Our major finding is that SNAP-II and SNAPPE-II predict death at the population level, not at the individual level, among infants born at extremely low gestational ages, even after adjusting for gestational age. Thus, illness-severity scores intended to predict death among the whole range of newborns admitted to an intensive care nursery appear to be applicable to the narrow gestational age range of 23–27 weeks.
SNAPs and gestational age
Because the SNAP-II distributions were not normally distributed, we have avoided means and standard deviations and have instead emphasized centiles of the distribution. The median (50th centile) SNAP-II, as well as the 25th and 75th centiles decreased progressively with increasing gestational age between 23 weeks and 26 weeks, when it appears to have leveled off (). This, as well as the sharp reduction in odds of dying when SNAP-II and SNAPPE-II are adjusted for gestational age, allows for the inference that SNAP-IIs and SNAPPE-IIs provide information about gestational age, and that they can function, in part, as surrogates for gestational age.
When adjustment is made for gestational age, the odds ratios for the arbitrary cutoffs of SNAP-II and SNAPPE-II are reduced. This indicates that the odds ratios were confounded by gestational age. In contrast, the odds ratios for the gestational-age-defined SNAP-II and SNAPPE-II cutoffs increase with adjustment for gestational age. This phenomenon reflects the supplemental information provided by the gestational age groups.
SNAPs, mortality, and adjustment for gestational age
The observations that the lower the gestational age, the higher the risk of death and the higher the SNAP-II and SNAPPE-II prompted us to adjust for gestational age when evaluating the associations between the SNAPs and the risk of death. We did this in two ways for all SNAP-II and SNAPPE-II cut-offs. For all comparisons, the more extreme the cutoff, the stronger the prediction of death.
SNAP-II and SNAPPE-II decline with increasing gestational age more rapidly between weeks 23 and 26, than between weeks 26 and 27. Thus, these two illness severity scores carry with them information about gestational age and gestational age related phenomena. Since low gestational age, as well as each illness severity score, conveys information about mortality risk, adjusting for gestational age diminishes the SNAP contribution to mortality.
When the illness severity score is classified in relation to gestational age peers, such as whether or not it is in the highest quartile or decile for gestational age, adjusting for gestational age tended to increase the SNAP contribution to mortality. This suggests that gestational age was not a confounder here, but was probably an effect modifier 8
Our comparing SNAP-IIs of infants to those of their peers within narrow gestational age strata, dichotomizing them based on this comparison, and then evaluating the total sample crudely approximates what would be seen if each child was assigned a Z score and then classified as above or below an arbitrary Z score. This approximation is limited by the use of an internal standard rather than the preferred external standard for a Z score. Despite limitations, though, our classifying infants as above or below the quartile or decile mark within each gestational age stratum provided valuable death-discriminating information.
Higher odds ratios with SNAPPE-II than with SNAP-II
Seven of the eight mortality odds ratios comparisons were higher for SNAPPE-II than for SNAP-II (). This is most likely a consequence of SNAPPE-II, but not SNAP-II, including information about intrauterine growth restriction, which is an independent predictor of mortality in preterm newborns 9, 10
The predictive values positive (predicting death) were much smaller than the predictive values negative (predicting survival) (). This is typical of most predictors.
The higher the proportion of infants with a SNAP-II of 30 or more at an institution, the higher the mortality rate (). Some of the variation among NICUs represents phenomena demonstrated in (at all gestational ages, the higher the SNAP-II, the higher the risk of death).
The distribution of gestational ages and SNAP-II among newborns at each hospital of birth. The numbers in each of the cells are row percents. The hospitals are arranged in ascending order of the proportion of infants born at 23 and 24 weeks.
We believe that some of the inter-institutional differences in mortality reflect different recruiting procedures. At some institutions, women admitted to the high-risk obstetrical service were enrolled before delivery, whereas at other institutions the child had to have a first ultrasound scan before the mother was asked to consent. Consequently, all of the very early deaths are included at prenatal-enrollment institutions, while none of the very early deaths are included at postnatal-enrollment institutions.
Although SNAP-II and SNAPPE-II include an item for repetitive seizures, in our sample of extremely low gestational age newborns, not one newborn had “multiple seizures confirmed or highly suspected.” No single component or group of components appeared to account for the institutional differences in the proportion of babies who had a SNAP-II of 30 or higher (). Thus, it is unlikely that medical care practices influenced SNAP-II or SNAPPE-II in this population of ELGANs.
Strengths and limitations
Samples defined by low birth weight tend to include intra-uterine growth restricted newborns who are gestationally older than others of the same or similar birth weights. When gestational age related phenomena are the focus, such as SNAP-II and SNAPPE-II, low birth weight samples can contribute to biased inferences 11
. We have avoided these problems by enrolling newborns on the basis of gestational age and not birth weight.
Because of its large size, our study has the power to identify odds ratios of less than 2.0. Nevertheless, this was not an issue because all the odds ratios in exceed this value.
We view the diversity of care characteristics among the 14 participating institutions to be an asset, and not a limitation. With 14 diverse institutions, our findings can be viewed as generalizable.
A recent study from the Neonatal Research Network has shown improved prediction of unfavorable outcome among preterm infants born at 22–25 weeks gestation when the gestational age variable is supplemented with information on the infant’s birthweight, sex, singleton status, and exposure to antenatal glucocorticoid 12
. A simple calculator based on these data is available on the Internet (www.nichd.nih.gov/neonatalestimates
). A formal comparison between this method and SNAP-II/SNAPPE-II is beyond the scope of this paper.
In summary, SNAP-II and SNAPPE-II predict death in our sample of infants born before the 28th week, even after adjusting for gestational age. Thus, we provide strong support for the claim that the death-discriminating information contained in SNAP-II and SNAPPE-II applies to the least mature.