This study used Korean National Health Insurance Service National Sample Cohort (NHIS-NSC) cohort data, which include information about approximately 1 million patients and have been used for many other publications [6, 7]. The data are randomly sampled and stratified according to age, gender, region, health insurance type, income decile, and individual total medical costs from 2002 to 2013. All Korean citizens are obligated to join the National Health Security System, which comprises the NHI and Medical Aid and is overseen by the Ministry of Health and Welfare. The data include a unique anonymous number for each patient and summarize age, gender, type of insurance, a list of diagnoses according to the International Classification of Diseases (ICD-10), medical costs claimed, prescribed drugs, and medical history. In addition, the unique anonymous numbers are linked to information on mortality obtained from the Korean National Statistical Office.
We conducted a cohort study of newly diagnosed gastric cancer patients, with carcinoma in situ of the gastric (ICD-10 code: D00.2) and malignant neoplasm of the gastric (ICD-10 code: C16), using a 2.5 % stratified random sample (n = 1,025,340) of all citizens on December 31, 2002 (Fig. 1). From this pool, we selected 8,833 patients with a primary diagnosis of gastric cancer from 2002 to 2013. Among them, 7,162 patients with preexisting gastric cancer in July 2011 were removed because coverage of EMR by the national health insurance began in August 2011. After these patients were eliminated, 1,671 patients who were free of the disease before July 2011 were selected.
Demographic characteristics including age, gender, and residential area were analyzed, as well as medical history of hypertension, diabetes, and disabled status. The specific ICD codes were as follows: hypertension, I.10–I.15; diabetes, E.10–E.14.
In terms of residential area, we defined residential area by whether the administrative district is located in a metropolitan area. In Korea, there are seven metropolitan cities of more than 1 million in population: Seoul, Busan, Incheon, Daejeon, Daegu, Gwangju, and Ulsan. The patients living in these cities are categorized as ‘metropolitan’ and all others as ‘non-metropolitan.’
We also considered the level of hospitals where patients were treated with endoscopic resection and divided all hospitals using a hierarchy with three components including tertiary hospitals, secondary hospitals, and primary clinics. Primary clinics are mainly responsible for outpatient visits. Secondary hospitals are usually defined as the second level in the healthcare system: these hire the specialists and manage some complicated cases and acute inpatient cases that are not suitable for primary clinics. Tertiary hospitals are the tertiary referral centers with subspecialists and a large number of beds.
Individual-level income measures
Regarding individual income levels, the NHI premium was used as a proxy measure of precise income because it is proportional to monthly income, including earnings and capital gains. The income deciles of enrolled subjects were categorized into the following four groups: ‘lowest,’ first to second decile; ‘middle-low,’ third to fifth decile; ‘middle-high,’ sixth to eighth decile; and ‘highest,’ ninth to tenth decile.
Neighborhood-level socioeconomic status
A summary measure was used to characterize neighborhood-level deprivation. We used a modified Townsend and Carstairs index  for the Korean situation based on census and other administrative data . In previous studies, four variables from census data were used to calculate the Carstairs index: (1) residents in households headed by unskilled individuals, (2) unemployed males, (3) residents overcrowded, and (4) residents without a car [10–12]. However, because we could not obtain car ownership information from census data, we replaced ‘residents without a car’ with ‘residences not owner occupied.’ The values were derived for each area using 2 % microdata from the 2005 Population and Housing Census from the Korea National Statistical Office. A positive, higher score on the index denotes greater deprivation. This modified index displays a significant association with health and has been shown to be robust for consistency over time and over outcome variables in many previous studies in Korea [9, 13–15]. Thus, the use of this verified area deprivation index provides more reliable results and could help in developing an area-deprived index suitable for health-related studies in South Korea.
The neighborhood deprivation index was calculated at the level of Si (city), Gun (county), and Gu (borough) by merging these four basic indicators, similar to the method used to calculate the Carstairs index. Si, Gun, and Gu are geographic units covering all small areas in Korea. We calculated a z score at the level of Si, Gun, and Gu using the mean and standard deviation of the four indicators. The z score was calculated by subtracting the mean from the observed value for each indicator, dividing the standard deviation by this value, and then summing the four standardized z scores. The indexes were categorized into three groups: ‘low’ (the least deprived), ‘middle,’ and ‘high’ (the most deprived) neighborhoods.
Before statistical analyses, we assessed the distribution of the demographic characteristics among gastric cancer patients at baseline. Continuous variables were expressed as means and standard deviations and were compared using t tests or the Kruskal–Wallis test where appropriate. Baseline categorical variables were expressed as numbers and percentages and were compared using the χ
We also estimated the adjusted odds ratios (ORs) and 95 % confidence intervals (CIs) of the chance for endoscopic resection by applying a multiple logistic regression model. Model fitting was performed using the PROC LOGISTIC process in SAS version 9.3 (SAS Institute, Cary, NC, USA).