Retinopathy of prematurity: a risk factor analysis with univariate and multivariate statistics

Abstract

Seventy-three out of 639 ophthalmoscopically examined newborns (11.4%) showed varying degrees of retinopathy of prematurity (ROP). Three infants were blind (0.5%). Sixty-six patients were matched with controls according to gestational age, birth weight, birth date and neonatal unit. Using univariate statistics, the variables representing the extent of O2-exposure, elevated paO2, elevated paCO2, paCO2-fluctuations, acidosis, blood transfusions, and artificial ventilation were found to be significantly associated with ROP. In contrast, the variables representing hypocapnia, Hb-levels, parenteral fluid administration and fluid retention showed no correlation with ROP. When a multivariate statistical method was used, the variables representing gestational age, elevated paO2, elevated paCO2 and paCO2-fluctuations as well as the ones defining blood transfusions lost their association with ROP. Episodes of acidosis, multiple birth, birth weight, total hours with Fio2 greater than 0.4, and total duration of artificial ventilation remained in the regression, and hypocapnia gained a significant negative correlation with ROP. This comparative survey based on univariate and multivariate statistics demonstrates that significance levels of identically defined biological and therapeutic variables obtained from the same population of patients are to be interpreted with caution. This fact may also explain contrasting opinions on risk factors in the literature. Multivariate statistical analyses are useful for a relative comparison of significance within a given set of variables. Conclusions on a possible causal relationship as well as on prophylactic measures are not possible.