Objective: To evaluate the influence of confounding and sampling bias on the relationship between fetal growth restriction in a very-low-birthweight-defined cohort (VLBW, < or =1500 g) and bilateral spastic cerebral palsy (BSCP) at early school-age.
Methods: Three hundred twenty-four of 407 long-term survivors of a regional cohort of VLBW newborns were followed until age 6 years. We categorized as small for gestational age (SGA) all infants whose birthweight Z-score was below -2 relative to published reference values. Uni- and multivariable logistic regression models were fit to estimate the risk of BSCP associated with SGA in the total sample, in subsamples defined by gestational age, and in a gestational age-matched case-control sample.
Results: In the total sample, no child below 28 weeks was SGA, and no child above 32 weeks had an appropriate birthweight for gestational age (AGA). The prevalence of BSCP was 14% in AGA and 2% in SGA infants. In both uni- and multivariable logistic regression analyses of the total sample, SGA was associated with a prominently reduced risk of BSCP (odds ratios range from 0.1 to 0.2, all 95% confidence limits exclude 1.0). However, analyses performed in samples defined by different gestational age cutoffs (24--31 weeks, 28--31 weeks) and in a sample using three gestational age-matched controls per BSCP-case did not show a protection by growth restriction (odds ratios range from 0.8 to 2.2, all 95% confidence limits include 1.0).
Conclusions: In VLBW-defined samples, the apparent protective effect of SGA for BSCP can be explained, at least in part, by the highly skewed distribution of SGA over the available gestational age range. From this follows that study cohorts should be defined by gestational age and not by birthweight. In distorted samples like this one, even controlling for gestational age does not reduce the illusion of a reduced cerebral palsy risk for growth restricted infants. Only restriction of the sample by gestational age and/or matching for gestational age reveals the absence of this apparent protective effect.