LAUSR

Incorrect Data Used in Statistical Analyses To the Editor On behalf of our coauthors, we write to explain errors that occurred in our article, “Association of Environmental Toxins With Amyotrophic Lateral Sclerosis,” which was published in the July 2016 issue of JAMA Neurology.1 The article reported our case-control study of occupational and residential exposures to 122 neurotoxic organic pollutants, including organochlorine pesticides (OCPs), polychlorinated biphenyls (PCBs), and polybrominated diphenyl ethers (PBDEs), that were assessed using a self-administered survey derived from the Agency for Toxic Substances and Disease Registry, the National Health and Nutrition Examination Survey, and the National Human Exposure Assessment Survey, as well as biological samples obtained from 156 patients with amyotrophic lateral sclerosis (ALS) and a control group of 128 individuals. The objective of the study was to evaluate the association of occupational exposures and environmental toxins with the risk for developing ALS in Michigan. To obtain a sufficiently large sample size, we used archived and newly obtained biological samples. Most of the archived samples were whole blood, and most of the newly obtained samples were plasma. These samples were analyzed by the laboratory in 3 batches, each using separate gas chromatography: mass spectrometry run for the 3 compound groups. Batch 1 had 93 samples, of which 88 were whole blood and 5 were plasma; batch 2 had 97 samples, of which 47 were whole blood and 50 were plasma; and batch 3 had 106 samples, of which 27 were whole blood and 79 were plasma.1 Laboratory protocols, calibrations, and unit adjustments vary between blood and plasma samples, although we have demonstrated excellent reproducibility between whole-blood and plasma matrices for these compounds in a recent publication.2 Unfortunately, the initial analyses of the first 2 batches of PCBs and PBDEs used calibration and unit adjustments for blood only. While this was corrected in the laboratory, the statistical analyses reported in our article1 were based on the original and incorrect data for PCBs and PBDEs. In somecases,thedifferencesbetweenwhatwasoriginallyreported and the correct data are large (eg, factor of 10). The analyses for OCPs were correct in the original article. This situation was brought to our attention by representatives of the Environmental Chemistry Laboratory at the Centers for Disease Control and Prevention who reported discrepancies between the median pollutant concentrations in our control participants and the data from the National Health and Nutrition Examination Survey. These differences were discussed in depth with members of the Centers for Disease Control and Prevention’s Agency for Toxic Substances and Disease Registry and Environmental Chemistry Laboratory. The Exposure Assessment Core at University of Michigan reviewed the data and discovered that incorrect calibration and unit adjustments were provided for samples measured with whole blood versus plasma. We have revised the data set and rerun the statistical analyses that were presented in the original article and Supplement. The original article included errors in Table 3; Figure 1; and eTables 1, 3, 5, 7, 9, and 10 in the Supplement.1 The corrected Table 3, which presents results from the multiple regression model, shows small changes in odds ratios and significances for pesticide exposure, a change to no statistical significance for PBDE-47 and PCB-175, and changes in the direction of the odds ratios for PCB-151 and PCB-202. The single pollutant models presented in eTable 5 show similar results for PBDEs. Several PCBs in the corrected analysis had significant associations that were not seen in the original analysis. The descriptive summaries in eTable 1 changed for several PCBs and PBDEs. eTables 3, 5, 7, 9, and 10 in the Supplement, which present results of sensitivity analyses for the models, continue to demonstrate agreement between observed, imputed, and match datasets. After correcting these errors, our study’s major results and conclusions remain unchanged: that persistent environmental pollutants measured in blood were associated with ALS and that these exposures may represent modifiable ALS disease risk factors. We apologize to the readers and editors of the journal for any confusion this may have caused. We have requested that our article be corrected online.3