Abstract
Smoking has been suggested to increase the risk of nonโHodgkin's lymphoma (NHL) but the results of epidemiological studies have been inconsistent. The aim of this work was to assess whether the findings of individual studies might have arisen by chance, bias or confounding and whether any associations found between smoking and NHL represent causeโandโeffect. Reports of the association between smoking and NHL were identified from Medline. Confidence intervals on relative risks and odds ratios, use of multiple comparisons, and information on source, direction, actual existence and size of potential biases and confounding and features of any associations were abstracted. Four out of five cohort studies found no association between current smoking and NHL but three may have been biased against doing so. One found an association with follicular lymphoma but without a convincing exposureโrisk gradient. The fifth found a strong association and an exposureโresponse gradient with ever smoking but excluded living cases from the endโpoint. Only one study found an association with past smoking which lacked features of causality. Eight out of 14 caseโcontrol studies found no association between current and/or past smoking and NHL but five may have been biased against doing so. Of six positive studies, three involved multiple comparisons, the association of one became nonโsignificant after eliminating bias, four did not explore features of causality and one found an association only in heavy smokers, particularly under 45 years old. There are no grounds to reject the null hypothesis but associations should continue to be sought particularly in subgroups of smokers and with NHL subtypes. Copyright ยฉ 2001 John Wiley & Sons, Ltd.