Data source and ethical consideration
We used the Korean National Health Insurance Service-Health Screening Cohort (NHIS-NSC) data for this study; a comprehensive description of this cohort is provided elsewhere [17]. Briefly, these data constitute approximately a million participants, 2.2% of the total eligible population, who were randomly sampled from the 2002 and followed for 11 years, until 2013. These data include information about participants' demographics, health insurance claim codes, diagnostic codes, socio-economic status, and medical examination data. This study was approved by the Institutional Review Board (IRB) of Hallym University (IRB No: 2019-10-023), and the need for written informed consent was waived.
Participant selection
Among the 514,866 participants with 615,488,428 medical claim codes, participants were selected for the asthma group according to the definition in our study (n = 81,106). Asthma was defined if the participants from the cohort were diagnosed with asthma (J45, J46) based on the International Classification of Diseases, Tenth Revision (ICD-10). Among these individuals, we selected participants who were diagnosed with asthma more than two times by a physician and who were treated with asthma-related medications. Asthma medications include inhaled corticosteroids (ICS), ICS combined with long-acting β2-agonists (LABAs), short-acting β2-agonists (SABAs), systemic LABAs, leukotriene antagonists (LTRAs), xanthine derivatives, or systemic corticosteroids [18]. Participants in the asthma group who were diagnosed with asthma in 2002 were removed to select asthma participants diagnosed for the first time (washout periods, n = 14,044). Other participants were selected for the control group (n = 433,760). In the control group, participants were excluded if they died before 2003 or had no records after 2003 (n = 34). In addition, the control participants were excluded if they were treated for J45 or J46 ICD-10 codes with no record of an asthma medication prescription (n = 108,730). Participants who were diagnosed with hiatal hernia, Zollinger-Ellison syndrome, systemic sclerosis, achalasia or pyloric obstruction, who underwent gastric surgery, and who had no record of total cholesterol/blood pressure/body mass index (BMI, kg/m2) were excluded. The asthma participants were 1:1 matched with control participants for age, sex, income, and region of residence. To minimize selection bias, the control participants were selected with random number generation. The index date of each asthma participant was set as the time of treatment of asthma. The index date for individual control participants was set as the index date of their matched asthma participant. Therefore, each matched asthma participant and control participant had the same index date. Using the above matching rules, 253,128 control participants were excluded and finally, 64,809 asthma participants were 1:1 matched with 64,809 control participants.
Proton-pump inhibitor (exposure)
PPI users were defined as participants who were prescribed PPIs within the previous year (365 days) before the index date. PPI users were classified into three categories as follows: (1) PPI prescription history, (2) the summation of PPI prescription dates, and (3) PPI prescription dates for each generation of PPIs.
Participants with a PPI prescription were grouped as current PPI users if they were prescribed PPIs within 30 days before the index date. If the participants were prescribed PPIs within 31 days to 365 days before the index date, they were grouped as past PPI users. Others were grouped as PPI non-users.
The summation of PPI prescription dates was divided into four groups: (1) PPI non-users, (2) PPI prescription 1–29 days, (3) PPI prescription 30–89 days, and (4) PPI prescription ≥ 90 days in one year.
In the third category, the participants were recategorized into two categories according to the generations of PPIs [19]. The first generation of PPIs consisted of pantoprazole, omeprazole, and lansoprazole. Second-generation PPIs were esomeprazole, dexlansoprazole, rabeprazole, and ilaprazole. PPI prescription dates were summed according to each generation of PPIs.
Asthma (outcome)
The asthma group included participants who were treated for asthma (ICD-10: J45) or status asthmaticus (J46) with asthma-related medications ≥ 2 times from 2002 through 2015. This definition was adapted from that in a previously validated study [20].
Among the asthma participants, AE was defined if asthma participants visited emergency medical doctors or had admissions that were not their first visit for asthma [21]. Asthma participants who were prescribed steroids ≥ 20 mg a day ≥ 3 days within 2 weeks were also defined as having AE [21].
Allergic asthma was defined as having at least one of the following: 1) history of allergic rhinitis by ICD-10 code (J30) within 1 year or 2) therapy with oral antihistamines, leukotriene modifiers, intranasal corticosteroid spray, or intranasal antihistamines for ≥ 1 month within 1 year. All asthma patients who did not meet the definition of allergic asthma were classified as having non-allergic asthma [22, 23].
Based on the definition above, asthma participants were classified into AE (n = 12,707) and NAE (n = 52,102) groups and allergic asthma (n = 18,194) and non-allergic asthma (n = 46,615) groups. The history of previous PPI prescription was analyzed for an association with asthma by comparing asthma patients to the controls, the AE group to the NAE group, and the allergic asthma group to the non-allergic asthma group (Fig. 1).
To evaluate the effects of PPI on AE, we set a new index date for AE and NAE. The new index date for AE was defined as the first exacerbation date. For NAE, a random date between the first asthma onset and the last day of follow-up was selected.
Covariates
Age groups were divided into 5-year intervals, beginning at 40–44 years and ending at 85 + years [2]. Income groups were classified into 5 classes (class 1 [lowest income]-5 [highest income]). The area of residence was classified as urban/rural based on the definitions in a previous study [2]. Tobacco smoking, alcohol consumption, and obesity according to BMI were categorized the same way as in our previous study [24].
Total cholesterol (mg/dL), systolic blood pressure (mmHg), diastolic blood pressure (mmHg), and fasting blood glucose (mg/dL) were measured. The Charlson Comorbidity Index (CCI) was applied excluding respiratory diseases. GERD and chronic obstructive pulmonary disease (COPD) were defined as in our previous study [25]. Among GERD patients, GERD treatment within the 2 years (730 days) before the index date was quantified.
We summed each histamine-2 receptor antagonist (H2RA) and non-steroidal anti-inflammatory drug (NSAID) prescription date within a year (365 days) before the index date.
Statistical analyses
The demographic characteristics were compared between the asthma and control groups, and the differences were considered standardized difference (SDs).
To analyze the odds ratios (ORs) with 95% confidence intervals (CIs) of PPI prescription history for asthma, PPI prescription dates for asthma, and first generation/second generation PPI prescription dates for asthma, conditional logistic regression was used. In these analyses, crude and adjusted models were stratified for age, sex, income, and region of residence.
Regarding AE and allergic asthma, only asthma participants were included in the analyses. To analyze the odds ratios (ORs) with 95% confidence intervals (CIs) of PPI prescription history for AE/allergic asthma, PPI prescription dates for AE/allergic asthma, and first generation/second generation PPI prescription dates for AE/allergic asthma, unconditional logistic regression was used. In these analyses, various adjusted models were calculated.
Subgroup analyses were performed for the sensitivity analysis. In this study, we divided participants by age (< 60 years old and ≥ 60 years old), sex (males and females), income (low income and high income), and region of residence (urban and rural).
SAS version 9.4 (SAS Institute Inc., Cary, NC, USA) was used for the statistical analysis. We performed two-tailed analyses, and a P value of < 0.05 was considered statistically significant.