Higher mortality and hospital charges in patients with cirrhosis and acute respiratory illness: a population-based study

Both cirrhosis and acute respiratory illness (ARI) carry substantial disease and financial burden. To compare hospitalized patients with cirrhosis with ARI to cirrhotic patients without ARI, a retrospective cohort study was conducted using the California Office of Statewide Health Planning and Development database. To balance the groups, propensity score matching (PSM) was used. We identified a total of 46,192 cirrhotic patients during the three study periods (14,049, 15,699, and 16,444 patients, respectively). Among patients hospitalized with cirrhosis, the ARI prevalence was higher in older age groups (p < 0.001), the Asian population (p = 0.002), non-Hispanic population (p = 0.001), and among Medicare patients (p < 0.001). Compared to controls, patients with ARI had 53.8% higher adjusted hospital charge ($122,555 vs. $79,685 per patient per admission, p < 0.001) and 35.0% higher adjusted in-hospital mortality (p < 0.001). Older patients, patients with alcoholic liver disease or liver cancer were at particularly higher risk (adjusted hazard ratio = 2.94 (95% CI: 2.26–3.83), 1.22 (95% CI: 1.02–1.45), and 2.17 (95% CI: 1.76–2.68) respectively, p = 0.028 to <0.001). Mortality rates and hospital charges in hospitalized cirrhotic patients with ARI were higher than in cirrhotic controls without ARI. Preventive efforts such as influenza and pneumococcal vaccination, especially in older patients and those with liver cancer, or alcoholic liver disease, would be of value.

After propensity score matching, the study population included 1240 and 3004, 1311 and 3240, 1409 and 3382 of ARI and non-ARI patients, respectively in the 2011-12 (Table 1) The x axis represents percentage of patients with acute respiratory illness in different subgroups of cirrhotic patients. The y axis shows subgroups of age, sex, race, ethnicity, insurance, severity of cirrhosis, and cause of cirrhosis. The p-values on the right side shows whether the difference of the prevalence within subgroups is significant. *Worker's compensation, county indigent program, other government, other indigent program, other payer. **HBV patients (with or without HCV, ALD, NAFLD). ***HCV patients (with or without ALD, NAFLD). ****Autoimmune hepatitis, alpha-1-antitrypsin deficiency, hereditary hemochromatosis, hemochromatosis due to repeated red blood cell transfusions, other hemochromatosis, disorders of copper metabolism, cholangitis.  Fig. 1).
In the subgroup of patients who required mechanical ventilation, we found that ARI patients requiring mechanical ventilation had more than double (2.22-fold) the seasonal hospital charge per patient than those not requiring ventilation, with total charges of $388,578 vs. $175,370 per patient per season (p < 0.001) ( Supplementary Fig. 2).
Mortality. Mortality Rates. After PSM, a total of 621 cirrhotic patients who were diagnosed with ARI during the 2011-12 season died before December 21, 2013 while the number of deaths among those without ARI was 1,327. The overall mortality rate among ARI patients was 47.4% while that among non-ARI patients was 41.0%. Specifically, the in-hospital, 30-day and one-year post-discharge morality rates among ARI patients were significantly higher than that among patients without ARI (7.6% vs. 5.8%, p < 0.001; 14.6% vs. 11.5%, p = 0.023; and 35.9% vs. 31.3%, p = 0.001, respectively). As illustrated from the Kaplan-Meier analysis, it was clear that within a follow-up period of one year from the date of hospital discharge, the mortality rate among ARI patients was consistently higher than that among patients without ARI (log rank test, p = 0.023) (Fig. 4).
Cause of Death. The most common cause of death among both cirrhotic patients with ARI and non-ARI control groups were liver-related (18.2% vs. 18.5%, p = 0.82) (Supplementary Table 3). The second most common cause of death also for both of these groups was related to cardiovascular disease, with a higher rate in the ARI group (5.1% vs. 3.6%, p = 0.023). As expected, the number of deaths caused by respiratory diseases in ARI patients was significantly higher than that among non-ARI patients (5.0% vs. 2.3%, p < 0.001). The results were similar for the 2010-11 season.
Factors related to higher hospital charges. We adjusted for hospital days and other confounding factors while studying factors related to seasonal hospital charges for cirrhotic patients with ARI during the 2011-12 season (Supplementary Table 5). In cirrhotic patients with ARI, longer duration of hospitalization was significantly associated with increased seasonal hospital charges, and patients hospitalized for 3 to 15 days accumulated total hospital charges that on average were $66,194 higher than those among patients hospitalized for less than 2 days (p < 0.001). Cirrhotic patients with ARI and history of multi-organ failure accounted for an average of $30,132 in additional expenditures compared with ARI patients without history of multi-organ failure (p = 0.039).

Discussion
We investigated the health burden of cirrhotic patients with and without ARI utilizing a large population-based sample of patients in the U.S, in which almost the entire inpatient population from the state of California was included. In addition to being the most populous state in the nation, with a total population of approximately 39 million, California's population is also among the most ethnically and socioeconomically diverse. To our knowledge, this is the first study to investigate the financial and mortality burden associated with ARI in patients with cirrhosis. We found that the prevalence of ARI in hospitalized patients with cirrhosis ranged from 8% to 9% over the three consecutive influenza seasons investigated. The prevalence rate is significantly higher (almost 40-fold) than the prevalence rate of 0.22% for patients hospitalized with influenza as reported by CDC 15 . government, other indigent program, other payer. b HBV patients (with or without HCV, ALD, NAFLD). c HCV patients (with or without ALD, NAFLD). d Autoimmune hepatitis, alpha-1-antitrypsin deficiency, hereditary hemochromatosis, hemochromatosis due to repeated red blood cell transfusions, other hemochromatosis, disorders of copper metabolism, cholangitis.
The prevalence of ARI was also found to be significantly higher among older patients (>65), Asian and non-Hispanic patients, and those with Medicare or Medicaid insurance coverage. These findings are not surprising as acute respiratory illnesses strike the elderly in disproportionate numbers as well as those with chronic illnesses as often overly represented in these populations 16 .
We also found that the presence of ARI in patients with cirrhosis placed a tremendous burden on the health care system. Though, the total number of hospitalizations among cirrhotic patients with ARI was lower than that among cirrhotic patients without ARI, ARI patients had, on average, a significantly longer length of hospital stay, higher hospital charges per patient per admission, and higher hospital charges per patient per season than non-ARI cirrhotic patients. According to a 2011 study on cirrhotic patients requiring intensive care, the mean  Cirrhotic patients with ARI also experienced consistently higher mortality rates than cirrhotic inpatients without ARI. Compared with cirrhotic inpatients without ARI, cirrhotic patients with ARI had 35%, 13%, and 19% higher risk of in-hospital, 30-day and 1-year mortality, respectively. Furthermore, older age, HCC, and having ALD-related cirrhosis were each associated with higher risk of mortality.
Our results were also notable for a substantial risk of mortality among cirrhotic patients with ARI compared to the general population. In the general population, analysis of the NIS data previously reported a declining in-hospital mortality rate of 4.1% in 2005 from 8.9% in 1993 for community-acquired pneumonia 18 . In another study, in-hospital mortality of acute respiratory disease was also well below 10% at 5.0%, 4.3%, and 4.4% during the 2010-11, 2011-12, and 2012-13 influenza seasons respectively 19 . In contrast, the in-hospital mortality rate for cirrhotic patients with ARI in the current study was 7.6% during the 2011-2012 season, which was also higher than that among patients with cirrhosis but without ARI (5.8%). Therefore, it is important to prevent ARI among cirrhotic patients. CDC reports on the effectiveness of the influenza vaccination for the years 2015-2016 estimated that the influenza vaccination prevented an estimated 5.1 million illnesses, 2.5 million medical visits, 71,000 hospitalizations, and 3,000 pneumonia and influenza deaths 15 . In addition, the CDC anticipated significant declines in invasive pneumococcal disease with the introduction of the pneumococcal conjugate vaccines in the United States (PCV7 in 2000 and PCV13 in 2010, 2012 for adults aged 19 years and older with immunocompromising disease and 2014 for adults >64 years of age), and estimates that invasive pneumococcal disease in adults 19 through 64 years old would decrease from 16 cases per 100,000 people in 1998 to only 7 cases per 100,000 people in 2015 20 . However, despite these impressive results, influenza vaccination coverage among adults aged ≥19 years was only 43.2% during the 2012-2013 influenza season while the pneumococcal vaccination coverage was only 20.3% among high-risk persons aged 19-64 years and 61.3% among adults aged ≥65 years during this same time 21 .

In-hospital/30-days/1-year number of death (n = 289/564/1486)
A significant strength of our study is that it is a population-based study utilizing a large and validated database and thus enhances its accuracy. We reported the updated prevalence, mortality and inpatient charge for cirrhotic patient overall and by specific etiologies to inform targeted public health measures for particularly high-risk groups such as older cirrhotic patients or those with alcoholic liver disease.
There are also several limitations to our study. This first is that the mortality rate from ARI might have been underestimated as a result of ARI diagnosis codes being miscoded or not recorded. Because the California Office of Statewide Health Planning and Development (OSHPD) database is an administrative database, the precision of diagnoses can also be decreased by additional coding errors. However, OSHPD has been a widely used and validated database and the accuracy of diagnoses is generally higher in inpatient than outpatient settings 22,23 . In addition, as the OSHPD database only contains data from patients in California, it may not represent the entire nation. However, California as the largest and most populous state in the U.S. and is worth studying in itself and with the most diverse population in the U.S., it also holds lessons that may be generalizable to multiple other regions in the country. Furthermore, California has a more temperate climate than many other states, which may have underestimated the overall impact of ARI and may have biased the outcomes towards the null hypothesis, and the impact of ARI is likely to be even greater in regions with more severe winter weather.

Conclusions
Our findings suggest that ARI accounts for substantial morbidity, mortality and financial burden among hospitalized, cirrhotic patients in the U.S., especially among older, Medicare, Medicaid, Asian, and non-Hispanic patients, as well as those with history of HCC, ALD, or severe comorbidities. Cirrhotic patients, especially the higher risk groups, should be targeted for preventive measures such as influenza and pneumonia vaccination as well as early and effective outpatient management of those who have contracted ARI in order to prevent costly and high-risk hospital admissions. Definitions. Cirrhosis was determined by ICD 9 codes and clinically defined as the presence of clinical, radiologic, endoscopic, manifestations of cirrhosis and/or portal hypertension (thrombocytopenia, splenomegaly, presence of varices) or symptoms of clinical hepatic decompensation (ascites, hepatic encephalopathy, jaundice, variceal hemorrhage).
Severity of liver disease was categorized into compensated cirrhosis, decompensated cirrhosis, and HCC. Cause of cirrhosis covered the liver diseases of HBV, hepatitis C virus (HCV), ALD, non-alcoholic fatty liver disease, and other liver diseases. Sample Selection. All adult patients (≥18 years) diagnosed with cirrhosis (ICD-9 codes: Supplementary We excluded patients with missing RLNs or co-infection with the human immunodeficiency virus. Patients without known cause of cirrhosis and who could not be categorized into one of the listed categories were excluded. We also excluded all patients who received a liver transplant at any time during our study period prior to any analysis. Propensity Score Matching. We performed a 1:3 propensity score matching of cirrhotic patients with ARI vs. cirrhotic patients without ARI in each of the three influenza seasons to adjust for differences in background risks (age, sex, race, ethnicity, insurance type, hospital site, severity of cirrhosis, cause of cirrhosis, comorbidities, and history of severe comorbidities) between ARI cirrhotic patients and non-ARI cirrhotic controls. Nearest neighbour matching using the caliper of 0.009 was applied. To get the optimal result, 0.2 of standard deviation of the logit of propensity score was used to calculate the caliper as previously recommended 28 . We also examined the balance of measured covariates between the matched cohorts using standardized differences 29 . Outcomes. In patients with cirrhosis, we report the prevalence of ARI, number of hospitalizations, length of hospital stay, hospital charges per patient day, hospital charges per patient admission, hospital charges per patient season, and in-hospital as well as post-discharge mortality. Mortality analysis was not performed for the most recent influenza season (2012-2013) due to relative lack of follow-up data for this cohort. All charges were inflation (3%) adjusted to 2013 US dollars. Causes of death were identified by ICD 10 codes in OSHPD and we categorized the causes as related to liver disease (alcoholic cirrhosis of liver, HCV, other cirrhosis of liver, liver cancer, non-alcoholic liver disease), cardiovascular disease (coronary artery disease, heart failure, ST elevation myocardial infarction, stroke, etc.), respiratory disease (chronic obstructive pulmonary disease, pneumonia), other malignant neoplasms, diabetes mellitus and kidney disease.
Other comorbidities examined were obtained using ICD-9 codes and included cardiovascular disease, diabetes mellitus, hypertension, any cancer other than HCC, alcohol use/abuse, drug abuse, hyperlipidemia, chronic obstructive pulmonary disease, chronic kidney disease, mental illness, renal failure, cardiac failure, major neurologic events, severe hematologic conditions, multi-organ failure, sepsis, hepatic failure, and respiratory failure 16 .
The severity of liver disease, cause of cirrhosis, and all comorbidities were obtained prior to the beginning of each influenza season (ICD-9 codes: Supplementary Table 7).

Statistical analysis.
Chi-square test was used to perform the descriptive analysis on the prevalence of ARI since all demographic and clinical features were dichotomous. The multivariate generalized linear model (GLM) with a gamma distribution and log-link function was conducted to compare hospital charges. GLM with a Poisson distribution and log-link function was used to compare number of hospitalizations and length of hospital stay. We adjusted for age, sex, race, ethnicity, insurance type, severity of cirrhosis, cause of cirrhosis, comorbidities, and history of severe comorbidities. We also performed subgroup analysis to compare the seasonal hospital charge in patients who required mechanical ventilation and those who did not.
Kaplan-Meier survival analysis was used to describe the cumulative mortality rates of cirrhotic patients with and without ARI. The censor date was death or the end of follow up, whichever came first. We also performed a log rank test to examine the significance of the difference in mortality between ARI cases and non-ARI controls. The Cox proportional hazards regression model was used to estimate hazard ratios and its 95% confidence interval relating ARI with mortality for cirrhotic inpatients adjusted for age, sex, race, ethnicity, insurance type, severity of cirrhosis, cause of cirrhosis, comorbidities, and history of severe comorbidities.
All statistical analyses were 2-tailed, with a significance level of 0.05. We performed all analyses using STATA version 14 (Stata Corp., College Station, TX).