MATERIALS AND METHODS
The Nationwide Readmissions Database (NRD) from 2013 was utilized for study development and analysis utilizing the International Classification of Diseases, Ninth Revision, Clinical Modification diagnosis and procedure codes. This database is a part of the Healthcare Cost Utilization Project State Inpatient Databases. The unique aspect of this database is its ability to track patients with patient-specific linkage numbers statewide in all nonfederal acute care hospitals over the course of one year. This database provides unweighted information for roughly 14 million discharges per year and is based on weighting to represent the United States national readmission rates, and it accounts for nearly 36 million discharges annually.
The study schema is represented in Figure 1. Hospitalizations for the 13 most common gastrointestinal diseases were identified based on data from the 2012 Nationwide Inpatient Sample and included GI hemorrhage, cholelithiasis with cholecystitis, acute pancreatitis, intestinal obstruction, appendicitis, chronic liver disease (CLD) including viral hepatitis, diverticulitis without hemorrhage, non-infectious gastroenteritis, obesity, C. difficile infection, GI infection, functional/motility disorder, and inflammatory bowel disease (IBD). The corresponding ICD-9 codes found in Supplementary Table 1 were used to query the 2013 NRD database. Patients were excluded from index admissions analysis if they were (1) 17 years old or younger, or (2) pregnant at the index admission. Further exclusion criteria were employed for analyzing thirty-day readmissions, including (1) death during the index admission, or (2) discharged in December 2013 since they could not be followed for thirty-day.
Figure 1 Study schema.
Institutional Review Board approval is not necessary for a population-based public data set based on The Ohio State University Data and Specimen Policy and Human Subjects Research Policy.
Patients and outcomes
For each index admission, patient characteristics were gathered, including age, sex, median income quartile based on zip code, and insurance status based on primary payer (Medicaid, Medicare, private insurance, other) indicated on discharge record. Comorbidity classification was carried out using the Agency for Healthcare Research and Quality-Elixhauser Index classifying patients as having less than 3 or greater than or equal to 3 comorbidities for risk adjustment. Hospital characteristics were also collected, including hospital size (small, medium, large) based on the Healthcare Cost Utilization Project developed algorithm and hospital location and teaching status as a single variable (rural, urban nonteaching, and urban teaching hospital). Lastly, length-of-stay was classified as less than or equal to 3 d, 3 to 7 d or greater than 7 d for index admission and disposition to home/home health/against medical advice or other was indicated.
The primary outcome measures were calculated for each of the 13 conditions: total index admissions, thirty-day readmission rate, mean length of stay, mean hospitalization cost of index hospitalization, and frequency and percent mortality at index admission and any subsequent admission during that calendar-year. The secondary outcomes were predictors of thirty-day readmissions for each of the most common GI and liver diseases.
Index admission characteristics, including the total number of admissions, mean length of stay, mean cost, and mortality rate were calculated for the 13 most common GI diseases in addition to thirty-day readmission and calendar-year mortality rates; data were presented with their corresponding 95% confidence intervals. Thirteen different multivariable logistic regression models were fit to evaluate predictors of thirty-day readmission for each of the most common GI and liver diseases. The models adjusted for all patient demographics and hospital characteristics and results were represented using odds ratios (ORs) and 95% confidence intervals. To examine the relationship between thirty-day readmission rates and any subsequent mortality within the calendar-year, Pearson’s correlation (r) with corresponding significance was calculated. Statistical significance was defined by P < 0.05. SAS 9.4 (SAS, Cary, NC, United States) was used for all analyses. All results are weighted and account for the complex survey design to represent national estimates.
In this analysis of a national inpatient database, we found that emergent conditions such as GI hemorrhage, cholecystitis due to cholelithiasis, intestinal obstruction, and acute pancreatitis were most frequent and accounted for nearly 60% of all index hospitalizations. In terms of inpatient deaths, CLD and GI bleeding together contributed to approximately 60% of index hospitalization and readmission-related mortality. Notably, the mortality rate due to CLD and viral hepatitis during index and subsequent admissions was more than twice higher than any other disease state. Unfortunately, mortality rates due to acute reversible disorders such as C. difficile infection and GI hemorrhage, albeit elevated, nearly tripled during subsequent hospitalizations (from 2% to 5%-6% of admissions). While chronic GI diseases such as those involving the liver, functional/motility disorders, and IBD had high thirty-day readmission rates (15%-25%), a significant opportunity existed to decrease readmission rates (13%-23%) in preventable and non-chronic GI diseases such as C. difficile infection, GI hemorrhage, and acute pancreatitis. In addition, readmissions within thirty-day correlated with increasing rates of mortality during subsequent hos-pitalizations in the same calendar-year.
Gastrointestinal hemorrhage, CLD and viral hepatitis, intestinal obstruction, and C. difficile infection attributed to an alarming 81.64% of all index admission mortality and 76.77% of all subsequent admission mortality. The mortality associated with CLD and viral hepatitis was the highest among the GI diseases examined at 6.07% of index admissions and 12.62% at any subsequent hospitalization. Medical complications following a diagnosis of liver disease are well known, including decompensated cirrhosis, fulminant liver failure, hepatocellular carcinoma, sepsis, acute renal failure, variceal bleeding, and hepatic encephalopathy. The overall mortality burden from CLD and viral hepatitis are expected to continue to increase. In the case of cirrhosis, observational studies have demonstrated a higher burden of mortality with United States annual deaths from cirrhosis increased by 65% from 1999 to 2016, largely driven due to alcoholic cirrhosis and NAFLD[10,11]. Similarly, an assessment of the global burden of viral hepatitis from the global burden of disease study 2013 also demonstrated global viral hepatitis deaths increased from 0.89 million to 1.45 million, an increase far outpacing other communicable diseases. Hospitalizations with a primary diagnosis of GI hemorrhage accounted for 1 of 5 index admissions and 1 of 3 deaths at index or subsequent (same calendar-year) hospitalization. The mortality rate for GI hemorrhage was amongst the highest, notably incurring twice the index mortality rate at any subsequent hospitalization within one-calendar year. Our study encompassed all categories of GI bleed and did not distinguish between the types and sources of GI bleed. A review of the primary literature revealed similarly high mortality rates in other studies. In-hospital mortality rates for non-variceal Upper GI hemorrhage were approximately 2.1% in a 2009 nationwide database analysis. A nationwide study of lower GI bleeding in the United Kingdom revealed in-hospital mortality of 3.4% with a 7.0% subsequent readmission mortality rate, and these findings are comparable to our study. Another alarming finding is the high mortality rates of C. difficile infection. Mortality is initially at 2.32% for index admission; however, mortality quickly jumps to 6.06% for readmissions. The high morbidity and mortality associated with C. difficile infections are well known[15,16]. Prior studies of the nationwide inpatient sample from 1993-2003 revealed a case fatality rate of around 3%-4% for discharged patients with a primary diagnosis of C. difficile. The high initial and subsequent mortality rates for preventable and non-chronic GI diseases such as C. difficile and GI bleeding underscore a key opportunity for providers to focus on preventative care to reduce initial and subsequent readmissions.
Costs for hospitalizations with GI diseases are significant. Mean index admission costs are $15000 for CLD and viral hepatitis, $12000 for cholelithiasis with cholecystitis, and $10000 for preventable GI diseases such as GI hemorrhage, C. difficile infection and acute pancreatitis. In comparison, the average hospitalization costs were $8000 for chronic obstructive pulmonary disease exacerbation, $10000 for Pneumonia, and $11000 for congestive heart failure exacerbation (CHF) in 2013 adjusted dollars. Total index admission costs for GI hemorrhage were $4.6 billion, followed by $4.2 billion in cholelithiasis with cholecystitis. Total costs for C. difficile infection in index admissions were approximately $1 billion and $2.4 billion for CLD and viral hepatitis, whereas, in large part due to sheer numbers, total index admissions cost was roughly $10 billion each for pneumonia and CHF exacerbations.
The average national length of stay is approximately 4.5 d. Among the thirteen most common GI diseases examined, 5 have a longer duration of stay than the national average and seven are below. Although not uniformly consistent, conditions with higher index mortality rates were generally paired with a longer initial length of stay. Among the four diseases with the highest index admission mortality (GI Hemorrhage, CLD and viral hepatitis, C. difficile, and intestinal obstruction), three of which had LOS higher than the national average. Our data suggest that there is a correlation between length of stay and mortality rate. However, our analysis does not examine patients in different stages of a disease state. We suspect that patients with higher mortality burden exist in more severe disease states, thus leading to more prolonged hospital admissions.
Readmission rates for GI diseases ranged from 5% to 25%. Chronic GI diseases such as CLD and IBD had high readmission rates, but preventable and non-chronic GI diseases such as C. difficile, GI hemorrhage, and acute pancreatitis also presented with significant readmission rates (13%-23%). Thirty-day readmission rates for GI diseases were comparable to other disease categories in previously examined database analysis: CHF exacerbations 23%, chronic obstructive pulmonary disease exacerbation 20.1%, 18.1% septicemia, and 14.2% for acute myocardial infarction.
Age, gender, income, hospital size, type of hospital, and disposition were not uniformly predictors of thirty-day readmission for all thirteen diseases examined. However, our analyses showed that patients with Elixhauser comorbidity indices ≥ 3, a longer length of stay, and Medicare/Medicaid were significantly more likely to be readmitted in the conditions examined. Patients with multiple comorbidities are often known to be at increased risk of readmission due to their overall poor baseline health status, which is congruent with findings from other studies[21-23]. As our analysis is unable to stratify the severity of a patient's illness, we believe that longer length of stay serves as a surrogate marker for the severity of a patient’s illness; longer hospitaliz-ation indicates more critically ill patients who are more prone to readmission. Our data should not be interpreted as evidence for reducing the length of stay as we are unable to compare patient disease severity and duration of stay due to the absence of objective clinical severity data.
Possessing Medicare and/or Medicaid were also predictors of readmission. Detailed analysis as to why Medicare and Medicaid patients had higher rates of readmission is challenging to discern while utilizing the NRD. Further studies should examine the reasons for readmissions in this population and to investigate strategies to reduce readmissions.
Disposition of patients to other than home or home-health was associated with early readmissions for GI hemorrhage, acute pancreatitis, cholelithiasis with cholecystitis, diverticulitis without hemorrhage, GI infection, and IBD. However, there was no association with C. difficile. Increasing age was also associated with early readmissions for diverticulitis, cholelithiasis with cholecystitis and CLD and viral hepatitis.
Our multivariable analyses show that factors such as disposition, age, type of insurance, length of stay, and comorbidities are predictors of readmission risks. Our analyses highlight an area of improvement for hospitals, discharging providers, and outpatient providers to not only recognize patients at increased risk for readmission with the aforementioned predictors but also to develop better post-discharge care coordination in patients with these conditions to reduce their readmissions. As the readmission rate is correlated with increased mortality, there is a clear window of opportunity to reduce the readmission rates, especially for preventable and non-chronic diseases. In the case of C. difficile, frequent antibiotic use, particularly clindamycin, cephalosporins and fluoroquinolones, and long-term proton-pump inhibitor use can increase the risk of recurrent C. difficile and should be carefully monitored post-discharge. Newer medications such as Bezlotoxumab, a Food and Drug Administration approved monoclonal antibody against toxin B, offer cost-effective therapeutic option in reducing the incidence of recurrent C. difficile in high-risk patient populations[25,26]. In acute pancreatitis, alcohol consumption, medications, hypertriglyceridemia, and gallstones are known triggers. Readmissions of acute pancreatitis can occur due to uncontrolled symptom management, recurrence of acute pancreatitis, and post pancreatitis complications such as pancreatic necrosis and pseudocyst formation. Readmission reduction strategies can include appropriate management of pain symptoms and close monitoring and treatment of underlying etiologies. Patients with alcohol-induced pancreatitis should have close management of alcohol use disorder with appropriate counseling and consideration of pha-rmacologic therapies[29,30]. In cases of severe hypertriglyceridemia, in addition to non-pharmacologic therapies such as weight loss and dietary changes, statin therapies were associated with a reduction in the number of patients developing pancreatitis. There is some evidence that routine plasmapheresis may also reduce the incidence of recurrent pancreatitis, though there remains a paucity of data, and additional data is needed. For patients with GI hemorrhage, strategies to reduce index admission and readmission include strict adherence to proton-pump inhibitors therapy, avoidance of NSAIDs, and treatment of underlying infections such as Helicobacter pylori. Patients on oral anticoagulation should have frank conversations with their providers weighing the risks and benefits of anticoagulation, as well as using validated scoring systems such as HAS-BLED scores in the case of atrial fibrillation.
There were inherent limitations with nationwide databases that rely on accurate ICD-9 coding[34,35]. The NRD is based on administrative data and lacks clinical variables, which preclude our ability to stratify patients based on the severity of their disease state. Important measures such as race/ethnicity and geographic location were removed as they were considered as patient identifiers. We also did not examine specific disease states or specific conditions associated with some of these GI diseases as predictors of early readmission. Furthermore, the NRD does not link patients for more than a single calendar-year. Finally, our analysis did not focus on predictors of mortality since it has been previously examined in prior papers.