Hospital Factors Associated With Care Discontinuity Following Emergency General Surgery

Key Points

Question  What hospital factors are associated with readmission to a nonindex hospital following emergency general surgery?

Findings  In an analysis of the 100% Medicare inpatient claims file from January 1, 2008, to November 30, 2011, patients who underwent emergency general surgery and were subsequently readmitted to a nonindex hospital were significantly more likely to have had their index surgery at a large, teaching, safety-net hospital.

Meaning  With the adverse outcomes associated with care discontinuity, there is a need for sustained efforts in increasing continuity of care among these hospitals.

Importance  Although there is evidence that changes in clinicians during the continuum of care (care discontinuity) are associated with higher mortality and complications among surgical patients, little is known regarding the drivers of care discontinuity among emergency general surgery (EGS) patients.

Objective  To identify hospital factors associated with care discontinuity among EGS patients.

Design, Setting, and Participants  We performed a retrospective analysis of the 100% Medicare inpatient claims file, from January 1, 2008, to November 30, 2011, and matched patient details to hospital information in the 2011 American Hospital Association Annual Survey database. We selected patients aged 65 years and older who had the most common procedures associated with the previously defined American Association for the Surgery of Trauma EGS diagnosis categories and survived to hospital discharge across the United States. The current analysis was conducted from February 1, 2016, to March 24, 2016.

Main Outcomes and Measures  Care discontinuity defined as readmission within 30 days to nonindex hospitals.

Results  There were 109 443 EGS patients readmitted within 30 days of discharge and 20 396 (18.6%) were readmitted to nonindex hospitals. Of the readmitted patients, 61 340 (56%) were female. Care discontinuity was higher among patients who were male (19.5% vs 18.0%), those younger than 85 years old (19.0% vs 16.6%), and those who lived 12.8 km (8 miles) or more away from the index hospitals (23.7% vs 14.8%) (all P < .001). Care discontinuity was independently associated with mortality (adjusted odds ratio [aOR], 1.16; 95% CI, 1.08-1.25). Hospital factors associated with care discontinuity included bed size of 200 or more (aOR, 1.45; 95% CI, 1.36-1.54), safety-net status (aOR, 1.35; 95% CI, 1.27-1.43), and teaching status (aOR, 1.18; 95% CI, 1.09-1.28). Care discontinuity was significantly lower among designated trauma centers (aOR, 0.89; 95% CI, 0.83-0.94) and highest among hospitals in the Midwest (aOR, 1.15; 95% CI, 1.05-1.26).

Conclusions and Relevance  Nearly 1 in 5 older EGS patients is readmitted to a hospital other than where their original procedure was performed. This care discontinuity is independently associated with mortality and is highest among EGS patients who are treated at large, teaching, safety-net hospitals. These data underscore the need for sustained efforts in increasing continuity of care among these hospitals and highlight the importance of accounting for these factors in risk-adjusted hospital comparisons.

Surgeons and hospitals are increasingly being held accountable for postoperative and postdischarge outcomes of the patients they treat.^1-3 Changes in care professionals at any point along the continuum of care have been associated with adverse outcomes among patients.^4-6 Outcomes, such as longer hospital lengths of stay, higher mortality rates, increased risk for complications, and lower patient satisfaction, have been associated with discontinuity in care.^4-6 Surgical patients who are readmitted to the same hospitals where they were initially treated have been shown to have better outcomes compared with their counterparts readmitted to nonindex hospitals.^7-9 Therefore, ensuring continuity of care may be an important step toward reducing adverse outcomes associated with surgery.

Emergency general surgery (EGS) accounts for more than 7% of all hospitalizations and an estimated $28 billion in annual hospital costs in the United States.^10,11 Compared with surgical patients who have elective procedures, those emergently treated are known to have significantly worse outcomes including higher mortality and complication rates.^12,13 Although the worse outcomes in these patients are often attributed to their inherent clinical conditions, other factors—such as differential transfusion practices, patients’ payer status, and hospital organizational structures at the time of presentation—have been shown to play important roles in the outcomes among these patients.^14-16 With the growing emphasis on the provision of high-value care to patients,¹⁷ sustained efforts must be made to identify ways to improve outcomes among EGS patients.

Despite the fact that nearly one-fifth of readmitted EGS patients are readmitted to nonindex hospitals,¹⁸ little is known regarding hospital factors that contribute to this care discontinuity following discharge for EGS. Differences in hospital characteristics have been shown to account for some of the variations in surgical readmission and understanding how these characteristics influence outcomes is useful for creating expectations of patient care, patient triage, and risk-adjusted hospital comparisons.^19,20 Therefore, the objective of this study was to identify hospital characteristics associated with care discontinuity, defined as readmission to a nonindex hospital, in EGS. We hypothesized that specific hospital characteristics will be associated with higher care discontinuity among EGS patients.

We used the 100% Medicare inpatient claims files from January 1, 2008, to November 30, 2011, to conduct this analysis from February 1, 2016, to March 24, 2016. The Medicare inpatient claims files contain hospital admission information collected by Medicare from the more than 45 million beneficiaries across the United States.^21,22 Unique patient identifiers are assigned to individual patients, making it possible to link multiple admissions during the study.

Using the American Association for the Surgery of Trauma definitions of EGS,²³ we selected patients 65 years and older with 1 of 467 unique admitting International Classification of Diseases, Ninth Revision, Clinical Modification diagnosis codes corresponding with EGS. The diagnosis categories and the International Classification of Diseases, Ninth Revision, Clinical Modification codes associated with them have been previously published.²³ The categories include upper gastrointestinal tract (UGT; upper gastrointestinal bleeding, peptic ulcer disease, fistulae, gastrostomy, small intestinal cancers, ileus, Meckel diverticulum, bowel perforation, and appendix); hepatic-pancreatic-biliary (gallstones and related diseases, pancreatitis, and hepatic abscesses); soft tissue (cellulitis, abscesses, fasciitis, wound care, pressure ulcers, and compartment syndrome); hernia (inguinal, femoral, umbilical, incisional, ventral, and diaphragmatic); intestinal obstruction (adhesions, incarcerated hernias, cancers, volvulus, and intussusceptions); cardiothoracic (cardiac tamponade, empyema, pneumothorax, and esophageal perforation); colorectal (lower gastrointestinal bleeding, diverticula disease, inflammatory bowel disease, colorectal cancers, colitis, colonic perforations, megacolon, regional enteritis, colostomy/ileostomy, hemorrhoids, perianal and perirectal fistulas and infections, anorectal stenosis, and rectal prolapse); vascular (ruptured aneurysms, acute intestinal ischemia, acute peripheral ischemia, and phlebitis); general abdominal (abdominal pain, abdominal mass, peritonitis, hemoperitoneum, and retroperitoneal abscesses); resuscitation (acute respiratory failure and shock); and others (tracheostomy, foreign bodies, and bladder rupture). Because the American Association for the Surgery of Trauma definition of EGS is based on diagnosis rather than procedure, it would be possible for a single procedure to be listed under more than 1 category.

To ensure we were examining patients who underwent surgical operations, we determined the most frequent single procedure performed in each of the diagnosis categories and restricted our analyses to patients who underwent these procedures. We did not include patients in the American Association for the Surgery of Trauma diagnosis category of “others” because the group was not homogenous and each procedure did not necessarily represent procedures commonly associated with diagnoses in the group. Patients discharged to home, home nursing care, rehabilitation, or skilled nursing facility were included. Owing to the fact that readmission patterns of patients transferred or discharged to postacute care facilities may be directly influenced by the subsequent facilities at which they were treated, we performed a sensitivity analysis by analyzing only patients who were discharged home. This study examined deidentified data and was deemed exempt from full review by the institutional review board of Partners Healthcare.

We included patient demographic data such as age (<85 years vs ≥85 years), sex, race/ethnicity (white, black, or other), and rural-urban commuting area (urban, suburban, large rural town, or rural). Because home distance from the index hospital is a known predictor of readmission destinations,⁷ we accounted for this by calculating the distance between index hospitals and the zip code centroids of patients’ homes. We categorized distance from index hospitals as less than 12.8 km (8 miles) and 12.8 km or more based on prior research showing that a differentially high proportion of readmitted patients returned to the index hospitals if the hospitals were less than 12.8 km from home.⁷

We calculated the Charlson Comorbidity Index score, which is a validated score that uses 19 possible diagnoses to predict mortality and perioperative complications in longitudinal data.^24-26 The Charlson Comorbidity Index score was classified as 0, 1, 2, and 3 or greater (higher being worse). We also measured hospital length of stay as a continuous variable.

To identify hospital characteristics, we linked the patient records to hospital information contained in the American Hospital Association database and the Medicare Final Impact Files for 2008 to 2011. We identified safety-net hospitals as hospitals in the top quartile of Medicare Disproportionate Share Hospital patient percentage. We used this definition because unlike other definitions, it accounts for the socioeconomic level of hospitals’ patient populations even if the patients are insured.²⁰ Hospital bed size (<200 vs ≥200 beds), teaching status (teaching vs nonteaching), and trauma center status (designated trauma center vs nondesignated trauma center) were also assessed. We assessed hospital location by extracting information on the core-based statistical area (metropolitan, micropolitan, division, or rural) and the region (Northeast, Midwest, South, and West).

Our primary outcome of interest was care discontinuity defined as readmission to a nonindex hospital within 30 days of discharge home. This was calculated as the proportion of patients who were readmitted to nonindex hospitals among the patients who were readmitted within 30 days of discharge. The secondary outcome was mortality within 30 days of discharge following the index hospitalization. Death was obtained from the Medicare Beneficiary Summary File.

Characteristics of patients who were readmitted to nonindex hospitals were compared with those readmitted to index hospitals using χ² tests. Proportions of patients who were readmitted to nonindex hospitals were also compared among hospital categories. We then used univariate logistic regression models to determine the patient and hospital factors that were associated with readmission to nonindex hospitals. Only factors that had P values of less than .10 in the univariate models were used in building a multivariate logistic regression model. All variables had complete data for nearly all patients (approximately 99%) and complete case analyses were used for the regression analyses. Adjusting simultaneously for patient factors in the model, we identified index hospital factors that are independently associated with care discontinuity and overall risk for mortality. Using this model, we also compared the odds of readmission with nonindex hospitals among the different EGS diagnosis categories. We then calculated the adjusted rates of readmission to nonindex hospitals for each of the diagnosis categories. To account for the nonindependence of patients treated at the same facilities, models were built with robust standard errors clustered at the level of individual hospitals. All analyses were done using SAS version 9.4 for Windows (SAS Institute) and the significance level was set as P < .05.

There were 109 443 EGS patients readmitted within 30 days of discharge home. Demographic and readmission data for the entire cohort are shown in the eTable in the Supplement. The overall readmission rate was 12.6%. Readmission rates stratified by diagnosis category ranged from 10.4% for hepatic-pancreatic-biliary to 27.7% for resuscitation. The most common procedure associated with each of the diagnosis categories were as follows: laparoscopic cholecystectomy (hepatic-pancreatic-biliary); open and other right hemicolectomy (UGT); other lysis of peritoneal adhesions (intestinal obstruction); open and other sigmoidectomy (colorectal); other incision with drainage of skin and subcutaneous tissue (soft tissue); other open incisional hernia repair with graft or prosthesis (hernia); temporary tracheostomy (resuscitation); angioplasty or atherectomy of other noncoronary vessel(s) (vascular); other incision of pleura (cardiothoracic); and other laparotomy (general abdominal). Overall mortality for the cohort was 6.1%.

Among readmitted patients, 20 396 (18.6%) were readmitted to nonindex hospitals. Unadjusted mortality for patients readmitted to nonindex hospitals was 7.2% compared with 5.8% for patients readmitted to the index hospital (P < .001). After controlling for all patient and hospital variables, readmission to a nonindex hospital was independently associated with mortality (adjusted odds ratio [aOR], 1.16; 95% CI, 1.08-1.25).

Patients readmitted to a nonindex hospital were more likely to be male, younger than 85 years old, and of black race (P < .001) (Table 1). The median (SD) distance from home to index hospital was 9.6 (146) km (6 miles) for patients readmitted to the index hospital and 15 (210.5) km (9.4 miles) for those readmitted to a nonindex hospital (P < .001). The median (SD) distance from home to readmitting hospital was 9.6 (146) km (6 miles) for those readmitted to the index hospital and 16.2 (184) km (10.1 miles) for those readmitted to a nonindex hospital (P < .001). Compared with patients who lived less than 12.8 km from the index hospitals, readmissions to nonindex hospitals were more common among patients who lived at least 12.8 km away from the index hospitals (23.7% vs 14.8%; P < .001). Patients readmitted to the index hospital had shorter index length of stay compared with patients readmitted to a nonindex hospital (mean [SD], 10.6 [9.0] vs 12.2 [13.3] days; P < .001) and shorter readmission length of stay (mean [SD], 7.6 [7.2] vs 7.9 [7.8] days; P < .001). Care discontinuity was highest among patients in the resuscitation diagnosis category (39.7%) and lowest in the hernia category (14.8%).

When hospital characteristics were examined, safety-net hospitals (22.5% vs 16.7%; P < .001), teaching hospitals (20.7% vs 18.1%; P < .001), and large (≥200 beds) hospitals (22.6% vs 17.1%; P < .001) had significantly higher proportions of their patients being readmitted to nonindex hospitals (Table 2). Hospitals in rural areas had much more care discontinuity compared with those in metropolitan areas (32.9% vs 17.6%; P < .001). Among US regions, hospitals in the Midwest had the highest care discontinuity (19.3%) while those in the Northeast had the lowest (17.2%) (P < .001).

Of all the patients readmitted at nonindex hospitals, 78.6% were readmitted to nonteaching hospitals, 70% of them were readmitted to large hospitals, and 61.2% were readmitted to hospitals in the metropolitan areas.

Multivariate analyses showed that hospital characteristics independently associated with care discontinuity included large size (aOR, 1.45; 95% CI, 1.36-1.54), safety-net status (aOR, 1.35; 95% CI, 1.27-1.43), and teaching status (aOR, 1.18; 95% CI, 1.09-1.28) (Table 3). Designated trauma centers had less care discontinuity compared with nondesignated trauma centers (aOR, 0.89; 95% CI, 0.83-0.94). Compared with hospitals in the Northeast, hospitals in other regions had significantly higher rates of care discontinuity.

Compared with patients in the hepatic-pancreatic-biliary diagnosis category, those in the UGT, intestinal obstruction, colorectal, and hernia categories had significantly lower odds of care discontinuity (Table 4). Those in the resuscitation, soft tissue, and vascular diagnosis categories had significantly higher care discontinuity. The adjusted rates of care discontinuity among EGS patients ranged from 20% (colorectal) to 64% (resuscitation) (Figure). The results did not significantly change on sensitivity analysis of only those patients discharged to home (data not shown).

In this study, we have shown that more than 12% of elderly EGS patients will be readmitted within 30 days and that nearly one-fifth of readmitted patients do not return to the index hospitals. These findings are in agreement with previously published work indicating a high degree of care discontinuity in this population.¹⁸ We have further demonstrated that care discontinuity among EGS patients is independently associated with short-term mortality. Patients treated at large hospitals, teaching hospitals, and safety-net hospitals were more likely to have care discontinuity. With the disproportionately poor outcomes associated with EGS, this study highlights a potential area for intervention to ensure improved postdischarge outcomes of EGS care.

Safety net hospitals are more likely to be teaching hospitals and to have large bed sizes,²⁰ meaning that patients in this unique cluster of hospitals will have disproportionately higher likelihoods of care discontinuity. It is not clear what the major drivers of these hospital-based differences are. However, safety-net hospitals have been shown to have lower performances on measures of patient experience such as satisfaction scores,²⁷ which might influence patients’ decision to return to the hospital for their readmission. Although the patient characteristics of these hospitals are unique in that they are more likely lower-income patients, the hospitals themselves are also likely to be poorly resourced.²⁰ Most of these hospitals are funded by limited government support and privately insured higher-income patients are more likely to stay away from these hospitals because of their reputations.²⁰ In addition, these hospitals are consistently disadvantaged under policies that place penalties on hospitals for lower comparative hospital ratings,^28,29 which further reduces their available funds, leading to a vicious circle of underperformance that could potentiate care discontinuity. Because a significant proportion of EGS patients are treated in these hospitals,³⁰ EGS patients are especially disadvantaged in both their hospital experiences and their postdischarge outcomes. Understanding strategies to improve satisfaction of patients in these hospitals and adjusting reimbursement policies to reduce the undue disadvantages these hospitals face may be beneficial in reducing care discontinuity among EGS patients.

Interestingly, many of the patients who were readmitted to nonindex hospitals were readmitted to large safety-net hospitals. The patterns of care provided at these hospitals may potentially predispose the patients to bad outcomes. Safety net hospitals are known to have low rates of meaningful use of electronic health records,^31,32 which makes it more difficult to share information with prior clinicians. Patients in these hospitals have also been reported to be less comfortable with the level of communication they have with their doctors compared with patients in other hospitals.²⁷ Communication gaps have been identified as one of the strongest factors associated with the negative effects of care discontinuity, and change of clinicians even within the same hospital has been shown to lead to higher surgical mortality rates.^4,9 Therefore, the impaired ability to access past medical records coupled with the inherent limitations in the available resources at these readmitting hospitals exposes EGS patients to a potentially different level of care on readmission to a nonindex hospital.

Despite the fact that hospital characteristics may drive some of the differences in care discontinuity, it is plausible that patients’ choices play roles in their decision to go to nonindex hospitals. While we adjusted for important patient characteristics, there may be other unmeasured factors that lead to care discontinuity. While living more than 12.8 km from the index hospital was associated with being readmitted to a nonindex hospital, we did not find a trend to being readmitted to hospitals closer to home. We found regional differences in care discontinuity as hospitals in the Northeast had significantly lower care discontinuity compared with other regions. A prior study also found regional variations in care discontinuity, which were seemingly unrelated to the urbanicity or size of such locations.⁷ This means that common cultures, perceptions, or practices of the patients or the treating hospitals may influence decisions to be readmitted at nonindex hospitals. Educating patients on the potential adverse outcomes associated with care discontinuity and ensuring that hospitals encourage patient follow-up prior to discharge may reduce care discontinuity among EGS patients.

Although we examined a distinct cohort of patients, there was significant heterogeneity in care discontinuity among EGS patients. Patients in the colorectal, hernia, intestinal obstruction, and UGT categories had low care discontinuity while patients in the resuscitation category had the highest care discontinuity. In another study, patients in the resuscitation category were shown to have significantly higher complication and mortality rates than other EGS patients.¹⁵ As the most common procedure performed in the resuscitation group was tracheostomy, it is likely that the increased morbidity and mortality found in the resuscitation group was a result of the underlying condition of the patient rather than the risk of the procedure itself. It is possible that the acuity of the presenting diagnoses during the index admissions could determine the likelihood of patients being admitted to the nearest hospitals and that the subsequent readmissions will be to the hospitals with which they are more familiar. However, these are only speculations and the differences in care discontinuity among EGS categories deserve further investigation.

There are a number of limitations to this study. The Medicare inpatient file claims database represents a group of relatively homogenous patients (all >65 years and all with insurance), which may limit the generalizability of the study findings. Like all administrative databases, the Medicare inpatient file claims do not contain physiologic data for more detailed risk adjustment. We chose to categorize age as a binary variable (65-85 or >85 years) based on initial data that showed this was an inflection point for our outcome (data not shown); however, it is possible that smaller subgroups of age have differing associations with care discontinuity. Additionally, we do not have detailed information on the reasons patients were readmitted to nonindex hospitals. It is possible that other factors, such as established relationships between 2 hospitals or situations where surgeons treat patients in multiple hospitals, may account for some of these trends. We also examined a limited spectrum of hospital characteristics and there may be further contributing factors not analyzed. However, the characteristics we include form large parts of the factors routinely included in hospital comparison models.¹⁹ Last, as with all observational studies, we cannot prove causation, only association.

Nearly 1 in 5 older EGS patients is readmitted to a hospital other than where their original procedure was performed. This care discontinuity is independently associated with mortality. Emergency general surgery patients who are treated at large, teaching, safety-net hospitals are likely to have high rates of discontinuity in their postdischarge care. With the adverse outcomes associated with care discontinuity, it is important to ensure continuity of care following emergency operations. Increased efforts to ensure patient satisfaction, improved communication between patients and clinicians, and integration of medical information across different hospitals are possible strategies to reduce the negative outcomes associated with care discontinuity among EGS patients.

Corresponding Author: Joaquim M. Havens, MD, Division of Trauma, Burns and Surgical Critical Care, Department of Surgery, Brigham and Women’s Hospital, 75 Francis St, Boston, MA 02115 (jhavens@partners.org).

Accepted for Publication: July 26, 2016.

Published Online: November 16, 2016. doi:10.1001/jamasurg.2016.4078

Author Contributions: Dr Havens had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Concept and design: Havens, Olufajo, Cooper.

Acquisition, analysis, or interpretation of data: Olufajo, Tsai, Jiang, Columbus, Nitzschke, Salim.

Drafting of the manuscript: Havens, Olufajo.

Critical revision of the manuscript for important intellectual content: All authors.

Statistical analysis: Olufajo, Tsai, Jiang.

Administrative, technical, or material support: Tsai, Salim.

Study supervision: Havens.

Conflict of Interest Disclosures: None reported.

Previous Presentation: This study was presented at the 26th Annual Scientific Meeting of the Society of Black Academic Surgeons; April 30, 2016; Columbus, Ohio.