Two speeds of increasing milk feeds for very preterm or very low-birthweight infants: the SIFT RCT

Article history

The research reported in this issue of the journal was funded by the HTA programme as project number 11/01/25. The contractual start date was in February 2013. The draft report began editorial review in February 2019 and was accepted for publication in September 2019. 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

Jane Abbott, Janet Berrington, Elaine Boyle, Ursula Bowler, Jon Dorling, Nicholas Embleton, Kenny McCormick, William McGuire, Edmund Jaszczuk, Samantha Johnson, Madeleine Hurd, Oliver Hewer, Andrew King, Alison Leaf, Louise Linsell, Christopher Partlett, David Murray, Ben Stenson, Judith Rankin and Tracy Roberts report funding from the National Institute for Health Research (NIHR) for the trial. Jon Dorling, Janet Berrington, Elaine Boyle, Nicholas Embleton, Edmund Jaszczuk, Samantha Johnson, Andrew King, Louise Linsell, William McGuire, Christopher Partlett and Tracy Roberts report receipt of funding from NIHR, outside the submitted work. Jon Dorling reports grants from Nutrinia (Nazareth, Israel) outside the submitted work; specifically, he was funded for part of his salary to work as an expert advisor on a trial of enteral insulin. Furthermore, he was a member of the NIHR Health Technology Assessment (HTA) General Board (2017–18) and the NIHR HTA Maternity, Newborn and Child Health Panel (2013–18). Elaine Boyle reports grants from the Medical Research Council and East Midlands Specialised Commissioning Group outside the submitted work. Janet Berrington reports grants and personal fees from Danone Early Life Nutrition (Paris, France) and grants from Prolacta Biosciences US (Duarte, CA, USA) outside the submitted work. Nicholas Embleton reports grants from Prolacta Biosciences US and Danone Early Life Nutrition and personal fees from Nestlé Nutrition Institute (Vevey, Switzerland), Baxter (Deerfield, IL, USA) and Fresenius Kabi (Bad Homburg vor der Höhe, Germany) outside the submitted work. Samantha Johnson reports grants from Action Medical Research (Horsham, UK), EU Horizon 2020 (Brussels, Belguim), the Medical Research Council (London, UK), Sparks (London, UK) and the Nuffield Foundation (London, UK) outside the submitted work. William McGuire is a member of the NIHR HTA Commissioning Board (2013 to present) and the HTA and Efficacy and Mechanism Evaluation Editorial Board (2012 to present). Edmund Juszczak was a member of the NIHR HTA General Board from 2016 to 2017 and the HTA funding committee (commissioning) from 2013 to 2016.

Copyright statement

© Queen’s Printer and Controller of HMSO 2020. This work was produced by Dorling et al. under the terms of a commissioning contract issued by the Secretary of State for Health and Social Care. 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.

2020 Queen’s Printer and Controller of HMSO

Outcomes affected by feeding strategies

In the UK, 1–2% of newborn infants are very preterm or have a very low birthweight (VLBW). Preterm birth is the major risk factor for infant mortality, with 73% of neonatal deaths in the UK occurring in infants born before 37 completed weeks of gestation. 2 As survival, especially of very preterm infants, has increased in recent years,3 the high prevalence of morbidity associated with preterm birth means that the assessment of long-term outcomes has become increasingly important. 4 Short- and long-term outcomes for preterm infants are affected by strategies that reduce infection rates, lower necrotising enterocolitis (NEC) rates, promote adequate growth and maintain access to tertiary-level facilities.

Optimising infant feeding strategies offers the potential to improve all of these outcomes. Benefits are likely to arise from both the individual and the combined effects of identifying the optimum feeding strategy, as the rates of such complications in very preterm infants are high. NEC that is severe enough to cause death or require surgery affects approximately 7.5% of infants born before 29 weeks of gestation and is the cause of death in 11% of the deaths of infants born before 32 weeks’ gestation. 5 Late-onset sepsis (LOS) affects around 25% of very preterm infants and is responsible for 10% of deaths in the same population. Long-term data following LOS or NEC suggest that these conditions almost double the risk of poor neurodevelopmental outcome. 6 Preterm infants are at significant risk of poor long-term neurodevelopmental problems: almost 12% of infants have moderate or severe disability,7 with both sepsis and NEC dramatically increasing this risk. 8–13

Nutritional support of preterm infants and speed of increasing milk feeds

Every year in the UK, around 8000 infants are born so preterm that they cannot initially be fed milk and, therefore, require intravenous nutrition. Milk feeding is gradually increased as the immature gut begins to tolerate milk and intravenous nutrition is correspondingly reduced, but there are few data determining how quickly this is best achieved. 14 One of the most serious complications of intravenous feeding is LOS, which occurs in 27% of infants born weighing < 1500 g at birth or under 29 weeks’ gestation. 14 LOS is known to cause poor long-term cognitive outcomes, liver damage and sudden death from cardiac problems resulting from misplaced catheters. 15–17 One of the most common late-onset infections is ‘catheter-related bloodstream infection’; the risk of bloodstream infection is directly related to the time the catheter is indwelling in the bloodstream. 18–20 The more rapid advancement of enteral feeds described in this study will, in principle, reduce exposure to intravenous nutrition by causing infants to reach full milk feeds (tolerating 150 ml/kg/day for 3 consecutive days) approximately 4 days earlier than the slower advancement. Reducing exposure by this amount could reduce the number of infections by between 5 and 15 cases per 250 infants, which is an absolute risk reduction of 4%. This is possibly an underestimate of the reduction, as infection risk increases with the length of time a catheter is in place. 21,22

However, faster increases in milk feed volumes may increase the likelihood of NEC that, as well as being potentially fatal, may provoke intolerance of feeds or gut dysfunction, which could result in longer times to achieve full feeds rather than shorter. Survivors of NEC also have significantly worse long-term outcomes across multiple developmental domains than those who are unaffected. 6,23 Therefore, although emerging data suggest that better health outcomes may be achieved with faster feeding increments, there are possible disadvantages of this and a randomised controlled trial (RCT) is required to support a change in clinical practice. 14

Existing evidence

Existing trial data are insufficient to determine whether or not advancing enteral feed volumes slowly (typically < 24 ml/kg/day) or more quickly (daily increments of 30–40 ml/kg) affects outcomes, including the risk of neurological impairment, LOS or NEC in very preterm or VLBW infants. 14,24–32 The Cochrane review14 included nine RCTs with a total of 949 participants (Box 1). None of the studies prior to The Speed of Increasing milk Feeds Trial (SIFT) published neurodevelopmental outcomes in early childhood and the Cochrane review authors concluded ‘that advancing enteral feed volumes at daily increments of 30 to 40 ml/kg (compared to 15 to 24 ml/kg) does not increase the risk of NEC or death in VLBW infants’. 14 They also concluded that ‘advancing the volume of enteral feeds at slow rates results in several days of delay in establishing full enteral feeds and increases the risk of invasive infection’. 14 ‘The applicability of these findings to extremely preterm, extremely low-birthweight or growth-restricted infants is limited’ due to the participants studied and ‘further randomised controlled trials in these populations may be warranted to resolve this uncertainty’. 14 SIFT provided this information by recruiting a large number of infants, including those at highest risk.

Objective

The study aimed to assess if faster (30 ml/kg/day) or slower (18 ml/kg/day) daily feed increments improve survival without moderate or severe disability at 24 months of age [corrected for gestational age (CGA)] and other morbidity and mortality in very preterm and/or VLBW infants.

Design

The SIFT was a multicentre, two-arm, non-blinded, parallel-group RCT in very preterm and/or VLBW infants in the UK and the Republic of Ireland. 1,33

Ethics approval and research governance

The SIFT protocol33 was approved by the National Research Ethics Service (NRES) Committee East Midlands – Nottingham 2 on 31 January 2013 (reference 13/EM/0030). Local approval and site-specific assessments were obtained from the NHS trusts for trial sites. The trial was registered with the International Standard Randomised Controlled Trial Register (https://doi.org/10.1186/ISRCTN76463425).

Patient and public involvement

The planning and delivery of the SIFT was facilitated by close engagement with infant and family representatives who were experienced in service-user representation. Bliss (www.bliss.org.uk/; accessed 29 August 2019), the UK national charity for ‘babies born premature or sick’, was the most heavily involved charity. Parents of children who had received neonatal intensive care contributed directly and via Bliss to both the development of trial materials [e.g. parent information leaflets (PILs) and consent forms] and training research staff (e.g. in simulated ‘consent-seeking’ sessions). INVOLVE good practice guidelines were followed to ensure service-user leadership in the trial delivery and dissemination of the findings. INVOLVE is a national advisory group established and funded by the National Institute for Health Research (NIHR) to support active public involvement in NHS, public health and social care research (www.invo.org.uk/about-involve/; accessed 9 December 2019).

Participants

Setting

The setting was neonatal units caring for very preterm infants in the UK and the Republic of Ireland:

• Recruiting sites – parental consent was obtained, infants were enrolled by randomisation and participation in the trial was commenced (n = 55; see Appendix 1).

• Continuing care sites – clinicians continued to administer the intervention and collect data if a participant was transferred from a recruiting or another continuing care site (n = 78; see Appendix 2).

Infants were able to participate in other clinical trials at the same time as taking part in the SIFT, depending on the nature of the interventions in the other trials. The Enteral Lactoferrin In Neonates (ELFIN) trial was designed alongside the SIFT to allow enrolment of infants into both trials. 1,34 The SIFT and the ELFIN trial shared some procedures including some joint data collection forms and other documents. Other trials running concurrently were discussed by the chief investigators or their delegated representative, who agreed whether or not joint recruitment was appropriate.

Screening and eligibility assessment

The local health-care team identified potential participants who met the eligibility criteria. Exclusion criteria were defined and assessed by clinicians. Assessment of eligibility was accepted to be within the scope of competency of appropriately trained and experienced neonatal nurses, as no specific medical assessments were required. Competency was formally delegated by the principal investigator (PI) on the delegation log.

Informed consent and recruitment

Parents of potential participants were approached only after receiving a PIL, which gave a full verbal and written explanation of the trial. Parents who did not speak English were approached only if an adult interpreter was available and if they were likely to be resident in the UK or the Republic of Ireland for at least 2 years.

The consent-seeking process included informing parents of the possible benefits and risks as a staged process. 35 If the anticipated infant was likely to be eligible to participate in the trial, preliminary verbal information and the PIL were offered prior to birth and this was followed up after birth. For infants who were not identified antenatally or if other issues took precedence, information was provided after birth.

Written informed parental consent was obtained by means of a dated parental signature and the signature of the person who obtained the informed consent; this was the PI or the health-care professional with delegated authority. The parents were given a copy of the signed informed consent form. A copy was retained both in the infant’s medical notes and in the site file by the PI and the original copy was posted to the trials unit co-ordinating centre.

No financial or material incentive or compensation was given to the participants or parents to take part. It was highlighted to parents that they were free to withdraw their infant from the trial at any time without the need to provide an explanation or reason. The PIL explained that such a decision would not affect any aspect of clinical care.

The trial entry form was completed after informed consent was received. Information on the form was then entered in the randomisation website hosted by the National Perinatal Epidemiology Unit (NPEU) Clinical Trial Unit (CTU) (https://rct.npeu.ox.ac.uk/; accessed 30 May 2015). Randomisation took place when the clinicians were ready to increase the feeds to > 30 ml/kg/day. Infants were considered to have been enrolled once they were allocated a study number and one of the two rates of feeding increment.

Interventions

Trial participants were allocated randomly to receive daily increments in milk feed volume of either 30 ml/kg or 18 ml/kg.

All other aspects of feeding and care followed routine clinical practice in the individual units, including the capacity to stop or alter the rate of the increase in feeds if clinically indicated.

Randomisation

Randomisation was performed by computer through a secure website hosted by the NPEU CTU, University of Oxford. A minimisation algorithm was used to balance prognostic factors: hospital, multiple birth, gestational age ranges and birthweight of < 10th centile for gestational age. The algorithm included a random component that minimises with 80% probability of reducing predictability. Multiple births were allocated to the same feeding increment rate.

Primary and secondary outcomes

The primary outcome was survival without moderate or severe neurodevelopmental disability at 24 months of age CGA. Moderate or severe neurodevelopmental disability was defined as any of:

• moderate or severe visual impairment (i.e. reduced vision uncorrected with aids, blind in one eye with good vision in the contralateral eye or blind/perceives light only)

• moderate or severe hearing impairment (i.e. hearing loss corrected with hearing aids, some hearing loss but not corrected by hearing aids, or deaf)

• moderate or severe gross motor impairment (i.e. unable to walk or sit independently)

• moderate or severe cognitive impairment, assessed using the Parent Report of Children’s Abilities – Revised (PARCA-R).

A total PARCA-R score of < 44 was used to identify children with moderate or severe cognitive impairment. 38 The definition is summarised in Box 2.

Secondary outcomes included:

• mortality

• moderate or severe neurodevelopmental disability at 24 months CGA

• death before discharge home

• microbiologically confirmed (Box 3) or clinically suspected late-onset invasive sepsis (Box 4) from trial entry to discharge home

• NEC (Bell’s stage 2 or 3) from trial entry to discharge home40–42

• time taken to reach full milk feeds (tolerating 150 ml/kg/day for 3 consecutive days)

• growth (change in weight and head circumference z-score for gestational age) from birth to discharge home

• duration of parenteral feeding

• duration of time in intensive care

• duration of hospital stay to discharge home

• diagnosis of cerebral palsy by a doctor or other health professional

• the individual components of the definition of moderate or severe neurodevelopmental disability.

Diagnoses of moderate or severe neurodevelopmental disability, LOS and NEC were confirmed by the blinded end-point review committee (BERC) using standard definitions (see Appendix 3 and the published protocol1,33). All of the data collection forms were assessed independently by pairs of clinicians who were unaware of allocation. We noted that owing to ‘rounding’ of the feed rate to the nearest 0.5 ml or to small changes in a daily weight in the clinical setting, some infants on ‘full feeds’ received only 146–149 ml/kg/day. We therefore considered an infant to be on full feeds if ≥ 145 ml/kg/day was tolerated for 3 consecutive days. Infants who did not meet these criteria were reviewed by the BERC to determine if a sustained level of feeding at a level below this had been achieved before discharge. Examples of this included feeds being stopped during transfer or for a procedure, use of higher-calorie formula or fluid restriction after 150 ml/kg/day had been reached.

Sample size

It was estimated that 80% of infants would survive to 24 months of age and 11% of survivors would have moderate or severe neurodevelopmental disability. 7 It was expected that the proportion with the primary outcome would be 71% in the comparator (slower increment) group. With a total sample size of 2500 and allowing for a questionnaire response rate of 80%, there would be 90% power to detect an absolute difference of 6.3% with a two-sided 5% significance level. Similarly, a sample size of 2500 infants would have 90% power to detect an absolute difference of 5.4% (from 25.0% in the comparator group) in the incidence of LOS43 and an absolute difference of 3.5% (from 6.0% in comparator group) in the incidence of NEC (Bell’s stage 2 or 3). 40–42

Subsequently, an inflation factor of 1.12 was applied to the sample size to allow for multiple births as they received the same allocation and would probably have correlated outcomes. This adjustment assumed the proportion of multiple births to be 25% and an intraclass correlation coefficient of 0.9 for the primary outcome at 24 months CGA, based on a previous study. 44 The total target sample size was therefore increased to 2800.

Statistical analyses

Demographic factors, baseline clinical characteristics and outcomes were summarised with counts and percentages for categorical variables, means [standard deviations (SDs)] for normally distributed continuous variables and medians [interquartile ranges (IQRs) or simple ranges] for other continuous variables. Outcomes were analysed according to allocation, using the slower fed group as the comparator.

Risk ratios (RRs) and 95% confidence intervals (CIs) were calculated for the primary outcome at 24 months CGA and for the discharge outcomes of LOS and NEC, with a 99% CI used for all other dichotomous outcomes to allow for multiple comparisons. For normally distributed continuous outcomes, the mean difference (99% CI) was presented; for skewed continuous variables, the median difference (99% CI) was presented. Adjusted risk ratios were estimated using log-binomial regression, or log-Poisson regression with a robust variance estimator if the binomial model failed to converge. Linear regression was used for normally distributed continuous variables and quantile regression was used for skewed continuous variables. The primary inference was based on the analysis adjusting for the minimisation factors at randomisation. Centre was fitted as a random effect and all other factors were fitted as fixed effects. The mother’s identification was nested within centre to take account of the additional level of clustering due to multiples and siblings. This adjusts the standard error to allow for the lack of independence in trial allocation and the potential correlation in outcome.

The consistency of the effects of advancing milk feeds on the incidence of the primary outcome, LOS and NEC across specific subgroups of infants was assessed using the statistical test of interaction. Prespecified subgroup analyses included (1) week of gestation at birth, (2) birthweight of < 10th centile versus ≥ 10th centile for gestational age and (3) type of milk received during the hospital stay (i.e. breast milk only/formula only/mixed) (see Figures 3 and 4). A non-prespecified analysis assessed the effect of the speed increments on sepsis and NEC in infants with abnormal Doppler ultrasounds (see Table 4). Other deviations from the protocol included the use of quantile regression instead of Cox regression to analyse time to full feeds (as the Cox proportional hazard assumption was not satisfied) and mixed-effect log-binomial-Poisson models instead of generalised estimating equations (owing to the ease and flexibility of these methods, which were not in common use when the study was conceived). We performed a sensitivity analysis to examine the affect of missing data at 24 months on the primary outcome by considering different scenarios departing from the assumption that data were missing completely at random.

Data collection

All outcome data for this trial were routinely recorded clinical items that could be obtained from the clinical notes or local microbiology laboratory records. Information was collected using the data collection forms (see Appendix 5).

A BERC, masked to participant allocation, reviewed all case report forms (CRFs) that reported moderate or severe impairment at 24 months of age CGA, episodes of LOS, episodes of NEC or episodes of other gastrointestinal pathology. Two members who were blind to allocation independently assessed adherence to case definitions and resolved any disagreements or discrepancies by discussion or referral to a third committee member, or both. Persisting uncertainties were discussed with the site PI or research nurse or both until resolved.

We noted that owing to ‘rounding’ of the feed rate to the nearest 0.5 ml or to small changes in a daily weight in the clinical setting, some infants on ‘full feeds’ received only 146–149 ml/kg/day. We therefore considered an infant to be on full feeds if ≥ 145 ml/kg/day was tolerated for 3 consecutive days. Infants who did not meet this criterion were reviewed by the BERC to determine if a sustained level of feeding at a level below this had been achieved before discharge. Examples of this included feeds being stopped during transfer or for a procedure, use of higher-calorie formula or fluid restriction after 150 ml/kg/day had been reached.

Adverse event reporting

Adverse events were defined as serious if they:

• resulted in death

• were life-threatening

• required inpatient hospitalisation or prolongation of existing hospitalisation

• resulted in persistent or significant disability/incapacity

• were a congenital anomaly/birth defect.

The term ‘life-threatening’ refers to an event in which the participant was at risk of death at the time of the event; it does not refer to an event that hypothetically might have caused death if it were more severe. Serious adverse events (SAEs) were to be reported from randomisation to discharge home.

Expected SAEs were those that could be reasonably expected to occur in the population of eligible infants during the course of the trial or form part of the outcome data. These did not require reporting by the SIFT co-ordinating centre and referred to the following SAEs:

• death (unless unexpected in this population)

• NEC or focal intestinal perforation

• microbiologically confirmed or clinically suspected late-onset invasive infection

• bronchopulmonary dysplasia (mechanical ventilator support or supplemental oxygen at 36 weeks’ postmenstrual age)

• intracranial abnormality (i.e. haemorrhage, parenchymal infarction or white matter damage) on cranial ultrasound scan or other imaging

• pulmonary haemorrhage

• patent ductus arteriosus requiring treatment (non-steroidal anti-inflammatory drugs or surgery)

• retinopathy of prematurity.

Reporting procedures

All expected SAEs (detailed above) were recorded on a CRF and were reviewed by the Data Monitoring Committee (DMC) at regular intervals throughout the trial. Any unexpected SAEs (a SAE that was not included in the list of expected SAEs) were reported by trial sites to the SIFT co-ordinating centre as soon as possible after the event had been recognised. Information on each SAE was recorded on a SAE reporting form, which was faxed to the SIFT co-ordinating centre. Additional information received for a case (follow-up or corrections to the original case) were faxed to the SIFT co-ordinating centre on a new SAE CRF. A standard operating procedure (SOP) that outlined the reporting procedure for clinicians was provided with the SAE form and in the trial handbook. The SIFT co-ordinating centre processed and reported the events, as specified in the CTU SOPs. The chief investigator informed all of the investigators concerned of the relevant information about unexpected SAEs that could adversely affect the safety of participants. Once per year throughout the recruiting period of the trial, a safety report was submitted to the sponsor and ethics committee.

Economic analysis

A health economic analysis of the two speeds of milk feed increments was performed in this study and is described in Chapter 4.

Governance and monitoring

Structured training for site investigators, local research nurses and other clinical staff was provided during initiation meetings. Training covered areas such as seeking consent, protocol details and processes and governance requirements. These events were supported with bespoke written and online training materials that were available to all staff via the trial website (www.npeu.ox.ac.uk/sift/neonatal-staff; accessed 9 January 2020). Staff in continuing care sites were directed to online training and advised to access support from the trial team as needed.

Ongoing monitoring included review of the investigator site files that contained delegation logs, good clinical practice certificates and research curricula vitae of staff. Best-practice data management procedures and data monitoring at the study data centre and trial centres were followed to achieve quality assurance. Data management was in accordance with SOPs at the trial co-ordinating centre (NPEU CTU) and a prespecified plan. Data monitoring included review of consent forms and participant eligibility. Additional validation checks of data were carried out regularly, with data queries issued to study sites for resolution. Final data validation checks were carried out before database lock, with questions being resolved by discussion with the site PI or local research nurse where possible.

During the trial, the study statisticians produced reports for the independent Trial Steering Committee (TSC) and the independent DMC. Data quality concerns that were identified by study statisticians were reported to study data management staff and were queried when appropriate or included in future routine data validation checks, or both. Opportunities for external, independent review of summary data were provided by the DMC and the TSC meetings.

Summary of changes to the study protocol

A summary of the changes made to the original protocol is presented in Appendix 7.

Recruitment and retention

Patient flow, including recruitment to and retention in the trial, is detailed in Figure 1. The trial recruited infants from June 2013 to June 2015 in 55 hospitals. In total, the trial recruited 2804 infants; 1400 infants were allocated to faster daily feed increments (30 ml/kg/day) and 1404 were allocated to receive slower feed increments (18 ml/kg/day). The trial was closed on reaching the sample size. All infants received the allocated intervention, but 69 infants discontinued the intervention as a result of clinician or parental preference (see Figure 1). For 11 of these infants, parental consent was withdrawn and their data were not available for analysis, but the remainder were included in the intention-to-treat analysis. Outcome data at discharge home were not available for eight infants; their data were included in analyses except when knowledge of discharge or the date of discharge was required. In total, 68 (4.9%) infants in the faster increment group and 77 (5.5%) in the slower increment group died before 24 months CGA. Outcome data on disability at 24 months CGA were available for 1156 (87.2%) surviving infants in the faster increment group and 1169 (88.4%) in the slower increment group. The primary outcome (mortality or disability) was therefore known for 1224 (87.8%) infants in the faster increment group and 1246 (89.0%) in the slower increment group (see Figure 1 and Appendix 8).

FIGURE 1.

Flow of participants through the trial. Adapted from New England Journal of Medicine, Dorling et al. 45 Controlled trial of two incremental milk-feeding rates in preterm infants. N Engl J Med 2019;381:1434–43. Copyright © 2019 Massachusetts Medical Society.

Adapted from New England Journal of Medicine, Dorling et al. 45 Controlled trial of two incremental milk-feeding rates in preterm infants. N Engl J Med 2019;381:1434–43. Copyright © 2019 Massachusetts Medical Society.

Demographic and other baseline characteristics

The baseline characteristics and demographic features of the participating infants were well balanced between the two feeding increment groups (Table 1). The median gestational age was 29 weeks in both groups (36% at < 28 weeks). The mean birthweight was 1144 g in the faster increment group and 1142 g in the slower increment group. Overall, 60% of infants were born via caesarean section; 24% of infants were born following rupture of maternal amniotic membranes for > 24 hours; and 16% of infants had evidence of absent or reversed end diastolic flow in the fetal umbilical arteries. The allocation groups were well balanced in individual recruiting sites, as per the minimisation algorithm (see Appendix 9).

Adherence

All of the infants received the allocated intervention but 69 infants discontinued the intervention: 66 from clinician or parental preference and three from transfer to a non-participating hospital (see Figure 1). For 11 of these 66 infants, parental consent was withdrawn and their data were not available for analysis. The remainder were included in intention-to-treat analyses. Outcome data at discharge home were not available for eight infants; their data were included in analyses except when knowledge of discharge or the date of discharge was required. In total, 68 (4.9%) infants in the faster increment group and 77 (5.5%) in the slower increment group died before 24 months CGA. Primary outcome classification at 24 months CGA was possible in 1224 (87.8%) infants in the faster increment group and 1246 (89.0%) in the slower increment group.

Outcomes

The estimates of effect for the primary and secondary outcomes are presented in Table 2 for outcomes at 24 months of age CGA and Table 3 for outcomes at hospital discharge.

Subgroup analyses

A subgroup analysis showed a significant interaction (p = 0.045) with the primary outcome for the type of enteral milk received: human, formula or both (Figure 2). No significant interaction was seen with the primary outcome for completed weeks of gestation at birth (p = 0.076) or birthweight < 10th centile or ≥ 10th centile for gestational age (p = 0.18). A post hoc analysis was also undertaken to assess the interaction of the presence of absent or reversed antenatal umbilical Doppler studies with the two incremental feed rates (see Tables 4 and 5).

FIGURE 2.

Subgroup analyses for survival without moderate or severe disability to 24 months of age CGA. n/N refers to the number of infants with the primary outcome/number of infants in that category. a, The primary outcome was survival without moderate or severe neurodevelopmental disability (CGA). p-values for interaction were adjusted for minimisation factors: collaborating hospital, single or multiple birth, gestational age at birth, and whether or not the birthweight was below the 10th percentile for gestational age, when technically possible. p-values and CIs were not adjusted for multiple comparisons and should not be used to infer definitive treatment effects. Adapted from New England Journal of Medicine, Dorling et al. 45 Controlled trial of two incremental milk-feeding rates in preterm infants. N Engl J Med 2019;381:1434–43. Copyright © 2019 Massachusetts Medical Society.

Adapted from New England Journal of Medicine, Dorling et al. 45 Controlled trial of two incremental milk-feeding rates in preterm infants. N Engl J Med 2019;381:1434–43. Copyright © 2019 Massachusetts Medical Society.

The subgroup analyses did not show any significant interactions with NEC (Figure 3) for:

• completed weeks of gestation at birth (p = 0.63)

• birthweight < 10th centile or ≥ 10th centile for gestational age (p = 0.25)

• the type of enteral milk received (human, formula or both) (p = 0.53)

• the presence of absent or reversed antenatal umbilical Doppler studies (p = 0.09). This was a post hoc analysis.

FIGURE 3.

Subgroup analyses for NEC from trial entry to discharge from hospital. a, n/N refers to the number of infants with one or more episodes of NEC from trial entry to hospital discharge/number of infants in that category. b, Combined with the 30⁺⁰ to 31⁺⁶ weeks’ category for calculation of RR. c, Combined with mixed category for calculation of RR. ARRs (faster/slower) and p-values for an interaction between allocation and category are shown. Adapted from New England Journal of Medicine, Dorling et al. 45 Controlled trial of two incremental milk-feeding rates in preterm infants. N Engl J Med 2019;381:1434–43. Copyright © 2019 Massachusetts Medical Society.

Adapted from New England Journal of Medicine, Dorling et al. 45 Controlled trial of two incremental milk-feeding rates in preterm infants. N Engl J Med 2019;381:1434–43. Copyright © 2019 Massachusetts Medical Society.

The subgroup analyses did not show any significant interactions with confirmed or suspected LOS (Figure 4) for:

• completed weeks of gestation at birth (p = 0.07)

• birthweight < 10th centile or ≥ 10th centile for gestational age (p = 0.51)

• type of enteral milk received (human, formula or both) (p = 0.56)

• presence of absent or reversed antenatal umbilical Doppler studies (p = 0.16). This was a post hoc analysis.

FIGURE 4.

Subgroup analyses for confirmed or suspected LOS from trial entry to discharge from hospital. a, n/N refers to the number of infants with one or more episodes of LOS from trial entry to hospital discharge/number of infants in that category. Adapted from New England Journal of Medicine, Dorling et al. 45 Controlled trial of two incremental milk-feeding rates in preterm infants. N Engl J Med 2019;381:1434–43. Copyright © 2019 Massachusetts Medical Society.

Adapted from New England Journal of Medicine, Dorling et al. 45 Controlled trial of two incremental milk-feeding rates in preterm infants. N Engl J Med 2019;381:1434–43. Copyright © 2019 Massachusetts Medical Society.

Adjusted risk ratios (faster/slower) and p-values for an interaction between allocation and presence of absent or reversed antenatal umbilical Doppler studies are shown in Tables 4 and 5.

Safety and adverse events

Sixty-two SAEs were reported, including follow-up reports, relating to 52 separate incidents. Of these 52 events, 34 were not related or relevant to the trial. Fourteen were deemed ‘possibly’ related, but were listed in the protocol as expected SAEs. 33

Four SAEs were deemed ‘possibly’ trial related and were not on the list of expected SAEs. All were reported to the Research Ethics Committee (REC), the DMC and the TSC in previous communications. These four SAEs were two cases of intracardiac thrombosis, one case of prolonged conjugated jaundice and one case of dehydration when a central venous line extravasated. Table 6 summarises the reported adverse events (definitions of adverse reactions and events are presented in Appendix 6).

Post hoc analyses

Post hoc exploratory analyses did not show any differential effects of the faster or slower feeding increments (see Tables 4 and 5).

Introduction

This chapter reports the economic evaluation conducted as part of SIFT. The objective of the economic evaluation was to compare the relative cost-effectiveness of two rates of enteral feed advancement, faster feed increments (30 ml/kg/day) with slower feed increments (18 ml/kg/day), on the principal outcome of survival without moderate or severe disability at 24 months of age CGA.

Methods

A within-trial cost-effectiveness analysis (CEA) was performed from the perspective of the NHS and Personal Social Services (PSS) in line with recommended practice. 48 The CEA results are expressed in terms of additional cost per survivor without disability at 24 months of age CGA.

Outcomes

The primary outcome of the CEA was disability-free survival at 24 months of age CGA. Secondary outcomes of the trial included microbiologically confirmed or clinically suspected LOS and NEC. These secondary clinical outcomes are presented here as part of a cost–consequence analysis (CCA). All statistical analyses were conducted in Stata^® version 15 (StataCorp LP, College Station, TX, USA).

Data

Economic analysis

A CCA was conducted as a preliminary measure to compare the disaggregated costs with the outcomes for both feeding increments. CCA is a form of economic evaluation where disaggregated costs and outcomes (consequences) are presented in their natural units. 57 To calculate costs, the quantity of resource use per infant was multiplied by unit costs. Mean costs per infant were estimated and the mean cost differences between the two feeding allocations were calculated. To address the skewness often present in cost data, a bootstrapping approach58 was carried out to calculate CIs around the mean costs. If the CIs of the difference in mean resource use and the costs between groups do not cross zero, this indicates a significant difference. In bootstrapping, repeated random samples of the same size as the original sample are drawn with replacement from the data. 58 The statistic of interest (mean) is calculated from each resample and these bootstrap estimates of the original statistic are then used to build up an empirical distribution for the statistic. 58

The primary base-case economic analysis took the form of a CEA from the perspective of the NHS and PSS. A CEA is a method for assessing the gains in health relative to the costs of different health interventions. 59 In the current study, health consequences are measured as a clinical outcome rather than in the form of health-related utilities, such as quality-adjusted life-years (QALYs). An incremental analysis was conducted, comparing incremental (additional) costs with the outcomes between the two feeding allocations. Costs and clinical outcomes associated with each feeding allocation were combined by calculating incremental cost-effectiveness ratios (ICERs). An ICER is expressed as the incremental cost (£) per incremental gain in a natural unit. 60 Cost data were discounted at 3.5% but discounting is not applied to outcomes in natural units; thus, outcomes were not discounted. Cost-effectiveness was based on the principal outcome of additional cost per survival without moderate to severe disability at 24 months of age CGA and from the perspective of the NHS and PSS.

The cost-effectiveness estimates are presented on a cost-effectiveness plane, to illustrate the incremental cost and effect of the intervention (faster feed increments). The cost-effectiveness plane has four quadrants, each with a different implication for the decision of implementing the intervention. 61 This was used to determine the relative position of the results. For example, faster feed increments might be said to ‘dominate’ slower feed increments if its position on the plane showed that the cost of the intervention was lower and the effectiveness in achieving the outcome was greater, when compared with slower feed increments. In other words, if faster feed increments were cheaper and more effective than slower feed increments, faster feed increments would be said to dominate.

Sensitivity analysis

Statistical uncertainty around the difference in ICERs was estimated by 5000 bootstrap replications, represented as scatter points on the cost-effectiveness plane. Results were presented using cost-effectiveness acceptability curves (CEACs) to reflect sampling variation and uncertainties in the cost-effectiveness value where appropriate. A CEAC shows the probability that an intervention (e.g. faster feed increments) is cost-effective compared with the alternative (e.g. slower feed increments) given the observed data, for a range of maximum monetary values (thresholds) that decision-makers might be willing to pay for a particular unit change in outcome. 61 Unlike economic evaluations, in which outcomes are expressed in QALYs, there is no set monetary threshold here for the primary outcome. Therefore, we examined cost-effectiveness at a range of monetary willingness-to-pay (WTP) thresholds (£0 to £40,000 per additional prematurity-adjusted survivor without disability at 24 months).

To explore the cost of missing resource use data for those infants who did not complete follow-up (15.4%), multiple imputation was performed. Multiple imputation replaces each missing observation with a set of plausible imputed (predicted) values, drawn from the posterior predictive distribution of the missing data given the observed data. 62 Because the missing values are replaced with predicted values, calculated in this way, this method is more favourable than simpler methods, such as replacing missing values with the means of all available values. Costs were imputed at the total cost level and the imputation model used 500 imputations.

Results

Costs

Mean resource use was combined with unit costs (see Table 7) to derive the health service costs accruing from each feeding strategy (Table 9). Average costs of the initial hospital stay did not differ very much between the two groups; this is not surprising given that mean resource use was very similar between groups. Mean costs throughout the 2-year follow-up are also presented in Table 9. The most sizeable cost difference was in the cost of inpatient stays; this cost £267 more in the faster increment group. Throughout the follow-up period, health service costs were generally slightly higher for the faster increment group, particularly for primary care services, such as general practitioner, health visitor and community nurse visits.

Mean total costs for each group are presented in Table 10. Faster feed increments cost approximately £1043 less per infant than slower feed increments during infants’ initial hospital stay; however, within the 2-year follow-up the faster increment group was more costly by approximately £349 per infant. Overall, the intervention of faster feed increments was less costly.

Sensitivity analysis

The scattered points on the cost-effectiveness plane (Figure 5) show that the vast majority of points lie in the south-west and north-west quadrants. This shows that at higher monetary thresholds, faster feed increments become the more expensive strategy but remain less effective. At the higher thresholds (north west), the existing treatment (slower feed increments) dominates the intervention. This finding is confirmed by the CEAC (Figure 6), which shows a decreased probability that the intervention is cost-effective as the WTP threshold increases. This is because the more monetary value that is placed on a disability-free life, the less we value this intervention, given its detrimental effects for this outcome.

FIGURE 5.

Cost-effectiveness plane (faster feed increments vs. slower feed increments).

FIGURE 6.

The CEAC (faster feed increments vs. slower feed increments).

Table 12 shows the mean resource use for all infants once missing data were accounted for using multiple imputation. When the missing values were accounted for, faster feed increments were slightly more costly than slower feed increments (£378 more per infant). This is in contrast to the baseline results that showed the reverse; however, given the poorer clinical outcomes associated with the faster feed increment, this finding is in line with the expectations. We would expect more resource use in the faster increment group as a result of more health-care utilisation, for example primary care visits and readmissions.

TABLE 12 - Mean resource use calculated with multiple imputation (£) (2016–17 prices)

Cost	Faster feed increment group (n = 1224), mean (SD)	Slower feed increment group (n = 1246), mean (SD)
Total costs of health service use after initial discharge from hospital to 24 months of age	109,410 (92,266)	109,032 (89,763)

Discussion

Supplementary data

Background

Complex neonatal care must be underpinned by a strong evidence base. This requires RCTs, often with enrolment shortly after birth when parents are likely to be extremely anxious and where there may be concerns about longer-term survival and prognosis. Health-care professionals must weigh up the additional stress caused to parents by being approached for RCT enrolment against their desire to support trial recruitment; there are likely to be a range of factors that may affect recruitment. 70 Preterm infants are a vulnerable group because they are unable to consent for themselves and, from an ethics perspective, proxy decisions about enrolment by parents and health-care professionals should always be made in the infant’s best interests. The vast majority of parents want to be involved in decisions about whether or not their infant should participate in a RCT, although it is common for parents to also seek guidance from health-care professionals. 71 Studies of such parents suggest that most feel that they made well-informed decisions, although there is debate as to whether or not fully informed consent can really be achieved in this context. 72–74 The anticipated level of risk and benefit from each individual RCT has been established as a key driver of parental desire to consent. 75,76 In many large, research-active neonatal intensive care units (NICUs), it is common for parents to be approached to enrol their infant in more than one trial or study. This presents important ethical and scientific challenges, many of which have been outlined in a recent review focused on neonatal practice,77 although similar challenges do exist in adult intensive care settings. 78 To achieve high-quality research that determines best practice of relevance to these families, it is important for research teams to understand the parental perspective when planning, designing, conducting and analysing research studies. Where co-enrolment does not compromise scientific integrity, it might be considered unethical to deny parents the choice of supporting studies and co-enrolment may also make studies more representative of the populations and questions that they aim to address. 77

Research aim

The aim of this study was to describe the perceptions and experiences of parents of infants who were approached to participate in at least one UK NIHR collaborative RCT, as well as one or more additional RCT or observational study. This study was designed to explore the experiences of parents whose infants had been approached to join either or both of two clinical trials of investigational medicinal products (CTIMPs), which were both funded by the UK NIHR Health Technology Assessment (HTA) programme. These trials were (1) the Speed of Increasing milk Feeds Trial (SIFT),1 a non-blinded trial comparing two different rates of increase in the amount of enteral milk feeds the infant received and (2) the Enteral Lactoferrin in Neonates (ELFIN) trial that was a double-blinded, placebo-controlled trial of supplemental enteral bovine lactoferrin (a milk protein) added to milk feeds. 39 Both of these trials recruited preterm infants who were born at < 32 weeks’ gestation and who were free of major congenital anomalies in the first few days of life. At the development stage of both trials, it appeared likely that there would be considerable overlap of these two trials, although owing to other factors, in the end there was in fact minimal overlap during recruitment. However, at all of the three units that took part in this qualitative study (PARENT), a number of NIHR or local RCTs and/or observational studies were under way,79,80 including one further HTA-funded RCT, Baby-OSCAR (see www.npeu.ox.ac.uk/baby-oscar; accessed 9 January 2020). At the time of writing, the Baby-OSCAR trial is still actively recruiting preterm infants who were born at < 29 weeks’ gestation and is a double-blinded, placebo-controlled study of the use of ibuprofen in the treatment of patent ductus arteriosus. We took advice from parents when planning this study and only sought ethics permission to interview parents whose infants had joined at least one trial or study, even if they had declined enrolment in other studies or trials. This meant that we did not approach parents who had declined participation in all of the studies or trials that they had been offered.

Literature search/review

Where the prospect of enrolment to more than one RCT or study exists, additional medical, practical and ethical issues arise for parents, clinical staff and research teams. 77,81 Large NICUs may be actively recruiting to several studies at any one time, which may include multicentre collaborative or single-centre RCTs and observational studies. Recent studies with parents suggest that many are receptive to being approached to be involved in more than one RCT or study, but this area has not been subject to detailed qualitative study in the context of multiple RCTs in neonates, despite the importance to future research progress. 82,83 A recent study in a paediatric intensive care unit setting suggested that co-enrolment does not adversely affect recruitment, but this did not include direct exploration of the reasons behind parental decision-making. 84

Methodology (including any changes to the protocol) and data sources

There were no changes to the original protocol.

Sampling

Parents were recruited when their infant was no longer receiving intensive care and prior to their infant being discharged home from one of three NICUs in the north of England. We included parents who had been approached to join at least one RCT that was a CTIMP and one additional RCT or observational study and who had subsequently consented for their infant to join at least one trial or study. We used purposive sampling based on actual RCT participation, age and educational or employment factors in an attempt to reflect a broad range of opinions and experiences. The initial approach for this study was made by a research nurse who had knowledge of study approaches and enrolment.

Interviews

We reviewed the existing literature, had discussions with doctors and nurses on a NICU and developed an initial interview topic guide with the aim of provoking discussion rather than seeking specific answers. The guide was iterative and flexible, which allowed participants to have space to define the issues that they regarded as significant and to articulate their experiences in their own words. 85 We estimated that we would need 15–20 interviews to reach thematic saturation, that is the point at which no new themes emerged from the data, and we reached this after 17 interviews, which was similar to our previous work with parents of preterm infants. 86 Interviews were tape-recorded, lasted approximately 1 hour and took place in a location convenient for the participants, either in hospital or in their own home. They were conducted by a single trained qualitative researcher (Judy Richards) who had experience of interviewing parents of preterm infants but who was not a health-care professional. Interviews were conducted either shortly before or shortly after the infant was discharged home. Interview recordings were transcribed verbatim and fully anonymised and participants were offered the opportunity to read and review the interview transcript.

Data analysis

Data analysis was carried out using a thematic approach85 in order to identify patterns of meaning across our data that are relevant to the conduct of research on a neonatal unit. The interview transcripts were analysed by two research team members to ensure that there was standardisation of thematic coding. Significant themes and subthemes were extracted from the data and their inter-relationships were considered alongside contradictions and overlap that emerged between them. The themes were discussed and agreed with three members of the research team (Judy Richards, Judith Rankin and Nicholas Embleton) and we include illustrative quotations for the identified themes. Participants are identified as P1, P2, etc.

Governance

This SIFT received approval from the NRES Committee East Midlands, Nottingham, and this PARENT substudy received approval from the Office for Research Ethics Committees Northern Ireland: reference 15/NI/0021, 2 February 2015.

Results

Seventeen semistructured qualitative interviews were carried out with parents. There were eight parents of twins, eight parents of singletons and one set of triplets. Nine interviews were with the mother alone, one was with the mother and a grandparent and seven interviews were with both parents. Interviews took place either in the hospital or at home. Data analysis identified four overarching themes, which outlined factors that influenced parents’ decisions regarding their infants’ participation in more than one RCT or study.

Theme 1: ‘just another little thing’

The main finding that emerged from the data was that parents did not perceive the enrolment of their infants in more than one study as a major issue in the context of their infant’s daily medical care (Box 5). Similarly, the actual number of studies or trials (range 1–5 studies or trials) that their infant participated in was typically not raised as significant for parents. For the most part, being introduced to several trials or studies soon after the birth was not seen as problematic. The majority of parents held a strong belief in the benefits of research and had faith that the health-care professionals caring for their infant would ensure that trial participation would not compromise care in any way. Parents generally viewed their infant’s participation in more than one study positively. For parents of some sick infants where survival was uncertain, trial participation was perceived as a ‘gift’ or a way of making their infant’s potentially short life matter. A small minority of parents were concerned about perceived issues that may arise when two trial interventions take place concurrently. However, most often, parents focused on the daily medical care that their infant was receiving and many forgot about the trials or studies once they were under way.

Theme 2: information gathering

Most parents focused on the specific details of each individual trial and attempted to gain information in order to make a decision that was as fully informed as possible (Box 6). Parents preferred a succinct PIL that provided just enough information in a clear ‘jargon-free’ fashion and also valued the opportunity to talk through any concerns they had with a health-care professional. On the whole, parents tended not to seek advice from other parents on the ward but did look for ‘stickers’ found on an infant’s cot or incubator that signified trial participation. Conversely, others did not read the PIL in detail and simply agreed to participate, driven largely by feelings of moral obligation to take part in trials for the benefit of future infants; the perception that their infant was benefiting from past research; or parents’ desire to repay the health-care professionals for the care their infant was receiving.

Although parents felt that they were given all of the information that they needed regarding each trial, some parents admitted that they sometimes felt too upset or unwell to fully take in trial details, especially very soon after the birth of their infant. Indeed, some parents suggested retrospectively that they could not recall the trial names and did not fully understand the goals of the studies or processes involved, but none suggested that they had any regrets about the decisions they made. Parents often used their own names for trials to help them remember specific details of each one (the most common being the ‘poo and wee trial’).

Parents not only placed importance on the information they received from the PIL but also on the manner of the health-care professional who initially approached them, an issue that they regarded as very influential in their decision-making process. They appreciated in particular a friendly and informal, but confident, approach. The majority of parents said that they would have disliked feeling ‘pressured’ (although most did not say that they felt pressured) into making a decision and always appreciated being given as much time as feasibly possible to think about the trial implications. Although parents often could not remember the names, numbers and nature of the trials or studies, they remembered clearly the health-care professional or researcher who introduced them to a trial and supported them throughout.

Theme 3: making decisions – ‘weighing up the pros and cons’

Parental decisions around participation were not based on the number of trials or studies, but rather on the perceived risks of each individual trial that they were asked to join (Box 7). Parents considered issues such as the well-being of their infant at the time of enrolment, the invasiveness of the trial, the perceived importance of the trial goals and a number of personal motivational factors. Most parents acknowledged that it was not possible to make a fully informed decision on behalf of their infant and so in the absence of certainty they drew on more personal, non-medical risk assessments in order to make a decision. These included wondering what decision ‘baby’ would make; perceiving that trial participation would help their infant to get well; welcoming ‘another pair of eyes’ over their infant’s care; a sense of purpose that helped them to overcome feelings of helplessness; a way of getting their infant home sooner; and, as mentioned above, a debt of gratitude.

Theme 4: saying ‘no’

Although parents were happy to be approached about multiple trials or studies, this did not mean that they agreed to enrol their infant on all of them (Box 8). Some parents admitted that they felt a degree of guilt attached to declining participation, largely because they felt that they owed it to future infants to take part in research. Although very few parents could remember how many trials their infant had participated in, taking part in several trials did provide perceived justification to refuse future trials, as it gave the impression to health-care professionals that they had ‘done their bit’. Parents who did say ‘no’ to a trial were divided in whether or not they felt it necessary to give a reason to justify their decision.

Parents articulated a number of specific reasons for refusing an individual trial; one of the main reasons concerned the time that they were approached about participation. If parents had a possible discharge home date for their infant, they were often keen to safeguard this and saw enrolment on a trial close to discharge as a possible threat to their infant leaving the hospital. In addition, a small minority of parents were approached when their infant was unwell; in these instances they felt that it was inappropriate that they had been approached and that hospital notes had not been read properly. Importantly, parents highlighted that it takes confidence to say ‘no’ to health-care professionals who are caring for their infant. Some parents were relieved to be supported by nursing staff in their decision to refuse a trial, whereas others felt that their decision was not initially accepted by health-care professionals, which left them feeling slightly pressurised and guilty.

Discussion including the robustness of the results and limitations

Enrolment of preterm infants into more than one RCT is a common challenge on many research-active neonatal units. 77 Previous studies have suggested that this may not be a significant issue for many parents; however, there are few data or in-depth qualitative studies. 75,81,82,87 Our study provides an in-depth analysis of parental experiences and motivations as well as supporting the idea that enrolment into multiple studies is not necessarily problematic for most parents. Although most parents offered complex personal reasons that motivated them to see research in a positive light and subsequently enrol their infant, others provided important insights into their decision-making and how it made them feel, including their feelings of guilt or obligation.

Our study has several strengths. The semistructured nature of the interviews gave parents the space and time to reflect on their experiences of enrolling their infant on multiple trials and the initial approach to parents to take part in this qualitative study typically took place several weeks after their infant was born and enrolled into these studies. Participants were enabled to articulate in their own words the issues that were important to them. A further strength was the impartial nature of the qualitative researcher (Judy Richards) who was not a health-care professional, had no prior involvement in clinical trials and was able to provide valuable insight as a ‘naive observer’. Nevertheless, the findings of the study require careful interpretation. The design and parameters of the research that may affect the generalisability to other populations, settings and situations include the:

• specifics of the hospital sites

• timing of approach for research

• interventional nature of the trials and studies

• parent experience of having an infant who survived and had participated in at least one study

• attitudes, behaviours and beliefs of the clinical and research teams involved.

Although we utilised purposive sampling, we did not record detailed demographic, socioeconomic or educational information from the parents and did not explore how those factors, or other factors such as ethnicity or religion, affect decision-making. Around half of our interviewees were parents of twins or triplets, yet we do not explore in this report how that may affect decision-making. This may be important, as recent studies suggest that such parents have important views on the co-randomisation of their infants. 88 In addition, we did not seek the views of parents who had declined participation in all of the studies or trials that they had been offered or the feelings of parents whose infants had enrolled in a study or trial but had subsequently died. The recent BRACELET study89 explored some of these issues for parents when bereavement occurs in the context of RCTs in neonatal and paediatric intensive care.

Our study supports existing data that suggest that parents make separate decisions about each individual trial, judging each by its own perceived ‘pros and cons’. 75,90 Our findings emphasise that parents want to make final decisions about their infant’s participation and contrast with earlier data that suggested that some parents wanted or expected health-care professionals to make the final decision. 71,91 Although most parents felt that they made good decisions that they did not later regret, some parents suggested that they had not fully taken in or understood the trial information at the time that they were approached and, as a result, some subsequent trial procedures came as a surprise. This highlights the challenges of gaining fully informed consent and the importance of continued and regular involvement of the research team to explain the nature and conduct of the study to parents, as well as the rights of parents to withdraw or to decline specific procedures. 71,74,75,82,91 Importantly, however, most parents acknowledged that despite feeling stressed, they were still happy with their decision to participate, frequently citing their faith in health-care professionals to protect their infant from harm. Many parents in our study suggested that they would have liked information about the trials and studies that were available before their infant was born and although many acknowledged that there were time restraints on decision-making, they did not perceive these decisions to have been made in haste or inadvisably either at the time or with hindsight.

Conclusions including implications for health care and recommendations for additional research

This in-depth exploration of parents’ views on enrolling their infants into more than one RCT or study highlights the need for research teams to be aware of a range of factors when approaching parents on the NICU, especially in the context of co-enrolment. Although parents felt that they had made good decisions and had few regrets, our study highlights that for some this may have been motivated by guilt, a debt of gratitude or a lack of confidence to refuse. However, although parents acknowledged that fully informed consent was not possible, they were still happy that their decisions were ‘good enough’ given the situation they were in. A deeper awareness of these factors would potentially enable health-care professionals to support parents more fully throughout the process of enrolling their infants on multiple trials.

The themes identified here could been used as part of training in good clinical practice and as part of ongoing education for researchers, students, nurses, doctors and allied health professionals who are, or may in the future be, involved in designing or supporting research. The study also suggests that careful consideration is needed when PILs are developed and emphasises the importance of involving patients, parents and/or members of the public in all aspects of research design. Modifications to the PIL may be appropriate where co-enrolment is likely. The study also highlights the confusing and ‘overwhelming’ nature of having a sick infant who requires intensive care and the additional confusion when parents are approached to join research studies. Providing such parents with additional opportunities to further discuss and learn about the research objectives and procedures after they have consented and eliciting their continued assent for their infants’ participation in the study will enhance their satisfaction and understanding of the need for research that aims to improve patient care. Future research could assess these modifications to see if they improve patient satisfaction and experience.

Background

Loss to follow-up has a detrimental effect on research studies. It can compromise internal and external validity of results for a number of reasons. These include reduction of statistical power owing to smaller sample size, negative impact on generalisability and potential to bias results. 92 Therefore, it is important to implement strategies to effectively maintain retention of participants.

Organising clinical assessments for follow-up can often be very costly in perinatal trials, so research questionnaires are often used to capture follow-up data, including the primary outcome. Therefore, the need to mitigate the loss to follow-up and maximise the return rate of questionnaires is essential.

Systematic reviews suggest that incentives are effective at improving response rates for research questionnaires in clinical trials. 93,94 Brueton et al. 93 and Khadjesari et al. 95 both reported that offering a monetary incentive improved questionnaire return rates compared with no incentive, in both postal and online settings. Edwards et al. 94 reported that unconditional incentives (i.e. a reward given in advance as a goodwill gesture) led to superior response rates compared with conditional incentives (i.e. the promise of a reward on receipt of a questionnaire). However, there was significant heterogeneity among these results (p < 0.00001). Dillman96 and Singer et al. 97 also reported results that favoured unconditional over conditional incentives.

However, use of financial incentives is costly and studies may not have the funds or resources available to provide unconditional incentives. A conditional incentive, promised on receipt of a completed questionnaire, could be a more cost-effective means of enhancing retention. Both direct costs and resource costs could be reduced by rewarding only those participants who respond. It would minimise wastage created by the monetary incentives sent out to non-responding participants. In studies of incentive versus no incentive, similar gains are reported for both conditional and unconditional incentives. Khadjesari et al. 95 reported an increase of 9% when using a conditional incentive, whereas Dillman96 described increases of 7–14% in studies using unconditional incentives. Furthermore, Brueton et al. 93 reported two studies (297 participants) that suggested that unconditional monetary incentives provided no greater effect than conditional incentives (entry into a prize draw).

There is also a lack of evidence for the most effective incentive methods in the context of perinatal RCT follow-up. There is sensitivity around the population (parents of vulnerable infants) and often a lengthy time period between recruitment and longer-term outcomes that are captured at follow-up (in this case 2 years). Hardy et al. 98 successfully carried out a study of incentives at 1-year follow-up and reported it to be the only known study of incentives in this population. Kenyon et al. 99 investigated the effects of incentives in a perinatal trial; however, this was for follow-up when the children were aged 7 years. It is, therefore, prudent to use the available opportunities to narrow the evidence gap as to what works best. 100 Given the significant heterogeneity in results in previous trials94 and the potential serious effects of loss to follow-up, the need is still present to investigate which method of incentive provides the greatest return, particularly in perinatal RCTs.

The main objective of the SIFT study within a trial (SWAT) was to establish in the parents of preterm infants whether offering an unconditional incentive in advance (with the first mailing of a questionnaire) or promising an incentive (in the first mailing) on completion of a 2-year follow-up postal questionnaire (conditional) improves the response rate. This randomised controlled SWAT was nested within the SIFT: a multicentre RCT carried out in neonatal units in the UK and Republic of Ireland caring for very preterm or VLBW infants. The primary outcome of the SIFT was survival without moderate or severe disability at 24 months of age (CGA). This primary outcome was assessed by a questionnaire sent directly to parents (principally to infants’ mothers). Questionnaires were sent both by post and by a link to an online submission form via e-mail and SMS (short message service) message where these contact details were available. The SWAT investigated whether or not an unconditional incentive (a monetary voucher given before completion of a questionnaire) was more effective than a conditional incentive (the promise of a monetary voucher on receipt of a completed questionnaire).

Changes to the SIFT protocol

The SIFT protocol was amended in 2015 to account for the use of financial incentives. The amendment was approved by the NRES Committee and implemented in 2016. Funding was used from surplus funding for direct research nurse funds on the trial, as recruitment finished 11 months early.

Consent

No additional consent from parents was sought for this SWAT. The PIL and original consent process clearly indicated that the trial primary outcome was to be ascertained by a questionnaire at 24 months of age CGA.

Sample size

Recruitment for the SIFT was completed on 30 June 2015, with 2804 infants randomised. Given that there was an expected overall mortality rate of 5% for the population, it was estimated that 1250 of these infants would survive to the projected start date of the study (originally 1 December 2016) and that about 10% of the parents of these infants would be excluded from the follow-up owing to withdrawal or to a lack of information on contact details or survival status. This would result in approximately 1100 infants eligible for the SWAT, giving 550 infants per group.

It was estimated that the response rate with no incentives would be approximately 66%, based on past experience from the BOOST-II UK trial,101 which had a similar patient population and 2-year follow-up method.

Based on the most applicable studies that investigated incentives, it was anticipated that the addition of an incentive would result in an absolute increase of 10% in the response rate. 95,99 In total, 550 infants per group would allow detection of an absolute difference in response rate of around 7% at 90% power and a two-sided 5% level of significance.

Methods

Eligible participants were those who were recruited to the SIFT who were due to be sent a questionnaire at the age of 24 months (CGA). Participants were recruited to SIFT in 55 recruiting centres in the UK and the Republic of Ireland. Eligibility criteria for inclusion in SIFT have been described elsewhere. 1 Participants were traced to confirm survival and current residence. Where these details could not be ascertained, parents were not contacted. In addition, parents who had withdrawn consent to the 2-year follow-up were excluded.

Participants were allocated randomly to one of two groups:

1. After group – the first paper letter to parents would include a promise of an incentive (£15 gift voucher redeemable at high-street shops) after receipt of a completed form.

2. Before group – the first paper letter to parents would enclose the incentive (£15 gift voucher redeemable at high-street shops) before the receipt of a completed form.

Participants were randomised in a 1 : 1 allocation ratio by permuted block randomisation (using variable block sizes) and stratified by original SIFT allocation (faster feed increment/slower feed increment) and by singleton/multiple birth. Infants from multiple births were allocated to the same incentive group. Vouchers were allocated per questionnaire, so parents of multiple births received a voucher for each infant.

The SIFT office staff at the NPEU CTU were aware of participant allocation owing to the nature of the interventions and the practicalities involved in sending out the letters and the vouchers.

The incentive was a high-street shop voucher valued at £15 (€15 for participants recruited in the Republic of Ireland), which was sent via post. Reminder letters in both groups mentioned the incentives. Letters to those allocated to the after group reiterated the promise of an incentive. Letters to those allocated to the before group tactfully mentioned the incentive that was sent with the first letter. Parents were also contacted via text and/or e-mail to give reminders during the follow-up, although these contacts were not included as part of the analysis of the number of reminders sent. All of the parents were provided with an option of completing the questionnaire online or, as a last resort, via the telephone.

The incentives SWAT was implemented midway through the SIFT follow-up. All SIFT participants who had returned their follow-up questionnaire prior to the incentives study being implemented were sent a £15 voucher out of courtesy to ensure that there was fairness.

Outcomes

The primary outcome was the rate of questionnaire return, defined as receipt of a completed or partially completed questionnaire at the SIFT office. A questionnaire was considered completed or partially completed if the first three out of the five subsections of the questionnaire were completed (as these sections were applicable to the derivation of the primary outcome for the main SIFT).

The secondary outcomes included the:

• primary method of completion (paper, online or telephone)

• total cost

• number of reminders.

The total cost included the postage, receipt of material via prepaid Freepost, cost of envelopes, supplementary materials (e.g. sticker sets sent with questionnaires for infants to play with) and value of gift vouchers. It did not include Freepost licence fee, printing, telephone calls and trial staff time. All costs for participants were calculated in GBP. The cost of the €15 vouchers that were sent to the participants in the Republic of Ireland was converted to GBP using the exchange rate (via xe.com) on 10 May 2017: the date of the invoice for these vouchers. Costs for these participants also included the higher air mail postage fees.

Statistical analysis

The baseline demographic information was summarised by randomised group using frequency counts and percentages for categorical data, means and SDs for normally distributed continuous data or medians with IQRs for other continuous data.

The comparative analysis entailed calculating the absolute difference in the proportion responding with the corresponding 95% CI and the difference in mean cost (plus 95% CI). In addition, the cost in GBP per 1% increase in response rate was calculated factoring in administration costs such as the number of reminder letters, as well as the monetary value of the incentive. For other outcomes relating to the method of completion and the reminder letters, a similar strategy was used based on the distributions/type of data collected.

The principal comparison was the incentivised before group versus the incentivised after group.

The prespecified subgroup analysis examined the consistency of effect of the timing of the incentive for SIFT original allocation (slower feed increment vs. faster feed increment) and singleton versus multiple births using the statistical test of interaction.

The prespecified exploratory analysis examined the response rate in the period prior to the incentives study starting, during the incentives study and overall (i.e. irrespective of incentive group allocation) with a 95% CI. In addition, an analysis exploring ‘regional’ variation was performed.

No adjustment was planned for multiple testing as this SWAT involves a very small number of focused hypothesis tests.

Results

Participant flow and baseline characteristics

In total, 923 infants were randomised to the SWAT (799 women). A total of 459 infants were allocated to receive the incentive before completion. Out of these, three were not sent questionnaires because addresses and survival status could not be confirmed. A total of 464 infants were allocated to receive the incentive after completion. Out of these, 11 were not sent questionnaires because addresses and survival status could not be confirmed and two of the infants were randomised in error. In these two cases, both were later found to have died after they were randomised to the SWAT. All 923 infants were included in analysis (Figure 7). Baseline comparability was balanced across the infant and maternal characteristics at trial entry (Table 14).

FIGURE 7.

Flow of participants through the incentives SWAT. a, Included in the analysis where data are available.

TABLE 14 - Infant and maternal characteristics at trial entry

b.p.m., beats per minute.

Minimisation factor.

Sometimes only one infant from a multiple pregnancy met the inclusion criteria and was recruited.

The number of infants from multiple pregnancies where the other fetuses were aborted, miscarried or stillborn.

The number of infants who were one of twins.

The number of infants who were one of triplets.

Unless otherwise stated the table gives the percentages of infants with data in that group of the trial who had (or whose mother had) the stated characteristic.

Characteristic	Before group (N = 459)	After group (N = 464)
Number of centres, n	51	51
Allocated to faster group in SIFT,a n/N (%)	226/459 (49.2)	231/464 (49.8)
Male sex, n/N (%)	229/459 (49.9)	257/463 (55.5)
Missing, n	0	1
Infant age at randomisation (days)
Median (IQR)	4 (3–6)	4 (3–6)
Birthweight of < 10th centile for gestational age, n/N (%)	106/459 (23.1)	77/463 (16.6)
Missing, n	0	1
Gestation at delivery (weeks)
Median (IQR)	29 (27–31)	29 (27–30)
23+0 to 23+6, n/N (%)	4/459 (0.9)	7/464 (1.5)
24+0 to 24+6, n/N (%)	21/459 (4.6)	26/464 (5.6)
25+0 to 25+6, n/N (%)	31/459 (6.8)	44/464 (9.5)
26+0 to 27+6, n/N (%)	101/459 (22.0)	107/464 (23.1)
28+0 to 29+6, n/N (%)	129/459 (28.1)	117/464 (25.2)
30+0 to 31+6, n/N (%)	146/459 (31.8)	145/464 (31.3)
32+0 to 36+6, n/N (%)	27/459 (5.9)	18/464 (3.9)
Birthweight (g)
Mean (SD)	1139.5 (331.8)	1131.8 (319.4)
< 500 g, n/N (%)	0/459 (0.0)	2/464 (0.4)
500 to 749 g, n/N (%)	59/459 (12.9)	57/464 (12.3)
750 to 999 g, n/N (%)	111/459 (24.2)	114/464 (24.6)
1000 to 1249 g, n/N (%)	124/459 (27.0)	114/464 (24.6)
1250 to 1499 g, n/N (%)	91/459 (19.8)	115/464 (24.8)
≥ 1500 g, n/N (%)	74/459 (16.1)	62/464 (13.4)
Infant heart rate > 100 b.p.m. at 5 minutes, n/N (%)	421/455 (92.5)	418/456 (91.7)
Missing, n	4	8
Infant temperature on admission (°C)
Mean (SD)	36.8 (0.7)	36.8 (0.8)
Missing, n	4	2
Infant worst base excess within the first 24 hours after birth (mEq/l)
Mean (SD)	–6.0 (3.9)	–6.1 (4.1)
Missing, n	8	10
Infant ventilated via endotracheal tube at randomisation, n/N (%)	90/458 (19.7)	107/463 (23.1)
Missing, n	1	1
Infant had absent or reversed end diastolic flow, n/N (%)	76/452 (16.8)	70/455 (15.4)
Missing, n	7	9
Mother’s age at randomisation (years)
Mean (SD)	30.7 (5.8)	31.2 (6.5)
Multiple pregnancy,a,b n/N (%)	144/459 (31.4)	139/464 (30.0)
Singles,c n	0	1
Twins,d n	136	116
Triplets,e n	8	22
Caesarean section delivery, n/N (%)	284/459 (61.9)	272/464 (58.6)
Membranes ruptured before labour, n/N (%)	167/453 (36.9)	159/460 (34.6)
Missing, n	6	4
Membranes ruptured > 24 hours before delivery, n/N (%)	119/454 (26.2)	104/458 (22.7)
Missing, n	5	6

Primary outcome

In total, 381 infants allocated to the before group had a questionnaire returned (out of 459, 83.0%) compared with 353 infants allocated to the after group (out of 464, 76.1%) (Table 15). An unconditional £15 incentive upfront (before group) led to a statistically significantly higher response rate than a conditional incentive (after group), with an absolute difference of 6.8% (95% CI 1.6% to 12.0%; p = 0.01), adjusted for stratification factors (trial allocation and single or multiple birth).

Secondary outcomes

Prespecified subgroup analysis

There was no evidence of a differential effect of the incentives strategy between the original SIFT allocations (faster feed increment/slower feed increment) or between single and multiple births. However, the response rate from parents of multiples was higher in the before group than the after group (see Table 16 and Figure 8), but the test of interaction is not statistically significant and the finding may be, in part, simply due to the increased amount (double or triple) received unconditionally.

FIGURE 8.

Subgroup analyses for response rate at 2 years.

Discussion

In this SWAT, an unconditional incentive (before group) produced a significantly more favourable response rate than a conditional incentive (after group), with an absolute increase of 6.8% (83.0% vs. 76.1%, respectively). This mirrors results from previous studies. 94,97 The overall response rate was significantly higher after the SWAT was implemented (79.5%, 95% CI 76.8% to 82.0%) than before it was implemented (72.1%, 95% CI 69.9% to 74.1%).

The majority of questionnaires were returned by post and this method saw the significant increase between groups. There was little difference between groups for the online or telephone-supplied questionnaires but there were relatively small numbers of responses overall via these two methods.

Both of the incentive methods are cost-effective in terms of increasing response rate. However, providing an unconditional incentive comes with an additional cost, in this case almost £3 per infant. The before group led to an absolute increase in response rate of 11%, at an average additional cost of £14.80 per infant. The after group led to an absolute increase in response rate of 4%, at an average additional cost of £11.84 per infant. Therefore, although the unconditional incentive is more expensive, it represents good value for money with a cost of £1.35 per infant for every 1% increase in response rate. This is in contrast to the conditional incentive that costs an additional £2.95 per infant for every 1% increase in response rate.

There was no evidence of a differential response rate between trial allocations or between single and multiple births; however, there was a greater response from parents of multiples in the before group. We could postulate that there is a cumulative factor involved here, as parents of multiples received greater upfront incentives (£30 for twins, £45 for triplets) than parents of singletons (£15 for one infant). Previous systematic reviews have reported that higher incentive amounts lead to higher response rates. 93

Strengths and limitations

This SWAT, nested within a perinatal RCT, contributes to the evidence base for methods of incentive in this context. This study was run efficiently and required minimal resource, as it was added to the existing SIFT study and was run without additional staff time. Delays in ethics approval and logistical management meant that the SWAT study did not begin until February 2017, which resulted in fewer infants due to receive their questionnaire and thus participate. Despite this, results showed a significant increase in the response rate in the group allocated an unconditional incentive (before group).

This study is limited to a particular study population: parents of infants born prematurely. It is also limited to the follow-up period of 2 years. Given the scarcity of studies in the context of perinatal RCTs, it is important to bear this in mind. Longer follow-up periods (e.g. at school age), different modes of data collection and assessment (e.g. clinical assessment) and other incentive amounts may produce different results.

Implications

This SWAT provides strong evidence that incentives can be a cost-effective strategy to maximise follow-up in perinatal RCTs. It also demonstrates that giving the incentive voucher before receiving the questionnaire is more effective than giving the incentive voucher after. Indeed, an unconditional incentive strategy, although more expensive than a conditional incentive strategy, pays dividends with a significantly higher response rate. This SWAT gives a clear indication of the potential benefit, which may be helpful to other triallists considering the same approach.

Summary of main findings

The SIFT shows that advancing enteral feed volumes at daily increments of 18 ml/kg versus 30 ml/kg does not affect survival without moderate or severe neurodevelopmental disability at 24 months CGA or affect the risks of LOS or NEC in very preterm or VLBW infants. Advancing feeds more quickly reduced the duration of parenteral nutrition by 2 days but was associated with an unexpected increase in the frequency of abnormal motor outcomes at 24 months CGA.

To our knowledge, the SIFT is the first study to assess the neurodevelopmental outcome in relation to the speed of increasing feeds. Although the composite primary outcome was not statistically different between the two groups, the unexpected finding of an observed increase in the risk of moderate or severe motor impairment in the faster feed increment group requires consideration. This makes it unclear which increment to adopt in clinical practice and was notable as we anticipated that those infants in the faster increment group would have better outcomes at 24 months CGA than those infants in the slower group, unless there was an increased risk of NEC that was not outweighed by a reduction in LOS for faster-fed infants. The result is more perplexing in that there were numerically fewer cases of LOS and NEC in the faster group. It is possible that this is a chance finding, but there are biologically plausible mechanisms that could explain it, such as increased cardiorespiratory events as a result of pressure on the diaphragm,102 impaired cerebral blood flow autoregulation103 or the inability to digest and absorb enteral nutrition. 104–107

The quality and power of the trial enhances the validity of the findings. Practices were used to limit bias, including central web-based randomisation for allocation concealment. We obtained a high rate of follow-up and assessed the trial cohort with intention-to-treat analyses based on a prespecified statistical analysis plan. The trial achieved recruitment of 2804 participants, as per protocol based on the a priori sample size estimation. This sample was inflated before the trial commenced to allow for correlation in the primary outcome between twins and multiple siblings who received the same intervention as their siblings. This approach was taken after consulting parents of multiple birth siblings and representative groups who raised concerns about feeding siblings at different rates.

At randomisation, demographic and prognostic characteristics were well balanced between the two groups, with a minimisation algorithm ensuring that there was balance for major known or anticipated prognostic indicators (i.e. hospital, multiple birth, gestational age ranges and birthweight of < 10th centile for gestational age) or potential confounding influences, such as recruiting site. Interim analyses by the trial’s independent DMC used strict criteria to minimise the chances of spurious findings due to data fluctuations before a sufficient sample size was achieved. Adherence to the allocated interventions was high, the incidence of protocol violations was low and primary outcome data were available for 89.8% of the trial cohort.

The PARCA-R is a parent-completed assessment of cognitive and language development at 24 months of age. 37 It has good concurrent validity and test–retest reliability and, using published cut-off scores for identifying preterm children at risk of developmental delay, it has excellent diagnostic utility (sensitivity and specificity of > 80%) for identifying those with moderate to severe delay, as determined by scores of < –2 SD on a gold-standard developmental test. 37 It is widely used both as an assessment in routine developmental follow-up (e.g. as recommended for use in neonatal follow-up in the NICE guideline108 for ‘developmental follow-up of children and young people born preterm’ in the UK) and as an outcome measure in observational studies and clinical trials. 37 The resources that are required to carry out gold-standard tests prohibit their use in large-scale studies, especially where there are large geographically dispersed cohorts as in SIFT. Thus, the PARCA-R provides a cost-efficient alternative to examiner-administered tests for assessing cognitive and language outcomes at 24 months of age. However, like many developmental screening tools, lower positive predictive values (PPVs) may result in over-referrals or false positives. In studies comparing the PARCA-R with a gold-standard test, PPVs have ranged from 23% to 65% in samples of children born preterm. As such, a higher rate of positive screens is found on the PARCA-R than the rate of children identified by an examiner using a standardised test. 37,109,110 Therefore, the proportion of children with moderate to severe developmental delay as measured by scores of < –2 SD is likely to be lower than observed in SIFT. However, studies have shown that preterm children with false-positive responses represent a high-risk group and have significantly poorer developmental outcomes than those with true-negative screens and, thus, have a higher risk of long-term intellectual disability. 37,110,111

Maximising developmental follow-up and minimising participant attrition is a key concern for researchers, given that selective loss to follow-up can bias results. 111,112 In the UK this is increasingly challenging and completion rates for questionnaire-based follow-up have recently been reported at approximately ≤ 60%. 44,113,114 Intensive efforts were therefore made to maximise response rates at 24 months of age CGA in SIFT, as detailed in Appendix 10, which shows the return rate of questionnaires alongside the timings of the interventions. The 72% response rate of completed, classifiable, SIFT questionnaires, with the addition of classifications of outcomes obtained from blinded end-point review of routine clinical follow-up data (11.2%), meant that we achieved an excellent overall follow-up rate of 83.2%, where a classification of disability was possible in survivors at 24 months of age.

Event rates for the primary and secondary outcomes were different from those we anticipated when estimating the sample size required for the trial. 2,33 There were fewer deaths than expected, which was probably a result of the timing of randomisation that took place in a median of 4 days of age for both groups, by which time most neonatal deaths have already occurred. 115 The rates of LOS and NEC were similar to previous studies; however, NEC occurred in 5.3% of SIFT participants compared with 6.95% of participants in previous studies and invasive (confirmed) infection was seen in 18.4% versus 18.8%, respectively. 13

To our knowledge, almost three times as many infants participated in SIFT than in all of the existing trials combined; this suggests that we were able to produce more precise estimates of effect size than those available before SIFT. Analysis of secondary outcomes demonstrated that the number of days to reach full milk feeds and the number of days of parenteral nutrition were reduced with faster increments. Although the 95% CI for the relative risk estimate for the primary outcome was not statistically significant, it excludes a > 1% reduction in risk and a ≥ 18% increase in risk for those fed at 30 ml/kg/day compared with those fed at 18 ml/kg/day.

These results substantially outweigh data from previous trials,24–32 as large numbers of high-risk infants were recruited, including 1020 extremely low-birthweight infants, 994 extremely preterm infants and 435 infants with absent or reversed end diastolic flow in the umbilical artery on antenatal Doppler studies.

Limitations

A limitation of the trial is that is was not blinded, as it would have been difficult to safely and completely blind caregivers and parents to the feed rate. This is unlikely to have influenced the ascertainment of the most important outcomes, which were reviewed by BERCs. It is possible that knowledge of allocation could alter clinician practice, for example stopping feeds more often or diagnosing suspected NEC in faster increment infants. We did, however, see fewer cases of NEC in the faster increment group, suggesting that this did not occur often.

Infants were a median age of 4 days at commencement of the intervention and some clinicians may have been less likely to enrol the highest-risk infants. The trial does not, therefore, allow conclusion about the safety of different feed advancement increments in the first few days after birth.

Subgroup analyses

The subgroup analyses were informative, as they did not show significant findings but, as with all such analyses, they are inherently underpowered. The one statistically significant finding in the subgroup analysis was evidence of excess adverse outcome in the small number of faster increment infants (18/30) who received formula milk alone compared with slower increment infants (12/40). Higher-risk infants (including those with abnormal Dopplers) did not do worse with faster feed increments (see Figures 3 and 4 and Table 4). As the size of the group that received formula alone comprised 93 infants only, with 24-month outcomes not being known in 23 infants and including 16 infants in the faster increment group, it is likely that survival without moderate or severe disability being lower in the faster group is a chance finding. Further analysis is planned given that 1666 infants received a mix of breast and formula milk. It is possible that a higher proportion or volume of breast milk could protect against a poor outcome through anti-infective or anti-inflammatory mechanisms. It was a limitation of the study that additional granularity was not used in this subgroup analysis.

In post hoc analyses, we did not show any evidence that absent or reversed end diastolic flow that was detected by antenatal Dopplers of the umbilical arteries increased the risk of NEC with the faster feed increment.

Cost analyses

Given the apparently worse motor outcome at 24 months CGA, increasing the milk feed volumes at a faster rate in very preterm or VLBW infants is not a cost-effective strategy. Although it may reduce the burden on scarce health-care resources in the short term, the cost–consequences of this strategy in the long term are likely to be too severe to recommend this clinical practice.

Qualitative analysis of parent views

This in-depth exploration of parents’ views on enrolling their infants into more than one RCT or study highlights the need for research teams to be aware of a range of factors when approaching parents about multiple studies on the NICU. Although parents felt that they had made good decisions and had few regrets, our study highlights that, for some, this may be motivated by guilt, a debt of gratitude or a lack of confidence to refuse. A deeper awareness of these factors would enable health-care professionals to support parents more fully through the process of enrolling their infants on multiple trials.

The study also suggests that careful consideration is needed when PILs are developed and emphasises the importance of involving patients, parents and/or members of the public in all aspects of research design. The study also highlights the confusing and ‘overwhelming’ nature of having a sick infant requiring intensive care and the additional confusion when they are approached to join research studies. Providing such parents with additional opportunities to further discuss and learn about the research objectives and procedures after they have consented and eliciting their continued assent for their infants’ participation in the study may be important mitigation.

Incentives trial

The incentives study was an important part of increasing the follow-up rate in the trial to ensure that the primary outcome was known in > 90% of cases after other data sources were also taken into account. The unconditional incentive given before receiving the questionnaire produced a more favourable response rate than a conditional incentive, with an absolute increase of 6.8% (83.0% vs. 76.1%, respectively). The overall response rate was significantly higher after the SWAT was implemented (79.5%, 95% CI 76.8% to 82.0%) than before it was implemented (72.1%, 95% CI 69.9% to 74.1%), suggesting that both methods of incentives can be an effective strategy in maximising return rates.

Applicability

The SIFT findings are likely to be applicable both in the UK and internationally. Participants were enrolled in 55 neonatal units across the UK and in the Republic of Ireland, which gave a broad geographical, social and ethnic representation. Many infants who were enrolled in a recruiting site were subsequently transferred to another neonatal unit of ongoing care, usually closer to home. Trial participation continued in another 78 neonatal units, which mirrored care pathways for preterm infants in managed clinical networks in the UK.

The trial population was representative of very preterm or VLBW infants who were cared for within health-care facilities in well-resourced health services and included a substantial proportion of extremely preterm infants (36%) and of infants with other putative risk factors for neonatal morbidity, such as prolonged rupture of maternal amniotic membranes (24%) and evidence of absent or reversed end diastolic flow in the umbilical artery (16%). Overall, 30.4% of participants acquired a microbiologically confirmed or clinically suspected late-onset infection and, in total, 18.4% had a microbiologically confirmed infection, which was consistent with rates reported from cohort studies and other RCTs. 18–20,22,24–32 Similarly, the incidence of NEC (5.3%) was similar to rates reported in large, population-based surveillance and cohort studies and RCTs. 72

Implications for practice

The SIFT does not conclusively support the routine use of faster or slower enteral feed increments at daily increments of 18 ml/kg versus 30 ml/kg. Neither rate affected the primary outcome of survival without moderate or severe neurodevelopmental disability at 24 months of age or the risk of LOS or NEC in very preterm or VLBW infants. Advancing feeds more quickly reduced the duration of parenteral nutrition use by 2 days but was associated with an unexpected increase in the frequency of abnormal motor outcomes at 24 months CGA. Clinicians reviewing these data will need to weigh up the relative importance of the secondary outcomes when deciding how fast to increase the milk feeds of infants in their care.

Implications for research

Further randomised trials and examination of feeding data sets (either existing or prospectively collected) may provide further information. Research efforts should continue to investigate the aetiology, epidemiology and pathogenesis of complications from faster and slower feeding, with a particular focus on neurodevelopmental outcomes in relation to milk-feeding practices. They should also aim to develop, refine and assess feeding interventions that may prevent or reduce adverse acute and long-term consequences for very preterm infants and their families. Assessing different incremental feeding increments may be of benefit, as different groups of infants may respond differently to different increments (e.g. extremely premature infants). It may also be worthwhile to assess different increments in other groups, as these increments may also be better for those exposed to associated risk factors such as formula milk or placental dysfunction. Further detailed analysis of the association between milk types and outcomes is planned using data from this trial. Interventions to improve parental understanding, experience and acceptance of multiple research studies also warrant study.

Patient and public involvement

The SIFT was facilitated by close engagement with infant and family representatives experienced in service-user representation. Bliss, the UK national charity for ‘babies born premature or sick’, and parents of children who had received neonatal intensive care contributed to development of trial materials (e.g. PILs and consent forms) and to training of research staff (e.g. in simulated ‘consent-seeking’ sessions). This assistance was extremely beneficial and an important part of the successful delivery of the study.

Funding and sponsorship

The SIFT was funded by the NIHR Health Technology Assessment programme (11/01/25) and sponsored by the University of Oxford. The funder provided advice and support and monitored study progress but did not have a role in study design or data collection, analysis and interpretation.

Independent Trial Steering Committee

The members of the TSC were Denis Azzopardi (chairperson), Alison Baum, Mike Bradburn, Kate Costeloe, Mark Dolman, Andrew Ewer, Fan Hutchison and Michael Millar.

Independent Data Monitoring Committee

The members of the DMC were Richard Cooke (chairperson), Steve Kempley and Andrea Marshall.

Contributions of authors

Jon Dorling (https://orcid.org/0000-0002-1691-3221) (Chief Investigator, Neonatologist) was responsible for data collection and management, study design, data interpretation and writing the report.

Oliver Hewer (https://orcid.org/0000-0002-6251-7174) (Trial Manager) was responsible for data collection and management and writing the report.

Madeleine Hurd (https://orcid.org/0000-0002-2797-2358) (Administrator and Data Co-ordinator) was responsible for data collection and management and writing the report.

Vasha Bari (https://orcid.org/0000-0001-8183-2455) (Administrator and Data Co-ordinator) was responsible for data collection and management.

Beth Bosiak (https://orcid.org/0000-0002-6856-879X) (Trial Manager) was responsible for data collection and management and study design.

Ursula Bowler (https://orcid.org/0000-0002-0100-0155) (Senior Trials Manager) was responsible for data collection and management and study design.

Andrew King (https://orcid.org/0000-0001-7175-2718) (Head of Trials Programming) was responsible for data collection and management.

Louise Linsell (https://orcid.org/0000-0003-3205-6511) (Lead Medical Statistician) was responsible for study design, data analysis and report writing.

David Murray (https://orcid.org/0000-0001-9010-2905) (Senior Trials Programmer) was responsible for data collection and management.

Omar Omar (https://orcid.org/0000-0001-7582-8337) (Trial Statistician) was responsible for study design and data analysis.

Christopher Partlett (https://orcid.org/0000-0001-5139-3412) (Trial Statistician) was responsible for data analysis.

Catherine Rounding (https://orcid.org/0000-0002-1376-7572) (Trial Manager) was responsible for data collection and management and study design.

John Townend (https://orcid.org/0000-0001-5455-2562) (Trial Statistician) was responsible for data analysis.

Jane Abbott (https://orcid.org/0000-0002-5847-6824) (PPI representative) was responsible for study design.

Janet Berrington (https://orcid.org/0000-0002-6185-2843) (Co-investigator, Neonatologist) was responsible for study design and data interpretation.

Elaine Boyle (https://orcid.org/0000-0002-5038-3148) (Co-investigator, Neonatologist) was responsible for study design.

Nicholas Embleton (https://orcid.org/0000-0003-3750-5566) (Co-investigator, Neonatologist) was responsible for study design and data interpretation.

Samantha Johnson (https://orcid.org/0000-0001-8963-7881) (Co-investigator, Developmental Psychologist) was responsible for study design and data interpretation.

Alison Leaf (https://orcid.org/0000-0002-6068-7188) (Co-investigator, Neonatologist) was responsible for study design and data interpretation.

Kenny McCormick (https://orcid.org/0000-0001-8378-2690) (Co-investigator, Neonatologist) was responsible for study design and data interpretation.

William McGuire (https://orcid.org/0000-0001-8572-3467) (Co-investigator, Neonatologist) was responsible for data collection and management, study design, data interpretation and writing the report.

Mehali Patel (https://orcid.org/0000-0001-6824-0551) (Patient and public involvement representative) was responsible for study design and data interpretation.

Tracy Roberts (https://orcid.org/0000-0002-0624-0537) (Co-investigator, Health Economist) was responsible for study design and data interpretation.

Ben Stenson (https://orcid.org/0000-0003-2645-7458) (Co-investigator, Neonatologist) was responsible for study design, data interpretation and writing the report.

Warda Tahir (https://orcid.org/0000-0002-4410-1757) (Health Economist) was responsible for data collection and management, data analysis, data interpretation and writing the report.

Mark Monahan (https://orcid.org/0000-0002-1175-9421) (Health Economist) was responsible for data collection and management, data analysis, data interpretation and writing the report.

Judy Richards (https://orcid.org/0000-0003-1287-5923) (Sociologist) was responsible for data collection and management, data analysis, data interpretation and writing the report.

Judith Rankin (https://orcid.org/0000-0001-5355-454X) (Epidemiologist) was responsible for data collection and management, study design, data analysis and data interpretation.

Edmund Juszczak (https://orcid.org/0000-0001-5500-2247) (NPEU CTU Director) was responsible for data collection and management, study design, data analysis, data interpretation and writing the report.

All of the authors approved the final draft of the manuscript.

Publications

Dorling J, Abbott J, Berrington J, Bosiak B, Bowler U, Boyle E, et al. Controlled trial of two incremental milk-feeding rates in preterm infants. N Engl J Med 2019;381:1434–43.

Tahir W, Monahan M, Dorling J, Hewer O, Bowler U, Linsell L, et al. Economic evaluation alongside the Speed of Increasing milk Feeds Trial (SIFT) [published online ahead of print April 2 2020]. Arch Dis Child 2020.

Data-sharing statement

All data requests should be submitted to the corresponding author for consideration. Please note exclusive use will be retained until the publication of major outputs. Access to anonymised data may be granted following review.

Patient data

This work uses data provided by patients and collected by the NHS as part of their care and support. Using patient data is vital to improve health and care for everyone. There is huge potential to make better use of information from people’s patient records, to understand more about disease, develop new treatments, monitor safety, and plan NHS services. Patient data should be kept safe and secure, to protect everyone’s privacy, and it’s important that there are safeguards to make sure that it is stored and used responsibly. Everyone should be able to find out about how patient data are used. #datasaveslives You can find out more about the background to this citation here: https://understandingpatientdata.org.uk/data-citation.

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 and Social Care. 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 and Social Care.

Review conventions

The four components will be defined as follows:

1. Moderate/severe vision impairment – reduced vision uncorrected with aids, blind in one eye with good vision in the contralateral eye or is blind or can perceive light only.

2. Moderate/severe hearing impairment – hearing loss corrected with aids, some hearing but loss not corrected by aids or deaf.

3. Moderate/severe gross motor impairment – unable to walk independently or unable to sit independently.

4. Moderate/severe cognitive impairment – child’s development is between 6 and 12 months behind corrected age, child’s development > 12 months behind corrected age, child has fewer than five meaningful words, vocalisations or signs or child is unable to understand words or signs.

These definitions are taken from the protocol apart from the definition of moderate/severe cognitive impairment. In the protocol it was stated that moderate/severe cognitive impairment would be assessed using the PARCA-R, a parent-reported measure of non-verbal cognitive and language development. Total PARCA-R scores of < 44 would be used to identify children with moderate/severe cognitive impairment. Participants whose data are reviewed will be unlikely to have a fully completed valid PARCA-R questionnaire.

If the child is classed as having moderate/severe impairment in one of the four domains above, then they will be classed as having moderate/severe disability in terms of the primary outcome.

If the only data available for the child are scores on the Bayley Scales of Infant and Toddler Development, third edition (Bayley-III), from an assessment performed between 22 and 28 months of age (CGA), these should be interpreted as follows:

• Standardised composite score of ≤ 84 (equivalent to a scaled score of ≤ 6) in any domain denotes moderate to severe impairment.

• If the child was able to complete the Bayley-III, particularly the cognitive and language scales, with scores in the normal range, it should be assumed that the child has mild/no visual impairment (even if this is not stated).

If Bayley-III scores are available in addition to clinical information, the scores should be interpreted in the light of that information, particularly where scores are close to the cut-off points for moderate/severe impairment listed above. Bayley-III is not the primary outcome measure and the cut-off points above are not formally agreed standards. Thus, reviewers should interpret in the light of other information where scores are close to the cut-off.

Where there is a detailed clinic letter that provides evidence of no concerns regarding developmental progress (i.e. cognitive, language, motor), it should be assumed that there is also no moderate to severe hearing impairment, unless otherwise stated.

Intensive care

High-dependency care

Special care

Serious adverse event

Adverse events are defined as serious if they:

• result in death

• are life-threatening

• require inpatient hospitalisation or prolongation of existing hospitalisation

• result in persistent or significant disability/incapacity

• are a congenital anomaly/birth defect.

The term ‘life-threatening’ refers to an event in which the participant was at risk of death at the time of the event; it does not refer to an event that hypothetically might have caused death if it were more severe.

The SAEs are to be reported from randomisation to discharge home.

Expected serious adverse events

The following are SAEs that could be reasonably expected to occur in this population of infants during the course of the trial or form part of the outcome data. They do not require reporting by the SIFT co-ordinating centre as SAEs:

• death (unless unexpected in this population)

• NEC or focal intestinal perforation

• microbiologically confirmed or clinically suspected late-onset invasive infection

• bronchopulmonary dysplasia (mechanical ventilator support or supplemental oxygen at 36 weeks’ postmenstrual age)

• intracranial abnormality (haemorrhage, parenchymal infarction or white matter damage) on cranial ultrasound scan or other imaging

• pulmonary haemorrhage

• patent ductus arteriosus requiring treatment (non-steroidal anti-inflammatory drugs or surgery)

• retinopathy of prematurity.