Weaning (or discontinuation) from mechanical ventilation is definitive cessation of mechanical ventilation and differs from extubation, which is removal of the endotracheal tube 12. Determining the optimal time at which to discontinue mechanical ventilation must not be based simply on clinical impression because weaning depends on multiple factors 23: central drive and peripheral nerves; mechanical respiratory loads, ventilatory muscle properties and gas exchange properties; and cardiac tolerance and peripheral oxygen demands. Premature weaning places the patient at risk for reintubation and airway trauma, whereas delayed weaning exposes them to risk for nosocomial infection and increases hospitalization costs. Indeed, 'the complexity of the decision to extubate provides a strong rationale for developing accurate predictors of extubation outcome' 3.

Extensive efforts have been made to identify predictors of successful weaning in adults 45 as well as in children 678910111213141516171819202122232425. The following indices were proposed by the Collective Task Force of the American College of Chest Physicians (TF) as the most promising weaning predictors 5: spontaneous respiratory rate, paediatric rapid shallow breathing (RSB) 13, RSB occlusion pressure (ROP) 26 and maximal inspiratory pressure during an occlusion test (Pi_max) 2327. Although none of these predictors appears to be sufficiently sensitive or specific in predicting weaning success, paediatric studies have used integrated indices, including respiratory drive, respiratory load, muscle strength and quality of gas exchange.

Fatigue (for example, diminution of endurance) of the inspiratory muscles is defined as reduction in capacity to develop force and/or velocity of a muscle, which results from muscle activity under load and is reversible with rest 2282930. The two reference techniques for assessing endurance (for example, detecting fatigue) are analysis of the change in the electromyographic power spectrum and in the force response of muscles to electrical stimulation 2. Another approach uses the tension-time index (TTI) of the diaphragm, calculated from the mean transdiaphragmatic pressure, which correlates well with reference techniques. However, because determination of transdiaphragmatic pressure is invasive, noninvasive TTI estimated from the measurement of mouth occlusion pressure (P_0.1) and from the integral of the airway pressure curve over time during spontaneous ventilation were used as predictors of weaning success 2. Finally, Jabour and coworkers evaluated ventilatory endurance using a modified TTI, namely the pressure-time index (PTI), which was calculated from peak airway pressure during mechanical ventilation 28.

Respiratory muscle endurance, which has been reported to have particular significance in predicting weaning success in adults 282930, has never been investigated in children. Thus, the aims of the present prospective study were as follows: to evaluate the ability of indices proposed by the TF 5 and endurance indices to predict weaning outcome; and to determine whether a combined index, including the most accurate TF and endurance indices, could enhance the ability to predict weaning success.

This prospective study was approved by the local hospital institutional review board, and parents provided informed consent before the study began. All children admitted to the paediatric intensive care unit at our university-affiliated hospital from March 1999 to July 2001, and who required mechanical ventilation for more than 6 hours were eligible for inclusion in the study. Children with chronic neuromuscular disease, who had undergone tracheostomy, or who were aged under 30 days were excluded. The primary physician was responsible for the weaning decision and process, and was blinded to the results of the measurements performed during the short period of spontaneous breathing (see below). All ventilatory variables were collected by the same investigator, and the weaning procedures were not delayed in his absence (for example nights and weekends); those patients who did undergo weaning procedures in his absence were excluded from the study.

Patients were enrolled in the study if they met all the following criteria, as defined for adults by the French Society of Critical Care Consensus Conference 31: improvement or resolution of the underlying cause of acute respiratory failure; core temperature <38.5°C; satisfactory renal function; no signs of infection; neuropsychological state compatible with autonomous breathing; correction of electrolyte disorders; a haemoglobin level above 9 g/dl; absence of left ventricular dysfunction or cardiac arrhythmia; no sedation or mild sedation (for example, midazolam <0.5 μg/kg per minute, morphine sulphate <1 mg/kg per day, or fentanyl citrate <0.25 μg/kg per hour); efficient cough; and adequate gas exchange. Moreover, all of the patients were weaned using pressure support ventilation (PSV) on Servo 300 or Servo 900C ventilators (Siemens-Elema, Solna, Sweden), with a pressure support below 15 cmH₂O, a positive end-expiratory pressure below 5 cmH₂O and fractional inspired oxygen (FiO₂) below 40% 31. After the patient's primary physician made the decision to wean, the patient's respiratory variables were recorded by the investigator, first during mechanical ventilation and then during a short period of spontaneous breathing. Then, the children underwent a spontaneous breathing trial (SBT) through a T-piece circuit or Canopy device for 30 minutes 32. Then, children were extubated, depending on their clinical status and blood gas determination 31. Patients who failed SBT were excluded. Weaning failure was defined as reinstitution of invasive or noninvasive mechanical ventilation within 48 hours of extubation, for a reason than upper airway obstruction.

Of the 220 patients who were eligible for inclusion in the study, 56 patients were enrolled (25%) and underwent 59 episodes of weaning (median age 29 [11–94] months, median weight 13 91011121314151617181920212223 kg). The median duration of ventilation was 4 (2–9) days, and 39 patients received mild sedation during the weaning procedure. Two patients who did not tolerate SBT were excluded. Of the remaining 54 children, 17 patients had neurologic disorders (30%), 28 had respiratory failure (49%), three had both neurologic disorders and respiratory failure (5%) and nine underwent surgery (16%). Of the remaining 57 episodes of weaning (in the 54 children), 45 attempts were successful (80%) and 12 were unsuccessful (20%; Table 2). The causes of reintubation were respiratory failure (n = 6; one child with spinal amyotrophy, one with myopathy and one with mucopolysaccharidosis), pulmonary oedema of cardiac origin (n = 3), bronchial obstruction (n = 2; 1 burned child) and atelectasis (n = 1). Three of the 12 patients who failed weaning needed noninvasive ventilation within 48 hours after extubation (20%) and nine required reintubation and mechanical ventilation for a reason other than upper airway obstruction (80%). Two patients who were successfully extubated required reintubation for severe upper airway obstruction (4%). Age, sex, weight, PRISM score, duration of ventilation and internal tube diameter were not statistically different between the weaning success and weaning failure groups (Table 3). Also, diagnoses were not statistically different between the groups. Finally, ventilator settings and spontaneous breathing parameters at the beginning of the weaning procedure were not significantly different between the two groups (Table 4).

The best single TF index was ROP, the best combination of TF indices was represented by the expression (0.66 × ROP) + (0.34 × Pi_max), and the best endurance index was the TTI₂. However, their performance was statistically poor (Table 5). The best model resulting from the combination of these indices was defined by the expression: (0.6 × ROP) – (0.1 × Pi_max) + (0.5 × TTI₂). This integrated index fitted the data well (P = 0.364, Hosmer–Lemeshow test) and was a good weaning predictor (P < 0.01, Kruskal-Wallis test). The threshold value that best discriminated between two groups was 1.3, with a value <1.3 predicting weaning success (P = 0.007, Fisher test). Although LR⁺ was 6.4 and the LR⁺/LR^- ratio was greater than 10, LR^- remained above 0.2, meaning that this integrated index was only predictive of weaning success.

In a population of 54 critically ill children, we found that the indices proposed by the TF were insufficient in predicting weaning outcome. Among the combined indices, including endurance in order to enhance prediction of weaning outcome, the best combination was defined by the expression: (0.6 × ROP) – (0.1 × Pi_max) + (0.5 × TTI₂). However, this index was not a valuable predictor of weaning outcome because it only predicted weaning success.

The best combination of TF indices included ROP and Pi_max. Although Pi_max was considered by the TF 5 to be one of the most promising weaning predictors in adults (based on LR⁺ and LR^-), no data concerning LRs were available for ROP in the adult or paediatric literature 26. In the present study ROP alone was only predictive of weaning success, and this finding is in agreement with that obtained in a population of 45 adult patients by Sassoon and Mahutte 26, who found that ROP may be useful and more accurate in predicting weaning outcome when ROP components (P_0.1 and RSB) alone were indeterminate 26. The AUC (0.80 ± 0.9) and LRs (LR⁺ 2.42, LR^- 0.05, and LR⁺/LR^- 48), recalculated from their published data, appear to be better than those in our study. Pi_max, which is considered a global index of respiratory muscle strength 35385152 and was adapted to children by El-Khatib and coworkers 23, was evaluated in five paediatric studies 6891012, but LRs were calculated in only one study designed to predict weaning failure 12. The LRs were recalculated to predict weaning success, and the results (LR⁺ 1.09, LR^-0.35, and LR⁺/LR^-3.11) appeared similar to those obtained in the present study (Tables 5 and 6) and in five adult studies 5. AUC values obtained in the five paediatric studies were similar to those in our study and ranged from 0.53 (95% confidence interval 0.47–0.67) 8 to 0.57 (95% confidence interval 0.47–0.67) 12. However, in a recent adult study that included 52 patients 50, Pi_max was a good weaning predictor (LR⁺ 1.40, LR^- 0.12, and LR⁺/LR^- 11.67).

In our study, TTI₂ was found to be the best endurance index, although it was not very predictive of weaning success (Table 5). Numerous studies indicate that duration of inspiration relative to the duration of the total respiratory cycle (for example, T_i/T_tot ratio) should be an important determinant of diaphragmatic fatigue 35535455. Hence, any increase of indices that include T_i/T_tot ratio such as PTI 2836, TTI₁ 3738 and TTI₂ 39 could reflect decreased endurance. In adults, Ramonaxto and coworkers 56 reported that TTI₂ highly correlated with TTI of the diaphragm, which was proposed by Bellemare and Grassino 5455 for quantifying the magnitude and duration of respiratory muscle contraction. However, TTI₂ has never been used as a predictor of weaning outcome either in adults or children, whereas TTI₁ and PTI appeared to be useful predictors of weaning outcome in adults 2856.

Several adult and paediatric studies 79131415 have attempted to define integrated weaning indices that include parameters in relation to respiratory muscle function, such as respiratory drive, respiratory load and muscle strength 355357, but none of these integrated indices was found to be predictive of weaning outcome. Respiratory muscle endurance has not been investigated in paediatric patients; however, the diaphragm is less endurant in children than in adults because of anatomic and physiologic differences 58. Furthermore, in adults respiratory muscle endurance was reported to be of particular significance in predicting weaning outcome 282930. Taking respiratory muscle endurance into account should improve prediction of weaning success. In the present study TTI₂ was found to contribute strongly to the integrated index, and its association with ROP and Pi_max improved LR⁺, LR^- and LR⁺/LR^- ratio.

Comparing the performance of weaning indices is difficult because few studies have calculated LRs, which are considered as the best tests 4. Indeed, only one paediatric study calculated LRs 12, and most other paediatric studies did not report the number of true-positive and false-positive findings. Furthermore, large discrepancies exist in the literature regarding population characteristics, the definition of weaning and extubation 59, inclusion and exclusion criteria (for example aetiology of respiratory failure and duration of ventilation), the definition of weaning success and weaning failure, and the use (or not) of noninvasive mechanical ventilation after extubation 1623.

Three factors may explain why integrated indices do not improve upon the accuracy of prediction of weaning outcome of single indices. First, most paediatric studies included small numbers of patients with a relative low rate of weaning failure, and it has been demonstrated that a large sample including more than 1,000 patients with at least 100 extubation (or weaning) failures would be necessary to assess adequately the ability of parameters to predict extubation (or weaning) success in children 60. Second, weaning indices are often measured too late, when patients meet all of the clinical criteria for weaning, and so threshold values are not very discriminatory 5. Third, SBT is considered to be the most direct and simplest way to assess patient performance without ventilator support 2, and so a failed SBT is synonymous of weaning failure. Thus, the number of false-negative findings (for example, patients who failed SBT but who could have been weaned) is unknown, and the specificity, NPV and LR^- of SBT cannot be determined 2. Among 105 extubations in adults considered to be false negative because of a pre-extubation respiratory rate above 30 breaths/minute, DeHaven and coworkers 17 observed that 97 were successful. Furthermore, Epstein and coworkers 3 noted that 30–70% of unplanned extubation in adult patients did not result in reintubation.

Weaning is usually delayed by clinicians 3, and a weaning index should be more predictive of weaning failure (for example, it should be more specific, and have a good NPV and a lower LR^-). Thus far, all paediatric studies of weaning indices, which determined their sensitivity, specificity, LRs, or AUCs, included a SBT. As long as a SBT is defined as a necessary pre-condition for extubation by the TF 2, comparisons between SBT and weaning indices will be limited because of the underestimation of false-negative findings and NPV.

Our study has two other limitations. First, there was no defined protocol for decreasing mechanical ventilation, although there is no evidence that such a protocol must be used in children 461. Moreover, we employed no set protocol for decreasing sedation, which is supposed to improve weaning success in children 59. Second, during the study period 220 patients were eligible according to the inclusion criteria, but only 27% of these were included. Like the study conducted by Thiagarajan and coworkers 14, in which 227 children from among 472 admitted to their paediatric intensive care unit (48%) were included, our study design required the presence onsite of the investigator in charge of data collection in order to decrease variability in measurement of the different parameters and to guarantee that the patient's physician remained blinded to the results.

The weaning failure rate was 20%, which is similar to that reported by others in paediatric patients, ranging from 1.4% 62 to 34% 12. The median PRISM value, calculated within 24 hours of admission, was close to that reported in most previous paediatric studies 11121663, although PRISM values were found to be lower in two studies 6264.

In our population of 54 critically ill children, indices proposed by the TF were insufficient in predicting weaning outcome. The best combined index, incorporating endurance, was defined by the expression: (0.6 × ROP) – (0.1 × Pi_max) + (0.5 × TTI₂). This is the first report of the value of endurance parameters in predicting weaning success in children. The index had a LR⁺ of 6.4 and a LR⁺/LR^- ratio above 10; however, at a threshold value <1.3 it was only predictive of weaning success (LR^- >0.2). Although this index cannot be considered sufficient for making decisions regarding weaning, the reason for this is probably that it was measured, like in other studies, when children met all clinical criteria for weaning. At the present time, clinical judgement associated with SBT still seems superior in the weaning process in children.

• Determining the optimal time to discontinue mechanical ventilation is important, because premature weaning places the patient at risk for reintubation and delayed weaning exposes the patient to risk for nosocomial infection.

• The indices proposed by the TF are not sufficiently sensitive or specific in predicting weaning outcome.

• Until now, indices that reflect respiratory muscle endurance have not been investigated in children.

• Our best index, combining ROP, Pi_max and TTI₂, was only predictive of weaning success.

• Such indices are unable to assist with decisions regarding weaning probably because they are measured when patients meet all clinical criteria for weaning.

AUC = area under the curve; FiO₂ = fractional inspired oxygen; LR = likelihood ratio; NPV = negative predictive value; Pi_max = maximal inspiratory pressure during an occlusion test; PRISM = Paediatric Risk of Mortality Score; PSV = pressure support ventilation; PTI = pressure-time index; ROP = RSB occlusion pressure; RR = respiratory rate; RSB = rapid shallow breathing; SBT = spontaneous breathing trial; TF = Task Force of the American College of Chest Physicians; T_i = inspiratory time; TTI = tension-time index; T_tot = total respiratory cycle time.

The authors declare that they have no competing interests.

BG performed the statistical analysis and interpretation of the data. ON, FL, BG, YR conceived the study, participated in the design and execution of the study, the analysis of data and writing of the manuscript. BG performed the statistical analysis and interpretation of the data. AS performed respiratory investigations. ON, AD, CF, RC, SL performed clinical data collection. All authors read and approved the final manuscript.