Results

Table 1 compares crude survival between small and large units for babies ⩽32 weeks gestation not involved in transfer of any kind, the method adopted in the 1987 study. In table 2 infants are shown categorised by birthweight, gestation, and whether they received ventilation. Among the whole group of infants ⩽32 weeks gestation mortality in the small units was significantly less (16.9%vs 11.5%; p<.001). Data from these two tables suggest that this effect occurs in infants ⩾27 weeks gestation; however, the tables also show that the group cared for in small units contained a far higher proportion of more mature infants. Straightforward statistical comparison of the two groups showed the population from large units to have been significantly smaller, more immature, given a lower 5 minute Apgar, more likely to be multiple births, more likely delivered by the breech and more likely to develop respiratory distress.

A similar crude comparison of mortality for infants: ⩽28 weeks gestation cared for in the two types of unit reveals a higher survival rate among infants whose care was entirely within the large units, but this difference was not significant.

Also shown in table 1 are CRIB data relating to the same two groups of babies (non- transfers ⩽32 weeks and non-transfers ⩽28 weeks) split between small and large units. In each case actual mortality falls within the 95% confidence interval of the CRIB estimate—that is, in both settings and in both gestation groups mortality falls within the acceptable range as defined by CRIB.

Tables 3 and 4 include information about actual and expected mortality for those babies involved in transfer. Again, there is no suggestion that any of the subgroups identified have a worse than predicted outcome, irrespective of whether the transfer occurred before or after delivery.

None of the above adequately quantifies the extent to which selection affects the work carried out in the different settings. For example, small units may choose to transfer all of their most immature infants if they appear viable and retain only those where death is considered inevitable; large units, by providing high risk pregnancy services, may attract particularly complex cases; both types of unit may adopt a policy of admitting all live born infants even if they are considered non-viable and may or may not offer those babies intensive care. The CRIB data, by quantifying disease severity, highlight phenomena which are almost certainly the end product of such selection. The whole population of infants ⩽32 weeks gestation cared for by the larger units had significantly higher disease severity scores than those in the smaller units (expected mortality 15.9% (14.3–17.6)vs 12.6% (11.3–14.1). The group of babies ⩽32 weeks gestation transferred had a significantly greater predicted mortality than those who received all their neonatal care in one unit (expected mortality 18.7% (17.1–20.3) vs14.1% (13.2–15.4)—as might be expected the process of transfer appeared to select for a population of babies who were more severely ill). This was particularly apparent among postnatal flying squad transfers. But babies ⩽28 weeks gestation had no significant difference in disease severity between those cared for in large and small units. In this group those transferred tended to be less severely ill compared with those cared for in a single unit, and to have a survival rate significantly better than predicted.

Discussion

The figures for survival corrected for initial disease severity are of prime importance. They indicate quite clearly that outcome was equally good for all babies ⩽32 weeks gestation, irrespective of the type of unit that provided their neonatal care. Taking the study population as a whole, there is evidence that the service is working in an integrated manner. The larger units, appropriately, attract a group of infants who are sicker than those whose care is entirely in the smaller units. The babies transferred for neonatal care tend to be sicker than those that normally receive treatment in the small units.

Focusing on babies: ⩽28 weeks gestation reveals a slightly different pattern. Here CRIB scores indicate that those infants not involved in transfer are similar, in terms of disease severity, in both types of unit and outcomes are equally good. Those babies that are transferred do not have a higher predicted mortality. This suggests that, among these infants, selection for transfer relies on different criteria; possibly, some infants are felt to be too sick or too immature to move from the smaller units.

In our 1987 study we identified worse outcome in terms of crude mortality for non-transferred infants ⩽28 weeks gestation when care had been delivered in one of the small units. In the current study a similar comparison of all non-transferred infants ⩽32 weeks gestation has shown a significantly better outcome for infants whose care was in the small units. In this current study CRIB puts in context the observed differences in mortality, something that was not possible as part of the 1987 investigation. However, we feel that the 1987 findings remain valid in relation to the service that existed at that time. Although we have not been able to apply retrospectively the CRIB score to the 1987 cohort, we know that descriptive variables associated with increased disease severity (presence of respiratory distress, lower birthweight, and lower 5 minute Apgar score) were significantly more common in the infants cared for in the large units at that time. For the 1987 findings to be invalid disease severity among babies in the smaller units would have to have been significantly higher than those in the larger centres, and this seems unlikely. These arguments show that a qualitative approach to assessing disease severity is limited in its usefulness and CRIB provides a much more objective framework to deal with these issues in future.

Given the improved levels of specialist medical and nursing input since 1987, it is plausible that differences in survival should have disappeared between both types of unit. But it is crucial to understand that these excellent results have not occurred by chance and that adequate resourcing of staff, equipment, and infrastructure are essential. The small units in Trent that undertake intensive care still rely on the large referral centres to help with peaks in demand, to provide subspecialty support, and for most specialist training. It is important, therefore, that the role of the large units within the current system is preserved to maintain the present co-operative working arrangements. We welcome the increased availability of good local care, but the current funding arrangements do not satisfactorily address issues of complexity of care, training, and service development. Dealing with these remains a challenge for the NHS.

Recent data from other parts of the developed world identify a different situation. Based on data from the United States in the early 1990s, Phibbs et al 8 noted that neonatal units with a high throughput of intensive care appeared to have lower mortality rates. However, their report contains important methodological differences relative to the data we present here, especially with regard to the measurement of disease severity, which in the US study relied heavily on diagnosis. There are also major differences in the US and UK health care systems, which limit the extent to which the study can be usefully considered in relation to neonatal care in the UK.

Phibbs et al identified a higher mortality for infants transferred for intensive care, consistent with previously published work9; this finding is often a feature of neonatal units’ annual reports. This poorer outcome has been variously ascribed to inadequate care before transfer or the process of transfer itself. Our data seem to indicate that, for babies ⩽32 weeks, the transferred infants simply represent a selected sicker group whose outcome is as predicted. To understand this effect more clearly, the CRIB score should be calculated at various points during the first 12 hours of life and the values obtained for infants involved in transfer compared with babies whose care is entirely in one unit. This approach differs from normal, in that CRIB is usually calculated just once on information related to the whole of the first 12 hours of life.

The use of CRIB remains controversial; some clinicians are concerned it does not adequately reflect complexity of care and clinical practice. Indeed, the group that originally developed CRIB have recently identified problems when CRIB is used to compare individual hospitals over short time spans.10 Their study, however, endorsed the type of approach taken here—that is, assessing groups of hospitals involving large numbers of infants over significant periods of time. We accept that the use of CRIB does not adequately reflect all aspects of policy. In particular, policies of admitting non-viable infants for terminal care and decisions about which infants to transfer—for example, whether to move the sickest or only those thought to have a reasonable chance of survival—distort the figures. The use of CRIB, as described here, permits, a broad assessment of policy in terms of service delivery and provides opportunities both to understand survival figures better and also to know how they might be improved.

We have not considered morbidity, which is clearly of major importance. Comparisons based on later outcomes will not be possible until a simple standardised method of review is more widely available across the UK.11