SMRs, from lowest- to highest-exposure septile, were 1.87, 2.17, 1.36, 0.92, 1.33, 1.10, and 0 for diabetes as the underlying cause of death. The trend for cumulative exposure, with p = .10, was not significant; the logarithm of cumulative exposure, p = .09. Risk ratios for any mention of diabetes on the death certificate were 1.00 (the referent), 1.27, 0.92, 0.81, 0.98, 0.72, and 0.54, respectively; this is a statistically significant negative trend with cumulative exposure, p = .02. The logarithm of cumulative exposure was not, however, significant (p = .12). The Cox regression model was controlled for year of birth (quartiles) and age (the time variable).

The consistency of results from the two sets of analyses—the underlying cause of death and any mention on the death certificate —could be the result of the low power and limitations common to analyses relying on death certificates for ascertaining a chronic condition such as diabetes. The exposure likelihood matrix assembled by the authors is a noteworthy attempt to rank-order people without the confounding of age and obesity. The caveats remain, though, that misclassification of exposure is likely and that there is potential for systematic underascertainment related to exposure level. The absence of weight data on death certificates also complicates the assessment of diabetes. Additionally, because the chloracne cohort had a markedly higher median cumulative exposure score (11,546) than workers without chloracne (77), the fact that there was no significant increase in the SMR for diabetes greatly dampens the hypothesized dioxin–diabetes association.

The SMR calculation made in this paper employs a methodology described in an earlier paper (Steenland et al., 1992) that presented the rationale for using multiple-cause-of-death data, especially when examining diseases and conditions that are often present at death, but are not the cause of death, yet are “serious enough to be noted by the physician on the death certificate,” such as diabetes. The authors created a ratio of the number of deaths for which a cause is listed at all and the number of deaths for which that cause is listed as the underlying cause of death. At one extreme are the ratios for transportation accidents and lung cancer—1.02 and 1.09. The other extreme is made up of diseases that are less likely to be fatal but likely to be present and significant at death, such as arthritis and hypertension without heart disease—10.70 and 12.10. Diabetes lies near the center, with a ratio of 3.82, and is one of the examples the authors present to demonstrate the multiple-cause-of-death approach.

This 1992 paper—which was not previously reviewed in a Veterans and Agent Orange series report—uses data regarding deaths in a U.S. worker cohort exposed to dioxin during the manufacture of herbicides and other chemicals to demonstrate the utility of examining multiple-cause mortality information. The authors calculated SMRs for underlying cause of death and for any mention of diabetes. The more narrow underlying-cause-of-death categorization for diabetes yielded 16 deaths, while the broader multiple-cause-of-death categorization yielded 58. This validated the attempt to capture more diabetes-related cases but did not substantially change the SMR—1.07 (0.61–1.75) for underlying