This paper discusses inference from hypothesis-generating studies on occupational risks for lung cancer based on routine hospital records. A hospital-based case-control study on 567 male lung cancer patients and 906 controls provided a practical example. Among possible causes of bias, the effects of poor detail in the occupational information, of a large proportion of exclusions due to incomplete information, of cardiovascular diseases as the prevalent diagnosis among controls, of selecting cases and controls from different hospitals with likely differences in referring areas, and the problem of multiple comparisons are highlighted. Significant excess lung cancer risks were found for farmers, miners, crushers, stonemasons and cement plant workers, and stock handlers and stevedores. Positive findings with small numbers of observations are more likely to be artificially generated, but also the precision of more robust risk estimates may be affected. The limits to inference from hypothesis generating case-control studies based upon routine hospital records, such as in the example described, outweigh the advantage of the ready availability of these data-bases.
Problems of inference from hypothesis-generating studies on lung cancer and occupation based on routine hospital records.
COCCO, PIER LUIGI;
1993-01-01
Abstract
This paper discusses inference from hypothesis-generating studies on occupational risks for lung cancer based on routine hospital records. A hospital-based case-control study on 567 male lung cancer patients and 906 controls provided a practical example. Among possible causes of bias, the effects of poor detail in the occupational information, of a large proportion of exclusions due to incomplete information, of cardiovascular diseases as the prevalent diagnosis among controls, of selecting cases and controls from different hospitals with likely differences in referring areas, and the problem of multiple comparisons are highlighted. Significant excess lung cancer risks were found for farmers, miners, crushers, stonemasons and cement plant workers, and stock handlers and stevedores. Positive findings with small numbers of observations are more likely to be artificially generated, but also the precision of more robust risk estimates may be affected. The limits to inference from hypothesis generating case-control studies based upon routine hospital records, such as in the example described, outweigh the advantage of the ready availability of these data-bases.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.