Esophageal Thermal Exposure to Hot Beverages: A Comparison of Metrics to Discriminate Distinct Consumption Habits

Consumption of hot beverages is a putative risk factor for esophageal cancer—the seventh commonest cancer worldwide, with the sixth highest number of cancer deaths in 2018 (1). Reports dated as early as 1939 speculate thermal irritation as a risk factor for esophageal cancer (2). In 2016, the International Agency for Research on Cancer (IARC), within their Monograph Program on the Identification of Carcinogenic Hazards to Humans, classified drinking “very hot beverages” (>65°C) as “probably carcinogenic to humans” (Group 2A; ref. 3) and recently listed it as a high priority for reevaluation (4). Positive associations between hot beverage consumption and esophageal cancer—namely squamous cell carcinoma (ESCC)—have been reported in most of the world's notable high incidence regions, including prospective evidence from the Chinese Kadoorie Biobank cohort (5) and the Iranian Golestan cohort, the latter reporting an HR of 1.6 for first sip temperature of tea drinking >60°C compared with <60°C (6).

Thermal injury exposure assessment methods are expected to have a large impact on the magnitude of its association with ESCC; however, there is no “gold-standard” exposure assessment method or metric. Previous epidemiologic studies, with the exception of the Golestan cohort, share a common limitation—they relied on self-reported perceived drinking temperatures, e.g., “very hot” versus “warm” beverages. In the few studies to objectively measure temperatures, either for descriptive purposes or in relation to disease, temperature at first sip was commonly utilized as the metric of thermal exposure (6–8). However, a reliable metric of thermal exposure might quantify the heat transfer from the swallowed liquid to the esophageal mucosa, i.e., heat being a function not only of temperature of the liquid, but also its volume. Indeed, in a 1972 publication in Gut by de Jong and colleagues (9), the authors performed a unique study of the relationship between beverage temperature and temperature measured in the lower esophagus during drinking through placement of a temperature sonde within the esophagus. In addition to a positive relationship between what the authors called “intra-esophageal temperature” and initial drinking temperature, they observed the expected positive association with sip volume, for example, with intraesophageal temperature reaching approximately 46°C upon either drinking 15 mL of a 65°C coffee or 20 mL of 60°C coffee. Despite these important findings, we are not aware of a thermal exposure measurement protocol that has quantitatively incorporated both sip temperature and volume.

In terms of the practicalities of measuring beverage drinking temperatures, studies have employed a variety of methods. In a UK study (10), a single tea temperature was recorded at first sip—defined as the point at which the beverage had cooled enough to drink in successive gulps. In a study of coffee drinkers (11), consumers were given a choice of hot or cooler coffee and presented with options to alter the temperature. A single temperature was recorded at the desired sipping temperature. In a high ESCC incidence region of Brazil, maté temperatures were assessed by single temperature measurements of maté inside the gourd (12). In the abovementioned Golestan cohort (8), participants were asked to sip their tea at 5°C intervals—determined from a proxy cup—until the desired drinking temperature was reached and recorded. In another high ESCC risk area in Tanzania, a cross-sectional study (7) of tea drinking habits in Kilimanjaro employed the Golestan protocol, additionally recording the times and temperature when half and the entire cup had been drunk, which demonstrated large differences in cooling curves by type of drink. The Iranian study observed agreement between self-reported and measured drinking temperatures (8). First sip temperature was also employed in a recent Kenyan cross-sectional study (13) of tea drinking temperatures, evidence that this metric is beginning to be routinely adopted.

Measuring temperatures throughout the drinking episode is possible; and accurate exposure assessment is necessary for prospective studies, cross-sectional community surveys (i.e., to enable reliable inter- and intrapopulation comparisons), and studies investigating the as yet unknown carcinogenic mechanism of thermal injury. Therefore, as part of ongoing investigations into the etiology of ESCC, we:

• (I) developed an improved measurement protocol of hot beverage/semiliquid thermal exposure which incorporates both sip volume and temperature;

• (II) used this protocol to record thermal exposure measurements in two groups of people who were expected to differ substantially in their thermal exposures: Group A (low exposure) of hot beverage drinking by Europe-based international researchers; Group B (high exposure) of hot porridge consumption in a high ESCC incidence region of China;

• (III) examined how the following metrics of thermal exposure discriminated Groups A and B: the temperature at first sip (the most commonly used metric), mean drinking temperature (across sips), sip-weighted mean drinking temperature, and, using de Jong and colleagues' (9) results, predicted intraesophageal liquid temperatures (IELT) as a function of beverage temperature and sip size.

Ethical approval for the study was received in China from the Ethics Committee of the Cancer Institute and Hospital, Chinese Academy of Medical Sciences (approval number NCC2017YJZ-001), and at the IARC (IEC Project No. 17-04-A1). Written-informed consent was provided by all participants, and the study was conducted in accordance with the International Ethical Guidelines for Health-related Research Involving Humans developed by the Council for International Organizations of Medical Sciences.

Thermal exposure metrics in two groups of individuals, expected to differ considerably, were compared. Group A was comprised of volunteer staff, invited via internal email, at IARC in Lyon, France, during the months of December 2017 to January 2018 (8 am–6 pm) who drank a hot beverage in an indoor office with an ambient temperature of 23°C. Volunteers were asked to choose from a selection of beverages (black tea, green tea, or black coffee with or without the addition of sugar only). For Group B, a subset of 52 participants of an ongoing population-based cohort study (14), residing in two rural villages in Cixian County, Hebei Province, China (Supplementary Fig. S1), were recruited. During November 2017 and January 2018, participants were visited at their homes when taking their morning porridge. Ambient temperature was recorded where the porridge was being consumed (sometimes outdoors) and ranged from 5 to 23°C.

Participants in both groups underwent an interview-administered questionnaire to obtain age, sex, and self-reported perceived hot beverage consumption habits.

In China (Group B), participants were asked to prepare two identical servings of porridge in their usual manner. This watery porridge is similar to the congee consumed in other regions of China, but made with maize instead of rice. One bowl was placed on a digital weighing balance for consumption and the other used for temperature measurement with a stainless steel temperature probe (Checktemp, HANNA Instruments). At the time of serving, a digital timer was started, and participants were instructed to begin eating their porridge when they normally would and in their usual manner. The time and temperature of the first sip were recorded, and the weight of the porridge before and after the sip to estimate sip volume (assuming 1 g of porridge ≈ 1 mL). This was repeated in what fieldworkers judged to be the middle of the meal and just before the end—giving measurements for three sips in total, in addition to the time taken to complete the meal. At IARC (Group A), the same protocol was employed for tea/coffee, with the exception that measurements were recorded for every sip throughout the drink to enable a detailed investigation of exposure metrics. To ease the capture of temperature at every sip, a Bluetooth-enabled thermometer (BlueTherm One, ETI Ltd.) with a rapid response penetration probe and smartphone application (HACCP Mobile) were used—capable of logging timestamped temperature readings on demand. This device is comparable with the one used in China, with the addition of data logging capability, and near-perfect agreement was found between both instruments when used simultaneously for validation (mean absolute temperature difference <0.3 ± 0.42°C between readings, n = 15 measurements; temperature range: 95–59°C). In addition, the procedure was repeated once more for each IARC participant on a different day to assess intraparticipant variability of different exposure metrics.

Prior to fieldwork, we preliminarily tested the performance of two alternative types of thermometer (Supplementary Fig. S2) on boiling water left to cool—a noncontact infrared thermometer (TFI 250, Ebro, Xylem Analytics) and a submersible temperature sensor and data logger (Tempo Disc, Blue Maestro Ltd.). However, the former was susceptible to interference from evaporating vapor, and the latter had slow instrumental response due to insulation from a waterproof casing (Supplementary Fig. S3).

In addition to the previously used exposure metrics of temperature and time at first sip and mean temperature across the drink, we investigated the utility of previously untested exposure metrics—mean temperature weighted for sip size and predicted IELT.

To estimate IELT, we extracted data from de Jong and colleagues' previous investigation of measured “intra-esophageal temperature” (9) and used it to model IELT. We note that the nomenclature “intra-esophageal liquid temperature” is used henceforth so as not to imply that the metric refers to tissue temperature, but likely reflecting the temperature of the liquid passing through the esophagus. We modeled IELT as a function of sip size (V_s in g) and sip temperature (T_s in °C), constraining IELT to be body temperature (37°C) when no sip was taken (V_s = 0), and having a theoretical asymptote at temperature T_s for large volumes, as follows: IELT = 37 - (T_s - 37)+2(T_s - 37)/(1 + exp (βV_s(T_s - 37)) for which predicted values are shown in Supplementary Table S1. However, its fit (Supplementary Fig. S4) could only be examined in the range of V_s from 5 to 20 g and T_s from 55 to 65°C, thus, as most sipping temperatures (T_s) exceed 65°C, many predictions of IELT are beyond the observed volume and temperature range. Random errors were then added to point IELT predictions, drawn from a normal distribution with mean = 0 and SD = 2.69, where 2.69 is the estimated between-participant SD in IELT for a given T_s and V_s, based on that observed for Jong's measured IELT values. A schematic of the different steps of deriving IELT is presented in Fig. 1. Sip-weighted mean temperature for each person was calculated across a single drinking episode as: (∑T_s × V_s)/∑V_s.

We estimated differences in the aforementioned metrics of drinking habits between the two exposure groups (China and IARC), using an unequal variance t test (Welch test) to compare sample means for continuous variables, and a χ² test for categorical variables. For the analysis of changes in sip weight over a drinking sitting, a multilevel model was fitted with sips nested within an individual's sitting, and taking a natural log-transformation to normalize the distribution of sip weights, thus the change in sips by temperature is reported on a relative scale. We also examined the correlation [Pearson (r)] of drinking metrics (natural log-transformed) with each other (separately for the two groups), and, for IARC participants only, correlations of a given metric between repeat sittings for the same participant.

Group A at IARC had 20 participants, each of whom had two drinks measured, giving 406 time-temperature-sip size measurements in total, and Group B from China had 52 participants who each consumed one bowl of porridge, with 156 measurements in total. The study groups differed in their demographic characteristics (Table 1). An equal sex and younger age distribution (all <40 years) was present for the IARC volunteers compared with those from China, who were >70% female and all >40 years old. There was a greater ethnic and cultural diversity among the IARC group, which included 55% White Europeans and volunteers originating from 4 different continents. Conversely, all Chinese participants were residents of the study area. Self-reported perceived consumption temperatures were higher at IARC. Although among Chinese participants approximately 60% categorized themselves in “hot” and approximately 40% in “warm,” 20% of the IARC volunteers categorized themselves in “very hot,” 60% in “hot,” and 20% selected “warm.” Similarly, no participant in China recalled having burned their mouth in the last month, whereas 90% of IARC volunteers selected at least once. Chinese participants opted toward faster self-reported consumption speeds compared with IARC—62% “fast” and 20% “fast,” respectively.

Visual observations were made of the manner in which individuals consumed their beverages. After pouring into mugs, IARC beverages drinkers typically waited before testing with their lips how hot drinks were before proceeding to take sips when comfortable. In China, porridge was served into bowls directly from the heat source. The porridge was of a watery consistency and drunk directly from the bowl like a beverage. Chopsticks were used to sweep the liquid into the path of the mouth, particularly toward the end of the meal. The visible temperature of the liquid at first sip, sometimes immediately after serving, was notable, as porridge could still be seen bubbling/steaming when visibly large sips were taken. Following these observations, a subjective assessment was made that thermal exposures were much higher in Chinese porridge consumers.

Agreeing with visual observations, statistically significant differences were found in objectively measured characteristics of thermal exposure between IARC beverage drinkers and Chinese porridge consumers (Table 2). IARC beverage drinkers started drinking after a mean 7.7 ± 3.5 minutes after pouring, at a mean temperature of 70.5 ± 7.8°C, 9.5°C [95% confidence interval (CI), 6.2–12.7] cooler than Chinese porridge consumers (79.9 ± 8.9°C), who took their first sip after a mean of just 2.1 ± 1.8 minutes after serving, 5.6 (95% CI: 6.8, 4.3) minutes sooner than the IARC volunteers. Chinese drinkers also finished their porridge more quickly, when it had cooled to a mean temperature of 75 ± 8.3°C, still 4.5°C hotter than the mean first sip temperature at IARC. Mean drinking durations (from first sip) were 6.3 ± 2.6 and 13.9 ± 4.4 minutes for China and IARC, respectively. This is of further note given the large difference in mean total serving size between both types of beverage, which was approximately 380 g for a serving of porridge and 190 g for a mug of tea or coffee. This resulted in a crude mean across-sip temperature of 77.1 ± 8.4°C in China, compared with 58.9 ± 4.3°C at IARC, i.e., a difference of 18.3°C (95% CI, 15.6–20.9), double the difference in the mean temperature at first sip.

Despite consuming at a higher temperature range, on average Chinese porridge consumers also took larger sips than IARC beverage drinkers, both at first sip (mean 20.9 g vs. 3.9 g) and across all sips (mean 22.3 vs. 16.1 g), but the variability around these mean sip sizes was also larger (e.g., SD 8.5 vs. 4.5 g). This resulted in significantly higher predicted IELTs (e.g., mean 61.0 ± 8.3°C vs. 42.4 ± 2.1°C; mean difference 18.3, 95% CI, 16.1–20.9) for Chinese porridge consumers compared with IARC beverage drinkers. Sip-weighted mean temperatures (76.9 ± 8.3°C vs. 56 ± 4°C; mean difference 21.0°C, 95% CI, 18.3–23.5) were also higher in this group. A characteristic drinking behavior was observed among IARC volunteers, whereby sip size increased by 11.4% (95% CI, 10.8–11.9) per 1°C decrease in temperature. This is illustrated in Fig. 2A and B, in which sip sizes can be seen to increase with decreasing sip temperatures, resulting in stable predicted IELTs throughout the drink as IARC volunteers took small sips in reaction to hot liquid early on. This tendency of increasing sip size was weaker in magnitude [1.8% (95% CI, 0.8–2.8) increase per degree decrease] among Chinese porridge consumers (Fig. 2C).

To visualize the difference in group separation according to alternative metrics, we plotted each individual's sip temperature in proportion to its sip size, with each person represented horizontally (Fig. 3A). For IARC beverage drinkers (blue points), sip sizes (graduated by point size) can be seen increasing as the drink cools, whereas the effect is small in China (red points). In Fig. 3A, individuals on the y axis are ordered by the commonly used metric of temperature at first sip, where there is appreciable overlap in exposure rank between the two groups. Alternatively, when ranked by the predicted IELT at first sip (Fig. 3B), the two groups are almost completely distinguishable from one another. The two metrics give different ranks as many IARC first sips were very hot, but they were too small in volume to have an appreciable effect on thermal exposure as measured by IELT or a sip-weighted temperature. The same degree of separation between the two groups was achieved by ranking by mean predicted IELT across sips, but to enable like-for-like comparison, first sip IELT is plotted in Fig. 3B.

Another reflection of the greater discrimination after incorporation of both size sip and temperature into metrics is that among the key metrics (bold in Table 2), the largest between-group difference was for sip-weighted temperature [China–IARC difference 21.0°C (18.3–23.5), t = 15.9], similar to the difference in mean IELT [18.5°C (16.1–20.9), t = 15.4] or mean temperature (18.3°C, t = 13.6), and all were double that of the temperature at first sip (9.5°C, t = 5.8).

To assess how alternative exposure metrics varied between-sittings (within-person), correlation was investigated with repeat sitting measurements, which was possible only for IARC volunteers. Overall, metrics incorporating (i) sip size and (ii) measurements from multiple sips in their derivation had slightly stronger correlations than those with only temperature (e.g., within-person correlation of mean predicted IELT: r = 0.76 for both vs. mean temperature: r = 0.71) or measurements for a single sip (e.g., r = 0.37 for first sip predicted IELT and r = 0.69 for first sip temperature). Furthermore, intraperson differences in first sip temperature of up to 10°C were observed between sittings. Intraperson correlations are shown in Table 3 and plotted in Supplementary Fig. S5. Between-metric correlations are also reported in Table 3 and plotted in Supplementary Fig. S6; they tended to be stronger for the Chinese sample because of the smaller within-person variation in sip size.

The relationships between perceived self-reported drinking temperatures and objective metrics are presented separately for IARC beverage drinkers and Chinese porridge consumers in Supplementary Table S2. Although temperatures were perceived to be higher among IARC beverage drinkers (Table 1) and the contrary was measured objectively, within each group, agreement was found between perceived and objective metrics. Relative to “warm,” participants reporting “hot” had 1.9°C and 5.1°C higher mean IELTs at IARC and in China, respectively; 2.8°C and 6.9°C higher for those reporting “very hot” (P for trend: 0.047 and 0.03 in Groups A and B, respectively).

Despite longstanding and growing evidence of the probable carcinogenicity of hot beverage consumption to the squamous cells of the esophagus, there exists no validated protocol for routine exposure assessment. There are no studies having comparatively assessed alternative metrics of thermal exposure or having verified whether a commonly used metric, temperature at first sip size, accurately measures thermal exposure in the esophagus. To our knowledge, our study is the first to objectively measure both sip temperatures and sip sizes throughout the duration of hot beverage/semiliquid consumption and use them to derive and comparatively assess different exposure metrics. In summary, although the commonly used temperature at first sip indeed discriminates two groups of hot beverage/porridge drinkers to some extent, use of mean temperature, and more so, sip-weighted mean temperature or mean IELT demonstrated much greater discriminatory ability. Hence, these novel metrics provided a more comprehensive thermal exposure assessment, and their application will reduce exposure misclassification in future studies. However, it should be noted that these two metrics do not account for sip size and temperature in similar manners. IELT is a metric that incorporates differentials in both average sip size and temperature between individuals. In contrast, the sip-weighted temperature is an average temperature alone (weighted to the larger sip sizes relative to all sips for that individual) and does not capture differences in average sip size between individuals. Thus, had the IARC and Chinese groups been consuming at similar temperatures, sip-weighted temperature would not have captured the large difference in thermal exposure experiences brought about by the larger sip sizes of the latter group.

We observed two contrasting thermal exposures in our study by comparing a group of Chinese hot porridge consumers, who took larger sips of hotter liquid sooner after serving, with a mixed group of international hot beverage drinkers, who waited longer to start drinking and took smaller sips which increased in size as temperature decreased, thus exhibiting a protective behavior which reduced their predicted esophageal thermal exposure.

In this study, we found that the previously employed metric of temperature at first sip is not sufficient in capturing relevant exposure contrasts, depicted by the large degree of overlap between both groups in Fig. 3A. In the knowledge that sip size is an important determinant of esophageal thermal exposure (9), by incorporating it, either by deriving predicted IELT, or calculating sip-weighted temperatures, it was possible to better characterize thermal exposures and make more meaningful comparisons between populations. A first-sip IELT discriminated the two groups almost as well as mean IELT or sip-weighted temperature; however, the comparatively weak within-person correlation (r = 0.37 at IARC) found for first sip IELT suggests that, to reflect habitual exposures, measurements at several time points throughout the drink/meal should be taken. This is a small logistical/time addition to the measurement protocol, while providing valuable information—as shown here—on the behavior of people to modify, or not, sip sizes in response to temperature. In addition, although mean temperature also discriminated well, the thermal exposure groups in this example were extremes apart, thus the improved sip temperature + size protocol may offer advantages when drinking temperature ranges are similar but sipping or gulping habits differ. Furthermore, a measurement at the beginning, middle, and just prior to the end of the drink is logistically feasible. Therefore, with no previous studies incorporating sip size into exposure assessment protocols, we have presented both a practical methodology for doing so and proposed exposure metrics to be derived from the data generated.

Our study had some limitations. Although we benefited from previous data (9) on the relationship between liquid temperature, volume, and IELT, thus enabling temperatures relevant to the target organ to be predicted, we were limited by the narrow range of sip temperatures (55–65°C) and volumes (5–20 mL) assessed by de Jong and colleagues. Therefore, we could only examine the fit of our model in this range, which only 49% and 24% of IARC and Chinese participants were within, respectively (Supplementary Table S3). Predicted IELTs for sips beyond this range of are subject to assumptions and should be interpreted cautiously. Further, we assumed that this model—based on liquid (coffee) drinking—equally applied to the liquid and porridge consumption in our study. The higher viscosity of porridge may however lead to different transfer and cooling rates through the esophagus, but with no porridge-specific IELT experiment, we assumed the same IELT prediction for both substances. For these reasons, researchers might prefer to use the sip-weighted mean temperature which simply reflects the temperature at which most volume is consumed; and our protocol affords this option. Further uncertainty of drinking temperature accuracy comes from the need to measure temperature in a proxy drinking vessel (Fig. 1B)—a necessary trade-off to allow inconspicuous measurements while not interfering with normal drinking behavior or contaminating the beverage. Temperature discrepancies between proxy and drinking vessels likely depend on several factors, including drinking duration, vessel material, ambient temperature, and volume change. It is advised to fill two vessels of the same material, at the same time with the same volume of liquid in the absence of a more direct approach. We acknowledge that the sample of individuals at IARC was not representative of a specific population; however, we emphasize that the two groups were primarily chosen to capture two samples with contrasting esophageal thermal exposures, for which both were adequate. Our study does not overcome the challenge of temperature measurement in case–control designs as, due to their dysphagia, ESCC cases cannot be measured with our protocol or any of the others.

Our approach has the potential to strengthen future descriptive cross-sectional studies and those seeking to identify biomarkers of exposure and effect, as well as prospective studies. We recently highlighted deficiencies of the self-reported/perceived approach to drinking temperatures (e.g., “very hot”/“hot”/“warm”), which is vulnerable to interviewer and recall bias (15). In the present study, self-reported temperatures were higher for IARC beverage drinkers, who had much lower objectively determined thermal exposures. However, prospective studies conducted in Iran (6) and China (5) have found significantly elevated ESCC risks for self-reported drinking temperatures. Indeed, in the present study, we report within-group agreement between perceived and measured exposures (Supplementary Table S2), particularly for mean IELT. This suggests that responses (i) are relative to the norm within a given population and (ii) reflect heat as opposed to temperature alone, and may be of value to within-population but not between-population comparisons.

Given the demonstrated findings of the importance of both sip size and sip temperature, it is unfortunate that the only large cohort to have measured drinking temperatures, in Golestan (6), did not also include sip sizes, ideally multiple sip sizes, but at least the size of the first sip. If an inverse sip size–temperature relationship was present in Golestan, misclassification of thermal exposures may have been large based on first sip temperature alone, perhaps contributing to the relatively low reported HR (∼1.4). Interestingly, the HR for self-reported “very hot” drinkers reported prospectively, which is not subject to recall bias, was 2.4, perhaps capturing a subgroup with truly high thermal exposure that was not captured in the measured first-sip temperature. The most profound implications of our study are how the findings of previous cross-sectional studies reporting first sip temperatures are interpreted; and how subsequent study designs are improved to yield meaningful descriptive data and interpopulation comparisons. Such information is essential to elucidating the carcinogenic mechanism of thermal injury and its lower threshold (i.e., maximum safe temperature at which any size sip can be taken), which is needed to design effective communication for cancer prevention strategies.

In conclusion, our study shows that temperature at first sip is suboptimal for assessing human exposure to hot beverages/semisolids, and future studies should include sip size measurements in exposure assessment protocols. In particular, comparisons across populations are questionable as drinking habits following first sip can be very distinct and result in different thermal exposure to the esophagus not captured in the information on the temperature at first sip alone. This finding endorses in general the responsibility of epidemiologists to question, verify, and continuously improve their applied exposure assessment methods, as the detection of especially moderate associations requires as little exposure misclassification as possible.

No potential conflicts of interest were disclosed.

Where authors are identified as personnel of the IARC/World Health Organization, the authors alone are responsible for the views expressed in this article, and they do not necessarily represent the decisions, policy, or views of the IARC/World Health Organization.

Conception and design: D.R.S. Middleton, W.-Q. Wei, V.A. McCormack

Development of methodology: D.R.S. Middleton, G. Byrnes, V.A. McCormack

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): D.R.S. Middleton, S.-H. Xie

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): D.R.S. Middleton, S.-H. Xie, L. Bouaoun, G. Byrnes, V.A. McCormack

Writing, review, and/or revision of the manuscript: D.R.S. Middleton, S.-H. Xie, L. Bouaoun, G. Byrnes, J. Schüz, W.-Q. Wei, V.A. McCormack

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): D.R.S. Middleton, G.-H. Song, W.-Q. Wei

Study supervision: G.-H. Song, W.-Q. Wei, V.A. McCormack

The authors thank all study participants, and the staff and fieldworkers at the Cancer Institute/Hospital of Ci County for their valuable contributions. Illustrations used in Fig. 2 were produced by Morena Sarzo, a Visual Designer from the IARC Communications Group. The work reported was undertaken during the tenure of a Postdoctoral Fellowship awarded to D.R.S. Middleton from the IARC, partially supported by the European Commission FP7 Marie Curie Actions—People—Co-funding of regional, national and international programmes (COFUND) and the World Cancer Research Fund International (grant no. 2018/1795). Work conducted in China was funded by a grants awarded to W.-Q. Wei by the National Key Research and Development Program (Precision Medicine Research; grant no. 2016YFC0901404) supported by the Ministry of Science and Technology of the People's Republic of China; the National Natural Science Fund (grant no. 81573224) supported by the National Natural Science Foundation of China; and the Innovation Fund for Medical Sciences of Chinese Academy of Medical Sciences (grant no. 2016-I2M-3-001).

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.