Study design and subjects
By the end of 1995, approximately 96% of total Taiwanese population had enrolled in NHI Program [8], a universal health program which has been implemented by NHI Administration under the jurisdiction of Ministry of Health and Welfare. NHI Administration has had contracted 97% of hospitals and 90% of clinics all over Taiwan [9]. In addition, the NHI Administration performs quarterly expert reviews on a random sample for every 50 to 100 ambulatory and inpatient claims to ensure the accuracy of claim files so that information available is considered to be complete and accurate [10]. We used data of ambulatory care claims (2006–2016), inpatient claims (2006–2016), registry for beneficiaries (2007–2009), and death certificate registry (2007–2016) for this study. The ambulatory care claims record all outpatient (including emergency room visit) related information including personal identification number (PIN), date of birth, sex, and date of outpatient visit with a maximum of three leading diagnostic codes while the inpatient claims include all hospitalization information including PIN, date of birth, sex, date of admission and discharge, with a maximum of five leading discharge diagnostic codes and four operation procedure codes. All the dataset can be inter-linked through PIN. The study proposal was approved by the Institutional Review Board of National Cheng Kung University Hospital (A-EX-104-008).
An individual was classified as a type 2 diabetic patient if he or she had an initial type 2 diabetes diagnosis (ICD-9-CM 250.x0, ICD-9-CM 250.x2 or ICD-10-CM E11) in ambulatory care and inpatient claims between 2007 and 2009 and then experienced another one or more diagnosis within the subsequent 12 months. Additionally, the first and last outpatient visits during the 12-month period had to be separated by at least 30 days to avoid accidental inclusion of miscoded patients. Initial diabetic cohort consisted of 1,431,903 patents. We excluded 4,023 subjects with missing information of sex or year of birth, 31,549 patients with type 1 diabetes, and 2,206 patients with gestational diabetes diagnosis between 1 January 2006 and the date of first type 2 diabetes diagnosis in 2007–2009 (i.e., the index date). We also excluded some patients recorded with cardiovascular risk factors for cardiomyopathy in either ambulatory care or inpatient claims before index date. After further excluding 284,255 patients with prior histories of ischemic heart disease, 628,713 patients with prior histories of hypertensive disease, 1,848 patients with prior histories of rheumatic heart disease, 4,025 patients with prior histories of valvular heart disease, 452 patients with prior histories of congenital heart disease, 51 patients with prior histories of acute myocarditis and 513 patients with prior histories of cardiomyopathy, the final diabetic cohort consisted of 474,268 patients (Fig. 1). Respective ICD-9 and ICD-9 codes are shown in Table 1.
Our control group was collected from registry of beneficiaries which contains information of PIN, date of birth, sex, geographic area of each member's NHI unit, and date of enrollment and withdrawal from NHI each time. Between 2007 and 2009, there were 23,328,994 individuals in the registry of beneficiaries. We excluded 1,029,105 subjects with missing information of sex or year of birth, 1,890,133 patients with either type 1 or type 2 diabetes, 168,681 patients with gestational diabetes, 1,041,366 patients with prior histories of ischemic heart disease, 1,880,906 patients with prior histories of hypertensive disease, 29,739 patients with prior histories of rheumatic heart disease, 199,060 patients with prior histories of valvular heart disease, 121,996 patients with prior histories of congenital heart disease, 1,227 patients with prior histories of acute myocarditis, 3,187 patients with prior histories of cardiomyopathy recorded in either ambulatory care or inpatient claims between 1 January 2006 and the index date (Fig. 1).
Using the individual matching technique, we randomly selected 1 control by matching 1 type 2 diabetes patient on age, sex, and the index date of type 2 diabetes diagnosis, totally 474,266 controls were selected from the 16,963,594 potential controls. The index date for subjects in the control group was the same as his/her matched type 2 diabetes.
The difference in time between the index date and the date of birth were set as the age of each study subject. We grouped the township/city of each member's NHI unit, either the beneficiaries' residential area or location of their employment, into two urbanization statuses (urban and rural) according to the classification scheme by Liu et al [11].
Follow-up, study end-points, and covariate
With the unique PIN, we linked study subjects to both ambulatory and inpatient claims from the index date to the last day of 2016 to identify the primary or secondary diagnostic codes of the following idiopathic cardiomyopathy diagnoses as the end point of this study: other primary cardiomyopathies (ICD-9-CM: 425.4), hypertrophic obstructive cardiomyopathy (ICD-9CM: 425.1; ICD-10-CM: I42.1), dilated cardiomyopathy (ICD-10-CM: I42.0), other hypertrophic cardiomyopathy (ICD-10-CM: I42.2) or other restrictive cardiomyopathy (ICD-10-CM: I42.5). Each study subject was followed from the index date to date of idiopathic cardiomyopathy occurrence, death censoring, or the last day of 2016, whichever came first. The information on various cardiovascular risk factors for cardiomyopathy including ischemic heart disease, hypertensive disease, rheumatic heart disease, valvular heart disease, congenital heart disease, acute myocarditis were retrieved from ambulatory care and inpatient claims between index date and date of end-of-follow-up, and were considered as potential confounders.
Statistical Analysis
The age- and sex-specific incidence density estimate was calculated with person-years as the denominator under the Poisson assumption. To assess the independent association of type 2 diabetes with the risk cardiomyopathy, we conducted Cox proportional hazard regression model with Fine and Gray’s method and adjusted age, sex, urbanization status and cardiovascular risk factors simultaneously in the model. Adjustment for the geographic variables may help reduce the presence of an urban-rural difference in accessibility to medical health services in Taiwan [12]. We adjusted cardiovascular risk factors that occurred after baseline type 2 diabetes, which might results in a potential for over-adjustment of these comorbidities as some of these cardiovascular risk factors could play a role of mediator located on the causal pathway from type 2 diabetes to cardiomyopathy. To address this potential problem, we conducted a sensitivity analysis that removed adjustment for these confounders.
All statistical analyses were performed with SAS (version 9.4; SAS Institute, Cary, NC). A P value < 0.05 was considered statistically significant.