We also assessed whether cases could have differentially discon tinued aspirin use for any reason, biasing our results by artifactually increasing risk in the unexposed.
We repeated the multivariate anal yses in Table 4, restricting the referent group to 5047 persons who stated that they had used no aspirin for 5 years or more.
The trend tests remained statistically significant for men and women combined for cancers of the esophagus, stomach, and colon, suggesting that for these cancers aspirin avoidance by cases did not explain the relation ships.
Only for rectum cancer was the trend weakened by restricting the referent group to long-term nonusers. We also tested whether the relative risk of fatal digestive tract cancers was higher among persons who had recently discontinued aspirin use ( 5 years).
No consistent pattern was seen for the four sites.
The risk was slightly higher among the short-term nonusers
for rectum cancer (relative risk, 1.28) but this was not statistically significant (P = 0.8) and was not seen for other digestive tract sites.
Therefore, changes in medication status due to undiagnosed cancer do not explain our findings.