Assessment of short readmissions following elective pulmonary lobectomy

Abstract

Background:

Reducing readmissions is critical for improving patient care and lowering costs. Despite this, few studies have assessed length of readmission following pulmonary lobectomy.

Methods:

Using the Healthcare Cost and Utilization Project New York State Inpatient Database, we identified adult patients undergoing elective pulmonary lobectomy (2007–2015) and assessed readmission within 30 days of hospital discharge. We analyzed the relationship between length of readmission and post-operative morbidity and mortality as well as primary diagnoses at readmission.

Results:

Of 19947 included patients, 2173 (10.9%) were readmitted within 30 days. The median (IQR) length of readmission was 5 (2–8) days. Longer length of readmission was associated with significantly higher likelihood of major complication (for every 1-day increase, aOR=1.14, 95% CI=1.12–1.17, p<0.001) and mortality (aOR=1.03, 95% CI=1.02–1.04, p<0.001) within 90 days. Primary diagnosis codes at readmission differed significantly with length of readmission.

Conclusions:

Interventions that target short readmissions may help to prevent a proportion of readmissions following elective lung resection.

Introduction

Reducing readmissions after pulmonary lobectomy is a growing priority for thoracic surgeons. A major driver of this is the increased emphasis that the Centers for Medicare and Medicaid Services (CMS) and other agencies have placed on readmissions as a quality metric¹, even going as far as penalizing institutions with high readmission rates through decreased reimbursement². Pulmonary lobectomy is one of the most common procedures in thoracic surgery³. Since it is often performed on elderly, high-risk patients with significant comorbidities in the semi-urgent setting of newly diagnosed malignancy⁴, it is unsurprising that readmissions after lobectomy are relatively common, occurring in 5–15% of cases⁵. However, readmissions are not just a statistic with reimbursement consequences; rather, readmissions often carry significant medical consequences and are associated with greater risk of short- and long-term mortality as well^5,6.

Despite clear importance, efforts to predict and more importantly reduce readmissions have been challenging. Several studies using various tumor registry and administrative databases have attempted to quantify predictors of readmission following lobectomy^4–10. The most resonant conclusion from these studies is that readmissions are associated with a complex mixture of patient-, treatment-, and hospital-specific factors that are often difficult to modify. However, implicit to any discussion of “preventing” readmissions (and CMS’ practices of penalized reimbursements) is that a proportion of readmissions must in fact be preventable. For example, patients readmitted with sepsis are fundamentally different than patients readmitted with mild pain. Additionally, factors like being readmitted to a non-index facility or being readmitted for very short hospital stays may be markers of potentially preventable readmissions or associated with less severe readmission diagnoses.

We hypothesized that a subset of short readmissions (i.e., short lengths of stay at readmission) may be associated with less severe causes (i.e., readmission diagnoses). In this study of patients undergoing elective pulmonary lobectomy, we examined the relationship between readmission length of stay and various outcomes. We also assessed the relationship between readmission length of stay and primary diagnosis codes at readmission, hypothesizing that short readmissions may be related to less severe diagnoses.

Patients and Methods

Study Design and Data Sources

We performed this retrospective cohort study using data from the Healthcare Cost and Utilization Project (HCUP) New York State Inpatient Databases (SID) through the Agency for Healthcare Research and Quality (AHRQ). The HCUP SID is an all-payer administrative database that encompasses almost 97 percent of all hospital discharges¹¹. The dataset provides over 100 variables including diagnosis codes, procedure codes, admission and discharge dates, patient demographics, hospital characteristics, and cost information for all inpatient discharges in New York hospitals. The datasets also assigns unique identifiers to every patient, which allows researchers to longitudinally assess for readmissions (including to other non-index facilities in the same state)¹². The study protocol was reviewed by the Washington University School of Medicine Human Research Protection Office and was deemed exempt given the de-identified nature of the database.

Patient Population

We included all patients undergoing elective pulmonary lobectomy between 2007 and 2015 using International Classification of Diseases, Ninth Edition, Clinical Modification (ICD-9-CM) codes for both open (32.49, 32.4) and thoracoscopic (32.41, 34.21) lobectomy. Robotic lobectomies (17.xx) remained relatively rare during the study period and were analyzed in the thoracoscopic group¹². Exclusion criteria were patients who lived out-of-state (i.e., unlikely to be readmitted in the state of New York), age <18 years old, emergent lobectomy (34.03), mortality during index hospitalization (i.e., unable to be readmitted), and non-elective operations (defined as lobectomies that occurred >3 days after admission to the index hospital).

Variable Definitions and Exposures

We extracted demographic and treatment-related variables including age, sex, race, median income (by residential zip code), insurance status, and comorbidities. Comorbidities were assessed using ICD-9-CM codes within 1 year prior and including the index admission to generate an Elixhauser comorbidity index¹³. Hospital characteristics were extracted from the American Hospital Association Annual Survey (Health Forum, Chicago, IL), including hospital size and academic teaching status. Unique hospital identifiers were also used to determine if patients were readmitted to the same index hospital or to a different non-index hospital. Major and minor complications following surgery were determined using ICD-9-CM diagnosis and procedure codes (available in Supplemental Table 1). Major complications included myocardial infarction, acute respiratory insufficiency/tracheostomy, pulmonary embolism, cerebrovascular accident, hemorrhage, empyema, and sepsis or septicemia¹². Other complications included supraventricular arrhythmia/atrial fibrillation, pneumothorax, postoperative air leak, pulmonary edema, pulmonary collapse, pneumonia, deep vein thrombosis, urinary tract infection, and wound infection. To distinguish post-operative complications from chronic conditions, we used present-on-admission (DXPOAn) codes⁴.

Outcomes

Our primary outcome of interest was hospital readmission within 30 days of index-hospital discharge. We chose date of discharge (as opposed to surgery) to ensure that all patients had equal amount of follow-up regardless of the index-hospitalization length of stay. We identified readmissions by using the unique patient identifier numbers to link all discharge records for each patient. After linking these records, we then calculated time to readmission and readmission length of stay. We also evaluated the relationship between readmission length of stay and several outcomes, including in-hospital mortality (within 30 and 90 days), any complication (within 30 and 90 days), and major complication (within 30 and 90 days).

Reasons for Readmission

To investigate our hypothesis regarding the relationship between reasons for readmission and readmission length of stay, we dichotomized length of stay into “short” and “long” groups. We defined “short readmissions” as ≤3 days and “long readmissions” >3 days, based on the mode of the distribution so that groups were relatively equal in size. We also assessed several different timepoint cut-offs in sensitivity analyses. We then compared reasons for readmission between the short and long readmission groups by manually searching primary diagnosis ICD-9-CM codes (DX1) from the discharge records of each unique readmission and grouping these codes into categories of readmission.

Statistical Analysis

Continuous variables were presented as means (standard deviation, SD) with Kruskal-Wallis tests. Categorical variables were presented as absolute numbers (percent) with χ² test statistics. Factors associated with readmission were assessed using multivariable hierarchical logistic regression models with clustering at the hospital-level (using the GENMOD function in SAS), controlling for age, sex, income quartile, insurance status, year, hospital size, hospital teaching affiliation, comorbidities, and surgical approach. The associations between readmission length of stay (continuous variable) and mortality, major complication, and any complication (within 30 and 90 days) were assessed in separate multivariable logistic regression models controlling for the same aforementioned covariates. Missing variables were reported in the descriptive analyses and separate unknown variables were used in the multivariable models. P-values of less than 0.05 were considered statistically significant and all P-values were 2-tailed. Data analyses were performed using SAS version 9.4 (SAS Institute, Cary, NC).

Results

Patient demographics

The study included 19947 patients undergoing pulmonary lobectomy who met inclusion criteria (Figure 1). The mean (SD) patient age was 66.1 (11.1) years old. A majority of patients were female (n=10971 [55.0%]) and white race (n=15210 [76.3%]). Most patients were insured through Medicare (n=10846 [54.4%]) or private (n=7140 [35.8%) plans. The most common comorbidities were hypertension (n=11803 [59.2%]), chronic lung disease (n=9130 [45.8%]), and diabetes mellitus (n=3638 [18.2%]). The most common surgical approach was open thoracotomy (n=10349 [51.9%]). Additional patient demographics are shown in Table 1. The rate of any complication and major complications within 30 days was 38.1% (n=7604) and 9.9% (n=1971). Specific post-operative complications are shown in Table 2.

Figure 1.

Study consort diagram

Table 1.

Study cohort patient demographics and hospital characteristics.

	Overall cohort		Readmission within 30 days
Variable	N=19947	%	Not readmitted N=17774	%	Readmitted N=2173	%	p-value
Age, y	66.1 ± 11.1	n/a	66.0 ± 11.1	n/a	67.4 ± 11.4	n/a	<0.001
Female	10971	55.00	9921	55.82	1050	48.32	<0.001
Race							0.60
White	15210	76.25	13557	76.27	1653	76.07
Black	1474	7.39	1299	7.31	175	8.05
Other	3147	15.78	2815	15.84	332	15.28
Unknown	116	0.58	103	0.58	13	0.60
Comorbidity
CHF	890	4.46	728	4.10	162	7.46	<0.001
Valve disease	1265	6.34	1092	6.14	173	7.96	0.001
PVD	1598	8.01	1344	7.56	254	11.69	<0.001
COPD	9130	45.77	7908	44.49	1222	56.24	<0.001
DM	3638	18.24	3172	17.85	466	21.45	<0.001
HTN	11803	59.17	10437	58.72	1366	62.86	<0.001
Renal disease	972	4.87	789	4.44	183	8.42	<0.001
Liver disease	411	2.06	355	2.00	56	2.58	0.07
Obesity	1503	7.53	1312	7.38	191	8.79	0.02
Weight loss	571	2.86	462	2.60	109	5.02	<0.001
Alcohol abuse	698	3.50	593	3.34	105	4.83	<0.001
Drug use	353	1.77	303	1.70	50	2.30	0.05
Median income							0.43
Q < 25%	3329	16.69	2948	16.59	381	17.53
Q 25%-50%	4583	22.98	4067	22.88	516	23.75
Q 51%-75%	5240	26.27	4699	26.44	541	24.90
Q > 75%	6385	32.01	5698	32.06	687	31.62
Unknown	410	2.06	362	2.04	48	2.21
Insurance status							<0.001
Medicare	10846	54.37	9538	53.66	1308	60.19
Private	7140	35.79	6457	36.33	683	31.43
Medicaid	1520	7.62	1372	7.72	148	6.81
Other	441	2.21	407	2.29	34	1.56
Academic hospital	16386	82.15	14624	82.28	1762	81.09	0.17
Hospital size							0.02
<300 beds	3762	18.86	3303	18.58	459	21.12
300–499 beds	6626	33.22	5898	33.18	728	33.50
500–799 beds	4749	23.81	4252	23.92	497	22.87
>800 beds	4810	24.11	4321	24.31	489	22.50
Surgical approach							<0.001
Thoracoscopic	9598	48.12	8659	48.72	939	43.21
Open	10349	51.88	9115	51.28	1234	56.79

CHF=congestive heart failure; PAH=pulmonary circulatory disease; PVD=peripheral vascular disease; COPD=chronic lung disease; DM=diabetes mellitus; HTN=hypertension

Table 2.

Post-operative outcomes and complications following pulmonary lobectomy.

	Overall cohort		Readmission within 30 days
Variable	N=19947	%	Not readmitted N=17774	%	Readmitted N=2173	%	p-value
Length of stay
Median (IQR), d	5 (4–7)	n/a	5 (4–7)	n/a	6 (4–9)	n/a	<0.001
Prolonged (≥14 days)	1311	6.57	1034	5.76	287	13.21	<0.001
Any complication (30 days)	7604	38.12	5961	33.54	1643	75.61	<0.001
Major complication (30 days)	1971	9.88	1278	7.19	693	31.89	<0.001
MI	102	0.51	50	0.28	52	2.39	<0.001
CVA	82	0.41	35	0.20	47	2.16	<0.001
ARI	1124	5.63	805	4.53	319	14.68	<0.001
PE	177	0.89	62	0.35	115	5.29	<0.001
Hemorrhage	460	2.31	367	2.06	93	4.28	<0.001
Empyema	255	1.28	77	0.43	178	8.19	<0.001
Sepsis or septicemia	407	2.04	195	1.10	212	9.76	<0.001
Minor complication (30 days)
AF/SVA	2359	11.83	1775	9.99	584	26.88	<0.001
Air leak	779	3.91	673	3.79	106	4.88	0.01
Pneumonia	1282	6.43	751	4.23	531	24.44	<0.001
Pneumothorax	2435	12.21	1952	10.98	483	22.23	<0.001
Pulmonary edema	50	0.25	35	0.20	15	0.69	<0.001
Pulmonary collapse	1879	9.42	1530	8.61	349	16.06	<0.001
Wound infection	200	1.00	74	0.42	126	5.80	<0.001
DVT	205	1.03	66	0.37	139	6.40	<0.001
UTI	559	2.80	324	1.82	235	10.81	<0.001

Readmission

Readmission within 30 days of hospital discharge was identified in 2173 (10.9%) patients. The median [IQR] time between hospital discharge and readmission was 10 (4–18) days. On multivariable analysis, factors associated with higher likelihood of readmission included history of CHF (adjusted odds ratio [aOR] 1.32, 95% CI 1.11–1.56, p=0.001), COPD (aOR 1.47, 95% CI 1.33–1.62, p<0.001), and renal disease (aOR 1.55, 95% CI 1.27–1.90, p<0.001) (Supplemental Table 2). Factors associated with lower likelihood of readmission included female sex (aOR 0.80, 95% CI 0.74–0.87, p<0.001), larger hospital size (>800 beds vs <300 beds, aOR 0.86, 95% CI 0.74–0.99, p=0.04), more recent surgical year (2011–2015 vs 2007–2010, aOR 0.84, 95% CI 0.76–0.92, p<0.001), and minimally invasive surgical approach (aOR 0.86, 95% CI 0.78–0.94, p<0.001).

Length of Readmission

The median (IQR) length of readmission was 5 (2–8) days (Figure 2). Longer length of readmission was associated with significantly higher likelihood of 30-day complication (for every 1-day increase, aOR 1.08, 95% CI 1.06–1.11, p<0.001, Table 3), 30-day major complication (aOR 1.12, 95% CI 1.1–1.14, p<0.001), 90-day complication (aOR 1.14, 95% CI 1.10–1.17, p<0.001), 90-day major complication (aOR 1.14, 95% CI 1.12–1.17, p<0.001), and 90-day mortality (aOR 1.03, 95% CI 1.02–1.04, p<0.001). However, readmission length of stay was not associated with 30-day mortality (aOR 0.96, 95% CI 0.90–1.02, p=0.16).

Figure 2.

Distribution of readmission length of stay following pulmonary lobectomy

Table 3.

Association between readmission length of stay and various outcomes

	30-day			90-day
Outcome	aOR	95% CI	P-value	aOR	95% CI	P-value
Complication	1.08	1.06–1.11	<0.001	1.14	1.10–1.17	<0.001
Major complication	1.12	1.10–1.14	<0.001	1.14	1.12–1.17	<0.001
Mortality	0.96	0.90–1.02	0.16	1.03	1.02–1.04	<0.001

All models adjusting for age, sex, income, insurance status, year, academic hospital status, hospital size, non-index readmission, and history of CHF, valve disease, pulmonary circulatory disease, PVD, COPD, DM, HTN, renal disease, liver disease, obesity, weight loss, alcohol abuse, and drug use.

Reasons for Readmission

To assess the relationship between length of readmission and primary diagnoses at readmission (i.e., reason for readmission), we dichotomized length of readmission into “short” (≤3 days) versus “long” groups (>3 days, based the mode in Figure 2). Of the readmitted patients, 826 (38.0%) had a short readmission. The most common diagnosis at readmission in the entire cohort was pneumonia (8.7%). Reasons for readmission differed significantly between groups (p<0.001, Figure 3). Among individuals with short readmissions, the most common primary diagnoses were pneumothorax/pulmonary collapse (7.5%), pleural effusion (6.5%), and chronic obstructive pulmonary disease (5.7%). Among individuals with long readmissions, the most common primary diagnoses were pneumonia (10.5%), pneumothorax/pulmonary collapse (9.7%), acute respiratory failure (6.5%), and wound infection (6.4%).

Figure 3.

Primary diagnoses at readmission comparing short (≤3 days) and long (>3 days) readmissions

On multivariable analysis, factors associated with short readmission included insurance status (Medicaid vs Medicare, aOR 1.57, 95% CI 1.07–2.30, p=0.02) and being readmitted to a non-index facility (aOR 1.52, 95% CI 1.29–1.78, p<0.001) (Supplementary Table 3).

Comment

Readmissions have been proposed as a quality metric to assess hospital performance following common operations like lobectomy. Inherent to this discussion is that hospitals and providers have some opportunity to control (and hopefully improve) their readmission rates, including by preventing a certain subset of readmissions. Our study explores short readmissions as a possibly preventable occurrence after lobectomy, especially those related to less severe diagnoses at readmission. We found that the most strongly associated factor with short readmission was presenting to a non-index facility (i.e., different hospital) for readmission and type of insurance coverage. Despite this, short readmissions were associated with a wide variety of primary diagnosis codes, highlighting the complexity of predicting or preventing these readmissions. In general, these findings should inform providers on a subset of readmissions that may deserve higher scrutiny, since addressing readmissions should be a high priority of thoracic surgeons.

“Non-index readmissions” are a well-studied event particularly in the “failure-to-rescue” and “fragmentation of care” literature¹⁴. Non-index readmissions occur when a patient is readmitted to a facility other than where the surgery was performed. Such readmissions are relatively common following major cancer operations (occurring in 10%–47% of cases)^15,16. Our study adds to these previous findings by showing that non-index readmissions following pulmonary resection are often associated with short length of stay and, in turn, less severe outcomes. Several factors may contribute to this observation. First, non-index facilities are often staffed by non-surgical providers or surgeons who may be inexperienced with the specifics of pulmonary lobectomy and therefore may have a lower threshold for readmission. Second, non-treating providers may be less familiar with typical post-operative findings on imaging or physical exam. For example, routine atelectasis on chest x-ray may be confused for pneumonia or typical wound erythema may be confused for wound infection. Finally, while we did not examine long-term outcomes, it is also important to note that other types of non-index treatment, like fragmentation of post-operative oncologic care (i.e., receiving adjuvant therapy through a different facility than where surgery was performed), have been associated with worse outcomes¹⁷. Further efforts to address fragmentation of care, including non-index readmissions, are important given the increasing regionalization of complex surgeries.

Our study also supplies several patient- and system-specific factors that may be targeted to reduce readmissions, particularly short readmissions. Obviously, the length of readmission is an impossible factor to know at the time of presentation, so other factors need to be targeted. First, individuals with a high likelihood of non-index readmission (like those who live further from the index hospital) may warrant closer follow-up. Post-operative discharge instructions could also specify who to contact if the patient experiences alarming symptoms. Second, enhanced recovery after surgery (ERAS) protocols should be followed as precisely as possible in order to reduce readmissions¹⁸. For example, we found that minimally invasive approach was strongly protective against readmissions. Third, comorbidities should be comprehensively addressed prior to surgery, as we found that these were highly predictive of readmission. The pre-operative period should be used to allow for multidisciplinary optimization, especially in heavily comorbid patients^19,20. Additionally, such patients should be followed closely in the post-operative period given their elevated readmission risk. Fourth, while we did not examine surgical volume metrics, larger hospital size was associated with lower risk of readmission. Conversely, while prior studies have demonstrated a relationship between academic affiliation and risk of readmission, we did not observe one in this study⁸. Nonetheless, additional efforts to characterize readmissions in low-volume, non-academic centers are likely needed to reduce readmissions in this unique practice environment, and vice versa^21,22. Finally, it is notable that we did not observe certain diagnoses in our analysis. In particular, we did not observe a large number of patients being readmitted with primary diagnoses related to uncontrolled pain. While it is possible that such events were under-coded, this further emphasizes the need for de-implementing the over-prescription of opiate narcotics, including in the setting of lobectomy^23,24. Addressing several of these factors together through quality improvement endeavors may help to reduce readmissions following lobectomy²⁵.

Several other studies have assessed readmissions following lobectomy, but few have explicitly examined the length of readmission. Brown and colleagues performed an analysis of the Society of Thoracic Surgery General Thoracic Surgery Database of adults undergoing elective lobectomy for lung cancer¹⁰. With a comparable readmission rate (8.2%), they found that readmission was most strongly associated with major post-operative complications (like pulmonary embolism and empyema). Our study adds to this finding by showing that major complications after lobectomy typically result in prolonged readmission (as expected), whereas short readmissions are typically associated with less severe primary diagnosis codes. Stiles and colleagues performed an analysis using administrative claims data, demonstrating that the most common diagnoses at readmission were due to pulmonary or cardiac etiologies⁴. Our study supports these findings but also shows that such readmissions typically have longer lengths of stay with less ideal outcomes.

This study has several strengths. Most notably, it uses administrative data from the New York SID which captures all readmissions to any acute care community hospital facility (i.e., index- and non-index hospital readmissions) in the state. Such non-index readmissions are often uncaptured by other clinical databases¹⁰. Second, we assembled a homogenous cohort of over 19,000 patients undergoing elective lobectomy. Conversely, this study also has several limitations. First, there is a substantial body of evidence that has examined so-called “planned” vs “unplanned” readmissions, mostly using the uniquely compiled data point in the American College of Surgeons NSQIP database²⁶. We did not assess whether readmissions in our study were planned. Despite this, prior studies have found that planned readmissions within 30 days are exceedingly rare following lobectomy²⁷. Second, we did not assess how various disease-specific indications for lobectomy may affect readmission. Though it is notable that a vast majority of lobectomies are performed for lung cancer, lobectomies for benign conditions may affect readmissions and therefore introduce bias into our study. Third, due to the age of the cohort, we were unable to examine robotic lobectomies. Additionally, other data elements important to readmission – such as geographic distance from the index hospital – were unavailable. Finally, we could not examine if the type of lobectomy (upper vs lower, right- vs left-sided) affects outcomes. For example, prior studies have demonstrated that readmissions are more common following lower lobectomy, since the lower lobes typically contribute more to gas exchange than upper lung fields¹⁰. Such differences may be even further exacerbated in patients with COPD (45.8% of this cohort) given the propensity of COPD to heterogeneously affect the upper lobes.

In conclusion, short readmissions after pulmonary lobectomy may be an avoidable phenomenon in select situations. Several factors, particularly readmission to non-index facilities, may help surgeons identify patients at high risk for such readmissions. Interventions that reduce readmissions are critical to mitigate costs and improve patient outcomes following lobectomy.

Glossary of Abbreviations

AHRQ

Agency for Healthcare Research and Quality

HCUP

Healthcare Cost and Utilization Project

SID

State Inpatient Database

CMS

Centers for Medicare and Medicaid Services

ICD

International Classification of Diseases