To the Editor: Dr Duncan and colleagues1 analyzed brainstem serotonin in children dying from sudden infant death syndrome (SIDS) and in controls. We are concerned about 2 possible errors in their statistical approach.
First, all statistical theory is based on probability sampling.2 However, the authors studied a small convenience sample of SIDS and non-SIDS infants. Although it may be difficult to draw a probability sample for this outcome, the authors need to provide a convincing argument of why they can apply statistical theory to their convenience sample. Their results may apply only to the small cohort sampled and may have a bias of unknown sign and magnitude when making inference to the SIDS population not studied.2 Therefore, there may be a higher probability of rejecting their hypotheses than they estimated. Moreover, the use of a small convenience sample may help explain why this study did not replicate some findings that were statistically significant in the authors' 2006 study3 (lines 4, 5, and 10 of Table 3 in the 2010 study) or their previous finding that their data may help explain why boys are approximately 50% more vulnerable to SIDS than girls.
Mage DT, Kollander M. Brainstem Serotonin in Sudden Infant Death Syndrome. JAMA. 2010;303(18):1810-1811. doi:10.1001/jama.2010.541