CIGARETTE SMOKING, ASBESTOS EXPOSURE, LUNG CANCER, AND SAMPLE SIZE

To the Editor:

I am concerned about one conclusion in the otherwise excellent article "Association of Cigarette Smoking and Asbestos Exposure with Location and Histology of Lung Cancer" by Lee and colleagues (1).

The authors state in the abstract: "The proportion of patients with significant exposure to asbestos was lower among those with adenocarcinomas but was not statistically significant (9.5 versus 15.3%, p = 0.09)." On page exlnk0 they state that "Among those with adenocarcinomas, 9.5% reported exposure to asbestos in contrast to 15.3% of those with other types of lung cancer, but the difference was not statistically significant (OR = 1.72; 95% CI: 0.95-3.12, p = 0.09). Stated another way, adenocarcinomas comprised 46.9% of the tumors among those exposed to asbestos compared to 60.3% of those not exposed to asbestos (p = 0.09)."

A study such as this relies heavily on statistical analyses. All the conclusions made are based on statistical inference which, in turn, is based on probabilities. Within this framework, we must accept that a comparison is different only if it is statistically different. This may sound like a tautology, but if we accept that premise, then there is no such thing as a difference that is not statistically significant.

There are a few reasons why two groups may look to be different but are not. One is simply that appearances can be deceiving, which is why we perform statistics in the first place; large scatter often contributes to this appearance of dissimilarity. The second is probability; we do have a certain probability of being wrong, which we accept based on our p value, usually 0.05, as the authors have done. And finally, the other reason is insufficient sample size; using power analysis, we can calculate, prior to the study, how large the sample size should be in order to detect a certain difference. Then, if no difference is detected, we can state with a certain probability that the two groups are the same, in much the same way as we state with a certain probability using the p value that two groups are different. If the sample size is not large enough, then it may not be possible to determine whether an apparent difference is real or not (2).

Although it is difficult to determine, without the raw data, which one of these possibilities is the case for the comparison in this article, I suspect that it may be the small number of patients with asbestosis who presented with tumors. But given the prevalence of the disease, this lack of power may be unavoidable. I believe that this uncertainty about the potential difference in tumors between those who were and those who were not exposed to asbestos, properly stated, does not detract from the other conclusions of the study.

MARCY F. PETRINI

Department of Medicine

Division of Pulmonary and Critical Care Medicine

University of Mississippi Medical Center

Jackson, Mississippi

	References

1. Lee, B. W., J. C. Wain, K. T. Kelsey, J. K. Wiencke, and D. C. Christiani. 1998. Association of cigarette smoking and asbestos exposure with location and histology of lung cancer. Am. J. Respir. Crit. Care Med. 157: 748-755 [Abstract/Free Full Text].

2. Cohen, J. 1988. Statistical Power Analysis for the Behavioral Sciences. Lawrence Erlbaum Associates, Hillsdale, NJ.

From the Authors:

Dr. Petrini suggests that the "small number of patients with asbestos [exposure]" may have been responsible for the lack of a statistically significant association between asbestos exposure and adenocarcinoma in our study. Although we cannot exclude this as a possibility, we believe that lack of power is not the most likely explanation for our finding. First, our results are in agreement with most prior studies reporting no specific association between asbestos exposure and adenocarcinomas (1). Furthermore, few studies that have demonstrated an association have reported more adenocarcinomas with asbestos exposure (6) which is in contrast to our study which showed a trend toward fewer adenocarcinomas with asbestos exposure (46.9% versus 60.3%, p = 0.09). Thus, it is more likely that the apparent near statistical significance of negative association between adenocarcinomas and asbestos exposure is due to multiple comparisons.

BURTON W. LEE

DAVID C. CHRISTIANI

Department of Environmental Health

Harvard School of Public Health

Boston, Massachusetts

	References

1. Hiraoka, K., A. Horie, and M. Kido. 1990. Study of asbestos bodies in Japanese urban patients. Am. J. Ind. Med. 18: 547-554 [Medline].

2. Kannerstein, M., and J. Churg. 1972. Pathology of carcinoma of the lung associated with asbestos exposure. Cancer 30: 14-21 [Medline].

3. Karjalainen, A., S. Antilla, L. Heikkila, P. Kyyronen, and H. Vainio. 1993. Lobe of origin of lung cancer among asbestos-exposed patients with or without diffuse interstitial fibrosis. Scan. J. Work Environ. Health 19: 102-107 .

4. Warnock, M., and W. Isenberg. 1986. Asbestos burden and the pathology of lung cancer. Chest 89: 20-26 [Abstract/Free Full Text].

5. Roggli, V., P. Pratt, and A. Brody. 1986. Asbestos content of lung tissue in asbestos associated diseases: a study of 110 cases. Br. J. Ind. Med. 43: 18-28 [Medline].

6. Whitwell, F., M. Newhouse, and D. Bennett. 1974. A study of the histological cell types of lung cancer in workers suffering from asbestosis in the United Kingdom. Br. J. Ind. Med. 31: 298-303 [Medline].

7. Johansson, L., M. Albin, K. Jakobsson, and Z. Mikoczy. 1992. Histologic type of lung carcinoma in asbestos cement workers and matched controls. Br. J. Ind. Med. 49: 626-630 [Medline].

8. Hourihane, D., and W. McCaughey. 1966. Pathological aspects of asbestos. Postgrad. Med. J. 42: 613 [Free Full Text].