End stage liver disease (ESLD) is a global health burden and one of the leading causes of mortality around the world [¹],[²]. Majority of ESLD patients frequently get admitted to hospital due to related complications of hepatic encephalopathy, spontaneous bacterial peritonitis (SBP) and gastrointestinal bleeding [³]. ESLD is also associated with autonomic dysfunction and increase levels of circulating neuropeptides such as vasoactive intestinal peptide (VIP) and galactin-3 [⁴],[⁵],[⁶]. These physiological perturbations are proposed in pathogenesis of atrial fibrillation (AF) in ESLD patients [⁷]. Studies have shown that ESLD is a predictor for new onset AF with advanced ESLD as manifested by worsening Model for End-Stage Liver Disease (MELD) score associated with further increased risk [⁸]. Till to date, there is no data on how AF affects inpatient outcomes of ESLD patients who are at greatest risk for frequent hospitalizations. In this paper, we aim to study these parameters from a national United States population database.

Data were collected from National Inpatient Sample (NIS). NIS is part of Healthcare Cost and Utilization Project (HCUP) databases and is made possible by a Federal-State-Industry partnership sponsored by the Agency for Healthcare Research and Quality (AHRQ). The NIS is derived from all States for national estimates of healthcare utilization, cost and outcomes [⁹]. Since NIS is compiled annually, the data can be used for analysis of disease trends over time. The study was deemed exempted from Institutional Review Board approval given the de-identified nature of the NIS database and public availability.

We analyzed NIS database from January 2005 to August 2015 using the International Classification of Diseases, 9th Revision, Clinical Modification (ICD-9-CM) codes. Patients ≥ 18 years of age were included. Patients with ESLD were identified using Goldberg’s third algorithm [¹⁰], a well-validated method for identifying ESLD from administrative datasets, generating a positive predictive value of 89.3%. Based on this algorithm, first ICD-9-CM codes were used to select chronic liver disease patients. (ICD-9-CM of 070.20-21, 070.23, 070.30-33, 070.40, 070.42, 070.49, 070.52, 070.59-60, 070.70-71, 070.90, 571.1, 571.40-41, 571.8, and 571.9), then a concurrent diagnosis code of cirrhosis was added (ICD-9-CM of 571.2, 571.5 or 571.6) and finally at least 1 concurrent diagnostic code for hepatic decompensation event (ICD-9-CM of 456.0-2, 789.5, 789.59, 572.2, 567.2, 567.21, 567.29, 567.8-9 or 572.4). Using this algorithm, we were able to extract a total of 309,959 ESLD patients that were included in final analysis. Please see [Figure 1] for detailed methodology of patient inclusion criteria.

Figure 1. Flow sheet of our paper

---

---

Baseline characteristics and hospital outcomes were derived and compared among ESLD patients with and without AF. To account for potential confounding factors and selection bias, a propensity score-matching model was developed using logistic regression to derive two matched groups for comparative outcomes analysis. A nearest neighbor 1:1 variable ratio, parallel, balanced propensity-matching model was made using a caliper width of 0.2. Descriptive statistics were presented as frequencies with percentages for categorical variables and as means with standard deviations for continuous variables. Baseline characteristics were compared using a Pearson𝜒2 test and Fisher's exact test for categorical variables and independent samples t-test for continuous variables.

Logistic regression was performed to estimate odds ratios (ORs) with 95% confidence intervals (CIs) to determine predictors of mortality in ESLD. Initially, binomial logistic regression model was used to identify variables from demographic data [Table 1] that were significantly associated with patient mortality (P value ˂ 0.10). These variables were then subsequently utilized in a multiple logistic regression model to identify predictors of mortality. A type I error rate of <0.05 was considered statistically significant. All statistical analyses were performed using statistical package for social science (SPSS) version 26 (IBM Corp) and R 3.5 for propensity matching

A total of 309,959 patients with ESLD were identified from the NIS dataset. Out of these 32,858 patients had AF (10.6%). Baseline characteristics of the study population are shown in [Table 1]. ESLD patients with AF were older (68.46 vs. 58.14 years, p < 0.01) and had higher burden of key co-morbidities such as diabetes (29.9% vs. 25.3%, p < 0.01), hypertension (53.8% vs. 40.1%, p < 0.01) and congestive heart failure (31.2% vs. 9.3%, p < 0.01). Overall, about 29,487 (9.5%) ESLD patients died at discharge (see [Table 2]). Mortality was 9.2% in ESLD without AF when compared to 12.3% in ESLD with AF (p < 0.01). In a propensity-matched cohort, this mortality difference continues to remain significant (12.2% vs. 10.8%, p < 0.01, supplemental []). A gradual downtrend trend in mortality was noted in ESLD patients with and with out AF over our study years (see [Figure 2]). This downward trend is same across both genders although male patients had higher mortality when compared to female patients over our study period (see [Figure 3]).

Figure 2. Mortality in end stage liver disease with and with out atrial fibrillation over study years

---

---

Figure 3. Mortality in end stage liver disease by gender and atrial fibrillation over study years

---

---

ESLD patients with AF have longer length of stay (9 vs. 7.36 days, p < 0.01) and increase costs of hospitalization (78,246 $ vs. 63,403 $, p < 0.01) when compared to ESLD patients with out AF. Please see [Figure 4] for length of stay and costs of hospitalization trends over our study years. Predictors of mortality in ESLD are shown in [Figure 5]. Advanced age, AF and African American race were independently associated with increased mortality. Urban and large hospitals were associated with lower mortality. Patients with metabolic acidosis, coagulopathy, pulmonary circulation disorders, congestive heart failure and cancers were also associated with increased mortality in our study cohort.

Figure 4. Mean cost of hospitalization and length of stay over our study period

---

---

Figure 5. Predictors of mortality in end stage liver disease patients

---

---

The main findings of our current investigation are: (1) ESLD patients with concomitant AF have increased mortality when compared to ESLD patients with out concomitant AF (12.30% vs. 9.20%, p < 0.01) and this difference persisted despite balancing co-variates in a propensity matched model (12.2% vs. 10.8%, p < 0.01). (2) The presence of AF is an independent predictor of mortality in ESLD patients and about 10.6% patients in our cohort have AF. (3) ESLD and AF patients have increased cost of hospitalization as well as length of stay when compared to ESLD patients with out AF.

ESLD is a rising global health burden and represents a final sequel in natural history of liver cirrhosis [¹],[²]. Frequent hospitalizations are common in ESLD patients due to concomitant complications of hepatic encephalopathy, SBP and gastrointestinal bleeding [³]. ESLD patients are more prone to developing AF even in absence of structural heart disease. This increased propensity of developing AF is proposed to be due to autonomic dysfunction and increased levels of circulating neuropeptides such as VIP and galactin-3 that exercise their effect either through modulating autonomic system or inducing fibrosis with in the heart muscle [⁴],[⁵],[⁶],[⁷]. The prevalence of AF in our cohort is about 10.6%, which is consistent with earlier studies. The study by Huang et al. [⁸] on 1727 consecutive ESLD patients awaiting liver transplantation showed AF prevalence to be about 11.2%. They also found that liver disease is an independent predictor of new onset AF and increased MELD scores are subsequently associated with worsening risk of new AF development. Similarly, another study by Lee et al. [¹¹] has found 46% relative risk of developing AF in patients with liver cirrhosis. In comparison to our study, they did not find AF to be significantly associated with all-cause mortality. It is pertinent to mention here that Lee et al. primarily enrolled patients with various stages of cirrhosis while our study exclusively focused on ESLD cohort which is comparatively more sicker and morbid and that may explain difference in mortality between both studies. In another study on ESLD patients undergoing liver transplantation [¹²], the occurrence of peri-procedural AF was associated with worsened mortality (HR 5.097, 95% CI 2.189-11.86). In our national cohort of ESLD patients, we have demonstrated that AF is associated with worse in-patient survival and that difference persists despite accounting for confounding variables. We also demonstrated that AF is an independent predictor of mortality in ESLD patients.

The strong association of worse outcomes of ESLD patients with AF poses unique management challenges. Stroke is a leading cause of mortality and disability in AF patients and anti-coagulation is often recommended to mitigate those risks [¹³],[¹⁴]. The utilization of anti-coagulation can be especially challenging in ESLD patients due to increased bleeding risk associated with platelet dysfunction and esophageal varices [¹⁵]. Additionally, there are studies showing increased propensity of hemorrhagic stroke in ESLD patients and that risk in some cases exceeds those of ischemic stroke [¹⁶],[¹⁷]. Our dataset, unfortunately, is not designed to ascertain causes of mortality but whether embolic or bleeding events contributed to poor outcomes needs further studies. Additionally, the association of worse mortality in ESLD patients with AF also calls into question measures to screen for AF in this patient population. Timely detection of AF and subsequent implementation of relevant therapeutic measures could result in improved outcomes in ESLD patients.

NIS is an administrative claim-based database that uses ICD-9-CM codes for diagnosis that may be subjected to error. The hard clinical points such as liver cirrhosis and mortality are, however, less prone to error. There are no well-defined ICD-9-CM codes for ESLD and we have used Goldberg’s third algorithm for stratifying these patients as mentioned in methods section. This method yields a positive predictive value of about 89.3% for ESLD but it is still plausible that some patients with ESLD may not have been captured using this methodology. NIS collects data on in-patient discharges and each admission is registered as an independent event. It is possible that same patient may have more than one subsequent admission over time. NIS samples are not designed to follow patients longitudinally so long-term outcomes could not be assessed from the present dataset. Additionally, data on AF management is lacking from NIS which have important implications on conclusions drawn from the study.

Our study shows AF to be associated with worse outcomes in ESLD patients. It is therefore imperative that treating physicians should have a strong clinical suspicion for AF in this specific patient cohort as timely AF detection could result in improved outcomes.