Patient population and treatment details
The Institutional Review Board Committee at MD Anderson Cancer Center approved our request to review the medical records of these patients. The need for informed patient consent was waived, as this was a retrospective review and no identifiable patient information is included in this report. This study was conducted in accordance with the ethical standards of the Declaration of Helsinki and its later amendments.
After construction of the BED-conversion treatment planning algorithm tool we sought to validate its use in a cohort of patients who had received re-RT. We tested its utility in early-stage NSCLC patients who underwent re-RT (i.e., re-SABR for isolated lung parenchymal recurrence or repeat fractionated radiotherapy (RT)) for isolated LRR after initial SABR. Patients were extracted from an institutional SABR database of over 900 patients [16]. This specific population was chosen because the safety and efficacy of re-RT after primary SABR, for which few experiences have been published to date, has been deemed an active area of interest by the International Association for the Study of Lung Cancer [1].
Our institutional practice for the treatment of lung tumors in the re-RT setting has been previously reported. [1] All patients in the current study received either: (1) 50 Gy/4 fractions (fx) followed by 70 Gy/10fx, (2) 50 Gy/4fx followed by 50 Gy/4fx, (3) 70 Gy/10fx followed by 70 Gy/10fx, or (4) 50 Gy/4fx followed by 60 Gy/30fx. Doses were typically prescribed to the 70–90% IDL covering the planning treatment volume (PTV). For intensity-modulated radiotherapy (IMRT) or volumetric modulated arc therapy (VMAT), an integrated boost to the gross tumor volume (GTV) was typically used to generate a high-dose region. Given this approach, dose distribution was inherently inhomogeneous; target coverage and OAR sparing were prioritized over homogeneity.
Plan deformation
All CT datasets from previous treatment courses were exported to Velocity software for dose (Gy) distribution deformation (VelocityAI 3.0.1, Velocity Medical, Atlanta, GA). Within Velocity, one of the treatment planning CTs was selected as the reference on which the other plan’s nominal dose distribution (Gy) would be deformed. As such, the selected reference CT acted to hold all independent dose distributions for each plan, which were overlaid but not yet summed. The selection of the reference RT course was arbitrary; the deformation process is designed to address anatomical differences.
During the deformation process, rigid registration was first performed to align the CTs according to bony structures. Then, deformable image registration was carried out between the two CTs to obtain the transformation matrix, which was used to deform the nominal distribution (Gy) onto the reference CT. This produced two independent nominal dose distributions (Gy) on a single anatomically validated CT set, one nominal dose distribution from the original reference CT and one nominal dose distribution from the CT that was deformed (Gy). Subsequently, two plans were superimposed on one CT (unsummed), (Fig. 1a). The voxel size used for deformation was 2.5 mm3.
BED conversion
The LQM was utilized to convert the nominal physical doses (Gy) to BED distributions in a voxel-by-voxel manner for each of the two individual data sets. For each voxel i, the BED was calculated as:
$${BED}_{i}= n{D}_{i} \left(1+ \frac{{D}_{i}}{\alpha /\beta }\right)$$
where n is the number of fractions, Di is the physical fractional dose, and α/β is the OAR-specific ratio (designated as 3 for normal tissues). Although we selected an α/β of 3 for simplicity given our interest in normal tissue tolerance, it should be noted that any α/β value could theoretically be used. Since this study aimed to assess normal tissue toxicity risk, evaluation of BED for tumor voxels was not performed. The individual BED maps from each of the two plans were then subsequently summed to yield an overall BED map for all OARs. A workflow of the algorithm process is shown in Fig. 1b.
BED summation and OAR evaluation
Converting dose (Gy) to BED allowed us to account for nonlinear biological response to differing dose per fraction, as BED distributions are additive according to the LQM. Thus, this novel process allowed the nominal distributions (Gy) from separate courses of treatment to be converted and summed to quantify the cumulative BEDs for each voxel in each OAR. The BED dose distribution was calculated by converting the nominal dose distribution (Gy) using an institutionally developed Python script. All BED dose distributions from the two different treatment courses for each patient were imported on the reference CT datasets in the Eclipse treatment planning system (Varian Medical Systems, Palo Alto, CA) and added together to obtain the cumulative BED dose distributions using the plan sum feature of the Eclipse system. The cumulative BED distributions to each OAR from both plans were then evaluated.
Statistical analysis
The Kaplan–Meier method and life tables were used to evaluate overall survival (OS), which was calculated from completion of re-RT to death from any cause. Treatment-related toxic effects were scored with the Common Terminology Criteria for Adverse Events (CTCAE), version 4.0. Reporting of other continuous and categorical data is descriptive and comparison between groups was not indicated. Data were analyzed with SPSS, version 21.0 (IBM Corp, Armonk, NY).