Re-irradiation for recurrent glioma- the NCI experience in tumor control, OAR toxicity and proposal of a novel prognostic scoring system

Survival following concurrent radiation (RT) and temozolomide (TMZ) as per the EORTC 26981/22981-NCIC CE3 trial remains poor with median survival ranging from 39 months in RPA (Recursive Partitioning Analysis) class I patients to 5.2 months in RPA class VI patients [1]. Despite improvements in the detection of distinct molecular signatures [2], increased precision in the administration of radiation therapy and the increasing availability of targeted and non-targeted systemic agents, additional improvements are still needed.

Brain tumor recurrences can be identified by the development of new neurological symptoms, radiographic changes or both [3]. Upon recurrence, the treatment recommendations can vary widely and are partially based on the patient’s performance status, tumor location, and time interval since last treatment. Depending on these factors, options may include re-resection, chemotherapy, re-irradiation (re-RT) or enrollment on a clinical trial [4]. However, tumor re-resection is possible in less than 50% of patients [5] and the response to systemic treatment, if it occurs, is short lived with overall survival (OS) from 7.1 to 9.6 months [4]. For a significant proportion of patients with recurrent glioma in whom re-resection is not possible and for whom systemic options have been exhausted, re-RT has emerged as a possible treatment option.

In multiple retrospective trials, re-RT for brain tumors has been shown to be feasible, with a median OS benefit of 8 months, a progression free survival (PFS) of 5 months and minimal reported significant toxicities [6‐15]. Acute toxicities including alopecia, headaches and nausea/vomiting, were mild and well managed with medical therapy, while late CNS toxicity was reported in most retrospective studies at a rate of less than 5%.

Existing scoring systems that may help guide patient selection for re-RT have been validated in some cohorts [10] but not in others [11]. Whilst they all include some common features rated as important by glioma re-RT experts (personal communication), they also differ with respect to inclusion of resection status, recurrent tumor size, patient symptoms and steroid use.

Toxicity attribution in patients with CNS tumors is challenging as both tumor progression and symptomatic radiation necrosis can cause significant clinical decline in the setting of short survival which limits long term evaluation. The current toxicity reporting structure is the Common Terminology Criteria for Adverse Events (CTCAEv4) [16]. For the three organs at risk (OAR) involved when contemplating CNS re-RT, the optic nerves, chiasm and brainstem, the toxicity scales remain imprecise or absent all together. In addition, most re-RT for recurrent glioma is by necessity administered in hypofractionated schedules where toxicity data and the ability to estimate normal tissue complication probabilities is limited. Since 50% of superiorly selected patients can live up to 2 years following re-RT as the systemic therapies become more effective, toxicity estimation in the setting of re-RT will become increasingly important.

In this study, we examined patient outcome using patient and recurrent tumor related factors as well as OAR dose and clinical toxicity information with NTCP calculations in patients who underwent re-RT with commonly used dose fractionation schemes. Despite the lack of a clear definition of toxicity [17‐19], While NTCP calculations suggest the possibility of significant toxicity in some patients, clinical data suggests minimal toxicity following re-RT. Our novel scoring system based on the ability to control recurrent tumor, risk of toxicity as well a more traditional scoring parameters (age, histology, symptomatic, disease free interval) with re-RT, may represent a step forward in the selection of patients for re-RT.

With the use of advanced therapy techniques and careful treatment planning, for the most part, physicians are able to minimize dose to previously radiated OAR allowing safe re-RT of brain tumor recurrences with an improvement in OS and PFS (6–15). According to the nomogram developed by Combs et al., 50% of patients with a score of 2 (i.e. GBM patients <50 year old) survive at least another 12 months from the time of re-RT [1]. Thus, a more thorough understanding of the potential toxicities that may occur after tumor re-RT are needed. Although acute toxicities in the form of alopecia, headaches and nausea/vomiting have been reported, they are self-limited and controlled with medical management. Likewise, late CNS toxicity remains underreported but in most retrospective studies is less than 5%. In our cohort of 31 patients we had no grade 3, 4 or 5 acute or late toxicities with a median PFS of 4 months (range 0.5–103 months) and OS of 6 months (range 0.7–103 months), a range in keeping with both existing validated prognostic scores [1, 11] and retrospective data.

When working on a re-RT plan for a CNS patient, the physician can use two normal tissue dose reports Emami, 1991 [27] and Quantec 2010 [17‐19] to help them decide the safety of the proposed treatment as well as the side effect profile to report to the prospective patient. Additionally, mathematical models using the Lyman–Kutcher–Burman NTCP [21] based on dose have been implemented in an attempt to calculate NTCP [28]. However, several problems arise with toxicity attribution. These include the lack of reliable toxicity estimation, the relatively short PFS and OS times, the overlapping symptoms of re-RT toxicity and tumor recurrence and the concurrent or adjuvant administration of systemic agents that can alter radioresponse and toxicity measurements. Our analysis suggests that the Emami and Quantec papers and the NCTP calculations appear to overestimate the rates of clinical toxicity.

To our knowledge ours is the only paper examining NTCP and cumulative dose in the setting of re-RT for gliomas. Calculation of the cumulative dose to OAR by creating a “plan sum” for the two treatment plans was found to often result in erroneous cumulative dose to OAR hence manual summation of dose delivered was employed to calculate NTCP in this study. Nonetheless, the calculated NTCP using maximum doses to OAR between the two different treatment plans was as high as 100% in the case of all four major OAR. An expected difference in NTCP was observed when mean dose (EUD) [29] versus max dose were employed for its calculation. Considering the lack of toxicity observed, we propose that NTCP calculations based on EUD may represent a more accurate estimation of risk. Mean dose may be both more accurate and more realistic [30, 31] due to its decreased dependency on biologic parameters, including alpha/beta ratio. This raises the question as to whether the OAR in question behaves as serial, parallel or serial-parallel in terms of toxicity and this is yet unclear [32, 33].

Additionally, neither dose per fraction nor time between the two radiation treatments, or the use of concurrent agents all of which may play a significant role in the development of treatment related toxicity, are currently accounted for when calculating NTCP. The impact of concurrent chemotherapy is unclear in the re-RT setting and may further defined by results of ongoing trialssuch as RTOG1205) which explores concurrent bevacizumab, NCT02709226 (Krauze et al.) allowing both concurrent temozolomide and bevacizumab and upcoming trials that explore concurrent temozolomide. Furthermore, most patients who receive re-RT tend to be younger with superior performance status and patient age has been shown to have some correlation with the development of toxicity or lack thereof [17]. This is an important factor when considering the lack of toxicity noted in this study. Limitations to our data include long recruiting times for the patient cohort and inability to carry out NTCP calculations in all patients Long recruiting times for the patient cohort add to the heterogeneity of the data, they are however unavoidable as patients who are referred for re-RT are often referred by virtue of the fact that they have a long disease free interval and a change in the treatment technique is therefore also more likely as technology evolves. However, our data does not suggest that technique played a role. We did note that patients who received 3D conformal RT the first time were more likely to receive IMRT on re-RT in order to decrease dose to OAR. NCTP calculations were only possible for 25 of 31 patients due to the following limitations: 1) inability to obtain the first radiation treatment plan often due to a longer time period since previous treatment (>5 years) or loss of the previous planning data and 2) use of a non conventional fractionation scheme, ie a hypofractionated scheme wherein usage of the linear quadratic equation for BED calculation is not generally accepted [20]. This does limit the results of the study in that it reduced the overall data available, however overall the results obtained do reflect a more homogenous set of dose fractionation schemes, which are employed in ongoing and upcoming prospective trials. It is likely that the inclusion of patients who were treated with hypofractionated schemes would potentially alter the conclusions since a higher level of late toxicity may be postulated. Of note, previous treatment plans are often labor intensive to difficult to obtain and integrate with the re-irradiation plan and thus re-irradiation may be practiced in the community without evaluation of the previous plan raising the issue of cumulative dose to OAR and hence the lack of retrospective data to produce superior models for NTCP in the re-RT setting.

The identification of true toxicity and its relationship to dose, will require 1) greater reporting of dose to OAR after re-RT and 2) robust testing for potential OAR toxicity, including visual field testing, audiology, neurocognitive function and quality of life assessment. Baseline visual field testing although inconsistently obtained is more commonly carried out and thus may more readily provide information on OAR toxicity in the short term. However, visual evoked potentials may represent a more accurate modality of assessing toxicity to the visual apparatus and bears consideration for inclusion in prospective clinical trials [32]. The lack of such data in the literature delegates estimation of risk to retrospective studies and models that are inadequate thus, making already challenging patient discussions and decision making, even more so.

When applying either the Carson 2007 [34] or the Combs 2013 [1] scoring system to our data, we did not achieve a statistically significant separation of the curves (Additional file 3: Figure S2) as has been noted in other published dataset comparisons [9], this however may reasonably be related to the small sample size in our study as compared to that of other authors Unlike the Carson 2007 scoring system, the Combs 2013 scoring system does not include KPS or steroid use but does include time since previous RT in addition to age and histology. Our scoring system considers the likelihood of controlling the recurrent tumor (target control) based on tumor size, tumor location and presence of diffuse disease parameters derived from our own practice and review of the practice and that of other radiation prominent oncologists (personal communication). In our analysis, the tumor control subgroup was highly significant for OS. In addition, we have shown that while KPS matters in terms of OS with re-RT, it is the patient’s symptomatic or asymptomatic status, which may be a more important surrogate of both PFS and OS. This indicates that parameters, currently not included in existing scoring systems may represent reasonable additions to a universally validated scoring system. The use of steroids is a challenging parameter for OS or PFS or for that matter as a study endpoint, as a significant proportion of patients is on steroids prior to re-RT, some for symptoms, others prophylactically. We did not find steroid use to be a useful parameter in our study. Steroid use may be helpful although the heterogeneity of its use in terms of doses, dose increases and the inconsistent capture of steroid usage details may make its consistent comparison and inclusion in scoring systems challenging.

This scoring system was developed out of the need to find a better way to select optimal patients for re-RT who stand to gain palliative benefit and for whom the risk of re-RT is considered acceptable. Further validation in larger cohorts will be required to validate and refine target based scoring systems such as this one.

No treatment related complications were observed in re-RT patients included in our study. Existing NTCP calculations based on maximal point dose and mean dose (EUD) to OAR may overestimate NTCP in the re-RT of high grade glioma. Ongoing heterogeneity in the approach to patient selection for re-RT underscores the need for a universally validated tumor and patient characteristics based scoring system.