Data source
This nationwide cohort study was based on patient data obtained from the National Health Insurance Database (NHID), which is managed by the Taiwan National Health Research Institute (NHRI). The NHID contains health care data for 99% of the population of Taiwan (approximately 23 million people). The NHI sample files, which are established and managed by the NHRI, consist of comprehensive use and enrollment information for a randomly selected sample of 1,000,000 NHI beneficiaries, representing approximately 5% of all enrollees in Taiwan in 2000. The NHRI is the only institute that is approved to conduct samplings of a representative portion of the entire population. Although privacy protections are maintained, the reimbursement data for sampled patients were retrieved and used for academic research after obtaining approval. The NHID contains comprehensive information, including demographic data, dates of clinical visits, diagnostic codes, and details of prescriptions. The International Classifications of Diseases, Revision 9, Clinical Modification (ICD-9-CM) was used to define diseases during the period of this study. This study has been approved by the NHRI.
Study design
A nested case–control approach is a useful alternative to cohort analysis to study time-dependent exposure [21]. The risk estimates from cohort and nested case–control analysis should be similar if confounding factors are controlled in both analyses. The strength of the nested case–control study design may be particularly useful in rare cases [22].
Study patients
Diabetes mellitus (DM) patients were identified using inpatient discharge records, or by 3 or more ambulatory care claims with a diagnosis of ICD-9-CM: 250. From the NHID, patients who had DM, and were using pioglitazone and rosiglitazone between January 1, 2000 and December 31, 2010 were compared to DM patients who were not treated with pioglitazone or rosiglitazone. Patients who had ever received a gastrectomy or vagotomy were excluded from the analyses. Patients with a previous diagnosis of gastric cancer or Zollinger–Ellison syndrome, and those who were less than 30 years old and more than 99 years old were also excluded. We further excluded those who had a hospital admission with a discharge diagnosis of insulin-dependent diabetes mellitus (ICD-9-CM 250.x1, 250.x3) or received a catastrophic illness certificate issued by the Department of Health for Type 1 Diabetes.
Exposure to pioglitazone or rosiglitazone
Information on all TZDs prescription was extracted from the NHRI prescription database. The defined daily dose (DDD) is the assumed average maintenance dose per day for drugs administered to adults and used according to their main indications. The DDDs recommended by the World Health Organization (WHO) [23] were used to quantify the use of TZDs. Cumulative DDD was estimated as the sum of dispensed DDDs of any TZDs (pioglitazone or rosiglitazone) from January 1, 2000 to the index date. The gathered data comprised the date of prescription, daily dosage, and the number of days of drug supply. The main exposure of interest was the use of pioglitazone or rosiglitazone, which entered Taiwan’s market in June 2001 and March 2000, respectively.
Definition of gastric cancer
All patients aged 30–99 years in the study cohort, with the first occurrence of stomach cancer ICD-9-CM 150.0-150.9 during the 11-year period, were included as cases based on inpatient discharge records. Patients with a previous diagnosis of gastric cancer were excluded. A diagnosis of gastric cancer in the NHID required histologic confirmation to be reported to the registry of the Catastrophic Illness Patient Database. All potential cases were validated by a linkage through the National Cancer Registry.
Definition of control group
A risk-set sample (control sample from those in the original study cohort who remained free of outcome at the time when a case occurred) matched by age (within 5 years), sex, and the number of days of follow-up, was used as controls for the cohort. For newly diagnosed Type 2 Diabetes patients, case and controls were also matched based on antidiabetic treatment duration (within 30 days) at cancer diagnosis. For newly diagnosed diabetic patients, the scheme that matched follow-up duration also considered diabetic duration. For prevalent patients with unknown duration, we selected controls with the same follow-up duration to reduce the confounding effect of diabetes duration. Up to four controls were selected for each patient [24].
Definition of peptic ulcer history and ulcer bleeding history
All endoscopically-diagnosed peptic ulcers in DM patients prior to the date of gastric cancer diagnosis, according to ambulatory care and inpatient discharge records, were used for peptic ulcer history. Peptic ulcers were defined as gastric ulcers (ICD-9-CM 531), duodenal ulcers (ICD-9-CM 532), and nonspecific peptic ulcers (ICD-9-CM 533) following endoscopic confirmation from January 1, 2000 to the index date. Based on inpatient discharge records prior to the date of gastric cancer diagnosis, peptic ulcer bleeding (following endoscopic confirmation) was used as ulcer bleeding history. Peptic ulcer bleeding was defined using ICD-9-CM codes 531.0, 531.2, 531.4, 531.6, 532.0, 532.2, 532.4, 532.6, 533.0 533.2, 533.4 and 533.6 following endoscopic confirmation from January 1, 2000 to the index date.
Definition H. pylorieradication rate
Patients placed into the category of H. pylori eradication therapy were defined as those who received triple or quadruple therapy during the same inpatient discharge record or outpatient visit from January 1, 2000 to the index date. The duration of therapy was between 7 and 14 days. H. pylori infection is treated with multidrug regimen that consists of proton pump inhibitors (PPIs) or histamine receptor-2 blockers (H2-blockers), clarithromycin or tetracycline, amoxicillin or metronidazole, and potentially bismuth. The PPIs administered to patients that were evaluated in this study were lansoprazole, esomeprazole, omeprazole, pantoprazole, and rabeprazole and the H2-blockers were cimetidine, famotidine, nizatidine, ranitidine, and roxatidine [25].
Definition of comorbidities
Patient comorbidities were identified using inpatient discharge records or by 3 or more ambulatory care claims with the diagnosis of coronary artery disease (CAD): ICD-9-CM 410–414, cerebral vascular disease (CVD): ICD-9-CM 430–438, chronic liver disease (CLD): 070.2x, 070.3x, V02.61, 070.41, 070.44, 070.51, 070.54, V02.62, 571.4, 571.2, 571.5, 571.6, 571.0x, 571.1x, 571.2, and 571.3x, chronic obstructive pulmonary disease (COPD): ICD-9-CM 490–492, 494, and 496, chronic kidney disease (CKD): ICD-9-CM 580–589, 250.4, 274.1, 283.11, 403.x1, 404.x2, 404.x3, 440.1, 442.1, 447.3, 572.4, 642.1x, 646.2x, and 794.4, and gastroesophageal reflux disease (GERD): 530.81 or erosive esophagitis (EE): 530.11.
Use of medication
Patients were categorized by their use of metformin, sulfonylurea, glucosidase inhibitors, meglitinides, dipeptidyl peptidase 4 (DPP-4) inhibitors, insulin, statins, angiotensin receptor blockers (ARBs), angiotensin-converting enzyme (ACE) inhibitors, aspirin, cyclooxygenase-2 (COX-2)-specific inhibitors, and non-steroidal anti-inflammatory drugs (NSAIDs) with at least 2 prescription prior to the index date of gastric cancer diagnosis [26].
Statistical analysis
For comparisons of proportions, χ
2 statistics were used. A conditional logistic regression model was used to estimate the relative magnitude in relation to the use of TZDs. Exposure was defined as patients who received at least 2 prescriptions for a TZD at any time between January 1, 2000 and the index date [26]. In the analysis, the participants were categorized into one of 2 TZDs exposure categories: nonuse, past use, and recent use. Furthermore, we divided the person-time-product into recent use (including current medication and discontinuation of medication prior to gastric cancer diagnosis < 6 months), past use (drug discontinuation to gastric cancer diagnosis ≥ 6 months), and non-use. The participants were categorized into users of dosages less than the median (< 260 DDDs) and users of dosages equal or greater than the median (≥ 260 DDDs). In the dose- and duration- response analysis, we calculated the odds ratios (OR) for higher (≥ 260 DDDs) or lower (< 260 DDDs), and for cumulative treatment duration ≥ 1 year or < 1 year. The OR and their 95% confidence intervals (CI) were calculated using patients with no exposure as the reference. All statistical analyses implemented in the present study were performed using an SAS statistical package (SAS System for Windows, version 9.2; SAS Institute, Cary, NC, USA).