The undesirable consequences of controlling for birth weight in perinatal epidemiological studies

Objective - To compare the effects of controlling for birth weight with those of controlling for gestational age at delivery in perinatal epidemiological studies using two examples. Setting - Western Australia. Subjects - Population data: all white births born at 20-46 weeks of gestation in Western Australia during 1985-91 inclusive (n = 147 564). Example 1: All Western Australian births from 1980-89 born either at 33-36 weeks inclusive (n = 13 607), or born with a birth weight of 2050-2900 g (n = 34 107). Example 2: 160 singleton cases of spastic cerebral palsy born to white mothers in Western Australia from 1975-80 and whose gestational age was known, compared with (a) 480 controls individually matched for gender and birth weight and (b) singletons with known gestational age liveborn to white mothers in Western Australia from 1980-81, or 1979-82 if < 30 weeks' gestational age at birth (n = 32 031). Measurements and main results - The risks of cerebral palsy associated with two separate exposures in groups defined by birth weight were compared with those in groups defined by gestational age, The origin of the differences are explained using total population data. The estimates of risk differ when exposure and outcome are both associated with appropriateness of fetal growth. The difference varied with gestational age, being greatest in the moderately preterm (33-36 weeks' gestation). Conclusion - Epidemiological studies in which appropriateness of fetal growth is an important variable should be based on gestational age at birth rather than birth weight, whatever the neonatal size or maturity.