Efficiency of sampling: birthweight and gestational age distributions in two cohorts, < 31 weeks and 500-1499 grams.

ABSTRACT

We studied the efficiency of two common sampling strategies used to assemble cohorts to study the long-term problems of preterm infants infants with birthweights of 500-1499 g, and infants with gestational ages (GA) of < 31 weeks. Birthweight, GA and 2-year outcome data from a population based study of infants < 2001 g, the Central New Jersey Brain Hemorrhage Study (NBH), were used to define the birthweight and GA distributions, at enrollment and at the age of 2 years, of overlapping subsets infants 500-1499 g (n = 599) and infants < 31 weeks of age (n = 522). Using frequencies from the NBH study, we estimated that 1000 infants of 500-1499 g enrolled at birth would produce 712 infants at the age of 2 years, 498 below 31 weeks and 214 above. Enrolling 1000 infants < 31 weeks would produce a cohort of 697 infants at the age of 2, all of whom were < 31 weeks. Neither sampling strategy maximised the statistical power to investigate the pathophysiological determinants of long-term outcomes associated with short GA. Both methods oversampled older GAs. A stratified sampling technique based on GA, designed to produce equal numbers of subjects at each week of GA, would improve statistical power to study long-term outcomes. As we move from descriptive to analytical studies of preterm infants, we need to devise efficient, GA-based, sampling strategies that maximise statistical power to test pathophysiological hypotheses.