Diabetes Poses a Higher Risk of Hepatocellular Carcinoma and Mortality in Patients With Chronic Hepatitis B

A Population-based Cohort Study

Yu-Chiau Shyu; Ting-Shuo Huang; Cheng-Hung Chien; Chau-Ting Yeh; Chih-Lang Lin; Rong-Nan Chien

Disclosures

J Viral Hepat. 2019;26(6):718-726. 

In This Article

Methods

Data Source

This study protocol conformed to the Helsinki Declaration and approved by the Institutional Review Board (IRB; no. 102-2430B) of Chang Gung Memorial Hospital. Patient records and information were anonymized and de-identified prior to analysis, and the need for written informed consent was waived by the IRB.

The database used in this study was the ambulatory claims data of the Longitudinal Health Insurance Database 2000 (LHID2000) and NHIRD. Implemented in 1995, Taiwan's National Health Insurance (NHI) programme is a compulsory universal health insurance programme, and the NHI Bureau is the sole payer of healthcare services. The electronic claim forms include such information as the patient's gender, as well as the medical institution visited, diagnostic codes, the date of any prescriptions given, the drugs prescribed and any claimed medical expenses. The reliability of diagnostic codes in the NHIRD has been proven by a previous study.[10]

Study Population

The target population in this study was patients with CHB and with/without DM aged 40 to 90 years old. A CHB diagnosis was defined as at least two NHI claim records in any diagnosis codes per visit with the presence of the International Classification of Diseases, 9th revision, Clinical Modifications (ICD-9-CM) codes of ICD-9 070.2, 070.3 and V02.61. The strategy of extraction of the sample in this study was as depicted in Figure 1.

Figure 1.

Study selection process of participants

The DM patients (n = 123 466) who were alive before 1 January 2000 and the newly diagnosed with DM during the period of 1 January 2000 to 31 December 2000 in the LHID2000 were selected. The LHID2000 used newly diagnosed DM codes based on the International Classification of Diseases, ninth revision (ICD-9-CM) as previously described.[11] In this study, patients newly diagnosed with DM in 2000 were included if they met the following conditions: (a) at least one hospitalization for which DM was diagnosed as one of the discharge codes (A-code A181 before 2000 or ICD-9 code 250 after 2000) or taking DM drugs; (b) at least two outpatient visits to the DM clinic in the same year; (c) no hospitalizations or outpatient visits for DM from 1996 to 1998; and (d) 20 years of age or older. In addition, we excluded any patients diagnosed with cancer (International Classification of Diseases, ninth revision [ICD-9] codes 140-208 and A-code A08x-A14x) and human immunodeficiency virus (HIV) infection (ICD-9 codes 042, 07953, V08, V6544, 79571) prior to 2000 from the study.

The non-DM comparison group were from NHIRD after exclusion of those who were ineligible, and we carried out a two-stage matching process. First, the subjects were matched 1:1 according to age and gender, and then, a 1:1 ratio propensity score was used to match participants with chronic HBV infection (ICD-9 codes 070.2, 070.3 and V02.61), and with or without DM, by age, gender, alcohol-related liver disease and baseline liver cirrhosis. Finally, this study cohort was continuously monitored to determine HCC development (ICD-9 codes 155, 155.0 and 155.2, and A-code A095) through December 2011 or censorship due to death. HCC diagnoses were confirmed by linking them with the Registry for the Catastrophic Illness Patient Database (RCIPD), a sub-section of the NHIRD, which is approved by the Bureau of National Health Insurance with pathological reports and other supporting documents, such as laboratory and image studies.[12] The date of death was identified in the RCIPD, while the date of discharge was identified by insurance coverage within 1 month of discharge as "critical and against medical advice" or after an emergency department visit during which intravenous epinephrine was administrated.

Covariate Assessment

In addition to DM, we adjusted all comorbidities in the Deyo-Charlson Comorbidity Index for their potential effect.[13,14] ICD-9, Clinical Modification (ICD-9-CM) codes derived from the NHI claim data were screened for comorbidities using a revised mapping algorithm developed by Quan et al[14] Moreover, being obese or overweight was defined by ICD-9 codes V77.8, 783.1, 278.00, 278.01 and 278.1; liver cirrhosis was defined as nonspecific cirrhosis (ICD-9 codes 571.5, 571.6), alcoholic cirrhosis of the liver (ICD-9 code 571.2), ascites (ICD-9-CM: 789.5), varices (ICD-9-CM: 456.0, 456.1, 456.2), encephalopathy (ICD-9-CM: 572.2) and hepatorenal syndrome (ICD-9-CM: 572.4).[15] All comorbidities were recorded as time-dependent variables for the subsequent statistical analysis.

Statistical Analysis

The propensity score matching used the nearest neighbour matching method with a 1:1 ratio between the diabetes and nondiabetes groups.[16] We included age, gender, alcohol-related liver disease and baseline liver cirrhosis in our propensity score model. All parametric data are reported as a median with an interquartile range (IQR), while dichotomous data are summarized as a frequency and percentage. In the univariate analysis, baseline characteristics were compared between the DM group and non-DM group using the chi-squared test, Fisher's exact test or Wilcox rank-sum test, as appropriate. The reasoning and theoretical principles for applying the multi-state model approach to reflect the transitions from "start-to-HCC," "start-to-death" and "HCC-to-death" are as previously described.[17–23] To investigate the potential causal affects between DM and HCC development, we constructed directed acyclic graphs (DAGs) to represent the hypothesized causal relationships among the variables (Figure S1).[24–26] In order to estimate the overall influence of DM on HCC development, we used age, gender, alcoholic liver diseases, baseline liver cirrhosis and overweight/obesity for the "start-to-HCC" transition. To model the transitions of "start-to-death" and "HCC-to-death," we adopted the AIC stepwise algorithm for the contributed R package "MASS".[27] All reported confidence intervals (CI) and tests were two-tailed, and a P-value < 0.05 was considered statistically significant. All statistical analyses were performed with R version 3.3.0 (R Foundation for Statistical Computing, Vienna, Austria) with "MatchIt",[16] "mstate",[17,18,24] "survival"[21] and "simPH"[23] packages.

Supplemental Figure 1.

The hypothesized causal relationships between variables were represented by the directed acyclic graph (DAG). The blue nodes represent the mediators, while the red nodes represent confounders. The green lines indicate causal paths, while the red lines indicate biased paths. The total causal effects of DM on HCC can be broken down into five paths (through post-DM NASH, post-DM CHB treatment, DM medications, LC after DM, and direct on HCC). We were able to determine the minimal sufficient adjustment sets for estimating the total effect of DM on HCC (Age, Gender, Alcohol, BMI, and LC before DM) using the DAG.

Comments

3090D553-9492-4563-8681-AD288FA52ACE

processing....