes of CMV and 171 of NIV with 42 patients receiving both types of respiratory support. Thirty-seven cases were excluded because of postextubation and obstructive sleep apnoea NIV indications. Also 18 episodes were excluded because an Infant Flow system was used. Then, 116 episodes were analyzed (Fig. 1). Type 1 ARF was indication for NIV in 38 episodes (32.8%), while type 2 ARF was indication in 78 (67.2%). Baseline characteristics of included cases are shown in Table 1.

Fig. 1
figure 1

Flow chart of patients and episodes included in the study and its classification in type 1 or type 2 acute respiratory failure groups. ARF acute respiratory failure

Table 1 Base-line characteristics of the non-invasive ventilation episodes and ARF causes

Interfaces and ventilators

Face masks were used in 89 episodes and nasal masks in 27 cases. Face mask was chosen in 71.1% of ARF type 1 cases and in 79.5 % of ARF type 2. No statistical relation was found between type of interface and age, weight or NIV indication.

We employed in all episodes specific ventilators for NIV: in one case a Vivo (Breas, Mölnlycke, Sweden), in two cases BiPAP® Harmony® (Respironics, Pittsburgh, PA) and in the rest of them BiPAP® Vision® (Respironics, Pittsburgh, PA). In only three cases CPAP was used.

NIV duration

NIV median duration was 41.0 h (range: 0.5–375); 45.0 h (range 11–375) in successful cases, and 13.0 h (range: 0.5–77) in failure episodes.

Pressures applied

Mean maximum EPAP was significantly higher in type 1 ARF group (8.2 ± 1.7 vs. 6.7 ± 1.2 cmH2O; p < 0.001). No significant differences were observed in PSV between ARF groups.

Evolution of clinical parameters

A progressive decrease of FiO2, HR and RR was seen during NIV therapy. Evolution of RR is shown in Fig. 2. HR decreased from 158.1 ± 30.0 before NIV to 127.0 ± 27.3 at 6 h in type 1 ARF, and from 165.0 ± 28.5 to 138.5 ± 21.0 in type 2 ARF. FiO2 decreased from 0.51 ± 0.30 before NIV to 0.41 ± 0.14 at 6 h in type 1 ARF, and from 0.39 ± 0.19 to 0.37 ± 0.14 in type 2 ARF.

Fig. 2
figure 2

Evolution of respiratory rate at the hours studied in type 1 and 2 acute respiratory failure groups. ARF acute respiratory failure


Midazolam boluses were administered in 12.2% of the cases and continuous perfusion of midazolam was needed in 52.2%. Median age of patients who received midazolam boluses or perfusion was 8.5 months (range 0.6–163.8), significantly lower than children’s who were not given any sedative (34.2 months, range 0.8–169.7; p = 0.001). There were no secondary effects due to sedatives and no failure could be attributed to this medication.

NIV complications

Complications were detected in 23 cases (20.2%). A non-deep skin lesion, without producing skin necrosis, was found in 18 cases, three developed a pneumothorax, one had an upper airway bleeding, and another one had a gastric distension. Skin injuries were directly related with NIV duration (92.5 h, range 16–248, vs. 39.0 h, range 0.5–375; p < 0.001). None of the children with a skin lesion needed analgesics and no failures could be attributed to this cause.


Ninety-eight episodes succeeded (84.5%). Success was more frequent in type 2 than in type 1 group [92.3% (CI 95%: 98.2–86.4) vs. 68.4% (CI 95%: 83.1–53.7); p = 0.001]. Baseline characteristics of successful and failed episodes are shown in Table 2. NIV failure before 6 h was significantly more frequent in ARF type 2 group: three out of six episodes, while only two out of 12 cases in ARF type 1 group (p = 0.024). Causes of NIV failure were hemodynamic instability, apneas and hypoxemia (five episodes for each cause), hypercapnia in two episodes, and pneumothorax in one.

Table 2 Base-line characteristics of non-invasive ventilation episodes in success and failure groups

PICU stay

Median PICU length of stay was 7 days (range 2–157). In failure episodes it was 14.5 days (range 3–134), while it was 7 days (range 2–157) in successful cases (p = 0.003).


Five patients died. None of the deaths were related to the use of NIV. Nevertheless NIV had failed in four out of five of these patients. In the case in which NIV had not failed, death was due to a massive intestinal necrosis in a patient with leukemia (47 days after NIV). In three episodes, death cause was a septic shock, and in the last one it was a malignant arrhythmia. None of the last four deaths happened in the first 45 h after intubation (median 48.0 h, range 45–312).

Predictive factors for NIV failure

Failure group showed lower age, weight and RR decrease from initial RR at hours 1 and 24. In other hand, they had higher PRISM score, HR and RR at 24 h, EPAP at 24 and 48 h, FiO2 at 1, 6, 24 and 48 h, and PCO2 at 6 and 24 h when compared to success group (Table 3).

Table 3 Parameters with significant differences between success and failure groups expressed in mean ± standard deviation (except for age and weight, expressed in median and range)

Multiple regression analysis performed including age, weight, HR, RR, FiO2, ARF type and PRISM score (and EPAP at hours 6, 12, 24 and 48), identified type 1 ARF and higher PRISM score as independently associated variables with NIV failure. Moreover, it identified lower RR decrease (at 1 h and at 6 h) also independently associated with NIV failure (Table 4).

Table 4 Multivariate analysis before non-invasive ventilation initiation, and after 1, 6, 12, 24 and 48 h


To our knowledge, this is the largest prospective study of NIV in paediatric ARF which tries to identify NIV outcome predictors in paediatric population. As it occurs in the vast majority of paediatric studies, the lack of a control group obligates us to evaluate carefully the conclusions obtained. However, this work is based on daily clinical practice even though a NIV protocol has been used.

Our study included a typical, heterogeneous critically ill paediatric population similar to that of the majority of European multidisciplinary PICUs. All patients considered to be candidates to receive NIV according to our inclusion criteria entered the study, without any exception. The posterior exclusion of obstructive sleep apnoea patients was done because it is not an ARF, and its treatment with nocturnal CPAP is well established. We also excluded postextubation cases, as their characteristics are proved to be very different to those of children who had not received CMV. Postextubation ARF in adults has been studied separately from other NIV indications [11, 12].

Previous studies have shown success rates for the use of NIV in paediatric patients to be between 57 and 92% [2022, 2931, 34]. Our results are comparable to those reporting a high success rate (over 84%). The improved results of Fortenberry et al [22] and Padman et al [30] in comparison to ours could be related to a higher patients’ age. Our sample had a low median age: 10.3(0.6–169.7), compared to other series which included older children [2022, 29, 30, 34]. Median age in Campion et al [32] and Larrar et al [31] studies was lower than ours because they only included infants with bronchiolitis.

Neuromuscular disease was the most frequent underlying disease. Because of their lack of strength these children usually achieve a good patient–ventilator synchrony, with a good outcome [29]. Immunodeficiency was also a usual underlying disease, and our good outcome (five successful NIV episodes out of seven) goes in accordance with literature [7].

Status asthmaticus was the most frequent diagnosis, and it is noteworthy that only one of these patients required CMV, in accordance with the case series reported by Akingbola [27] and Carroll [24]. Pneumonia was also quite frequent, with a 22.6% failure rate, higher than 13% reported by Essouri et al [20]. NIV failed in only five out of 33 bronchiolitis, a better outcome than the one reported by Campion and Larrar [34, 35].

Interfaces mostly used were the face masks, as little infants do not cooperate and air leaks through the mouth could occur. This would result in ventilator–child asynchrony and could be a cause of NIV failure. Due to the lack of cooperation, sedative use was quite frequent (64% of the episodes, similar to 56–76% reported by Essouri and Campion) [20, 32].

NIV was a safe therapy, and the most frequent damage (skin-lesion) usually has a good outcome. The five deaths during our study were not related to NIV use.

The main objective of our study was to identify predictive factors of NIV failure in PICU patients. Three independent risk factors of NIV failure were type 1 ARF, PRISM score and lower RR decrease during NIV initial period. Type 1 ARF patients are potentially more likely to suffer a NIV failure. This finding agrees with most of previous studies, in which higher NIV failure risk was found in hypoxemic cases (similar to type 1 ARF). ARDS diagnosis was independently associated with failure in Essouri’s study [20] and a primary pulmonary parenchymal disease was a risk factor for NIV failure in Joshi’s report [21]. Therefore, we consider extremely important to classify all ARF patients in one of these groups before NIV therapy is started. PRISM score has already been identified as a predictive factor for NIV outcome [31, 32]. Pediatric Logistic Organ Dysfunction, another prognostic severity score was reported in Essouri’s study [20]. However, Bernet et al [34] did not find differences in Paediatric Index of Mortality score between responders and non-responders. The other risk factor was related to RR emphasizing the importance of close RR monitoring when performing NIV. Once NIV is started, lower RR decrease at 1–6 h was associated with NIV failure. A similar finding was made by Essouri, who found a significantly higher RR decrease at 2 h in the success group [20]. In summary, patients with type 1 ARF and high PRISM score are candidates to CMV, as well as children who do not reduce RR in the first 6 h of NIV therapy.

In univariate analysis, we found that smaller children are in higher risk of failure. An age under or equal to 6-year-old was associated with NIV failure in Joshi’s study [21]. Our result could be related to the difficulties in achieving a good ventilator–child synchrony due to gas leakage, the lack of adequate mask sizes [25] and not sensitive enough triggers. A higher EPAP was found at 24 and 48 h in failure group. This result has not been reported in any paediatric study, but Antonelli found that a higher level of EPAP was independently associated with NIV failure in adult patients with ARDS [9].

This study has several limitations. First, an observational study does not allow us to make any conclusion about gasometric determinations as predictive factors of NIV outcome. In the protocol we followed, blood gas sampling was not mandatory. We tried to use clinical data, performing blood gas analysis only in case of doubts about patient clinical evolution. If child’s comfortability with good patient–ventilator synchrony, adequate thorax motion and auscultation, and decreasing heart and respiratory rate are achieved, there is no point on blood gas sampling. We think that in paediatric patients on NIV, blood sampling could produce patient/ventilator asynchrony. In the opposite side, when clinical parameters did not improve, we performed blood gas determinations. Second, the study was conducted in only one PICU. This limits the generalization of the results, although our PICU is likely comparable to most European multidisciplinary PICUs.

This study has several strengths. First, it is a prospective study, following the protocol published by “Spanish Respiratory Group of the Spanish Society of Paediatric Intensive Care” [39] conducted in all consecutive patients admitted in PICU. Second, it is the only paediatric prospective cohort done in a 4-year period, which allowed the findings to be representative and accounted for possible seasonal variation. Third, it is the biggest study focusing on predictive factors of NIV success in critically ill children. Finally, our study provides recent descriptive NIV data, not available in the current PICU literature.


Paediatric NIV is a new respiratory support technique that in our experience is useful in more than 84% of the cases. Type 1 ARF group classification, PRISM score and lower RR decrease were independent risk factors for NIV failure which suggest that focus should be given to the correct classification of patients, PRISM score, and close RR monitoring.