Data source and research samples
A CKD thematic data was used from the NHIRD maintained by the Health and Welfare Data Science Center, Ministry of Health and Welfare, Taiwan, of which the detail was prescribed in the previous publication [17]. NHIRD contains the claims records of health care utilization of 99.9% of Taiwan’s 23 million population enrolled in the Taiwan National Health Insurance program[25]. The deidentified information retained in the data included date of birth, sex, residency area, diagnostic codes, medical procedures, and drug prescriptions. Before the NHIRD data was released for research, the personal information of all beneficiaries was de-identification and anonymous to ensure the privacy following rigorous secrecy guidelines. Therefore, informed consent was waived by the Ethics Review of Board Shin-Kong Wu Ho-Su Memorial Hospital (IRB approval number: No. 20200806R).
Study design and study population
The study was designed as a population-based longitudinal cohort study. The selected population were CKD patients diagnosed with hypertension (ICD-10: I10–I15) and diabetes (ICD-10: E08–E13) from January 1, 2006, through December 31, 2006. (n = 18,434). Patients who had previously been diagnosed with malignancy (ICD-10: C00-C96,D45, D47.Z9, E31.22, Z51.12), dementia (ICD-10: F01.50, F01.51, F03.90, and F03.91), cerebrovascular diseases (ICD-10: I60- I63, I65-I69, and G45-G46) were excluded. Patients under regular hemodialysis, peritoneal dialysis, or previously underwent renal transplantation were also excluded. Moreover, patients with ages under 20 years or above 80, or with missing information were also excluded. Finally, a total 11,758 CKD patients with hypertension and diabetes were chosen into analysis (Fig. 1). The index date was defined as January 1, 2007. The data was analyzed from the index date to the first instance of the desired outcome, dementia, or to the end of the observation date (December 31, 2017).
Exposure to study drugs
We used medication possession ratio (MPR) to calculate the degree how the patients had been prescribed the medication of insulin during the following period. MPR is measured as the sum of the days of a given drug in observation period, divided by the total number of days in the observation period [26]. We also defined insulin non-users as the MPR = 0% and insulin users as the MPR > 0%.
Covariates
Baseline comorbidities, including ischemic heart disease (ICD-10: I10–I15), hyperlipidemia (ICD-10: E78), atrial fibrillation (ICD-10:I48), chronic heart failure (ICD-10: I50), peripheral artery disease (ICD-10: I70.2–I70.9), asthma (ICD-10: J45), chronic obstructive pulmonary disease (ICD-10: J44), major depression disease (ICD-10: F32–F33), Parkinson’s disease (ICD-10: G20–G21), rheumatic arthritis (ICD-10: M05–M06), hyperthyroidism (ICD-10: E05), hypothyroidism (ICD-10: E01.8 and E02–E03), gout (ICD-10: M10), and insomnia (ICD-10: G47.0, and F51.0) was defined as having at least 3 outpatient diagnoses or 1 inpatient diagnosis usage within 1 year before the index date. Usage of other drugs, including benzodiazepines, anticoagulants, non-steroidal anti-inflammatory drugs (NSAIDs), acetaminophen, statins, calcium channel blockers, angiotensin converting enzyme inhibitors (ACEIs), angiotensin receptor blockers (ARBs), metformin, beta-blockers, and diuretics was defined as having at least 3 months of usage within 1 year before the index date.
Outcome measurement
Dementia, the main outcome of this study, was identified if a person qualified for a Severe Illness IC Card for dementia from the Taiwan National Health Insurance Administration. This was based on diagnoses by a licensed neurologist or psychiatrist.
Statistical analysis
Continuous data of baseline characteristics is expressed as mean ± standard deviation (SD), while categorical data is expressed as counts with proportions. The difference between groups (insulin users vs non-users) was compared using Chi-squared tests for portion and t-tests for means of continuous variables. Competing risk regression analyses were conducted for dementia developing (all-cause death was considered as a competing event) using the method described by Fine and Gray [27]. The stratified proportional sub-distribution hazard ratio (HR) was calculated to estimate the exposure and covariate effects on the cumulative incidence function.
To assess the robustness of our findings, we added the variables of demographic data, comorbidities, and medications into multivariable regression models step by step. We also did the analyses of dosing effect of insulin on the prevention of dementia, which a Cochran-Armitage test was used for testing the dosing effect of insulin. We stratified insulin users by MPR rates, sorting them in the following groups: those with ≤ 20% MPR, those with MPR > 80% and sorted those with 20–80% MPR into groups per 20% MPR. We used the group with ≤ 20% MPR as our reference point.
Finally, we performed a subgroup analysis by the age, gender, commodities and medications to test the consistency of the insulin protective effect on dementia in CKD population with diabetes and hypertension. If the groups with a portion of patients number < 10% were excluded from subgroup analysis.
All statistical analyses were created by version 9.4 of SAS (SAS Institute, Cary, NC, USA) software. For all tests, a 2-sided p-values < 0.05 was considered as indicative of statistical significance.