Nonsignificant But Suggestive Results?

To the Editor: I would like to comment on a study in the April issue by Stephen V. Faraone, Ph.D., et al. (1). It is surprising to me that a study that did not produce a single positive result could elicit such broad-sweeping conclusions from its authors. Is it really acceptable science to tout findings that are nonsignificant but suggestive? We have seen nearly two decades of genetic linkage analysis studies and, as the author of another article in the same issue of the Journal concedes, there has yet to be a “confirmed gene related to bipolar disorder” (2, p. 595). One would think that this fact would make genetic researchers more circumspect.

Yet this mountain of failed studies is now cited as evidence for the genetics of bipolar disorder, including the concept of “susceptibility” genes and now the idea of genes influencing when the onset of the disorder is likely to occur. All of this is nothing more than wild speculation. I suggest that those conducting future genome-wide scans for linkage make a good-faith effort to demonstrate that elevated lod score (logarithm of the odds ratio for linkage) results are not merely additional false positives before presenting their findings.

In fact, the very use of lod scores should be brought into question when performing a genome-wide scan. It should hardly be surprising if a few regions of the several hundred being examined have elevated lod scores simply because of random variance from the large sample of loci “scanned.” This point could easily be confirmed, and I challenge the authors to take the same subjects, randomly assign the group without regard to diagnosis, and repeat the genome-wide scan. I venture to guess that this will also produce a few elevated lod scores. Would these random results also be described as nonsignificant but suggestive?