Reevaluation of mortality risks from leukemia in the formaldehyde cohort study of the National Cancer Institute

Objective. To determine whether the National Cancer Institute's (NCI) recent suggestion of a causal association between formaldehyde exposure and mortality from leukemia and myeloid leukemia (ML) is robust with respect to alternative characterizations and categorizations of formaldehyde exposure and to alternative methods of data analysis. Methods. The original authors provided the cohort data. We computed US and local county rate-based standardized mortality ratios (SMRs) and internal cohort rate-based relative risks (RR) by categories of four formaldehyde exposure metrics (highest peak, average intensity (AIE), cumulative, and duration), using both NCI categories and an alternative categorization based on tertiles of deaths from all leukemia among exposed subjects. For highest peak exposure, we computed RRs by the duration of time worked in the highest peak category and the time since highest peak exposure. For AIE, we computed RRs by the duration of exposure and the time since first exposure. Results. Our external comparisons revealed that the elevated leukemia and ML RRs and associated trends reported by NCI for highest peak and AIE occurred because null (or slight) to moderate mortality excesses were compared with statistically significant baseline category deficits in deaths. Our alternative categorization of AIE yielded leukemia and ML SMRs close to 1.0 in the highest exposure category, and revealed weaker evidence of a trend in RRs for leukemia and ML. We corroborated NCI's finding of no association for cumulative and duration of formaldehyde exposure. We found no consistent evidence that leukemia or ML risks increased with increasing duration of time spent in a given highest peak exposure (or for AIE, duration of exposure in a given AIE category). We also found no consistent evidence that leukemia or ML risks were greater in the more relevant shorter (less than 20 years) versus longer (20+ years) periods of time from the first highest peak exposure (or for AIE, first exposure). Conclusions. Our reanalysis provided little evidence to support NCI's suggestion of a causal association between formaldehyde exposure and mortality from leukemia and ML. NCI's key findings for highest peak exposure and AIE do not adequately account for the inordinately large deficits in deaths in the categories used as the baselines for internal rate-based RRs. The NCI findings also do not adequately account for the duration of time subjects spent in the highest peak category (or for AIE, duration of exposure) or the time since their first peak exposure (or for AIE, time since first exposure). Our finding that NCI's suggestion of a causal association is not robust with respect to alternative categorizations of formaldehyde exposure and methods of data analysis casts considerable additional uncertainty regarding the validity of this suggested association. (C) 2004 Elsevier Inc. All rights reserved.