A JAMA journal is retracting and replacing a 2020 paper which linked exposure to nonionizing radiation — think cellphones, Bluetooth devices and microwave ovens — during pregnancy to the risk for attention deficit disorder later in childhood after a reader pointed out a critical error in the study.
The paper, “Association Between Maternal Exposure to Magnetic Field Nonionizing Radiation During Pregnancy and Risk of Attention-Deficit/Hyperactivity Disorder in Offspring in a Longitudinal Birth Cohort,” appeared in JAMA Network Open and prompted a significant amount of media coverage, as well as activity on social media.
According to the authors:
In this study, in utero exposure to some, but not all, high levels of MF nonionizing radiation was associated with a higher risk of ADHD. However, the associations observed were inconsistent and nonlinear.
The corrected study doesn’t abandon that claim entirely, but it does question the implications of the association. Here’s the notice, which is attributed to De-Kun Li, a reproductive epidemiologist at Kaiser Permanente Northern California, in Oakland, and the first author on the paper:
On behalf of my coauthors, I write to report a concern raised about the methods and analyses for the Original Investigation, “Association Between Maternal Exposure to Magnetic Field Nonionizing Radiation During Pregnancy and Risk of Attention-Deficit/Hyperactivity Disorder in Offspring in a Longitudinal Birth Cohort,”1 that was published in JAMA Network Open on March 24, 2020. The concern was brought to the attention of the journal editors by a reader and the editors forwarded this concern to us. After discussion and assessment, we have determined that there were errors in the statistical analyses, which required use of different analyses and which result in changes to some of the original findings and limitations to the interpretations of this study. As a result, the editors have asked us to retract and replace this article.
The notice goes into detail about the methodology of the study, including how the researchers defined high exposure to magnetic fields:
To examine the association of high levels of MF nonionizing radiation with risk of ADHD, we used the 90th percentile of the 24-hour measurements as the MF index, which reflects the MF nonionizing radiation level at or above which a participant was exposed for 10% of the time during the day. We had originally classified participating mother-child dyads into low or high MF nonionizing radiation exposure groups based on the experience of previous studies. 2–4 However, we erroneously reported using a 1.3-mG cut point based on the 25th percentile of the MF index’s distribution. And we then conducted analyses with participants dichotomized to a low exposure level (those whose MF nonionizing radiation level was <1.3 mG) or a high exposure level (those whose MF nonionizing radiation level was ≥1.3 mG). These analyses led to the conclusions that “in utero exposure to high levels of MF nonionizing radiation was associated with an increased risk of ADHD, especially ADHD with immune-related comorbidity.” 1
But using different criteria — continuous exposures instead of cut-points and a dose-response analysis:
we continue to find in utero exposures to some, but not all, high levels of MF nonionizing radiation were associated with a higher risk of ADHD. However, the associations observed were inconsistent and nonlinear, limiting interpretations. Thus, changes are needed to the original article, including the following:
We removed the cut point–based results in the original Table 1 and replaced the data using mean (SD) based on continuous MF exposure levels. Tables 2 through 4, which previously reported findings based on dichotomous low and high MF exposure levels, have been replaced with findings from the analysis using continuous measure and categorical levels of MF exposure. We also replaced the eTable in Supplement 1 with similar analyses.
We have added an eFigure in Supplement 2 showing the findings of the dose-response analysis between MF exposure in quintiles of mother-child dyads and the risk of ADHD. In addition, we have replaced the original dichotomized Kaplan-Meier survival curve with multicategorical Kaplan-Meier survival curves for multiple maternal exposure levels.
Finally, we have added to the Discussion section the following limitations: “no corrections were made for multiple comparisons and for many of the significant associations observed among children with ADHD and concurrent immune-related comorbidities, the CIs were wide, indicating that these findings should be interpreted with caution.”
On behalf of my coauthors, I apologize for any confusion this has caused readers and have requested that the original article be retracted and replaced with a corrected version. The corrections affect the Abstract, text, Tables, Figure, and Supplement 1. The replacement article includes new supplements with a copy of the original article with the errors highlighted and another copy with the corrections highlighted.
Li told us:
Basically, we replaced the original results, which were based on a dichotomized MF exposure categorization, with results based on continuous MF exposure levels, as well as results based on categorical MF exposure levels, without using a cut point level. As stated in the letter you referenced, after the new analyses, we continued to find that in-utero exposures to some, but not all, high levels of MF non-ionization were associated with a higher risk of ADHD. Compared with the results in Tables 2-4 of the original article to those in revised Tables 2-4 in the replacement article, the associations became stronger for some MF exposure categories as well.
We appreciate that a reader took the time to contact the journal and look on it as a successful outcome of the scientific review process.
JAMA Network Open retracted its first paper just last month.