Alcohol, diet and body fatness as risk factors for stomach cancer and its subtypes: a pooled analysis

Dallas English-d.english@unimelb.edu.au
Robert MacInnis-macinnis@unimelb.edu.au
Pietro Ferrari-ferrarip@iarc.fr
Andrew Haydon-andrew.haydon@monash.edu
Others

A grant application will be submitted to the Australian National Health and Medical Research Council and will include funding for each cohort to help support data preparation. We propose to work with Stephanie Smith-Warner and mostly leverage the existing data repository and infrastructure of the Pooling Project of Prospective Studies of Diet and Cancer (Harvard), and also to invite other eligible cohorts within the NCI Cohort Consortium to participate in this project.

Stomach cancer is the fifth most common cause of cancer globally with an estimated 952,000 incident cases (6.8% of all cancers) in 2012. Stomach cancer is mostly incurable at diagnosis and progresses rapidly from disease to death for most. Despite a decline in incidence, it is still the third leading cause of death from cancer worldwide, due to its high case-fatality. Survival is modest even in high income countries, making prevention paramount.

Established risk factors for stomach cancer include Helicobacter pylori infection, cigarette smoking and industrial chemical exposure. The Continuous Update Project of the World Cancer Research Fund recently adjudged heavy alcohol use, consumption of processed meat and foods preserved by salting, and greater body fatness as probable causes of stomach cancer, all modifiable lifestyle factors attributable to a significant proportion of the global cancer burden (World Cancer Research Fund International/American Institute for Cancer Research, 2016). Alcohol use, processed meat intake and obesity are implicated in carcinogenesis. Ethanol in alcoholic beverages is causally linked to several cancer sites (but not stomach), with an estimated 5.5% of all cancer cases worldwide attributed to alcohol use (2012). While an association at intakes of ≥45 g/day for stomach cancer has been observed, evaluating risks using lifetime alcohol intake and for light-to-moderate drinking will provide more conclusive evidence. Processed meat is associated with an increased risk of non-cardia stomach while being overweight or obese is associated with an increased risk of cardia cancer. The roles of other commonly encountered dietary and non-dietary factors including red meat and fish, fruit, vegetable and cereal intake, dietary fiber intake, consumption of sugar-sweetened beverages, coffee and tea, and aspirin and NSAID use remain largely elusive.

Although a role in the aetiology of stomach cancer is mechanistically plausible for these largely modifiable factors, none at present is conclusively established as a causal agent. An individual observational study usually does not have sufficient statistical power to examine associations between alcohol use, dietary factors, medication use and body fatness and stomach cancer risk, meaningfully. Evaluating risk differences by subtype or for population subgroups is even harder. Despite evidence from meta-analyses showing associations between some of these exposures and risk of stomach cancer, problems inherent to the meta-analytic process and overtly weak strength of associations have undermined the progression to establishing an unambiguous link between exposure and outcome. In the absence of clinical trials, only a large-scaled pooled analysis of prospective studies will offer conclusive scientific evidence to confirm or rule out a role for these exposures in stomach cancer incidence, mitigating above limitations and generating more accurate and precise results.

Risky drinking, lack of fruit and vegetable intake, greater processed meat intake and body mass index of ≥30 kg/m2 are recognized as threats to optimum human health (Australian National Health and Medical Research Council Guidelines, 2013). We propose to investigate whether these and other potentially modifiable lifestyle factors increase or reduce stomach cancer risk, and we anticipate that this information would translate into better informed prevention strategies. Even modest reductions in population exposure to potentially modifiable causal agents would likely translate to sizeable reductions in stomach cancer risk and premature deaths from stomach cancer.

Our over-arching objective is to assess, with the maximum available precision, alcohol use; intake of meat and fish; intake of fruit, vegetable and cereal; dietary fibre intake; consumption of sugar-sweetened beverages, coffee and tea; use of aspirin and NSAIDs; and adult height, overweight and obesity, as risk factors for adenocarcinoma of the stomach; findings to date have been inconclusive.

Aim 1: To conduct a pooled analysis using data from prospective cohort studies to examine associations between above exposures and risk of stomach cancer
Aim 2: To assess whether these associations differ by anatomical location (cardia vs. non-cardia) and histologic subtype (diffuse-type vs. intestinal-type)
Aim 3: To investigate whether any associations between these exposures and stomach cancer risk are modified by cigarette smoking status/intensity or past Helicobacter pylori infection status

We will conduct a pooled analysis, of data from cohorts participating in the Pooling Project of Prospective Studies of Diet and Cancer and invite other cohorts within the Consortium to participate. The majority of the required questionnaire data has already been collected and harmonized, but some additional data will be requested. New follow-up data will be collected on incident stomach cancers. We will request updated follow-up for other cancer diagnoses and death (for calculation of person time), as well as relevant questionnaire data from more recent follow-up rounds. Data collation, cleaning and harmonization will be done through Stephanie Smith-Warner's group at Harvard. Data analyses will be conducted via remote access to the Harvard server after receiving the necessary approvals from participating cohorts. An analytical working group will be formed with interested cohort PIs/delegates.

Cox regression will be performed to calculate hazard ratios (HR) and 95% confidence intervals for associations between alcohol use; intake of red/processed/other meat and fish; intake of fruit, vegetable and cereal; dietary fiber intake; consumption of sugar-sweetened beverages, coffee and tea; adult height, overweight and obesity; use of aspirin and NSAIDs, and stomach cancer risk. The primary analytic approach will be a two-stage process involving (i) estimating study-specific HRs using harmonized data and (ii) pooling of these results using random effects models. As secondary analyses, we will use an aggregated approach in which the harmonized data from all studies are combined into a single data set and then analyzed. Heterogeneity in associations by disease subtype will be assessed using a joint Cox proportional model (based on a competing risk model) which allows different baseline hazard functions for each subtype, and direct comparison of associations by subtype. We will use a mixed-effects meta-regression model to test for effect modification, including by smoking and Helicobacter pylori infection status. We will routinely examine whether associations differ by sex and geographic region.

Evaluation of an association between these exposures and stomach cancer risk presents analytical challenges. Any relationship, if true, is likely to be weak or moderate; thus the statistical power to detect an association in a single study is likely to be very low plus requires the accumulation of information on a sizeable study population with a sufficiently large number of cases. This is likely only with a collaborative effort to jointly analyze existing information from large prospective epidemiological investigations.

Considering the relatively common incidence of stomach cancer, we will include studies with at least 100 incident stomach cancer cases.

This proposal focuses on incident adenocarcinomas of the stomach that arise from the gastric mucosal glands and make up over 90% of all stomach cancers [coded C16 according to the 10th revision of the International Classification of Diseases (ICD); cardia = C16.0; non-cardia = C16.1…C16.6; overlapping and unspecified lesions = C16.8 and C16.9 constitute around 50% of all stomach cancers in Australia, the United States and the United Kingdom]. The proposed study will have large enough statistical power to detect weak associations of public health and clinical significance between exposures and stomach cancer risk (approximately >10,000 cases), even after stratification by sex (~5,000 cases). We expect the following number of cases (minimum) for subtypes: non-cardia=3,333 and cardia=1,667, and intestinal-type=5,400 and diffuse-type=3,200. The study offers sufficient power to detect a significant linear trend in risk comparing extreme quintiles/categories for exposures for the subtypes.

This proposal focuses on incident adenocarcinomas of the stomach. Clinical information such as stage, grade, anatomical site and histology will be requested from studies. Case ascertainment for studies currently in the Pooling Project of Prospective Studies of Diet and Cancer generally is estimated to be >90%. Date of diagnosis with incident invasive cancer of another type (for censoring); date of death; date last known to be free of cancer; and date last known to be alive will also be requested. Outcome data will be harmonized during project.

Data mostly harmonized within Pooling Project of Prospective Studies of Diet and Cancer: alcohol use; red and processed meat and fish intake; fruit, vegetable and cereal intake; intake of dietary fiber; sugar-sweetened beverages, coffee and tea consumption; adult height, overweight and obesity; use of aspirin and NSAIDs; cigarette smoking.

Data on Helicobacter pylori will be harmonized during the project.

Baseline non-dietary risk factor data including, but not limited to, demographics, socioeconomic status and education, and dietary information such as energy intake will be requested. All studies will provide age and sex. Generally, confounding variables will be categorized similarly across studies.

No