The most important finding of our study was the significant increase in risk of bladder cancer associated with maté consumption. Most previous studies on maté drinking and human cancer [7–9, 13–15] were conducted in cancer sites on which maté ingestion may be in direct contact with the epithelium. This fact reinforces the hypothesis that maté drinking acts by thermal injury. A complementary mechanistic hypothesis is that maté could contain chemical carcinogens. An unpublished chemical analysis (R.D. Adams and D. Hoffmann, personal communication) failed to obtain evidences of the presence of N-Nitroso compounds. On the other hand, one study reported the presence of large amounts of benzo [a]pyrene in eight commercial samples of maté leaf bought in Germany; however, the beverage resulting from infusion of the leaves contained only 0.02–0.12 micrograms/liter . Other chemical studies suggested that maté could have carcinogenic effects due to its contents in tannins [17, 18]. Experimental studies in rats and mice showed that caffeic acid (a metabolite of cholorogenic acid which is abundant in maté leaves) has carcinogenic effects on the kidney . Experimental studies in Brazil suggested that maté displayed mutagenic and clastogenic activities in cell cultures [20, 21]. Finally, in an experimental study in animals, Roffo  was able to produce squamous cell carcinomas of the skin after painting this organ with "tar" from maté. Recently, Fagundes and colleagues (personal communication) studied the urine in maté drinkers and found evidences of high levels of PAH derived from tobacco and maté.
Since there exist the possibility that maté could contain chemicals with carcinogenic activity, it was suggested that epidemiologic studies on maté drinking and cancer sites not related with maté temperature in humans could be a useful model to test this possibility. Up to date five such studies were conducted. Three case-control studies conducted in Uruguay [2, 23, 24] displayed significant positive associations between maté drinking and lung, renal cell and bladder cancers. One additional case-control from Argentina study was characterized by a strong direct association between coffee consumption and bladder cancer, but no effect of maté drinking . Finally, in a recent case-control study conducted in Córdoba, Argentina, maté was associated with bladder cancer risk . No data were previously reported on effect of temperature of maté on bladder cancer risk.
Thus, it remains the strong possibility that maté drinking could be carcinogenic to cancer sites not related with a direct contact with the beverage. In other words, our findings of a strong direct association between maté drinking and bladder cancer risk are partially supported by previous chemical, experimental and epidemiological studies in cancer of the bladder and of other organs.
Another possibility is related with residual confounding of maté drinking by tobacco smoking. In the present study, the analysis of maté drinking was adjusted for smoking status, cigarettes per day and years since quit. Moreover, since there were a sizeable number of never smokers, this allowed to estimate the effect of maté drinking among this subset. Although results among never smokers were less precise than those obtained in the whole study population, the increased risk for heavy drinking of maté obtained in this subgroup detracts from the hypothesis of confounding by tobacco smoking.
Concerning the effect of other non-alcoholic beverages, both coffee and tea drinking were directly associated with bladder cancer risk, mainly when coffee is ingested with milk. The latter result was unexpected and can be due to chance or residual confounding. It might also indicate the presence of bladder carcinogens in black tea. Previous studies on tea intake and bladder cancer risk yielded controversial results. At least one prospective study  and two case-control studies [28, 29] reported significant increases in risk of bladder cancer, whereas the remaining studies failed to show an association between tea consumption and bladder cancer risk . Thus, there is no consistent evidence of a carcinogenic effect of tea on bladder mucosa. The possibility of residual confounding from smoking is very difficult to exclude since the OR's for tea drinking among never smokers was not clearly elevated, although the number of tea drinkers was rather small. Chance remains an additional possible explanation of these findings.
We did not ask about intake of water. However, this would create a bias only if drinking of water were associated with bladder cancer risk and is correlated to drinking of the beverages under study. Indeed, low overall fluid intake (represented mainly by water) has been suggested as a possible risk factor for bladder cancer , but this would act as a positive confounder in our study only if high intake of maté, coffee or tea would associated with low overall fluid intake. The lack of information on source of drinking water and potential exposure to drinking water contaminants such as chlorination by-products and nitrates is an additional limitation of our study.
The present study, as other hospital-based case-control studies, is subjected to several limitations. Selection bias is almost impossible to rule out. We tried to minimize this bias by frequency matching controls and cases on age, sex and residence. Although matching for the latter variable was not complete, we adjusted for it in all analyses. Furthermore, we have adjusted the risk estimates for possible determinants of selection of cases and controls, such as urban/rural status and education.
Misclassification bias is also difficult to exclude. It is important to note that the role of nonalcoholic beverages in bladder cancer risk is unknown, both by the general population, the hospital population and the interviewers. Thus, is rather unlikely that differential or nondifferential exposure to maté drinking has occurred in our study. Furthermore, although current intake of non-alcoholic beverages among controls might have been modified by their disease, we consider unlikely that the use of hospital controls has created a bias in the prevalence of lifetime exposure. A further limitation is the relatively small sample size of the study, which reduced the power of detecting interactions between risk factors (e.g., differences in maté -related OR according to smoking status) and differences according to subtle differences in exposure (e.g., whether the effect of coffee with milk is really different from that of black coffee, as suggested by our results). On the other hand our study has strengths. Perhaps the more important strength is related with the high response rate in both series of patients (cases and controls). Another strength is the absence of proxy interviews. Finally, the sizeable number of never smokers, which allowed to estimate OR's of maté drinking among these subset of patients, is a strength of the study.