Individual and combined effects of major environmental risk factors of esophageal cancer: the Golestan Cohort Study

Abstract

Background & Aims:

Northeast-Iran has one of the highest reported rates of esophageal squamous cell carcinoma (ESCC) worldwide. Decades of international investigations in this region indicate key roles for some local habits and environmental exposures. The aim of this study is to investigate the individual and combined effects of the major environmental risk factors of ESCC.

Methods:

this study is a population-based cohort of 50,045 individuals aged 40–75 years old from both urban and rural areas across Northeast-Iran. Detailed data on demographics, diet, lifestyle, socioeconomic status, temperature of drinking beverages, and different exposures were collected upon enrolment using validated methods, questionnaires, and physical examinations.

Results:

during an average 10 years of follow-up 317 participants developed ESCC. Opium smoking (HR:1.85, 95%CI:1.18–2.90), drinking hot tea (HR:1.60, 95%CI:1.15–2.22), low intake of fruits (HR:1.48, 95%CI:1.07–2.05) and vegetables (HR:1.62, 95%CI:1.03–2.56), excessive tooth loss (HR:1.66, 95%CI:1.04–2.64), drinking un-piped water (HR:2.04, 95%CI:1.09–3.81), and exposure to indoor air pollution (HR:1.57, 95%CI:1.08–2.29) were significantly associated with increased ESCC risk in a dose-response manner. Combined exposure to these risk factors was associated with a stepwise increase in the risk of developing ESCC, reaching a more than seven-fold increased risk in the highest category.

Conclusions:

Our results indicate a multi-factorial causal nature of ESCC in this population and support the hypothesis that the high ESCC rates are due to a combination of factors including thermal injury (from hot tea), PAH exposure (from opium and indoor air pollution), and a nutrient deficient diet, with additional effects linked to exposure to un-piped water and tooth loss.

Graphical Abstract

Introduction

Golestan province in northeastern Iran has long been known to have one of the highest reported rates of esophageal squamous cell carcinoma (ESCC).^1,2 The striking ESCC rates in this region led scientists from the International Agency for Research on Cancer (IARC) and the Institute of Public Health Research of Tehran University to begin a series of joint epidemiologic investigations in 1969 that indicated smoking and alcohol consumption did not explain these high rates, and rather they must be explained by other local habits.^1,3 Initial hypotheses included use of opium, hot tea consumption and low intake of fruits and vegetables.^3,4 Due to the sociopolitical changes of 1979 in Iran, these collaborative investigations stopped before reaching definitive conclusions.

In 2001 scientists from the Tehran University of Medical Sciences, IARC and the U.S National Cancer Institute initiated a new series of joint epidemiologic studies in Golestan to re-explore the underlying etiologic factors of the unusual ESCC rates in the region.^5,6 The Golestan Cohort Study (GCS), with more than fifty thousand individuals, was launched as the first large-scale, population-based prospective study in central and western Asia with detailed exposure assessments and biological samples to provide a major resource for studying ESCC.^2,5–7 The GCS provides a unique opportunity to investigate ESCC risk factors due to its large sample size, successful follow up with <1% loss after ten years, minimum amount of missing data, and limited confounding effects from alcohol consumption and smoking (especially in women) because of their low consumption rates.

Based on 505,865 person-years of follow up, we now provide the first report from the GCS of the individual and combined effects of the main ESCC risk factors in northeastern Iran. In this report, we evaluate the main suspected environmental exposures, including those found to be important in the 1970s studies in Golestan,^1,3,4 those found to be significantly associated with ESCC in our earlier Golestan case-control study,^8–11 and those reported to be important in other parts of the world.^12–15

Materials and Methods

Study population and design

The design of the GCS has been previously described.⁷ Briefly, the GCS is a prospective population-based cohort study of individuals aged 40–75 years old from Golestan province in northeastern Iran. During a four-year period starting from 2004, the GCS recruited 50,045 individuals from both rural and urban areas of Gonbad, Maraveh-tappeh, Kalaleh and Aq-qala districts (Figure1). The participants from urban areas were selected randomly by systemic clustering using the household numbers and were then contacted by trained healthcare workers and were invited to visit the GCS center and participate in the study. In the rural areas all eligible people residing in the villages of the study area (326 villages) were contacted and invited to participate in the study. This process was done using the network of health houses, smallest part of the primary healthcare system, that are present in each group of villages and usually staffed by two healthcare workers who also reside in those villages. Participation rates were 70% for women and 50% for men in the urban area and 84% for women and 70% for men in the rural area⁷.

Figure1.

Golestan province in northeastern Iran and the age standardized incidence rates (ASR) of esophageal cancer in different regions of the province. The presented ASR are calculated per 100,000 persons-years from both the Golestan Cancer Registry database (Roshandel et al. Arch Iran Med. 2012;15(4):196–200) and also the Golestan Cohort Study (GCS) database (unpublished). The circles illustrate the sampling districts in the GCS as 1. Maraveh-tappeh 2. Kalaleh 3. Gonbad and 4. Aq-qala

Those who had been diagnosed with upper gastrointestinal (GI) cancers, those who were unwilling to participate, and temporary residents were excluded. After explaining the study, a written informed consent was obtained from all participants. The GCS was approved by the institutional review boards of the Digestive Disease Research Institute of the Tehran University of Medical Sciences (Ref: FWA00001331), the International Agency for Research on Cancer (Ref: CN/23/3), and the US National Cancer Institute.

Exposure assessment

Participants were interviewed by trained general physicians and nutritionists. Two structured questionnaires that were validated in the pilot phase of the study,^5,7,16 were completed for each participant, including a detailed general questionnaire (collecting data on demographics, lifestyle, socioeconomic status, and various exposures) and a Food Frequency Questionnaire (FFQ). The FFQ was developed by a team of Iranian nutritionists and contains 116 food items, including 25 vegetable items and 15 fruit items, along with portion size photos, as well as inquiries about the frequency and amount of each item consumed.¹⁶ The validity and reliability of the FFQ has been confirmed previously through repeated administration of the FFQs and 24-hour dietary recalls over one year and comparing the questionnaire data with serum and urinary excretion levels of selected nutrients.¹⁶

In Golestan, tea is the main hot beverage consumed and is drunk in large quantities. We used a validated and reliable method to evaluate tea drinking temperature in the GCS.^5,7,17 At each session, the interviewer prepared two fresh cups of tea; one was given to the participant, and the other was hidden behind a screen with a digital thermometer inside. When the tea temperature was 75°C, participants were asked to sip the tea, and if it was their usual tea drinking temperature, it was recorded. Otherwise, tea was left to cool, and the procedure was repeated at 5°C intervals. The amount, frequency, and duration of consumption were also recorded for each participant.

Participants were asked about any regular consumption of opium, cigarettes, nass (a chewing tobacco product containing tobacco, ash and lime), and alcohol, including the duration, frequency and consumption amount of each agent. The reliability and validity of self-reported opium and tobacco consumption data were confirmed previously through re-interviewing a subgroup of participants and comparing the questionnaire responses with the presence of codeine, morphine, and cotinine in their urine samples.^5,18

Participants were asked about their oral health and hygiene practices, and their decayed, filled and missing teeth (DMFT) were counted by dentist-trained interviewers. Primary drinking water sources were evaluated, including current access to in-home piped water and current and past duration of drinking un-piped water from cisterns, wells or natural sources. The participants were also asked about their lifelong history, duration, and level of exposure to animals including ruminants, equines, dogs and poultry. Fuel sources for household heating and cooking, the duration of their use, and installation of chimneys were also assessed by relevant questions.

Defining level of exposures

Cumulative use of opium was calculated in nokhod-years (nokhod is a local unit for opium that equals 0.2 grams), cigarette smoking was assessed as pack-years (a pack includes 20 cigarettes), and nass chewing as nass-years by calculating the number of units used per day multiplied by the number of consumption years. In Golestan, opium is consumed mainly through smoking and ingestion. Because of the differences in the contents, level of exposure and carcinogenic byproducts of consuming opium through smoking and ingestion,^19,20 we analyzed these two routes separately. For evaluating opium smoking, we categorized the participants as either never smoked opium, and for smokers, the tertiles of the cumulative nokhod-years of smoked opium, while for assessing opium ingestion we categorized the participants as either never ingested opium or the two quantiles of the cumulative nokhod-years of ingested opium. Only 5% of opium users consumed opium through both routes. For these participants, we calculated the cumulative amount of ingested and smoked opium separately and included them in the corresponding categories of opium ingestion and opium smoking. For evaluating cigarette smoking we categorized the participants as never smokers, and the tertiles of the cumulative pack-years of smoked cigarettes, while for nass chewing we categorized the participants as never chewed nass, and the two quantiles of the cumulative nass-years of chewed nass. Because of the few numbers of participants who consumed alcohol, for alcohol consumption we only categorized the participants based on ever or never regular consumption of alcohol.

We defined consumption of hot tea as drinking tea at or above 60 degrees Celsius. To assess the effect of drinking hot tea we initially evaluated the dose-response association of drinking tea with ESCC in those who consumed tea at <60 °C and those who consumed tea at ≥60 °C. We did not observe any effect for drinking tea at <60 °C (Supplementary File1), therefore we categorized the participants based on the temperature and the amount of tea they consumed per day as “drinking only warm tea”, “drinking ≤3 cups of hot tea”, “drinking 4–6 cups of hot tea” and “drinking more than 6 cups of hot tea”. One cup of tea was defined as 250cc of tea. Intake of fruits and vegetables were translated into grams and were separately quantified based on the number of servings of each group as: less than one serving, one-two servings and more than two servings/day. For both fruits and vegetables one standard serving size was considered as 80 grams.²¹

Due to cultural preferences, many people in Golestan often choose to have their teeth removed if they have any dental problems, including caries. As such, on the oral examination at enrolment only a few participants had filled teeth, while most had many of their teeth removed previously. Thus, using tooth loss is more appropriate than using DMFT index for evaluating oral health in this population. Furthermore, because of the substantial confounding effect of age on oral health variables we calculated the predicted number of tooth loss by age using a LOESS model and then calculated the difference between the predicted values and the actual number of tooth loss for each participant and categorized the difference into quartiles (Supplementary File 2).²²

In Golestan, natural gas, kerosene, diesel, and biomass fuels are used for home heating and/or cooking purposes. We defined exposure to indoor air pollution as the household use of non-gas fuels without chimneys. To quantify drinking un-piped water and daily contact with ruminants we categorized the participants as “never exposed”, “exposure ended ≥ 20 years ago”, “exposure ended 1–19 years ago”, and “continued exposure”. Exposure to indoor air pollution included an additional exposure group of ‘ended 1–5 years ago’.

Socioeconomic status was grouped based on the quartiles of a composite wealth score that was created using multiple correspondence analysis on the following variables: ownership of car, motorbike, television, refrigerator, freezer, vacuum and washing machine, as well as house ownership, house structure and size, and having a bath in the residence. The method for creating this wealth score has been previously described.²³

Follow-up and outcome ascertainment

All participants were followed since recruitment by active annual telephone surveys, home visits (when telephone contact was not successful) and monthly review of provincial cancer and death registration data. In the case of death, a validated verbal autopsy questionnaire was administered to relatives of the deceased to collect information that could be used to determine the time, place and cause of death.²⁴ For both incident cancers and deaths, the team gathered copies of all available and relevant medical reports. To verify the diagnosis of cancer or cause of death, all gathered information was separately reviewed by two expert physicians, who were requested to identify the code for the outcome based on the 10th revision of the International Statistical Classification of Diseases and Related Health Problems (ICD-10).²⁵ In case of disagreement, a third expert physician was asked to independently review the information and make the final diagnosis. In our final analysis, we included histologically proven ESCC cases (90%) and those esophageal cancer cases that were identified through available medical reports other than histology and verbal autopsy (10%).

Statistical analysis

Cox proportional hazards regression models were used to estimate hazard ratios (HRs) and corresponding 95% confidence intervals (CIs) for the association between different exposures and ESCC. The entry time was defined as the date at which the participant was recruited to the GCS, and the exit time was the end of follow-up time, defined as the date of first diagnosis of esophageal cancer for cancer cases, the date of death for deaths from any other causes, and the date of last follow-up for other participants, through December 31, 2017.

We selected potential ESCC risk factors based on the existing literature on suspected environmental exposures in the 1970s Golestan investigations (opium consumption, drinking hot tea, low intake of fruits and vegetables),^1,3,4 our earlier Golestan case-control study (poor oral health, tobacco consumption, contact with ruminants, drinking un-piped water),^8–11 and studies from other high and low incidence areas worldwide (alcohol drinking, and exposure to indoor air pollution).^12–14 We then separately adjusted the selected exposures for baseline characteristics including age, sex, ethnicity, residence districts, and quartiles of socioeconomic status, as shown in Table1. In order to identify the main independent ESCC risk factors we built a fully adjusted model to include baseline characteristics and levels of all selected environmental exposures. Dose-response associations were evaluated by comparing ESCC risk across different levels of each exposure.

Table1.

Baseline characteristics of the entire cohort participants and participants who developed Esophageal Squamous Cell Carcinoma (ESCC) in the Golestan Cohort Study.

Baseline characteristics	ESCC Cases (n = 317) N (%)	Cohort Participants (n = 50,038) N (%)	P value
Age (years) †			< 0.001
	59.54 (± 9.06)	52 (± 8.88)
Gender			< 0.001
•Male	174 (54.89)	21,230 (42.43)
•Female	143 (45.11)	28,808 (57.57)
Residence districts			< 0.001
•Maraveh-tappeh	47 (14.73)	3,448 (6.89)
•Kalaleh	85 (26.81)	10,732 (21.45)
•Gonbad	173 (54.57)	30,314 (60.58)
•Aq-qala	12 (3.79)	5,457 (10.91)
Ethnicity			< 0.001
•Turkmen	278 (87.70)	37,247 (74.44)
•Non-Turkmen	39 (12.30)	12,791 (25.56)
Socioeconomic Status ‡			< 0.001
•First quartile (lowest)	140 (44.16)	13,933 (27.84)
•Second quartile	81 (25.55)	11,144 (22.27)
•Third quartile	60 (18.93)	12,584 (25.15)
•Fourth quartile (highest)	36 (11.36)	12,377 (24.74)

^†:Age displayed as Mean (± Standard Deviation)

^‡:We categorized socioeconomic status based on a composite wealth score that was created using multiple correspondence analysis on the ownership of house, vehicle, and some home appliances (Islami F, et al. Int J Epidemiol. 2009;38(4):978–88).

To assess the combined effects of the individual risk factors, we created a combined exposure score (CE-score) based on the number and level of risk factors to which each participant was exposed. For this purpose we used a points system method that was originally developed by Sullivan et al. to combine the results of complex multivariable models for more practical use.²⁶ Briefly we created the CE-score through the following five steps: 1) running the fully adjusted Cox regression model that includes baseline characteristics and levels of all suspected ESCC risk factors; 2) defining the regression coefficient of age as the constant for this method to represent the number of regression units that reflect one point in the final points system or CE-score; 3) calculating the individual risk point for each level of the included risk factors by dividing its respective regression coefficient by the selected constant; 4) rounding the risk points to the nearest integer; 5) calculating the CE-score by summing the individual risk points for each one of the seven identified independent risk factors for each GCS participant. We then categorized the CE-score into quintiles and compared the risk of developing ESCC across them. We also calculated the population attributable fraction (PAF) of the combined exposure to the seven identified risk factors using the adjusted Cox model with Greenland and Drescher methods for estimating PAFs.^27,28 993 participants (1.98%) had missing information on one or more of the seven risk factors included in the CE-score and therefore were excluded from the combined effects analysis. The principles of this method are described with hypothetical examples in the Supplementary File3.

To assess reverse causality, we performed a sensitivity analysis by dropping the first two years of follow-up. We also repeated the analysis restricted to those with histologically confirmed ESCC. Analyses were also conducted separately for men and women. All statistical analyses were performed using Stata statistical software version 14 (Stata Corporation, College Station, Texas, USA).

Results

Baseline characteristics

Fifty-thousand and forty-five participants were enrolled in this prospective cohort; of these, seven individuals (0.01%) had been diagnosed with upper GI cancers upon enrolment and therefore we excluded them, leaving 50,038 participants. The mean ± standard deviation of the follow-up years was 10 ± 2 years (median: 10, range: 0 – 14 years). Three-hundred and seventeen participants developed primary ESCC; of these 285 cases (90%) have histologic confirmation, while 32 cases (9.7%) were identified through verbal autopsy and available medical records other than a histology report. Baseline characteristics of the cohort and the subgroup of ESCC cases are shown in Table 1. Participants who developed ESCC tended to be older, have male gender, live in eastern Golestan (Figure1), belong to the Turkmen ethnicity, and have lower socioeconomic status (Table1).

Main ESCC risk factors

We analyzed the association between different environmental exposures and risk of developing ESCC in the unadjusted, minimally adjusted (by the baseline characteristics in Table1), and fully adjusted models. Opium smoking, drinking hot tea, low intake of fruits, low intake of vegetables, excessive tooth loss, drinking un-piped water, and exposure to indoor air pollution were all significantly associated with increased ESCC risk in the unadjusted (p<0.001), minimally adjusted (p<0.01), and fully adjusted (p<0.05) models (Table2, Figure2). Except for daily intake of vegetables and drinking un-piped water, which showed borderline dose-response associations with increased ESCC risk in the fully adjusted model (P_trend =0.10, P_trend =0.05, respectively), these associations were significant for all identified risk factors in all three models (P_trend<0.05) (Table2, Figure2). Smoking cigarettes, ingestion of opium, consumption of alcohol, chewing nass, and contact with ruminants were not associated with increased ESCC risk after adjustment for baseline characteristics or in the fully adjusted model (Table2).

Table2.

Association between different levels of exposure to potential environmental risk factors and esophageal squamous cell carcinoma (ESCC) in the Golestan Cohort Study

Environmental exposures	ESCC cases N (%)	Entire Cohort N (%)	Unadjusted HR (95% CI)	Adjusted † HR (95% CI)	Fully Adjusted ┤ HR (95% CI)
Opium consumption through smoking (nokhod-years)‡
Never	256 (81.27)	43,717 (87.48)	1	1	1
Lowest tertile (<10)	16 (5.08)	2,355 (4.71)	1.20 (0.72 – 2.00)	1.14 (0.69 – 1.91)	1.24 (0.74 – 2.09)
Middle tertile (10 – 42)	15 (4.76)	1,933 (3.87)	1.39 (0.82 – 2.34)	1.30 (0.77 – 2.21)	1.31 (0.75 – 2.29)
Highest tertile (>42)	28 (8.89)	1,967 (3.94)	2.63 (1.78 – 3.89)	1.96 (1.31 – 2.93)	1.85 (1.18 – 2.90)
P for trend			< 0.001	0.001	0.009
Opium consumption through ingestion (nokhod-years)
Never	285 (90.19)	47,382 (94.72)	1	1	1
Lower than median (≤30)	15 (4.75)	1,380 (2.76)	2.08 (1.24 – 3.50)	1.12 (0.66 – 1.89)	0.98 (0.55 – 1.74)
Higher than median (>30)	16 (5.06)	1,260 (2.52)	2.46 (1.48 – 4.07)	1.13 (0.67 – 1.89)	1.09 (0.62 – 1.91)
P for trend			< 0.001	0.551	0.836
Drinking hot tea at ≥ 60 ° C
Only warm tea (<60 ° C)	93 (29.34)	19,775 (39.52)	1	1	1
≤ 3 cups/day	37 (11.67)	7,178 (14.35)	1.11 (0.76 – 1.63)	1.02 (0.69 – 1.49)	1.00 (0.66 – 1.52)
4 – 6 cups/day	113 (36.65)	14,607 (29.19)	1.68 (1.28 – 2.21)	1.55 (1.17 – 2.04)	1.50 (1.12 – 2.01)
> 6 cups/day	74 (23.34)	8,478 (16.94)	1.91 (1.40 – 2.59)	1.61 (1.18 – 2.19)	1.60 (1.15 – 2.22)
P for trend			< 0.001	< 0.001	0.001
Daily intake of fruits *
≥ 3 servings	82 (26.62)	17,077 (34.75)	1	1	1
1 – 2 servings	116 (37.66)	17,655 (35.93)	1.43 (1.08 – 1.90)	1.49 (1.11 – 1.99)	1.28 (0.94 – 1.75)
< 1 serving	110 (35.71)	14,409 (29.32)	1.68 (1.26 – 2.24)	1.51 (1.12 – 2.05)	1.48 (1.07 – 2.05)
P for trend			< 0.001	0.008	0.018
Daily intake of vegetables *
≥ 3 servings	27 (8.77)	10,182 (20.71)	1	1	1
1 – 2 servings	140 (45.45)	21,973 (44.70)	2.46 (1.62 – 3.71)	1.91 (1.26 – 2.91)	1.58 (1.02 – 2.46)
< 1 serving	141 (45.78)	17,000 (34.58)	3.23 (2.14 – 4.88)	2.12 (1.39 – 3.24)	1.62 (1.03 – 2.56)
P for trend			< 0.001	< 0.001	0.102
Excessive tooth loss ¶
≤ predicted tooth loss	140 (44.16)	26,279 (52.52)	1	1	1
1 – 4 excess tooth loss	42 (13.25)	6,546 (13.08)	1.22 (0.84 – 1.72)	0.96 (0.68 – 1.36)	0.96 (0.67 – 1.39)
5 – 8 excess tooth loss	45 (14.20)	5,926 (11.84)	1.47 (1.05 – 2.06)	0.84 (0.59 – 1.19)	0.83 (0.58 – 1.21)
9 – 11 excess tooth loss	64 (20.19)	6,126 (12.24)	2.02 (1.50 – 2.71)	1.42 (1.05 – 1.93)	1.50 (1.08 – 2.07)
≥ 12 excess tooth loss	26 (8.20)	5,161 (10.31)	0.93 (0.61 – 1.41)	1.68 (1.08 – 2.61)	1.66 (1.04 – 2.64)
P for trend			0.010	0.015	0.013
Drinking un-piped water (untreated water from wells, cisterns and natural sources)
≥ 31 years ago	16 (5.05)	8,146 (16.28)	1	1	1
20 – 30 years ago	97 (30.60)	15,966 (31.91)	3.27 (1.92 – 5.56)	2.51 (1.46 – 4.33)	1.69 (0.93 – 3.06)
1 – 19 years ago	130 (41.01)	17,443 (34.86)	4.07 (2.42 – 6.84)	2.98 (1.73 – 5.13)	1.78 (0.97 – 3.24)
Continued	74 (23.34)	8,483 (16.95)	4.68 (2.72 – 8.04)	3.32 (1.89 – 5.85)	2.04 (1.09 – 3.81)
P for trend			< 0.001	<0.001	0.056
Exposure to indoor air pollution (using indoor non-gas fuel without chimneys)
Never	50 (15.77)	12,111 (24.21)	1	1	1
20 ≥ years ago	36 (11.36)	8,150 (16.29)	1.09 (0.71 – 1.68)	1.12 (0.70 – 1.74)	1.30 (0.82 – 2.08)
6 – 19 years ago	38 (11.99)	8,108 (16.20)	1.13 (0.74 – 1.72)	1.20 (0.78 – 1.86)	1.37 (0.86 – 2.20)
1 – 5 years ago	90 (28.39)	12,422 (24.83)	1.75 (1.24 – 2.48)	1.46 (1.03 – 2.06)	1.48 (1.02 – 2.13)
Continued	103 (32.49)	9,235 (18.46)	2.64 (1.88 – 3.70)	1.78 (1.26 – 2.51)	1.57 (1.08 – 2.29)
P for trend			< 0.001	< 0.001	0.012
Daily contact with ruminants
Never	10 (3.32)	4,061 (8.39)	1	1	1
≥ 20 years ago	40 (13.29)	9,666 (19.96)	1.72 (0.86 – 3.45)	1.03 (0.51 – 2.08)	0.79 (0.39 – 1.62)
1 – 19 years ago	61 (20.27)	10,272 (21.21)	2.56 (1.31 – 5.00)	1.37 (0.69 – 2.71)	0.90 (0.44 – 1.82)
Continued	190 (63.12)	24,424 (50.44)	3.26 (1.72 – 6.16)	1.60 (0.82 – 3.08)	1.01 (0.51 – 2.00)
P for trend			< 0.001	0.009	0.161
Alcohol drinking
Never	312 (98.42)	48,329 (96.58)	1	1	1
Ever	5 (1.58)	1,709 (3.42)	0.45 (0.18 – 1.10)	0.47 (0.19 – 1.15)	0.29 (0.09 – 0.94)
Cigarette smoking (cumulative pack-year)
Never	246 (77.60)	41,382 (82.70)	1	1	1
Lowest tertile (<5.6)	18 (5.68)	2,893 (5.78)	1.07 (0.66 – 1.73)	0.92 (0.56 – 1.51)	0.83 (0.49 – 1.43)
Middle tertile (5.7 – 20)	23 (7.26)	2,993 (5.98)	1.34 (0.87 – 2.05)	1.22 (0.78 – 1.92)	1.27 (0.78 – 2.05)
Highest tertile (>20)	30 (9.46)	2,770 (5.54)	1.97 (1.35 – 2.88)	1.35 (0.90 – 2.02)	1.28 (0.81 – 2.02)
P for trend			< 0.001	0.12	0.263
Nass chewing (cumulative nass/year)
Never	272 (85.80)	46,198 (92.33)	1	1	1
Lower than median (≤60)	18 (5.68)	2,016 (4.03)	1.63 (1.01 – 2.64)	1.05 (0.64 – 1.72)	0.86 (0.50 – 1.47)
Higher than median (>60)	27 (8.52)	1,824 (3.65)	2.81 (1.89 – 4.17)	1.12 (0.73 – 1.71)	0.96 (0.60 – 1.52)
P for trend			< 0.001	0.56	0.868

HR: Hazards Ratio, CI: Confidence Interval

^†:Adjusted for baseline characteristics including: age, gender, residence districts, ethnicity, and quartiles of the socioeconomic status.

^┤:the fully adjusted model includes age, gender, residence counties, ethnicity, quartiles of the socioeconomic status, and levels of exposure to all potential ESCC risk factors and confounders as illustrated in the table.

^‡:nokhod is a local unit that equals 0.2 grams of opium

^*:each serving is defined as intake of 80 grams of the corresponding dietary group

^¶:this number was calculated by subtracting the age-dependent predicted number of tooth loss from the actual number of tooth loss.

Data are missing on the amount and/or route of opium consumption for 82 (0.16%) cohort participants, including 3 (0.95%) ESCC cases; on fruits and vegetables consumption for 883 (1.76%) cohort participants, including 9 (2.84%) ESCC cases; on heating/cooking fuel sources for 12 (0.02%) cohort participants; and on history of contact with ruminants for 1,615 (3.22%) cohort participants, including 16 (5.04%) ESCC cases.

Figure2.

Main risk factors associated with increased risk of developing esophageal squamous cell carcinoma (ESCC), in the Golestan Cohort Study. This figure was adopted from the fully adjusted model that simultaneously included all baseline characteristics, confounders and risk factors for developing ESCC. The risk points, determined for the significantly associated modifiable environmental factors, are calculated by dividing the regression coefficient of each risk factor by the selected constant (regression coefficient of age), and then rounded to the closest integer. One risk point equals the risk of developing ESCC with each year increase in age in this population. For evaluating the ESCC risk associated with combined exposure to the illustrated risk factors, a combined exposure score was created by summing the individual risk points for each participant.

Despite all adjustments age, male gender, living in eastern Golestan, Turkmen ethnicity, and low socioeconomic status remained significantly associated with ESCC in all models (Figure2).

Combined effects of the main ESCC risk factors

The calculated risk point integers for each level of the main identified ESCC risk factors are illustrated in Figure2. Based on the formula; one risk point equals the risk of developing ESCC with each year increase in age in this population (Figure2, Supplementary File3). The CE-score ranged from 0–39. After adjustment for baseline characteristics, there was an average of 1.5-fold increase in the HRs of ESCC following each increment in the quintiles of the CE-score (P <0.001). Compared to individuals in the lowest quintile, those in the second quintile had a HR=2.66 (95%CI, 1.37 – 5.13, p=0.004), in the third quintile had a HR=4.23 (95%CI, 2.28 – 7.86, p<0.001), in the fourth quintile had a HR=5.15 (95%CI, 2.76 – 9.59, p<0.001) and in the highest quintile had a HR=7.12 (95%CI, 3.86 – 13.12, p<0.001) for developing ESCC (Figure 3). The PAF of ESCC due to the seven environmental risk factors was 76% (95% CI: 60% – 85%).

Figure3.

The hazards ratios of developing ESCC among the quintiles of the combined exposures score. The hazards ratios were adjusted for age, gender, residence counties, ethnicity, and quartiles of the socioeconomic status. Very Similar trends were observed when the analysis was repeated after excluding those cases without histologic confirmation or after dropping the first two years of follow-up.

Sensitivity and sex-specific analyses

When we repeated the analyses after excluding ESCC cases without histology reports, or after dropping the first two years of follow-up, the obtained results were comparable to the results without these exclusions (Figure3, Supplementary Files 4 and 5). When we repeated the analyses separately for men and women, comparable results were observed for most exposures, except for the exposure to indoor air pollution which revealed stronger effects in women (Supplementary File 6). This is potentially due to their longer and more intense level of exposure to indoor air pollution. Low intake of fruits also revealed strong effects in men but not in women (Supplementary File 6).

Discussion

After more than ten years of successfully following-up fifty thousand participants of the GCS, the results show that smoking opium, drinking hot tea, low intake of fruits, low intake of vegetables, tooth loss, drinking un-piped water, and exposure to indoor air pollution are all associated with increased ESCC risk in a dose-response manner. Furthermore, a higher combined exposure score based on the number and level of exposure to these seven risk factors is associated with a stepwise increase in ESCC risk, with the participants in the highest quintile having a more than seven-fold increased ESCC risk. While each of these factors were associated with ESCC risk after simultaneous adjustment, whether some or all of them are causal risk factors for the high rates of ESCC in this population is less clear. These results do however point to a combination of these risk factors driving the high risk of ESCC in this population, with 76% of ESCC cancer incidence being explained by the identified exposures, assuming that all are causal.

In our study a large proportion of ESCC cases were women. Although the ESCC incidence rate has declined over recent decades, it is still the second most common cancer in both men and women residing in this area with a male: female incidence ratio that approaches 1.^29,30 This sex ratio is similar to those from other high incidence areas including Africa,³¹ and China,¹⁴ , contrary to the lower incidence areas where ESCC is believed to follow an obvious male predominance.³² This sex pattern suggests the contribution of strong environmental risk factors that are shared by both men and women in the development of the disease³⁰. The results of the sex-specific subgroup analysis that revealed similar trends and comparable individual and combined effects for most identified risk factors among the male and female participants further support this hypothesis.

Our results show an 85% increased ESCC risk in the highest level of opium smoking that is independent of other risk factors and confounders, including tobacco smoking and chewing. Despite the fact that an estimated 18 million people illicitly consume opiates worldwide, and one third of the produced opium is consumed raw,³³ there has been little epidemiologic research on the relationship between opium and cancers.³⁴ A recent systematic review of the few available studies suggested a possibly causal relationship between opium consumption and cancer in different sites including the esophagus.³⁴ Opium might be linked to ESCC through a casual mechanism, due to the highly mutagenic and potentially carcinogenic compounds found in opium pyrolysates and opium smoke,^19,35 and through a facilitating mechanism, due to some of its alkaloid contents that induce relative stasis in the esophagus.³⁵

One unexpected and unexplained finding in this study was the lack of association between ingestion of opium and ESCC risk. Compared to opium smoking, opium ingestion exposes the users to less polycyclic and heterocyclic aromatic hydrocarbons and primary aromatic amines, but to almost ten times higher morphine.^19,20 In our previous case-control study of ESCC in Golestan, both ingesting and smoking opium were associated with increased risk of ESCC,⁹ and previous analyses in the GCS have shown consistently harmful associations between both smoking and ingesting opium and risk of pancreatic cancer,³⁶ digestive disease mortality,³⁷ and respiratory disease mortality.³⁸ However, the null result may be related to the small number of cases who ingested opium (total of 31 cases), and more prospective analyses of opium ingestion and ESCC risk are needed before a conclusion about this exposure can be made.

The accumulating evidence since the early 1970s on the association of consuming hot beverages (mainly tea, and maté in South America) and ESCC resulted in the classification of hot beverages as probably carcinogenic to humans by the IARC.³⁹ The main limitation in the current literature is that nearly all studies rely on self-reported perception of drinking temperature of beverages, which may vary across individuals and populations.^40,41 Further, most studies of this association have been case-control studies that are prone to recall bias.^39–41 To our knowledge the GCS is the only large-scale prospective cohort study in which actual tea drinking temperature has been measured by a validated method.^5,7,17 Although the exact underlying mechanism is still unknown, our findings regarding the significant dose-response association of drinking tea with increased ESCC risk at ≥ 60 °C and not at < 60°C, along with the null effect of drinking hot tea in low quantities (<3cups) on ESCC risk favor the hypothesis of repetitive thermal injury in developing esophageal cancer as cell injury may occur due to repetitive exposure to high temperature, and the sequelae of this injury may lead to cancer and/or increased susceptibility to other carcinogens.^39–41

Low intake of fruits and vegetables and tooth loss were also independently associated with increased ESCC risk. The protective effect of fruits and vegetables on ESCC risk has been shown in many studies conducted in populations with both high and low ESCC rates, probably due to their content of high levels of vitamins and phytochemicals with antioxidant and anti-tumor effects.^42,43 Different indicators of poor oral health have been also linked to ESCC and its precursor lesion, esophageal squamous dysplasia, in studies from Golestan and other regions.^8,44,45 Persistent low-grade systemic inflammation that is associated with periodontal disease and the accumulation of altered oral microbiota that produce potentially carcinogenic metabolites such as acetaldehyde and nitrite (a precursor of nitrosamines) are among the proposed mechanisms for the association of poor oral health and increased ESCC risk.^8,44–47

Drinking un-piped water has been linked to increased risk of upper GI cancers in high incidence regions including China,^14,15 and Iran.^11,48 Although we did not have data on the quality of drinking water in the current study, another study in Golestan showed higher concentrations of nitrates and some minerals in the un-piped drinking water in areas with higher ESCC incidence and mortality rates.⁴⁹ One of the possible mechanisms might be through the increased endogenous production of nitrosamines and nitric oxide following ingestion of additional nitrates in water.^50,51

Regarding the association of indoor air pollution and increased ESCC risk, these findings are consistent with the case control studies conducted in high incidence regions of Asia,⁵² Africa,¹² South America,¹² and Europe.¹³ Diesel, kerosene, and biomass are the non-gas fuels used for house heating and cooking in Golestan. The exposure to combustion products of these fuels have been recognized as carcinogenic (diesel), or probably carcinogenic (biomass) to humans,^53,54 due to their content of carcinogenic compounds including polycyclic aromatic hydrocarbons (PAH).^53,54 Our previous findings regarding high levels of a PAH metabolite in urine samples of people residing in high incidence areas of Golestan,⁵⁵ and higher levels of a PAH marker in non-tumoral esophageal epithelium from ESCC patients compared to controls,⁵⁶ provide further support to this hypothesis.

None of the previously recognized risk factors alone could explain the ESCC rates in Golestan and other high incidence areas.^2,3,32 The results of this study strongly suggest that ESCC in Golestan is a multifactorial disease, requiring a combination of exposures for its development.^3,32 Our findings support our original hypothesis that the high ESCC rates in this region are due to a combination of thermal injury (from hot tea), PAH exposure (from indoor air pollution and opium use) and a nutrient deficient diet. We also found, however, that drinking un-piped water and poor oral health were additional strong risk factors (probably through increasing the exposure to N-nitroso compounds). Finally, co-exposure to these identified risk factors was associated with a six to seven-fold increase in the risk of developing ESCC. The continuous decline in the ESCC rates in Golestan since 1970s, following the improvement of basic social infrastructure, provide further support for this hypothesis.

The strengths of this study are its large sample size, minimum (<1%) loss to follow up, small amount of missing data (3%), validated questionnaire data, and being the first prospective study that objectively measured ESCC risk factors, including the actual tea drinking temperature and oral health at enrolment. Reduction in ESCC rates in Golastan may be expected given the improvements in housing and local infrastructure, and dissemination of these results to the local population, coupled with positive actions related to drinking of hot tea, nutrition and opium use, may result in a further decline in ESCC rates.

Abbreviations:

CE-score

combined exposure score

ESCC

esophageal squamous cell carcinoma