Analysis of weekend effect in severe acute liver injury: A nationwide database study

Abstract Background and Aims Severe acute liver injury (ALI) can lead to poor outcomes without timely management. Comparatively worse outcomes in various severe, emergent conditions have been attributed to reduced hospital resources experienced by patient weekend admissions, a phenomenon termed “weekend effect.” To date, a weekend effect has not been studied in severe ALI, an emergency also necessitating timely management. We aimed to evaluate such an effect in this condition by analyzing a large national inpatient database in the United States. Methods We analyzed the Nationwide/National Inpatient Sample (NIS) 2000 to 2014, the largest inpatient, all‐payer database in the United States (US), containing sociodemographic, clinical, patient‐, and hospital‐level data. We identified severe ALI using International Classification of Disease, 9th Revision diagnosis codes for acute/subacute hepatic necrosis (570) with encephalopathy (572.2). Our primary outcome was in‐hospital mortality. Using a full‐model approach for covariate selection, we performed multiple logistic regression modeling to assess for weekend effect and identify predictors of in‐hospital mortality. Results We identified 15 762 eligible hospitalizations, with 12 182 (77.3%) having complete covariate data. This sample comprised 53.3% males, 69.3% White race, and had an average (± SD) age of 55.0 ± 14.1 years. We utilized a full‐model approach for covariate inclusion but did not include patient transfer data due to limited availability. We observed no significant mortality differences in weekend admissions (OR = 1.06, 95% CI: 0.97‐1.15, P = 0.02). However, significantly higher mortality was associated with male sex, older age, Black or Hispanic race, Northeast US hospitalization, urban teaching status, and larger hospital size. Sensitivity analyses using multiple imputation datasets and transfer covariates did not change our results. Conclusion We did not observe a weekend effect of in‐hospital mortality for weekend admissions for severe ALI, but our overall diagnosis ascertainment yield was low—indicating that lack of accurate documentation for the etiology of severe ALI may be masking an effect. Additionally, our findings suggest that racial differences and hospital‐level characteristics in the context of severe ALI may be associated with varying outcomes, regardless of admission day, which warrants further research.

and hospital-level characteristics in the context of severe ALI may be associated with varying outcomes, regardless of admission day, which warrants further research.

K E Y W O R D S
acute liver failure, health services, inpatient, mortality, nationwide/national inpatient sample, predictors 1 | INTRODUCTION Severe acute liver injury (ALI) and acute liver failure result in high morbidity and mortality in the absence of prompt diagnosis and causespecific management. [1][2][3] Acute liver failure causes diffuse cerebral dysfunction due to cerebral edema and elevated intracranial pressure, causing high neurological morbidity and mortality in otherwise healthy adults despite potential reversibility with prompt treatment. 4 Acetaminophen toxicity, whether by intentional or unintentional overdose, is responsible for 46% of all acute liver failure cases in the United States, and disease severity is potentially reversible with Nacetylcysteine provision. 5 Severe cases of Budd-Chiari syndrome, although rare, require prompt intervention and liver transplantation. 6 For patients with severe ALI, adverse outcomes may occur without prompt recognition and diagnosis.
Recent studies suggest that differences in patient outcomes may be associated with weekend admission, a phenomenon termed "weekend effect." A weekend effect has been reported in the United States and Europe for a variety of severe, emergent conditions requiring prompt diagnosis and intervention, such as peptic ulcer diseaserelated gastrointestinal (GI) hemorrhage, GI cancers, myocardial infarction, and stroke. [7][8][9][10][11] Meta-analysis for all-cause weekend versus weekday mortality with subgroup analysis by personnel staffing levels, procedure rates, times, and delays, and illness severity, has shown that patients admitted on the weekends have a consistently higher relative risk of mortality. 12 Several predictors for poor outcomes related to weekend admissions have been reported. Transfer status on the weekend for emergent cases like abdominal aortic aneurysm rupture may be associated with higher mortality than weekday transfers. 13 Inconsistency in quality and access to care on the weekends 14 and acute changes in patient status occurring on weekends requiring procedural intervention 15 have been proposed mechanisms for this weekend effect. For ALI, these processes may also contribute to differential patient outcomes.
To date, there has been no investigation for the possibility of a weekend effect for hospitalizations with severe ALI. Severe liver injury and failure with delay in provision of cause-specific treatment or liver transplantation may lead to poor outcomes. As ALI is an acute or subacute disease necessitating specific recognition, subspecialty management, and consideration of transfer for advanced therapies (such as liver transplantation), patients admitted on weekends may receive lower-quality care. To study this question, we analyzed a large, national inpatient database to assess if there is evidence of an effect of weekend admission for in-hospital mortality for severe ALI. states, corresponding to an approximate 20% stratified sample of US community hospitals. This database contains the individual hospitalization as the unit of observation and includes sociodemographic (sex, race, income, and insurance status), clinical (length of stay, ICD-9-CM codes for up to 25 diagnoses), and patient-level outcomes data (mortality, liver transplant status, and hospital transfer status). The NIS also contains hospital characteristics such as teaching status, region and urban/rural location, and hospital bed size. This database is available for purchase through the HCUP. The authors were adherent to the HCUP formal data use agreement guidelines.

| Patient confidentiality and institutional board review
To protect patient confidentiality, the NIS does not provide identifiers that link hospitalizations to unique individuals. Thus, individual patient consent was not sought for this analysis. The Yale School of Medicine (New Haven, CT) Institutional Review Board deemed that this study, using secondary data from a public, de-identified database, met criteria for exemption.

| Patient sample selection, outcomes, and covariates
We included patient hospitalizations with severe ALI, defined as having International Classification of Disease, Clinical Modification, 9 th revision (ICD-CM-9) diagnosis codes of acute and subacute necrosis of liver (570) with encephalopathy (572.2). We extracted sociodemographic and clinical characteristics that may have otherwise confounded the effect of weekend admission on mortality including age, sex, race (White, Black, Hispanic, Asian, Native American, or Other), income quartile, and insurance payer (Medicare, Medicaid, Private, Other). We studied liver-related characteristics that may also affect mortality rates, such as liver disease etiology, if known, and liver transplantation. Comorbidity burden was quantified through calculation of the Elixhauser Comorbidity Index. 16 The primary exposure of interest was admission on a weekend day, and the outcome of interest was in-hospital mortality. We also assessed length of stay as a secondary outcome. Both exposure (weekend admission) and outcomes (inhospital mortality and length of stay) were available as NIS data elements. Hospital-related characteristics included for analysis were geographic region (Northeast, Midwest, South, West) and teaching status (rural, urban nonteaching, urban teaching).

| Statistical analysis
Descriptive statistics on patient hospitalizations and hospital characteristics were extracted. Normality assessment of continuous variables (age) was made based on graphical assessment of the distribution histogram. Tests for differences between normally distributed variables were performed using Student's t-test, and between non-normallydistributed variables, with Kruskal-Wallis test. Differences between categorical variables were tested with chi-square tests. We subsequently performed complete-case univariable and multiple logistic regression modeling of in-hospital mortality by weekend admission and other covariates. All studied covariates were included in the final model (full-model approach) except for patient transfer status due to limited availability, as this data element was only available in the 2008 to 2014 datasets. To account for bias due to confounding by missing data or patient transfer status, we conducted additional sensitivity analyses by producing separate multiple regression models which also included: (a) patient transfer data (indicator for patients transferred in from another health facility) to account for potential confounding, and (b) 10 multiple imputation datasets to account for missing covariate data. Multiple imputation was performed using the MI command of Stata 14.2 statistical analysis software (College Station, TX, USA), utilizing fully conditional specification as to not require assumption of multivariate normal distribution. 17 Statistical significance was defined as a two-tailed P value of <.05. All analyses were performed on Stata 14.2.

| Descriptive data
Overall, demographic characteristics for severe ALI hospitalizations were similar regardless of weekend or weekday status (Table 1). There were no significant differences in race, income quartile, or types of insurance. In patient care, there were no significant differences in liver transplant status or palliative care consultations. However, there was a small but statistically significant difference in age among hospitalizations with weekend admissions (54.4 ± 14.2 vs 55.1 ± 14.0 years, ttest P value = 0.008) and a slightly higher proportion of hospitalizations at large hospitals (68.2% vs 66.1%, chi-square P value = 0.005).
In-hospital mortality was not significantly different between groups (42.3% weekday vs 42.9% weekend, chi-square There were inconsistencies in how this was presented throughout the manuscript.Sometimes, this was presented as "P value =", sometimes as "p", and sometimes as "P". Please be consistent and follow the selected reporting/typesetting guidelines.P value = 0.53). Median length of hospital stay was marginally shorter for weekend compared with weekday admissions (8 vs 9 days, respectively; Kruskal-Wallis P value = 0.0002).
We additionally evaluated the 3580 patients with any missing covariate data (Supplemental Table 1 Additionally, length of stay was shorter in excluded patients (median 8 vs 9 days, P value = 0.01, Kruskal-Wallis test).

| Multiple logistic regression
In the final multiple logistic regression model (  There are several potential explanations for this finding. Severe ALI can present variably over days and even months, unlike myocardial infarction, GI hemorrhage, or stroke, which often present as acute, life-threatening events. An exception to this could be observed in hyperacute liver failure (onset <7 days), and although our study cannot stratify by ALI timing, this is a less-common presentation. 18 Otherwise, severe ALI may evolve over a protracted time course, with antecedent symptoms up to 12 to 24 weeks before presentation. 1  Hospital-level characteristics such as urban and large hospitals were associated with higher mortality, which some studies have attributed to the possibility of sicker patients being transferred to larger referral centers for specialized care. 13  We recognize that our analysis likely does include patients with acute liver failure, recognizing it as the severe phenotype of ALI or progression of severe ALI. In a study of diagnosis code performance in acute liver failure, Lo Re and colleagues found a positive predictive value of 67% using a combination of diagnosis codes; however, this only identified 3% of acute liver failure cases. 22 Furthermore, the nature of the NIS database only reports on hospitalization-level data and does not account for multiple hospitalizations by the same individual, although this would be a larger concern with more rare diseases with very small sample sizes. We did not perform statistical analysis using statistical weights or survey-based analytical methods, as this study did not explore nationwide or temporal trends. Due to our focus on inhospital mortality and length of stay as outcomes of interest, we are not able to make observations on other important clinical outcomes.

| Sensitivity analysis
Previous studies have identified drug-induced or acetaminopheninduced liver injury as the most common culprits for ALF, 23

CONFLICT OF INTEREST
The authors report no affiliation or involvement in potential conflicts of interest. The funding source had no involvement in study design, collection, analysis, interpretation of the data, writing of the report, nor the decision to submit the report for publication.

TRANSPARENCY STATEMENT
The lead author (Albert Do) affirms that this manuscript is an honest, accurate, and transparent account of the study being reported; that no important aspects of the study have been omitted; and that any discrepancies form the study as planned (and, if relevant, registered) have been explained.

DATA AVAILABILITY STATEMENT
The authors confirm that the data supporting the findings of this study are available within the article or its supplementary materials, with original source material from the appropriately cited database.