点击显示 收起
1 From the Department of Preventive Medicine, Division of Biostatistics and Epidemiology, The University of Tennessee, Memphis College of Medicine, Memphis, and the Department of Dietetics and Nutrition, Schools of Allied Health, Medicine and Nursing, The University of Kansas Medical Center, Kansas City.
See corresponding article on page 120
2 Address reprint requests to SE Carlson, University of Kansas Medical Center, Department of Dietetics and Nutrition, School of Allied Health, 3901 Rainbow Boulevard, Kansas City, KS 66160-7600. E-mail: scarlson{at}kumc.edu.
Studies of preterm infants have uniformly shown higher visual function at some time during infancy with experimental formulas containing the n-3 fatty acid docosahexaenoic acid (DHA) than with conventional formulas containing -linolenic acid, the essential fatty acid precursor of DHA (1). In studies of term infants, however, the evidence for a favorable effect of higher DHA status on visual function is inconsistent (1). When the significance of findings appears to depend on the particular population selected, it is important to consider that differences among studies in variability of the outcome measure (eg, visual acuity) and the antecedent variable (eg, DHA status of the infant at enrollment) may be involved. Both can affect required sample sizes, required power to reject the null hypothesis, and generalizability of findings.
The report by Makrides et al (2), which compared the growth and visual acuity of term infants whose DHA status was varied by changing the ratios of dietary essential fatty acids in formula, provides the first evidence that demographic variables influence the visual function of these infants through some unknown mechanism. Makrides et al report that paternal smoking, head circumference at birth, postconceptual age at the time visual function was assessed, and paternal social status were significant, independent predictors of visual acuity among the term infants in the study population, thereby systematically increasing the variability of visual acuity. No previous studies of the effect of DHA status on visual function controlled for parental smoking status or head circumference at birth either statistically or by design (through stratification). Postconceptual age at assessment was controlled for in all of the studies of preterm infants but in only 2 of the studies of term infants. Perhaps not coincidentally, these 2 studies are among those showing a positive effect of feeding DHA on visual acuity in term infants (3, 4).
Another factor that increases the variability of an outcome measure is the number of testers. In multicenter trials, it is essential to have more than one tester. Single-center trials may also use more than one tester for practical reasons. Regardless, interrater reliabilities should be determined in advance of the study, and a high interrater reliability (ie, a correlation >0.9) should be a condition of combining data from multiple testers. Even a high correlation among testers does not preclude the need for increasing the numbers of infants enrolled, however. Unless the regression of one rater's scores on the other's falls along the line of identity (ie, has a slope of 1.0), the differences among raters will systematically inflate the variation in visual acuity, thereby potentially masking or enhancing the effect of the dietary intervention.
Only recently has the high degree of variability in the DHA status of infants at enrollment (the antecedent variable) been appreciated, although the initial evidence has been available for some time. Clandinin et al (5) showed that most intrauterine accumulation of brain DHA occurs in the last trimester, but also showed a large interindividual variability in accumulation. Subsequently, a 4-fold variation in plasma phospholipid DHA was reported in preterm (6) and term infants (7). In preterm infants, this variability was independent of DHA intake throughout the first year of life (6). At 6 wk of age, dietary DHA intake accounted for only 32.4% of the variability in apparent DHA status of term infants, whereas DHA concentrations in plasma phospholipids at birth accounted for 51.8% of the variability (7).
The variability in DHA status at enrollment is probably even greater among centers and within multicenter studies that include infants of mothers with very different cultural patterns of food intake. Variability in DHA status may also explain why studies of preterm infants have consistently shown benefits of DHA intake on visual function. Unlike term infants, very preterm infants are born with little brain DHA (5). Consequently, the effect of supplementing these infants with DHA would be expected to be larger (increasing statistical power) than in term infants, even if the variables discussed above influenced visual development. The potential for antecedent DHA status to influence the ultimate response could mask or enhance the effects of improving DHA status within a given study and contribute to inconsistency in the results among studies.
Because information is lacking on the underlying mechanisms determining DHA status in term infants and how such status may affect visual acuity and other outcome variables, one strategy is to direct investigations toward these researchable questions before initiating additional studies of dietary interventions. However, if an investigator does study the relation between diet and a developmental variable such as visual acuity, a stratified randomization scheme or covariate analysis can be used to reduce the apparently excessive variability in the outcome variable. For example, an investigator might stratify on the basis of infant DHA status and current smoking status before randomization.
Because the studies published to date did not collect data on these potentially important variables, we are left to guess what influence these variables might have had on reducing the power or biasing the results. At best, we can conclude that some of these confounders may have been influential enough to produce type II errors (accepting the null hypothesis in error).
In focusing on variables that may be hidden within the published studies, we have ignored known methodologic differences among the published studies of DHA supplementation and visual function. These differences (eg, DHA dose and interval of DHA supplementation) are presumed to affect both the amount and timing of DHA accumulation in the infant's central nervous system Consequently, they too could differentially influence the statistical power of a given study. However, these differences can be compared among studies.
A statistically nonsignificant finding in a study implies a small, and presumably clinically unimportant, effect on outcome. Such a conclusion may be reached in error if sources of variability are not recognized and controlled for by appropriate means. The observations of Makrides et al (2) suggest there may be excessive variability in many studies of DHA status and visual function in term infants. This variability has likely contributed to the inconsistency of results among studies and prevents us from drawing any valid conclusions about the ability of term infants to benefit from improved DHA status. In future studies in infants, investigators would do well to consider antecedent DHA status in relation to developmental outcome. In addition, it is hoped that the report of Makrides et al (2) will stimulate investigators to begin collecting data on the variables they found influential.
REFERENCES
Related articles in AJCN: