1. Study design (population)
Korea has a single payer, universal health insurance system that covers more than 99% of its population. The National Health Insurance Service (NHIS) has a comprehensive health database of both inpatients and outpatients. It includes all medical claims data for prescription drugs, codes of diagnosis and treatment, and all reimbursed medical procedures. In addition, the claims data include information such as age, sex, accompanying diseases, and medications. This study on HBV was conducted using insurance claims from the NHIS containing any codes referencing the diagnosis or treatment of CHB and related disease in Korea, an HBV-endemic country. The Institutional Review Board/Ethics Committee of “Blinded for peer review” approved this study and granted a waiver of informed consent because of the retrospective study design (“Blinded for peer review”). This study was performed in accordance with local law and the Declaration of Helsinki. All authors had access to the study data and reviewed and approved the final manuscript.
We collected data for all CHB patients in Korea who were newly diagnosed with ML between January 1, 2002, and December 31, 2016. As a control group, we used the propensity score matching method to collect data on 10 times that number of CHB patients who did not develop ML (Supplement Figure 1). The use of antivirals was determined based on claim records for lamivudine, adefovir, telbivudine, ETV, and TDF. The treatment periods were calculated based on medication claim date records. Analyzing those data also allowed us to infer the severity of the accompanying disease. In that way, baseline liver function was divided into chronic hepatitis and cirrhosis. In cirrhotic patients, compensation or decompensation was determined by a severity assessment. The index date was the time when ML was diagnosed, when antiviral drugs administered, when treatment drugs administered according to liver function, when a procedure performed to treat complications, or when the patient died.
To compare the effects of antivirals (ETV vs. TDF) on ML development, we performed a subgroup analysis. In Korea, patients with CHB were given ETV from 2007 and TDF from 2012, following reimbursement policies of the NHIS. Treatment-naïve CHB patients prescribed TDF or ETV as initial treatment from 2012 to 2014 were followed up until 2017 to observe the occurrence of ML. To apply the same observation period, the follow-up durations of patients diagnosed with ML (the time taken to diagnose ML after drug administration) were randomly assigned to control patients. CHB patients who started antivirals within the first 3 months before their diagnosis of ML as prophylactic therapy were excluded.
Patients who met one or more of the following criteria were excluded: chronic hepatitis C, chronic hepatitis D, acute viral hepatitis, HIV infection, liver transplantation, malignant neoplasm in the liver including HCC and cholangiocarcinoma, and age less than twenty or more than eighty (limitation of age was used only in survival analysis) (Supplement Table 1).
2. Data collection
All patients were classified using the International Classification of Disease, version 10 (ICD-10) codes. CHB was defined as B18.0 or B18.1. ML was classified by subtype using ICD-10 codes. Mature B cell type was determined as C82, C83, C85, and C88.4. The mature T-cell and NK type was classified as C84, Hodgkin lymphoma as C81, and unknown as C88.0, C88.2, C88.3, C88.7, and C88.9. LC was defined as K74.0, K74.1, K74.2, K74.6. Decompensation was defined as LC taking diuretics for ascites control, vasoconstrictors for variceal bleeding, or a non-selective beta blocker for varix. In addition, in the presence of an abdominal paracentesis code or endoscopic esophageal and gastric varices treatment code, the patient was diagnosed with decompensated cirrhosis (Supplement Table 2).
In addition to liver disease, co-morbidities used in the analyses were diagnosed as the ICD-10 codes for DM, HTN, and CKD (Supplement Table 3). Antiviral treatment was defined as prescription codes for lamivudine, adefovir, telbivudine, ETV, or TDF for more than 6 months (Supplement Table 4).
3. Statistical analysis
To make the control group, we used a propensity score–matching analysis to reduce the effect of selection bias and potential confounding factors. Propensity scores were conducted using the following variables: age, sex, LC, DM, HTN, and CKD. A logistic regression model was used to compare the baseline characteristics between the ML and non- ML groups. The study endpoints were survival and the development of ML in HBV patients. The survival rates and cumulative incidence of ML were estimated using the Kaplan-Meier method and compared using a log-rank test. Univariate and multivariable Cox proportional hazards models were used to compare clinical outcomes. All statistical analyses were performed using R statistical software, version 3.3.3 (R Foundation Inc; http://cran.r-project.org/). The propensity score–matching analysis used the “Matchit” package. Survival estimation and comparisons between groups used the “survival” package, and the cumulative incidence estimates used the “cmprsk” package. All reported P values are two-tailed, and P values less than 0.05 were considered statistically significant.