We identified 222 patients who received SBRT for primary lung cancer at our institution between January 2007 and October 2013. The inclusion criteria were (1) presence of newly diagnosed primary NSCLC with or without biopsy confirmation, regardless of prior treatment for lung cancer; (2) clinical T1–3N0M0 (UICC TNM classification, 7th edition); (3) tumor diameter less than 5 cm; (4) received SBRT with 48 Gy in 4 fractions; and (5) follow-up time of at least 6 months. The exclusion criteria were patients with cytologically or histologically diagnosed small-cell lung cancer and those who received other dose fractionation regimens or adjuvant chemotherapy. Twenty-two patients were excluded, and 200 patients who met our inclusion criteria were enrolled.
This retrospective study was approved by our institutional review board (approval number 19164). All patients provided written informed consent for use of their data in clinical research before the administration of SBRT and had the opportunity to opt-out of the study.
All patients were immobilized with the BodyFix double-vacuum immobilization system (Medical Intelligence, Schwabmuenchen, Germany) and observed using four-dimensional computed tomography (CT). An internal target volume (ITV) was determined using four-dimensional CT images encompassing the gross tumor volumes (GTVs) in all respiratory phases. A planning target volume (PTV) was generated by expanding the ITV by 5–8 mm in all directions. A total dose of 48 Gy in 4 fractions was prescribed at the isocenter using 6–9 non-coplanar static conformal beams with a 5 mm multi-leaf collimator margin. All treatment plans were generated using the Eclipse treatment planning system (Varian Medical Systems, Palo Alto, CA, USA). The radiation doses before January 2009 were calculated using a pencil beam algorithm with an inhomogeneity correction and thereafter with an analytical anisotropic algorithm. Image guidance before September 2007 was based on bony anatomy matched with orthogonal kV imaging and thereafter, tumor matching with online three-dimensional cone-beam CT imaging was used. All treatments were delivered using a linear accelerator (Clinac 23EX, Varian Medical Systems, Palo Alto, CA, USA).
Patient follow-up and evaluation of local failure
Follow-up examinations after SBRT typically consisted of a chest CT every 3 months for 1 year, and thereafter, every 6 months. When relapse was suspected, 18F-fluorodeoxyglucose-positron emission tomography-CT (18 F-FDG-PET/CT) was performed. Local failure was defined as recurrence within 1 cm of the PTV as described by the Radiation Therapy Oncology Group (RTOG) 0236 trial  and diagnosed using biopsy confirmation of viable carcinoma or the accumulation of FDG to maximum standardized uptake value (SUVmax) ≥ 5.0 or the pre-treatment SUVmax . When these examinations were not feasible, local failure was diagnosed based on the presence of high-risk CT features as proposed by Huang et al.  In patients who were not followed up in person, local failure was evaluated based on information from their new physicians.
We grouped patients into consecutive (OTT = 4 or 5 days, n = 116) or non-consecutive treatment groups (OTT = 6–10 days, n = 84) (i.e., in the consecutive group, SBRT was administered within a calendar week, whereas, in the non-consecutive group, SBRT was administered over two calendar weeks with a treatment-break because of the weekend). The primary and secondary outcomes of interest were LC and OS across the two groups. We performed all analyses at the patient level and restricted the follow-up period to the first 5 years after treatment. LC was defined as the time from the start of SBRT to the date of local failure. OS was defined as the time from the start of SBRT to the date of death from any cause.
The differences in baseline characteristics between the two groups were assessed using the Fisher’s exact test or chi-squared test. To account for the imbalance of baseline covariates between the two groups, we performed a propensity score (PS) analysis using the inverse probability of treatment weighting (IPTW). PS values were estimated from covariates using multivariable logistic regression and plotted as histograms. The concordance statistic (c-statistic), a measure of goodness of fit in logistic regression, was used to test the appropriateness of the model. Covariate balance was assessed using a standardized mean difference approach. A standardized mean difference of less than 0.10 for a given covariate was considered as an acceptable balance. Unadjusted LC and OS curves were shown using the Kaplan-Meier method and compared with the log-rank test, and the corresponding hazard ratios (HRs) were estimated. Adjusted LC and OS curves were shown using the Cox proportional hazards model with PS-weighting, and the adjusted HRs were estimated. Furthermore, to account for the residual covariate difference after PS weighting, direct covariates-adjusted survival curves  with PS-weighting were generated using the covariates that we used to estimate PS, and the corresponding adjusted HRs were estimated. We performed a sensitivity analysis comparing 4 days of consecutive treatment with 6–10 days of non-consecutive treatment (i.e., excluding patients treated with an OTT of 5 days) to validate the robustness of the results.
All statistical tests were two-sided, and a p-value of less than 0.05 was considered statistically significant. All analyses were performed using R version 3.5.3 (R Foundation for Statistical Computing, Vienna, Austria) or SAS version 9.4 (SAS Institute Inc., Cary, NC, USA).