Comparison of Rates and Outcomes of Readmission to Index vs Nonindex Hospitals After Major Cancer Surgery

Key Points

Questions  What proportions of readmissions after a major cancer operation are to a hospital other than the hospital where the surgery was performed and does this influence readmission outcomes?

Findings  Using data from the Nationwide Readmissions Database, this study found that 20.1% of readmissions after major cancer surgery were to a nonindex hospital. These nonindex hospital readmissions were associated with significantly higher risk-adjusted mortality and morbidity compared with readmissions to index hospitals.

Meaning  Increasing regionalization of major cancer surgery must account for the potential adverse effect of nonindex readmissions attributable to fragmentation of care; targeted interventions aimed at reducing nonindex readmissions may improve readmission outcomes overall.

Importance  Increasing regionalization of cancer surgery has the inadvertent potential to lead to fragmentation of care if readmissions occur at a facility other than the index hospital. The magnitude and adverse effects of readmission to a facility other than the one where the surgery was performed are unclear.

Objectives  To assess rates of readmission to nonindex hospitals after major cancer surgery and to compare outcomes between index and nonindex hospital readmissions.

Design, Setting, and Participants  In this multicenter, population-based, nationally representative study of adult patients undergoing a major cancer operation (defined as esophagectomies or gastrectomies, hepaticobiliary resections, pancreatectomies, colorectal resections, or cystectomies), retrospective analyses were performed using the Nationwide Readmissions Database (admissions from January 1 through September 30, 2013). Descriptive analyses were performed to determine 90-day readmission characteristics, including timing, cost, and outcomes. Adjusting for clustering by facility, the study used multivariate logistic regression to identify factors associated with nonindex vs index readmissions. The study also used regression models to identify differences in mortality, major complications, and subsequent readmissions between the 2 groups. Data analysis was performed from January 1 through December 31, 2013.

Exposures  Readmission to index vs nonindex hospitals (defined as any hospital other than the hospital where the major cancer operation was performed).

Main Outcomes and Measures  Proportion of 90-day readmissions and nonindex readmissions after major cancer surgery, factors associated with nonindex readmissions, and difference between in-hospital mortality, hospital costs, and subsequent readmissions for patients admitted to index vs nonindex hospitals.

Results  A total of 60 970 patients were included in the study (mean [SD] age, 67 [13] years; 7619 [55.6%] male and 6075 [44.4%] female). The 90-day readmission rate was 23.0%. Of the 13 695 first readmissions, 20.1% were to a nonindex hospital. Independent factors associated with readmission to a nonindex hospital included type of procedure, comorbidities (OR, 1.40; 95% CI, 1.15-1.70), elective admission (OR, 1.21; 95% CI, 1.06-1.37), discharge to a nursing facility (OR, 1.20; 95% CI, 1.07-1.36), and surgery at a teaching hospital (OR, 1.16; 95% CI, 1.00-1.34) (all P < .05). After risk adjustment, patients readmitted to nonindex hospitals had 31.2% higher odds of mortality (odds ratio, 1.31; 95% CI, 1.05-1.64) and 27.3% higher odds of having a major complication (odds ratio, 1.27; 95% CI, 1.14-1.42). Subsequent readmissions and hospital costs were not different between the 2 groups.

Conclusions and Relevance  Approximately one-fifth of readmissions were to a nonindex hospital and were associated with higher mortality and morbidity than readmission to index hospitals. Factors that influence nonindex readmissions have been identified to target interventions.

The implementation of the Hospital Readmissions Reduction Program in 2012 made reduction of readmissions a health care priority.¹ Under this program, readmissions are considered to be a quality metric by the Centers for Medicare & Medicaid Services, and hospitals incur penalties if readmission rates are higher than average.² In addition to the increased health care costs, readmissions may result in poorer outcomes for patients, especially when readmissions are to a hospital different from the hospital where the surgery was performed (nonindex hospitals). Readmissions to a nonindex hospital after major general surgical and vascular procedures are associated with 25% to 48% increased odds of mortality.^3,4

Readmission to a nonindex hospital is especially relevant in the care of patients undergoing major cancer operations. Optimum management of patients with cancer requires multidisciplinary care coordination, and fragmentation of such care can lead to worse outcomes.^3,5 Prior work has found that readmission after cancer surgery occurs in 30% to 50% of patients⁶ and nonindex readmissions occur in 20% to 34%.^7-9 However, studies thus far have been limited in their ability to capture such nonindex readmissions on a national level or across multiple cancer types. The introduction of the Nationwide Readmissions Database (NRD) has made nationally representative assessment of hospital readmissions possible. We aimed to quantify rates of readmission to nonindex hospitals after major cancer surgery and to compare outcomes between index and nonindex hospital readmissions across the United States. We hypothesized that short-term outcomes of in-hospital morbidity and mortality would be worse among patients readmitted to nonindex compared with index hospitals.

We analyzed the Health Care Utilization Project NRD (admission year 2013). The NRD contains nationally representative data on hospital readmissions for all payers and uninsured patients.¹⁰ It contains information from 21 geographically dispersed states that account for 49.3% of the total US resident population and 49.1% of all US hospitalizations. Each entry has a unique verified patient linkage number that is used to track patients across hospitals and therefore can be used to analyze readmissions. Variables in the data set include patient demographics, disease-specific diagnostic codes for each admission, calculated severity measures, outcomes, and hospital-level characteristics. The NRD is unique in its ability to not only identify readmissions within a given period but also study the characteristics of the readmission. The NRD is a publicly available deidentified database. This study was exempt from review by the institutional review boards at Howard University Hospital and the Mayo Clinic in Arizona.

We selected all adult patients (aged ≥18 years) from the NRD who were undergoing a complex cancer operation within the first 9 months of the year. All patients had at least 90 days of follow-up time. Complex operations were defined as esophagectomies or gastrectomies, hepaticobiliary resections, pancreatectomies, colorectal resections, lung resections, and cystectomies. These operations were chosen because they represent a broad sample of surgical procedures that are associated with a clinically significant readmission rate. International Classification of Disease, Ninth Edition, Clinical Modifications (ICD-9-CM) procedure codes were used to identify patients who underwent one of the above-mentioned procedures (eTable 1 in the Supplement). The ICD-9-CM diagnosis codes were used to identify patients with a cancer diagnosis that corresponded to the procedure being performed (eTable 1 in the Supplement). The Figure depicts our selection process. Data analysis was performed from January 1 through December 31, 2013.

Age was included as a categorical variable in deciles of age. Severity of illness was assessed using the all patient refined diagnosis related groups severity of illness variable. This variable, included in the data set for each admission, is calculated by 3M using regression analyses and categorizes patients into minor, moderate, major, and severe loss of function categories. Comorbidities included the Agency for HealthCare Research and Quality–generated Elixhauser comorbidity variables.¹¹ These variables were used to generate a comorbidity score for each patient and grouped patients into 4 categories (0, 1, 2, or >2 comorbidities). A comorbidity of malignant tumor was omitted during the scoring because all the patients had a diagnosis of malignant tumor.

In addition, we used ICD-9-CM diagnosis codes to generate variables for major complications, defined as pneumonia, pulmonary embolism, renal failure, cerebrovascular accident, myocardial infarction, cardiac arrest, adult respiratory distress syndrome, sepsis, and septic shock (eTable 1 in the Supplement). A binary variable (yes/no) was created to signify the presence of any of these complications during the admission, as described elsewhere.¹² Length of hospital stay was included as a continuous variable. Hospital costs were generated by merging hospital charges with the cost-to-charge ratio files provided with the data set. Our outcome variables were discharge disposition, length of hospital stay, cost, major morbidity, and in-hospital mortality.

Analyses were performed separately for the primary admission (ie, the admission during which the procedure was performed) and for readmissions. Discharge-level weights provided with the data set provided for national estimates.¹⁰ The NRD-provided time to readmission variable and the primary admission length of stay were used to calculate the number of days to readmission. All readmissions that occurred within 90 days of discharge were included in this analysis. We used 90 days not 30 days to capture the use of adjuvant therapy that may have contributed to the readmission. Readmission proportions were calculated only for patients who survived to discharge.

All 90-day readmissions were categorized as index vs nonindex. Index hospitals were those at which the patient had the initial major cancer operation performed; all other hospitals were nonindex hospitals. Descriptive analyses were performed for the first readmission by index vs nonindex hospital status and tabulated. Bivariate analyses were performed to compare differences in patient and hospital characteristics between these 2 categories. Patients who were transferred to an index hospital during an initial nonindex readmission are included in the index hospital category. Sensitivity analysis that excluded transferred patients was also performed. Multivariable logistic regression was used to identify factors present on the primary admission that were associated with readmission to a nonindex vs an index hospital. Variables in the model included age, sex, comorbidity score, procedure performed, resident of state vs nonresident of state, insurance, income, discharge month, severity of illness, elective vs nonelective admission, hospital characteristics, presence of major complication, prolonged length of stay, and discharge to a facility vs home. The model accounted for the stratified single-stage cluster sampling design of the NRD, and linearized SEs were used.⁶

To compare differences in outcome between readmission to index vs nonindex hospitals, we used only the first 90-day readmission of the patient. These analyses were performed on data specific to the readmission encounter. We performed multivariable regression analyses to determine adjusted outcome differences between the 2 hospital categories. Logistic regression was used for in-hospital mortality, major complications, prolonged length of stay, and additional readmissions (a second readmission) with a separate multivariable model for each outcome. Generalized linear modeling with log link and γ family were used to test for differences in hospital cost and length of stay. All variables listed above and the cause of readmission were included in the models. To adjust for the variation in outcomes by facility, all models were adjusted for clustering by facility using robust SEs. Sensitivity analyses were performed for mortality and morbidity separately for each of the major oncologic procedures.

The Hosmer-Lemeshow goodness-of-fit test was used to assess each model.¹³ In addition, the F test and linktest were performed, and the pseudo R² was checked.^14,15 All analyses were performed with Stata software, version 13 (StataCorp). Where applicable, the χ² tests and t tests were used to test for significance. P < .05 (2-sided) was considered to be statistically significant.

We analyzed 92 260 visits by 60 970 adult patients (mean [SD] age, 67 [13] years; 7619 [55.6%] male and 6075 [44.4%] female) who underwent a major cancer operation from January 1 through September 30, 2013. Patient and hospital characteristics are given in eTable 2 in the Supplement. For the primary admission, 16.2% experienced a major complication, median length of stay was 6 days (interquartile range [IQR], 4-10 days), and the overall in-hospital mortality rate was 2.4%.

Of the 59 493 patients who survived until discharge, 13 695 (23.0%) were readmitted within 90 days. The number of 90-day readmissions ranged from 1 to 8, with 4840 patients (8.1% of all patients) having more than 1 readmission. The median time to readmission was 30 days, with an IQR of 10 to 58 days (eFigure in the Supplement). Table 1 lists the number of eligible patients by procedure and their readmission rates. The 90-day readmission rates varied significantly by procedure and ranged from 18% for lung resections to 39% for cystectomies (Table 1). Of the 18 535 total 90-day readmissions, 4091 (22.1%) were to a different hospital. The median hospital cost of a 90-day readmission was $8353, with an IQR of $4971 to $15 262. The total weighted cost of all 90-day readmissions was US $554 million (range, US $542 million to US $566 million).

When analyzing only the first readmission (13 695 patients), 2757 patients (20.1%) were readmitted to a nonindex hospital. Differences in initial patient and hospital characteristics between index and nonindex hospital readmissions are presented in Table 2. On multivariate analysis, factors that were more likely to be associated with admission to a nonindex hospital were higher comorbidity scores (odds ratio [OR], 1.40; 95% CI, 1.15-1.70; for score of 3 vs 0), elective vs nonelective admission (OR, 1.21; 95% CI, 1.06-1.37), being a resident of the same state (OR, 3.18; 95% CI, 2.16-4.68), metropolitan teaching vs metropolitan nonteaching hospitals (OR, 1.16; 95% CI, 1.00-1.34), living in smaller counties (OR, 1.75; 95% CI, 1.35-2.28; for nonmetropolitan, nonmicropolitan counties compared with central metropolitan areas), and being discharged to a facility vs home (OR, 1.20; 95% CI, 1.07-1.36) (Table 3). Patients with higher median household incomes (OR, 0.76; 95% CI, 0.64-0.93; highest compared with lowest) and those who underwent surgery at large compared with small hospitals (OR, 0.73; 95% CI, 0.54-0.93) were less likely to be readmitted to a nonindex hospital. The model satisfied postestimation testing criteria, with the F test being significant at P < .001, the linktest at P = .66, the χ² test at P = .75, and an R² of 0.158.

The most frequent causes of index and nonindex hospital readmissions included infections, gastrointestinal complications, respiratory conditions, dehydration or acute kidney injury, and cardiovascular conditions (eTable 3 in the Supplement). Patients with conditions of a higher acuity or related to medical comorbidity had a higher proportion of nonindex readmissions.

On multivariate analysis, 90-day readmission to a nonindex hospital resulted in 27% higher odds of experiencing a major complication (OR, 1.27; 95% CI, 1.14-1.42) and 31% higher odds of mortality (OR, 1.31; 95% CI, 1.05-1.64) compared with index hospital readmissions (Table 4). The mean overall length of stay was slightly shorter at nonindex hospitals (OR, 0.92; 95% CI, 0.89-0.96). There was no difference between cost and additional readmission between the 2 groups. The results of stratified analyses by each procedure type are presented in eTable 4 in the Supplement. Patients who were transferred back to the index hospital accounted for only 0.5% of the sample. Sensitivity analysis that excluded these patients revealed similar findings. All models satisfied posttesting criteria.

Approximately one-fifth of readmissions after major cancer surgery were to nonindex hospitals in this nationally representative data set. Nonindex readmissions were associated with 27% higher odds of morbidity and 31% higher odds of mortality compared with readmissions to index hospitals. To our knowledge, this is the first nationally representative study on quantifying nonindex readmissions after major cancer surgery and the association of such care fragmentation with higher morbidity and mortality.

The high degree of coordination required for the care of patients undergoing complex cancer surgery presents a unique challenge if an unexpected event or exacerbation of an underlying medical condition occurs in the postoperative period. Often a multidisciplinary approach is necessary with detailed knowledge of the patient’s history, most recent surgery, and hospital course needed to optimize a management plan.^6,16 However, increasing regionalization of complex cancer operations means that such information may not be immediately available to local facilities that manage an acute problem, with the potential of worse outcomes—the classic scenario for health care fragmentation. Similar to our study, prior reports^7,8,10,17-19 have also highlighted a high degree of care fragmentation after major cancer operations. Zheng et al,⁸ in an analysis of the California State Inpatient Database, found 20% of readmissions were to nonindex hospitals. In their study, nonindex readmissions were also associated with 31% higher odds of mortality. Stitzenberg et al,¹⁷ in an analysis from the Surveillance, Epidemiology, and End Results (SEER)–Medicare linked database, found that one-third of readmitted patients were at nonindex hospitals. Also using the NRD, Chappidi et al⁷ found that 1 in 3 patients experienced care fragmentation after surgery for urologic cancers. Although our analysis allows only a study of in-hospital outcomes for readmissions, long-term outcomes have also been reported to be adversely affected. Fragmentation of care after major cancer surgery influences the timing of adjuvant therapy and decreases 1-year survival.^10,18,19

In our analysis, we found no difference between hospital cost for index vs nonindex readmissions. This finding is similar to those of a prior study²⁰ on readmission after radical prostatectomy. However, for complex vascular surgery, readmission to nonindex hospitals resulted in lower costs compared with index hospital readmissions.²¹ This variation is likely to be related to differences in the study population; however, further research is required.

Factors associated with readmission to a nonindex vs index hospital are multifactorial and broadly categorized as patient- and hospital-related factors. Some of these associations can be explained by travel distance because patients, especially those who are older and have more comorbidities, are more likely to present at a hospital closer to home in emergency situations. Stitzenberg et al,¹⁷ in their analysis of the SEER-Medicare data set, found readmission to a nonindex hospital to be significantly associated with travel distance. Patients who lived the farthest from the index hospital had 50% higher rates of emergency department visits and an almost 5 times higher rate of nonindex readmissions than did those living closest to the index hospital. Regionalization of care prompts patients to travel significant distances to undergo their surgical procedure^22,23; however, these patients may often go to the closest hospital for unplanned readmissions. This effect is especially true for patients with pancreatic cancer. Yermilov et al,²⁴ in an analysis of pancreaticoduodenectomies in the state of California, found that up to 47% of readmissions were to nonindex hospitals. We are unable to explore this further using the NRD because the distance traveled is not reported in the database.

In addition, causes of higher acuity, such as trauma, strokes, myocardial infarctions, and thromboembolic events, had a much higher proportion of nonindex readmissions than other causes, such as pain or wound complications. This finding has also been reported with readmissions after complex vascular surgery, in which readmissions related to medical comorbidities were more likely to be to nonindex hospitals.²¹ High-acuity conditions often generate a call to an ambulance that will invariably take the patient to the closest hospital. These patients often require early stabilization, and transport to the nearest hospital may be the most appropriate measure at the time. However, patients with procedure-related complications who are stable enough may benefit from transfer to their index hospital. The challenge for health care personnel is to differentiate between the 2 types of conditions to appropriately triage patients to the nearest emergency department vs the index hospital. Postdischarge systems of care may help navigate patients with complications.

Patients discharged to a facility vs home were 22% more likely to be readmitted to a nonindex hospital. Postdischarge care coordination between index hospitals and subacute facilities is another area of potential intervention. Hospitals and facilities should have policies and agreements in place to prevent such a high nonindex readmission rate, especially in the early postoperative period when continuity of care may make a substantial difference.

Our study has several limitations, most of which are inherent to retrospective analysis of large administrative data sets. The NRD, although designed to study readmissions, is derived from administrative data. It does not contain important clinical information, such as cancer stage, intent of surgery, or adjuvant therapy, which may have an influence on our results. In addition, we do not have information on other factors that may influence a patient’s decision to be admitted to an index vs nonindex hospital, such as travel distance, hospital accreditation, committee on cancer designation, or patient functional status. In addition, the NRD only picks up readmissions within the same state; thus, we do not identify patients who were readmitted to out-of-state hospitals. However, these patients are expected to be too few to influence our results. In addition, we include 90-day readmissions from all causes and do not perform separate analysis of high-acuity conditions, such as acute myocardial infarctions, and relatively less acute conditions, such as pain. It is possible that there is no outcome difference by hospital type for high-acuity conditions or even conditions unrelated to the primary procedure, such as trauma. However, the degree of subjectivity involved in categorizing high-acuity and related or unrelated readmissions using administrative data would not provide for accurate analysis. Finally, although we used a robust and widely used method for risk adjustment and our results are similar to those previously reported, it is possible that unaccounted risk may explain the differences in our study. For example, although type of operation is accounted for in the analysis (the extent of the surgery has variation), this along with many other important biological factors cannot be accounted for using an administrative data set.

Our study highlights a large readmission burden after major cancer surgery. Approximately one-fifth of readmitted patients presented to hospitals other than hospitals where their surgery was performed, which was associated with higher morbidity and in-hospital mortality likely because of loss of continuity of care. Interventions targeted at reducing nonindex readmissions and improving care coordination, such as early referrals, interhospital transfers, use of telemedicine technologies, and integration of electronic medical records, may result in improved outcomes for patients readmitted after a complex cancer operation.

Accepted for Publication: January 14, 2018.

Corresponding Author: Nabil Wasif, MD, MPH, Department of Surgery, Mayo Clinic, 5777 E Mayo Blvd, Phoenix, AZ 85054 (wasif.nabil@mayo.edu).

Published Online: April 11, 2018. doi:10.1001/jamasurg.2018.0380

Author Contributions: Drs Zafar and Wasif had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.

Study concept and design: Zafar, Shah, Raoof, Wilson, Wasif.

Acquisition, analysis, or interpretation of data: Zafar, Shah, Channa, Wasif.

Drafting of the manuscript: Zafar, Shah, Channa.

Critical revision of the manuscript for important intellectual content: Zafar, Shah, Raoof, Wilson, Wasif.

Statistical analysis: Zafar, Shah, Raoof, Channa.

Obtained funding: Wasif.

Administrative, technical, or material support: Zafar, Shah, Wasif.

Study supervision: Wilson, Wasif.

Conflict of Interest Disclosures: None reported.