More Stringent Threshold Needed

To the Editor: Cameron S. Carter, M.D., and colleagues report a brain activation study of schizophrenic patients performing a modified version of the Stroop task, which they interpret as providing evidence of anterior cingulate dysfunction (1). However, their data may also be interpreted as disproving their hypothesis. This ambiguity may not be apparent to readers unfamiliar with neuroimaging methodology.

Dr. Carter et al. used a statistical threshold of p<0.01, uncorrected for multiple comparisons, on the basis that their analysis was hypothesis-led. What may not be apparent to those not conversant with statistical parametric mapping is that the usual significance threshold justifiable in these circumstances is p<0.001. A more lenient threshold may allow for false positives (2). The most stringent analysis would correct for multiple comparisons; however, biologically meaningful but statistically insignificant activations might then be excluded. The thresholds applied in the two figures contained in Dr. Carter et al.’s article vary, and the spatial extents of the activations shown at p<0.001 (figure 2) are limited. There is no focus in the anterior cingulate gyrus.

Dr. Carter et al. are entitled to defend their hypothesis, but I would argue that other investigators, adopting the more stringent threshold (p<0.001), might have reported such findings as negative (or perhaps as a trend). This point may have been addressed by peer review and might have been acknowledged in the discussion. As Brodie noted in the Journal, there is “an extraordinary peer review responsibility…unevenly assumed in the publication of functional imaging papers” (3, p. 145).

If such a carefully designed study in a large group of patients (in functional imaging terms) failed to demonstrate anterior cingulate dysfunction, then this might suggest that functional abnormalities, when they occur in schizophrenia, are state- and not trait-related. Dolan et al. reported such dysfunction in acutely psychotic, nonmedicated patients (4). Dr. Carter et al.’s patients may have been “too well” when studied: “all patients were mildly ill…outpatients” (1, p. 1674).

Other brain regions have been identified as showing state-related dysfunction in the acutely symptomatic psychotic patient (e.g., language areas in the presence of auditory hallucinations). I argue that Dr. Carter and his colleagues have shown that anterior cingulate dysfunction is another state-related finding, although no less informative for that.