Statistics from Altmetric.com
We appreciate the recent contribution of Ulmet al 1 to the controversial question as to whether occupational exposure to crystalline silica, a classified carcinogen, increases the risk of lung cancer irrespective of silicosis. There are, however, some serious conceptual and methodological questions regarding the design, conduct, and analysis of the study which may affect the conclusions.
If silicosis is a surrogate for the internal dose or on the causal pathway between silica exposure and lung cancer, this question might only be addressed with detailed individual exposure information and classification of the silicosis grade in a population of silicotic and non-silicotic subjects.
The authors presented the pooled data from two matched case-control studies among non-silicotic subjects in two German industries. The study populations were not clearly defined with respect to the region. Cohorts for the recruitment of cases and controls as suggested by the authors cannot be reconstructed. Given the lack of a nationwide cancer registry in Germany, complete case ascertainment is questionable. In particular, the selection of potential cases from the workforce due to health problems poses a threat to internal validity.
Recruitment of controls and exposure levels varied between the two industries. Selection of more highly exposed or silicotic control subjects, in particular in the stone and quarrying industry, cannot be ruled out. We think that more detailed information on the data and results from the two industries, which have been presented in a previous German report of this study,2 is crucial for the interpretation of the results. Discrepant results between the two industries and between different publications of this study ought to be discussed.
The authors made a great effort to reconstruct the exposure to silica dust. However, the quality and completeness of the exposure data remain questionable because the exposure assessment was retrospective and no figures on missing dust measurements were presented.
The baseline risks of the cases and controls were high (cumulative exposure ⩾0.12 mg/m3 • years) which may render the detection of an additional risk of lung cancer in this study population difficult. Could the restriction to highly exposed individuals explain the low relative risks of lung cancer in this study?
In the previous report of this case-control study,2 which included subjects with silicosis, the risk estimates of lung cancer for the three exposure indices were relatively low and not significant. Why do the authors expect the risks and the significance to be higher after excluding those with silicosis?
The authors presented only adjusted odds ratios. We appreciate the control of confounding by other occupational exposures. However, if the duration of exposure is a surrogate for the investigated exposure, an additional risk might not be observed after adjustment for duration of exposure.
The authors matched subjects on smoking status yet, in the previous report, an index of smoking intensity and duration was a significant risk factor for lung cancer and the smoking adjusted risk estimates were slightly higher than the risk estimates reported in this paper.2 Why did the authors not adjust for smoking as in the previous report?
In general, details on excluded and deceased subjects, coding of variables, missing data, and statistical methods presented in this paper were not easily verifiable. In the discussion the authors concluded that the study showed no association between exposure to crystalline silica and lung cancer which might either be a true finding or be attributed to a lack of statistical power. Like previous studies of non-silicotic subjects3 this study lacked the power to detect risks of 1.4 or less. In addition, all potential sources of bias (non-differential misclassification of the exposure, selection bias, and healthy worker effect) lead to an underestimation of the risk estimates. We would therefore suggest that the results are interpreted with caution. It remains unclear whether silicosis is on the causal pathway between exposure to silica and lung cancer.
authors' reply We appreciate the comments of Baur and colleagues on our paper.1-1 In Germany it is difficult to perform epidemiological studies due to the lack of registries and some regulations, therefore in case-control studies complete case ascertainment can never be guaranteed. We checked all the sources available to identify all cases with lung cancer in the selected area of Germany. The main question, however, is whether the group of cases enrolled is selected and not a random sample. There are no indications for that bias with respect to the exposure.
The selection of controls can be a problem in case-control studies, especially if there is no file of all possible controls available, which is the situation in the stone and quarrying industry. As already mentioned in the paper, for this type of industry the so-called accident file has been used. The possibility of a selection bias can never be ruled out. However, most of the accidents are based on carelessness such as stumbling. Whether those accidents are related to any exposure levels is highly speculative.
Due to lack of space in the paper the data reported had to be condensed. In table 1-1 the distribution of the cumulative exposure for both industries are given separately. There is a difference between both industries with respect to higher exposures.
The level of exposure in both industries is comparable to the other studies and not exceptionally high as stated by the authors of the letter. In the monograph by IARC1-2 the geometric mean for the ceramic industry varied between 0.01 and 0.44 mg/m3with a range of 0.01–1.14 mg/m3. The level in the quarrying industry is higher with a mean value of 0.03–2.1 mg/m3 and peak values over 100 mg/m3.
Assessment of the exposure in occupational epidemiology is mostly based on retrospective estimates and not on measurements. Within this study great effort has been made to quantify the exposure. If measurements were available these have been taken into account.
If the smoking habits are included in the model the fit improves considerably. In the analysis of the average exposure the likelihood ratio value is 386.62 with smoking and 345.72 without. However, the estimates of the odds ratio for the average exposure remain nearly unchanged (with smoking: OR = 0.94, p = 0.82; without smoking: OR = 0.91; p = 0.71).
The exclusion of subjects with silicosis in the final analysis is due to increased knowledge acquired subsequently. After the decision by the IARC in 1996, the main interest was focused on the question of whether or not silicosis is a prerequisite for lung cancer. There are only a few studies available investigating the risk in subjects without silicosis.
Our study is in line with other publications. In the absence of silicosis there is either no additional risk caused by silica dust or only a very low risk. For practical purposes it seems justified to check the current threshold values for silica dust in order to prevent silicosis and therefore to exclude, or at least minimise, the associated risk of lung cancer.