Published online June 1, 2007
PEDIATRICS
Vol. 119
No. 6
June 2007, pp.
1251-1252
(doi:10.1542/peds.2007-0364)
Neonatal Brain Volumetric Studies: Regression Analysis and Interpretation
George T. Vasileiadis, MD, MSc
School of Human Development
University of Nottingham
Nottingham NG7 2UH, United Kingdom
Lawson Health Research Institute
Department of Imaging
London, Ontario, Canada N6A 4V2
To the Editor.—
Volumetric techniques are increasingly used to investigate the impact of injury or intervention on the neonatal brain. Parikh et al1 recently reported in an interesting article that postnatal dexamethasone therapy is followed by reduced cerebral tissue volumes. In view of relatively poor MRI signal contrast between gray and white matter in neonates, researchers overcome challenges in regard to image acquisition and segmentation techniques. However, there remains a big challenge in how best to analyze and interpret the volumes measured. When interpreting the results of neonatal brain volumetric studies, there is often a need to address possible confounding factors. The decision on what confounding factors/covariates to include in the regression model is critical for the conclusions of the study. Equally critical is the way that these possible confounding factors are being tested for and analyzed in the regression model.
With regard to the hypothesis tested, previous scientific evidence is usually a good way to start finding which possible covariates to test for. In a landmark article in 1998, Hüppi et al2 addressed clinical parameters that correlate with neonatal cerebral volumes. Moreover, depending on the selection criteria and hypotheses, different samples may display different significant covariates. The statistical significance of group differences on demographic and clinical characteristics may be of some help but is limited. Factors not significantly different between groups may well be proven statistically significant when entered in the regression model to test the primary hypothesis, and vice versa. Also, the influence of a covariate on the primary outcome measure may partially overlap with the influence of a different covariate, which makes the combination tested crucial.
In regard to the regression model, it is a good exploratory approach to enter and test the possible covariates initially 1 by 1. This can be followed by a stepwise regression that involves the more statistically significant ones. It is safer to use larger P values (eg, .1 or .2) for entry criteria in the stepwise regression than for exit (eg, .05 or .1). The covariates that have the most influence on the primary outcome (cumulative effect if >1) should stay in the model, and then the primary outcome measurements can be adjusted accordingly. It is important to note whether comparisons for the primary outcome were statistically significant before and after the regression analysis.
With respect to the above and in terms of cortical volume, along with postmenstrual age at scan, scaling effects and size differences between subjects and groups at scan could have an influence on neonatal cerebral volumes and be a significant covariate. Also, uncomplicated germinal matrix-intraventricular hemorrhage could be a significant covariate for the cortical volume.3 For the reader, understanding is benefited from adequate details on the demographic and clinical characteristics of the subjects. The presentation of the raw data and their statistics, before adjustments, is also essential for the interpretation of neonatal brain volume studies.
REFERENCES
- Parikh NA, Lasky RE, Kennedy KA, et al. Postnatal dexamethasone therapy and cerebral tissue volumes in extremely low birth weight infants.
Pediatrics. 2007;119
:265
–272[Abstract/Free Full Text]
- Hüppi PS, Warfield S, Kikinis R, et al. Quantitative magnetic resonance imaging of brain development in premature and mature newborns.
Ann Neurol. 1998;43
:224
–235[CrossRef][ISI][Medline]
- Vasileiadis GT, Gelman N, Han VKM, et al. Uncomplicated intraventricular hemorrhage is followed by reduced cortical volume at near-term age.
Pediatrics. 2004;114(3). Available at: www.pediatrics.org/cgi/content/full/114/3/e367
PEDIATRICS (ISSN 1098-4275). ©2007 by the American Academy of Pediatrics
