Bias and uncertainty can result from single institution non-randomized heterogeneous mixtures of patients with varying follow-up times and unknown censoring of competing risks. Throughout the past quarter of a century, over a million patients have been treated with radiosurgery on Gamma Knife alone (26), over a million more patients have been treated with SBRT on CyberKnife alone (27), and countless more have been treated on stereotactically capable linear accelerators. No excuse remains for there to be only two limited published datasets for an important critical structure like brachial plexus. It is imperative that the field of radiation oncology collects data more rigorously as highlighted by the lessons of QUANTEC (11, 10) and as continues to be emphasized by all the HyTEC papers (12, 28). In the meantime it is important to glean as much information as possible from the sparse datasets that do exist, and to pool them into increasingly larger datasets (10). A full de-identified database of 197 patients with dosimetric information and outcome for each patient was published more than 100 years ago (29), showing that it is possible to accomplish this without sophisticated algorithms. One of the first dose-response models was created more than 90 years ago from clinical data by hand on graph paper (30) even before the first electronic computer was invented; with modern automated algorithms there is no excuse to not save and analyze the data in properly designed studies with actuarial outcomes at specific time points in multiple institutions with large cohorts of data.
The dose-tolerance numbers for conventional fractionation from the Emami paper were based on expert opinion over 30 years ago, in terms of the radiation dose limits for 1/3, 2/3/ and 3/3 organ volume, with the probability of 5% (TD 5/5) or 50% (TD 50/5) risks of complications within a 5-year follow-up. The original paper did emphasize the need for more research and available data. Two decades later the ensuing accumulated published data was consolidated into QUANTEC (32) which was much more accurate owing to the growing body of cooperative trials and institutional studies. However, the improved accuracy of QUANTEC also came with increased complexity and varied format of the limits, which is difficult to use in a day to day clinical work. The goal of the DVH Risk Map (13) is to balance the convenience of a unified framework of dose tolerance limits in low-risk and high-risk categories, with the accuracy of dose-response modeling from all the emerging published clinical data, particularly in the setting of hypofractionated SBRT.
Brachial plexus dose tolerance for conventional fractionation has been studied (5, 33–34) and contouring guidelines are available (2, 35–36). The Emami limit for brachial plexus of EQD2 = 60Gy (5) corresponds to 26 Gy in 3 fractions, which is remarkably the same dose limit as recommended in the Indiana study (1). However, the paradigm has transformed from allowing 100% organ exposure at that dose in conventional fractionation (5), now all the way down to the 0% volume at the same dose for SBRT (1, 37).
About one third of the combined dataset had Dmax values in excess of 10 Gy per fraction, where the LQ model has been questioned (16). For this reason the Karolinska authors compared LQ to USC, and found no major difference for this data (2). The Indiana dataset being in terms of physical dose avoids the BED question, but is itself a major limitation of the pooled model.
Limitations of both studies include basing the data on a small cohort of patients with limited follow up. These data may not reflect the full incidence of toxicity after SBRT, because many patients might not survive long enough for the toxicity to develop or may be lost to follow-up for a variety of reasons. Another limitation is the usage of re-irradiation for some of the Karolinska cases, although this only caused one of the complications, so insufficient data were available to construct a model that could account for re-irradiation tolerance. Differences in grading of complications was acknowledged, which may contribute to inaccurate causal analysis. Half of the complications were grade 2 and only one potentially grade 4 paresis was reported in each of the two studies, but the studies did not indicate the specific grade for each Dmax value of the whole dataset so there is no way to create separate models for each grade, as was done in a brain dose tolerance study (38) for example. Furthermore, as noted in Table 1, the grading scales vary especially for the higher-grade events. A risk of 10% is higher than ideal for brachial plexus, but until the grade of each patient is reported in a consistent scale, clinicians must use their own judgement when interpreting the results.