Lack of a relation between human neonatal thyroxine and pediatric neurobehavioral disorders: neonatal total thyroxine is not a good proxy measure of maternal thyroid hormone insufficiency.

Abstract

WE WERE STRUCK by the title of the paper recently published by Soldin et al. in the February issue of Thyroid (“Lack of a relation between human neonatal thyroxine and pediatric neurobehavioral disorders” (1) for its provocative, if unintended, implication that neonatal thyroxine (T4) is not associated with specific neurobehavioral deficits. Clearly, neonatal thyroxine may be associated with a variety of neurobehavioral deficits (2–5), and new research is revealing associations between neuropsychological outcome and subtle maternal thyroid hormone insufficiency (6–9). In contrast to the implication of the title, the hypothesis under investigation was that “intrauterine thyroid dysfunction is an early factor in the development of neurobehavioral disorders” in humans. Clearly, this hypothesis has important implications for human health. To test this hypothesis, Soldin et al. (1) utilized a case-control study design in which the hypothesized causal pathway of interest (insufficient levels of free T4 during fetal brain development leading to deficiencies in central nervous system [CNS] development) was presumed to be represented adequately by relationships between neonatal levels of T4 and a hospital clinic diagnosis of any of the following: attention-deficit hyperactivity disorders (ADHD), autism spectrum disorder, behavioral disorder, cognitive disorder, developmental delay, emotional disorder, learning disability, or speech/language disorder. However, we contend that the choices of neonatal T4 as a proxy measure of intrauterine thyroid status, and the use of hospital diagnoses of broad and heterogeneous neurobehavioral disorders limits the ability to adequately test this hypothesized causal pathway of interest. In the first case, the choice of neonatal T4 as a proxy measure of intrauterine thyroid status is not appropriate for reason of (at least) the findings of Haddow et al. (8), who observed that neonatal T4 is not reflective of maternal thyroid status, and that of Calvo et al. (10) who observed that the fetus is exposed to biologically relevant levels of free T4 that are likely to be affected by subtle changes (e.g., hypothyroxinemia) in maternal free T4 rather than in neonatal T4 (11). The issues germane to the use of hospital diagnoses are more complex. Soldin et al. (1) identified cases based on the sole criterion of a hospital clinic diagnosis—rendering the cases a disproportionate representation of higher severity disease; this has implications for the generalizability of study findings. Although the controls were intended to be identified as children without a hospital clinic diagnosis of a neurobehavioral disorder and with neonatal T4 determinations made at the same time as those of the cases, there was no assessment of the “controls” to verify his or her control status as having no diagnosis of disease made elsewhere (e.g., pediatric practices) in any year up to the time when the matched case was diagnosed. Moreover, no information was collected on potential confounders and modifiers of the relationships between neonatal T4 and neurobehavioral disorder. Therefore, the data collected were biased toward the null in that, among the cases, some normal levels of neonatal T4 are likely “misrepresentations” of potentially low, or infrequently high, levels of free T4 in utero, rendering cases spuriously similar to controls; and the controls likely include some children with a neurobehavioral disorder, the diagnosis being made in a nonhospital setting, rendering the controls spuriously similar to the cases. Finally, the available data include no low levels of T4. This precludes the discovery of any of the relationships of interest and the reported odds ratios are estimates of relationships that are not of interest; namely those between disease and variations in normal levels of T4 on the second day after birth. Further confusing the interpretation of the reported odds ratios is that it is not known what they are actually