Effect of Missing Data.
We tested the possibility that the exclu sion of persons who did not specify both frequency and duration of aspirin use might bias our findings.
In the rate analyses, cancer mor tality among persons who gave no information on aspirin use was similar to or slightly below that of nonusers; while among persons who specified frequency but not duration of use, risk resembled that in users.
Exceptions were cancer of the esophagus and lung in men. Men who did not specify the frequency of their aspirin use had a relative risk of esophagus cancer of 1.10 (95% confidence interval, 0.82-1.46) and of respiratory cancers of 1.20 (95% confidence inter val, 1.10-1.30).
In multivariate analyses, the relative risk estimates for digestive tract cancers in persons who specified their aspirin usage became somewhat stronger (i.e., further below 1.0) when persons with missing or incomplete data were excluded from the models.
We present in Tables 4 and 5 only the relative risk estimates for aspirin from models which include the missing data categories, since in our data, these underestimate the association with aspirin.
Effect of Missing Data. We tested the possibility that the exclu sion of persons who did not specify both frequency and duration of aspirin use might bias our findings. In the rate analyses, cancer mor tality among persons who gave no information on aspirin use was similar to or slightly below that of nonusers; while among persons who specified frequency but not duration of use, risk resembled that in users. Exceptions were cancer of the esophagus and lung in men. Men who did not specify the frequency of their aspirin use had a relative risk of esophagus cancer of 1.10 (95% confidence interval, 0.82-1.46) and of respiratory cancers of 1.20 (95% confidence inter val, 1.10-1.30). In multivariate analyses, the relative risk estimates for digestive tract cancers in persons who specified their aspirin usage became somewhat stronger (i.e., further below 1.0) when persons with missing or incomplete data were excluded from the models. We present in Tables 4 and 5 only the relative risk estimates for aspirin from models which include the missing data categories, since in our data, these underestimate the association with aspirin.
การแปล กรุณารอสักครู่..