Predictive value of brain tumor reporting and data system scores for functional outcomes and progression-free survival in glioblastoma

Abstract

Background

The Brain Tumor Reporting and Data System (BT-RADS) is a standardized reporting system developed for routine clinical use for MRI interpretation in patients with primary brain tumors. This study evaluates the BT-RADS predictive value in glioblastoma (GBM) for patient functional outcomes assessed through the Karnofsky Performance Scale (KPS) score and PFS.

Methods

This retrospective study includes 94 adult patients with newly diagnosed GBM at Emory University Hospital from January 2019 to January 2023. Predictor variables included BT-RADS scores recorded from MRIs ∼3 months (BT-RADS_A) and ∼9 months post-resection or biopsy (BT-RADS_B), with scores classified as low (BT-RADS score 0-3a) or high (BT-RADS score 3b-4). Outcome variables included PFS and KPS scores at 1 year post-diagnosis, with scores of 70-100% classified as favorable and 0-60% classified as unfavorable. Covariates included O6-methylguanine-DNA methyltransferase (MGMT) methylation status.

Results

In a univariate logistic regression analysis, BT-RADS_A was a significant predictor (P = .04) in discriminating an unfavorable KPS from a favorable one. In a multivariate logistic regression, the best model fit was the model containing BT-RADS_A and tumor MGMT methylation status (P = .03). In a model of BT-RADS_A and BT-RADS_B, accounting for MGMT methylation status, the risk of tumor progression was 50% lower (P = .003) with a low BT-RADS (0-3a) in the BT-RADS_A group and 38% lower (P = .03) with a low BT-RADS (0-3a) in the BT-RADS_B group.

Conclusions

BT-RADS scores can predict functional outcomes and PFS in newly diagnosed GBM patients at ∼3 and ∼9 months post-diagnosis. Future studies are warranted to validate these findings and refine predictive models.

Key Points.

• BT-RADS scores predict functional outcomes and PFS in newly diagnosed GBM.

• BT-RADS scores from 3 months post-diagnosis MRIs predict favorable and unfavorable KPSs.

• Low BT-RADS scores from 3 and 9 months post-diagnosis MRIs predict GBM progression risk.

Importance of the Study

The Brain Tumor Reporting and Data System (BT-RADS) has been developed as a novel standardization system in primary brain tumor MRIs. The BT-RADS association with functional outcomes and progression-free survival (PFS) in GBM remains unclear. In this 94-patient single-center retrospective study, we evaluated the association of BT-RADS with Karnofsky Performance Scale (KPS) and PFS. We assessed BT-RADS scores at 3 months and 9 months post-treatment intervals in newly diagnosed GBM patients and examined their predictive value alongside MGMT methylation status. BT-RADS scores recorded from 3 months post-diagnosis MRIs can discriminate between an unfavorable KPS from a favorable one. Low BT-RADS scores recorded from 3 and 9 months post-diagnosis MRIs predict the risk for GBM progression. Our findings demonstrate that BT-RADS can identify patients at higher risk of functional decline and disease progression, underscoring its potential role in guiding prognostication and multidisciplinary care in neuro-oncology.

Glioblastoma (GBM) is the most common primary malignant tumor of the adult central nervous system (CNS) in adults.¹ According to the 2021 WHO CNS Tumor classification, GBM is defined as a diffuse glioma with a wild-type isocitrate dehydrogenase (IDH) plus at least one criterion among necrosis, microvascular proliferation, TERT promoter mutation, EGFR gene amplification, and the combined gain of chromosome 7 and loss of chromosome 10.² The standard treatment protocol for GBM consists of maximal safe surgical tumor resection, followed by 6 weeks of standard brain radiotherapy (RT) with concurrent adjuvant oral temozolomide (TMZ).^3,4 Once initial chemoradiation is completed, patients with stable disease may undergo up to 6-12 months of TMZ maintenance regimen.^5-7 O6-methylguanine-DNA methyltransferase (MGMT) influences response to the alkylating agent temozolomide and prognosis in GBM, with methylated MGMT predicting an overall survival (OS) of 15.5 months and progression-free survival (PFS) of 9.4 months, and unmethylated MGMT predicting an OS of 11.1 months and PFS of 6.3 months.^3,8 Patients with clinical or radiographic evidence of disease progression may undergo various treatment options, including re-resection, bevacizumab administration, or salvage therapy with the alkylating agent lomustine.^9,10

Posttreatment GBM MRIs characterize treatment response, identify tumor recurrence, and the potential need for treatment modification. The overlap in MRI findings between treatment effect and true tumor progression poses challenges to MRI interpretation.^11,12 Subjectivity in MRI interpretation and reporting underlies a limited inter-rater reliability and represents an additional variable challenging standardized longitudinal patient care.

To standardize treatment response, imaging interpretation criteria have been developed across the years in clinical trials. The oldest McDonald criteria, introduced in 1990, followed by RANO criteria in 2010, and its novel version, RANO 2.0, introduced in 2023, were radiologic criteria implemented to increase the reliability of response assessment.^1,3-15 However, these criteria are mainly used in clinical trials, and despite some centers having introduced RANO criteria in routine practice, there are limitations related to the complexity of the criteria, leading to inter-rater variability, the subjective nature of some criteria, and oversimplification of patient assessment.¹⁶ In response to this, the Brain Tumor Reporting and Data System (BT-RADS) was developed to standardize brain MRI reporting, facilitate consistency in post-treatment brain tumor evaluation, and achieve more widespread general use in radiology reports.^17-20 The BT-RADS quantifies MRI findings into scores ranging from 0 to 4, stratifying the level of suspicion for disease progression based on several factors, including changes from prior imaging, bevacizumab use, and recency of radiation therapy.¹⁸ Kim et al investigated the relationship between overall survival and BT-RADS and demonstrated that patients with higher BT-RADS scores, indicating disease progression, have lower median survival rates, with the risk of death 1.5 times higher with each stepwise increment in BT-RADS score.¹¹ The BT-RADS and related resources are available online (www.btrads.com).

Nonetheless, data on the relationship between BT-RADS and GBM functional outcomes or PFS are limited. The Karnofsky Performance Scale (KPS) is a risk stratification tool widely used in the neuro-oncological clinical practice for the assessment of GBM patients’ performance or functional status.²¹ KPS scores are an independent prognostic factor for PFS and OS, with higher KPS scores associated with longer OS and PFS, and are essential for treatment and follow-up planning and clinical trial enrollment.^22-27 As such, investigating the association between BT-RADS and KPS is relevant for prognosis assessment and GBM management.

In this study, we investigated the role of BT-RADS in predicting KPSs at one year post GBM diagnosis and the association between BT-RADS and PFS in the same patient cohort.

Methods

Study Design and Patient Selection

This HIPAA-compliant retrospective review was approved by the Institutional Review Board of our institution. Adult patients who underwent surgical resection for newly diagnosed GBM, IDH-wildtype (confirmed through pathology testing) at a single center between January 2019 and January 2023 were included. In selecting GBM patients, the authors followed the 2021 WHO CNS Tumor classification, which defines GBM as IDH wild-type diffuse gliomas.²

Inclusion criteria were completion of an initial 6-week chemoradiation treatment course, availability of post-treatment brain MRI reports with assigned BT-RADS scores, and assigned KPS scores 1 year post-tumor diagnosis. The exclusion criteria were patients who had a diagnosis of secondary or benign brain tumors, patients who had not completed the initial 6 weeks of chemoradiation post-surgery, and those without posttreatment MRIs or BT-RADS scores. Patients’ demographics and clinical information, including sex, age of diagnosis, race, tumor MGMT methylation status, and use of salvage chemotherapy, were also recorded.

Predictor Variables

Patients were scanned on one of several 1.5T or 3T MRI scanners, using a standardized brain tumor MR imaging protocol with and without intravenous contrast agent. The imaging protocol included: sagittal 3D T1w pre-contrast MP-RAGE 1 mm slices reformatted into axial and coronal images, sagittal 3D FLAIR 1 mm slices reformatted into axial and coronal images, axial DWI 5 mm slices, axial gradient echo T2w 5 mm slices, axial fast spin echo T1w 5 mm slices, axial T2w fast spin echo fat-suppressed 5 mm slices, and sagittal 3D T1w post-contrast MP-RAGE 1 mm slices reformatted into axial and coronal images. BT-RADS scores were assigned during routine clinical care by the interpreting radiologist and included in the radiology report. Reports were scored based on the published criteria. BT-RADS scores are assigned routinely to all brain tumor patients receiving clinical care in our institution. Predictor variables were BT-RADS scores assessed at 3 months post-diagnosis (named BT-RADS_A) and BT-RADS scores assessed at 9 months post-diagnosis (named BT-RADS_B). The date of diagnosis corresponds to the date of tumor resection or biopsy. In fact, even if the final pathology results at a later time than surgery, the date of diagnosis corresponds to the date the tumor tissue is obtained for pathology evaluation, instead of when the pathology final report is released.

BT-RADS scores were also classified as low (BT-RADS score 0-3a) and high (BT-RADS score 3b-4). In the original BT-RADS paper by Kim et al a score of BT-RADS 3b had the highest likelihood of worsening on follow-up scans, with a majority of patients (67%-81%) with higher grades of radiographic progression (3b, 3c, and 4) demonstrating worsening imaging findings on subsequent MRI reports.¹¹ On the other hand, patients classified as BT-RADS 3a, reflecting pseudoprogression, or worsening of imaging findings within 90 days of completing radiation therapy, more frequently demonstrated stable (35%) or improving (22%) findings on follow-up imaging. These findings established a natural inflection point in the scoring system, where categories 1-3a indicate stable or improving disease, whereas 3b-4 reflect an increased concern for true tumor progression. Based on this evidence, the groups of BT-RADS 1-3a, classified as low BT-RADS scores, and BT-RADS 3b-4, classified as high BT-RADS scores, have been chosen.

Outcomes

Two outcome variables were assessed—KPS score at 1 year post-tumor diagnosis, and PFS. KPS scores were assigned by the clinical neuro-oncology team at the time of clinical evaluation and were extracted from the patients’ records for this study. KPS scores range from 0% (deceased) to 100% (normal, no signs of disease) in ten-point increments²¹. KPS was recorded as “at least 70%” or as numeric variables in increments of 10% (0%, 10%, 20%, 30%, 40%, 50%, and 60%). As a retrospective study, we did not have enough information to reclassify the “at least 70%” group. Therefore, we defined a favorable KPS score as “at least 70%” and an unfavorable KPS score as 0%-60%, as was also used by Lutterbach et al.²⁸ PFS was calculated for each patient as the difference between the initial chemoradiation treatment start date and the date of diagnosis of disease progression. Disease progression was defined as radiographic progression, often classified as BT-RADS scores of 3c or 4, with the date of progressive disease defined as the first time that criteria were met. For patients not diagnosed with disease progression, the date of their final clinic visit was used in the PFS calculation.

Covariates

Covariates included age at tumor diagnosis, sex, race, tumor MGMT methylation status, and use of salvage chemotherapy. These were manually extracted from the electronic medical record (EPIC Health Systems).

Statistical Analysis

The associations between categorical variables and KPS outcome were analyzed using the chi-square test of association. The non-parametric Kruskal–Wallis test was used to determine the association between the continuous variables and KPS outcome. The correlation between BT-RADS_A and BT-RADS_B was assessed using Spearman correlation. Univariate and multivariate logistic regression analyses were done by modeling KPS outcome with stepwise addition of predictors and covariates. PFS curves were assessed by Kaplan–Meier methods, using BT-RADS as a predictor, and comparisons between low and high BT-RADS scores were done with log-rank tests. The Cox proportional-hazards regression model was used to evaluate the predictive value of each predictor on PFS independently, in combination with each other, and with MGMT methylation status. An alpha level of 0.05 was set as the level of significance. SAS software version 9.4 (SAS Institute) was used for all analyses.

Results

A total of 94 patients were evaluated in the study–median age, 59.5 years; interquartile range, 51.8-65 years. Forty-eight percent (45 of 94) of the patients were males, 51% (48 of 94) of the patient population had methylated MGMT tumor status. By the end of the study period, only 13% (12 of 94) had received salvage chemotherapy. Of those patients treated with salvage chemotherapies, 10 patients received lomustine alone, one initially received lomustine before transitioning to cisplatin/etoposide, and one received irinotecan alone. The median KPS score at one year was 60% (interquartile range, 60%-70%), and the median PFS was 240 days (interquartile range, 142-348 days). The most commonly assigned BT-RADS score in the BT-RADS_A group was 3a (39 of 94), and that for the BT-RADS_B group was 2 (24 of 94) (Figure 1).

Figure 1.

Bar graph showing relative frequency of BT-RADS scores (from original reports) among selected MRI exams (n = 94) 3 months (A) and 9 months (B) after completing chemo-radiation. BT-RADS = Brain Tumor Reporting and Data System.

Descriptive analysis indicated that when comparing an unfavorable KPS score to a favorable one, BT-RADS_A was the only statistically different variable between the 2 score groups (see Table 1). There was a weak negative correlation between BT-RADS_A and BT-RADS_B (r = −0.11; P value = .30). The patient’s co-variables of sex, race, age at diagnosis, MGMT methylation status, and salvage chemotherapy use had no statistically significant association with KPS scores (see Table 1).

Table 1.

Descriptive Statistics

Characteristic	Favorable KPS score (at least 70%)	Unfavorable KPS score (0-60%)	P value
Sex			.81
Male (n = 45), %	49.0	46.5
Female (n = 49), %	51.0	53.5
Race			.18
White (n = 68), %	66.7	79.1
Other (n = 26), %	33.3	20.9
Tumor MGMT methylation status			.22
Methylated (n = 48), %	56.9	44.2
Unmethylated (n = 46), %	43.1	55.8
Salvage chemotherapy			.36
Yes (n = 12), %	15.7	9.3
No (n = 82), %	84.3	90.7
BT-RADS_A score			.03*
Low [0-3a] (n = 63), %	76.5	55.8
High [3b-4] (n = 31), %	23.5	44.2
BT-RADS_B score			.64
Low [0-3a] (n = 44), %	49.0	44.2
High [3b-4] (n = 50), %	51.0	55.8
Age at diagnosis (n = 94), years	58 (51, 63)	61 (54, 66)	.18

50% median and interquartile ranges (in parenthesis) shown for age at diagnosis. BT-RADS = Brain Tumor Reporting and Data System, BT-RADS_A = Scores assessed at 3 months post-chemoradiation, BT-RADS_B = Scores assessed at 9 months post-chemoradiation, MGMT= 0 -methylguanine-DNA methyltransferase.

^*Denotes statistical significance at α = 0.05.

In a univariate logistic regression analysis, BT-RADS_A was the only significant predictor (P = .04) in discriminating an unfavorable KPS score from a favorable one (see Table 2). In a multivariate logistic regression, the best model fit was the model containing BT-RADS_A and tumor MGMT methylation status (model P value = .03) (see Table 2). Controlling for tumor MGMT methylation status, the odds of having an unfavorable KPS score with a high BT-RADS score (scores 3b-4) are 2.82 times the odds of having an unfavorable KPS score with a low BT-RADS score (scores 0-3a). The receiver operating characteristic (ROC) curve analysis indicates that this best model fit is better than chance in differentiating an unfavorable KPS score from a favorable one (model AUC = 0.65; 95% CI = 0.55, 0.76).

Table 2.

Multivariate Regression Models with Unfavorable KPS as Reference Outcome

Model with variables	OR (95% CI)	P value
Model 1		.04*
BT-RADS_A	2.57 (1.06, 6.23)	.04*
Model 2		.64
BT-RADS_B	1.25 (0.54, 2.74)	.64
Model 3		.07
BT-RADS_A	2.84 (1.14, 7.11)	.03*
BT-RADS_B	1.52 (0.64, 3.60)	.35
Model 4		.03*
BT-RADS_A	2.82 (1.14, 6.99)	.03*
MGMT status (unmethylated vs. methylated)	1.90 (0.81, 4.46)	.14
Model 5		.46
BT-RADS_B	1.08 (0.47, 2.52)	.85
MGMT status (unmethylated vs. methylated)	1.63 (0.71, 3.79)	.25
Model 6		.07
BT-RADS_A	3.00 (1.18, 7.64)	.02*
BT-RADS_B	1.34 (0.55, 3.27)	.52
MGMT status (unmethylated vs methylated)	1.79 (0.75, 4.28)	.20

Unfavorable KPS score (0%-60%). BT-RADS = Brain Tumor Reporting and Data System, BT-RADS_A = Scores assessed at 3 months post-chemoradiation, BT-RADS_B = Scores assessed at 9 months post-chemoradiation, MGMT = 0 -methylguanine-DNA methyltransferase.

^*Denotes statistical significance at α = 0.05.

In a Kaplan–Meier survival analysis, all the patients experienced disease progression or recurrence. Overall, the median PFS was 240.5 days (95% CI = 210, 259 days). In the BT-RADS_A group, the log-rank test indicated that the difference between the survival curves for low BT-RADS score (scores 0-3a) and high BT-RADS score (scores 3b-4) was not statistically significant (P value = 0.07) (Figure 2). In the BT-RADS_B group, the log-rank test indicated that the difference between the survival curves for low BT-RADS score (scores 0-3a) and high BT-RADS score (scores 3b-4) was statistically significant (P value = 0.04) (Figure 3).

Figure 2.

Survival function curve showing the number of patients at risk of tumor progression on the horizontal axis, and the 95% Hall-Wellner bands. BT-RADS = Brain Tumor Reporting and Data System. Blue curve [0] = low BT-RADS score (scores 0-3a) and red curve [1] = high BT-RADS score (scores 3b-4).

Figure 3.

Survival function curve showing the number of patients at risk of tumor progression on the horizontal axis, and the 95% Hall-Wellner bands. BT-RADS = Brain Tumor Reporting and Data System. Blue curve [0] = low BT-RADS score (scores 0-3a) and red curve [1] = high BT-RADS score (scores 3b-4).

In a Cox regression model analysis, without controlling for tumor MGMT methylation status in the BT-RADS_A group, a low BT-RADS score was associated with a 33% decrease in risk of disease progression compared to a high BT-RADS score, but this decrease was not statistically significant (hazard ratio, 0.67; P value = 0.07) (see Table 3). Controlling for tumor MGMT methylation status in the BT-RADS_A group, a low BT-RADS score was associated with a 44% decrease in risk of disease progression compared to a high BT-RADS score, and the difference was statistically significant (hazard ratio, 0.56; P value = 0.01) (see Table 3). For the BT-RADS_B group and without controlling for tumor MGMT methylation status, a low BT-RADS score was associated with a 35% decrease in risk of disease progression compared to a high BT-RADS score, but this decrease was not statistically significant (hazard ratio, 0.65; P value = .05) (see Table 3). Controlling for tumor MGMT methylation status in the BT-RADS_B group, a low BT-RADS score was associated with a 27% decrease in risk of disease progression compared to a high BT-RADS score, but the difference was not statistically significant (hazard ratio, 0.73; P value = .14) (see Table 3). In a final model accounting for BT-RADS_B and tumor MGMT methylation status, a low BT-RADS score in the BT-RADS_A group was associated with a 50% decrease in risk of tumor progression compared to a high BT-RADS score in the same BT-RADS_A group (hazard ratio, 0.50; P value = .003) (Table 3).

Table 3.

Cox Proportional Hazards Regression Model with Low BT-RADS Score (0-3a) as Favorable Outcome: Estimating Hazard Function

Model with variables	HR	P value
Model 1		.07
BT-RADS_A	0.67	.07
Model 2		.05
BT-RADS_B	0.65	.05
Model 3		.01
BT-RADS_A	0.56	.01*
BT-RADS_B	0.56	<.001*
Model 4		<.001*
BT-RADS_A	0.56	.01*
MGMT status (methylated vs.  unmethylated)	0.38	<.001*
Model 5		<.001*
BT-RADS_B	0.73	.14
MGMT status (methylated vs. unmethylated)	0.44	<.001*
Model 6		<.001*
BT-RADS_A	0.62	.03*
BT-RADS_B	0.49	<.001*
MGMT status (methylated vs. unmethylated)	0.40	<.001*

BT-RADS = Brain Tumor Reporting and Data System, BT-RADS_A = Scores assessed at 3 months post-chemoradiation, BT-RADS_B = Scores assessed at 9 months post-chemoradiation, MGMT= 0 -methylguanine-DNA methyltransferase.

^*Denotes statistical significance at α = 0.05.

Discussion

GBM remains a challenging and aggressive brain tumor, necessitating a comprehensive understanding of prognostic factors and how these can be applied to improve the clinical management of patients. RANO criteria have limited use in clinical practice given inter-rater variability and the subjective nature of some of their criteria, among other limitations.¹⁶ The BT-RADS criteria represent a structured method for interpreting brain MRI findings and have shown merit in predicting trends in GBM patients’ imaging findings as well as OS.^11,19 BT-RADS has also shown its value in improving report consistency amongst radiologists and overall physician communication, a particularly important feature considering the highly multidisciplinary quality of GBM management.¹⁷ In this study, we sought to determine the prognostic value of BT-RADS scores in predicting patients’ functional outcomes following one year of treatment as well as PFS, and to investigate new potential applications of this system in guiding clinical management of GBM patients.

Our results showed no statistically significant association between patient covariates, including sex/race and age of diagnosis, and functional outcomes. Prior population studies have established that older patients do worse compared to their younger counterparts, not only due to biological factors, such as the increased burden of comorbidities and worsened toxicity effects, but also due to less aggressive treatment approaches when managing elderly patients KPS original paper.²⁹ These patterns were not replicated in this study, as the age of diagnosis had no statistically significant association with KPS scores. This may be because there was an underrepresented proportion of younger patients in this sample population, with only 19 patients below age 50, and therefore any trends comparing patient age would not be well captured. The lack of prognostic value of race/ethnicity in this study does, however, align with existing literature, with several studies showing that while the incidence of GBM varies greatly by race/ethnicity, the impact of these factors on patient outcomes has been far more inconsistent.^30,31 Salvage chemotherapy use was also not found to be associated with KPS scores. While we might expect drug toxicity to more prominently affect functional outcomes at one year, the proportion of patients who received salvage therapies, at only 13%, may have been too limited to produce substantial shifts in outcome variables. Additionally, almost all patients who underwent treatment with salvage chemotherapies received lomustine, with its major side effect being drug-induced thrombocytopenia.³² While this certainly limits its use in treating recurrent GBM long term,³² hematological toxicities more likely have less of an impact on patient day-to-day functioning when compared to other toxicity profiles, such as gastroenterological toxicities linked to agents like irinotecan³³ and cisplatin/etoposide,³⁴ these being used in only a small minority of patients in this study. Additionally, only one patient remained on lomustine for a prolonged period, completing 4 doses, whereas all other patients were found to have evidence of disease progression on MRI and were subsequently taken off of lomustine, typically after the first or second doses. This trend was similarly seen in patients undergoing treatment with other salvage chemotherapies; therefore, the protracted treatment course might not have had as large an impact on patients’ daily living.

In the best fit model containing BT-RADS_A and tumor MGMT methylation status, BT-RADS_A score, as a predictor assessed at 3 months post-chemoradiation, was significantly better than chance in discriminating an unfavorable KPS (0-60%) from a favorable one (70%-100%). Furthermore, a higher BTRADS_A score (3b-4), indicating worsening imaging findings, was significantly associated with an unfavorable KPS. This finding highlights the potential utility of the early BT-RADS score post-chemoradiation as an early prognostic indicator of patient performance status, stemming from the demonstrated impact that early treatment response has on functional decline.³⁵ BT-RADS_B scores, on the other hand, assessed at 9 months post-chemoradiation, were not significant predictors of functional outcomes at 1 year. Furthermore, the average PFS of 284 days suggests a majority of disease progression had already occurred after 9 months, ie, after this subset of images is acquired, lending a possible explanation as to the low predictive value of the BT-RADS_B score with functional outcomes. In addition, the limited percentage of patients (13%) receiving salvage therapy may have underpowered the statistical analysis, and the prediction of functional outcomes by BT-RADS_B.

As aforementioned, an average PFS of 284 days was observed in our study. The survival curves for both BT-RADS groups show a steeper decline in the slope prior to 400 days, indicating increased rates of disease progression during this period, after which PFS stabilizes. With the literature describing a median survival time of 15 months (465 days) for GBM IDH-wildtype patients, we expect most disease progression to occur before this point, which is supported by this data.³⁶ Our study also indicates a statistically significant difference in survival curves between low and high BT-RADS_B, assessed 9 months post-chemoradiation but not for BT-RADS_A assessed at 3 months. This is likely related to the biological and radiographic complexity and heterogeneity of early post-treatment periods, during which treatment-related changes secondary to radiotherapy often occur. In fact, prior studies have similarly demonstrated the lower specificity of early imaging to predict disease progression.^13,37 On the other hand, the association between BT-RADS_B and PFS must be interpreted cautiously, given that a significant portion of patients demonstrated disease progression and initiated salvage therapy by the 9-month mark. In fact, recurrence rate remains high in GBM despite standard therapy.³⁸ As such, at 9 months post-chemoradiation, we expect different patients to have already developed progression, and this is reflected in the higher BT-RADS score (3b-4) at this time point. Furthermore, stratifying by salvage therapy may have been underpowered, given that only 13% of patients received salvage therapy. These findings highlight the need for future studies that stratify by interval progression status and the need for a higher number of patients managed with salvage therapy.

Overall, our study suggests that BT-RADS scores from both early- and late-stage imaging represent valuable clinical tools for neuro-prognostication. Early post-treatment BT-RADS showed a more significant association with functional status, suggesting that patients with worse disease burden following initial chemoradiation may experience a more challenging treatment course and poorer overall performance outcomes. On the other hand, late-stage BT-RADS was more predictive of PFS, which may be related to potential additive complications of deleterious treatment effects and co-morbidities. With various models used in this study, the inclusion of MGMT methylation status improved predictive accuracy, indicating MGMT methylation status may modulate the association seen between BT-RADS scores and various prognostic factors. This highlights the importance of integrating molecular biomarkers into predictive models for more precise prognostication, as existing literature suggests.

Several limitations should be acknowledged. The study had a relatively small sample size, which may have underpowered the analysis. The retrospective nature of the study, as well as the exclusion of patients without assigned KPS or BT-RADS scores, further limited the sample pool. In addition to absolute KPS at 1 year, the evaluation of a delta KPS, ie, a KPS at one year standardized on a baseline KPS in the post-op period, would require a larger sample size to address the inter-rater reliability of KPS at baseline and 1 year follow-up, and can be the subject of future work. Selection bias may have also impacted our results. In this study, we excluded patients who did not complete the standard protocol of 6 weeks of chemoradiation. These patients have advanced age or poor functional status at diagnosis or following surgical resection, have an overall worse prognosis, and undergo a different treatment regimen with hypofractionated radiotherapy with or without concurrent TMZ.^29,39,40 This bias limits the generalizability of our results. Our study is based on a single institution, further restricting the application of these findings to other healthcare settings. OS analysis was not included in this study due to the longer follow-up period required compared to PFS, and because OS association with BT-RADS has already been investigated in earlier publications.^11,41 Lastly, the extent of surgical resection on the first resection was not considered among the covariates and may represent a subject for future work. Despite these limitations, our study findings contribute to the growing body of evidence supporting the utility of BT-RADS scores in predictive models, with the hope of better informing the clinical management of GBM patients.

The BT-RADS criteria represent a valuable tool in standardizing MRI interpretation and assisting in the neuro-prognostication of GBM patients. Our work provides an exploratory analysis of the use of BT-RADS in predicting patient functional outcomes and solidifies its ability to predict disease progression and survival. Thanks to these properties, BT-RADS has the potential to help guide the longitudinal and multidisciplinary care of GBM patients. As the next step, we envision a multi-center study incorporating clinical and molecular variables into predictive models of survival and quality of life using the BT-RADS criteria. This study could also further refine the analysis by stratifying the BT-RADS criteria into its subcategories, namely by comparing the predictive value of scores 3a, 3b, and 3c specifically. This will enable further exploration of these models and their applications in guiding more personalized management of GBM patients, and will increase the statistical power and external validity of the currently available data.

Funding

The project was supported by the Radiological Society of North America (RSNA) Research and Education Foundation, through grant number RSCH0235 (PI: Weinberg).

Conflict of Interest Statement

The authors declare that they have no conflict of interest relevant to this work.

Author Contributions

Conceptualization, M.P., M.E., T.A., and B.D.W.; methodology, M.P., M.E., T.A., and B.D.W.; data curation, M.P., M.E., T.A., and B.D.W.; writing—original draft preparation, M.P. and M.E.; writing—review and editing, M.P., M.E., T.A., L.D., J.Z., and B.D.W.; supervision, B.D.W. All authors have read and agreed to the published version of the manuscript.

Ethics Approval

The authors have received approval from the Emory University Institutional Review Board, granting permission for the passive use of the patient’s medical records for research purposes.

Data Availability

The data supporting the findings of this study are available from the corresponding author upon reasonable request. The dataset includes de-identified clinical and imaging information from patients treated at Emory University Hospital between January 2019 and January 2023.