br Recently a retrospective pooled
Recently, a retrospective pooled analysis of 2 prematurely terminated randomized controlled trials suggested that SBRT is better tolerated than surgery and may lead to improved overall survival.7 However, several study limitations necessitate caution to avoid over-interpreting these results, and there remains a paucity of robust clinical data to support the above statement, given the heterogeneity of study cohorts.8,9 To address this AZD-5991 issue, a number of studies have used propensity score matching to minimize the risk of selection bias.10 The purpose of the present systematic review and meta-analysis is to assess the clinical outcomes of SBRT versus surgery for patients with early-stage NSCLC. Primary end points included overall survival in matched and unmatched cohorts. Secondary end points included cancer-specific survival, disease-free survival, freedom from locoregional recurrence, freedom from distant recurrence, and perioperative mortality and morbidity. Each end point was assessed using matched and unmatched cohorts to compare relative outcomes, whenever possible. Subgroup analyses of lobectomy and sublobar resection versus SBRT were also performed for overall survival.
MATERIALS AND METHODS
Literature Search Strategy
A systematic review was performed using EMBASE and Ovid Medline, from their dates of inception to January 2018. To identify all potentially relevant studies, we combined the search terms (‘‘SBRT’’ or ‘‘SABR’’ or ‘‘stereotactic’’ or ‘‘radiosurgery’’) and (‘‘NSCLC’’ or ‘‘non-small cell lung’’ or ‘‘carcinoma, non-small cell lung’’) and (‘‘surg*’’ or ‘‘resect*’’ or lobectomy) as either Medical Subject Headings or keywords. All identified articles were then assessed by applying the predefined selection criteria. A summary of search strategies and techniques has been described in detail previously.11
Selection Criteria and Data Appraisal
Eligible studies for selection in the systematic review were those in which comparative overall survival was reported for patients who
underwent SBRT or surgical resection for NSCLC. When institutions published duplicate studies with accumulating numbers of patients or increased lengths of follow-up, only the most complete or updated re-ports were included for meta-analysis. Abstracts, case reports, confer-ence presentations, editorials, expert opinions, and publications not written in English were excluded. Data were extracted from article texts, tables, figures, and supplementary material. Two investigators (D.W. and C.D.C.) independently reviewed each retrieved article. Dis-crepancies between the 2 reviewers were resolved by discussion and consensus. To assess the quality of the nonrandomized studies, the Newcastle-Ottawa scale was used to evaluate the selection, compara-bility, and outcomes reported in each study, with 0 to 3 stars indicating poor quality, 4 to 6 stars indicating moderate quality, and 7 or more stars indicating high quality.12
When more than 4 studies provided relevant data on the same predetermined end point, meta-analysis was performed by combining the reported clinical outcomes of individual studies using a random effect model. Odds ratio (OR) and standard error were extracted or calculated from each study using methods described by Tierney and colleagues13 and Parma and colleagues.14 When calculations were not possible because of inadequate data, ORs were estimated using Kaplan–Meier graphs. I2 statistic was used to estimate the percentage of total variation across studies attributable to heterogeneity rather than chance. Meta-analysis was performed using Review Manager (version 5.1.2, Cochrane Collaboration, Oxford, UK). All P values were 2 sided.
Individual patient survival data were reconstructed using Guyot’s iterative algorithm to solve the Kaplan–Meier equations originally used to produce the published graphs.15 This algorithm used digitalized Kaplan–Meier curve data to find numeric solutions to the inverted Kaplan–Meier equations, and food chain assumes a constant, noninformative censoring mechanism. The reconstructed patient survival data were then aggregated to form the combined survival curve. Reconstructed Kaplan–Meier analyses were conducted using R (version 3.2.5, R Core Team, Vienna, Austria).