Clinical and Economic Burden Associated with Prolonged Air Leaks Among Patients Undergoing Thoracic Resection: A Retrospective Database Analysis

Introduction

Although prolonged air leak (PAL) is one of the most serious and common complications following thoracic resection, there is no current standard of care for prevention. PAL is typically defined as an air leak persisting longer than five days post-operatively, a definition based on the expected length of hospital stay for a lobectomy.^1,2 A recent systematic review and meta-analysis of 39 international studies comprising 89,006 patients found that pooled PAL incidence was 15% after pulmonary surgery, ranging between 5% and 60%.³ Incidence, however, varies with PAL definition (ie, number of post-operative days) and procedural characteristics (eg, surgical approach, extent of resection, surgical indication).^1–3

The clinical and economic consequences of PAL include increased hospital and ICU utilization, longer hospital lengths of stay (LOS), and increased risk of clinical complications such as pneumonia and pleural space infections.^2,4 Lung sealants are widely used in the United States to reduce the incidence of PAL during thoracic resection. Both biologic-based sealants (eg, fibrin-based and collagen-fleece bound) and synthetic sealants (eg, polyethylene glycol-based and polyglycolic acid-based) have been approved by the Food and Drug Administration (FDA) and are marketed in the U.S.⁵ In various comparative studies (randomized controlled trials, retrospective studies, preclinical) several commercially-marketed sealants used in conjunction with standard of care (stitches, staples) have demonstrated clinical efficacy, lower PAL, shorter hospital LOS and lower healthcare costs compared with these outcomes in patients who received standard of care alone.^6–10 Although these sealants have been assessed individually using various study designs, to our knowledge no study has evaluated the impact of use of lung sealants in aggregate on the real-world incidence of PAL and associated healthcare resource utilization and cost outcomes. To address this knowledge gap, the current study estimated the incremental economic and clinical burden of PAL among patients with lung sealants used during thoracic resection in the United States (US).

Materials and Methods

Data Source and Patient Selection

This retrospective observational study examined data from inpatient and outpatient hospital billing records contained in the Premier Healthcare Database (PHD). The PHD is a population-based research database that contains healthcare records contributed by a convenience sample of nearly 1000 US hospitals and health systems that are members of the Premier Healthcare Performance Improvement Alliance, which represents approximately 25% of annual US inpatient discharges. This database includes discharge-level information on patient demographics, diagnoses, procedures, use of medications and devices, hospital borne costs (ie, costs of each hospital encounter from the hospital perspective), length of stay, discharge disposition, as well as information on hospital and provider characteristics. Although the database excludes federally funded hospitals, contributing hospitals are nationally representative with respect to bed size, geographic region, location (urban/rural) and teaching status.

This analysis of the Premier Healthcare database was conducted under an exemption from Institutional Review Board oversight for US-based studies using de-identified healthcare records, as dictated by Title 45 Code of Federal Regulations (45 CFR 46.104(d)(4)(ii)) (https://www.ecfr.gov/current/title-45/subtitle-A/subchapter-A/part-46/subpart-A/section-46.104). As Premier Healthcare data do not contain direct identifiers of individuals, employers, households, or providers, Institutional Review Board approval is not required.

The study population included adults (age ≥18 years) with inpatient thoracic resection (ie, lobectomy, segmentectomy, wedge resection) between October 1, 2015 and March 31, 2021 and lung sealant (Progel™, Coseal^®, TachoSil^®, Tisseel, Evicel^®, BioGlue^®) used during their procedure. The earliest qualifying admission date was considered the index date. For each patient, follow-up extended through 90 days post-discharge with the hospital in which the index admission occurred contributing data to the PHD throughout that period. Patients with any of the following were excluded: admission as a transfer from another facility, missing discharge disposition, receipt of more than one branded lung sealant, or concomitant cardiac surgery.

For analysis, each patient was classified into one of two mutually exclusive cohorts based the presence or absence of PAL during the index hospitalization. PAL was defined as the diagnosis of post-procedural air leak or post-procedural pneumothorax with an inpatient LOS exceeding 5 days after the resection procedure was completed.

Measurement of Patient and Hospital/Provider Characteristics

Patients and hospital/provider characteristics measured during the index admission included age, sex, race, ethnicity, marital status, payer type, admission type (elective/non-elective), urban vs rural hospital, hospital teaching status, hospital geographic region, hospital bed size, operating physician specialty, year of surgery/index admission, annual hospital surgical volume for thoracic resection, and an indicator for whether hospital costs are derived from a cost-to-charge ratio vs procedural costing. Patient clinical characteristics measured during the index admission included the individual components of the Charlson Comorbidity (CCI) and Elixhauser Comorbidity Indices,^11–13 and their aggregate index scores. Procedural characteristics included surgical approach, type of resection (lobectomy, wedge resection, segmentectomy), lobe location, and surgical indication (malignancy/non-malignancy). The CCI and Elixhauser were measured through the presence of ICD-10-CM codes, excluding those for which there was an indication that the comorbidity was not present on admission.

Measurement of Economic and Clinical Outcomes

Primary outcomes included the total intensive care unit (ICU) days during the index hospital stay, total hospital costs from the hospital perspective, all-cause hospital readmissions within 30-, 60-, and 90-days post-procedure, patients’ discharge status (home/home health vs other setting of care), and mortality. Hospital costs were inflation adjusted to 2021 US dollars using the Medical Care component of the US Bureau of Labor Statistics Consumer Price Index.

Statistical Analysis

Descriptive analyses, stratified by the absence or presence of PAL, were utilized to describe patient, procedure, and hospital/provider characteristics, and unadjusted outcomes. Data for categorical variables were summarized using counts and percentages of patients in each category. Data for continuous variables were summarized using the mean and standard deviation of the variable distribution; medians and other percentile information were reported for variables which are not normally distributed.

Multivariable generalized linear models (GLM) were used to quantify associations between incident PAL and each of the study outcomes, accounting for hospital-level clustering, and adjusting for differences in patient, procedure, and hospital/provider characteristics between the two patient cohorts. GLM with log link and negative binomial error distribution was used for ICU days, GLM with logit link and binomial error distribution was used for all-cause readmissions within 30, 60, and 90 days post-discharge, discharge status, and mortality, and a GLM with log link and gamma distribution was used for total hospital costs. Additionally, to confirm the face validity of the present study’s findings in relation to prior research, we fit a multivariable logistic regression model to examine the association of patient, provider, procedure, and hospital characteristics with the incidence of PAL. Statistical analyses were performed using StataSE 16 (StataCorp, College Station, Texas, US).

Results

A total of 62,220 patients met initial selection criteria; 9727 of these patients had one lung sealant of interest used during their index thoracic resection and were included in the final study population (Table 1). Tables 2-4 display information on patient demographics/clinical characteristics, procedure characteristics, and hospital/provider characteristics, respectively. Overall, the mean age of study patients was 66 years, 51% were female, and most (84%) were of white race. Approximately 63% of patients had Medicare coverage and nearly three-quarters of patients had an Elixhauser Comorbidity Score between 1 and 4, with most of the remaining patients having scores of 5 or higher (Table 2). Most (81%) patients’ surgical indication was for malignancy and surgical approach was evenly distributed (open 35%, video-assisted thoracoscopic surgery [VATS] 34%, robotic 32%). Out of all resections, 5647 (58%) were lobectomies, and the right upper was the most prevalent (31%) lobe position (Table 3). Patients were predominantly treated in urban hospitals (91%) and nearly half (48%) underwent surgery by a pulmonary/thoracic surgeon (Table 4).

Table 1 Patient Selection

Table 2 Patient Demographic Characteristics and Comorbidity Burden at Index

Table 3 Procedure Characteristics

Table 4 Hospital and Provider Characteristics

Incident PAL was identified in 12.5% (n=1213) of patients after the index procedure. In unadjusted analysis, mean (SD) age was not meaningfully different between patients with and without PAL (67.4 years [10.0] and 65.9 years [11.7], respectively) although the age distribution of patients with PAL was shifted upward with a larger proportion aged 65 years or older and covered by Medicare. Patients with PAL also carried a greater comorbidity burden; 76% had an Elixhauser Comorbidity Score of 3 or higher compared with 66% of patients without PAL.

After adjusting for differences in demographics, clinical characteristics, procedural and hospital characteristics, incident PAL was found to significantly increase resource utilization and costs (Table 5). Specifically, PAL was associated with significant incremental increases in ICU days (0.93 days, p<0.001) and incrementally higher total hospital costs ($11,119, p<0.001). The presence of PAL also significantly decreased patients’ chance of being discharged to home vs other setting (eg, skilled nursing facility) from 91.3% to 88.1% (3.6%, p<0.001). Patients’ risk of being readmitted also significantly increased with the presence of PAL from 9.3% to 12.6% at 30 days, from 11.7% to 15.4% at 60 days, and from 13.6% to 17.2% at 90 days, all p<0.01. Additionally, although the absolute risk of in-hospital mortality was low in the study population, it was two times higher in patients with PAL compared with patients without evidence of PAL (2.4% vs 1.1%, p=0.001).

Table 5 Adjusted Cost and Utilization Outcomes

In the multivariable logistic regression model fit to examine the association of patient, provider, procedure, and hospital characteristics with the incidence of PAL, male sex, resection of the right upper lobe, lobectomy (compared to wedge, segmentectomy), and multiple resections were significantly associated with increased risk of PAL (Table 6). Minimally invasive surgery was significantly associated with decreased risk of PAL.

Table 6 Predictors of Prolonged Air Leaks

Discussion

Greater risk of clinical complications, longer postoperative and ICU stays as well as higher costs have been documented previously for air leaks in general and PAL in various settings and patient populations,^8,14–17 but to our knowledge, this is the first real-world study to quantify the incremental healthcare resource utilization and cost impacts of PAL in patients who have received prophylactic lung sealants.

Although findings from these earlier studies cannot be directly compared with ours given differences in study populations and air leak definitions used, they do provide important historical and clinical context for our results. One 6-year retrospective study reported that post-operative air leaks lasting more than 6 days occurred in 6.7% of all patients who underwent lung resection between January 2002 and December 2007 in the Liverpool Heart and Chest Hospital NHS Foundation Trust (United Kingdom). In that study, PAL was associated with increased length of stay (p<0.0001), in-hospital mortality (p=0.003) and intensive care unit readmission (p=0.05).¹⁶ An administrative claims data study of patients who underwent a lobectomy, segmentectomy or wedge resection in 2009–2011 found that PAL added $15,000 per patient. Consistent with the current study’s findings, the authors also reported that video-assisted thoracoscopic surgery was associated with a reduced risk of PAL, which subsequently resulted in lower total hospital costs, length of stay, and risk of readmission.¹⁷

Another, more recent study, examined the impact of air leak complications, defined as any air leak plus pneumothorax in patients who had undergone primary lobectomy, segmentectomy, or wedge resections in 2012 through 2014 and found associated increased utilization burden, hospital costs and mortality risk.¹⁴ This study found adjusted mean hospital costs were approximately $6500 higher in patients with air leak complications compared to costs for patients without such complications.

Yotsukura et al quantified the impact of PAL in 2278 patients who underwent pulmonary resection for lung cancer from 2014 to 2018. PAL occurred in 4.0% of those patients and was associated with the development of additional complications (p < 0.001) and a 32% increase in hospital costs (p<0.001).¹⁵ In another retrospective analysis of 982 patients undergoing lobectomy or segmentectomy between 2014 and 2018, PAL occurred in 27% of patients. Incidence of 90-day readmission was twice that in patients with PAL compared to patients without PAL and those with PAL had 27% higher index hospital costs (p<0.0001).¹⁸ Significant risk factors for PAL in the current study are consistent with prior literature, including male sex, resection of the right upper lobe, and an open surgical approach.^15,16,18,19

Limitations of the present study include those inherent in using a hospital billing database for research purposes. Search/text mining techniques were used to identify the use of lung sealants; however, the absence of device identification through such techniques does not necessarily rule out the use of the device. It must be assumed that misclassification is non-differential between the group of patients using sealants who experienced PAL as compared to those who did not. The PHD does not include an ICD-10 diagnosis date, so the exact onset and duration of PALs is unknown. Additionally, the PHD does not capture nutritional status which is a significant prognostic factor for PAL.²⁰ Other unmeasurable variables such as provider skill, surgical technique, overall patient health, product availability in hospitals, hospital practices, and other factors (eg, forced expiratory volume, presence of incomplete or fused fissures) may lead to residual confounding after adjusted analyses. It is important to note that in the hospital readmissions that do not occur within the same hospital as the index hospital admission are not captured in the PHD and therefore may be underestimated. Lastly, the study results will not necessarily be generalizable to all hospitals in the US.

Conclusion

This analysis demonstrates that despite the use of lung sealants, PAL continues to put a burden on the healthcare system, highlighting an unmet need for improved sealant technology.