Results

The results from the five cities are presented in anonymous form.

Data were retrieved on 1098 infants. No child of birth weight 1251 g or more had ⩾ stage 3 ROP. The population studied was therefore limited to infants of less than 1251 g birth weight. This resulted in a secondary cohort of 620 infants, of whom 421 (67.9%) survived to the first eye check. Table 1 shows the number of live born infants and survivors in this latter cohort. The results of an eye examination were unable to be traced for 22 (3.5%) infants out of this group of 620 following transfer from one of the study hospitals. None of these infants later presented to the local ophthalmology teams with visual problems, and it was assumed that they did not have ⩾ stage 3 ROP. These missing data were spread throughout the five cities. CRIB data were unable to be obtained for 131/620 (21.1%) infants, 118 (90%) of whom were from city 2. None of these infants had ⩾ stage 3 ROP.

Table 2 illustrates a number of risk factors present in the different populations. No differences were seen between the cities when comparing unadjusted rates of survival of infants < 26 weeks' gestation or < 751 g birth weight. The percentage of infants entering the study with a poor CRIB score (> 10) similarly showed no difference. It is interesting to note the low number of infants < 26 weeks' gestation and < 751 g birth weight in city 3. This finding was not significant.

The incidence of ⩾ stage 3 ROP per 1000 infants surviving in city 1 was 167 (95% CI 96 to 260). This was higher than in the other four cities. The excess of severe ROP reached statistical significance in comparisons with city 2 (24 (6 to 59)) and city 4 (16 (0 to 84)). In city 3, incidence of ⩾ stage 3 ROP was 23 (1 to 120), and in city 5 the incidence was 74 (21 to 179). The relative risk of ⩾ stage 3 ROP in city 1 compared with the other four cities was 5.5 (2.5 to 11.9). We felt that this post hoc comparison was justified because the anecdotal reports that led to this study related to a high incidence of ROP in city 1.

Tables 3 and 4 show the distribution of cases by birth weight and gestation respectively. The small numbers in each city show no pattern, but the aggregated data showing the proportion of infants affected show a clear link to birth weight and gestation.

The logistic regression model (CRIB) disclosed that the actual death rate for city 1 (270/1000 live births ⩽ 1250 g) was significantly lower than that predicted by the model (385/1000 live births ⩽ 1250 g (95% CI 339 to 431)). The actual death rates for cities 2, 3, and 4 were within the confidence intervals of the rate predicted by the CRIB score (city 2, 316/1000 live births ⩽ 1250 gv CRIB estimate 334 (95% CI 289 to 382); city 3, 328/1000 live births ⩽ 1250 g vCRIB estimate 280 (95% CI 193 to 368); city 4, 370/1000 live births ⩽ 1250 g v CRIB estimate 327 (95% CI 280 to 376)). The actual death rate for city 5 was significantly higher than that predicted by the model (430/1000 live births ⩽ 1250 g v CRIB estimate 344 (95% CI 266 to 423)).

Within CRIB, gestation has only a small weighting. As the risk of ROP is known to be closely associated with low gestation, it seemed appropriate to investigate whether variation in the distribution of gestation between the centres was responsible for the observed differences. Therefore a “new model” was derived to predict mortality using week of gestation as an additional variable to adjust the CRIB prediction. Although the overall “fit” of the model was improved, the pattern seen with CRIB alone (with fewer than expected deaths in city 1 and excess deaths in city 5) remained (city 1, 270/1000 live births ⩽ 1250 g v new model estimate 391 (95% CI 323 to 456); city 5, 430/1000 live births ⩽ 1250 g v new model estimate 344 (95% CI 280 to 409)).

Discussion

The results presented show a significant variation in the incidence of ⩾ stage 3 ROP. This was highest in city 1 where the finding was accompanied by a death rate that was significantly lower than that predicted by the CRIB score. Cities 2, 3, and 4 had actual death rates within the range predicted by the CRIB score and an incidence of severe ROP far lower than that of city 1. These data suggest that improved survival and the development of severe ROP are associated. City 5 has a worse than predicted death rate but an incidence of severe ROP closest to that of city 1, albeit not significantly different from any of the cities. The major neonatal unit serving city 5 has a policy of assessing on the neonatal unit and recording as a live birth even the most immature infants born alive. CRIB was not developed to adjust for such an effect and we believe this policy is responsible for the higher than expected mortality.

We did not identify any infant over 1250 g birth weight with ⩾ stage 3 ROP; however, this does not indicate that current recommendations for screening (all babies ⩽ 1500 g) are too wide. It would not be appropriate to draw such a conclusion from a study that was not designed to investigate incidence in relation to birth weight.

Existing studies in the literature do not present a clear picture of the major influences on the incidence of ROP. Only two have reported conventional incidence studies making point estimates in geographically defined populations.13 ,21 The first used admissions to neonatal intensive care units as a denominator, and the second used all deliveries in the relevant birthweight category. It is known that differences in attitude with regard to the “limits of viability” have played an important part in determining access to neonatal intensive care.11 Therefore it is impossible to assess the extent to which decisions about viability made by individual hospitals within the two study boundaries resulted in the most immature babies, with the highest potential for developing severe ROP, being either included or excluded.

Differences in attitudes to viability are likely to have been of far greater importance in those studies that compared rates of ROP and or ROP blindness over time7-9 and which have drawn different conclusions over whether improved survival does or does not account for observed changes in incidence. Interpretation of these studies is made more difficult by additional factors. Over time, an increasing proportion of immature babies have been offered intensive care, but in addition a greater percentage of more mature babies have avoided the need for such support.10 As a result, the populations being compared in these investigations will be fundamentally different over time. Of equal importance in studies that have sometimes compared populations extending over 30 years are the major changes in the nature of intensive life support. Given these dramatic changes, it is possible that particular treatments exerted a specific influence on the incidence of ROP, independently of survival per se.

It has also been suggested that ROP is more common when preterm infants receive their care away from major centres.13 ,14 This concept may similarly not be sound. Population based studies in one region of the UK indicate that, currently, infants ⩽ 28 weeks' gestation transferred postnatally have a better outcome (corrected for disease severity) than similar infants whose care is entirely within one unit, small or large.22 This selection of infants may have an important influence on observed rates of ROP when transferred infants are analysed as part of the whole group cared for in the major centre(s) within a particular study.

In an attempt to deal with these various confounding issues, our study was based on the same 12 month period in all the units, and at a time when modern neonatal intensive care practice was well established and in general terms standardised in each of the participating centres. In addition, the comparisons were standardised in terms of survival using CRIB to adjust for case mix.

However, there are shortcomings within the approach we have taken. The study is limited by the small numbers of cases of severe ROP in cities 2, 3, 4, and 5 (10 in total) and by the fact that data are from one year only. We have used the CRIB score as a measure of initial disease severity and this scoring system was derived from infants who were generally of greater gestational age than infants in this study. We have made the assumption that basic differences in the population, such as levels of deprivation, will be manifest, and corrected for, through the severity of illness. Specific ethnic influences and variation in any genetic propensity to develop ROP have been ignored.21 ,23 ,24 In spite of these caveats, we feel the data show an important association between survival and the incidence of ⩾ stage 3 ROP.

It is our belief that in any one setting it is a complex interplay between the presenting population and both the quantity and quality of intensive care available that determines the measured rate of ROP. Improved understanding of the pathogenesis of ROP supports the importance of cardiorespiratory stability.25 Observational studies confirm the association of physiological instability and the incidence of ROP.26 Therefore, even where neonatal intensive care facilities are well established, in the most immature infants, who are inherently unstable, the risk of ROP is high and will be manifested if they survive. Poor quality intensive care in such a setting may produce a range of effects depending on whether survival is jeopardised or not. In a situation in which the population is changing, with a more positive attitude being taken with regard to the care of immature infants, one would expect an increase in the incidence of ROP until both the change and the learning curve of the perinatal service is complete. For example, Gilbert et al 27 studying blindness from ROP in “middle income countries” in Eastern Europe, Asia, Latin America, and Africa reported that, in those countries with lower infant death rates, ROP accounted for a greater proportion of the cases of severe visual impairment or blindness. The influence of new approaches to treatment, with no obvious direct effect on the eye, is more complex to predict. Higgins et al 28 in a single centre study showed a decrease in severe ROP associated with antenatal dexamethasone administration. They did not, however, comment on whether the dexamethasone was associated with a concomitant increase in survival. Previously, the effect of antenatal corticosteroids on ROP was reported as inconclusive.29 It is likely that the measured effect of dexamethasone on ROP will reflect the extent to which improved survival alters the study population either directly or through a change in attitude to viability.

There seems to be no doubt that poor care of any newborn infant, for example inappropriate and uncontrolled use of oxygen, can lead to the production of ROP. However, for the most immature infants, with homoeostatic mechanisms that are not fully developed, the aggressive interventions necessary for survival may be sufficient to initiate ROP. Hence excellent care with improved odds for survival may in fact carry with it an increased risk for development of ROP, and it is possible that, with current therapeutic strategies, we cannot always prevent this disease and continue to improve survival. The development of new interventions that will be able to deal with this specific risk has become the new challenge.30