In univariable analysis, we found the type of treatment and the EQD2 to be independently associated with radiation-induced effects. When the type of treatment was incorporated into the diagnostic model, only radiotherapy with adjuvant PCV remained statistically significant. Since monotherapy radiotherapy was chosen as reference category, this means that patients who received radiotherapy with adjuvant PCV have higher odds of developing radiation-induced effects than patients who received monotherapy radiotherapy. The association between type of treatment and radiation-induced effects is explained by the literature, which reports a three times higher risk of developing radiation-induced effects when radiotherapy is combined with chemotherapy. The explanation for this is that radiation causes failure of the blood-brain barrier. Hereby, the tumor area is better accessible for chemotherapeutics, resulting in higher chances of developing radiation-induced effects.9,22,23 Furthermore, Levin et al. treated glioma patients with hyperfractionated radiotherapy with concomitant carboplatin and adjuvant PCV and reported one of the highest incidence of radiation-induced effects known in literature of 21 percent.24 We also found the EQD2 to be significantly associated with radiation-induced effects in both univariable as multivariable analysis, meaning that the odds on developing radiation-induced effects are higher when a higher EQD2 is given. Throughout the literature it is extensively known that radiation-induced effects is directly associated to the received radiation doses.22,25
In univariable analysis, we found increased
marginal enhancement of the surgical cavity, multiple new enhancing lesions and new or increased subependymal enhancement to be significantly associated with tumor progression. The two latter MRI-characteristics are supported by the literature.6,7 Mullins et al. found subependymal enhancement and multiple new enhancing lesions to be predictive for tumor progression when combined with callosal involvement.7 Young et al. also found subependymal enhancement to be independently predictive for tumor progression.6 The association between increased marginal enhancement of the surgical cavity and the disease status was not investigated before. In this study, we provide evidence for association between increased marginal enhancement of the surgical cavity and tumor progression independently as well as when incorporated in our diagnostic model.
In multivariable analysis, we additionally found
time to progression to be significantly associated with tumor progression. This means that with every passing month the odds of developing tumor increase by a factor 1.112. The longer the progression-free period, the more odds of developing tumor progression are present. This result is in accordance with the prognosis of high-grade gliomas.26-28
In multivariable analysis of the subgroup of patients that were treated according to the Stupp-protocol5, we did not find an association between the EQD2 and radiation-induced effects. This is easily explained by the fact that every patients was treated according to Stupp-protocol and therefore received an equal EQD2 of 60 Gy.5
The multivariable analysis of the subgroup patients
that received clinical and/or radiological follow-up showed some differences from the multivariable analysis of the total study population. Time to progression and increased marginal enhancement of the surgical cavity did not show a significant association with tumor progression in this subgroup for which we do not have an explanation. Furthermore, enhancement crossing the midline revealed a significant association with tumor progression in this subgroup. This might be explained by the fact that patients with this determinant are not eligible for a re-resection and therefore relatively more often end up in this reference group, causing an almost significant association (p = 0.071) in the total study population to become just significant in this subgroup (p = 0.043).
Since we found multiple differences between the
univariable analyses of the subgroups regarding the reference tests, this leads us to question whether the use of two reference tests possibly introduces bias. The vast majority of patients with a progressive or new enhancement received clinical and/or radiological follow-up. However, when the progression scan is very suspicious for tumor progression, despite remaining doubt that it is caused by radiation-induced effects, the patient is discussed to undergo a re-resection and therefore histological follow-up. This leads to a pre-selection of the groups, of which the patient discussed for histological follow-up have a higher probability of having tumor progression. Even though histological follow-up is the reference standard throughout the literature, histological follow-up is risky for the patient and can also lead to sampling error. We believe that the use of two reference standards reflects the clinical setting best and is therefore justified.
Strengths and limitations
We reported the largest study population that was reviewed for differentiation between tumor progression and radiation-induced effects in patients with an irradiated high-grade glioma with conventional MRI-characteristics. Due to the relatively large amount of patients, gathered over the course of almost 6 years, we collected a representative group of high-grade glioma patients. This leads us to believe that, our diagnostic model will be generalizable. Thereby, when comparing the baseline characteristics of one of the largest glioblastoma randomized controlled trialswith the baseline characteristics of our study, the generalizability is supported as well.28 Moreover, the results of the multivariable analysis of the Stupp-subgroup5 and the grade 4-subgroup are in complete accordance with the results of our total study population, which also pleads for generalizability of our results. Our model has not yet been validated.
Our study had some important limitations as well.
First of all, the patients were enrolled consecutively, however, when patients eligible for inclusion had a strongly suspicious progression scan for tumor progression, resulting in directly changing the course of management, they were excluded from this study. This might have resulted in bias regarding the patient selection since the patients with obvious tumor progression, accompanied by certain MRI-characteristics, were excluded. We believe, however, that this choice is justified since the progression scans of these patients apparently did not provide hesitance regarding the disease status, while the purpose of this study is to review the MRI-characteristics of the progression scans that provide doubt between tumor progression and radiation-induced effects.
Secondly, collection of the patient characteristics
and the interpretation of the MRI were performed before retrieving the outcome. Nonetheless they were both interpreted by the same author (KvL) which might lead one to doubt this blinding strategy. Some characteristics were, however, not susceptible to interpretation, e.g. age, sex, histological tumor type, tumor size etcetera.
There was no blinding for the clinical information and the index test results at the time of the evaluation of the reference test. Nonetheless, we strictly pre-specified the criteria for both tumor progression as radiation-induced effects and believe that the evaluation of the reference tests were therefore not susceptible for personal interpretation.
Thereby, we tried to compensate for both these risks of bias by checking all uncertainties and a random sample size of MRI-characteristics and disease status by a blinded second observer (TJS).
Lastly, we did not take the use of dexamethasone in
the clinical and/or radiological follow-up group at the time of the reference scan into account. This might have led to an under- or overestimation of the radiation-induced effects, since dexamethasone is given for relieving symptoms by reducing edema due to radiation-induced effects.29,30 However, retrieving this data was not possible due to inconsistent notation of the doses dexamethasone in the electronic patient files.