Readmission Patterns After GI Cancer Hospitalizations: The Medical Versus Surgical Patient

Abstract

Purpose:

Readmission within 30 days has been used as a metric for quality of care received at hospitals for certain diagnoses. In the era of accountability, value-based care, and increasing cancer costs, policymakers are looking into cancer readmissions as well. It is important to describe the readmission profile of patients with cancer in the most clinically relevant approach to inform policy and health care delivery that can positively impact patient outcomes.

Patients and Methods:

We conducted a retrospective cohort study using linked Texas Cancer Registry and Medicare claims data. We included elderly Texas residents diagnosed with GI cancer and identified risk factors for unplanned readmission using generalized estimating equations, comparing medical with surgical cancer-related hospitalizations.

Results:

We analyzed 69,693 hospitalizations from 31,736 patients. The unplanned readmission rate was higher after medical hospitalizations than after surgical hospitalizations (21.6% v 13.4%, respectively). Shared risk factors for readmission after medical and surgical hospitalizations included advanced disease stage, high comorbidity index, and emergency room visit and radiation therapy within 30 days before index hospitalization. Several other associated factors and reasons for readmission were noted to be unique to medical or surgical hospitalizations alone.

Conclusion:

Unplanned readmissions among elderly patients with GI cancer are more common after medical hospitalizations compared with surgical hospitalizations. There are shared risk factors and unique risk factors for these hospitalizations that can inform policy, health care delivery, and interventions to reduce readmissions. Other findings underscore the importance of care coordination and comorbidity management in this patient population.

INTRODUCTION

Readmissions within 30 days of discharge are regarded by policy makers as failure of the transition of care process.^1,2 Since 2009, the Centers for Medicare and Medicaid Services (CMS) has mandated the reporting of hospital-level readmission rates for diagnoses such as acute myocardial infarction and pneumonia. Under the Affordable Care Act (ACA), penalties are now being incurred by hospitals with high readmission rates for these encounters.³

Current CMS specifications exclude from the readmission measure, hospitalizations that are primarily for the medical treatment of cancer.⁴ Recently, however, the National Quality Forum has started looking into cancer readmissions for endorsement as a quality measure.⁵ Increasing cancer costs, with a significant portion of the spending allotted toward hospitalizations,⁶ coupled with decreasing trends in readmission rates seen in the post-ACA era, provide the impetus for policymakers to consider this as a quality metric for hospitals caring for patients with cancer.⁷

As the validity of the readmission metric continues to be debated among different stakeholders, various cancer viders have investigated their own data to understand the nature of readmissions within their institutions.^8-10 Several studies have focused on the postsurgical period, which uses readmission outcomes as a quality measure.^11-14 Predictors of readmission have been investigated as well (eg, patient characteristics and some regional and hospital effects), although there is a lack of consistency across studies.^15-17 Age, area of residence, hospital size, and hospital type are examples of factors that are variably associated with readmission.

Clinical providers, hospital administrators, payers, policymakers and patients could all benefit from a better understanding of the readmission profile of patients with cancer. To this end, we conducted a population-based retrospective cohort study to examine the patterns and predictors of unplanned 30-day readmission among elderly patients in Texas who have GI cancer and compared medical patients with surgical patients. We chose to investigate elderly patients with GI cancer because hospitalizations are common in this population and the proportion of cancer care cost attributable to hospitalizations is known to be high for these patients.^6,11,15,18 Investigating readmissions for medical patients and surgical patients separately provides a different but relevant approach from both a clinical and policy perspective and can provide guidance for readmission programs in this patient population.

PATIENTS AND METHODS

The institutional review boards at The University of Texas Medical Branch at Galveston, The University of Texas MD Anderson Cancer Center, and the Texas Department of State Health Services and the Privacy Review Board of the CMS approved this study.

Data Source

We used Texas Cancer Registry (TCR) data linked with Medicare claims data in our study. The TCR is a population-based registry containing statewide data for cancers diagnosed between 1995 and 2013.¹⁹ The Medicare claims database contains utilization files for beneficiaries of the Medicare health insurance plan.²⁰ We used Medicare Provider and Review (MedPAR) files, which contain information on inpatient admissions, as well as outpatient and carrier files. Ninety-eight percent of patients age 65 years or older whose records are entered into TCR are matched with Medicare enrollment and claims files. The TCR-Medicare linkage is performed under the guidance of the National Cancer Institute, the TCR, and CMS, which collect the Medicare claims data. For our study, we examined Medicare claims from 2000 through 2012 for patients diagnosed with GI cancer (ie, esophageal, gastric, liver or intrahepatic ductal, pancreatic, colon, or anorectal cancer) between 2001 and 2010. We reviewed claims for each patient for up to 2 years from the date of initial cancer diagnosis (also the years of active cancer treatment) or until death, whichever came first.

Patients

Texas residents age ≥ 66 years with a first primary diagnosis of GI cancer were included. Patients had uninterrupted Medicare Part A and Part B coverage and no additional coverage through a health maintenance organization.^18,21 We excluded patients with uncommon histologic types of GI cancer and those whose cancer was diagnosed at autopsy or through a death certificate.

Identifying Unplanned Readmissions

We reviewed hospitalizations to acute care hospitals that were classified as short-stay hospitalizations in the Medicare claims database (< 25 days).²² An index hospitalization was considered to have resulted in a readmission if it was followed by an unplanned hospitalization (urgent or emergent admission type) within 30 days of discharge^16,23-25 and if the admission was not primarily for chemotherapy, radiotherapy, or inpatient rehabilitation.⁴ A readmission could be an index hospitalization for a subsequent hospitalization.⁴ We excluded hospitalizations that could not be followed for 30 days after discharge, such as those in which patients died during the index hospitalization or within 30 days of discharge, those in which patients were discharged after November 30, 2012 (because of data set limitations), and those that resulted in transfers to other acute care facilities. These exclusions were applied to give each discharged patient the full 30-day observation period of being at risk for our outcome.

Predictor Variables

To identify predictors of unplanned readmission, we studied the patient and hospitalization characteristics of each inpatient claim. These included age, sex, race and ethnicity, area of residence, census tract poverty level (used as a proxy for socioeconomic status), and state buy-in variable, which was used to determine dual eligibility for Medicare and Medicaid. Clinical characteristics included cancer type, disease stage, and Charlson comorbidity index (CCI). The CCI was determined for each patient by reviewing Medicare claims for the International Classification of Diseases, 9th Revision, Clinical Modification codes pertaining to Charlson comorbidities during the 12-month period before cancer diagnosis.^26-29 Hospitalization characteristics included any radiotherapy, chemotherapy, surgical procedure, emergency room (ER) visit, or outpatient visit within 30 days before an index hospitalization; admission type (unplanned or other); length of stay (LOS); and presence of an intensive care unit stay during the index hospitalization.

We determined the most common International Classification of Diseases, 9th Revision, Clinical Modification diagnoses that were responsible for a readmission and grouped them into more meaningful clinical categories using the Agency for Healthcare Research and Quality Clinical Classifications software.³⁰ We described the most common reasons for readmission on the basis of the principal diagnosis of the hospitalization and using the first coded non–primary cancer diagnosis in the claims data, to account for instances when the primary cancer is coded first if the reason for hospitalization was cancer related (eg, intestinal obstruction as a result of colon cancer).¹⁸ To ascertain potential relatedness of readmission to the index hospitalization, we estimated the proportion of readmission pairs (index and readmitted encounter) with similar diagnosis-related group (DRG).

Statistical Analysis

The hospitalization encounter was considered the unit of analysis. Readmission rates were calculated by obtaining the proportion of index hospitalizations resulting in an unplanned readmission within 30 days of discharge, among the total number of hospitalizations observed in our cohort.² Index hospitalizations were further categorized into medical and surgical hospitalizations. Surgical hospitalizations were identified using the surgery indicator within the MedPAR claims file, augmented by the presence of an operating room charge and an anesthesia charge. All other hospitalizations were considered medical hospitalizations. The median time to readmission and the most common reasons for readmission, as well as factors associated with readmission, were described for each subgroup (ie, surgical hospitalizations and medical hospitalizations).

We performed sensitivity analyses by comparing patients who were excluded as a result of 30-day mortality after discharge with patients who remained in the cohort using univariable analysis. We conducted univariable analysis to describe patients who experienced unplanned readmission. Multivariable logistic regression models using generalized estimating equations were used to fit the unplanned readmission data and identify factors associated with unplanned readmission. To determine whether there was interaction between some of our variables, we stratified the cohort into those with a CCI score of 0 and those with a CCI score ≥ 1 and examined for difference in estimates between strata. The odds ratio (OR) and 95% CI for unplanned readmission for the different predictors were similar in both groups (CCI = 0 and CCI ≥ 1). Differences in estimates observed were not statistically significant (P > .050), thus verifying that there was no significant interaction between CCI and other predictors included in our regression model. Statistical analyses were performed using Statistical Analysis Software version 9.3 for Windows XP (SAS Institute, Cary, NC).

RESULTS

A total of 36,188 patients met our inclusion criteria. After excluding hospitalizations that could not be followed 30 days after discharge, we identified 69,693 index hospitalizations from 31,736 patients in the TCR-Medicare database that made up our final cohort (Appendix Fig A1, online only).

Patient characteristics are listed in Table 1. Briefly, 34.6% of patients were ≥ 80 years old, 48.8% were male, 71.1% were white, 45.1% resided in big metro areas, and 20.1% had dual eligibility for Medicare and Medicaid. The most common cancer in our cohort was colon cancer (51.1%), and the least common was esophageal cancer (4.9%), following the general distribution of GI cancer in the state of Texas.¹⁹ Most patients (49.6%) had a CCI of 0 at the time of diagnosis (Table 1). Sensitivity analyses comparing the characteristics of excluded patients (ie, died within 30 days of discharge) with those of patients who remained in the cohort showed that patients who died within 30 days of discharge were a higher risk group (Data Supplement).

Table 1.

Demographic and Clinical Characteristics of the Study Population and Univariable Analysis of Unplanned Readmission

The overall unplanned readmission rate in our study was 17.8%. The median time to readmission was 11 days. Of the 69,693 index hospitalizations, 37,778 encounters were classified as medical, and 31,915 encounters were classified as surgical. The readmission rate after medical hospitalizations was higher than after surgical hospitalizations (21.6% v 13.4%, respectively; P < .001). After univariable analysis, only age was not associated with having an unplanned readmission (Table 1).

The results of the multivariable analysis are listed in Table 2. After medical hospitalizations, we found that patients who lived in the least affluent neighborhoods had increased odds of readmission (OR, 1.20; 95% CI, 1.09 to 1.31). Pancreatic cancer was associated with the highest odds of readmission compared with colon cancer (OR, 1.62; 95% CI, 1.49 to 1.75), followed by esophageal cancer (OR, 1.44; 95% CI, 1.29 to 1.62), liver or intrahepatic ductal cancer (OR, 1.43; 95% CI, 1.29 to 1.59), and gastric cancer (OR, 1.30; 95% CI, 1.18 to 1.43). Advanced disease stage was associated with increased odds of readmission (P < .001). Having a CCI of 1 increased the odds of readmission by 13% (OR, 1.13; 95% CI, 1.06 to 1.22), with odds increasing as the CCI increased. Radiotherapy and ER visit within 30 days before index hospitalization, unplanned index hospitalization, and LOS ≥ 7 days were all predictors of readmission after a medical hospitalization.

Table 2.

Multivariable Analysis of Risk Factors for Unplanned Readmission After Medical and Surgical Hospitalizations

Predictors of unplanned readmission after surgical hospitalizations differed from those observed after medical hospitalizations. Age ≥ 80 years was associated with an increased odds of readmission after surgical hospitalizations (OR, 1.17; 95% CI, 1.07 to 1.29), whereas age was not a predictor of readmission after medical hospitalizations. Dual eligibility for Medicare and Medicaid increased the odds for readmission (OR, 1.13; 95% CI, 1.03 to 1.25). Recent surgery and intensive care unit stay during the index hospitalization were associated with readmission after surgical hospitalizations (OR, 1.29 [95% CI, 1.11 to 1.49] and OR, 1.15 [95% CI, 1.07 to 1.23], respectively). Advanced stage of disease and increasing CCI were predictors of readmission after surgical hospitalizations, similar to medical hospitalizations (Table 2).

After medical hospitalizations, the top reasons for unplanned readmission were fluid and electrolyte disorders (5.9%), septicemia (5.8%), colon cancer (5.4%), secondary malignancies or metastatic disease (4.9%), and congestive heart failure (4.8%; Table 3). When we considered the first noncancer diagnosis as the primary reason for hospitalization, we found that urinary tract infection, renal failure, and complications of surgical procedures and medical care were among the top reasons for medical readmissions.

Table 3.

Most Common All-Cause and Non–Primary Cancer Reasons for Unplanned Readmission After Medical and Surgical Hospitalizations

The most common reasons for unplanned readmission after surgical hospitalizations were complications of surgical procedures or medical care (15.4%), fluid and electrolyte disorders (5.9%), septicemia (5.6%), intestinal obstruction (5.3%), and pneumonia (3.2%; Table 3). The top reasons for readmission after surgical hospitalizations did not change when we considered the first noncancer diagnosis as the primary reason for hospitalization.

Thirteen percent of medical hospitalizations that resulted in unplanned 30-day readmission had the same DRG as the readmitted hospitalization. Only 3.2% of surgical hospitalizations had the same DRG recorded for the subsequent readmission.

DISCUSSION

Our study provides a detailed description of readmission patterns after surgical and medical hospitalizations. This increases the evidence base upon which future policy recommendations regarding cancer readmissions can be based.

The overall unplanned readmission rate observed in our study was 17.8%. Across different GI cancers, readmission rates range from 5% to 29%.^8,9,12,31,32 The readmission rate after surgical hospitalizations (13.4%) was lower than after medical hospitalizations (21.6%). This trend was also observed in a study looking at related readmissions among patients with cancer admitted to US academic centers (4.2% surgical readmission rate v 6.6% medical readmission rate).¹¹ A Canadian study looking at several cancer types also found a lower surgical readmission rate compared with the medical readmission rate (9.3% v 19.6%, respectively).¹⁵

The median time to readmission in our cohort was 11 days. This information, along with the observed common reasons for readmission, could inform postdischarge interventions. To date, no intervention has been shown to singly and effectively reduce readmissions.³³ Knowing what problems commonly arise and the optimal time to intervene could augment available strategies.

We presented predictors of readmission after medical hospitalizations separately from those after surgical hospitalizations. Shared risk factors after surgical and medical hospitalizations, such as advanced stage of disease, increasing CCI, longer LOS, and an unplanned index hospitalization, have been noted in other studies and are useful for risk stratification and targeting interventions where the highest impact might be expected.^13,15,34,35 Other shared predictors such as radiotherapy and ER visit within 30 days before index hospitalization are novel findings and may represent opportunities for practice improvement at the time of these outpatient visits (eg, closer follow-up after these encounters). Recent chemotherapy was not associated with readmission, and this supports previous findings showing low rates of chemotherapy-related hospitalizations.^21,36 Comorbidity is a consistent predictor of readmission seen across different studies including ours. This finding may have policy implications because it underscores the importance of primary care and subspecialty care involvement during active cancer treatment and through survivorship. Standard health care delivery for this patient population should ideally include a well-coordinated and patient-centered multidisciplinary care team that is able to manage all comorbidities to achieve the best clinical outcomes.³⁷ The association of socioeconomic factors with unplanned readmission implies that these variables should be included in any risk adjustment methodology for readmission measures.³⁸

The most common reason for surgical readmissions was complication of surgical and medical care, similar to other studies.^15,34,39,40 The finding that only 3.2% of surgical readmissions are related on the basis of the same DRG on readmission is inconsistent with this finding. It has been suggested that unplanned readmissions that are related to the original hospitalization should be the focus of hospital improvement efforts impacted by the ACA. Our findings imply that using the DRG to identify an unplanned related cancer readmission may not be accurate.

Similarly, unplanned related medical readmissions on the basis of similar DRG on readmission (13.0%) may not be the best strategy to identify related and potentially avoidable cancer readmissions, because two of the most common reasons for readmission after medical hospitalizations are the primary cancer itself and metastatic disease.

Our study has limitations. First, inherent to any study using administrative data, our assumptions for defining our variables were limited to available data in our data source. As an example, inasmuch as Medicare claims data have an indicator for surgery within the MedPAR files, our definition for a surgical hospitalization had to be augmented by an operating room charge and an anesthesia charge to minimize misclassification of hospitalizations.

Potential bias, however, inevitably remains. In addition, when we examined the reasons for hospitalization, additional reasons for a medical readmission were noted after considering only noncancer diagnoses. This emphasizes the limitations of administrative data as they relate to coding, hence our application of additional strategies to appropriately interpret our findings. Nonetheless, our data source allowed us to study a larger cohort size, investigate variables that are available in most health records, and ascertain readmissions to any health care facility regardless of place of index hospitalization.

Second, we presented aggregate data for six GI cancers. Because this could be a heterogeneous group, we adjusted analyses where appropriate. Third, our study included only Texas residents. Regional variation in readmissions has been described.⁴¹ Our findings may not be generalizable to all patients with GI cancer across the United States. Fourth, using census tract poverty level to approximate socioeconomic status could be subject to ecologic bias because it is an area-level rather than an individual-level variable. However, this variable has been widely used in many studies using claims data and is fairly accepted. Fifth, although the state buy-in variable has been used in numerous health services studies to examine health disparities by dual status, some have challenged its sensitivity with regard to determining Medicare and Medicaid eligibility.⁴² We acknowledge that further validation of this variable may be needed. Finally, patients in our cohort had uninterrupted Medicare coverage and thus had access to health services. Elderly patients with GI cancer who do not have insurance or do not qualify for Medicare may have readmission patterns that are different from our cohort.

In conclusion, our study provides several policy- and practice-relevant implications. Our finding that readmission outcomes for surgical and medical hospitalizations are different suggest that policies concerning cancer readmissions should make a distinction between them. Predictors identified should be used to adjust readmission risk indices for this population. These predictors could also help improve current risk stratification strategies. Common reasons for readmission after medical and surgical hospitalizations were a mix of cancer-related and comorbidity-related diagnoses. This highlights the need to improve comorbidity management and care coordination efforts across different providers during care transition. Finally, although ascertaining preventability of readmissions is outside the scope of this study and remains an unanswered but important question, we do not recommend using DRG to identify avoidable readmissions based on relatedness because this could be prone to misclassification. Policymakers should use our findings as well as those of other investigators to guide their approach on how the readmissions metric might be applied to assess quality of cancer care delivery.

Appendix

Fig A1.

Algorithm for selection of the study cohort. HMO, health maintenance organization.

AUTHOR CONTRIBUTIONS

Conception and design: Joanna-Grace M. Manzano, Linda S. Elting, Marina C. George, Maria E. Suarez-Almazor

Provision of study materials or patients: Linda S. Elting

Collection and assembly of data: Joanna-Grace M. Manzano, Ming Yang, Hui Zhao, Linda S. Elting, Ruili Luo, Maria E. Suarez-Almazor

Data analysis and interpretation: All authors

Manuscript writing: All authors

Final approval of manuscript: All authors

Accountable for all aspects of the work: All authors

AUTHORS' DISCLOSURES OF POTENTIAL CONFLICTS OF INTEREST

Readmission Patterns After GI Cancer Hospitalizations: The Medical Versus Surgical Patient

The following represents disclosure information provided by authors of this manuscript. All relationships are considered compensated. Relationships are self-held unless noted. I = Immediate Family Member, Inst = My Institution. Relationships may not relate to the subject matter of this manuscript. For more information about ASCO's conflict of interest policy, please refer to www.asco.org/rwc or ascopubs.org/jop/site/ifc/journal-policies.html.

Joanna-Grace M. Manzano

No relationship to disclose

Ming Yang

Employment: Genentech

Stock or Other Ownership: Genentech

Hui Zhao

No relationship to disclose

Linda S. Elting

Consulting or Advisory Role: Izun Pharmaceuticals, Galera Therapeutics, Inform Genomics

Marina C. George

Patents, Royalties, Other Intellectual Property: UpToDate (I)

Ruili Luo

No relationship to disclose

Maria E. Suarez-Almazor

Consulting or Advisory Role: Endo Pharmaceuticals, Genentech, Pfizer, Eli Lilly

Research Funding: Pfizer