Study Data Source and Patient Selection
This study was undertaken from a US hospital perspective. The data source was the Premier Healthcare Database, which contains complete clinical coding, hospital cost, and patient billing data from more than 700 hospitals throughout the USA. Although the database excludes federally funded hospitals (e.g., Veterans Affairs), the hospitals included are nationally representative on the basis of bed size, geographic region, location (urban/rural), and teaching status. The database contains date-stamped records of all billed items including medications; laboratory, diagnostic, and therapeutic services; and primary and secondary diagnoses for each patient’s hospitalization. Additionally, the database also provides patient demographic and payer information.
Figure 1 shows the study’s patient selection process. Patients selected for the study underwent VATS lobectomy, as defined by International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) procedure coding and International Classification of Diseases, Tenth Revision, Procedure Coding System (ICD-10-PCS) coding (See Supplemental Appendix 1 for a listing of all codes used in the study) during a hospital admission between January 1, 2012 and September 30, 2016. The first hospital admission for VATS lobectomy during this period was defined as the index admission, and patients were required to be at least 18 years of age at the time of the index admission. Patients were excluded from the study if they had missing data on hospital supply, room and board, or total hospital costs, if they were transferred from another institution, or if they had a non-elective VATS lobectomy.
Use of either powered or manual staplers during the index hospitalization was identified from hospital administrative records by searching for various combinations of device names (e.g., iDrive, Powered Echelon Flex, Powered Vascular Stapler), model numbers (e.g., PCE45A, IDRVULTRA1, PVE35A), and/or descriptors of devices being “powered”. Staplers were also further classified by manufacturer (Ethicon/Johnson & Johnson; Medtronic/Covidien). Only patients for whom a stapler used during the index hospitalization could be identified as either powered or manual were retained for the study; patients with evidence of both powered and manual staplers were excluded from the study because of the inability to assign them to one of the two study groups: powered stapler group or manual stapler group. Patients were also excluded if da Vinci EndoWrist® surgical staplers (Intuitive Surgical, Sunnyvale, CA), which are part of the robotic da Vinci Surgical System, were used during the index hospitalization.
Patient and Hospital/Provider Characteristics
Patient demographics and hospital/provider characteristics measured during the index admission included age, gender, marital status, race, Hispanic vs. non-Hispanic indicator, payer type, urban or rural hospital, hospital teaching status, hospital geographic region, hospital bed size, procedural physician specialty, year of surgery/index admission, hospital surgical volume for VATS lobectomy, and an indicator for whether hospital costs are derived from a cost-to-charge ratio or procedural costing.
Patient clinical characteristics measured during the index admission included the Charlson comorbidity index (CCI) [15], the day on which the VATS lobectomy procedure was performed after hospital admission, cancer vs. non-cancer primary diagnosis, concomitant wedge resection, concomitant segmentectomy resection, robotic assistance, and several individual comorbidities from the Charlson and Elixhauser comorbidity indices [based on diagnoses recorded in any position: cancer (metastatic), cancer (non metastatic), cerebrovascular disease, chronic pulmonary disease, coagulopathy, congestive heart failure, connective tissue/rheumatic disease, depression, diabetes (w/complications), diabetes (w/o complications), hypertension (complicated), hypertension (uncomplicated), hypothyroidism, liver disease (mild), myocardial infarction, neurological disorders, obesity, peripheral vascular disease, renal disease, valvular disease, and weight loss] [15]. Comorbidities and the CCI were measured through the presence of ICD-9-CM/ICD-10-CM codes, excluding those for which there was an indication that the comorbidity was not present on admission. Certain comorbidities from the Charlson and Elixhauser comorbidity indices were excluded because of infrequency or non-occurrence (e.g., paralysis, human immunodeficiency virus/acquired immunodeficiency deficiency syndrome).
Outcomes
Economic and healthcare resource use outcomes were evaluated during the index admission (with the exception of readmissions) and included hospital length of stay (LOS); total hospital costs from the hospital perspective (including subcategories of medical/surgical supply costs, room and board costs, and operating room costs); operating room time; discharge to a non-home setting (e.g., skilled nursing facility, intermediate care facility) vs. discharge to home with or without home healthcare; and 30-, 60-, and 90-day all-cause readmissions to the hospital in which the VATS lobectomy procedure was performed. All costs were inflation adjusted to 2016 US dollars using the Medical Care component of the US Bureau of Labor Statistics Consumer Price Index. When we analyzed the operating room time outcome, patients were included only if their operating room time values recorded in the database fell between 30 min and 24 h (86.5% of patients in the primary analysis met this criterion); this step was taken to eliminate patients with implausible values. When we analyzed the all-cause readmission outcomes, patients were included only if the hospital in which their VATS lobectomy procedure was performed had (at an overall hospital level) discharge records extending throughout the observation period of interest (30, 60, or 90 days); this step was taken to eliminate patients for whom all-cause readmission would be unobservable because of non-contribution of discharge data at a hospital level.
Clinical outcomes were evaluated during the index admission and included a composite outcome of bleeding and/or transfusions; transfusions alone; acute posthemorrhagic anemia; air leak complications (based on the ICD-9-CM/ICD-10-CM diagnosis codes for pneumothorax or air leak, which includes but is not necessarily limited specifically to prolonged air leaks); pneumonia; and infection (comprising surgical site, septicemia, pneumonia, and infections of other sites [see Supplemental Appendix 1]).
Within the study protocol, the operating room time, all-cause readmission, and clinical complication outcomes were designated as exploratory, while other outcomes were designated as primary. The exploratory outcomes were designated as such because of (1) uncertainty regarding the accuracy of values recorded for operating room time; (2) readmissions to the hospital within the Premier Healthcare Database are captured only when the patient returns to the hospital in which the index admission took place, thereby introducing the potential for incomplete outcome data capture; and (3) uncertainty regarding the sensitivity and specificity of the clinical coding for complications.
Statistical Analyses
Bivariate analyses, stratified by the powered vs. manual stapler groups, were used to describe patient and hospital/provider characteristics and unadjusted outcomes. Standardized differences were used to assess the magnitude of differences in baseline characteristics between the study groups, where a standardized difference greater than 0.10 was considered to be imbalanced.
Multivariable regression models were used to compare outcomes between the powered and the manual stapler group, adjusting for all aforementioned patient and hospital/provider characteristics, regardless of the values of the standardized differences between study groups. The Box Cox and modified Park test were used, respectively, to select link functions and error distributions which were tailored to the empirical distributions of the outcome variables (e.g., log link and gamma error distribution for hospital costs; logit link and binomial error distribution for dichotomous outcomes) [16, 17]. Statistical clustering may arise among patients who receive treatment within the same hospital; thus, this was addressed using generalized estimating equations (GEE) models and mixed models. The former used an exchangeable working correlation structure—chosen on the basis of a qualitative understanding of the potential nature of clustering within hospitals. Mixed models were used when GEE models failed to converge, which was primarily the case for dichotomous outcomes. In the GEE models, inference was based on empirical (robust) standard error estimates. Adjusted outcomes were generated for each of the comparator groups using the recycled prediction (marginal standardization approach) [18]. A two-sided critical value of 0.05 was used to determine statistical significance. All statistical analyses were performed using SAS version 9.3.
Subgroup Analyses
First, to test whether the powered or manual findings were driven by manufacturer-level effects, post hoc subgroup analyses were completed wherein the predominant manufacturers within the powered and manual stapler groups were compared to one another for all study outcomes using the same statistical analysis approach as the primary analyses; specifically, Ethicon powered staplers, which accounted for 99.4% of the powered staplers, and Medtronic manual staplers, which accounted for 75.8% of the manual staplers.
Second, to examine the role of the thoracic vessel tissue-specific design of the Ethicon PVS stapler, post hoc subgroup analyses were completed wherein the PVS stapler was compared to Medtronic manual staplers for all study outcomes using the same statistical analysis approach as the primary analyses. These analyses are restricted to only the years in which the PVS stapler was present in the database (2015 and 2016). Furthermore, to increase the likelihood that the Ethicon PVS stapler was used for vascular transections, evidence of a second non-PVS Ethicon powered stapler being used during the VATS lobectomy procedure was required for the Ethicon PVS stapler group.
Finally, the prevalence of comorbid chronic obstructive pulmonary disease (COPD) is relatively high among patients undergoing lung lobectomies, and has been associated with perioperative morbidity and mortality [19]. Thus, analyses of hemostasis-related complications were repeated in the subgroup of patients with a comorbid diagnosis of chronic obstructive pulmonary disease (codified as chronic obstructive pulmonary disease and allied disorders, according to the ICD-9-CM/ICD-10-CM taxonomies) for the primary analysis and the post hoc subgroup analysis comparing outcomes between Ethicon powered staplers and Medtronic manual staplers.
Propensity Score Matched Sensitivity Analysis
Finally, a sensitivity analysis was conducted to determine whether the study’s primary findings are robust to use of an alternative statistical approach: propensity score matching. Specifically, the powered stapler group was propensity score matched to the manual stapler group using a variable ratio matching approach of up to three manual stapler patients per one powered stapler patient. The propensity score was estimated in multivariable logistic regression using all covariates from the main multivariable models as predictors of membership in the powered stapler cohort (vs. manual stapler cohort as reference). The matching was accomplished through greedy matching, with a caliper equal to 0.2 times the standard deviation of the logit of the propensity score. After matching, the post-match balance of covariates was examined via standardized differences, and variables which remained imbalanced (standardized difference > 0.10) across the propensity matched cohorts were entered into a second-stage multivariable outcome model in which the primary predictor was membership in the powered stapler cohort (vs. manual stapler cohort as reference). The second-stage multivariable outcome models were implemented in the same manner as the primary analysis models and adjusted outcomes were generated for each of the comparator groups using the recycled predication. As a result of the comparatively smaller sample sizes of the subgroup analyses, which after matching would have likely yielded inadequate statistical power, this sensitivity analysis was conducted only for the primary analysis.
Study Conduct and Protection of Human Subjects
This study was conducted per a prespecified protocol which was approved through scientific governance. As a result of the retrospective, observational nature of the study, it was not a registered trial. The study was planned and conducted in a manner consistent with the International Society of Pharmacoepidemiology principles of the Good Research for Comparative Effectiveness guidance and the PICO Model for Clinical Questions [20, 21]. The Premier Healthcare Database consists of de-identified healthcare records. In the USA, retrospective analyses of the Premier Healthcare Database data are considered exempt from institutional review board (IRB) oversight as dictated by Title 45 Code of Federal Regulations, Part 46 of the USA, specifically 45 CFR 46.101(b)(4) (http://www.hhs.gov/ohrp/humansubjects/guidance/45cfr46.html). In addition, in accordance with the HIPAA Privacy Rule, disclosed data from the Premier Healthcare Database are considered de-identified per 45 CFR 164.506(d)(2)(ii)(B) through the “Expert determination” method (http://www.hhs.gov/hipaa/for-professionals/privacy/special-topics/de-identification/). Throughout this research project, the study data remained de-identified and stored on encrypted, password-protected servers to protect patient confidentiality. This article does not contain any studies with human participants or animals performed by any of the authors.