United Kingdom Oscillation Study: long-term outcomes of a randomised trial of two modes of neonatal ventilation

Article history

The research reported in this issue of the journal was funded by the HTA programme as project number 08/116/10. The contractual start date was in March 2011. The draft report began editorial review in May 2013 and was accepted for publication in January 2014. The authors have been wholly responsible for all data collection, analysis and interpretation, and for writing up their work. The HTA editors and publisher have tried to ensure the accuracy of the authors’ report and would like to thank the reviewers for their constructive comments on the draft document. However, they do not accept liability for damages or losses arising from material published in this report.

Declared competing interests of authors

Anne Greenough has held grants from various ventilator manufacturers and received honoraria for giving lectures and advising various ventilator manufacturers.

Copyright statement

© Queen’s Printer and Controller of HMSO 2014. This work was produced by Greenough et al. under the terms of a commissioning contract issued by the Secretary of State for Health. This issue may be freely reproduced for the purposes of private research and study and extracts (or indeed, the full report) may be included in professional journals provided that suitable acknowledgement is made and the reproduction is not associated with any form of advertising. Applications for commercial reproduction should be addressed to: NIHR Journals Library, National Institute for Health Research, Evaluation, Trials and Studies Coordinating Centre, Alpha House, University of Southampton Science Park, Southampton SO16 7NS, UK.

United Kingdom Oscillation Study

The UKOS was a multicentre, randomised trial undertaken to determine whether or not use of HFO or conventional ventilation (CV) from within 1 hour of birth would reduce mortality and the incidence of BPD. (The earlier results of UKOS discussed below were published separately elsewhere.)1,10,11

Infants were eligible for the study if their gestational age was between 23 weeks and 28 weeks plus 6 days, they were born in a participating centre and they required endotracheal intubation from birth and ongoing intensive care. Infants were excluded if they had to be transferred to another hospital for intensive care shortly after birth or had a major congenital malformation.

A total of 25 centres participated in the study – 22 in the UK and one each in Australia, Ireland and Singapore. To ensure that each centre had adequate experience with high-frequency oscillatory ventilation (HFOV), we required participating centres to have used this type of ventilatory support in a minimum of 20infants before the study began. The quality of collected data was monitored and the statistical analyses were performed at the co-ordinating centre (St. George’s Hospital, London, UK). Both the South Thames Multicentre Research Ethics Committee, London, UK, and the Local Research Ethics Committee at each participating centre approved the protocol.

Women at high risk of delivering an infant before 29 weeks of gestation were invited, before delivery, to participate in the trial, and oral or written assent was obtained. Randomisation occurred either when delivery was imminent or immediately after the infant was born. Written confirmation of consent was obtained from one or both parents within 24 hours after the birth, as directed by the Multicentre Research Ethics Committee. If consent was refused, the infant was excluded and the mode of ventilation was left to the discretion of the clinician.

After assent or consent had been obtained, infants were randomly assigned, in blocks of four, to either CV or HFOV, with stratification according to gestational age (two strata) and according to centre (25 strata). Procedures were implemented to ensure balanced assignment within strata at each participating centre. Each centre kept a log of all eligible infants and reasons for non-recruitment.

Within 1 hour after birth, eligible infants were assigned to receive either CV or HFOV as their primary mode of respiratory support. Unless the infant could be extubated electively, switching from the assigned mode of ventilation was permitted only during the first 120 hours after birth, if the clinical condition for a minimum of 1 hour met the criteria for treatment failure. These criteria were a partial pressure of oxygen (PaO₂) of < 49 mmHg in an infant receiving a fraction of inspired oxygen (FiO₂) of 1.0 following changes in the mean airway pressure or peak inspiratory pressure, or a partial pressure of carbon dioxide (PaCO₂) of > 60 mmHg despite interventions to improve ventilation, or both. If the infant still required ventilation after 120 hours of age, clinicians were free to use whichever mode of ventilation they wished. No changes in clinical management except those indicated below were specified as part of the trial. Conventional ventilation was delivered by time-cycled, pressure-limited ventilators starting with a rate of 60 breaths per minute and an inspiratory time of 0.4 seconds. Subsequently, ventilator settings were adjusted at the discretion of the attending clinician to maintain a PaO₂ between 49 and 75 mmHg and a PaCO₂ between 34 and 53 mmHg. HFOV, with optimisation of lung volume, was delivered by one of three models of high-frequency oscillator [the Dräger Babylog 8000 (Dräger Medical, Lubeck, Germany), the SensorMedics 3100A (CareFusion, San Diego, CA, USA), or the SLE 2000HFO (SLE Ltd, South Croydon, UK)], all of which have been shown to have similar performance characteristics at the frequencies recommended in this trial. Ventilation was begun at a mean airway pressure of 6–8 cmH₂O and a frequency of 10 Hz, and the amplitude was increased until the infant’s chest was seen to be ‘bouncing’. The ratio of inspiration to expiration was fixed at either 1 : 1 (with the Dräger or SLE ventilator) or 1 : 2 (with the SensorMedics ventilator), in accordance with the manufacturers’ recommendations. The FiO₂ was initially set to ensure adequate oxygenation (PaO₂ > 48 mmHg), and, when the FiO₂ was > 0.3, the mean airway pressure was increased by 0.5–1.0 cmH₂O every 10–15 minutes until it was possible to decrease the FiO₂. The FiO₂ was reduced to 0.3 before the mean airway pressure was decreased, provided that the lungs were not hyperinflated (a condition defined by the flattening of the diaphragm below the margin of the ninth rib on chest radiography). Settings were then adjusted to maintain a PaO₂ between 49 and 75 mmHg and a PaCO₂ between 34 and 53 mmHg. Oxygenation was managed by adjustment of the mean airway pressure and the FiO₂; PaCO₂ was managed by adjustment of the oscillatory amplitude, but if difficulties in the management of the PaCO₂ persisted after a change in the amplitude alone, the ventilator frequency was also adjusted. If pulmonary interstitial emphysema developed, the strategy was changed to one of low volume and high FiO₂ with the reduction in the mean airway pressure to the lowest level compatible with a PaO₂ of 49–75 mmHg, even if this strategy resulted in an increase in the FiO₂ to the range of 0.7–0.9. No simultaneous positive-pressure breathing was used. The protocol recommended that infants receive exogenous surfactant as soon as possible after birth. A subsequent dose (given approximately 12 hours later) was recommended for infants receiving CV if the FiO₂ was > 0.3 and for infants receiving HFOV if the mean airway pressure was > 10 cmH₂O.

Statistical analysis

An independent committee reviewed statistical analyses performed 12 and 18 months after recruitment began and found no reason to stop the trial early. Analyses were adjusted to preserve an overall level of significance of 0.05. For the secondary outcomes (both main effects and interactions), we used the Bonferroni method to correct for multiple testing, which resulted in the use of a p-value of 0.004 to indicate significance. All reported p-values are uncorrected unless otherwise stated.

Unadjusted relative risks or hazard ratios, as appropriate, with 95% confidence intervals (CIs) were calculated to estimate the relative effect of HFOV as compared with that of CV for all outcomes. Logistic regression or Cox regression was used to investigate treatment effects, with the use of gestational age (23–25 weeks or 26–28 weeks) and location (UK and Ireland, Australia or Singapore) as covariates. Interaction terms were fit in the model in order to assess differences in treatment effects according to gestational age and location. Baseline variables with the potential to be important prognostic factors were identified in advance of the analysis. We decided to include them in the model only if a clinically important imbalance was observed. All statistical analyses were performed according to the intention-to-treat principle, with the use of Stata software (StataCorp LP, College Station, TX, USA; version 12).

Between August 1998 and January 2001, 870 infants underwent randomisation; 804 were subsequently enrolled in the trial and data from 797 were analysed ( Figure 1 ).

FIGURE 1.

United Kingdom Oscillation Study Consolidated Standards of Reporting Trials (CONSORT) flow diagram.

The two treatment groups were well balanced in terms of maternal characteristics. A total of 91% of the women received antenatal corticosteroids. The groups were also closely matched in terms of characteristics of the infants; 96% of infants were given surfactant replacement therapy at a median of 28 minutes after birth (range 0–1232 minutes).

Pulmonary function at follow-up of very preterm infants from the United Kingdom Oscillation Study

There were similar rates of chronic lung disease, defined as oxygen dependency at 36 weeks postmenstrual age (BPD), in the two ventilator groups of UKOS,1 as reported above. A diagnosis of BPD, however, has been poorly associated with long-term respiratory outcome. Potential differences in lung function between the groups could become apparent as the infants grew older. Indeed, it has been reported that airway function may deteriorate during the first year after birth in prematurely born infants, regardless of whether or not they had initial lung disease. 12,13 A previous randomised study14 had included respiratory follow-up and measurement of pulmonary function in infancy. 15 No differences in lung function in infancy were found. 15 Infants in that study, however, were relatively mature compared with those recruited into UKOS; in addition, they did not receive antenatal steroids or exogenous surfactant and no strategies to optimise lung volume on HFO were employed. 16 The aim, therefore, of this study10 was to test the hypothesis that infants who had been exposed to antenatal steroids and exogenous surfactant and randomised to HFO in the UKOS trial would have superior pulmonary function at follow-up to those ventilated conventionally.

Pulmonary function assessments at 1 year corrected age were performed at a single centre in London, UK [King’s College Hospital (KCH)], and a subgroup of trial infants was recruited from the participating centres that were within reasonable travelling distance from that centre. Informed written consent from infants’ parents was obtained before testing, and the study was approved by both the South Thames Multicentre Research Ethics Committee and the Local Research Ethics Committee of KCH NHS Trust.

Infants were tested between the ages of 11 and 14 months corrected age. Before their appointment, parents were asked to complete a 2-week respiratory symptom diary card. Appointments were deferred if the infant developed symptoms of a respiratory tract infection during this period. All infants were seen in the paediatric respiratory laboratory at KCH. On arrival, a history was taken, and each infant was weighed, measured and examined. Parents were asked not to reveal the mode of ventilation to which their child had been initially assigned. The testing procedure consisted of measurement of tidal breathing parameters, functional residual capacity (FRC) by whole-body plethysmography (FRC_pleth), inspiratory and expiratory airway resistance (R _aw), and FRC by helium gas dilution (FRC_He). Additional detail on the method for making these measurements is provided in the online supplement.

Respiratory and neurological outcomes at 2 years of age of infants from the United Kingdom Oscillation Study11

In this study,11 the outcome for surviving infants up to 2 years of age corrected for prematurity who had been entered into UKOS was assessed to determine whether ventilatory modality was associated with either increased longer-term respiratory or neurodevelopmental morbidity.

Introduction

Recent meta-analyses of randomised trials of new modes of ventilation have demonstrated that only HFO use was associated with a significant reduction in BPD, but the effect was modest. 29 In that meta-analysis of 15 trials, overall, there were no significant differences in severe intracranial haemorrhage or periventricular leukomalacia rates, but the effects were inconsistent across the trials. Following the adverse results of one trial,30 a type of oscillator was withdrawn. Nevertheless, a survey showed that 40% of UK neonatal units regularly ventilating babies use HFO in addition to, or in place of, CV. 31 Before even more neonatal units adopt HFO into their routine practice, it is essential to determine, using clinically meaningful assessments, that is respiratory and neurodevelopmental status at school age, whether or not HFO is at least as safe and efficacious as CV techniques. Only if similar or better outcomes are found would it be appropriate to continue to use HFO.

The clinical implications of the systematic review29 are difficult to interpret, as the diagnosis of BPD does not correlate well with long-term pulmonary outcomes in prematurely born children. A better predictive measure is lung function assessment at follow-up, but this has rarely been incorporated into randomised trials of HFO. At school age, the data on lung function are limited and conflicting. Small airway function may decline over the first year after birth in prematurely born infants. 12 Whether or not there is catch-up growth has not been examined, but the results of a non-randomised study suggested that the decline does not occur if prematurely born infants are initially supported by high-volume HFO rather than CV. 13 Thus, it is important to determine whether or not the use of high-volume HFO in a randomised trial in infants at highest risk for adverse long-term respiratory outcomes, that is those born very prematurely, is associated with better lung function, particularly small airway function and other respiratory outcomes at school age. Longitudinal assessment of lung function is also required to determine if the use of HFO from birth influences catch-up growth in lung function.

Although, the meta-analysis of HFO trials demonstrated no significant excess of neurodevelopmental abnormalities,29 in some studies HFO has been associated with increases in severe intracranial haemorrhage and periventricular leukomalacia. The associations are biologically plausible as high-volume HFO could cause lung overdistension compromising cardiac output and cerebral perfusion. In addition, HFO could increase hypocarbia, which can also result in less severe, but clinically important, degrees of brain injury. Thus, it is important when assessing long-term respiratory outcomes of infants entered into randomised trials of HFO to also determine their long-term neurodevelopmental outcomes. Such data are essential to determine if HFO should continue to be used and be introduced even further into clinical practice or, conversely, its use be discontinued for very prematurely born infants, even if there are favourable respiratory outcomes.

Pulmonary hypertension (PH) complicates severe BPD, but even raised pulmonary vascular resistance (PVR), which can be present in older patients with BPD, can result in morbidity. There is some evidence that the degree of PVR may also depend on ventilator strategy, as it may be lower if fast rates and low tidal volumes rather than slow rates and high tidal volumes are used. Thus, it was important to determine if HFO use in very prematurely born infants might reduce the risk of PVR at school age. Diagnosis of PH is often difficult because the symptoms may be subtle and masked by coexisting respiratory problems. 32 Doppler echocardiography is commonly used to screen for PH in clinical practice and has been used to screen for PH in other groups of patients such as those with sickle cell disease. 33 The children in this study were assessed using Doppler echocardiography, which is an accurate and non-invasive technique. 33,34 Although tricuspid regurgitation is seen in only about 33% of normal children, if there is PH, 80% of patients will have tricuspid regurgitation which can be quantified by Doppler. 35

The aim of this follow-up study was to determine the long-term outcomes of children who had been recruited into UKOS and, in particular, to test the hypothesis that use of HFO in the newborn period would be associated with superior small airway function at school age. In addition, we wished to assess the effects of HFO compared with CV on a broad range of respiratory health and educational outcomes at age 11–14 years in children born very prematurely. The results of those follow-up assessments of children from the randomised trial would robustly inform the true risk–benefit ratio of use of HFO in very prematurely born infants. A null (no difference) finding would be as important clinically as any difference that might be observed, as it would resolve the uncertainty surrounding the long-term effects of HFO and CV and determine whether or not HFO could be safely used to support very prematurely born infants. A subsidiary aim was to track the lung function in the subset of children previously assessed at 1 year, as those results would highlight whether or not changes in lung function over time differed according to ventilation mode.

Study design

Comprehensive lung function and cardiac assessments were undertaken when the children were 11–14 years of age at KCH NHS Foundation Trust, London, UK. All assessments were made by a research fellow and research nurse blind to the child’s randomised mode of ventilation. Respiratory, health-related quality of life and functional assessment questionnaires were completed (see Appendix 1 ). Parents and their children who were unable to attend the London centre completed the questionnaires only.

Recruitment of children into the study and parent input

The UKOS children were followed to 2 years of age. Since then we have maintained contact with the families by sending birthday cards to the UK-based children. This included an information sheet, a stamped-addressed envelope and a request to inform us of any change in contact details. We also provided information about the study on our website. Families have spontaneously kept in touch with us and many have informed us of changes in their contact details. When funding for the follow-up at age 11–14 years was obtained, a newsletter was sent to all families. A mother of a UKOS child has been involved in the study design and its conduct as a member of the steering committee. Her input has been invaluable in advising us on recruitment strategies and communication with families.

Assessments

Sample size

The primary outcome was small airway function. A sample size of 320 allowed a difference of 0.36 SDs in the mean lung function results to be detected with 90% power at the 5% significance level. Differences in lung function of ≥1.00 SD have been demonstrated in children with and without adverse respiratory outcomes; thus, our sample size allowed detection of a clinically important difference in lung function. Secondary outcomes were other aspects of lung function, respiratory health and symptoms, multiattribute health status as assessed by HUI-3, the results of the SDQ, special educational needs (SEN) support and subject-specific educational attainment.

Statistical analysis

The study was analysed as a two parallel-group study in keeping with the original design. The general modelling approach was to use a mixed model for both continuous and binary outcomes with the mother/pregnancy as the random effect to allow for clustering due to the relatively high proportion of multiple births common in very preterm populations. Methodological work involving simulations by one of the UKOS investigators (JP) and colleagues has shown that for data sets with a similar structure to UKOS, that is with most children being singleton births (i.e. a cluster of size one), but with a proportion of children who are twins, triplets or quads, the best estimates are obtained from using a mixed model, even if the proportions of multiples is relatively low. 42 For the binary data, the Laplace method was used within Stata as our ongoing simulations have indicated that this method gives the most reliable estimates. All study outcome analyses were adjusted for observed baseline imbalances between the two ventilation groups by incorporating the unbalanced factors as fixed effects in the multifactorial model. Unadjusted and adjusted analyses have been presented to show the effects of adjustment as estimates with 95% CIs. As we had a clearly predefined single primary outcome, FEF₇₅, we have not adjusted for multiple testing of the secondary outcomes. In a few cases with very small proportions for secondary outcomes, for example, cerebral palsy, affecting 30 children overall, the mixed model with covariates would not converge and so a one-level logistic model was used with the clustering allowed for by obtaining a robust standard error. Where numbers of a binary event were very small, an adjusted analysis was impossible using any method and so, in such cases, a simple chi-squared test was used to provide an indicative p-value.

Neonatal baseline data were compared for the children recruited and not recruited at age 11–14 years to determine the representativeness of the group with follow-up data. Neonatal and follow-up data in the sample recruited for follow-up were compared by ventilation group to check for imbalance by group due to differential recruitment.

The primary analysis was to compare FEF₇₅ z-score by mode of ventilation at birth. The z-scores normalised lung function for the sex and height of the child using standard formulae incorporated into the lung function measuring equipment.

As some lung function results had a skewed distribution, those data were transformed, most frequently using a logarithmic transformation. A further sensitivity analysis was performed to adjust lung function for cotinine level and pubertal stage regardless of their statistical significance. This was done as cotinine is an indicator of exposure to environmental tobacco smoke and thus potentially affects respiratory function and pubertal stage is linked to growth rate. A further sensitivity analysis was performed on key secondary outcomes derived from the questionnaire data to include only those children with both lung function and questionnaire data as, as anticipated, some families were only able to complete questionnaires by mail and not able to attend for assessment.

Differences in mean lung function are often difficult to interpret clinically and, so, for the primary outcome we also calculated the proportions of children in each ventilator group who had results below the tenth centile as a criterion for ‘poor’ lung function. This was possible because the lung function z-scores follow a normal distribution. The fuller rationale and methods for this are given in Peacock et al. 43

Dealing with missing data

Some children were unable to complete all lung function tests and, so, multiple imputation using chained equations was used to impute missing data. The following variables, in addition to all lung function variables, were used in the imputation: ventilation group, birthweight, gestational age group, use of surfactant, multiple birth, mother’s ethnicity, child’s current height, a binary health indicator (any report of the following wheeze, antibiotics, chest medicine, hospital admission, seizure, diabetes, cerebral palsy, hydrocephalus, gastrostomy, bowel stoma indicating ‘yes’) and attention deficit hyperactivity disorder (ADHD) inattention score. Fifty data sets were imputed, and the imputation assumed that given these covariates, the data were missing at random. A further sensitivity analysis of the primary outcome was performed adjusting for the factors related to non-recruitment, namely ethnicity, Index of Multiple Deprivation (IMD) and maternal smoking during pregnancy. 44

Intraclass correlation coefficients

These have been calculated to aid other researchers.

Software

An online data collection system for clinical studies (MedSciNet; MedSciNet AB, Stockholm, Sweden) was used for data collection and data management. Statistical analysis was conducted using Stata.

Results

Recruitment

Seven hundred and ninety-seven infants were recruited into UKOS from 25 centres; 22 were in England, Scotland or Wales and one in each of Australia, Ireland and Singapore. Infants from the 22 UK centres were followed up at the age of 6, 12 and 24 months and a subset of 76 UK-based children, who were able to travel to KCH, underwent lung function assessment at age 12 months.

The target group for the current study included all 538 children in England, Scotland, Wales and Ireland surviving to hospital discharge ( Figure 2 ). Fifteen children had subsequently died and 57, despite vigorous efforts, could not be traced, and so the maximum available sample was 466. One hundred and forty-eight children either declined follow-up or failed to reply to multiple letters and telephone calls. A total of 319 children are the subject of this report. The planned sample size was 320 children completing all elements of the study. However, while our total was virtually at target (n = 319), only 256 of these completed full lung function tests as well as completing questionnaires, leaving 59 who took part by completing the detailed questionnaires but not the assessments, and four who completed the assessments but did not return the questionnaires. Comparison of the baseline characteristics of those who were and were not recruited demonstrated significant differences with regard to only the mother’s ethnic group, children who were recruited were more likely to have a Caucasian mother and less likely to have a mother who smoked during pregnancy (24% vs. 38%) ( Table 7 ). Differences in the birthweight z-score were of borderline significance; recruited children had on average a lower z-score than those not recruited (mean z = –0.59 vs. –0.41).

FIGURE 2.

United Kingdom Oscillation Study Consolidated Standards of Reporting Trials (CONSORT) flow diagram.

TABLE 7 - Comparison of baseline characteristics between children recruited and not recruited. The data are presented as the mean (SD) or number/total number (%) unless specified

IMD for those not recruited is based on last known address postcode. Higher values for IMD indicate greater deprivation. IMD is a UK measure of deprivation and, so, cannot be calculated for children from the three non-UK centres.

Baseline characteristics	Recruited	Not recruited	p-value
N	319	204
Male sex	162/319 (51)	109/204 (53)	0.550
Mother’s ethnicity
White	285/318 (90.0)	149/203 (73.0)	< 0.001 overall
Black	21/318 (6.6)	35/203 (17.0)
Other	12/318 (3.8)	19/203 (9.3)
IMD median (range)a	15.2 (1.0–68.1)	28.2 (1.1–70.0)	< 0.001
Birthweight (g)	895 (209)	914 (204)	0.310
Birthweight z-score (range)	–0.59 (–3.45 to 2.41)	–0.41 (–3.28 to 2.17)	0.050
Gestational age (weeks)	26.9 (1.33)	26.7 (1.39)	0.350
Multiple birth	76/319 (24)	45/204 (22)	0.640
Surfactant given	310/319 (97)	203/204 (99)	0.097
Mother smoked during pregnancy	69/292 (24)	72/188 (38)	0.001
Postnatal steroids	84/314 (27)	61/203 (30)	0.420
Oxygen dependency at 36 weeks postmenstrual age	183/319 (57)	121/204 (59)	0.660
Oxygen dependency at 28 days	262/319 (82)	164/204 (80)	0.620
Oxygen dependent at discharge	71/315 (23)	44/204 (22)	0.800

Lung function and allergy assessment results

There was a statistically significant difference in the primary outcome small airway function, FEF₇₅; the z-score was higher in the HFO group (mean FEF₇₅ z-score was –1.19 vs. –0.97) ( Table 10 ). This difference was significant in both the unadjusted model that allowed for multiple births, but did not include any covariates, and in the fully adjusted model which additionally adjusted for the baseline neonatal factors that had shown imbalance between the groups. The adjusted difference in mean z-scores was 0.23 (95% CI 0.02 to 0.45). The percentage of children with lung function below the tenth centile was 46% in the CV compared with 37% in the HFO group. There were similar differences in mean lung function between the groups for both FEF₅₀ and FEF₂₅. The histograms for FEF₇₅ shows that the two groups had a similar shape distribution and the CV distribution is simply shifted downwards, that is to say there was a reduction in FEF₇₅ in all children ( Figure 3 ).

TABLE 10 - Lung function and allergy testing results by ventilation group. Results are presented as the difference of means (HFO – CV) or odds ratioa (HFO/CV), unadjusted and adjusted for birthweight, gestational age groups and whether surfactant was given before birth

D _L,CO, diffusing capacity of the lung for carbon monoxide; FRC_SF6, functional residual capacity derived from LCI measurement using sulphur hexafluoride as the tracer gas; p.p.b., parts per billion; R5Hz, respiratory resistance at 5 Hz.

Odds ratios.

Estimates based on log-transformed FeNO. Differences are the ratio of geometric means.

Lung function tests			CV	HFO	Unadjusted difference or ORa (95% CI)	Adjusted difference or ORa (95% CI)	p-value for adjusted analysis
N			121	129
FEF75 z-score		248	–1.19 (0.80)	–0.97 (0.95)	0.21(0.00 to 0.42)	0.23 (0.02 to 0.45)	0.035
FEF50 z-score		248	–1.37 (0.85)	–1.07 (0.93)	0.28 (0.07 to 0.49)	0.30 (0.09 to 0.52)	0.006
FEF25 z-score		248	–1.16 (0.95)	–0.84 (0.90)	0.27 (0.05 to 0.49)	0.29 (0.07 to 0.51)	0.011
FEF25–75 z-score		231	–1.58 (1.05)	–1.34 (1.09)	0.18 (–0.07 to 0.44)	0.21 (–0.04 to 0.47)	0.100
FEV1 z-score		248	–0.95 (1.02)	–0.60 (1.08)	0.31 (0.06 to 0.56)	0.35 (0.09 to 0.60)	0.008
FVC z-score		248	–0.44 (0.89)	–0.29 (1.05)	0.11 (–0.12 to 0.35)	0.13 (–0.10 to 0.37)	0.270
FEV1/FVC z-score		248	–1.75 (1.78)	–1.16 (1.75)	0.54 (0.12 to 0.95)	0.58 (0.16 to 0.99)	0.007
PEF % predicted		247	80.3 (15.0)	86.3 (15.5)	5.58 (1.97 to 9.18)	5.85 (2.21 to 9.49)	0.002
Gas transfer
D L,CO z-score		210	–1.10 (0.92)	–0.81 (1.19)	0.30 (0.02 to 0.57)	0.31 (0.04 to 0.58)	0.023
VA (l)		210	3.44 (0.66)	3.40 (0.59)	–0.05 (–0.19 to 0.10)	–0.05 (–0.20 to 0.09)	0.480
D L,CO/VA (mmol/minute/kPa/l)		210	1.73 (0.20)	1.76 (0.21)	0.04 (–0.01 to 0.09)	0.04 (–0.01 to 0.09)	0.110
RV z-score		211	0.46 (1.19)	0.31 (1.35)	–0.05 (–0.38 to 0.27)	–0.09 (–0.42 to 0.24)	0.600
TLC z-score		213	0.20 (1.00)	0.36 (1.13)	0.16 (–0.11 to 0.44)	0.16 (–0.12 to 0.43)	0.260
FRCpleth z-score		218	–0.07 (1.26)	–0.11 (1.28)	–0.10 (–0.42 to 0.23)	–0.08 (–0.41 to 0.25)	0.630
VCmax z-score		213	–0.50 (0.88)	–0.17 (1.09)	0.29 (0.03 to 0.55)	0.31 (0.05 to 0.57)	0.020
FRCHe z-score		229	–0.62 (1.10)	–0.75 (1.05)	–0.15 (–0.41 to 0.10)	–0.18 (–0.44 to 0.08)	0.190
LCI		155	7.50 (1.18)	7.62 (1.39)	0.16 (–0.21 to 0.53)	0.17 (–0.21 to 0.54)	0.390
FRCSF6 (l)		163	1.77 (0.43)	1.73 (0.42)	–0.04 (–0.16 to 0.08)	–0.04 (–0.16 to 0.08)	0.530
R5Hz % predicted		237	99.6 (23)	92.5 (21)	–7.02 (–12.30 to –1.70)	–7.13 (–12.50 to –1.76)	0.009
R20Hz % predicted		237	95.5 (24)	90.2 (22)	–5.65 (–11.20 to –0.08)	–5.22 (–10.70 to 0.24)	0.061
Airways reactivity
Cold air challenge, positive response		193	24/95 (25%)	20/98 (20%)	0.76 (0.39 to 1.49)a	0.76 (0.38 to 1.53)a	0.450
Bronchodilator, positive response		37	7/21 (33%)	716 (44%)	1.56 (0.41 to 5.99)a	1.75 (0.38 to 7.98)a	0.470
FeNO (p.p.b.)b		207	15.4 (1.88)	14.7 (1.91)	0.96 (0.80 to 1.14)	0.98 (0.83 to 1.17)	0.840
Skin-prick test, positive		188	9/92 (9.8%)	11/96 (11.5%)	1.19 (0.43 to 3.28)	1.34 (0.45 to 3.99)	0.600
Number of positives	1		5/9	8/11
	2		4/9	2/11
	3		0/9	1/11

FIGURE 3.

Distribution of FEF₇₅ z-score by ventilation group. Dotted blue line indicates CV group. Solid green line indicates HFO group.

There were significant differences between the ventilation groups with regard to a number of the other lung function results: FEV₁ (difference = 0.35 SDs), FEV₁ : FVC (0.58 SDs), PEF (5.85% points), diffusing capacity of the lung for carbon monoxide (D _L,CO) (0.31 SDs), VC_max (0.31 SDs) and respiratory resistance at 5 Hz (R5Hz) (7.13% points) (see Table 10 ).

The results were all worse on average in the CV group. There were no significant differences with regard to airway hyper-reactivity and exhaled nitric oxide between the two groups.

Sensitivity analyses were performed on the lung function measurement results; pubertal stage and cotinine levels were added to the fully adjusted model ( Table 11 ). This further analysis demonstrated findings consistent with the previous analysis, with significant differences in the primary outcome and the above secondary outcomes with similar effect sizes. The results of multiple imputation used to address the incomplete lung function data demonstrated similar effect sizes between the HFO and CV groups ( Table 12 ), and the further analysis that adjusted for differences in the sample assessed and those not follow-up also gave almost identical effect sizes ( Table 13 , sensitivity analysis 2).

TABLE 11 - Sensitivity analysis of the lung function data, adjusting additionally for pubertal stage and cotinine level

p.p.b., parts per billion; R20Hz, respiratory resistance at 20 Hz.

Basic model: adjusted for birthweight, gestational age groups and whether surfactant was given before birth.

Sensitivity analysis: adjusted for birthweight, gestational age groups, whether surfactant was given before birth, puberty status (having reached stage 3 or not) and cotinine level (undetectable, passive smoker or active smoker).

Estimates based on log-transformed FeNO. Differences are the ratio of geometric means.

Lung function test	Basic modela		Sensitivity analysisb
	Adjusted difference (95% CI)	p-value	Adjusted difference (95% CI)	p-value
FEF75 z-score	0.23 (0.02 to 0.45)	0.035	0.30 (0.06 to 0.53)	0.013
FEF50 z-score	0.30 (0.09 to 0.52)	0.006	0.30 (0.06 to 0.53)	0.013
FEF25 z-score	0.29 (0.07 to 0.51)	0.011	0.25 (0.02 to 0.49)	0.034
FEF25–75 z-score	0.21 (–0.04 to 0.47)	0.100	0.17 (–0.11 to 0.45)	0.240
FEV1 z-score	0.35 (0.09 to 0.60)	0.008	0.32 (0.04 to 0.61)	0.027
FVC z-score	0.13 (–0.10 to 0.37)	0.270	0.10 (–0.16 to 0.36)	0.460
FEV1/FVC z-score	0.58 (0.16 to 0.99)	0.007	0.45 (0.02 to 0.88)	0.041
PEF (% predicted)	5.85 (2.21 to 9.49)	0.002	6.88 (2.77 to 11.0)	0.001
D L,CO z-score	0.31 (0.04 to 0.58)	0.023	0.35 (0.04 to 0.65)	0.028
VA (l)	–0.05 (–0.20 to 0.09)	0.480	–0.07 (–0.21 to 0.08)	0.360
D L,CO/VA (mmol/minute/kPa/l)	0.04 (–0.01 to 0.09)	0.110	0.03 (–0.03 to 0.08)	0.330
RV z-score	–0.09 (–0.42 to 0.24)	0.600	–0.03 (–0.37 to 0.31)	0.850
FRCpleth z-score	–0.08 (–0.41 to 0.25)	0.630	–0.09 (–0.44 to 0.26)	0.620
VCmax z-score	0.31 (0.05 to 0.57)	0.020	0.31 (0.02 to 0.60)	0.037
FRCHe z-score	–0.18 (–0.44 to 0.08)	0.190	–0.22 (–0.51 to 0.08)	0.150
LCI	0.17 (–0.21 to 0.54)	0.390	0.13 (–0.33 to 0.58)	0.580
FRCSF6 (l)	–0.04 (–0.16 to 0.08)	0.530	–0.06 (–0.18 to 0.07)	0.370
R5Hz (% predicted)	–7.13 (–12.5 0 to –1.76)	0.009	–7.45 (–13.40 to –1.50)	0.014
R20Hz (% predicted)	–5.22 (–10.70 to 0.24)	0.061	–5.42 (–11.30 to 0.43)	0.069
FeNO (p.p.b.)c	0.98 (0.83 to 1.17)	0.84 0	0.94 (0.78 to 1.14)	0.550

TABLE 12 - Multiple imputation on missing lung function data

p.p.b., parts per billion; R20Hz, respiratory resistance at 20 Hz.

Basic model (on complete data): adjusted for birthweight, gestational age groups and whether surfactant was given before birth.

Imputed values only given where data set was incomplete for that particular lung function measurement.

Estimates based on log-transformed FeNO. Differences are ratio of geometric means.

Lung function test	Available data	Imputed data	Complete cases (basic modela) adjusted difference (95% CI)	p-value	Imputed datab adjusted difference (95% CI)	p-value
FEF75 z-score	248	–	0.23 (0.02 to 0.45)	0.035	–	–
FEF50 z-score	248	–	0.30 (0.09 to 0.52)	0.006	–	–
FEF25 z-score	248	–	0.29 (0.07 to 0.51)	0.011	–	–
FEF25–75 z-score	231	17	0.21 (–0.04 to 0.47)	0.100	0.20 (–0.05 to 0.45)	0.12
FEV1 z-score	248	–	0.35 (0.09 to 0.60)	0.008	–	–
FVC z-score	248	–	0.13 (–0.10 to 0.37)	0.270	–	–
FEV1/FVC z-score	248	–	0.58 (0.16 to 0.99)	0.007	–	–
PEF (% predicted)	247	1	5.85 (2.21 to 9.49)	0.002	5.85 (2.22 to 9.48)	0.002
D L,CO z-score	209	39	0.31 (0.04 to 0.58)	0.023	0.29 (0.02 to 0.56)	0.037
VA (l)	209	39	–0.05 (–0.20 to 0.09)	0.480	–0.08 (–0.21 to 0.06)	0.280
D L,CO/VA (mmol/minute/kPa/l)	210	Not imputed	0.04 (–0.01 to 0.09)	0.110	–	–
RV z-score	211	37	–0.09 (–0.42 to 0.24)	0.600	–0.20 (–0.53 to 0.13)	0.240
TLC z-score	212	36	0.16 (–0.12 to 0.43)	0.260	0.11 (–0.16 to 0.37)	0.430
FRCpleth z-score	217	31	–0.08 (–0.41 to 0.25)	0.630	–0.11 (–0.43 to 0.21)	0.500
VCmax z-score	212	36	0.31 (0.05 to 0.57)	0.020	0.25 (0.00 to 0.50)	0.046
FRCHe z-score	228	20	–0.18 (–0.44 to 0.08)	0.190	–0.16 (–0.42 to 0.10)	0.230
LCI	153	95	0.17 (–0.21 to 0.54)	0.390	0.04 (–0.41 to 0.49)	0.870
FRCSF6 (l)	161	87	–0.04 (–0.16 to 0.08)	0.530	–0.08 (–0.20 to 0.04)	0.180
R5Hz (% predicted)	235	13	–7.13 (–12.50 to –1.76)	0.009	–7.63 (–13.10 to –2.14)	0.006
R20Hz (% predicted)	235	13	–5.22 (–10.70 to 0.24)	0.061	–5.49 (–11.20 to 0.23)	0.060
FeNO (p.p.b.)c	206	42	0.98 (0.83 to 1.17)	0.840	0.99 (0.83 to 1.19)	0.940

TABLE 13 - Additional sensitivity analyses of the lung function data, adjusting additionally for (1) pubertal stage, cotinine levels and oxygen dependency at 36 weeks; (2) ethnicity, IMD, mother smoked during pregnancy

p.p.b., parts per billion; R20Hz, respiratory resistance at 20 Hz.

Estimates based on log-transformed FeNO. Differences are the ratio of geometric means.

Lung function test	Basic model adjusted difference (95% CI)	p-value	Sensitivity analysis 1 adjusted difference (95% CI)	p-value	Sensitivity analysis 2 adjusted difference (95% CI)	p-value
FEF75 z-score	0.23 (0.02 to 0.45)	0.035	0.27 (0.04 to 0.50)	0.0230	0.27 (0.04 to 0.50)	0.022
FEF50 z-score	0.30 (0.09 to 0.52)	0.006	0.26 (0.03 to 0.49)	0.027	0.34 (0.11 to 0.57)	0.004
FEF25 z-score	0.29 (0.07 to 0.51)	0.011	0.21 (–0.02 to 0.44)	0.070	0.34 (0.11 to 0.58)	0.005
FEF25–75 z-score	0.21 (–0.04 to 0.47)	0.100	0.14 (–0.13 to 0.41)	0.300	0.29 (0.02 to 0.57)	0.034
FEV1 z-score	0.35 (0.09 to 0.60)	0.008	0.28 (0.00 to 0.56)	0.053	0.41 (0.14 to 0.67)	0.003
FVC z-score	0.13 (–0.10 to 0.37)	0.270	0.08 (–0.18 to 0.34)	0.550	0.17 (–0.07 to 0.41)	0.170
FEV1/FVC z-score	0.58 (0.16 to 0.99)	0.007	0.40 (–0.02 to 0.83)	0.062	0.69 (0.24 to 1.14)	0.003
PEF (% predicted)	5.85 (2.21 to 9.49)	0.002	6.54 (2.44 to 10.60)	0.002	6.64 (2.73 to 10.5)	0.001
D L,CO z-score	0.31 (0.04 to 0.58)	0.023	0.34 (0.03 to 0.65)	0.030	0.34 (0.05 to 0.62)	0.021
VA (l)	–0.05 (–0.20 to 0.09)	0.480	–0.06 (–0.21 to 0.08)	0.410	–0.02 (–0.17 to 0.14)	0.820
D L,CO/VA (mmol/minute/kPa/l)	0.04 (–0.01 to 0.09)	0.110	0.02 (–0.03 to 0.08)	0.410	0.05 (0.00 to 0.11)	0.069
RV z-score	–0.09 (–0.42 to 0.24)	0.600	0.01 (–0.32 to 0.34)	0.950	–0.09 (–0.44 to 0.26)	0.620
TLC z-score	0.16 (–0.12 to 0.43)	0.260	0.20 (–0.10 to 0.49)	0.200	0.17 (–0.12 to 0.45)	0.250
FRCpleth z-score	–0.08 (–0.41 to 0.25)	0.630	–0.07 (–0.42 to 0.28)	0.700	–0.10 (–0.45 to 0.24)	0.560
VCmax z-score	0.31 (0.05 to 0.57)	0.020	0.30 (0.01 to 0.59)	0.043	0.36 (0.10 to 0.61)	0.007
FRCHe z-score	–0.18 (–0.44 to 0.08)	0.190	–0.20 (–0.50 to 0.09)	0.170	–0.21 (–0.48 to 0.07)	0.140
LCI	0.17 (–0.21 to 0.54)	0.390	0.21 (–0.22 to 0.65)	0.340	0.29 (–0.10 to 0.68)	0.140
FRCSF6 (l)	–0.04 (–0.16 to 0.08)	0.530	–0.06 (–0.19 to 0.06)	0.340	–0.05 (–0.18 to 0.08)	0.450
R5Hz (% predicted)	–7.13 (–12.50 to –1.76)	0.009	–7.23 (–13.20 to –1.29)	0.017	–8.28 (–14.10 to –2.46)	0.005
R20Hz (% predicted)	–5.22 (–10.70 to 0.24)	0.061	–5.36 (–11.20 to 0.51)	0.073	–5.88 (–11.90 to 0.17)	0.057
FeNO (p.p.b.)	0.98 (0.83 to 1.17)a	0.840a	0.92 (0.76 to 1.11)a	0.390	0.99 (0.81 to 1.20)a	0.890

Analysis of the lung function results of children who had also been assessed at 1 year demonstrated that their small airway function had deteriorated, as demonstrated by an increase in gas trapping.

Discussion

We have demonstrated that schoolchildren born extremely prematurely who were supported by HFO in the neonatal period had an increase in mean lung function of 0.23 SDs on average compared with those supported by CV. The proportion of children with lung function results below the tenth centile was eight percentage points lower in the HFO group than in the CV group. Specifically, the HFO group had better small airway function (FEF₇₅), as we had hypothesised and, in addition, they also had superior large airway function. Those latter results are particularly compelling as there were similar findings from different assessments of large airway function (FEV₁, FEF₅₀, FEF₂₅), including from the non-volitional test impulse oscillometry. In addition, the HFO group had better D _L,CO results, suggesting they had a greater lung surface area for gas exchange. The groups did differ at baseline in mean birthweight, gestational age and administration of surfactant, but all differences favoured the CV group and adjustment for these factors had no effect on the differences in mean lung function that were observed. The difference in the mean FEF₇₅ results between the two groups was due to a shift in the entire CV group’s distribution downwards, rather than an effect only in certain children (see Figure 3 ). It was a whole-population effect, as first described by Rose,45 and arises when there is a small effect on each subject. Thus, these data suggest that the use of HFO would benefit all extremely prematurely born infants.

The differences in lung function, although statistically significant, were relatively small: on average, approximately 0.30 z-scores. This small effect and the respiratory reserve in childhood explain why there was no associated increase in respiratory morbidity, as documented by symptom status and need for medication on the parent-completed questionnaires. In addition, there was no significant difference in the number of hospital admissions between the two groups, but only three of the whole cohort had required admission to hospital for chest problems. The greater proportion of the CV group than the HFO group with small airway function results below the tenth percentile, however, may make them more vulnerable to subsequent lung function insults such as smoking. There were no significant differences in the echocardiographic results between the two groups; whether or not this reflects that few of the children had PH in the neonatal period is not known, as the centres did not undertake routine screening. It may be, however, that both groups have clinically important abnormalities in PVR. To address that question, a cohort of term-born children is currently being assessed.

The results of our subset who were also measured at 1 year suggests that their small airway function may have deteriorated, as they had greater evidence of gas trapping when assessed at 11–14 years of age than when they were assessed at 1 year corrected age. Those results are in keeping with the decline in small airway function seen in the first year after birth in moderately prematurely born infants12 and extremely prematurely born infants initially supported by CV. 13 Thus, it will be very important to reassess all the children to determine whether or not their lung function deteriorates further still with increasing age and they become symptomatic.

We did not recruit 320 children for full assessment, but recruited 319 overall and 256 children for lung function assessment. Maternal smoking was higher than found among mothers of a Swedish cohort of schoolchildren46 and likely reflects the lower socioeconomic class of prematurely born infants. Maternal smoking also differed between those who were and were not recruited. The lung function group’s results, however, showed consistent and statistically differences which were unchanged by any of the many adjustments we employed and so we are fully confident in our findings.

We were concerned that any respiratory benefit associated with use of HFO might have been associated with adverse neurodevelopmental outcomes as, in some trials, HFO has been associated with increases in severe intracranial haemorrhage and periventricular leukomalacia. Those adverse outcomes could be the result of lung over-distension compromising cardiac output and cerebral perfusion and/or hypocarbia. There were, however, no significant differences between the groups regarding the majority of assessments of functional outcomes. A significantly greater proportion of the HFO children recorded that they had emotional symptoms on the SDQ, but this difference was not found by either the parents or teachers and is probably a type 1 error. Indeed, there were significant differences between the two groups in educational attainment with regard to art and design, IT and design and technology, all favouring the HFO children. In addition, a borderline significantly greater proportion of the CV children were receiving SEN support at school.

There are now 17 trials of elective HFO compared with CV for acute pulmonary dysfunction in preterm infants included in the systematic review in the Cochrane database. 47 HFO use remained associated with a reduction in BPD, although the effect was of borderline significance, it was also associated with a significantly lower incidence of retinopathy of prematurity. HFO use was also associated with a significantly increased incidence of pneumothorax, but, overall, there was no significant difference according to ventilation mode with regard to short-term neurological morbidity. 47 The authors of the systematic review concluded that there was no clear evidence that elective HFO offers any important advantages over CV when used as the initial ventilation strategy to support preterm infants and future trials should target those infants at highest risk of BPD, extremely prematurely born infants and report important long-term neurodevelopmental outcomes. 47

We have undertaken a large randomised trial of HFO (UKOS) in extremely prematurely born infants, all born before 29 weeks of gestational age. 1 There were no significant differences in the short-term outcomes1 or at the 2-year follow-up,11 although certain respiratory outcomes favoured the HFO group. 11 It is, then, perhaps not surprising in retrospect that we now have identified clinically important differences in respiratory function favouring the HFO group compared with the CV group when they were assessed at 11–14 years of age.

Our results demonstrate the importance of long-term follow-up of children born very prematurely entered into randomised trials if the full impact of interventions delivered in infancy is to be robustly determined. A lack of a significant positive result in infancy may not mean the intervention had no effect, but rather it may become manifest later in childhood.

Conclusion

The follow-up of extremely prematurely born infants at 11–14 years of age entered into a randomised trial of HFO compared with CV has demonstrated significant differences in lung function in favour of HFO. There was no evidence that this was offset by poorer functional outcomes; indeed, HFO children did better in some school subjects.

Contributions of authors

Professors Anne Greenough and Professor Janet Peacock designed the overall study.

Professor Neil Marlow was responsible for which functional assessments were used and Dr Sandy Calvert was the principal investigator for UKOS.

All the above contributed to the ongoing monitoring of this study and interpretation of the results.

The UKOS follow-up team additionally consisted of:

Dr Sanja Zivanovic (research fellow) who was primarily responsible for the follow-up assessments and will be writing up this study as her PhD thesis.

Mrs Mireia Alcazar-Paris (research nurse) who assisted Dr Sanja Zivanovic to undertake the assessments, she also was responsible for contacting the parents and data entry.

Ms Jessica Lo (statistician) who was responsible for checking the data and for all data analysis overseen by Professor Peacock.

United Kingdom Oscillation Study steering committee

The external steering group was chaired by Professor Henry Halliday, Honorary Professor of Child Health, and included clinical experts, a senior statistician and a UKOS parent (see below). The steering committee met on two occasions during the study. The external members provided invaluable advice about the conduct of the study, particularly regarding contacting families and maximising the response rate. The external steering committee also commented on the data accrual in terms of completed visits and on the statistics analysis plan.

The other members were:

Professor John Henderson, University of Bristol, Professor of Paediatric Respiratory Medicine: independent member.

Dr Steve Cunningham, University of Edinburgh, Consultant and Honorary Reader in Paediatric Respiratory Medicine: independent member.

Mrs Sally Kerry, Queen Mary University of London, Reader in Medical Statistics: independent member.

Mrs Janie Dromgoole: UKOS parent representative.

Professor Anne Greenough, King’s College London, Professor of Neonatology and Respiratory Physiology: joint principal investigator.

Professor Janet Peacock, King’s College London, Professor of Medical Statistics: joint principal investigator.

Professor Neil Marlow, University College London, Professor of Neonatology: co-investigator.

Dr Sandra Calvert, St George’s University of London, Consultant Neonatologist: co-investigator.

Publications

Zivanovic S, Peacock J, Alcazar-Paris M, Lo JW, Lunt A, Marlow N, et al. United Kingdom Oscillation Study Group. Late outcomes of a randomised trial of high frequency oscillation in neonates. N Engl J Med 2014;370:1121–30.

Disclaimers

This report presents independent research funded by the National Institute for Health Research (NIHR). The views and opinions expressed by authors in this publication are those of the authors and do not necessarily reflect those of the NHS, the NIHR, NETSCC, the HTA programme or the Department of Health. If there are verbatim quotations included in this publication the views and opinions expressed by the interviewees are those of the interviewees and do not necessarily reflect those of the authors, those of the NHS, the NIHR, NETSCC, the HTA programme or the Department of Health.