This was a nationwide, population-based, historical cohort study using the National Health Information Database (NHID) established by the National Health Insurance Service (NHIS) in the Republic of Korea. The NHIS is a single-payer insurance program providing compulsory national health screening since 1995, covering almost the entire Korean population of 50 million. Any insured Korean adults aged 40 years or older are subject to receive a health check-up every 2 years (once a year for manual workers), and the participation rate was 74.1% in 2019. The health check-up consists of anthropometry, a self-administered questionnaire on past medical history or health-related behaviour such as drinking or smoking, and laboratory tests.
The NHID incorporates data from the NHIS, and the database includes all information on reimbursement for outpatient visits and hospital admissions. This database has been widely used in numerous studies, since it has been proven to be representative of the Korean population.[22,24,25] The information in the database includes the medical diagnoses recorded in the 10th Revision of the International Classification of Diseases (ICD), along with information on treatments, prescriptions by the physicians and costs incurred. Information on sociodemographic status, clinical risk factors or comorbidities and laboratory findings is also available.
This study was approved by the Institutional Review Board of Kyung Hee University, Seoul, Korea (INB No. KHSIRB-19-180) and was performed in accordance with the ethical guidelines of the 1975 Declaration of Helsinki.
The study recruited a historical cohort of 4 899 493 adult subjects aged 40–69 years who underwent the NHIS health check-up ≥2 times between January 2004 and December 2007. The ICD code and clinical information were used to exclude the study subjects (Table S1). Subjects meeting one or more of the following criteria from 2004 to 2007 were excluded: diagnosis of hepatitis B, hepatitis C, acute viral hepatitis or human immunodeficiency virus infection (n = 345 861); diagnosis of alcoholic liver disease or self-reported excessive alcohol consumption (alcohol intake ≥30 g/day for men and ≥20 g/day for women) (n = 124 448); diagnosis of liver cirrhosis (n = 9830); diagnosis of hepatocellular carcinoma or other malignancies (n = 201 382); diagnosis of Wilson's disease (n = 1191); diagnosis of stroke (n = 166 071) and dementia or death (n = 18 762). A total of 4 031 948 subjects were included in this study (Figure 1).
Assessment of NAFLD and Clinical Outcomes
This study defined NAFLD using the hepatic steatosis index (HSI), which is calculated as 8 × (ALT/AST ratio) + BMI (+2, if female; +2, if diabetes mellitus). HSI was adopted since it is the scoring system developed and validated on the largest cohort of Korean individuals. An HSI >36.0 indicates NAFLD with a specificity of 92.4%, while an HSI <30.0 indicates absence of NAFLD with a sensitivity of 93.1%. The study subjects were categorized into three groups: non-NAFLD, NAFLD and intermediate. Subjects with HSI <30 at all health check-ups were classified into the non-NAFLD group, while those with HSI >36 at one or more health check-ups were classified into the NAFLD group. The remaining subjects were classified into the intermediate group. For the subgroup analysis, NAFLD subjects were further categorized into two groups: NAFLD group 1 (HSI >36 at all health check-ups) and NAFLD group 2 (HSI >36 at one or more, but not all, health check-ups).
The primary outcome of interest was the development of dementia, which was confirmed by ICD-10 codes of dementia (F00-F03 or G30–32) with the history of prescription of antidementia drugs, such as donepezil, galantamine, rivastigmine or memantine. The date of dementia diagnosis was the first date of outpatient or inpatient records that included dementia as the primary diagnosis. The follow-up period for each subject was calculated from January 1, 2008, to the date of dementia diagnosis, death or the last follow-up on December 31, 2017.
Baseline characteristics, concomitant medications and clinical outcomes were obtained from the NHIS database. The data of each patient at the time closest to January 1, 2008 were used as the baseline data. The diagnostic codes based on the NHIS database are presented in Table S1. Anthropometric assessment of each subject was performed by healthcare professionals. Height, weight and blood pressure (systolic and diastolic) were assessed. Body mass index (BMI) was calculated as weight in kilograms divided by height in meters squared. Patients were categorized as non-obese (BMI <25 kg/m2) and obese (BMI ≥25 kg/m2) based on their BMI.[27,28] Baseline comorbidities, such as hypertension and diabetes, were defined by the responses to the medical history questionnaire. When the examinees were registered as disabled in the NHIS database, they were defined as having a disability. Information on vital status was obtained by accessing the death certificate database from the National Population Registry of the Korea National Statistical Office. The death registration was nearly complete, and 98.3% was confirmed by the diagnoses of the physician.
All patients who met the eligibility criteria at baseline were included in the analyses. Categorical and continuous variables were compared using the Chi-square test and t-test respectively. The incidence rates of dementia were computed by dividing the number of newly diagnosed dementia cases during the study period by 1000 person-years (PY). The Cox proportional hazard model was used to compare the clinical outcomes between the groups. We calculated the crude and adjusted hazard ratios (HRs) with 95% confidence intervals (CIs). The multivariable analysis included the following variables: age, sex, income, disability, residence area, hypertension, total cholesterol, fasting blood glucose, systolic blood pressure and diastolic blood pressure. Since death without dementia can lead to informative censoring in the assessment of the risk of dementia, competing risk analysis was performed using Fine and Gray's proportional sub-distribution hazard model.[30–32]
Propensity score-matching analysis was performed to reduce the effect of selection bias and potential confounding factors between the NAFLD and non-NAFLD groups. Propensity scores were derived using the following variables: age, sex, income, disability, residence area, hypertension, total cholesterol, fasting blood glucose, systolic blood pressure and diastolic blood pressure. For propensity score matching, a nearest-neighbour 1:1 matching scheme with a calliper size of 0.1 was employed.
All statistical analyses were performed using SAS (version 9.1, SAS) and R, version 3.3.1 (https://cran.r-project.org/). All reported p values are two-sided, and p values <.05 were considered statistically significant.
Liver International. 2022;42(5):1027-1036. © 2022 Blackwell Publishing