Background
Optimization of stereotactic body radiotherapy (SBRT) plan quality is crucial to minimize normal tissue dose and hence toxicities for inoperable early stage non–small–cell lung cancer (NSCLC). In lung SBRT small photon fields are widely used, which are known to introduce significant lateral electronic disequilibrium (LED) in heterogeneous tissue caused by out–scattering electrons not being compensated by in–scattering electrons [
1]. Using the LED phenomenon for optimizing lung SBRT (LED–SBRT), which is based on the differential reductions of lung and tumor doses caused by the LED in order to steepen the dose gradients and thereby increasing the dose within the tumor while reducing the dose in the normal lung, has been recently proposed by Disher et al. [
1]. However, in order for LED–SBRT to be robust against dosimetric errors, Monte Carlo (MC) simulation techniques [
2] must be employed to accurately model the particle transport.
Besides direct MC simulation, the LED phenomenon was also investigated implicitly through the relationship with the prescription isodose line (PIL) by comparing the dose results from type–A pencil beam (PB), which cannot model the LED, and type–B MC based dose calculations [
3]. Using multiple dynamic conformal arcs (DCA), Oku et al. showed a clear dependence of the PIL on the dosimetric plan quality [
4]. They found higher lung dose and lower dose conformity at 50% downwards to 20% PIL after the optimal improvement at 60% PIL with respect to the reference of 80% PIL. Similar results of optimal PIL at ~ 60–70% and much lower at ~ 40–50% were found for forward planned linac–based and inverse planned robotic–based SBRT lung treatments, respectively [
5].
The clinical implication of lowering the PIL to 60% vs. 80% PIL in 5–fraction lung SBRT was initially studied by Takeda et al., also using DCA techniques [
6]. After 6 months post–SBRT, they found no local recurrences and only a limited number of incidents of radiation pneumonitis ≥ Grade 2 (1 out of 15 patients) with lower PIL. Furthermore, Guckenberger et al. found in their retrospective large–scale multi–center analysis with low PIL (< 80%) a significant higher freedom from local progression as compared to higher PIL (86.8% vs. 69.1%,
p = 0.005), again with lower toxicity for lower PIL [
7]. Further evidence of the clinical effects of varying PIL was supported by a number of recent published series from European and Japanese SBRT working groups that demonstrated the iso–effectiveness of SBRT treatments between those prescribing the biological effective dose (BED) > 100 Gy
10 (
α / β = 10 Gy for NSCLC) to the isocenter, to the PTV periphery with 95% coverage (
D95%), and to the GTV mean or median dose [
8‐
13].
Despite the similar local control rates reported based on different prescription concepts, the inconsistency of the conventional PTV prescription concept has been well acknowledged in the latest published International Commission on Radiation Units and Measurements (ICRU) report 91 [
14], pointing at increased variability of the internal GTV dose for lung SBRT. Although the report hinted at a possible solution by using a GTV–based prescription and a few other studies [
15‐
17] coherently showed more consistent GTV dose when re–normalizing or prescribing the treatment dose to the GTV mean or median dose or
D99%, the implications of such GTV–based re–normalization/re–prescription with respect to the PIL in the conventional PTV prescription concept has never been studied.
In this work, we focused on the technical feasibility of enhancing the dosimetric quality of inverse VMAT optimization by assessing the optimal PIL and with that the potential for dose de–escalation. Further on the hypothesis that the potential dependence of plan quality on the PIL is related to the LED, retrospective PB re-calculations were performed for all direct MC–optimized VMAT plans to assess the relationship of the PIL with the dosimetric changes. Ultimately, we tried to investigate the GTV–based re–normalization / re–prescription concept with respect to the PIL for VMAT LED–SBRT.
Discussion
This study demonstrated the technical feasibility of optimizing the PIL through its implicit relation with the lateral electronic disequilibrium (LED) to increase the dose gradient outside the target and hence improving the overall dosimetric quality in VMAT–based SBRT planning. This implicit relation was initially investigated in terms of the PTV dose differences between the PB and the MC dose calculation results for DCA–based lung SBRT [
3]. This study followed the same methodology to understand the hypothesized LED origin of the dosimetric improvement. However, following the suggestion by Dish et al. [
1] that the LED phenomenon can also be exploited in inverse optimization, our results were the first to show the dependence of PIL on LED for VMAT–type lung SBRT.
Using a reverse operation to re–calculate the type–B MC optimized dose plans with the type–A PB algorithms, as opposed to Zheng et al. [
3] who re–calculated the PB–optimized plans by the MC algorithms, we found significant increase of (negative) dose differences of mean dose in the low density lung tissue embedded in the PTV (i.e., PTV minus GTV) with decreasing PIL plans indicative of increased magnitude of LED at lower PIL. The net negative dose difference can be explained by the increased fluence using type–B algorithms at 60% PIL in striving to compensate the LED deficiency at the low density tissue dominant field edge. The increased fluence led to increased overall dose deposition that was assumed by the type–A PB algorithms. The
D
2%
of the PTV border zone also showed similar trends of increasing (negative) dose differences with decreasing PIL despite insignificant statistical differences.
Unlike forward planning with DCA techniques, the advantage of inversely optimized VMAT is that the optimal MLC aperture relative to the tumor size can be solved through an intuitive adjustment of optimization parameters that are directly related to the clinical goals. Given the set of clinical goals, the optimizer would implicitly determine the optimal extent of LED adjusting for the photon beam energy as well as the variation of lung density between patients, avoiding the manual iteration of changing mostly the isotropic MLC margin to arrive at the desired PIL level, thus improving the planning efficiency. The direct incorporation of Monte Carlo dose engines further ensured that the dose distribution was robust against dosimetric errors caused by LED. Furthermore, this optimization approach is quite simple as we aimed at optimization of the PIL by adjusting the maximum and minimum dose (volume) to the target and the constraints of several dose controlling shell structures around the PTV.
Using this inverse optimization approach, the optimal PIL showing the most rapid dose falloff, which is significantly system design and planning technique dependent, was found to be between 60% and 70%. These results were consistent with previous reports by Oku et al. [
4] showing best plan at 60% PIL. Despite the theoretical benefits of lower PIL than 60%, the associated toxicity profiles and local control remain largely unknown because it produced “hot spot” in the target beyond the acceptable range by most trial protocol [
20,
21,
27] and therefore very limited clinical data are available. With the optimized PILs at 60% to 70% in present study, the GTV mean dose almost doubled in comparison to the reference 85% PIL without increasing lung dose, although the dose between the PTV and GTV was on average higher at 70% and 60% PIL by 6% and 10%, respectively. Considering this fact, it may be of genuine concern that the normal tissues embedded in this region, i.e. the non–tumorous margin region where motion and system inaccuracies are compensated for, receive higher doses. Given the treatment prescription of 54 Gy in 3 fractions delivered in 2 weeks for this study, 95% of the PTV will receive at least a biological effective dose (BED) of 151 Gy
10, which is likely high enough to sterilize not just the tumor cells but also all other normal tissue cells as well. Yet, the increased chance of developing high grade radiation pneumonitis or fibrosis inside this small volume between GTV and PTV may be small as incidents of radiation pneumonitis or fibrosis are generally correlated with mean lung dose or low dose lung volume which were not increased at the lower 70–60% PIL [
28]. Such assumption also has support from the recent DEGRO guidelines published by the German SBRT working group for early stage NSCLC which recommended a maximum dose of 150% to the PTV (i.e., ~ 65% PIL) based on the clinical evidences from a large–scale multi–center study [
9].
Furthermore, this study clearly demonstrated that besides lung doses at optimized PIL as low as 60% doses to other serial OARs such as esophagus, heart, bronchus and trachea, major vessels, and spinal cord did not increase even for central tumors because the optimizer would automatically determine the set of anisotropic MLC margins variable with the gantry angle to achieve the specified dose constraints. This is generally impractical and labor–intensive in forward planning with DCA techniques. One of the exceptions could be tumors having the PTV overlapped with the chest walls. Nevertheless, this problem can be partly addressed by imposing more stringent dose–volume constraints on the chest wall and rib structures to push away the high dose. Also, the plan quality metric of the target dose conformity showed no statistical differences between different PILs. The dose gradient index was better for plans with 60% PIL than 80 and 85% PIL (p < 0.05) and was comparable between 60 and 70% PIL plans. The monitor units were found to increase with decreasing PIL in a linear manner likely resulting in minimally increased treatment time.
The additional advantage that comes along with optimized PIL at ~ 60 to 70% is the potential of further margin reduction. In this study, the mid–ventilation PTV was based on the van Herk’s margin recipe and calculated assuming a reference 85% PIL. The theoretical margin may be decreased by 1 to 2 mm from 85 to 60% PIL for our patient cohort whose observed motion was up to 2 cm, mainly due to the smaller
β value of the inverse cumulative standard normal distributions at the prescribed PTV minimum dose level [
19]. This margin reduction, although small, may leverage the dosimetric benefits of inverse optimization in VMAT even further, though we acknowledge that further studies are required in this regard and a discussion if the van Herk’s margin recipe can be used for inhomogeneous dose distributions is beyond the scope of this work.
Clinically, a BED of 100 Gy
10, assuming an
α / β ratio of 10 Gy, has been universally recognized as the approximate threshold dose to achieve adequate local control in early stage NSCLC, however, only based on a prescription to the PTV periphery [
7,
29,
30] and with unclear PIL. On the other hand, Guckenberger et al. [
7] recently hinted at lower PIL (< 80%) being significantly superior in local control in their data (86.8% vs. 69.1%,
p = 0.005). Recently, the same retrospective large–scale multi–center study have reported on local control being also significantly dependent of the maximum isocenter dose [
9,
10] which strongly supports our investigation of decreasing the PIL through inverse VMAT optimization and with that increasing the GTV doses. Following that concept and going one step further, a PTV prescription dose of 3 × 18 Gy at 85% PIL (151 Gy
10 to PTV D
95% and
D
max
= 198 Gy
10, reference dose level in our study) may very likely be reduced to 3 × 14 Gy at 60% PIL (101 Gy
10 to PTV D
95% and
D
max
= 233 Gy
10), resulting in substantial reductions in dose to OARs and even possibly in an increase in tumor control probability (TCP).
Based on this idea, this study further investigated the potential of significant dose de–escalation for SBRT by making best use of the physics derivable from the LED phenomenon. When we prescribed the dose in a way that GTV
D
50%
equals 54 Gy (BED = 151 Gy
10) for all PIL with the aim to keep the TCP constant, the resulting BED in the PTV
D
95%
ranged from 71.0 to 79.5 Gy
10, and 87.1 to 92.7 Gy
10 at 60% and 70% PIL, respectively. While such PTV doses may seem low in comparison to previous publications, a recent investigation of PTV prescription dose reduction with constant high GTV mean doses found high local control for lung tumors even with low PTV prescription doses (PTV D
95% BED ≥89.7 Gy
10 vs. < 89.7 Gy
10, hazard ratio 0.077, confidence interval 0.012–0.503,
p = 0.001,) [
13]. Hence, we may hypothesize that 3 × 18 Gy prescribed to the mean GTV dose at 60 to 70% PIL may still result in high LC (> 90% [
13]) while substantially reduce the dose to the lungs and other OARs.
In practice, prescription based on the GTV mean dose can be optimized for different combinations of dose fractionation schedule and PIL. For example, 54 Gy GTV mean dose could be delivered in 4 fractions at 60% PIL, producing PTV
D
95%
of 79.5 Gy
10 that is roughly equivalent to 48 Gy in 4 fractions prescribed to the isocenter as commonly practiced in Japan [
27]. For dose fractionation schedules with more fractions, 54 Gy GTV mean dose could be delivered in 5 fractions, but at slightly higher PIL at ~ 67%, producing PTV
D
95%
of 66.2 Gy
10 equivalent to 50 Gy in 5 fractions reported by Aoki et al. [
11] Alternatively, the prescription can also be re–normalized to higher median GTV dose than 54 Gy to achieve the desired PTV
D
95%
covered by the optimal PIL. Table
3 is provided to predict the 3–year local control rate from different dose schedules and prescription methods reported in the literature and in this study.
Table 3
Summary of the biological effective dose (BED) to 95% of the planning target volume (PTV D
95%
) and the 3–year local control (LC) rates in the literature
| Isocenter | 80% | Type-A | 50 | | | 66.2 | 95.0% |
| | | | | | | | 95.0% |
| Isocenter | 80% | Type-A | 48 | | 62.3 | | 90.0% |
| | | | | | | | 87.0% |
| PTV edge | 80% | Type-A | 45 | 99.0 | | | 87.8% |
| PTV edge | 80% | Type-A | 60 60 | 145.7 | | 108.6 | 89.3% 89.3% |
This study | GTV mean dose | 70% | Type-B | 54 | 89.2 | 76.7 | 69.1 | |
60% | Type-B | | 74.3 | 64.4 | 58.4 | |
Nevertheless, it is worthwhile to note that the concept of GTV–based prescription for lung SBRT and outcomes for such method are exclusively limited to robotic SBRT so far [
13,
15], although the data is part of and fits nicely to recent TCP modeling [
9,
10]. Clinical implementation of the proposed dose de–escalation approach by optimized PIL at ~ 60% with other techniques such as VMAT must be taken with great cautions, and further clinical studies are warranted to validate its efficacy and safety. It is admitted that this study did not fully address the implementation issues of GTV–based prescription. Future studies would be required to develop a link of the GTV prescription with the conventional concept of PTV prescription isodose line, the dose encompassing level of the PTV and their interactions with other patient–dependent / treatment technique–specific factors such as tumor motion range, PTV definition, etc. Further limitations to this study come from the limited number of presented cases as other parameter such as lesion location, volume and dimensions or density of the lesions itself could not have been statistically investigated.