Maté Drinking and Esophageal Squamous Cell Carcinoma in South America: Pooled Results from Two Large Multicenter Case–Control Studies

Maté tea is an infusion made from leaves of the tree Ilex paraguariensis, a member of the Aquifoliaceae (holly) family (1, 2). It is a nonalcoholic beverage consumed throughout southern South America, and is gaining broader acceptance in other areas of the world as a tea and dietary supplement based on purported health benefits, such as lowered cholesterol levels, improved cardiovascular health, and obesity management (2, 3). However, studies have linked maté consumption with esophageal squamous cell carcinoma (ESCC), as well as cancers of the oral cavity, pharynx, larynx, lung, kidney, and bladder (4–13). The International Agency for Research on Cancer (IARC) designated hot maté drinking a probable human carcinogen (group IIA; 1). Proposed carcinogenic mechanisms include thermal injury from repeated high temperature exposure and exposure to polycyclic aromatic hydrocarbons (PAH), a production-related contaminant (1, 14–17).

Although studies have associated maté consumption with ESCC, there has been no quantitative evaluation of the relationship between ESCC and total exposure, as measured by liters/day–year (LPDY), the product of lifetime mean liters/day (LPD) and years of consumption. In addition, evaluations of potential effect modifiers, such as age, sex, cigarette smoking, and alcohol consumption, have been limited.

We pooled data from two large case–control studies of ESCC, one an aggregation of five component studies. Our goals were to evaluate (i) the quantitative relationship between ESCC and LPDY of maté consumption, (ii) the impact of LPD on the strength of association, and (iii) potential effect modifiers, including maté temperature, sex, age, cigarette smoking, and alcohol consumption.

Cases included patients who were incident between 1990 and 2004, ages 35 to 85 years in medical records of the Oncology Institute Cancer Registry with histopathologically confirmed ESCC. Patients had to be mentally competent for interview, diagnosed within the previous 4 months and resident in Uruguay for ≥10 years (5).

Controls included patients admitted to the same institution during the same period with conditions unrelated to tobacco smoking and alcohol consumption and with comparable residency, and were frequency matched to cases by age and sex. Within the frequency matched groups, investigators enrolled greater numbers of female controls.

Interviews occurred shortly after hospital admittance. Questionnaires collected information on demographic and socioeconomic characteristics, personal and family history, maté drinking, alcohol consumption, and tobacco smoking. For alcohol intake, we calculated mL of ethanol per day by summing ethanol per day for a standard serving of each beverage type.

Between 1986 and 1992, investigators conducted four hospital-based case–control studies of ESCC in Argentina, Brazil, Paraguay, and Uruguay, the latter independent of the Uruguay study described above. Investigators further extended this Uruguay component, which represented a fifth study. IARC coordinated the studies, which we have collectively denoted as the IARC multinational study. Results have been published previously (4, 18–22). The components used similar protocols and questionnaires, allowing for local adaptations.

Cases included histologically confirmed patients with ESCC (the Paraguay component also accepted cytologic or radiologic diagnoses), diagnosed within the previous 3 to 6 months, resident in the area for ≥5 years and competent for interview. In Argentina, cases were ascertained from the 10 main hospitals of greater La Plata (19). In Brazil, cases were ascertained from the 8 main hospitals and 3 radiotherapy units of Port Alegre and Pelotas, at the time the 2 largest cities in the state of Rio Grande do Sul (21). In Uruguay, cases came from the 4 main hospitals in Montevideo, which covered about 45% of the population of the city and about 55% of the population of the rest of the country (4, 22). Investigators checked identification numbers and names to ensure there was no duplication or overlap of cases among the various Uruguay component studies. In Paraguay, subjects were ascertained from the 4 hospitals, private clinics, pathology laboratories, and radiology clinics in Asuncion (20). The case participation rate in all studies was high (90.0%–99.2%).

Controls who were admitted during the same period were frequency matched to cases on hospital, gender, age, and residence period, and included patients with diseases unrelated to alcohol or tobacco. Diagnostic categories were given previously (13, 19–22). Investigators replaced controls that refused to participate, except in Paraguay, although control participation rate was high (97.0%).

Questionnaires ascertained information on demographic and socioeconomic characteristics, tobacco smoking, alcohol drinking, and consumption of hot beverages, including maté, tea, and coffee. Beverage temperatures were self-assessed. Proxy interviews were not accepted.

The Institutional Review Board or Research Ethics Committee for each study approved data collection and, if required, participation in the pooling.

For categorical variables, we computed ORs using standard logistic regression (23). For continuous LPDY, d, ORs were not log-linear. We thus fitted the model |${\rm OR}(d, z)\, = \,\exp (\alpha z) \times {\rm OR}(d)$|⁠, where

z was a vector of adjustment variables with parameters α, whereas β was the excess OR per LPDY (EOR/LPDY), a measure of strength of association. We replaced d with |$d \times \exp \{\theta {\rm ln}(d)\} \, = \,d^{1 + \theta}$| and used the likelihood ratio to test no departure from linearity |${\rm (}\theta \,{\rm =}\,{\rm 0)}$|⁠.

We evaluated effect modification by examining variations in the trend of ORs by LPDY across a categorical factor (f). For factor f with S categories, s = 1,…,S, we fitted

where β_s parameters replaced β and d_s equaled d within category s and zero otherwise. If f was unrelated to maté consumption, for example, sex, then z included f as an adjustment variable. If f was related to maté consumption, for example, LPD or beverage temperature, then z did not include f because no adjustment in never-drinkers was required. We compared deviances for model (1) and model (2) to test homogeneity of slopes, |$\beta _1 \, = \, \ldots \, = \,\beta _s \, = \,\beta$|⁠, i.e., no effect modification. We replaced d_s with |$d_s \times \exp \{\theta _s {\rm ln}(d_s)\}$| to test no departure from linearity within category |$(\theta _s \, = \,0)$|⁠.

We used the Epicure program to estimate ORs and 95% confidence intervals (CI), fit models, and derive likelihood-based CIs for β estimates (24).

Analyses adjusted for joint categories of study/component (6 levels), cigarette smoking in pack-years (0, <30, 30–39, 40–59, 60–79, and ≥80) and alcohol consumption in mL ethanol/d–y (0, <1,170, 1,170–2,439, 2,440–4,679, 4,680–9,359, and ≥9,360), and for age (<55, 55–64, 65–74, and ≥75 years), sex, cigarettes/d (<10, 10–19, 20–29, and ≥30), mL ethanol per day (<32, 33–77, 78–155, and ≥156), years of education (<3, 3–5, and ≥6 for the Uruguay study and <4, 4–6, and ≥7 for the IARC study), and for the Uruguay study income (<US$120, ≥US$120, missing) and residency (urban, rural).

ORs by LPDY increased linearly in the IARC data and in the Uruguay data, but only among maté drinkers in the latter. For Uruguay data, we defined a fixed offset to adjust for ever and never maté drinkers using the model |${\rm OR}(d)\, = \,\exp \{\alpha {\rm I}(d)\} \times \{1 + \beta d\}$|⁠, where I(d) equaled one for d > 0 and zero otherwise. The estimate, |$\exp \{\alpha \}$|⁠, was 2.42 (95% CI, 1.5–2.9), and represented the LPDY-adjusted OR of ever relative to never-consumed maté. A detailed examination identified a small subgroup responsible for the excess. The subgroup included male (3 cases and 53 controls) and female (1 case and 61 controls) urban residents who abstained from alcohol, with ORs for ever-consumed maté of 4.24 (95% CI, 1.1–16.7) and 13.8 (95% CI, 1.8–105.8), respectively. We fixed the offset to −ln(4.24) and −ln(13.8) for Uruguay male and female urban residents who never consumed alcohol or maté and zero otherwise. The offset essentially served to replace the observed case to control odds with the expected odds, eliminating the nonlinearity. See details in Supplementary Materials and comments in the Discussion. The use of a fixed offset was an a priori decision, due to a concern about the possibility of broad impact on ORs from this small, highly influential subgroup. Alternatively, we could have introduced an indicator variable for this subgroup and estimated its effect, or have omitted these subjects. Regardless of approach, inference was unaffected.

There were 1,400 cases (1,085 males and 315 females) and 3,229 controls (2,279 males and 950 females; Table 1). ORs increased with pack-years of smoking, cigarettes/d, cumulative alcohol consumption, and alcohol intensity in both studies (P < 0.01). ORs increased with use of mixed/black-only tobacco cigarettes compared with blond-only tobacco cigarettes, achieving statistical significance in the IARC study and the pooled data.

In the Uruguay study, 95.6% of cases and 87.5% of controls and in the IARC study 92.9% of cases and 87.0% of controls ever drank maté. Among drinkers, Uruguay cases had greater mean intensity, duration, and total intake (1.2 LPD, 52.3 years, and 64.8 LPDY, respectively) compared with IARC cases (1.1 LPD, 47.0 years, and 54.3 LPDY). For controls, these maté-related metrics were also greater in the Uruguay study (1.1, 50.8, and 55.3) than in the IARC study (0.9, 44.9, and 40.8). Intake for the two Uruguay components of the IARC study was comparable with intake for the Uruguay study, and generally exceeded intake for the Argentina, Brazil, and Paraguay components of the IARC study (see Supplementary Table SB1).

The overall adjusted OR for ESCC by ever compared with never use of maté was 1.60 (95% CI, 1.2–2.2; Table 2), ORs by cumulative maté consumption and maté intensity increased in each study and the pooled data, with stronger associations in the IARC data. The offset modification greatly influenced the Uruguay results, as without the offset ORs were 1.0, 2.3, 2.8, 3.5, 2.0, and 3.3 for LPDY (P < 0.01) and 1.0, 2.5 2.8, 3.2, and 2.6 for LPD (P = 0.04) for their respective categories.

We evaluated ORs by self-reported maté temperature, warm, hot, or very hot, and found that ORs increased significantly with temperature, although the OR for warm maté consumption was not statistically significant. Few users ceased consumption (8.4% and 12.0% in Uruguay and IARC controls, respectively) and ORs varied inconsistently with years since cessation. Among drinkers, ORs were increased at younger age at first use in the Uruguay data and unrelated in the IARC data.

OR trends were homogeneous across studies for maté temperature (P = 0.85) and cessation (P = 0.84), but differed for cumulative intake, intensity and age at first use (P < 0.01; not shown).

We first examined LPD as a modifier of the LPDY association that is whether the strength of association varied by intensity or alternatively whether for a fixed total LPDY low intensity for long duration resulted in greater, equal, or lesser risk compared with high intensity for short duration. For joint categories of LPDY and LPD, ORs relative to never-drinkers increased with LPDY within each LPD category (Fig. 1A–D, solid symbol), with trends consistent with linearity, except for 1.0–1.9 LPD (P = 0.03). EOR/LPDY estimates for the 4 LPD categories were 0.001, 0.006, 0.008, and 0.007, respectively, revealing minimal variation in strength of association (test of homogeneity; P = 0.35). Across the full range of continuous LPDY, a linear relationship described maté-related ORs (Fig. 1E; P = 0.76 for the test of no departure from linearity). The EOR/LPDY estimate with 95% CI was 0.009 (0.005–0.014). After omitting low-intensity drinkers (LPD < 0.5), model fit improved slightly (dash line), and the EOR/LPDY estimate was 0.012 (95% CI, 0.007–0.020).

There was significant variation of EOR/LPDY estimates with temperature, years since maté cessation and age at first consumption (P < 0.01; Table 3). The EOR/LPDY estimates increased with temperature, 0.004 (95% CI, −0.002–0.013), 0.007 (95% CI, 0.003–0.013), and 0.016 (95% CI, 0.009–0.027) for warm, hot, and very hot consumers, respectively. In warm maté consumers, ORs by LPDY categories increased monotonically; however, the test of no trend did not reject (P = 0.14). EOR/LPDY estimates varied with years since cessation, 0.009 (95% CI, 0.004–0.015), 0.020 (95% CI, 0.006–0.044), and 0.005 (95% CI, −0.003–0.022) for 0, 1 to 4, and ≥5 years since cessation, respectively, but was not monotonic. Because prodromal symptoms may have influenced consumption, we applied a post hoc categorization <5 and ≥5 years since cessation and found that the EOR/LPDY estimate was greater in current and recent former drinkers, 0.009 (95% CI, 0.004–0.015) than in long-term former drinkers 0.005 (95% CI, −0.003–0.021), with the difference nearly significant (P = 0.08). Subject ages younger than 12 years at first maté consumption exhibited the strongest association, 0.012 (95% CI, 0.006–0.020), compared with older initiators, 0.005 (95% CI, 0.001–0.011) at 12 to 16 years and 0.008 (95% CI, 0.001–0.017) at ≥17 years (P = 0.01).

The LPDY association was statistically homogeneous by sex (P = 0.29), although category-specific ORs by LPDY were larger in females and EOR/LPDY estimates were 0.013 (95% CI, 0.004–0.032) in females and 0.007 (95% CI, 0.003–0.013) in males. For attained ages younger than 65, 65 to 74, and ≥ 75 years, EOR/LPDY estimates were 0.015 (95% CI, 0.007–0.029), 0.006 (95% CI, 0.001–0.015), and 0.006 (95% CI, 0.001–0.016), respectively, suggesting an enhanced trend at younger ages (P = 0.14). A post hoc categorization of ages younger than 65 and ≥65 years resulted in EOR/LPDY estimates of 0.015 (95% CI, 0.005–0.033) and 0.006 (95% CI, 0.001–0.013), which differed significantly (P = 0.05).

Modification of the EOR/LPDY was inconsistent for smoking related variables. ORs by LPDY varied by cigarette smoking status (P = 0.02), with the strongest association in never-smokers, 0.018 (95% CI, 0.007–0.038), decreasing in former, 0.009 (95% CI, 0.002–0.022), and current-smokers, 0.003 (95% CI, −0.001–0.009). However, ORs at higher LPDY levels drove the variation, as ORs for less than 70 LPDY were similar (Table 3). For less than 70 LPDY, the EOR/LPDY estimates for never, former, and current smokers were 0.006 (95% CI, −0.004–0.023), 0.007 (95% CI, −0.003–0.026), and 0.018 (95% CI, 0.005–0.042), respectively, and homogeneous (P = 0.32). Tobacco type was a significant effect modifier, with trends increased in never-smokers and in mixed/black-only tobacco users, but not in blond tobacco users (P < 0.01). However, for less than 70 LPDY, ORs for blond-only tobacco smokers were elevated and the EOR/LPDY estimate was 0.013 (95% CI, 0.001–0.035), consistent with estimates for never and mixed/black-only tobacco users (P = 0.22).

Category-specific ORs by LPDY were highest in never alcohol drinkers, with EOR/LPDY estimates for never, former, and current alcohol drinkers of 0.017 (95% CI, 0.006–0.037), 0.004 (95% CI, −0.001–0.014), and 0.008 (95% CI, 0.002–0.018), respectively. However, the test of homogeneity of trends was not rejected (P = 0.12). A post hoc evaluation of EOR/LPDY estimates for never and ever alcohol drinkers rejected homogeneity (P = 0.04).

ORs by LPDY were consistent with linearity among exposed in the Uruguay study (P = 0.91) and among all subjects in the IARC study (P = 0.11), with EOR/LPDY estimates of 0.003 (95% CI, −0.001–0.009) and 0.015 (95% CI, 0.008–0.025), respectively. Homogeneity of EOR/LPDY estimates was rejected (P = 0.01; Supplementary Table SB2).

Adjusted for study differences, variations in EOR/LPDY patterns across the potential modifiers were consistent for the studies, and tests of homogeneity of effect modification did not reject, except for age at first maté consumption (P < 0.01). The EOR/LPDY estimate was largest for ages younger than 12 years in the Uruguay study and for ages 12 to 16 years in the IARC study (Supplementary Table SB2.)

This analysis represents the first detailed assessment of the exposure–response association for maté consumption and ESCC risk and of the potential modifying effects for a broad range of factors. In particular, we evaluated (i) the relationship of ESCC and cumulative maté consumption, (ii) the influence of maté consumption intensity on the strength of association, and (iii) the impact of potential effect modifiers. The pooled results were consistent with the constituent studies (4, 5, 18–22), with marginal ORs increasing significantly with ever use, cumulative intake, and intensity. Our pooled results extended prior analyses to demonstrate that ORs increased linearly with LPDY (Fig. 1), rising to 2.0-fold for ≥100 LPDY consumers. Moreover, maté intensity did not alter the linear association with LPDY, suggesting that the main determinant of risk was cumulative intake and that for a given intake, higher intensity consumption for shorter duration, or lower intensity consumption for longer duration resulted in comparable ORs. We could not however, rule out an enhanced association in low (<0.5 LPD) intensity drinkers, although this enhancement may have reflected differential misclassification, with lower maté intensity cases underreporting cumulative intake.

Epidemiologic studies have linked ESCC to repeated ingestion of high-temperature liquids, such as tea, coffee, and maté (7, 15, 25–27), implicating thermal injury as a carcinogen. Although estimates of the association have varied, increased ORs with beverage temperature are observed in many countries and across diverse beverage types (15). Intraesophageal temperatures are sensitive to initial fluid temperature, time between sips, and sip volume, suggesting substantial inherent variability (28, 29). Moreover, temperatures were self-assessed, further increasing misclassification. Despite the substantial misclassification, the strength of association in the current analysis increased with temperature; EOR/LPDY estimates were 0.004, 0.007, and 0.016 for consumption at warm, hot, and very hot temperatures, respectively (Table 3), consistent with thermal injury damaging the epithelial lining of the esophagus and thereby directly affecting risk or enabling other factors. Experimental animal studies involving high-temperature liquids support this pattern (30–32). Nonetheless, risks for warm maté drinkers remain uncertain. Although category-specific ORs increased monotonically, the test of no trend was not rejected (P = 0.14).

Although there have been relatively few studies and results to date are not conclusive, studies have associated maté consumption with diverse cancer sites, including oral cavity, pharynx, larynx, lung, kidney, and bladder (4–12, 16). Thus, the etiology of ESCC may potentially involve maté-associated nonthermal factors. Attention has focused on PAHs, in particular benzo[a]pyrene, a possible production-acquired contaminate (16, 33), which IARC has classified as a human carcinogen (34, 35). Because cigarette smoke contains PAHs, residual confounding may have influenced maté-related ORs (2, 8). However, substantial confounding in the current analysis seems unlikely because among users the Pearson correlation between LPD of maté and cigarettes/d was small (0.11 in controls), urinary measurements of 1-hydroxypyrene glucuronide, a stable PAH metabolite, correlated positively with maté consumption (14) and, importantly, we observed significant trends in ORs with LPDY in never-smokers and in smokers after extensive smoking adjustment.

Conclusions were not definitive about modification by other maté-related variables. Cessation of maté drinking significantly modified EOR/LPDY estimates (P < 0.01); however, the largest estimate occurred in recent (1–4 years) former drinkers (P = 0.020), with lower estimates in both current (P = 0.009), and long-term (≥5 years) former drinkers (P = 0.005; Table 3). Because prodromal symptoms may have influenced responses, we recalculated EOR/LPDY for less than 5 and ≥5-year cessation and found estimates of 0.009 and 0.006, respectively, indicating reduced maté effects with increased cessation (P < 0.01). This result agreed with two previous studies that found higher ORs in former compared with current drinkers (4, 7), but not another, which found monotonically decreasing ORs with cessation (13). Our analyses were necessarily limited due to few long-term quitters (74 cases and 167 controls). Younger ages at initiation increased the strength of the LPDY association; however, interpretation was problematic because variations in EOR/LPDY estimates were inconsistent across studies (P < 0.01; Supplementary Table SB2).

Cumulative maté effects were statistically homogeneous by sex for each study and the pooled data; however, category-specific ORs with LPDY and EOR/LPDY estimates were greater in females. These results corresponded to previous findings for the IARC study (4). Although not significant, consistency in the enhanced effects in females suggested the need for further evaluation in other study populations.

No definitive conclusions were possible for the roles of age, cigarette smoking and alcohol as effect modifiers. The largest EOR/LPDY estimate occurred for ages <65 years in each study and in the pooled data (Supplementary Table SC2); however, homogeneity of EOR/LPDY estimates was not rejected (P = 0.14). Only under post hoc evaluation did EOR/LPDY estimates vary significantly for ages <65 years. In the pooled data, smoking status and type of tobacco were significant modifiers of the maté association, but higher LPDY consumers drove results. For <70 LPDY (representing 83% of controls), EOR/LPDY estimates were 0.006, 0.007 and 0.018 for never, former and current smokers, respectively, and homogeneous (P = 0.32), which was concordant with a previous result (13). Estimates were −0.002, 0.013, 0.020 in never-smokers, blond-only, mixed/black-only tobacco users (P = 0.22). Finally, although ORs and the EOR/LPDY estimate were greatest in those who never consumed alcohol and homogeneity was not statistically rejected (P = 0.12), the differential EOR/LPDY estimates occurred only in the IARC study (Table 2).

Initial analyses revealed that ORs by LPDY increased linearly in both the IARC and Uruguay datasets, with linearity in the latter dataset occurring only in maté consumers. Although maté-related ORs could vary in populations due to different methods of preparation and consumption, trends with consumption should be roughly comparable. Exploratory analysis of the Uruguay dataset identified a small subgroup of urban residents who never consumed alcohol (4 cases and 114 controls) with significant ORs by ever-consumed maté of 4.2 for males and 13.8 for females. The inclusion of a fixed offset eliminated nonlinearity in the Uruguay data. An alternative approach could have specified a nonlinear relationship for ORs with LPDY in the Uruguay data, and derived an offset for the IARC data that induced a curvilinear pattern to mimic the Uruguay data. We did not apply this approach, as it increases model complexity and because linearity typically represents the preferred first-order approximation (Occam's Razor). A second alternative could have omitted the offset and used a combined linear relationship for the IARC data and a curvilinear relationship for the Uruguay data. Under this approach, the inference in Table 3 was largely unchanged, except EOR/LPDY variations were not significant for attained age (P = 0.87 and P = 0.63 for post hoc categories of ages <65 and ≥65) but were significant for alcohol status (P < 0.01; not shown).

In summary, our results confirmed the hypothesis that drinking maté increases risk of ESCC, with ORs consistent with a linear relationship in cumulative intake. Moreover, the strength of association with cumulative intake was not influenced by consumption intensity, so that greater daily consumption for a shorter duration or less daily consumption for a longer duration resulted in comparable ORs. The increased ORs also occurred at all beverage temperatures, but were greater with higher maté temperatures.

No potential conflicts of interest were disclosed.

Conception and design: J.H. Lubin, C.C. Abnet, P. Boffetta, B.I. Graubard, N. Muñoz, S.M. Dawsey

Development of methodology: J.H. Lubin, G. Acosta, C. Victora, X. Castellsagué

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): E. De Stefani, G. Acosta, P. Boffetta, C. Victora, N. Muñoz, H. Deneo-Pellegrini

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): J.H. Lubin, C.C. Abnet, P. Boffetta, B.I. Graubard, X. Castellsagué

Writing, review, and/or revision of the manuscript: J.H. Lubin, E. De Stefani, C.C. Abnet, G. Acosta, P. Boffetta, C. Victora, B.I. Graubard, N. Muñoz, H. Deneo-Pellegrini, S. Franceschi, X. Castellsagué, A.L. Ronco, S.M. Dawsey

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): X. Castellsagué

This work was supported by the IARC, Lyon, France [Uruguay study; to E. De Stefani, H. Deneo-Pellegrini, A.L. Ronco, and G. Acosta (IARC/CRA ECE/98/17)]. X. Castellsagué, C. Victora, and E. De Stefani, the IARC study and its components were supported with grants from IARC to Fondo de Investigaciones Sanitarias (Spain; FIS 97/0662), Fondo para la Investigacion Cientifica y Tecnologica (Argentina), Institut Municipal d'Investigacio Medica (Barcelona), Fundacao de Amparo a Pesquisa no Estado de Sao Paulo (01/01768-2), and from the European Commission (IC18-CT97-0222), the Comision Honoraria de Lucha contra el Cancer (Montevideo, Uruguay; 5471-066), the International Union Against Cancer, the International Cancer Research Data Bank Program of the National Cancer Institute, and the NIH (Bethesda, MD; N01-CO-65341). J.H. Lubin, C.C. Abnet, B.I. Graubard, and S.M. Dawsey were supported by the Intramural Research Program of the NIH, National Cancer Institute, Division of Cancer Epidemiology and Genetics.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.