Li et al. Respond to "No Link between Hot Tubs and Miscarriage"

Abstract

Observational epidemiologic studies are subject to limitations that are often difficult to quantify and therefore are subject to interpretation. It was our judgment that despite the limitations of our study (1), the observed association between the use of hot tubs or Jacuzzis (any type of whirlpool bath; Jacuzzi Brands, Inc., West Palm Beach, Florida) during pregnancy and the risk of miscarriage was valid enough to be published. This judgment was based not only on the findings presented in our paper but also on the findings of other associations (prenatal exposure to magnetic fields and use of nonsteroidal anti-inflammatory drugs) derived from the same study cohort (2, 3). The findings of the risk of miscarriage associated with other exposures (magnetic fields and nonsteroidal anti-inflammatory drugs) have been corroborated by other studies (4, 5). Even some of the findings reported in our study (e.g., a lack of association with fever) were confirmed by a recent cohort study (6). Nonetheless, reasonable minds differ. In this issue of the Journal, Hertz-Picciotto and Howards (7) present their judgment on the impact of potential biases and conclude that those limitations undermine our conclusion (1). Our paper addressed many of those criticisms. The new concern of residual confounding due to maternal age was also not supported by our data: the hazard ratio remained essentially unchanged after a linear term for maternal age of ≥25 years was used, as suggested by Hertz-Picciotto and Howards (hazard ratio = 1.9, 95 percent confidence interval: 1.2, 2.9). In addition, we realize that the difference in judgment of the impact of those limitations stems not only from the different assumptions for the potential biases but also from the choice of interpreting the results. For example, it is puzzling that, given the results presented in their table 2 (derived from our paper), they would have concluded that “the prospective portion of the study shows essentially no association at all” (7, p. 939). The hazard ratios were consistent regardless of whether the interviews were conducted before or after miscarriage. The hazard ratio of 3.2, based on 2,331 persondays from the exposed women and 24,214 person-days from the unexposed women, remained borderline significant (95 percent confidence interval: 1.0, 10.3) despite the reduced sample size due to splitting women into prospective and retrospective groups. Furthermore, while Hertz-Picciotto and Howards attributed our findings to recall bias, they failed to explain why the recall bias only existed for miscarriage before 10 weeks of gestation, for there was no association for miscarriage beyond 10 weeks of gestation (refer to their table 2). A higher risk associated with use of a hot tub or Jacuzzi for early miscarriage (<10 weeks of gestation) makes good biologic sense; however, it does not support the argument for recall bias. Ultimately, the validity of our findings (1) will be determined by the findings of replication studies that overcome some of the limitations. Therefore, we would encourage other investigators, including critics of our findings, to conduct better studies to confirm or refute our findings.