The original trial

The methods have been fully described in the original trial reports. In brief, during the recruitment period, November 1993 through March 2001, women with fetal growth restriction between 24-36 completed weeks, where an umbilical artery Doppler waveform had been recorded, and the responsible clinician was uncertain whether to deliver the baby immediately, were randomly allocated to either “deliver now” or “defer delivery” until it could safely be delayed no longer. Mode of delivery and monitoring strategies for the defer delivery group was left up to the attending obstetrician.

The current study

Children delivered to mothers in 5 of the original recruiting countries (Germany, The Netherlands, Italy, Slovenia, and United Kingdom) were followed up. In the United Kingdom, children were flagged by the Office of National Statistics, which provided the name of the family’s current general practitioner. The practice was contacted to confirm its current contact details. We maintained contact with the families with annual birthday and Christmas cards, enclosing a change-of-address card. Participants were contacted by mail or telephone and invited to bring the child to an appointment with the psychologist at their school. In The Netherlands, a similar tracing system was used. In Germany, Italy, and Slovenia we relied on the knowledge of the pediatrician who had initially cared for the baby. All parents gave written informed consent, and the study was approved by the North West Region (United Kingdom) Multicenter Research Ethics Committee and the appropriate local committees in the other countries. No attempt was made to follow up participants from Belgium, Cyprus, Czech Republic, Hungary, Greece, Poland, Portugal, and Saudi Arabia. This was because of the small numbers of participants in each country and the difficulty in identifying participants and skilled assessors.

The original participants had been recruited over a 6-year period, but the present evaluation was carried out over 3 years. As far as possible therefore, the earlier recruits were evaluated first to minimize the age range.

In the United Kingdom, children were evaluated at school by 1 of 3 psychologists during a single day using a series of directly administered tests (cognition, language, motor) and parent report instruments (behavior and health utility). Teachers also rated children’s behavior and completed a comprehensive rating scale of academic attainment. This design for data collection has been previously validated. In other countries, validated translations and adaptations of the following tests were administered, and assessments were carried out in a variety of settings (clinic/home/school).

The Kaufman Assessment Battery for Children (ABC) is a well-validated assessment of populations of preterm children.

The mental processing composite (MPC) is a global measure of the child’s cognitive ability or generalized IQ (age standardized to mean: 100; SD: 15) and comprises 2 subareas of sequential and simultaneous cognitive processing.

The Clinical Evaluation of Language Fundamentals–Revised (CELF-R) [UK] is a standardized test of language that complements the Kaufman, which evaluates nonlanguage-based cognition. Four CELF-R subtests were selected covering receptive (2 subtests) and expressive (2 subtests) language. Each subtest was standardized to give a mean of 10, SD 3, and a high score representing better language skills. This was only delivered in English-speaking countries.

The Movement-ABC comprises 8 tests of motor impairment grouped as manual dexterity, ball skills, and dynamic balance; these are summed to give an overall impairment score, with higher scores representing greater impairment.

Teachers rated the scholastic performance of the index and comparison children against the national expected level of attainment for a child of the same age in English, mathematics, science, technology, geography, history, and information technology. This was combined as a Total Academic Achievement Score.

Teachers and parents completed the respective versions of Strength and Difficulties Questionnaire from which are derived subscales for emotional symptoms, conduct problems, hyperactivity, peer problems, prosocial behavior, and overall difficulties. Teachers and parents also rated the impact of the behavior, including chronicity, distress, social impairment, and burden to others caused by child behavior. Additional items were included to assess the 2 dimensions of attentional and overactivity/impulsivity problems and school adaptation with the Connors test.

The Health Utilities Index comprises parental reports of the child’s health state in respect to hearing, vision, dexterity, and ambulation. These are summated to a utility score, where 1 corresponds to optimal health and 0 represents death. Negative scores are allowed for health states considered worse than death, so the full range of scores runs from –0 ^. 34 to 1 ^. 0. Apart from the CELF, which was only administered in English-speaking countries, all questionnaires and tests were identical in each country, apart from translation.

Data were double entered onto a database to avoid entry errors. After data checking and cleaning the database was merged with the original study database for analysis using STATA version 10 (StataCorp, College Station, TX).

Statistical analysis

The primary outcome was the Kaufman-ABC MPC scores for survivors. Where this outcome was missing for otherwise evaluated participants, a value was imputed as follows: if the child was too developmentally impaired to complete the assessment, a value of 40 (the lowest measurable value) was assigned. Otherwise, the predicted value from regressing the measured MPC values on gestational age and, if recorded, Griffiths developmental quotient at 2 years was assigned according to a prespecified analysis protocol. Values were imputed for 5 children: 3 who were too impaired to complete the assessment (1 severe mental retardation; 1 Prader-Willi syndrome; 1 autistic with severe learning difficulties); 1 whose blindness prevented completion; and 1 with only a teacher’s assessment at the long-term follow-up visit. No other values were imputed.

A secondary outcome was the composite endpoint of death or severe disability. For this purpose, severe disability was defined as ≥1 of the following outcomes: severe blindness, severe deafness, cerebral palsy, or Kaufman MPC <70 points.

The analysis strategy was fully described for the original trial. The primary analysis compared the randomized groups using classic regression to control for end-diastolic frequency (reversed, absent, reduced, normal) and gestational age at randomization. Estimates and their SEs from these classic analyses were used to update prespecified skeptical and enthusiastic prior distributions for Bayesian interpretation.

In brief, the Bayesian approach allows direct consideration of the probability of treatment superiority. Rather than interpret the trial data in isolation, the Bayesian analysis formally allows the trial data to modify an individual’s prior beliefs regarding relative treatment effect. For this trial obstetricians’ prior beliefs varied widely. We therefore, adopted stylized formulations of prior beliefs to reflect those who were enthusiastic for either treatment and those who were skeptical regarding the existence of any difference. The beliefs are presented as odds ratios with 95% confidence intervals. The trial data are then used to modify these beliefs into a posterior odds ratio with 95% confidence interval.

For the primary outcome the skeptical prior was that immediate delivery would be unlikely to alter MPC scores by >5 points in those of ≥31 completed weeks’ gestation, and by >10 points in those at earlier gestational ages. For the composite endpoint, the skeptical prior was that immediate delivery would be unlikely to more than double or halve the odds ratio.

Separate analyses for the Kaufman-ABC, Movement-ABC, and composite endpoint were undertaken for the United Kingdom only, and for all countries excluding Italy (because of the low follow-up rate). This long-term follow-up study was not originally planned in the trial for which 2-year outcomes were primary. Original sample size calculations were based on the 2-year assessments. The 95% credibility intervals demonstrate the range of effect size that is consistent with the data.