Page 88 - 16Neonatal Jaundice_compressed

Early prediction of serious hyperbilirubinaemia

all the cases were included, early jaundice (OR 7.3, 95% CI 2.8–19), gestational age per week
(OR 0.6, 95% CI 0.4–0.7), breastfeeding at discharge (OR 6.9, 95% CI 2.7–17.5), Asian race
(OR 3.1, 95% CI 1.5–6.3), bruising (OR 3.5, 95% CI 1.7–7.4) , cephalohaematoma (OR 3.2,
95% CI 1.1–9.2), and maternal age ≥ 25 years (OR 2.6, 95% CI 1.1–9.2) were all
independently associated with hyperbilirubinaemia. After excluding cases with early jaundice,
similar findings were reported, with two exceptions – history of jaundice in a newborn was
statistically significant in the second model and black race was not included in it as all the early
jaundice cases were black. A simple risk index was then developed by assigning points to the
risk factors (approximately equal to their OR in the second model) that were found to be
statistically significant after exclusion of early jaundice cases. The accuracy of the risk index in
predicting hyperbilirubinaemia was good (c = 0.85). With a threshold risk score > 10 points,
the likelihood ratio of babies having serum bilirubin levels ≥ 428 micromol/litre was 2.2 but it
increased to 18.8 when a score of > 20 points was used as the threshold. [EL II]
35
In the fifth study, from the USA, a risk index score for predicting hyperbilirubinaemia was
validated, and a subset of this index was combined with pre-discharge serum bilirubin measured
at < 48 hours for predicting subsequent hyperbilirubinaemia. To validate the risk index score in
predicting serum bilirubin ≥ 427 micromol/litre, 67 cases and 208 randomly sampled controls
were selected from a cohort of 53 997 babies using similar study design, case definitions and
9
selection criteria to the previous study. The baseline characteristics of the 1997–98 cohort
study group were similar to those of the previous (1995–96) cohort. After excluding family
history of jaundice, a modified risk index was developed and it showed accuracy in predicting
significant hyperbilirubinaemia (serum bilirubin levels ≥ 427 micromol/litre) with a c-statistic of
0.83 (95% CI 0.77 to 0.89). The results were similar to those from the previous study (c-statistic
0.84, 95% CI 0.79 to 0.89). In the second part of the study, the records of 5706 babies born in
the same setting over a period of 4 years and who had serum bilirubin measured at < 48 hours
of age were reviewed retrospectively. A partial clinical risk index was derived by deleting family
history of jaundice, breastfeeding and bruising, and by substituting scalp injury with
cephalohaematoma. The serum bilirubin levels measured at < 48 hours were classified into
age-specific percentile groups (< 40th centile, 40th to < 75th centile, 75th to < 95th centile,
≥ 95th centile), and then transformed into hour-specific z scores but subtracting the observed
value from the calculated median for that age and dividing by the calculated standard deviation.
Significant hyperbilirubinaemia was defined as maximum serum bilirubin levels
≥ 342 micromol/litre. Pre-discharge serum bilirubin levels expressed as hour-specific centiles
showed better accuracy for predicting hyperbilirubinaemia than the partial risk index score (c-
statistic 0.79 versus 0.69). Within each percentile category there was a five- to fifteen-fold
increase in the risk of hyperbilirubinaemia for those with a risk index score > 10 compared
with those with a score > 4. Transforming the pre-discharge serum bilirubin centiles into the
hour-specific z scores improved their predictive ability (c-statistic from 0.79 to 0.83), but the
best results were obtained by combining pre-discharge serum bilirubin z scores with the partial
risk index score (c-statistic 0.86). [EL II]
Another retrospective cohort study, conducted in a urban teaching hospital in the USA, 12;14 also
compared the predictive performance of combined clinical risk factor assessment and pre-
discharge serum bilirubin measurement. The study population (n = 899) and the methodology
has been described in detail in Chapter 3 on risk factors. The population was the same as that
used in a previous study but for this study it was restricted to the time interval when ≥ 75% of
36
babies had both samples collected. Out of 996 eligible babies, 899 (90%) were finally included.
Hospital records were reviewed retrospectively to collect information on risk factors. Their
association with hyperbilirubinaemia was explored by univariate analysis and a risk factor score
was derived from regression modelling using the factors independently associated with
significant hyperbilirubinaemia. The final risk factor model included birthweight, gestational age
< 38 weeks, oxytocin use during labour, vacuum delivery, breastfeeding, and combined breast-
and bottle-feeding. Pre-discharge serum bilirubin levels were expressed as risk zones on the
hour-specific bilirubin nomogram. Significant hyperbilirubinaemia (serum bilirubin > 95th
centile on the nomogram) was present in 98 of 899 (11%) babies. The predictive accuracy of
pre-discharge serum bilirubin risk zone (c-statistic 0.83, 95% CI 0.80 to 0.86) was better than
the clinical risk factor score (c-statistic 0.71, 95% CI 0.66 to 0.76). By decreasing the thresholds
of a positive test for the risk factor score (higher to lower score) and pre-discharge serum

59