Readmission following lung cancer resection is associated with a six-fold increase in 90-day postoperative mortality

Abstract

Objectives

Postoperative readmission impacts patient care and healthcare costs. There is a paucity of nation-wide data describing the clinical significance of readmission following thoracic operations. The purpose of this study was to evaluate the relationship between postoperative readmission and mortality following lung cancer resection.

Methods

Data were extracted for lung cancer resection patients from the linked SEER-Medicare registry (2006-2011), including demographics, comorbidities, socioeconomic factors, readmission within 30 days from discharge, and 90-day mortality. Readmitting facility and diagnoses were identified. A hierarchical regression model clustered at the hospital level identified predictors of readmission.

Results

We identified 11,432 lung cancer resection patients discharged alive from 677 hospitals. The median age was 74.5 years and 52% of patients received an open lobectomy. 30-day readmission rate was 12.8%, and 28.3% of readmissions were to facilities that did not perform the original operation. Readmission was associated with a six-fold increase in 90-day mortality (14.4% vs. 2.5%, p <0.001). The most common readmitting diagnoses were respiratory insufficiency, pneumonia, pneumothorax, and cardiac complications. Patient factors associated with readmission included resection type, age, prior induction chemoradiation, preoperative comorbidities including congestive heart failure and COPD, and low regional population density.

Conclusions

Factors associated with early readmission following lung cancer resection include patient comorbidities, type of operation, and socioeconomic factors. Metrics that only report readmissions to the operative provider miss one-fourth of all cases. Importantly, readmitted patients have an increased risk of death and demand maximum attention and optimal care.

Introduction

Lung cancer is the leading cause of cancer death in the United States (1), and resection remains the treatment of choice for appropriate surgical candidates with early stage disease (2). Not only is early postoperative readmission clinically-relevant (3), it is an important predictor of increased resource utilization (4). Beginning in October of 2012, the Affordable Care Act established the Hospital Readmissions Reduction Program, reducing Medicare payments for excess readmissions for acute myocardial infarction, pneumonia, and heart failure (5). Because most early postoperative readmissions are perceived as preventable, incentives to reduce their occurrence are likely forthcoming in the future.

Currently, there is a paucity of national data describing the frequency of postoperative readmissions following lung cancer resection. Data from the American College of Surgeons National Surgical Quality Improvement Program (ACS-NSQIP) reports 30-day readmission following all thoracic operations to be 11.9% (6), but further granularity is necessary to elaborate a pragmatic quality metric. While the development of a risk-adjusted readmission metric for coronary artery bypass surgery is under way through the Society of Thoracic Surgeons (STS) (7), such a process is nonexistent for pulmonary resections. The STS General Thoracic Surgery Database includes postoperative outcomes up to 30 days for member providers, but because the STS is comprised primarily of thoracic surgery specialists (8), the Surveillance Epidemiology and End Results (SEER)-Medicare database may better represent operative experiences nationwide (9).

The primary objectives of this study were to determine the frequency and associated risk factors of early readmission following lung cancer resection, and to assess the impact of readmission on 90-day outcomes. We hypothesized that readmission within 30 days of discharge is associated with an increased risk of subsequent mortality.

Methods

SEER-Medicare Database

The SEER registry is a population-based collection of incident cases, and includes cancer diagnostic, descriptive, and therapeutic information relevant to the time of diagnosis. The National Cancer Institute links the SEER registry to Medicare data for eligible patients to provide comprehensive information on survival, inpatient admissions, outpatient events, and other healthcare claims for 93% of patients 65 years old or older (10). These data accurately account for postoperative readmissions, and are not limited to readmissions taking place at the facilities providing the index operation. The combined SEER-Medicare database encompasses approximately 26% of the population, and provides an opportunity for longitudinal studies broadly generalizable to the Medicare population.

Patient Selection

The 2006 to 2010 SEER-Medicare database was used to identify records for all patients age 66 years or greater with non-small cell lung cancer (NSCLC) of any stage by American Joint Committee on Cancer criteria who received surgical resection (11). Exclusionary criteria included enrollment in a Medicare HMO, lung cancer diagnoses made at autopsy, prior lung cancer diagnoses within one year of index diagnosis, missing date of diagnosis, wedge resection for stage IV disease, and death prior to discharge from the operative admission. To ensure that all patients had adequate pre-surgical records to identify comorbid diseases present at the time of surgery, we additionally excluded patients who were not eligible for Medicare during the three months prior to surgery, or who were diagnosed in 2006.

Demographic information available in the SEER-Medicare data included age, gender, race, and the operative and readmitting facilities. Clinical data included year of operation, final pathologic stage, procedure type and approach, preoperative comorbidities, and readmission diagnoses. To differentiate between readmissions for operative complications and unrelated readmissions, a panel of ICD-9 codes comprised of respiratory, infectious, cardiac, wound, and renal diagnoses were matched to readmission diagnoses recorded in SEER-Medicare (Supplemental table). Comorbidities were identified using the Deyo modification of the Charlson index (12), and were defined using criteria provided by the National Cancer Institute for use with diagnoses reported within inpatient files (MEDPAR), outpatient files (OUTSAF), and physician claims data (NCH) (13, 14). Mortality measures were based on Medicare death certificate records within the SEER-Medicare database. The primary outcomes were readmission within 30 days of discharge from the operative admission and mortality within 90 days of surgery.

Statistical Analyses

Mortality rate within 90 days of surgery was compared between the readmitted and non-readmitted patient groups. Subgroup analyses included mortality rate comparisons based on the readmitting facility (operative vs. non-operative) and readmitting diagnosis (operative complication vs. unrelated). To compare mortality rates between patients who were readmitted and those who were not, a hierarchical generalized linear model for 90-day mortality using previously-reported model predictors in addition to 30-day readmission and restricted to patients who were discharged alive was created (15). A single hierarchical generalized linear model was used to estimate risk of readmission, with adjustments for data clustered by treatment provider. Model predictors were selected a priori based on literature review and frequency of occurrence within our dataset, and included both clinical and socioeconomic risk factors. The statistical significance of each predictor of readmission included in the models was assessed using the F test statistic. Test for provider covariance was performed to determine if risk-adjusted readmission rates varied significantly between operative providers. All outcomes data were analyzed using SAS (version 9.3; SAS Institute, Inc, Cary, NC) and R (version 3.0.2) statistical software (16). The University of Virginia Institutional Review Board for Health Sciences Research approved this study.

Results

For patients diagnosed with NSCLC between 2007 and 2009, SEER-Medicare captured 11,432 patients who were discharged alive following resection and who met all inclusion criteria. The median age was 74.5 years at the time of surgery and most patients presented with stage I disease (70.4%). Roughly half of patients were female (51.5%) and the predominant race was white (89.7%). The most common procedure performed was an open lobectomy (51.6%), and thoracoscopic approaches accounted for 26.8% of resections (Table 1). The most common preoperative comorbidities were chronic pulmonary disease, diabetes, and peripheral vascular disease. Patients included within the final study population were treated at 677 hospitals.

Table 1.

Patient demographics and preoperative comorbidities

Demographics	n (%)
Age
65-69	2653 (23.2%)
70-74	3437 (30.1%)
75-79	3045 (26.6%)
80-84	1751 (15.3%)
85+	546 (4.8%)
Gender
Male	5545 (48.5%)
Female	5887 (51.5%)
Year of Diagnosis
2007	3871 (33.9%)
2008	3823 (33.4%)
2009	3738 (32.7%)
Race
White	10252 (89.7%)
Black	607 (5.3%)
Asian	257 (2.3%)
Hispanic	79 (0.7%)
Other	229 (2.0%)
Tumor Stage
I	7906 (70.4%)
II	1270 (11.3%)
III	1682 (15.0%)
IV	366 (3.3%)
Not available	201 (1.8%)
Operation
Open lobectomy/bilobectomy	5904 (51.6%)
Open wedge resection	1244 (10.9%)
Open segmentectomy	803 (7.0%)
Open pneumonectomy	334 (2.9%)
Chest wall resection with lung	88 (0.8%)
VATS lobectomy	1625 (14.2%)
VATS wedge resection	940 (8.2%)
VATS segmentectomy	494 (4.3%)
Comorbidities
Induction Chemoradiation	563 (4.9%)
Myocardial Infarction	1172 (10.3%)
Heart Failure	1262 (11.0%)
Peripheral Vascular Disease	1743 (15.2%)
Cerebral Vascular Disease	1129 (9.9%)
Chronic Pulmonary Disease	7078 (61.9%)
Diabetes	3476 (30.4%)
Chronic Renal Failure	1012 (8.9%)

The overall readmission rate within 30 days of discharge was 12.8% (1461/11432), and 28.3% (414/1461) of readmissions were to facilities that did not perform the index operation. The median length of stay during postoperative readmissions was 4 days (IQR 2 – 7). Readmissions were more frequent within the first two weeks following discharge (Figure 1). Mortality rate within 90 days of surgery was nearly six-times higher among patients who experienced at least one early readmission (OR 6.6; 14.4%, (210/1461) vs. 2.5%, (249/9971), p < .001). Among readmitted patients, those who underwent two or more readmissions within 60 days of discharge did not have a significantly higher 90-day mortality rate than those readmitted only once (16.2%, 58/358 vs 13.8%, 152/1103, p = .295). In a hierarchical generalized linear model for 90-day mortality, postoperative readmission within 30 days had the largest contribution to predicting mortality risk (OR 5.79, p < .001, F-test statistic 291). The model C-statistic was 0.80 (Table 2).

Figure 1.

Time to readmission after discharge from lung cancer resection. Although early readmissions occur most frequently during the first two weeks following discharge, occurrences persist beyond three months.

Table 2.

Hierarchical regression for risk factors associated with 90-day mortality. Model C-statistic: 0.80

Variable	OR	CI	p-value	F test
30-day Readmission	5.79	4.73 – 7.09	<0.001	290.7
Procedure Type			<0.001	13.3
Pneumonectomy	3.97	2.41 – 6.56
Chest wall resection with lung	6.18	3.08 – 12.4
Lobectomy/bilobectomy	1.06	0.72 – 1.55
Segmentectomy	1.50	0.92 – 2.43
Wedge resection	158	1.02 – 2.47
VATS lobectomy	0.74	0.46 – 1.18
VATS segmentectomy	0.62	0.31 – 1.26
VATS wedge resection	REF
Surgical Year			0.028	3.0
2007	0.60	0.36 – 0.99
2008	0.75	0.46 – 1.23
2009	0.87	0.53 – 1.40
2010	REF
Age			0.001	4.7
85+	2.28	1.49 – 3.49
80-84	1.67	1.20 – 2.32
75-79	1.29	0.96 – 1.74
70-74	1.22	0.91 – 1.64
65-69	REF
Gender			<0.001	29.6
Female	0.56	0.46 – 0.69
Male	REF
Race			0.263	1.3
Asian	0.39	0.14 – 1.07
Black	0.94	0.60 – 1.49
Other	0.75	0.39 – 1.47
White	REF
Comorbidity
Induction Chemoradiation	1.44	0.99 – 2.10	0.055	3.7
Acute MI	0.85	0.62 – 1.17	0.329	1.0
CHF	1.95	1.52 – 2.51	<0.001	27.1
PVD	1.07	0.83 – 1.39	0.604	0.3
Cerebrovascular Disease	1.10	0.82 – 1.49	0.532	0.4
COPD	1.27	1.02 – 1.58	0.035	4.4
Diabetes	1.07	0.86 – 1.34	0.532	0.4
Renal Failure	1.15	0.86 – 1.56	0.347	0.9

Ninety-day mortality did not differ between patients readmitted to the operative facility versus those readmitted to an alternate facility (13.6% vs. 16.4% p = .16). However, of patients initially readmitted to a non-operative facility, 11.8% (49/414) were transferred before discharge, compared to 0.86% of patients readmitted to the operative facility (9/1047, p < .001). Readmitted patients who were subsequently transferred had a higher 90-day mortality rate than those who were not (25.9%, 15/58 vs 13.9%, 195/1403, p = .011), suggesting that transfers of care were likely related to clinical acuity. Based on primary admitting diagnoses, 54.1% (791/1461) of readmissions were directly related to postoperative complications, the most common of which were respiratory insufficiency (24.0%, 190/791), pneumonia (16.7%, 132/791), cardiac complications (14.8%, 117/791) and pneumothorax (13.7%, 108/791, Figure 2). Mortality at 90 days did not differ significantly between patients readmitted due to operative complications and those readmitted with unrelated primary diagnoses (OR 1.22, p = .21).

Figure 2.

Distribution of primary readmitting diagnoses over time. Operative complications comprised 54.1% (791/1461) of all readmissions (left). Among pulmonary operative complications, the most common readmitting diagnoses were respiratory insufficiency, pneumonia, and pneumothorax (right).

Hierarchical generalized regression models were used to estimate the risk of readmission within 30 days of discharge. Hospital provider was included in the models as a random effect, accounting for the clustering of procedures within hospitals. The C-statistic for the predictive accuracy of the model was 0.604. Predictors of early readmission with the highest F-statistic in descending order were COPD, congestive heart failure (CHF), prior induction chemoradiation, recent myocardial infarction, and chronic renal failure (Table 3). Of the procedure types evaluated, pneumonectomy and combined lung and chest wall resection had the highest odds ratios of early readmission. Interestingly, when compared to video-assisted thoracoscopic (VATS) wedge resections, more extensive VATS resections were associated with a lower 30-day readmission rate. Among socioeconomic factors, regional population density was associated with readmission, while regional high school non-graduation rate trended toward an association. The test for provider covariance was not significant (δ² =0.085, p= 0.41), suggesting that, after adjusting for clinical and socioeconomic risk factors, there was no evidence of significant differences in readmission rates across hospitals.

Table 3.

Clinical risk factors for 30 day readmission. Model C-statistic: 0.604.

Variable	OR	CI	p-value	F test
Procedure Type			0.018	2.4
Pneumonectomy	1.21	0.85 – 1.72
Chest wall resection with lung	1.15	0.62 – 2.13
Lobectomy/bilobectomy	0.82	0.66 – 1.01
Segmentectomy	0.96	0.72 – 1.27
Wedge resection	0.78	0.60 – 1.01
VATS lobectomy	0.74	0.58 – 0.95
VATS segmentectomy	0.69	0.49 – 0.98
VATS wedge resection	REF
Surgical Year			0.584	0.7
2007	1.08	0.78 – 1.49
2008	1.17	0.85 – 1.61
2009	1.14	0.83 – 1.57
2010	REF
Age			0.025	2.8
85+	1.47	1.11 – 1.94
80-84	1.22	1.00 – 1.48
75-79	1.21	1.03 – 1.44
70-74	1.07	0.91 – 1.26
65-69	REF
Gender			0.049	3.9
Female	0.88	0.78 – 1.00
Male	REF
Race			0.190	1.6
Asian	0.80	0.52 – 1.24
Black	0.75	0.57 – 1.01
Other	0.88	0.60 – 1.28
White	REF
Comorbidity
Induction Chemoradiation	1.52	1.19 – 1.93	<0.001	11.5
Acute myocardial infarction	1.25	1.01 – 1.50	0.015	5.9
Congestive Heart Failure	1.56	1.32 – 1.83	<0.001	27.9
Peripheral Vascular Disease	1.14	0.98 – 1.34	0.093	2.8
Cerebrovascular Disease	1.18	0.99 – 1.42	0.070	3.3
COPD	1.47	1.29 – 1.67	<0.001	34.3
Diabetes	1.15	1.01 – 1.31	0.035	4.5
Renal Failure	1.25	1.04 – 1.51	0.018	5.6
Married	0.96	0.85 – 1.09	0.530	0.4
Regional Median Income			0.570	0.7
Q1 (lowest income)	0.89	0.67 – 1.18
Q2	0.95	0.74 – 1.22
Q3	1.06	0.85 – 1.32
Q4	1.05	0.86 – 1.29
Q5 (highest income)	REF
Regional Population Density			0.032	2.7
Q1 (least dense)	0.92	0.75 – 1.13
Q2	1.15	0.95 – 1.39
Q3	1.14	0.94 – 1.38
Q4	1.24	1.03 – 1.50
Q5 (most dense)	REF
Regional High School Non-Graduation			0.067	2.2
Q1 (fewest non-graduates)	0.79	0.60 – 1.03
Q2	0.73	0.57 – 0.93
Q3	0.91	0.73 – 1.13
Q4	0.97	0.79 – 1.18
Q5 (most non-graduates)	REF

OR: odds ratio CI: 95% confidence interval REF: reference value Q1-5: quintiles 1-5 VATS: Video-assisted thoracic surgery COPD: chronic obstructive pulmonary disease

Discussion

The present study is the first to report the relationship between early readmission and postoperative mortality within a national lung cancer resection database. Our results suggest that readmission within 30 days of discharge is common, and is substantially underestimated by metrics which only report readmissions to the operative facility. Even more important, early readmission following lung cancer resection is associated with a six-fold increase in the risk of death. Our results highlight the acuity of patients readmitted following lung resections, and argue for intensified resource allocation and clinical awareness. When presented with these patients, clinicians should maintain a low threshold for admission to intensive care units and transfer to tertiary care centers.

Our study using SEER-Medicare data encompasses a large, diverse group of hospitals. Within this group, overall 30-day readmission rate following lung cancer resection was 12.8%. This finding is in concordance with outcomes derived from 1992-2002 SEER-Medicare data showing a 15% 30-day readmission rate (4), suggesting that risk of postoperative readmission in this population has not changed significantly over the last fifteen years. Freeman et al recently reported a 7% operation-related, 90-day readmission rate and an 8.3% overall readmission rate within a high-volume, multi-institutional database (17). Similarly, Varela and colleagues reported an emergency 30-day readmission rate of 6.9% (18). Within our dataset, a higher overall readmission rate may be related in part to the older median age of the Medicare population; this is reflected by the high rate of readmissions which were unrelated to the index operation (45.9%). Isolating readmissions related to postoperative complications revealed an occurrence rate of 6.9%, which is in concordance with prior studies. A higher rate of readmissions unrelated to the index operation within our dataset may also be due to inclusion of more low-volume centers in areas with less abundant outpatient resources. However, testing for provider covariance within our dataset showed that risk-adjusted readmission rates did not differ significantly between hospitals. This finding likely reflects the wide range of provider case volumes, which greatly reduces the utility of readmission rate as a differentiating quality metric.

Our analyses demonstrate that 30-day readmission following lung cancer resection is the strongest risk factor for 90-day mortality, out-performing all commonly-reported preoperative predictors. Furthermore, this association was independent of the readmitting facility (operative vs. non-operative) and the readmission category (operative complication vs. unrelated). Readmitted patients who were transferred to another facility had a higher 90-day mortality rate than those who were not. Rather than suggesting a causative effect, we believe this association is reflective of the appropriate transfer of high-acuity patients to tertiary care facilities. Single- and multi-institutional data have previously shown a significant decrease in long-term survival among lung resection patients experiencing early readmission (4, 19). Our results indicate that this effect is even more critical over the initial months following surgery. While mortality two to five years after resection may not be regarded as surgically-related or preventable, death within 90 days from surgery is more likely to be attributable to the perioperative experience, and may offer opportunities for early intervention.

Hierarchical regression demonstrated that socioeconomic factors such as rate of regional high school non-graduation rate and population density may be related to likelihood of postoperative readmission. A recent study within the cardiac surgery community also identified lower education level as a significant risk factor for postoperative readmission (20). Although similar results in general thoracic surgery have borne out in a multi-institutional European study relating regional deprivation scores to readmission (21), a link between socioeconomic factors and readmission after lung resection has not previously been demonstrated in the United States. Due to protocols protecting patient confidentiality, a direct analysis of participant census data is not available through SEER-Medicare, thus socioeconomic surrogate covariates were chosen. The association between regional socioeconomic factors and risk of readmission highlights some of the difficulties with patient education and outpatient care access, and advocates for further development of healthcare infrastructure in disadvantaged regions.

Among preexisting clinical risk factors, COPD, CHF, induction chemoradiation, recent myocardial infarction, renal failure, and age were all significantly associated with rate of readmission. Not surprisingly, pneumonectomy was the operation associated with the highest risk of postoperative readmission. These results are in concordance with most published series (4, 17, 19, 21), and expand upon existing evidence through utilizing a large national database and implementing a hierarchical multiple regression model that adjusts for the provider. Awareness of the most important comorbid risk factors may assist with postoperative management and discharge planning. Among patients who are at high risk for readmission, early post-discharge telephone calls to assess pain control, home oxygen requirements, and adequate performance of activities of daily living may be a simple and inexpensive means to reduce the likelihood of unnecessary readmission (22). Such a protocol may also allow early identification of warning signs to promote proactive intervention when indicated. Our finding that more extensive VATS resections are associated with a reduced readmission rate compared to VATS wedge resections may be reflective of the fact that more extensive resections tend to be performed on patients with better preoperative functional reserve (23). Of note, VATS resections only constituted 26.7% of the lung resections within our dataset. Minimally-invasive procedures may comprise a higher proportion of the operative volume within specialized thoracic centers such as those represented by the STS-GTSD (24); the frequency and predictors of postoperative readmissions in this setting deserve further, focused study.

There are several limitations to this study. First, the SEER-Medicare database excludes patients under 65 years of age. Thus, the median age of our dataset is older than the general lung cancer resection population, and may lend selection bias to both the rate of readmission and the rate of 90-day mortality. However, the median age of patients included in this study is only slightly older than the median age at diagnosis for non-small cell lung cancer nationally (70 years) (25). Furthermore, the average age of lung cancer resection patients included in the STS database is 67 years, which falls within the Medicare population (24). Second, despite accounting for seventeen variables, our hierarchical model for early readmission achieved a C-statistic of only 0.604, indicating that the model had only modest ability to predict future readmissions. Due to the presence of many interrelated and often immeasurable factors which contribute to postoperative readmissions, this outcome has historically been difficult to predict (26). For example, an existing 18-variable readmission model used in a large, single-institution study of cardiac surgery patients had a C-statistic of only 0.7 (27). Similarly, subspecialty-specific ACS-NSQIP prediction models for readmission also perform modestly, with C-statistics between 0.65 and 0.71, despite incorporating a number of relevant perioperative clinical data (6). Because the SEER-Medicare database does not enable separation between preoperative and postoperative diagnosis codes within index operative admissions, we were unable to reliably account for postoperative complications within our predictive models. The moderate C-statistic notwithstanding, the present hierarchical multivariate regression model augments existing univariate analyses within the thoracic literature and highlights preoperative factors that warrant increased awareness of readmission risk.

Conclusions

Readmissions within 30 days of discharge following lung cancer resection are common, and are underestimated by metrics that focus only on the operating facility. Factors that increase the risk of readmission include preoperative comorbidities, procedure type and socioeconomic factors. Although accurately predicting a future readmission in the preoperative setting is difficult, consideration of relevant risk factors should be a requisite component of discharge planning. Most importantly, patients who present with an early readmission are at high risk of subsequent death and demand maximum attention and care.