Neuropsychological performance and solvent exposure among car body repair shop workers.

The statement by Steenland and Stayner that we have analysed the data from the study conducted by NIOSH is a gross exaggeration. Our study was based on employment records microfilmed by NIOSH, additional employment records from member companies (microfiche, hard copies, and computer tapes), vital status information from various sources (SSA, NDI, DMF, member companies), and death certificates from state health departments. All these data sources were equally important in our study, and we have properly identified and acknowledged every source in our paper. The discrepancy in cohort size between our study' and the NIOSH study2 is most likely the result of additional information we obtained from the participating companies. As stated earlier, the NIOSH microfilms contained both illegible and incomplete work history information. It was necessary for us to obtain additional information from member companies to resolve these data gaps. To our knowledge, NIOSH had never gone back to the member companies for additional information. Steenland and Stayner criticised our study for lack of detailed exposure information. Although we recognised the value of valid exposure data in epidemiological studies, that is not to say that a study is better simply because it has some exposure estimates, valid or otherwise. One of us (OW) participated in some of the walk through surveys and in reviewing the NIOSH exposure classification. Although much resources (from both NIOSH and HIMA member companies) were spent on historical exposure estimates, the validity of the estimates provided by the NIOSH model is questionable. Several member companies have expressed their concerns regarding the inaccuracy of such estimates. For example, at one facility, the NIOSH model predicted that the exposure for a steriliser/operator in 1977 was 19-3 ppm, but industrial hygiene measurements based on 17 samples indicated that the actual exposure was 45-2 ppm. We are strongly of the opinion that the NIOSH exposure estimates were inaccurate and would have been a major source of misclassification if they were incorporated in the analySiS. As we stated in our paper,' average duration of exposure was only about one year shorter than average duration of employment. Thus duration of employment was a close surrogate measure for exposure. We defined latency as time since first employment. NIOSH defined latency as time since first exposure. The difference was minor. The statement by Steenland and Stayner that we failed to observe a "trend" for all haematopoietic cancer by latency was inaccurate. Our data did show an "upward trend" for all haematopoietic cancer. As this broad International Classification of Disease (ICD) category consists of several heterogeneous diseases, however, we attached little interpretation to it. It would be far more meaningful to examine the individual cancer categories within this broad category. Such analyses were done. In particular, as noted in our paper, as there was a significant increase of non-Hodgkin's lymphoma in men, we performed a latency analysis of non-Hodgkin's lymphoma for men, and no trend was detected. Similar to our results, NIOSH's analyses also failed to show any trend by latency for more specific individual disease categories. The statement by Steenland and Stayner characterising our study as "an essentially duplicate analysis" of the data is inaccurate. Data sources aside, we have presented far more analyses than the NIOSH paper. To start with, we presented standardised mortality ratios (SMRs) for 50 causes of death in most analyses, whereas a much smaller number of causes of death was presented by NIOSH (most tables in the NIOSH's paper had only 10 causes of death). Furthermore, as previously pointed out by one of us (OW), the category "non-Hodgkin's lymphoma" used in the NIOSH study (ICD 202 only) was incorrect.3 In the 8th ICD, nonHodgkin's lymphoma consists of both codes 200 and 202. Our analysis of non-Hodgkin's lymphoma was based on the proper ICD categories. Finally, we also provided analyses for all four major histological cell types of leukaemia.' NIOSH did not analyse leukaemia data by cell type. Steenland and Stayner accused us of "downplaying" the carcinogenic risk of ethylene oxide. It should be pointed out that nowhere in their own paper did Steenland and Stayner conclude that their data had shown a carcinogenic effect of ethylene oxide. In fact, even though Steenland and Stayner found that "overall there was no significant increase in mortality from any cause in the study cohort," they concluded that, because of the small sample size, "our (NIOSH's) findings are therefore not conclusive."2 Thus with regard to whether ethylene oxide is carcinogenic to humans, the only conclusion that NIOSH could offer was no conclusion. NIOSH's conclusion of their study being not conclusive begs the question why NIOSH conducted the study in the first place. As stated in the letter by Steenland and Stayner, a feasibility study was conducted by NIOSH before the actual mortality study. Based on the results of the feasibility study, NIOSH estimated that the mortality study would have adequate statistical power (80% at a = 0 05) to detect a risk ratio as small as 2-0 for all haematopoietic cancer and 2-7 for leukaemia. Our calculation indicated that the NIOSH study actually had adequate power to detect risk ratios as small as 1-47 and 1-79 for all haematopoietic cancers and leukaemia, respectively. Thus the actual power of the NIOSH mortality study was much higher than what was anticipated based on the NIOSH feasibility study. We can only assume that the NIOSH mortality study was given a "go ahead" only after a careful consideration of the anticipated statistical power.