Introduction

Top Abstract Introduction Methods Results Discussion References

The speciality of neonatal intensive care has developed over the past 30 years. This period has seen a rapid growth in provision of care and dramatic improvements in the survival of premature infants.¹ However, these developments have been accompanied by concern that the falling mortality may have been achieved at the cost of high rates of disability in survivors.

Information on late morbidity in survivors of neonatal intensive care is needed by several groups of people: by parents, so that they can understand the possible consequences of survival in their baby and take informed decisions about their child's care; by the clinical team, for sharing with parents, for evaluating their service, and for research; by commissioners of neonatal care and other services for children, so that they can make informed decisions and plans; and by the general public, so that they can take part in an informed debate on priorities in health care.

Although official reports since 1992 have highlighted the need for neonatal units and health authorities to collect information on later morbidity,^2-5 it remains generally unavailable on a population basis or outside specialist centres. However, there is broad agreement on the data that should be collected and on the features of systems to collect such data.⁶ Such systems should be simple, standardised between units, based on existing data collection systems (and hence be capable of being implemented at little additional expenditure), and capable of achieving high levels of ascertainment.⁶ We carried out a pragmatic cluster randomised controlled trial to compare two approaches to the collection and collation of information on preterm infants who had required neonatal intensive care and had survived to 2 years of age. Both systems had the potential to fulfil these criteria. One method was based on a parental assessment of the child's health at 2 years. The other method relied on collation and review of clinical information collected as part of routine service delivery.

	Methods

Top Abstract Introduction Methods Results Discussion References

We collected information on a group of babies at high risk who had been admitted to the neonatal units of the nine collaborating hospitals in the former Trent and Wessex NHS regions over one calendar year (1997). All units provided intensive care, but this was a bigger component of the workload of the large units. Babies were eligible for inclusion in the study if they were born at 32 completed weeks of gestation (that is, up to and including 32 weeks and 6 days) and if their mothers lived in the health authority in which the hospital was located. A hierarchical dating algorithm on the basis of last menstrual period, ultrasound scanning, and other clinical information was used to assess gestation.

Before the start of the study we randomised neonatal units to one of the two intervention arms. We used minimisation⁷ to ensure that the two arms of the trial were about equal by region, city (Leicester and Nottingham, which each had two participating neonatal units), and size of unit (Wessex had three small and two large units; all units in Trent were large). The assignment of units within each region was performed by a statistician who was not part of the study team using tables of random sampling numbers. The identity of neonatal units was concealed in sealed envelopes.

Information collected on all babies, regardless of intervention arm, was based on the Oxford minimum dataset.⁶ The Oxford dataset provides a framework for describing health status at 2 years of age (corrected for gestation at delivery) focusing particularly on major health problemsfor instance, if the child is blind or has severe developmental delay. It also includes basic demographic and clinical information about the child's early clinical management. All babies in the study were "flagged" with the NHS central registry, which allowed us to be aware of children who had moved or died and hence enabled us to avoid contacting parents whose child had died after discharge from the neonatal unit.

Interventions
The essence of both interventions was that a small amount of clerical time was funded to enable follow up of eligible babies by using a core protocol and standardised data collection materials but with considerable flexibility to tailor activity to local circumstances. The study was coordinated on a regional basis, with one coordinator in Trent and one in Wessex. To prevent the coordinators becoming part of the intervention, their role with clerks was restricted to training, developing the local system, and, thereafter, triggering requests for information.

Procedure
Although our primary aim was to test two different methods of data collection, we thought it was necessary to show that the items of information we collected were valid. To do this, we selected a 10% sample of children in each method by taking every child recorded as having a major health problem at 2 years and every fifth child who was reported as not having a major health problem, starting with births from 15 April 1997, until a quota of 47 children had been achieved. These children were visited at home within six weeks of their 2nd birthday (corrected age) by one of the study coordinators (either a health visitor or a neonatal nurse). The coordinator, who had not seen the information already recorded for the child at 2 years, reviewed the same areas of development by interview of the parents and assessment of the child. She then checked for any discrepancies between her information and that obtained in either intervention arm and attempted to obtain further information from the parents to resolve discrepancies.

	Results

Top Abstract Introduction Methods Results Discussion References

We recruited 236 infants to each arm of the study (table 1). There were no significant differences between the two arms in terms of clinical characteristics. Of those infants eligible for recruitment, two in each arm were not approached because of a failure of procedures. Three families refused to join the parent questionnaire arm of the study, and in two of these cases this seemed to be because of language difficulties. Seventeen families declined to take part in the community follow up arm. In most cases no reason was stated, but some families expressed concern that the study might provide information to social services.

Four children in the community follow up arm died between recruitment and a corrected age of 2 years. None of the children in the parental questionnaire arm of the study died. As outcome in these children was known we included them in the appropriate numerators and denominators.

Some information was obtained from parents of 214 (90.7%) children, while community clerks reported on 223 (94.5%). The 214 responses from parents were largely complete. In each of the nine fields of development covered by the Oxford minimum dataset, health status was described in no less than 210 cases. In contrast, community returns showed more variation, particularly in relation to motor development (193 complete) and communication (197 complete). As a result of these differences complete data were available for a far higher proportion of children in the parental arm (85.2%, 95% confidence interval 76.0% to 94.3% v 67.0%, 42.9% to 91.0%), although in 60 of the 65 cases in the community arm where data were missing this was limited to just one or two fields of development. Most of these infants were from just one centre.

Table 2 shows a summary comparison of the two methods. Percentages relate to ascertainment in relation to the 236 infants recruited to each arm. The proportion of returned questionnaires, with details of their completeness, is provided for each method with 95% confidence intervals, which are adjusted for the cluster design.

Accuracy
More of the parents' questionnaires had all fields completed compared with the community questionnaires (difference 18%, P=0.048). Of the 214 questionnaires returned by parents, only 51.4% of were returned, as requested, within six weeks of their child reaching a corrected age of 2 years. This compared with 71.2% in the community arm. Data from the community arm were often not current, with clerks having access only to information about the child from the last time he or she was seen. Comments about the child's abilities were based on information recorded before 18 months of corrected age in 28% of cases. Because of the timing of routine screening this was a particular problem in relation to vision. However, even after we excluded this item the proportion of variables reported on the basis of data before 18 months of corrected age was still 22%. Information was retrieved from various sources available to the clerks. However, about two thirds was obtained either by contacting the health visitor or by reviewing the records of earlier assessments held on the community database.

Validation
Study coordinators visited and assessed 47 children to determine whether health status had been correctly assigned. All 23 for whom information had been provided by parents were found to have been correctly allocated by parental report (eight had severe functional loss, 15 were normal). Of the 24 children assessed from their community records (10 with reported severe functional loss, 14 normal), five were found to have been incorrectly allocated by the community clerks. All five were children recorded as having severe functional loss when in fact they were normal. In three cases this occurred because the clerk completed the form with data about development that had been obtained when the child was much younger and the record had not been updated in relation to later progress. In two cases the records were ambiguous in terms of the child's health.

Financial analysis
The costs of implementing the two approaches were an average of £2271 per centre (range £1520-£3170) with the parental questionnaire and £3709 (£2430-£4047) with the community review. The variation by centre largely reflected the working practices of the clerks involved, with marked differences in the time spent on each case. The equivalent figures for average cost per case recruited were £37 (£25-£52) for the parental questionnaire and £61 (£40-£67) for the community review.

	Discussion

Top Abstract Introduction Methods Results Discussion References

Several reviews and inquiries have commented on the importance of information about long term health outcomes for children who receive neonatal intensive care. ^{3 4 14} While this is self evident in relation to the children concerned, the comments have generally been based on the wider importance of such information. "Quality care" in relation to neonatal care should equate with high rates of survival among children who need this type of support and who then go on to function normally in society. This type of information would clearly be of interest to parents, who want to know not just if their premature infant will survive but the chances that he or she will be normal. Community services and specialist education provision could be delivered much more efficiently if planning was informed by accurate and ongoing data about the health status of the target population. Despite this clear need for information, apart from isolated cohort studies funded as research, it is simply not available in the United Kingdom or indeed in much of the developed world.

Past problems
The failure to make progress in this area seems to result from several factors. The initial focus of neonatal intensive care was on improving survival. Therefore, although outcome was seen as important, procedures to gain these data were not established in a systematic fashion. Where cohorts of infants have been reviewed in later childhood the approach has generally been to use a detailed series of tests to look for minor as well as major variations from normal. These types of assessment require the use of trained professionals and hence are costly.¹⁵ The community services in the United Kingdom perform health surveillance as part of their core activity and collect information about the health status of all children.¹⁶ Currently there is no routine system for identifying subgroups, such as infants who have required intensive care, to allow their data to be abstracted. More importantly there is no national protocol with regard to how and when children are reviewed by community services. Similarly the approach to documentation and recording varies widely.⁸

New opportunities
We wanted to identify a simple method of obtaining data about the later health status of this group of children, who are at high risk of long term neurodevelopmental problems (although the methods should apply to any other groupfor instance, those below a certain birth weight). We chose a corrected age of 2 years as we thought this would provide feedback to clinicians in a time frame that was still relevant to their practice. The level of detail was selected to be informative for all interested groups while being cheap to collect.

Financial context
Given the financial constraints of the NHS it is often difficult to convince commissioners of health care that expenditure on obtaining follow up data is justified. The cost of neonatal intensive care is around £1000 a day and about 1.5% of births in the United Kingdom are at 32 weeks' gestation. The additional costs associated with the methods used in this study (highest estimate £67 per case) are trivial by comparison and might lead to information on the success or otherwise of neonatal care, which at present can be measured only by early mortality.