Data source
This population-based nested case–control study was based on data extracted from the Urumqi Hypertension Database (UHDATA) established by the Hypertension Research Institute of People’s Hospital of Xinjiang Uygur Autonomous Region, China. This database, which is updated continuously, was created in December 2019 by our hospital and Yidu Cloud Company, using customized natural language processing (NLP) software. The database includes all patients with definite diagnosis of hypertension who have visited our hospital since 2004 when our hospital began using an electronic medical records system. The database contains 1458 elements. including data from outpatient and inpatient electronic medical records, parameters of Hospital Information System, Laboratory Information System, Picture Archiving and Communication System (PACS), and Radiology Information System; and nursing records.
Study Population
The study population was assembled using the UHDATA. Patients were eligible for inclusion if they 1) were ≥ 18 years old; 2) had hypertension complicated by hyperlipidemia and/or were classified as high risk; 3) had been longitudinally observed from January 1, 2006, to December 31, 2018; and 4) had at least two hospitalization records. Hyperlipidemia and high-risk hypertension were defined per the 2019 ESC/EAS dyslipidemia management guidelines. From among the 52,146 patients that met these criteria, 11,466 patients were excluded because of 1) previous history of aortic dissection, 2) hepatic insufficiency, 3) vasculitis, 4) Marfan syndrome, and 5) incomplete data. The remaining 40,680 patients were observed from January 1, 2006, until the diagnosis of aortic dissection, death, or end of the study period (April 30, 2022), whichever came first. (Fig. 1).
Identification and definition of outcomes
The primary outcome was the first mention of aortic dissection in UHDATA, i.e., International Classification of Diseases 10 revision (ICD-10) diagnostic codes 171.0, I71.004, I71.0002, and I71.404 (aortic dissection); I71 .0011 (type A aortic dissection); I71.0021 (type B aortic dissection); I71.404 (abdominal aortic dissection); I71.900 and I71.902 (aortic aneurysm); I71.201 (ascending aortic aneurysm); I71.207 (descending aortic aneurysm); I71.210 (aortic root aneurysm); I71.211 (thoracic aortic aneurysm); I71.600 (thoracoabdominal aortic aneurysm); and I71.400 and I71.402 (abdominal aortic aneurysm) [19]. The results of aortography and computed tomography and the classification of cases were cross-checked by two investigators to exclude false positive cases. A total of 647 patients who developed aortic aneurysm were identified; these patients comprised the case group. The hospital admission date or the event date was the index date of case diagnosis. For each case, 10 controls matched for age, sex, and index date of case diagnosis were selected from among the 40,033 patients who did not develop aortic aneurysm.
Exposure
Exposure to statins was defined as use of any statin prior to the date of aortic dissection diagnosis (or the corresponding date in controls) either during hospitalization (i.e., long-term medical orders in inpatient records) or as outpatient (i.e., statin prescription more than once in outpatient records). Statins were classified as lipophilic or hydrophilic using the drug distribution pharmacokinetic parameter logP (partition coefficient). Lipophilic statins (atorvastatin, fluvastatin, lovastatin, simvastatin, and pitavastatin) were defined as those with logP > 0, and hydrophilic statins as those with logP < 0 [20–22].
Data collection and covariables
Selection of covariates in this study was based on previous literature and included the following: general data (age, sex, smoking status, alcohol consumption, body mass index, systolic blood pressure [SBP], diastolic blood pressure [DBP], pulse rate, duration of hypertension); biochemical parameters (total cholesterol [TC], triglycerides, and high-density lipoprotein cholesterol [HDL-C], low-density lipoprotein cholesterol [LDL-C], serum creatinine, alanine aminotransferase [ALT], aspartate aminotransferase [AST]); comorbidities such as coronary artery disease [CAD], cerebrovascular disease, diabetes, and renal insufficiency; and use of medications such as antihypertensive drugs, statins, aspirin, and so on).
Statistical analysis
Continuous variables (age, pulse rate, SBP, DBP, body mass index, ALT, AST, serum creatinine, TC, triglycerides, HDL-C, LDL-C) were expressed as means ± standard deviation and compared using the t test (for normally distributed variables) or the Mann–Whitney U test (for non-normally distributed variables). Categorical variables were expressed as frequencies and proportions and compared using the chi-square test. The variance inflation factor (VIF) was used to test for collinearity among TC, triglycerides, HDL-C, and LDL-C (tolerance < 0.1 and VIF > 10 was taken as indication of multicollinearity). Conditional logistic regression models were used to calculate odds ratios (ORs) and 95% confidence intervals (CIs). Additionally, stratified analysis was carried out to evaluate the association between statins and aortic dissection in subgroups of age (< 50 years/≥50 years), sex, body mass index (BMI; <24/≥24), smoking (never/current), alcohol drinking (never/current), duration of hypertension (5 years/5–10 years/>10 years), SBP (< 140 mmHg/140–159 mmHg/160–179 mmHg/≥180 mmHg), DBP (< 90 mmHg/≥90 mmHg), LDL-C (< 1.56 mmol/L/≥1.56 mmol/L) and comorbidities such as CAD and diabetes (present/absent). In addition, sensitivity analysis was performed after excluding patients taking different types of antihypertensive drugs. Statistical analysis was performed using SPSS 26 (IBM Corp., Armonk, NY, USA) and R 4.2 (https://www.r-project.org/).
Missing data
Missing data is an inevitable feature of observational studies. In this study, 7.88% of covariate data were missing. To minimize resulting bias, the missing data were filled using multiple interpolations, and five imputations were filled employing multiple interpolations. The imputation data were not significantly different from the original data (Table S1). The Rubin method was applied for statistical analysis [23].