Abstract
The lessons to be learned from considering why the stratified analysis of Table 1 may have been inappropriate are as follows: (1) In a prospective investigation, one must be concerned with the changing status of the individuals involved. Yet, in situations, biassed results could occur from attempts to take changing status into account. An alternative effect of changing status could be to change the interpretation of the results obtained. In any case, some minimal amount of information on changing status has to be collected. (2) For a prospective study, it can be important that the analysis take the passage of time into account. Not only may time in itself be having an effect, but the decreasing with time of the populations at risk has to be taken into account. (3) Heterogeneity of the study population must be considered, with the analysis made on more homogeneous strata. Whatever other bases are used for stratification, it is ordinarily essential for most disease processes, and particularly for cancer, to stratify according to initial age. Note that if we have stratified according to initial age, the fact that age thereafter changes can be ignored, as ages remain homogeneous within a stratum (but with certain assumptions, the analysis could be simplified so as to add together from all the initial-age strata the total person-years at risk at a particular age). (4) Making the analysis specific for the disease supposed to be related to the study factor law can increase chances of identifying associations between disease and the suspect factor, notwithstanding the diminution in the number of cases available for analysis. The conducting of a more specific analysis does not preclude also making an analysis for the broader disease category. (5) Next, it is essential that results be reported in such a way as to make clear just what was the nature of the statistical analysis and also to show up certain other features of the study. The nature of Hirayama's investigation was such that the only logical analysis that could have been made would have been along the lines I have indicated above. The absence of indications in Hirayama's report that he did such a logical analysis, e.g., his not stating what age groupings were made of nonsmoking wives, may have been just carelessness. On the other hand, it may reflect that Hirayama never did make an appropriate analysis and so missed his opportunity to present a persuasive case for the lung cancer effects of passive smoking. He should give it another try. (6) Finally, it is important to be suspicious. Things may not be what they seem. Ostensibly, Hirayama was working with a population of nonsmoking wives. Actually, he was working with a population of wives who reported themselves to be nonsmokers. The distinction is important and should influence our interpretation of the results.