Potential Clinical Significance of Overall Targeting Accuracy and Motion Management in the Treatment of Tumors That Move With Respiration: Lessons Learnt From a Quarter Century of Stereotactic Body Radiotherapy From Dose Response Models

Abstract

Objective: To determine the long-term normal tissue complication probability with stereotactic body radiation therapy (SBRT) treatments for targets that move with respiration and its relation with the type of respiratory motion management (tracking vs. compression or gating).

Methods: A PubMed search was performed for identifying literature regarding dose, volume, fractionation, and toxicity (grade 3 or higher) for SBRT treatments for tumors which move with respiration. From the identified papers logistic or probit dose-response models were fitted to the data using the maximum-likelihood technique and confidence intervals were based on the profile-likelihood method in the dose-volume histogram (DVH) Evaluator.

Results: Pooled logistic and probit models for grade 3 or higher toxicity for aorta, chest wall, duodenum, and small bowel suggest a significant difference when live motion tracking was used for targeting tumors with move with respiration which was on the average 10 times lower, in the high dose range.

Conclusion: Live respiratory motion management appears to have a better toxicity outcome when treating targets which move with respiration with very steep peripheral dose gradients. This analysis is however limited by sparsity of rigorous data due to poor reporting in the literature.

Keywords: dose response; normal tissue complication probability; radiosurgery; stereotactic body radiation therapy; tracking.

Figure 1

PRISMA diagram showing that less than half a percent of the published stereotactic body radiation therapy (SBRT) literature was found to have normal tissue complication probability (NTCP) models, and only four papers were found (–29) that had enough detail to compare NTCP dose-response with and without motion tracking.

Figure 2

Pooled logistic (19) or probit (20) models for aorta and major vessels (A), chest wall (B), duodenum (C), and small bowel (D). In each graph the red squares represent planned critical structure doses at which complications occurred and the blue dots represent planned doses that did not result in a complication, on a per-patient basis. The solid blue curve is the maximum-likelihood fitted logistic or probit model (–29). The dashed green curves are the confidence intervals based on the profile-likelihood method (22). The yellow highlighting shows the region of comparison, as summarized in Table 1 . AE, adverse event; NfxED, N-fraction equivalent dose; DVH, dose volume histogram; Dx, DVH level corresponding to volume x; VEGFI, vascular endothelial growth factor inhibitor; MLE, maximum-likelihood estimate.

Figure 3

The Emami table (A) (34, 35) for conventional fractionation as compared to (B) the dose-volume histogram (DVH) Risk Map for aorta dose tolerance in 1 to 5 fractions, with estimated risk levels from the model in Figure 2A (26) as the second number in each cell of the table, when available. Like the Emami table, the DVH Risk Map has both low-risk limits and high-risk limits. The DVH Risk Map additionally plots the data graphically and has a separate row for each degree of fractionation. Dashed green lines represent the low-risk limits and green dots represent individual patient data that is below the low-risk limits. Red lines represent high-risk limits and red dots represent individual patient data that is above the high-risk limits. Yellow dots are patient data between low- and high-risk limits. Red squares represent doses at which grade 3 (G3) or higher adverse events (AE) occurred. Blue diamonds represent published dose tolerance limits (–37), and representative well-established limits have been circled and labeled.

Figure 4

Since radiosurgery uses beam collimation to block adjacent critical structures from receiving high dose, a typical linear accelerator off-axis profile for a 4cm collimator shows that a targeting error of (A) 5mm could result in four times and (B) 1.5mm could result in 1.5 times, higher than planned dose to the critical structure. This figure is a simplistic single-beam illustration to explain the dose gradient concept, whereas more realistic examples with multiple beams are shown in Appendix Figure A4 , where 2mm away from the surface of the spinal cord the dose is 50% higher, and in Appendix Figure A5 where 1.5mm away from the surface of the chiasm the dose is 50% higher.

Figure 5

“The current (data-loss) paradigm” as depicted in QUANTEC (46), reproduced with permission. “Data are effectively lost to the wider scientific community after publication. Capturing key datasets in query-able data repositories would accelerate the discovery of causative factors and increase the accuracy of parameter estimates” (46).