Table 5.

Possible contributors leading to false positive results in NCI cohort studies.

Generic issue	Suggested examples
Chance	Statistically significant deficits in leukemia mortality in internal non- and low-exposed comparison groups that substantially influence RR calculations. This single issue alone, if properly addressed and accounted for, would appear to undermine any conclusion about a causal association between formaldehyde exposure and leukemia.
Systematic errors in design of study	Questionable reliance on an unconventional measure of exposure, (i.e., peak); all significant associations with leukemia mortality disappeared when assessed using cumulative number of peaks ≥4 ppm as a more appropriate metric of potential exposure.
Analysis of study	Failure to consider using an external comparison group in light of significant mortality deficits in internal groups; as shown by the Marsh et al. (2004) analysis, significant effects disappeared when this was done. In addition, the implications of >1000 deaths missed in the 1994 follow-up, and the substantial effects of these missed deaths on the RRs as shown in Table 2, should have been addressed.
Inadequate accounting for confounding variables	Previous employment in occupations with exposure to known risk factors for NPC.