Over the past few decades, a number of risk factors for ESCC have been identified, including tobacco smoking, alcohol drinking, dietary and micronutrient deficiency, high temperature of beverage and food consumption, poverty, history of head and neck cancer and other miscellaneous factors (such as fast eating habits and polycyclic aromatic hydrocarbons (PAH) exposure) [4]. Tobacco smoking and alcohol drinking are major risk factors for ESCC in the US population, explaining over 90% of cases in men [5]. But several epidemiological studies have shown that smoking and alcohol drinking play a much less significant role in the etiology of esophageal cancer in the high risk areas of China [6], [7].

Several studies have reported the attributable causes of cancer in western populations [8], [9], [10]. For example, the burden of cancer mortality and incidence attributable to risk factors was estimated in France in 2000 [8], and it was found that the fraction of esophageal cancer deaths caused by tobacco smoking, alcohol drinking, overweight and obesity was 79.2% for men and 49.4% for women. Only one case control study, focusing on one province, has estimated the proportion of esophageal cancer attributable to lifestyle risk factors in China, and it found the combined population attributable fraction (PAF) to be 62.6% [11].

In our study, 17.9% of esophageal cancer deaths among men and 1.9% among women were attributable to tobacco smoking, respectively. These estimates were comparable to the few previous estimates from Chinese studies. A large-scale retrospective proportional mortality study by Liu and colleagues [27] estimated that tobacco smoking was responsible for 27.9% of esophageal cancer deaths in middle-aged men and 2.8% in middle-aged women. A more recent prospective cohort study, conducted by Gu and colleagues [28], showed that the fraction of esophageal cancer deaths caused by tobacco smoking was 19.4% in men and 1.6% in women. Our estimates are lower than those of Liu’s study [27], but similar to those of Gu’s study [28]. The reasons for the differences in these estimates are probably related to differences in the RRs and smoking prevalences used in calculating them (Table 3). The RR used in our study was obtained from a prospective study with 29,584 adults participating in general trial cohort in China, and was the same RR as that found in men in Gu’s study (RR=1.34). The RR in Liu’s study was higher in men (1.61), possibly because the smoking information was derived retrospectively from the surviving spouse or family members, which may have led to misclassification or recall biases. In our study, smoking prevalence was estimated by linear interpolation using the results of the 1984 and 1996 national smoking surveys, while in Gu’s study it was abstracted from a cross-sectional study conducted in 2000–2001 [29]. In Liu’s study, information of smoking habits was derived from the family members.

Alcohol drinking is another important risk factor for esophageal cancer. We estimated that the fraction of esophageal cancer in our population that was attributable to alcohol drinking was 15.2% for men and 1.3% for women. Danaei and his colleagues [30] calculated mortality from 12 types of cancer attributable to nine risk factors in seven World Bank regions in 2001, and showed that 24% of all esophageal cancer deaths (men and women combined) in low and middle-income countries and 41% in high income countries were attributable to alcohol drinking. Our estimate among men was lower than the estimate of this global analysis for both men and women in low and middle income countries, and was much lower than that in high income countries. For women, our PAF was much lower than the estimate from a previous population based case-control study in the US (64.4%) [31] and the estimate from a previous European population based prospective cohort study (25% for cancers of the upper aerodigestive tract) [32]. In China, however, an increase in the prevalence of alcohol drinking among women between 1991(2.6%) and 2002 (4.5%) was observed. Thus it is expected that the contribution of alcohol drinking to the cancer burden will increase in the future.

Most previous studies about the cancer burden in global or national studies have evaluated the combined effect of total fruit and vegetable intake [30], [31], [33]. Our report separately estimates the esophageal cancer burden attributable to low intake of fruit and low intake of vegetables. Fruit and vegetables contain different nutrients and have different nutritional values [34]. For example, the contents of vitamins, minerals, fiber, and phytochemicals are higher in most vegetables, especially in dark vegetables, while carbohydrates and organic acids are higher in fruits. Danaei and his colleagues [30] calculated that 19% of esophageal cancer deaths in low and middle-income countries were attributable to low vegetable and fruit intake. Our estimates indicated a little higher proportion of esophageal cancer deaths were attributable to low vegetable and fruit intake, reflecting the different sources of RRs and exposure rates in the two studies.

We compared our estimates of the combined effects of smoking, drinking, and low fruit and vegetable intake with similar estimates published from previous studies in different countries (Table 4). PAFs in our analysis were much lower than the corresponding figures in the USA [31], France [8], the Nordic countries [9] and the UK [10]. Tobacco smoking and alcohol drinking are the major risk factors for esophageal cancer in these western populations, accounting for around 80% of the population attributable risk. In China, on the other hand, smoking and drinking are less important risk factors for esophageal cancer.

This study had several limitations. First, we did not evaluate the effects of some other known risk factors, including overweight/obesity, tooth loss and poor oral health, and consumption of hot drinks. BMI has been inversely associated with esophageal squamous cell carcinoma in various countries, including China [35]. At least two prospective studies in China have also shown that tooth loss increases risk of developing upper gastrointestinal cancers [36], [37]. And in the Golestan Case-Control Study in Iran, drinking hot and very hot tea were associated with 2-fold and 8-fold elevated risks of esophageal cancer, respectively [38]. Therefore, there is a need for additional studies to evaluate the contribution of these and other risk factors to the burden of esophageal cancer in China. Second, there were some sources of uncertainty in the relative risks we used in making our PAF estimates. The RRs of alcohol drinking and of low vegetable and fruit intake were obtained from meta-analyses, and some of the original studies included in these meta-analyses may not have been adjusted for confounding factors. Moreover, the RRs of low vegetable and fruit intake were derived from non-Chinese Asian studies, and there was uncertainty in the extrapolation of these RR to the Chinese population. It was difficult to evaluate the direction and potential magnitude of these potential biases. Also, RRs in men were used for both genders because of a lack of available data among women or the statistical instability of risk estimates in women, but RRs for men and women might be different. Third, we were not able to collect information on the amount and duration of smoking or alcohol drinking in the Chinese population at large, and epidemiological studies have demonstrated that these variables are important to fully assess the impact of these exposures on esophageal cancer [39], [40]. Fourth, we did not adjust our estimates of PAF for the possible interaction because too little data are available in studies from China to provide valid estimates of the ORs of interaction between different risk factors. Finally, another potential limitation was the lack of measured esophageal cancer incidence data in China, which meant that we had to estimate these numbers from mortality data and a mortality-to-incidence ratio. This M/I ratio was calculated by Poisson regression models, adjusted for age, gender and cancer registry site, but it was not adjusted for other relevant factors such as socioeconomic status because of a lack of available data.