OVERVIEW
The ultimate goal of radiotherapy treatment planning is to find a treatment that will yield a high tumor control probability (TCP) with an acceptable normal tissue complication probability (NTCP). Yet most treatment planning today is not based upon optimization of TCPs and NTCPs, but rather upon meeting physical dose and volume constraints defined by the planner. It has been suggested that treatment planning evaluation and optimization would be more effective if they were biologically and not dose/volume based, and this is the claim debated in this month’s Point/Counterpoint.
Arguing for the Proposition is Joseph O. Deasy Ph.D. Dr. Deasy obtained his Ph.D. in Physics from the University of Kentucky in 1992 and subsequently was a postdoctoral fellow at the University of Wisconsin, Madison in Rock Mackie’s group, where he met Jack Fowler, who stimulated his interest in predicting outcomes. He subsequently held several faculty positions, including professor from 2008 to 2010 at Washington University in St. Louis, where he established a division of Bioinformatics and Outcomes Research within the Department of Radiation Oncology, before moving to Memorial Sloan Kettering Cancer Center in 2010 as Chair of the Department of Medical Physics. His research interests include applying statistical modeling to the analysis of large complex datasets in order to understand the relationship between treatment, patient, and disease characteristics and the probability of local control and normal tissue toxicity and applying statistical methods to information derived from medical images, a growing field called “radiomics.” Dr. Deasy is a very active in the AAPM, ASTRO, and chairs several committees and task groups. He has held numerous grants to support his research and has published over 100 papers in peer-reviewed journals.
Arguing against the Proposition is Charles S. Mayo, Ph.D. Dr. Mayo obtained his Ph.D. in Physics from the University of Massachusetts, Amherst in 1991 and was a post-doctoral fellow from 1990 to 1993 specializing in proton therapy at the Harvard University Cyclotron Laboratory, Cambridge, MA. After holding several positions in New England hospitals, he moved to the Department of Radiation Oncology, Mayo Clinic, Rochester, MN in 2010, where he is currently Assistant Professor of Medical Physics. He has been very active on AAPM Committees and Task Groups and currently serves as chair of TG 263 on Standardizing Nomenclature for Radiation Therapy. He has served as President of the New England Chapter and is a fellow of the AAPM. Dr. Mayo’s major research interests include effects of reduced treatment time on tumor radiobiology, normal tissue tolerance to radiation, modeling effects of motion on TCP/NTCP, design and analysis of patient outcomes databases, and design of web interfaced/database solutions for managing clinical practice, on which he has published extensively.
FOR THE PROPOSITION: Joseph O. Deasy, Ph.D.
Opening Statement
The goal of radiotherapy should be to give a highly effective tumor treatment with an acceptably low risk of toxicity. I believe that we have reached the tipping point, where predictive NTCP and TCP models, when validated against relevant datasets, should be placed in the hands of clinical practitioners, alongside accepted dose–volume metrics, such as those in QUANTEC reviews.1,2
Today, prescriptions for standard external beam radiotherapy are often written in categories of “one-prescription-dose-fits-all,” with little personalization to the patient’s disease, despite the fact that we know very clearly that normal tissue tolerance is strongly related to organ/tissue volumes irradiated. To give just one example, most locally advanced lung cancer patients treated in the U.S. receive 60 Gy in 30 fractions, despite wide variations in tumor volume, shape, and location within the lungs. Just as importantly, our current obsession with dose flatness, and with overly generous margins, has little radiobiological support and is likely to be counterproductive.
In many cases, commonly used dose–volume planning limits (for normal tissues) and dose–volume goals (e.g., the PTV D95, the maximum dose in the coldest 5% of the planning target volume) are likely to be poor substitutes for NTCP and TCP prediction models derived from adequately powered datasets. Dose–volume constraints typically have higher uncertainty with respect to their impact on outcome, because they are inherently less general and only represent part of the (dose and fractionation) picture that biological/NTCP models attempt to integrate into a useful estimate. It follows that driving IMRT optimization using outcome-based functions could result in significantly more effective dose distributions for many patients.
The radiobiological modeling in TCP and NTCP functions throws the foundations of the conventional planning paradigm into question. In particular, the practice of worshipping flat dose distributions, accompanied by large margins overlapping with normal tissues (“paranoid target volumes”), should lose its unearned credibility and be superseded by a more rational approach to choosing dose limits.
Currently, there is a good reason to consult relevant NTCP models (and in a few cases, TCP models) on a routine basis, but this is not as easy to do in commercial treatment planning systems as it should. Furthermore, to properly use NTCP and TCP models, new protocols need to be written and established which allow for greater treatment customization based on the anatomical details of each patient’s case.
Other technologies will be crucial for optimizing radiotherapy treatment plans. Probabilistic treatment planning, which accounts for residual geometrical uncertainties relevant to a given treatment situation, is another piece of the puzzle for rationally optimizing radiotherapy, and can help avoid overly generous margins. Together with increasingly accurate online tissue localization (for example, using in-room MRI), a much more aggressive approach could be taken: steep dose gradients could be routinely placed close to the edge of the tumor (or regions of suspected occult disease), thus increasing the likelihood of complete disease sterilization due to the hotter dose over most of the target, while lessening the impact on normal tissues spared by the rapid dose falloff.
Although TCP and NTCP models are improving significantly with continued publications, there are many areas where better models are needed, and the imperative to pool the data necessary to improve the models is greater than ever.
Even today, routine treatment planning would benefit from referencing TCP and NTCP functions, alongside accepted dose–volume constraints. It is logical to expect that the increasing integration of TCP and NTCP models into clinical practice, over time, will result in more therapeutically effective (nonuniform) dose distributions, and a rationalized personalization (within limits) of prescription dose and fractionation.
AGAINST THE PROPOSITION: Charles S. Mayo, Ph.D.
Opening Statement
Can we accept the proposition that we know enough about the reliability and errors associated with biological models and the resultant impact on dose distributions (among the many treatment planning systems) produced by using these models in optimization and evaluation of patient plans, to take the extreme position of discarding several decades of clinical experience based on DVH metrics as the fundamental benchmarks? Dr. Deasy et al. have elegantly summarized one argument against the Proposition: “Despite a large number of dose-volume-outcome publications, made possible by the revolution in three-dimensional treatment planning, progress in normal tissue complication probability (NTCP) modeling to date has been modest. The QUANTEC reviews, though helpful, have demonstrated the limited accuracy of existing risk protection models.”1
Models are still evolving to improve and demonstrate predictive ability.2,3 There can be wide variability among existing models and challenges in accurately predicting clinical outcomes. Classic models often do not have means to adequately reflect the clinically observed impact of nondosimetric factors such as age, chemotherapy, surgery, setup reproducibility, or even fractionation. Efforts to improve models have highlighted a richer set of interactions to explore such as heart–lung codependence4 or nerve–vasculature interactions. Development of radio-genomics databases have helped pave the way for models factoring in single-nucleotide polymorphisms to improve predictive power of NTCP models.5 Imaging based biomarkers6 may be used to improve models. Monte Carlo approaches to TCP/NTCP estimations and in-silico modeling of outcomes and mechanisms are very promising.7 Concluding that we are ready to specify which models the community should use may be premature.
Is it more efficient or reliable to set constraints using radiobiological metrics during optimization rather than DVH metrics? After all, is not an experienced treatment planner intuitively factoring in an EUD calculation when considering dose–response in setting constraints on dose regions of the DVH curve? However, when evaluating the plan, consumers of the results are mindful of extensive literature, trial results, and personal experience correlating DVH metrics with outcomes. A few clinics have long experience examining correlations of NTCP and DVH metrics with outcomes to shape their perspective on clinical decisions.8 In the larger community, when a benchmark DVH metric has not been met but the radiobiological metric has, replanning to also meet the DVH metric is the likely outcome. Relying only on biologically based metrics rather than considering both for plan optimization and evaluation is less effective.
To reach the goal of more reliance on radiobiologically based metrics, we need to greatly improve the community’s personal experience with use of these metrics. A useful strategy would be to routinely calculate these metrics along with DVH metrics in plan evaluations, saving the results in databases to correlate with outcomes monitored in those clinics, pooling results among institutions to explore the impact of variations from one clinic to another in order to find consensus. The Proposition sets forth a great vision, but one that the community as a whole is not ready for right now.
Rebuttal: Joseph O. Deasy, Ph.D.
It is admittedly awkward to argue against quotes from my work with my QUANTEC coauthors, but my own position has evolved, given the steady stream of TCP and NTCP models and analyses that have been published since QUANTEC.9–14 We are in an exciting age of predictive model development. This is an opportunity that we should not ignore: given the mathematical nature of these models, only medical physicists are equipped to lead the effort of clinical adoption (e.g., see the Biosuite software system developed by Nahum and colleagues).15 Meanwhile, standard practice creeps along, based on rather arbitrary, DVH-based measures of treatment plan quality which have long been understood to have limited predictive value.16 As one example: there is currently little clinical evidence that either the D95 or D98 of the PTV is a good predictor of tumor control. I am happy to concede to my debate opponent that our knowledge of NTCP and TCP models is incomplete, with many holes requiring further improvements, and that the clinical use of predictive models should proceed cautiously.3 A wholesale replacement of standard DVH guidelines with NTCP and TCP models is not to be recommended. Nonetheless, the use of published, validated models to inform physicians and planners concerning tradeoffs in toxicity risk against disease control is highly desirable and would, I believe, amount to a real (though admittedly incremental) improvement in treatment planning.
Consider the historical parallel of the argument previously advanced against using heterogeneity-corrected dose calculations in lung because our clinical experience was (at that time) based on water-equivalent calculations. It is now well-accepted that those simplified algorithms were dangerously misleading (resulting in incorrect doses and undersized apertures). Analogously, we can continue to rely on relatively arbitrary traditional DVH metrics, or we can make a serious, sustained, concerted effort to put validated outcome prediction models into the hands of physicians, physicists, and treatment planners, who could use the resulting predictions (alongside other considerations, such as age, performance status, and the patient’s input) to improve clinical decision making.
Rebuttal: Charles S. Mayo, Ph.D.
There is too little multi-institutional data available on tumor control or normal tissue complications documenting correlations of traditional DVH metrics with TCP/NTCP models and contrasting confidence intervals for predicting outcomes. Without this clinical data, the arguments in favor of prioritizing biological models over DVH metrics for clinical decision making are weaker than they should be. Using models to add to what we know from the DVH metrics falls short of the goal of the Proposition, but it is a very good and plausible first step.
Treatment planning systems should make it easy to routinely calculate, report, and record sets of DVH and radiobiological model metrics. If this had been true for the last decade, it is unlikely that we would be having the current debate and speculation. QUANTEC might then have been able to report on a literature summary table of recommend radiobiological constraints in addition to one of DVH metrics.
We should be cautious about the potential for overstating the impact of use of these models in optimization and evaluation. Sculpting steep dose gradients and margin reduction have steadily improved with improvements in MLC design, IMRT/VMAT optimization algorithms, and IGRT technologies. Radiobiological models are less the limiting factor than the physics of the beams and the planning systems’ practical abilities to push limits. For proton beams, models that reflect effects of variation of LET in range of the Bragg peak may result in change for how we evaluate some plans. TCP models produce less uniform dose distributions, but the range of values that will pass is still going to be limited. Until there are data demonstrating outcomes as good or better, it is unlikely that a clinician would accept extremely cold regions in a CTV or hot regions in a PTV, despite having acceptable TCP values.
REFERENCES
- 1.Deasy J. O.et al. , “Improving normal tissue complication probability models: The need to adopt a “data-pooling” culture,” Int. J. Radiat. Oncol., Biol., Phys. 76, S151–S154 (2010). 10.1016/j.ijrobp.2009.06.094 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Bentzen S. M.et al. , “Quantitative analyses of normal tissue effects in the clinic (QUANTEC): An introduction to the scientific issues,” Int. J. Radiat. Oncol., Biol., Phys. 76(3), S3–S9 (2010). 10.1016/j.ijrobp.2009.09.040 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Li X. A.et al. , “The use and QA of biologically related models for treatment planning: Short report of the TG-166 of the therapy physics committee of the AAPM,” Med. Phys. 39, 1386–1409 (2012). 10.1118/1.3685447 [DOI] [PubMed] [Google Scholar]
- 4.Cella L.et al. , “Complication probability models for radiation-induced heart valvular dysfunction: Do heart-lung interactions play a role?,” PLoS One 9(10), e111753 (2014). 10.1371/journal.pone.0111753 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Tucker S. L.et al. , “Incorporating single-nucleotide polymorphisms into the Lyman model to improve prediction of radiation pneumonitis,” Int. J. Radiat. Oncol., Biol., Phys. 85(1), 251–257 (2013). 10.1016/j.ijrobp.2012.02.021 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Nazem-Zadeh M. R.et al. , “Response-driven imaging biomarkers for predicting radiation necrosis of the brain,” Phys. Med. Biol. 59(10), 2535–2547 (2014). 10.1088/0031-9155/59/10/2535 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.El Naqa I., Pater P., and Seuntjens J., “Monte Carlo role in radiobiological modelling of radiotherapy outcomes,” Phys. Med. Biol. 57(11), R75–R97 (2012). 10.1088/0031-9155/57/11/r75 [DOI] [PubMed] [Google Scholar]
- 8.Kong F. M., Pan C., Eisbruch A., and Ten Haken R. K., “Physical models and simpler dosimetric descriptors of radiation late toxicity,” Semin. Radiat. Oncol. 17(2), 108–120 (2007). 10.1016/j.semradonc.2006.11.007 [DOI] [PubMed] [Google Scholar]
- 9.Jeong J., Shoghi K. I., and Deasy J. O., “Modelling the interplay between hypoxia and proliferation in radiotherapy tumour response,” Phys. Med. Biol. 58, 4897–4919 (2013). 10.1088/0031-9155/58/14/4897 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Chi A., Tomé W. A., Fowler J., Komaki R., Nguyen N. P., Mehta M. P., and Welsh J. S., “Stereotactic body radiation therapy in non–small-cell lung cancer: Linking radiobiological modeling and clinical outcome,” Am. J. Clin. Oncol. 34, 432–441 (2011). 10.1097/COC.0b013e3181df4b3f [DOI] [PubMed] [Google Scholar]
- 11.Partridge M., Ramos M., Sardaro A., and Brada M., “Dose escalation for non-small cell lung cancer: Analysis and modelling of published literature,” Radiother. Oncol. 99, 6–11 (2011). 10.1016/j.radonc.2011.02.014 [DOI] [PubMed] [Google Scholar]
- 12.Fenwick J. D., Nahum A. E., Malik Z. I., Eswar C. V., Hatton M. Q., Laurence V. M., Lester J. F., and Landau D. B., “Escalation and intensification of radiotherapy for stage III non-small cell lung cancer: Opportunities for treatment improvement,” Clin. Oncol. 21, 343–360 (2009). 10.1016/j.clon.2008.12.011 [DOI] [PubMed] [Google Scholar]
- 13.Tucker S. L.et al. , “Late rectal toxicity on RTOG 94-06: Analysis using a mixture Lyman model,” Int. J. Radiat. Oncol., Biol., Phys. 78, 1253–1260 (2010). 10.1016/j.ijrobp.2010.01.069 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Christianen M. E. M. C.et al. , “Predictive modelling for swallowing dysfunction after primary (chemo) radiation: Results of a prospective observational study,” Radiother. Oncol. 105, 107–114 (2012). 10.1016/j.radonc.2011.08.009 [DOI] [PubMed] [Google Scholar]
- 15.Uzan J. and Nahum A. E., “Radiobiologically guided optimisation of the prescription dose and fractionation scheme in radiotherapy using BioSuite,” Br. J. Radiol. 85, 1279–1286 (2012). 10.1259/bjr/20476567 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Deasy J. O., Niemierko A., Herbert D., Yan D., Jackson A., Ten Haken R. K., Langer M., and Sapareto S., “Methodological issues in radiation dose–volume outcome analyses: Summary of a joint AAPM/NIH workshop,” Med. Phys. 29, 2109–2127 (2002). 10.1118/1.1501473 [DOI] [PubMed] [Google Scholar]