Results

Sixteen published articles of randomised controlled trials,6-10 16-26 one article in preparation11 and two abstracts,26 of which only one had been published, were identified. Two studies22 24 were excluded because study patients consisted mainly of term neonates with pulmonary diseases other than RDS. In two studies,17(Ramanathan R, Ruiz I, Tantivit P, Cayabyab R, deLemos R. High frequency oscillatory ventilation compared to conventional mechanical ventilation in preterm infants with respiratory distress syndrome. Abstract, San Diego, CA: Society for Pediatric Research;1995) HFV was used for a limited time (48 hours and 96 hours, respectively), after which according to the study protocol patients were switched back to conventional ventilation. Three studies19-21 used HFV as a rescue treatment rather than as an elective mode of ventilation. One abstract was excluded because insufficient data were available on the methods of the trial at the time of data analysis.27 Nine published studies and one study in press met inclusion criteria6-11 16 18 23 25; the characteristics of these studies and the results of the methodological assessment are shown in tables 1 and 2.

Nine of the 10 studies mentioned mortality at the age of 28 to 30 days. Both in the overall analysis as well as in the subgroup analyses no significant difference was found (table 3). For the overall analysis the relative risk was 1.02 with a 95% CI between 0.78 and 1.35. Nine studies explored the effect on bronchopulmonary dysplasia at the age of 28 to 30 days. There was no significant difference in BPD in the overall analysis [RR 0.91, (95% CI: 0.80, 1.03)], nor in the subgroup analyses (table 4). For the HLVS subgroup the RR was 0.84, with a 95% CI between 0.69 and 1.02.

Chronic lung disease at a postconceptional age of 36 weeks was described in five, more recent studies. There is a significant reduction in CLD both in the overall analysis [RR 0.50, (95% CI: 0.32, 0.78)] and in the subgroup analyses (table 5). The reduction is most evident in the subgroup of high frequency oscillators [RR 0.35, (95% CI: 0.16, 0.77)], and in the HLVS subgroup [RR 0.44, (95% CI: 0.27, 0.73)].

The incidence of IVH was mentioned in all but one trial, and the incidence of severe (grades 3 and 4) IVH was mentioned in eight trials. There was no difference in risk in the overall analysis [RR 1.09, (95% CI: 0.96, 1.23)] or in the subgroup analyses. The risk of severe IVH was significantly higher in the HFV group, however, both when using the fixed effects model [RR 1.31, (95% CI: 1.04, 1.66)], and the random effects model [RR 1.34, (95% CI: 1.05, 1.70)] (table 6). Study outcomes all differed, however. The increased risk of severe IVH was only found in the HIFI trial [RR 1.41, (95% CI: 1.06, 1.88)], which dominates this overall analysis here. The other trials did not find a higher risk. In the subgroup analyses the difference for severe IVH was not significant, although in the subgroup of high frequency oscillators, once again dominated by the HIFI trial, a trend towards an increased risk occurred [RR 1.30, (95% CI: 1.00, 1.69)].

Of the 10 studies, seven report on the incidence of periventricular leucomalacia. In the overall analysis using the fixed effects model the RR is 1.39 (95% CI: 0.96, 2.02), and with the random effects model the RR is 1.28 with a wider non-significant 95% confidence interval (95% CI: 0.51, 3.19) (table 7). Again, this outcome varies among the studies. A higher risk for cystic PVL was evident in the study by Wiswell [RR 5.00, (95% CI: 1.19, 21.04)]. In the HIFI trial a similar trend was present, with an RR of 1.61 (95% CI: 0.99, 2.60). The other trials did not observe a difference.

All 10 studies report on the incidence of air leaks. There is no significant difference in risk for gross air leak and/or interstitial pulmonary emphysema in the overall analysis [RR 1.06, (95% CI: 0.91, 1.23)], nor in the subgroup analyses. In the HIFI study an RR of 1.20 (95% CI: 1.00, 1.43) for air leaks is observed in the HFV group, due to a higher incidence of pneumoperitoneum (3% vs 1% in the conventional ventilation group).

Due to the fact that values on the duration of mechanical ventilation, oxygen dependency, and hospital stay were expressed differently in the studies—mean plus standard deviation or median with range—these data could not be pooled. Only in the study by Gerstmann was a significant reduction in duration of oxygen dependency for infants weighing more than 1 kg evident (median of 13.2 days vs 27.6 days for the HFV and CV group, respectively). In the same study the number of infants weighing 1 kg or less still on a ventilator on day 28 of life was significantly lower in the HFV group. In the other studies no significant differences were found in duration of oxygen dependency, mechanical ventilation, or hospital stay.

Only two studies report on the long term pulmonary outcome. In the HIFI trial a total 432 of 524 surviving study patients (82%) were followed up and pulmonary function tests performed in 223 of them (43%).27 There were no significant differences in incidence of episodes of wheezing, respiratory tract infections, or hospital readmissions. Impairment of pulmonary function was similar in both groups. Ogawa performed a chest x-ray of all his study infants at the age of 1 year. He found typical evidence of bronchopulmonary dysplasia in 2% of the HFV treated patients vs 4% in the conventionally ventilated patients, which was not significantly different.8

The same two studies comment on the long term neurodevelopmental outcome. In the HIFI trial 386 of the 504 survivors (77%) were neurologically tested between the corrected age of 16 and 24 months.28 The proportion of children with a normal neurodevelopmental status (Bayley score of more than 83 and no major neurological defect) is significantly lower in the HFV group (54%vs 65% in the conventional ventilation group). This is associated with a significantly higher incidence of hydrocephalus in the HFV group (12% vs 6% in the conventional ventilation group). No difference is observed in risk for, or severity of, cerebral palsy. Ogawa reports an identical incidence of developmental delay at the age of 1 year of 9% in both study groups.8

Discussion

We used meta-analysis to address the question of efficacy and safety of the elective use of HFV in the treatment of RDS in preterm infants. All trials were rigorously assessed for their methodological quality. Several comments can be made. First, treatment allocation may not have been truly concealed, as this was done by the care provider in the centre where the patient was treated. In open trials true concealed allocation is only possible with central allocation by telephone and by a person not taking care of the patient, so that the patient is registered as a study patient before being allocated to a treatment group. Second, in these types of intervention studies blinding of the care providers is not possible, so all the included trials are sensitive for performance bias. Moreover, only two trials provided explicit guidelines for general patient management in the methods,6 8 reducing cointervention bias. In one study it was even stated that no attempt was made to control non-respiratory aspects of care.10 Blinding of outcome assessment can be achieved to a certain extent if the persons interpreting chestx-ray pictures or cerebral ultrasound scans are unaware of the treatment assignment of the patient. This was evidently not the case in all the studies and thus could lead to outcome detection bias in some studies.9 10 15 16 24

The results of the overall analysis are strongly influenced by the HIFI trial, which dominates the analysis due to its large study population. This trial has been criticised because the high lung volume strategy was not used.29 Investigators also probably lacked experience of the ventilators at that time. Animal studies confirm the advantages of a high lung volume strategy to minimise lung injury,2 30 and more recent, smaller clinical trials using this strategy show better pulmonary outcomes, without any increase in intracranial morbidity.9-11

It was suggested in the HIFI trial that the increased risk of severe IVH could be explained by obstructed venous return because of the continuous positive intrathoracic pressure. Yet this could not be confirmed by animal or human studies.31-33 Wiswell’s HFJV trial showed a marked increase in periventricular leucomalacia in the HFV group. Mean PaCO₂ concentrations were significantly lower in the HFV group, however, compared with the conventional ventilation group. The negative effects of hypocarbia on cerebral blood flow are described by Volpe and have been confirmed by Fujimoto and Wiswell.33-36 The results of the individual trials indicate that the therapeutic margin of this ventilation technique is very narrow, and that harm can be done if it is handled without sufficient experience or extensive monitoring.

By pooling all the data we increased the power of the evidence; heterogeneity between included studies, however, can impair the qualitative aspect of the analysis. Therefore, we tested for heterogeneity and performed subgroup analyses. Meta-analyses were performed by using both the Mantel-Haenszel approach14 and the DerSimonian and Laird technique.15 The former refers to the fixed-effects model and assumes that there is only one true effect of the intervention estimated by each study. The latter refers to the random-effects model and assumes that the studies are randomly taken from a population of studies and that there is a distribution of treatment effects, instead of a single true effect. When studies are homogeneous, the results of both fixed effects and random effects model will be similar; if there is significant heterogeneity between studies, however, the DerSimonian and Laird approach (random effects model) will be more conservative, resulting in larger confidence intervals around the estimate of the pooled RR, thus being more likely to find no significant difference between treatments.

For some of the outcomes we did find a strong heterogeneity in separate study results. As Marlow mentioned in his annotation, there are important differences among trials concerning the intervention and the study population, which could explain the variations in the outcomes.37 Different ventilators, as well as different ventilatory strategies were used for both types of ventilation. Four HFV frequencies going from 4 Hz18 up to 15 Hz6 8 16 were used, and in the study by Wiswell and Keszler a low frequency background conventional ventilation was used. Conventional ventilation frequencies as low as 25 breaths per minute6 7 and as high as 60 breaths per minute9 10 were used. Mean birthweight and gestational age of the study infants varies between 900 g at 26 weeks and 1500 g at 31 weeks.9 11 The children’s’ postnatal age at which HFV was started also varies between 1.2 hours and 15.5 hours.18 25 Marlow rightly stated that, supported by animal data,2 the very early use of HFV with a high lung volume strategy could be a very important factor in preventing lung injury and thus reducing chronic lung disease.37

We conclude from the meta-analysis that HFV does not improve survival of premature infants with RDS, but it does reduce the risk of chronic lung disease at 36 weeks of postconceptional age. It may be associated, however, with an increased risk of severe intraventricular bleeding. Yet, from the subgroup analysis there is evidence that, if a high lung volume strategy is used, the reduction of chronic lung disease may be more pronounced, and this favourable effect is achieved without increasing the risk of severe intraventricular haemorrhage.

Several areas of uncertainty remain. First, the hypothesis that a high lung volume strategy is associated with better outcome still requires confirmation with controlled trials of sufficient size, investigating both short term as well as long term pulmonary and neurological outcomes. Cohorts should be followed up long enough and at comparable postconceptional ages, and should be assessed with reproducible and standardised tests. Second, the outcome in very preterm infants (less than 30 weeks gestational age) and very low birthweight infants (birthweight less than 1000 g), who are at highest risk of developing CLD and are most susceptible to neurological complications, is still to be investigated. Although most studies stratify their study patients according to birthweight, the results are usually not presented according to weight strata. Finally, as high frequency flow interrupters become widely used, randomised controlled trials investigating efficacy and safety of this technique of HFV are needed.

We suggest that new clinical trials be done evaluating the usefulness of the elective use of HFV, started as soon as possible after birth, applied with a high lung volume strategy, in very premature infants (less than 30 weeks gestational age and/or weighing less than 1 kg) and using the high-frequency flow interruptors. An optimal study design is important, such as centralised concealed treatment allocation, avoidance of cointervention bias, and blinding of outcome assessment. A consensus should be made on relevant outcomes so that results of these trials are of optimal quality and comparable with each other.