 |
|
 |
|
 |
 |
|
 |
 |
 |
 |
 |
 |
 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
LETTER TO THE EDITOR |
Robert M. Hamer, PhD
Departments of Psychiatry and Biostatistics
University of North Carolina
Chapel Hill, NC 27599-2719
To the Editor.—
We are writing to you about a recent article that appeared in Pediatrics.1 The authors of this very interesting article claimed that posttraumatic stress disorder (PTSD) "symptoms and cortisol levels at baseline are associated with changes in hippocampal volume over an ensuing 12- to 18-month interval" after a stressful life event. However, we believe that this strong claim is not supported by the data the authors presented.
The first problem is that the methods the authors used to measure hippocampal volume are of low precision. In all studies that use MRI to measure brain volume, experimental variance can be introduced during data acquisition (eg, patient motion, changes in scanner hardware or software, scanner field variation) and during data analysis (eg, partial-volume problems, voxel misclassification, manual delineation error).2 Among the 15 children who they evaluated, the authors acknowledged (in line 3 of "Results") that there was 1 child who showed an increase in right hippocampal volume that amounted to a full SD, or 
15%. This child was excluded from additional evaluation. However, the existence of 1 outlier this extreme suggests that the method of measuring hippocampal volume was relatively imprecise. This inference is consistent with the fact that the authors measured interrater reliability but did not report the results. If we hypothesize that the volume-measurement methods used were imprecise and that hippocampal volume of these subjects did not actually change over the time interval studied, what results would have been obtained? We would expect that hippocampal volume change scores would have fluctuated randomly and would not have been significantly different from 0. This is precisely the outcome shown in Table 1, in which change scores are all small and nonsignificant. The only way to compensate for measurement imprecision is to increase the size of the study population.2
|
The results shown suggest that, for a longitudinal study of hippocampal volume in adult patients and controls, a minimum sample size of 58 to 66 subjects is required for 80% statistical power, even if we assume that there is a 5% change in patient hippocampal volume. Our results suggest that the Carrion et al study1 had only 25% of the subjects needed for 80% statistical power, even if hippocampal volume had changed by 5%, which it did not (average change reported in this study was +2.9%). It could be argued that the authors actually presented a longitudinal 1-group study, which would require a sample size of only 16 to 18 subjects, but we think that such a study design is grossly inappropriate in subjects who are likely to be undergoing volumetric brain growth.
The third problem is that no data from healthy controls were presented, so we have no way of knowing what "normal" looks like by the methods that the authors used. This is a particularly serious error of omission in the case of children (the average age of subjects in their study was 10.4 years [range: 8–14 years]), in whom brain volume is likely to be changing.4 Furthermore, it would be a mistake to assume that hippocampal volume must be increasing simply because brain volume is still increasing in children under 12 years of age.5 In the absence of control data, we cannot assume that the reported decrease in hippocampal volume was abnormal. We note that it could be argued that analysis in this article was entirely focused on whether changes in hippocampal volume are correlated with PTSD symptoms, so that brain changes in healthy controls are irrelevant. Yet, such an analysis would require a sophisticated use of statistics.
A fourth problem is that there were several problematic uses of statistical tests. For example, there were 8 reported hippocampal measures (see their Table 2), and all 8 measures were tested for correlation with PTSD symptoms, with hyperarousal, and with cortisol levels. This means that 24 separate tests were performed without correction for multiple comparisons. If correction for multiple comparisons was performed, we think that none of the reported results would be significant. Also, Pearson correlation coefficients were used to compare hippocampal volume at baseline and at follow-up. There are several ways to determine the equivalence of multiple sets of measurements, or the extent to which 1 set of measurements can be substituted for the other. Pearson correlation coefficients are a particularly poor choice, because they only reflect the extent to which 2 sets of measurements are linearly related. More appropriate measures are intraclass or concordance correlations, because they reflect the extent to which measurements are fungible, or invariant in terms of mean and SD. For example, if 2 sets of brain-volume measurements are available, and all volumes in 1 set are exactly fivefold larger than those in the other set, the Pearson correlation would be 1.00, whereas a concordance correlation would be much lower, showing that the 2 sets of measures are not interchangeable. Last, an outlier was arbitrarily deleted from analysis when the statistical test used was not sensitive to outliers. Once a decision has been made to use Spearman correlations, there is no need to exclude outliers; all scores are turned into ranks, so a large change in absolute value has little impact.
A final problem with this article is that, although it was not rigorously done, it confirms 1 of 2 competing theories about how PTSD affects the hippocampus in children. As such, it may be given inordinate weight, although it is a weak study. An earlier longitudinal study of children that involved 18 subjects (rather than the 15 reported here) concluded that hippocampal volume did not differ between PTSD patients and controls at baseline, at follow-up, or over time.6 An earlier study by Carrion et al7 also failed to find a difference in hippocampal volume between pediatric patients with PTSD and controls. Among 37 adult trauma survivors who were imaged prospectively within 1 week of trauma, there was no significant difference between the subjects who did and did not develop PTSD.8 Given that at least 1 earlier study included both pediatric patients and controls and followed the patients for a longer time period,6 the weight of evidence suggests that PTSD is not associated with a change in hippocampal volume, at least among children.
We agree with Carrion et al that longitudinal studies are inherently stronger than cross-sectional studies and that the effects of PTSD on hippocampal volume in children can only be understood in the context of a strong longitudinal study. However, we do not believe that an underpowered, uncontrolled, and overanalyzed study is a step toward that goal.
ACKNOWLEDGMENTS
Dr Steen was supported by the National Alliance for Research on Schizophrenia and Depression as a Hofmann Trust Investigator.
REFERENCES
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
 |
||
|
||
Home | My Pediatrics | Journal Information | Current Issue | Past Issues | Subscriptions & Services | Contact Us Click here for faster international access © 2007 American Academy of Pediatrics. All rights reserved. |
||
|
||
