In our series, the overall 3, 5, and 10-year local tumor control rates were 98% (95%, 100%), 92% (85%, 98%), and 77% (66%, 89%) respectively. Our cohort of patients were mainly females as meningioma are more common in females [18].To our knowledge, our series is the first series to specifically relate BED and BED/margin ratio to local tumor control of para-sellar meningioma after stereotactic radiosurgery with a median follow-up of 46 months. We found that both of these variables were significantly associated with local tumor control.

Parasellar meningioma are the most common tumors of parasellar region and constitute 15% of all meningioma [19]. A multidisciplinary approach is strongly recommended for meningioma of such a location as gross total resection is usually not feasible without incurring significant morbidity [14, 20]. Management options include watchful observation, open surgical resection, endoscopic resection, or stereotactic radiosurgery (SRS) [14]. With a reported resection associated mortality of 7% and morbidity of 60% for parasellar meningiomas, SRS has become a favored option as a primary, adjuvant, or salvage treatment of parasellar meningioma [15, 21, 22]. Based on a recent International Stereotactic Radiosurgery Society (ISRS) Practice Guideline of cavernous sinus meningiomas, SRS was considered a safe and effective option, with 5-year progression free survival ranging from 86 to 99% and 10 year progression free survival ranging from 69 to 97% [23]. The ISRS guideline recommended margin doses from 11 to 16 Gy (level III evidence) [23]. The margin doses and local control rates in our series are consistent with these results.

However, the small percentage of local failures in both the ISRS review and our current series requires further investigation. Multiple factors have been reported to affect outcome of SRS in parasellar meningioma including: tumor volume, prior surgeries, margin dose, histological grade, progression after microsurgery, Ki-67 index, and extent of resection [24]. For this study, we hypothesized that BED may also be associated with the probability of local control.

Fractionated radiation therapy has long relied on the linear-quadratic (LQ) model to describe the relationship between the proportion of cell-kill and absorbed [24]. While the mechanisms underlying the LQ-model remain uncertain [25], as a model it has proven to be a useful mechanism to describe the relative radiosensitivity of different tumors and normal tissue types. The BED concept is derived directly from the LQ-model and is defined as “the total dose required to give the same log cell kill as the schedule being studied, at an infinitely low dose rate or with infinitely small fractions well‑spaced out” [24, 26]. The BED model has been generalized to account for cell-repopulation effects by adding terms to describe repopulation that occurs during and between radiation dose delivery.

However, until recently there has been little interest in investigating the BED concept in the setting of SRS. SRS is conceptually quite different from fractionated radiosurgery in that an ablative dose of radiation is delivered in one or a small number of treatment fractions, and not at infinitely low dose rates [27]. SRS has proven to result in high probability of local control across a variety of indications, yet there are still a small subset of sub-optimal responders, even when all patients are treated with uniform margin doses of radiation [9, 28, 29].

GKRS in particular has time/dose-rate dependencies which could hypothetically impact the radiobiological effectiveness of a treatment. GKRS uses cobalt-60 as a radiation source, which is a radioactive isotope with a half-life of 5.26 years. Thus, in GKRS the dose rate is variable, and treatments which are given the same margin absorbed dose but delivered 5.26 years apart will have dose rates that differ by a factor of two. In addition, the shot-by-shot delivery of a Gamma Knife system means that there are pauses in radiation delivery as the system prepares to move the patient from one stereotactic location to another [30]. The amount of “non-irradiation” time this takes depends on the Gamma Knife system used. For instance, the model U Gamma Knife required a manual procedure to set a stereotactic coordinate, mount the correct tertiary collimator on the unit, and delivery the radiation [31]. This procedure could take many minutes to accomplish. By contrast, the model Icon Gamma Knife currently used at our institution moves the patient and changes collimator settings in a completely automated fashion that takes only seconds [32]. It seems reasonable to hypothesize that the amount of cellular repair of radiation damage could significantly differ between historical and recent Gamma Knife technologies, and this may explain some of the cause for local failures in GKRS for a given margin absorbed dose [9]. While the high doses and small number of fractions that characterize SRS depart from the assumptions of the LQ-model and BED models, there is evidence that demonstrates dose rate may still play an important role in SRS efficacy. For example, irradiation of 10 cm of spinal cord of pigs with two different dose rates (25 Gy in 25 min vs. 140 min) resulted in 0% radiation induced myelopathy in the longer time plan1. Simply put, with a longer treatment time, the lower the biological effective dose and the more the repair of sublethal radiation induced changes [1, 9].

There have been several prior attempts to create BED models tailored for GKRS delivery, and prior studies have found mixed results when applying these models to observed clinical outcomes. Millar et al. calculated BED based on an equation including dose rate, inter-isocenter time, exposure time, sublethal radiation damage repair slow and fast rates, total dose [1, 33,34,35]. Other models were used to calculate BED by Jones and Hopewell [16] as well as Graffeo et al. [29]. Initial studies by Graffeo et al. [29] and Balossier et al. [28] found an association between BED and biochemical remission in acromegaly patients which was statistically significant only in the 2nd study. Dumot et al. concluded that BED is a strong predictor of endocrine remission in acromegalic patients with a cut off value of > 170 Gy2.47 [9]. Hopewell et al. found variation of 15% for a physical dose for an individual patient with vestibular schwannoma [7]. The same group suggested that treatment with smaller number of iso-centers and shorter treatment time achieved the highest biological effective dose [7].

Tuleasca et al. found a clear association between BED and the incidence of hypesthesia in trigeminal neuralgia patients with incidence of 42% after approximately 2600 Gy2.47 [36]. In another study, Tuleasca et al. showed BED as a strong predictor of obliteration of unruptured AVM after GKRS [13]. In our series, while margin dose was not statistically significant as a predictor of local control, BED and the BED/margin ratio did show a statistically significant association. Huo et al. found significant association between BED > 50 Gy2.47 and local control for intracranial meningioma but their series wasn’t specific for para-sellar meningioma [12]. The same group showed that margin dose > 12 Gy did not achieve better control, but it was associated with more incidence of edema and radiation induced changes [12, 37]. In our series, we found a BED threshold of 68Gy2.47 was correlated with better tumor control specifically for parasellar meningiomas. As such, optimization of BED at the time of radio-surgical planning for parasellar meningiomas may serve as an avenue for improving long-term tumor control.

While our study is the first to address specifically relation of BED to parasellar meningioma outcome, we recognize the study has several limitations. Most importantly, our study is retrospective, and it is limited in sample size. This may introduce bias into the results in the form of selective inclusion and limited statistical power.

The assumptions that we used for parameters in the two-compartment BED model each have some uncertainty, Significantly, the α/β ratio for meningioma and slow/fast repair half-lives are estimates from the simplified BED model used in our investigation. The inter-shot times we utilized for each Gamma Knife are estimates from our experience. However, the BED model is fairly insensitive to small differences in these parameters [1, 38].

As mentioned above, the many fraction/low dose rate assumptions of the BED model depart significantly from the way GKRS is delivered. Also, the simplified 2-compartment BED model employed in our study assumes the entire tumor receives a uniform dose (and thus a uniform BED), and that the entire tumor is receiving the same dose/BED whenever radiation is being delivered, rather than in the shot-by-shot method employed during GKRS. The shot-by-shot delivery of the Gamma Knife creates a complicated relationship between a given tumor’s volume and shape and the treatment delivery time, which in turn is related to the BED. This relationship depends on many factors, including the current activity of the Gamma Knife 60Co sources, the number and size of the shots used in the treatment, and the prescribed margin dose. However, several studies have now demonstrated that BED may be significantly associated with outcome, and our study is consistent with those results. Our results showed an association between BED and local control as well as BED normalized to margin dose and local control, while failing to demonstrate an association between margin dose and local control. In addition, there is a colinear relation between margin dose and BED which may affect confidence intervals and significance. However, further investigation using larger datasets and across a wider range of dose rates and BEDs is required to better understand the magnitude of any association.

One practical limitation of our study is that the BED calculation is performed off-line, as there is not currently functionality within GammaPlan to perform real-time BED calculations. While calculation of the simplified model used in this study has negligible calculation time, there is some data collection overhead that makes it unlikely to become a standard clinical practice without further refinement and integrated functionality within GammaPlan.