Nordic Gastric and Esophageal Tumor Study

PROJECT DESCRIPTION

Theory and method

Study design We will conduct a nationwide and population-based multi-national case-control study in the five Nordic countries during the total study period from 1994 through 2022, thus allowing up to 28 years of follow-up. The start of the study period will vary between countries depending on when their national drug registries started, i.e. in 1994 (Denmark), 1994 (Finland), 2002 (Iceland), 2004 (Norway) and 2005 (Sweden).

Participants The study will include all adult residents in Sweden, Denmark, Finland, Iceland and Norway with a first gastric adenocarcinoma diagnosed during the study period according to the national cancer registries. The control participants are selected from the general populations of the five countries and frequency-matched for age, sex, calendar year, and country. The index date for matching is the date of gastric adenocarcinoma diagnosis. For each case of gastric cancer, 10 control participants will be included. Because the controls are matched within each country by age and calendar year, the time-window used to ascertain the exposure will be the same in cases and controls. We will have the same detailed information on all study variables in cases and controls.

Exposure The exposure is long-term PPI-use. Two definitions will be used. The first, high PPI-consumption, is defined as >180 defined daily doses (DDD) of PPI per year during the time from start of the drug registry or entry in the cohort until the index date. The second definition is cumulative exposure to PPI, defined by the total number of DDDs consumption during the time window of exposure. The DDDs are retrieved from the prescribed drug registries. From a Swedish cohort of 8,276,316 adult individuals of the general population that we use for research purposes, 496,579 (6.0%) were exposed to long-term PPI-use (>180 DDDs).

The comparison exposure is long-term use of an H2RB (>180 DDDs), measured using the same two definitions described for PPI, i.e., high H2RB consumption and cumulative exposure. DDDs for H2RB are also retrieved from the prescribed drug registries.

Outcome The study outcome will be gastric adenocarcinoma. Adenocarcinoma represents >95% of all gastric malignancies. We will not assess rare histologies of gastric malignancy because these have different aetiology, including the potential association with PPI-use. We will also conduct separately sub-analyses of cardia adenocarcinoma and non-cardia adenocarcinoma.

Confounders A confounder is by definition associated with both the exposure and the outcome (and is not in the causal pathway between these).29 The main risk factor for gastric adenocarcinoma is 1) Helicobacter pylori-infection, but also other less strong risk factors might also have some confounding effects in this project, i.e., 2) age (as for most cancer types), 3) sex (a majority of patients with gastric adenocarcinoma are men), 4) calendar year (there have been changes in diagnostic ability of gastric adenocarcinoma), 5) country (there are some variations in incidence rates of gastric adenocarcinoma between the Nordic countries), 6) comorbidity (some severe diseases included in the Charlson comorbidity index might increase the risk of gastric adenocarcinoma), 7) long-term medication with aspirin or non-steroidal anti-inflammatory drugs (such medication may decrease the risk of gastric adenocarcinoma) and 8) long-term medication with statins (could also decrease this risk).

All these 8 variables will be controlled for. Age, sex, calendar year, and country will be controlled for by matching. Helicobacter pylori, comorbidity, long-term medication (>180 defined DDDs per year) with aspirin or non-steroidal anti-inflammatory drugs and statins will be adjusted for using multivariable logistic regression. Regarding Helicobacter pylori, we will use treatment of this infection as a marker of infection, which is the most complete and valid way to assess presence of this infection in the dataset. For comorbidity we will use the most recent version of the well-validated Charlson comorbidity index.30 We have excellent data of both medication variables from the drug registries.

Data sources For each participant, both cases and all controls, we will collect data from the following nationwide registries in all five countries: 1) Prescribed drug registries, 2) cancer registries, 3) patient registries, 4) cause of death registries and 5) registries of the total population. This will allow data on demographic characteristics (age, sex, calendar year, country), use of all relevant medications (including PPI, H2RB, Helicobacter pylori treatment, aspirin or non-steroidal anti-inflammatory drugs and statins with substance, brand name, ATC-code, formulation, package size, amount, dosage and dates for prescription and dispensation), comorbidity (all diagnoses and surgical procedures), and mortality.

Project organisation

Research team Ten people are directly involved in this project. The project originates and is organised from our research group at Karolinska Institutet (KI). The local team includes 7 people: The chief investigator Jesper Lagergren (main applicant), the biostatistician Giola Santoni, the statistician Jacinth Yan, the postdoctoral researcher Cecilia Radkiewicz, the PhD candidate we hope to recruit, and the administrator Giulia Marras. In addition, we have representatives of each of the four non-Swedish countries included in the project, i.e., associate professor My von Euler-Chelpin (Denmark), professor Joonas Kauppila (Finland), associate professor Helgi Birgisson (Iceland) and associate professor Eivind Ness-Jensen (Norway).

The main roles of these team members are outlined here:

1. Jesper Lagergren will lead the work and the team conducting the work, be responsible for the progress of the project, and be supervisor of the postdoc and PhD candidate.
2. Giola Santoni will guide and supervise in the data complex data management and statistical analyses and co-supervise the PhD candidate.
3. Jacinth Yan will conduct complex data management and analyses.
4. Cecilia Radkiewicz will help with the planning and co-supervise the PhD candidate.
5. The PhD candidate will draft the detailed study protocol, do data management and statistical analyses with support of Giola Santoni, and draft the paper.
6. My von Euler-Chelpin is responsible for the Danish data collection.
7. Joonas Kauppila is responsible for the Finnish data collection.
8. Helgi Birgisson is responsible for the Icelandic data collection.
9. Eivind Ness-Jensen is responsible for the Norwegian data collection.
10. Giulia Marras will help organise the collaboration and administer the project.

Data storage and management Because Danish law does not allow their data to leave the country, data will be stored at a safe server at the governmental agency Statistics Denmark, and managed by remote access. Giola Santoni is responsible for the data management and statistical analyses.

Main applicant The main applicant, Jesper Lagergren, will spend approximately 10% of his work-time for this specific project. He is the chief investigator of this multi-national project.

DATA ANALYSIS AND STATISTICS

Design This will be a population-based case-control study with prospectively collected data on all variables for all cases and controls in the five Nordic countries. The data collection is prospective, which avoids recall bias and selection bias. For each case of gastric adenocarcinoma, 10 control participants from the general populations will be recruited. The control participants will be frequency-matched to the cases regarding age, sex, calendar year and country of residence. The additional possible confounders, i.e., Helicobacter pylori, comorbidity and medication with aspirin or non-steroidal anti-inflammatory drugs as well as statins will be adjusted for by modelling.

Statistical analysis Conditional logistic regression will be used to estimate the relative risk of gastric adenocarcinoma comparing long-term PPI-users with non-users, providing odds ratios with 95% confidence intervals. Three models will be applied: 1) A crude model where only the matching variables are controlled for; 2) an adjusted model additionally adjusted for Helicobacter pylori, comorbidity and medication with aspirin or non-steroidal anti-inflammatory drugs as well as statins, and 3) a model with multivariable adjustment for all 8 confounders listed under 'Confounders' above, which will guarantee that these variables are completely adjusted for. We will also do stratified analyses for each of these 8 confounders. In addition, we will assess whether increased length of PPI-use (>10 years) among long-term users is associated with higher risk estimates, which would argue in favour of causality. Except for conventional confounding, confounding by indication (particularly protopathic bias) is a special threat to the validity of many pharmaco-epidemiological studies. However, long-term PPI-use is not an indication for gastric adenocarcinoma. Thus, this should not be a problem in the present project.

Statistical power One of the advantages of the project is the excellent statistical power, which is due to inclusion of cases from five entire countries, long and complete follow-up and relatively high prevalence of long-term PPI-users. In order to further optimise the power, we include 10 controls per case, after which further controls will not improve power.31 The main analysis will include all approximately 47,000 cases and 470,000 controls, and sub-analyses of non-cardia adenocarcinoma will include 35,300 cases and 353,000 controls, while the sub-analysis of cardia adenocarcinoma will include 11,700 cases and 117,000 controls. The exposure prevalence of long-term PPI-use is 6% (and 2% for H2RB use). The main analysis has 100% power to verity an odds ratio already of 1.10, and we expect a much stronger association. For the analysis of cardia adenocarcinoma, the study has 94% power to verify an odds ratio of 1.15.