Edge found 19 deaths from lung cancer in 429 Barrow men with plaques and only 4 in 429 Carlisle controls. He recognized that 198 men with plaques were found in hospital or clinic sources and that this may have biased the results. Indeed, 13 of the 19 had evidence suggestive of lung cancer at the time of the first x-ray films; these 13 should have been excluded. However, he compared 231 of the men with plaques discovered on “routine films” with the 429 men in Carlisle, all found by examination of “routine films.” The lung cancer death frequency was 2.6 percent in the study group, compared with 0.9 percent in the control group. This difference is not statistically significant at the 0.05 level by x2 with Yates’ correction.
Edge then reduced the number of men with plaques from 429 to 407 by excluding 3 lost to follow-up and 19 who died within the first year of follow-up, including 3 with lung cancer. Calculating observed and expected (based on national rates) numbers of lung cancer deaths, he found the following ratios by period of follow-up: 0 to 2 years, 8:1.8; 2 to 4 years, 0:1.9; 4 to 6 years, 2:2.0; 6 + years, 6:2.8. Thus, 8 of the 16 patients with lung cancer died within the first 2 years, indicating serious selection bias. The ratio for the period 2 or more years after the start of observation was 8:6.7=1.19, with 95 percent CLs I calculated to be 0.51 to 2.35. While his Table 3 shows the ratio rising from 0:1.9 during the period 2-4 years to 6:2.8 6 or more years after the start of observation, the last ratio is not statistically significant at the 0.05 level. Furthermore, given the figures cited above for the number of subjects with evidence of lung cancer at the start of observation (n = 13), several of the 8 patients who died more than 2 years after the start of observation must have had evidence of lung cancer at the start of observation and should have been excluded.
The studies by Fletcher and by Edge were done in the same city, with considerable overlap of the time periods (1960 through 1970 and 1964 through 1971, respectively) in which chest films were reviewed; therefore, it is likely that many subjects with plaques were common to both studies. Both were subject to selection bias. Essentially, the two studies taken together represent the initial cohort research that generated the hypothesis that workers with pleural plaques are at increased risk of lung cancer.
The other four cohort studies may be considered as a group to test the validity of this hypothesis. There are some differences in design (Table 1). Two* had cohorts derived from community x-ray surveys in Finland, one dealt with shipyard employees in Sweden, and one included employees of two asbestos cement plants in New Orleans. The types of asbestos to which these cohorts were exposed were specified in three studies: mixtures of chrysotile, amosite, and crocidolite in two* and mainly anthophyllite in one. The numbers of persons with pleural plaques varied. In the study by Hughes and Weill, 56 percent of the 62 subjects had plaques with or without difluse pleural thickening, and the rest had only diffuse pleural thickening, each subgroup contributing 1 of the 2 lung cancer cases (J. M. Hughes, personal communication, 1992).
Table 2 shows that the completeness of follow-up was given in only two of the four studies,* both close to 100 percent. The periods of follow-up varied a little. Only two of the studies had a limited attempt to control for smoking habits. In three studies, the investigators controlled for age and sex. There is doubt about adjustment for differences in age and sex between the plaque group and the controls only in the study by Kiviluoto et al. Two of the four studies controlled for latency by limiting the results for the plaque groups to observation at 20 or more years after beginning of exposure to asbestos.
Table 3 shows the statistical methods and results in the cohort studies. Kiviluoto et al used cumulative incidence of lung cancer from a cancer registry for the plaque group and the control group, while the other investigators used SMRs or a proportional hazards model to obtain the relative risks. None of the relative risks in these 4 studies is statistically significantly different from 1.00. They vary from 0.81 to 1.33, the latter being based on only 2 observed cases. If one sums the observed and expected numbers of lung cancers for these 4 studies, they total 51 and 50.4, respectively, giving a summary ratio of 1.01, which is a reasonable measure of overall relative risk weighted by the expected numbers. The 95 percent CLs for this ratio based on the Poisson distribution are 0.75 and 1.33.