some of the results support the hypothesis that diesel exhaust (DE) exposure increases the risk of lung cancer, some aspects of the results and potential limitations in the DEMS should be taken into consideration in the interpretation of the evidence. In the analysis of continuous exposure variables (1), the hazard ratio for one unit of cumulative DE exposure (one µg/m 3 -y respirable elemental carbon [REC]) was 1.001 among underground miners and 1.02 (ie, 20-fold higher) among surface miners. Corresponding HRs for one logtransformed unit of average REC intensity were 1.26 and 2.60, respectively. The authors interpret these results as indicating a stronger carcinogenic potential of “aged” DE. In the corresponding analysis of the nested case–control study, however, the risk of lung cancer among surface miners was not consistently increased [Table 4 in (2): a nonstatistically significant dose-dependent decrease in risk was shown for two of the variables]. The DEMS authors interpret these results in the light of the small number of lung cancer cases and the low levels of DE exposure in this group. The reader, however, remains uncertain as to whether surface miners, compared with underground miners, have between twofold and 20-fold higher DE-related risk or no increased risk at all. These results cast doubts on the validity of exposure assessment in the DEMS (3). The anomalous result of a higher overall standardized mortality ratio (SMR) among surface miners than among underground miners in the cohort analysis is attributed to confounding by smoking. No clear evidence, however, is provided to support this interpretation. In the nested case– control study, the distribution of controls by smoking habit [Table 2 in (2)] suggests a lower amount of smoking in surface vs underground miners. Although controls are not representative samples of the two groups of miners, this finding contradicts the hypothesis of tobacco smoking being a stronger confounder for surface miners than for underground miners. More importantly, the odds ratios for tobacco smoking were different in the two groups of miners; whereas the odds ratios in surface workers were consistent with previous studies (4), those in underground miners were much lower. The latter result is hardly credible and puts in question the way information on tobacco smoking was collected and analyzed in the DEMS. The DEMS authors quote a number of previous studies of DE-exposed workers as supporting evidence of their results (5–11). However, the results of some of these studies only weakly support the hypothesis of a causal association [eg, (7)], and several welldesigned studies that did not support the authors’ conclusions were not quoted [see (12) for a detailed review]. More importantly, these previous studies included drivers, machine operators, and railroad workers whose circumstances of DE exposure were closer to those of surface miners than to those of underground miners in the DEMS, for whom there was no clear evidence of an effect of DE exposure. The results reported in the two articles do not match the plan of the statistical analysis outlined in the DEMS protocol (13). One particularly striking example concerns the use of lag in exposure–response analyses. In the study protocol, only a brief mention of lagged analyses is made: “In addition, lagged estimates of exposure will be explored to account for the latent period relating to lung cancer development” [(13), page 21], whereas strong emphasis is given to results of lagged analyses in both articles. The exclusion of miners with less than 5 years of employment and the exclusion of the category at highest exposure in dose– response analyses were not mentioned in the protocol, but the key results of the study are based on these exclusions. This is not to say that data-driven analyses should not be conducted and reported. When a study protocol is prepared, typically on the basis of limited data from a pilot study, it is impossible to figure out all the nuances and complexities of the final data, but to ignore systematically the plans outlined in the protocol is not good practice and may open the door to an arbitrary selection of results. At a minimum, the reports should have distinguished between a priori and a posteriori analyses. The DEMS authors state (2) that exposure to REC in the range of 2–6 μg/m 3
Read full abstract