• 2019-10
  • 2019-11
  • 2020-03
  • 2020-07
  • 2020-08
  • br The data within the NCDB are anonymous


    The data within the NCDB are anonymous; therefore, the study was exempt from institutional review board review and no consent was required. The American College of Surgeons and the Commission on Can-cer have not verified and are not responsible for the analytic or statistical methodology employed or the conclusions drawn from these data by investigators.8 
    that mortality was directly related to the reporting institution, patients were excluded if the reporting facility did not match the facility where the primary surgery took place. Patients were also excluded if the primary surgical procedure was localized tumor destruction, including laser abla-tion, cryosurgery, electrocautery, and/or fulguration. Variables such as age, gender, ethnicity, comorbidities (represented as Charlson/Deyo score), insured status, median income quartiles from 2008 to 2012, treatment facil-ity city size, treatment facility type, procedure volume, primary tumor site, analytic stage, histology, grade, neoadjuvant therapy, definitive procedure type, and procedure approach (eg, video-assisted thorascopic surgery, open, or robotic) were obtained from the database. Extent of surgery was document as removal of <1 lobe, a lobectomy, pneumonectomy, or not otherwise specified. Pathologic tumor characteristics such as analytic stage and grade were extracted from the NCDB. Stage was reported in accordance with American Joint Committee on Cancer seventh edition.9 Histology was extracted and simplified to categories: adenocarcinoma, squamous cell can-cer, neuroendocrine, bronchoalveolar cancer, and other as classified in In-ternational Classification of Disease for Oncology third revision.10 Some patient demographic information such as age and ethnicity were grouped for analytic purposes. Neoadjuvant therapy was defined as any SCH-772984 and/or chemotherapy before surgery. Hospital procedure volume was calcu-lated as the procedure volume recorded by a single institution divided by the number of years of data entry provided by that same institution.
    Some NCDB variables of interest contained missing data. This is demonstrated by the results of the descriptive analyses (Table 1 and missing data analysis in Tables E1 and E2). Variables with<5% missingness were imputed with median values to be included in the regression analyses. Although imputation does introduce a potential source of ambiguity, the benefit is maintaining power and sample size to detect associations.11
    Statistical Analysis
    Descriptive statistics were calculated for patient characteristics for both
    30-day mortality and 90-day mortality. The c2 test was used to compare categorical variables and the Kruskal-Wallis test was used to compare continuous variables.
    Operative mortality rates were stratified into 2 groups: 30-day mortality and 90-day mortality. Within each group, sequential mixed-effects logistic regression models using a hospital identifier as a random effect were gener-ated under 4 conditions using R (R Foundation for Statistical Computing, Vienna, Austria). Model 1 was built using a hospital identifier as a random effect without any covariates. Model 2 added patient and tumor factors such as demographic characteristics and tumor stage. Model 3 added facility type. Model 4 added procedure volume within specific ranges. Variation be-tween these 4 models was used to calculate mortality attributable to patient and tumor factors (Model 2) and hospital level factors (Models 3 and 4). Model 4 was ultimately used as the final regression model to identify predic-tors of 30-day and 90-day mortality. Model significance for the odds ratios (ORs) were based on 95% confidence interval (CI) and the P value.
    Subsequently, risk-adjusted mixed effect logistic regression models were used to evaluate hospital performance based on calculated mortality at prespecified time points every 30 days between 30-day mortality and 360-day mortality. Because this analysis was primarily used to assess qual-ity of care at each center, center-level variables, including urban versus ru-ral population size, procedure volume, and facility type, were excluded in this model.12 Each hospital was ranked using an Empirical Bayes predic-tion of center effects for each time point. Similar to STS public reporting ranking system, hospitals were ranked into the highest (2.5%), middle (95%), and lowest (2.5%) performance groups.3 
    Study Population RESULTS
    Patients diagnosed with NSCLC between 2004 and 2013 were identified
    in the NCDB. Patients were excluded if they did not undergo surgical resec-
    went surgery identified within the NCDB at 1230 facilities
    tion at the reporting facility, had unknown vital status, or were missing