Mortality Risk and Decompensation in Hospitalized Patients with Non-Alcoholic Liver Cirrhosis: Implications for Disease Management

Here we aimed to assess the mortality risk and distribution of deaths from different complications and etiologies for non-alcoholic liver cirrhosis (NALC) adult inpatients and compare them with that of the general hospitalized adult population. Hospitalized patients with a primary diagnosis of NALC and aged between 30 and 80 years of age from 1999 to 2010 were identified using a population-based administrative claims database in Taiwan. They were matched with a general, non-NALC population of hospitalized patients. Causes of death considered were variceal hemorrhage, ascites, hepatic encephalopathy, spontaneous bacterial peritonitis, hepatocellular carcinoma, jaundice, and hepatorenal syndrome. A total of 109,128 NALC inpatients were included and then matched with 109,128 inpatients without NALC. Overall mortality rates were 21.2 (95% CI: 21.0–21.4) and 6.27 (95% CI: 6.17–6.37) per 100 person-years, respectively. Among complications that caused death in NALC patients, variceal hemorrhage was the most common (23.7%, 11.9 per 100 person-years), followed by ascites (20.9%, 10.4 per 100 person-years) and encephalopathy (18.4%, 9.21 per 100 person-years). Among all etiologies, mortality rates were highest for NALC patients with HBV infection (43.7%, 21.8 per 100 person-years), followed by HBV-HCV coinfection (41.8%, 20.9 per 100 person-years), HCV infection (41.2%, 20.6 per 100 person-years), and NAFLD (35.9%, 17.9 per 100 person-years). In this study, we demonstrated that mortality risks in NALC patients may differ with their etiology and their subsequent complications. Patients’ care plans, thus, should be formulated accordingly.


Introduction
Cirrhosis of the liver is a highly prevalent disease in Asian-Pacific countries, accounting for an estimated of 328,000 deaths per year [1]. The main causes of this condition in western countries are primarily alcohol intake and non-alcoholic fatty liver disease (NAFLD), whereas viral hepatitis B (HBV) infection is the most common cause in most parts of Asia such as Taiwan [2]. Taiwan still has a relatively high proportion of HBV-and HCV-infected population despite its decline since the implementation of the national viral hepatitis therapy programs [2,3]. Additionally, more health-damaging lifestyles such as high sugar intake and physical inactivity are observed, leading to metabolic diseases and

Data Source and Study Design
We used a population-based administrative claims database in Taiwan, the National Health Insurance Research Database (NHIRD), to identify all inpatients aged between 30 and 80 years of age with a primary diagnosis of NALC (International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) codes 571.5 and 571.6) within a period of 12 years from 1 January 1999 to 31 December 2010. NHIRD is an administrative claims-based database for the country's single-payer national health insurance. It holds a comprehensive set of claims data for >99% of the population and their uptake of services including outpatient, hospitalization, and emergency care, as well as any associated prescriptions and laboratory tests. Patients with the following characteristics were excluded from the study: (a) pre-diagnosed with cancer before index date, or (b) history of alcoholicrelated conditions like alcoholic fatty liver (571.0), alcoholic hepatitis (571.1), alcoholic liver cirrhosis (571.2), and alcoholic liver damage (571.3). Our process for subject selection has been illustrated in Figure 1.
We designed a retrospective cohort study in which the NALC subjects were matched 1:1 to the general inpatient population without any diagnosis of NALC (control cohort). The two cohorts were matched based on age, sex, Charlson Comorbidity Index (CCI), etiology, smoking status, and index date (i.e., incident date of hospital admission). Etiologies considered in this context were chronic viral hepatitis B (HBV) and C (HCV), HBV-HCV coinfection (HBV+HCV), and non-alcoholic fatty liver disease (NAFLD). NAFLD was defined if subject had a diagnostic code of ICD-9-CM code 571.8 on their medical record before index date. Since lifestyle factors such as smoking cannot be determined in a claims database but can be a confounding factor for mortality, we used the status of chronic obstructive pulmonary disease (COPD) (ICD-9-CM: 490, 491, 492, 494, 496) as a proxy for long term smoking. All study subjects were observed until 31 December 2011, event of death, or withdrawal from the National Health Insurance, whichever occurred first. We designed a retrospective cohort study in which the NALC subjects were matche 1:1 to the general inpatient population without any diagnosis of NALC (control cohor The two cohorts were matched based on age, sex, Charlson Comorbidity Index (CCI), e ology, smoking status, and index date (i.e., incident date of hospital admission). Etiologi considered in this context were chronic viral hepatitis B (HBV) and C (HCV), HBV-HC coinfection (HBV+HCV), and non-alcoholic fatty liver disease (NAFLD). NAFLD was d fined if subject had a diagnostic code of ICD-9-CM code 571.8 on their medical reco before index date. Since lifestyle factors such as smoking cannot be determined in a claim database but can be a confounding factor for mortality, we used the status of chronic o structive pulmonary disease (COPD) (ICD-9-CM: 490, 491, 492, 494, 496) as a proxy f long term smoking. All study subjects were observed until 31 December 2011, event death, or withdrawal from the National Health Insurance, whichever occurred first.
As admitted patients have higher probability of multiple and more severe conditions, comorbidities were also considered and defined in the form of CCI score and chronic smoking (i.e., COPD). All considered etiologies and comorbidities were defined in the subjects' medical history one year prior to their index dates.

Statistical Aanalysis
For analysis, continuous data are expressed using means and standard deviations, whereas categorical data are presented as counts and percentages. Crude mortality rates are expressed in rates (per 100 person-years, PYs) with 95% confidence intervals (CIs). Chi-squared test was used to test the difference in the proportion of deaths between two cohorts. Kaplan-Meier curve was used to show the cumulative survival probability of the study cohorts over time while log-rank test was performed to test the significance of difference between the two survival curves. Finally, Cox proportional hazards model was used to estimate hazard ratios (HRs) for mortality after adjusting for confounding factors. A two-sided p-value of <0.05 was considered statistically significant.

Subgroup Analysis
To test if risks of death differed with etiology and complication, subgroup analysis was performed for both the excess mortality risks and adjusted hazard ratios for mortality. HBV and HCV were considered as etiologies since they are the most prevalent hepatitis infections in Taiwan. NAFLD was chosen because it has become increasingly prevalent as a result of changing lifestyles and diet leading up to metabolic syndromes such as obesity and diabetes mellitus.

Ethical Considerations
This study was approved by the research ethics committee at China Medical University and Hospital (CMUH107-REC2-105). Informed consent was waived due to the retrospective nature of this study and our use of secondary data for analyses.

Baseline Characteristics of the Study Cohorts
Our final cohorts comprised 109,128 NALC subjects and 109,128 matched control patients without NALC. Their total follow-up was 355,486 PYs and 583,423 PYs, respectively. Baseline attributes of the two cohorts were similar post-matching: male (68.3%), mean age of 59.4 years, etiology (HBV-only (13.6%), HCV-only (10.9%), HBV-HCV coinfection (1.31%), and NAFLD (3.80%)) and CCI scores (1: 56.1%; ≥2: 43.9%). However, as expected, the prevalence of decompensation was much higher in NALC patients. General characteristics of the study cohorts are presented in Table 1.  Table 2 illustrates the proportions and crude mortality rates for the cohorts with and without NALC. In general, 42.4% (46,297 of 109,128) of inpatients with cirrhosis died compared with 12.5% (13,684 of 109,128) in their counterparts during the follow-up period. Overall mortality rates were 21.2 (95% CI: 21.0-21.4) and 6.27 (95% CI: 6.17-6.37) per 100 PYs, respectively, and the difference was statistically significant (p < 0.0001). Among complications that caused death in NALC patients, variceal hemorrhage was the most common (23.7%, 11.9 per 100 PYs), followed by ascites (20.9%, 10.4 per 100 PYs) and encephalopathy (18.4%, 9.21 per 100 PYs). All other considered causes of death accounted for mortality of less than 10% each in NALC patients: SBP 8.8%, HCC 8.2%, jaundice 2.7%, and HRS 2.2%. Figure 2 provides a graphic representation of the comparison of mortality rates between the two cohorts.  Table 2 illustrates the proportions and crude mortality rates for the cohorts with and without NALC. In general, 42.4% (46,297 of 109,128) of inpatients with cirrhosis died compared with 12.5% (13,684 of 109,128) in their counterparts during the follow-up period. Overall mortality rates were 21.2 (95% CI: 21.0-21.4) and 6.27 (95% CI: 6.17-6.37) per 100 PYs, respectively, and the difference was statistically significant (p < 0.0001). Among complications that caused death in NALC patients, variceal hemorrhage was the most common (23.7%, 11.9 per 100 PYs), followed by ascites (20.9%, 10.4 per 100 PYs) and encephalopathy (18.4%, 9.21 per 100 PYs). All other considered causes of death accounted for mortality of less than 10% each in NALC patients: SBP 8.8%, HCC 8.2%, jaundice 2.7%, and HRS 2.2%. Figure 2 provides a graphic representation of the comparison of mortality rates between the two cohorts.  For the control cohort, the number of deaths from decompensation was negligible, with only variceal hemorrhage with a slightly higher mortality rate of 1.50 per 100 PYs (3.0%). Kaplan-Meier curves for all-cause mortality also highlight the significantly lower cumulative survival probability in the NALC cohort relative to the control (log-rank test, p < 0.0001). The sharp decrease within the first year of follow-up was especially noticeable (Figure 3). For the control cohort, the number of deaths from decompensation was negligible, with only variceal hemorrhage with a slightly higher mortality rate of 1.50 per 100 PYs (3.0%). Kaplan-Meier curves for all-cause mortality also highlight the significantly lower cumulative survival probability in the NALC cohort relative to the control (log-rank test, p < 0.0001). The sharp decrease within the first year of follow-up was especially noticeable (Figure 3).   Table 3 presents the excess risk of death further stratified by etiology among NALC patients. Overall mortality rates were highest for NALC patients with HBV-only etiology (43.7%, 21.8 per 100 PYs), followed by HBV-HCV coinfection (41.8%, 20.9 per 100 PYs), HCV-only (41.2%, 20.6 per 100 PYs), and lastly, NAFLD (35.9%, 17.9 per 100 PYs). Among the complications, variceal hemorrhage, ascites, and encephalopathy were the top three causes of death across all etiologies. We also noticed that although mortality rates were generally lowest in NAFLD among the four etiologies, they were still considerably higher than that of the control inpatient cohort.

Adjusted Mortality Risk in Patient Subgroups
To examine under which attributes do NALC patients have significantly elevated mortality risk than non-NALC patients, we stratified our subjects into subgroups of etiology, sex, and age (Table 4).
We first compared their all-cause mortality risk after adjusting for etiology, age, sex, CCI score, smoking status, and index date. We found that NALC patients of HBV-only etiology had the highest adjusted hazard ratios (aHRs) versus the control patients within the etiology subgroup (aHR: 8.38 (95% CI: 7.87-8.93)), and <40 years of age had the highest aHR among all age groups (aHR: 13.3 (95% CI: 11.9-15.0)). This suggests that among hospitalized patients with HBV infection, those with NALC are 8.38 times more likely to die from any cause compared with those without NALC; among patients <40 years of age, those with NALC are 13.3 times more likely to die from any cause compared with those without NALC. It is also important to note that, among inpatients with HBV-HCV coinfection, the relative risk was also significantly high (aHR: 6.69; 95% CI: 5.53-8.10).
We subsequently presented the adjusted risks of cause-specific mortality (variceal hemorrhage, ascites, encephalopathy, SBP, HCC, jaundice, HRS). Results indicate that the risk of death from variceal hemorrhage in NALC patients with HBV was 16.1 times higher than non-NALC patients with HBV (95% CI: 14.3-18.0). Similarly, aHR for death by ascites was 83.3 times higher in NALC patients with NAFLD compared with non-NALC patients with NAFLD (95% CI: 45.9-151.0). When examining encephalopathy as the cause of death, females with NALC had an aHR of 164.5 (95% CI: 127.3-212.6) relative to females without NALC. The increased complication-specific mortality risks for NALC patients compared with non-NALC individuals were also more profound in the younger age groups of <40, 40-49, and 50-59. This confirmed that NALC ubiquitously elevated patient's mortality risk despite the cause and that the mortality risk can vary with patients' etiology, age and sex.

Main Findings
In this population-based retrospective cohort study, we demonstrated that variceal bleeding, ascites, and encephalopathy can potentially be the three leading causes of mortality among NALC patients. Our study also confirmed that the mortality risk varies with patient's etiology, age, and sex, as well as the complication developed in these patients when advancing to the decompensation state. Specifically, patients of HBV etiology had the highest excess risk of mortality from decompensation amongst all etiology subgroups (HBV, HCV, HBV + HCV, and NAFLD). This may be partly due to the insidious nature of clinical course and disease progression in HBV patients which is different from other etiologies, subsequently elevating their risk of death. Also, it is believed that the prognosis for most chronically HCV-infected patients tend to be relatively nonthreatening over a period of time despite their clinical progression [11]. Elevated complication-specific mortality risks for NALC patients relative to non-NALC individuals were also more profound in <40 years, 40-49 years and 50-59 years age groups.

Comparison with Other Studies
A similar investigation was conducted in England looking into deaths by etiology in liver cirrhosis patients relative to the general population [12]. In a total of 5118 cirrhosis patients, crude (all-cause) mortality rate for viral hepatitis-related cirrhosis was 14.1 per 100 PYs which is less than what was observed in our population (21.2 per 100 PYs). This may be attributable to the hospitalized status of our study cohorts. The increased mortality risk in cirrhosis patients relative to non-cirrhosis patients was also consistent with our findings but was more considerable in the English study (6-fold vs. 3-fold). Nevertheless, it is important to note that results from our study should be interpreted differently since we considered the complications as our possible causes of death while removing alcoholic-related factors.
Tapper and Parikh (2018) found in their US cohort study that age-adjusted cirrhosis mortality was 12.18/100,000 population; this rate has been escalating and the greatest rate of change was attributed to alcohol use in young people [13]. Although their outcome of interest is similar to that of this study, they examined distribution of mortality risk by race and geographical region which cannot be directly compared with our results. Also, unlike this study, they included alcohol-related cirrhosis. In another US study examining the trends in mortality of cirrhosis patients, mortality rate in viral hepatitis-related cirrhosis was 9.0 per 100 PYs which is much lower than what we estimated due to the reason noted above. However, similar to our observation, they detected a higher mortality risk from ascites (12.4/100 PYs), HE (5.3/100 PYs), and variceal bleeding (2.9/100 PYs) among the cirrhosis-related complications [14].
For NALC patients, we detected a very high probability of death within the first year following their hospital admission. This agrees with previous evidence from England which suggested that mortality could be as high as 8 times more than that in the general population at one year [15]. Another Danish study also showed that up to 38% of cirrhosis patients die within the first year of follow-up [16].
We also found that NAFLD and HBV-HCV co-infected patients had very high risks of death from complications even after accounting for confounding factors. While there is lack of evidence on the odds of death in NAFLD cirrhosis patients, much evidence did indicate an increased risk of disease progression into cirrhosis, decompensation, and HCC in HBV-HCV co-infected cases [17][18][19]. We included NAFLD in this study and demonstrated its plausible impact on the mortality risk in NALC patients. This is crucial for Asian countries since it has become an emerging epidemic imposed by dietary and lifestyle changes with serious health outcomes [20].
Among complications that led to mortality, variceal hemorrhage was most commonly observed, followed by ascites and encephalopathy. A cohort study conducted in Denmark confirmed that 56% of liver cirrhosis hospitalized patients die in the first year from associated conditions such as esophageal varices and hepatic coma [16]. Similar to our study, they also noticed increased mortality in patients of chronic hepatitis etiology. A descriptive study in India demonstrated that the most common cause of death among cirrhosis patients was HE (50%) followed by gastrointestinal bleeding (27%), and that HBV is a prevalent etiological factor in this Asian population [21]. Moreover, results from a Korean study have confirmed that mortality rate of decompensated cirrhosis patients is even significantly greater than patients with cancers [22].
Many of the complications of cirrhosis are the consequence of portal hypertension. Formation of varices may be ensued, as well as vascular and other abnormalities that contribute to the pathogenesis of ascites and other complications like encephalopathy [23]. All these are prognostic factors that can predispose to premature mortality in patients with cirrhosis.
For instance, variceal bleeding is a frequent condition in decompensated cirrhosis with a higher risk of mortality than the other causes of gastrointestinal bleeding [24]. Indeed, we showed that variceal bleeding is the primary cause of death among all cirrhosis-related complications, which agrees with the clinical staging of cirrhosis and its corresponding prognosis [25]. High mortality attributed to this complication is mainly due to the high fatality of its initial episodes, as well as the high probability for recurrent bleeding [24][25][26]. High mortality in subjects with ascites and hepatic encephalopathy is also supported by other studies [12,27,28].

Strengths and Limitations
For our study, we specifically targeted NALC patients, unlike many previous studies which dealt primarily with alcoholic cirrhosis. To the best of our knowledge, this is the first population-based cohort study that has been conducted to investigate cause-specific deaths by etiology in NALC patients. Furthermore, past studies were either conducted in hospital-based settings which could lead to overestimation of mortality, or they did not explore the complications as specific causes of deaths and their risks; they mostly examined neoplasm-, metabolic-, or circulatory-related factors as causes of death [12,15,16,21,29]. Moreover, NAFLD was considered as an etiology in our study which was rarely observed in earlier studies. This provides novelty and an update to our current knowledge about NALC epidemiology in adherence with modern lifestyle and social transition. We also considered the impact of ill-status baseline of hospitalized patients on their prognosis; this is why we performed rigorous matching. One of the criteria was patients' CCI that took into account the effect of other comorbidities such as cardiovascular diseases and metastatic solid tumor.
Nevertheless, this study is not without limitations. Clear separation of liver disease by etiology is unfeasible even in developed economies as suggested by a previous study [1]. For our analysis, the occurrence of decompensation is not mutually exclusive from one another for each patient, i.e., we identified each condition when its corresponding code is found in its admission record, irrespective of its order (primary, secondary, or tertiary diagnosis). Hence, one may be counted more than once when we stratified the analysis into subgroups by type of decompensation. Also, for the purpose of our study, we did not examine closely the heterogeneity within the HBV-HCV co-infected population which may be affecting the viral genotypes and/or activity status of the viruses. Lastly, there are several caveats using claims-based administrative data: First, laboratory details such as bilirubin levels, prothrombin time or viral load were not available in the administrative data sources to verify patients' diagnosis and classify the severity of disease (i.e., Child-Pugh-Turcotte score); secondly, NAFLD stage, which could influence cirrhosis prognosis, could not be defined without the presence of biopsy or histological data; and thirdly, our estimates of cirrhotic mortality should be interpreted with caution since they are based on hospitalized cases (i.e., overestimation). Many cases of liver cirrhosis in the early stages are asymptomatic and often diagnosed. COPD as a surrogate for smoking status could potentially misclassify or underestimate the prevalence of smoking in the subject sample since only those that were chronic smokers and developed COPD would be identified.

Implications on Disease Management
Our results have important implications to disease management for NALC patients. Our evidence on the varying mortality risk among hospitalized NALC patients should caution clinicians to be vigilant while caring inpatients of various etiologies (e.g., HBV, NAFLD) and age groups. Moreover, different clinical measures should be taken to prevent different complications, most likely in the order of their severity and risk of death. We hope that the evidence presented in this study can help further improve quality of care by guiding clinicians in their practice under the real-world setting.

Conclusions
In conclusion, results from this population-based retrospective cohort study demonstrated the significant elevated risk of mortality among NALC patients compared with general inpatients. Specifically, variceal bleeding, ascites, and encephalopathy can potentially be the three leading causes of mortality among these patients. To ensure quality of care and effective management of disease, patient etiologies and potential complications must also be taken into consideration when formulating patients' care plans.