The number of household members as a risk factor for peptic ulcer disease

Peptic ulcer disease (PUD) is caused by many sociodemographic and economic risk factors other than H. pylori infection. However, no studies reported an association between PUD and the number of household members. We showed the number of family members affected by PUD based on sex in a Korean population. This cross-sectional study used 1998–2009 data from the Korea National Health and Nutrition Examination Survey of the Korea Centers for Disease Control and Prevention. Multiple binary logistic regression models adjusted for confounders were constructed to analyze the association of PUD with the number of household members. The number of household members was associated with PUD, age, body mass index (BMI), waist circumference, systolic blood pressure, hemoglobin, glucose, location (urban/rural), income, education level, stress, current drinking, and smoking in both sexes. Men with other household members had a higher PUD risk compared to men or women living alone (reference), and the opposite was observed for women. Men with 4 household members had a higher PUD risk than men living alone in the model adjusted for age, BMI, income, location, education, and stress (OR = 2.04 [95% CI 1.28–3.27], p value = .003). Women with more than 6 household members had a lower PUD risk than women living alone in the adjusted model (OR = 0.50 [0.33–0.75], p value = .001). Women with more household members had a lower PUD risk. However, more men had PUD than women regardless of the number of household members.


Results
Characteristics of the subjects across categories of the number of household members. Table 1 indicates the sex differences between men and women and general characteristics of the PUD and non-PUD groups. Participant characteristics according to categories of the number of household members are indicated in Table 2 for men and Table 3 for women. A significant relationship between PUD and the number of household members was revealed for men. Age, SBP, location, education level, stress, current drinking, and smoking were also significantly associated with the number of household members. Men with 2 household members (7.14%) were more likely to have PUD and high SBP, to be less educated, and older than men in other household sizes. For women, PUD, age, SBP, location, education level, stress, current drinking and smoking had a statistically significant relationship to the number of household members. Women living alone (8.73%) were more likely to have PUD and high SBP, to be less educated, and older than women in other household sizes.
Associations between PUD and the number of household members. Table 4 presents the association of PUD with the number of household members for model comparison with adjustment for covariates. The number of household members was significantly associated with PUD risk for men and women with and without adjustment. These models showed very different trends according to sex. Men with other household members had a higher risk of PUD in all models compared to men living alone (reference group). Specifically, men with 4 household members had a higher risk of PUD than men living alone in model 1 (adjusted for age and BMI) (OR = 2.09 (1. 31-3.35), p value = 0.002), model 2 (adjusted for age, BMI, income, location, and education) (OR = 2.13 (1.33-3.40), p value = 0.002), and model 3 (for age, BMI, income, location, education, and stress) (OR = 2.04 (1. 28-3.27), p value = 0.003). Notably, the risk of PUD decreased for women as the number of household members increased compared to women living alone in most models. Women with more than 6 household members had a lower risk of PUD than women living alone in the crude model (OR = 0. 35

Discussion
Gastric and duodenal ulcer diseases have been studied for a long time worldwide. However, there were no previous studies on the number of household members and PUD. Therefore, we reviewed the literature on family affairs closely in relation to the number of household members, PUD, and sex differences, and we expected that women would have more PUD than men because women are more involved and experience more stress in family affairs than men 42 . The regularity of meal intake and skipping breakfast have a strong effect on PUD 11,15,16,25,43 , and the number of household members is closely associated with the regularity of meal intake, meal preparation, and the role of meal production due to the common activities among family members 11,25,43,44 . For example, Leblanc et al. 45 examined sex differences in eating behaviors and dietary intake based on a food frequency questionnaire and the Three-Factor Eating questionnaire, and they concluded that women engaged in meal preparation each week much more frequently than men. Ma 46 and Quelly 47 noted that women have more responsibility for meal preparation than men in many countries and cultures. Therefore, women play an important role in meal production, ingredient purchases, cooking methods, and decisions on the type, nutrition, and quantity of meals for adults and children in their family 14,19 . Many adult women in Korea secure a job to obtain income in addition to preparing most meals and handling family-related activities, such as house cleaning and washing. Therefore, women expend time and labor and are more stressed in the preparation of meals as the number of household members increases, but this finding is not universal. Despite some ongoing changes, Korea remains a patriarchal society. Therefore, we hypothesized that women would be more likely to have PUD than men as the number of household members increased. However, our results were the opposite of what we expected. Our findings indicated that women were less likely to have PUD as the number of household members increased, and men were more likely regardless of the number of family members. Further studies are needed to clarify this finding.
Several studies suggested that alcohol intake was a risk factor for PUD 30,33,34 or the occurrence of PUD 31,32 , but other studies argued that alcohol intake was not associated with PUD 22,27,29,[35][36][37] . Kato et al. 27 reported that alcohol intake was not a risk factor for gastric or duodenal ulcers in Hawaii. Chou 36 suggested that moderate alcohol intake minimally increased the odds of PUD in a large U.S. population study. Johnsen et al. 22 argued that alcohol and coffee intake were not associated with PUD in a 7-year follow-up study in Norway. Levenstein 16 argued that these controversial results may be due to the total amount of alcohol intake and indicated that moderate intake seemed to strengthen gastroduodenal mucosa, but heavy alcohol intake may cause PUD due to the direct mucosal and acid secretion stimulation. Liu et al. 48 argued that moderate alcohol intake was related to a reduction of H. pylori infection. Our findings are consistent with the results of previous studies 30,33,34 and indicated that alcohol drinking was highly associated with PUD in men and women in crude analyses and that this risk factor showed significant differences according to the number of household members.
Numerous studies reported that smoking was an important risk factor for PUD 10,13,14,22,23,[25][26][27][28][29] . However, some researchers disagreed with the association between smoking and PUD 35,37 . Aldoori et al. 35 demonstrated that current and past smoking was not associated with the risk of duodenal ulcers despite adjustment for age, BMI, dietary fiber, and the use of drugs, such as aspirin and nonsteroidal anti-inflammatory drugs, in a prospective study Our results showed that smoking was highly associated with PUD in men but not women. The relationship between PUD and abdominal adiposity or obesity is not clear 19,20 . Several studies suggested that high BMI was an independent risk factor for PUD 20,23,24 , and other studies reported that BMI was not related to PUD 49,50 . Our results showed that BMI was associated with PUD in men but not women, and waist circumference was strongly related to PUD in women but not men.
Many previous studies suggested that age was one of the most reliable risk factors for PUD 10,13,14,17,[19][20][21][22] . Although the incidence of PUD is decreasing in many countries due to new therapies 1 , the incidence of PUD and its bleeding complications is growing in the elderly population, and PUD mortality, management, and hospitalization are increasing due to the rapid population increase in most countries 1,21,51 . However, one study argued that age, sex, and abdominal symptoms were not risk factors for PUD in Japanese patients 37 . Our findings revealed that age was very strongly associated with PUD in men and women.
Limitations of the study. This study had several limitations. First, we did not determine why men were at higher risk of PUD than women regardless of the number of household members or why women were at lower risk of PUD as the number of household members increased. It is difficult to find causality in the results due to the cross-sectional nature of this study. Further longitudinal studies are needed to identify causal relations. Second, our findings may not be similar to other countries or ethnic groups due to the differences in socioeconomic, environmental, and psychological characteristics. Last, many studies reported that H. pylori infection was highly associated with PUD. However, our study did not consider the effects of H. pylori infection because information of H. pylori infection was not provided in the Korea National Health and Nutrition Examination Survey. Therefore, a limitation of this study is that H. pylori infection was not considered. Further study is needed to reveal the causes of different associations between the number of household members and risk of PUD in men and women and to consider the effects of H. pylori infection. Despite these limitations, this study also has strengths. The statistical results in this study are powerful because the KNHANES provides a nationally representative sample of the Korean population. To our knowledge, this study is the first report of a significant association between PUD and the number of household members in the world.

Methods
Sampling and data source. This cross-sectional study used data from the Korea National Health and Nutrition Examination Survey (KNHANES I-IV) from 1998 to 2009, which included PUD diagnosis. The KNHANES has been performed by the Korea Centers for Disease Control and Prevention (KCDC) since 1998. The purpose of the KNHANES is to produce representative and reliable statistics of national and municipal units of the health, food and nutrition intake of the population and to develop health promotion programs. The present study selected the KNHANES sample using the multistage stratified cluster sampling method, which is a complex sampling design method, to improve the sample representativeness and estimation accuracy.

Definition of PUD.
Subjects with PUD were identified via the question "Do you have PUD diagnosed by a physician?" in a self-administered questionnaire. The PUD group consisted of subjects who checked "Yes", and the non-PUD group consisted of subjects who checked "No" or "Not applicable" according to the KCDC guidelines. Specifically, the diagnostic definition of PUD included gastric duodenal ulcers and gastritis in the KNHANES I and gastric and duodenal ulcers in the KNHANES II-IV. Therefore, if subjects had at least one of the three ulcer types, they were included in the PUD group. www.nature.com/scientificreports/ Measurement and blood test. The variables in the two surveys, health interviews and health examinations, were used in this study to evaluate the association between PUD and the number of household members. We considered variables on socioeconomic status, medical condition and health behaviors in health interviews and variables on anthropometric measures, blood pressure, pulse rate and blood tests in health examinations. Information on socioeconomic status (the number of household members, income, and education) and medical condition (gastric cancer, liver cancer, colorectal cancer, peptic ulcer diabetes, and hypertension) was collected via a self-administered questionnaire. Subjects with a disease such as gastric cancer, liver cancer, colorectal cancer, diabetes or hypertension were identified if they had been diagnosed with one of these diseases by a physician. Information on health behaviors (drinking, smoking, and stress) was collected using face-to-face interviews in the health interview questionnaire 52 . Alcohol drinking was categorized as subjects who had been Table 3. General characteristics of the subjects across categories of the number of household members in Korean women. BMI body mass index, SBP systolic blood pressure, DBP diastolic blood pressure. *: p < .05; **: p < .01; † p < .001. *, ** and † indicate p values of the difference across categories of the number of household members. These p values were obtained from a general linear model for continuous variables and from Rao-Scott chi-squared tests for categorical variables. Continuous variables are summarized as the means ± SE (standard error). Categorical variables are summarized as percentages (SE). All statistical analyses were performed using weight, cluster and stratification parameters to account for the complex sampling design. www.nature.com/scientificreports/ drunk more than once during the last year or never drunk during the last year. Smoking was categorized as subjects who were smoking currently, quit smoking, or had never smoked. Anthropometric measures (BMI, waist circumference), blood pressure and pulse rate, and blood tests (hemoglobin, cholesterol, etc.) were examined according to standardized protocols by trained medical personnel. Blood tests were performed using blood samples obtained after fasting for at least eight hours (Advia 1650, Siemens, New York, USA; Hitachi Automatic Analyzer 7600, Hitachi, Tokyo, Japan). The equipment used was calibrated periodically. BMI was determined by weight and height, and waist circumference was measured at the midpoint between the iliac crest and the lowest rib. Blood pressure was calculated as the average value of the second and third values of three measurements using a mercury sphygmomanometer (Baumanometer; Baum, New York, USA). Statistical analysis. All statistical analyses were performed using complex sample procedures in SPSS 23 for Windows (SPSS Inc., Chicago, IL, USA) to account for the complex sampling survey data. All statistical analyses were performed using weights to obtain estimates that were representative of Korean population to account for the complex sampling design. The weights with stratification were provided by the KNHANES. The significance level of α = 0.05 was used for all statistical tests.
Continuous variables are summarized as the means ± standard error (SE), and categorical variables are summarized as percentages (SE). General linear models were used for continuous variables, and Rao-Scott chisquared tests were used for categorical variables to compare differences between the PUD and non-PUD groups. The same methods were used to compare differences across categories of the number of household members according to variable type.
Binary logistic regressions were used to investigate the association of PUD with the number of household members for each sex after the data were transformed by standardization. Multiple binary logistic regression Table 4. Adjusted odds ratios for PUD according to the number of household members. OR odds ratio, CI Confidence interval. Model 1: Adjusted for age and body mass index. Model 2: Adjusted for age, body mass index, income, location, and education. Model 3: Adjusted for age, body mass index, income, location, education, and stress. p values were obtained from multiple logistic regression analyses with adjustment. These analyses were performed using weight, cluster and stratification parameters to account for the complex sampling design. Odds ratios are presented with 95% confidence intervals. www.nature.com/scientificreports/ models were constructed to analyze the association between PUD and the number of household members with different combinations of adjustment variables, taking into account several confounders that affected PUD in previous studies. Model 1 was adjusted for age and BMI as covariates. Model 2 was adjusted for age, BMI, income, location, and education, and model 3 was adjusted for age, BMI, income, location, education, and stress as covariates. Odds ratios with 95% confidence intervals and p values were indicated by sex.

Data availability
Data are available from the Korea National Health and Nutrition Examination Survey of the Korea Centers for Disease Control and Prevention (http://knhan es.cdc.go.kr/ and https ://knhan es.cdc.go.kr/knhan es/sub03 /sub03 _02_02.do. www.nature.com/scientificreports/