Methods

Patient characteristics and data collection

This retrospective study was approved by the Institutional Review Board of the Erasmus MC, University Medical Center Rotterdam, Rotterdam, The Netherlands (reference number: MEC-2022-0760; date of approval: 03 January 2023). Between January 2012 and December 2022, 302 adult patients with pathologically confirmed newly diagnosed IDH-wildtype glioblastoma (in accordance with the 2021 WHO Classification of Tumors of the Central Nervous System [12]), were scheduled to receive radiotherapy with a total dose of 60 or 40.05 Gray (Gy) at the Department of Radiotherapy of the Erasmus MC Cancer Institute (Rotterdam, The Netherlands). Patients were excluded from this analysis if they did not complete radiation treatment (n = 11), they did not receive concurrent chemotherapy (n = 6), their radiotherapy treatment plan was unavailable (n = 3), progressive disease (PD) was not confirmed (n = 57), follow-up MRI was unavailable (n = 16), or pathological analysis after re-resection revealed PD was pseudoprogression rather than true tumor recurrence (n = 8). For the remaining 201 patients, we collected clinical and imaging data, the radiotherapy target delineations, and the delivered radiotherapy treatment plans. The following parameters were collected for multiple linear regression modeling: age (at the time of surgery), gender, tumor laterality (hemisphere), the presence of contrast-enhancement prior to surgery, extent of resection, O6-methylguanine methyltransferase (MGMT) methylation status, GTV volume, the 5th percentile of the pre-radiotherapy apparent diffusion coefficient (ADC) signal intensity distribution within the GTV (ADC_P5), and involvement of the subventricular zone (SVZ) and subgranular zone (SGZ). The GTV typically encompassed the resection cavity plus any residual contrast-enhancing tumor on post-contrast T1-weighted MRI. Deviations from this definition were possible at the discretion of the treating radiation oncologist. The choice of the 5th percentile of the ADC distribution (ADC_P5) was motivated by its robustness to outlier voxels with erroneously low signal intensities. The ADC_P5 is an indicator of diffusion restriction, with lower values associated with higher Ki-67 labeling indices [13]. The SVZ was considered involved when the contrast-enhancing tumor touched the lining of the lateral ventricles on preoperative imaging [14], [15], [16], [17]. For the SGZ, the pre-radiotherapy T1-weighted image was first skull-stripped (HD-BET [18]) and then registered to the MNI152 brain template (flirt [19], [20], FSL v6.0.7, Oxford, UK). The resulting transformation matrix was used to register the GTV to MNI152 space. The SGZ was considered involved when the GTV and the SGZ mask from the Hippocampus and Subfields CoBrA atlas [21] overlapped.

Recurrence analysis

The timepoint of recurrence was defined as the moment when the multidisciplinary team or treating physician determined PD. Sequential MRI-scans obtained before and after this timepoint were reviewed in consultation with a neuroradiologist to confirm radiological evidence of progression and reduce the risk of misclassification due to pseudo-progression. The MRI-scan showing first radiological evidence of progression was rigidly registered to the planning CT in MIM Maestro® (v7.1.6), and alignment was visually verified for all cases. Subsequently, the contrast-enhancing recurrence volume was manually delineated in MIM Maestro® by a single observer. The recurrence distance was defined as the minimum isotropic expansion from the GTV required to encompass 80% of the recurrence volume (RD80). Thus, an RD80 of 15 mm indicates that a 15 mm isotropic margin around the GTV would encompass 80% of the recurrence volume. The choice for an 80% coverage of the recurrence volume was guided by various studies that performed pattern of failure analysis based on the high-dose radiation treatment field. In these studies, recurrences were classified as in-field if ≥80% of the recurrence volume was located within the 95% isodose line [7], [22], [23], [24]. The RD80 was calculated in two steps: First, the minimum Euclidean distance to the outer edge of the GTV was calculated for each voxel within the recurrence volume. Thereafter, the 80th percentile of these distances was extracted to quantify the RD80. A schematic representation of the RD80 is given in Fig. 1_._ In addition to the RD80, we performed pattern of failure analysis based on the high-dose radiation treatment field. Recurrences were classified as in-field, marginal, or distant if >80%, 20–80, <20% of the recurrence volume was covered by the 95% isodose line of the clinical radiotherapy plan, respectively.

Fig. 1: A schematic representation of the recurrence distance calculation. The RD80 represents the minimum isotropic expansion from the GTV required to encompass 80% of the recurrence volume. If a margin of RD80 was added to the GTV, 80% of the recurrence volume would be located within this margin. Note that we show a simplified 2-dimensional representation in this figure.

Regression analysis

Because visual assessment of the GTV volume and the RD80 showed right-skewed distributions, both variables were log-transformed using the natural logarithm. For the RD80, a small constant of 0.5 mm was added to all distances prior to log-transformation to account for zero values. We performed multiple linear regression analyses to identify relevant variables that were associated with the log-transformed RD80. First, all variables presented in Table 1, excluding the CTV-margin and number of adjuvant chemotherapy cycles until PD, were examined using univariable linear regression. Variables with a p-value <0.20 were retained for the initial multiple linear regression model. As involvement of neurogenic zones has been found to correlate to distant recurrences and dissemination [14], [15], [16], [17], [25], [26], [27], [28], the variables involvement of the SVZ and the SGZ were included in the initial multivariable model regardless of their statistical significance in the univariable models. We applied a backward elimination procedure, where the variable with the highest p-value was sequentially removed at each step. Backward elimination continued until solely variables with a statistically significant association (two-sided p < 0.05) remained. To assess the impact of using an alternative recurrence distance metric, the minimum isotropic expansion from the GTV required to encompass 95% of the recurrence volume (RD95) was additionally computed, and the regression analysis with backward elimination was repeated using the RD95. Regression analysis was performed using R (v4.3.2).

Characteristics | No. of patients (%)
Age (years)
Mean (range) | 58 (31–81)

Gender
Male | 134 (66.7%)
Female | 67 (33.3%)

Hemisphere
Left | 92 (45.8%)
Right | 100 (49.8%)
Bilateral | 9 (4.5%)

Contrast-enhancement prior to surgery
Present | 180 (89.6%)
Absent | 21 (10.4%)

Extent of resection
Biopsy | 33 (16.4%)
Partial resection | 141 (70.1%)
Gross total resection | 27 (13.4%)

MGMT methylation status
Unmethylated | 74 (36.8%)
Methylated | 58 (28.9%)
Heterogeneous | 11 (5.5%)
Unknown | 58 (28.9%)

GTV volume (cm3)
Median (IQR) | 42.9 (22.6–78.1)

ADC_P5 (10−6 mm2/s)
Median (IQR)⁎ | 648 (371–734)
 | 
Involvement of the SVZ
SVZ involved | 80 (39.8%)
SVZ not involved | 121 (60.2%)

Involvement of the SGZ
SGZ involved | 89 (44.3%)
SGZ not involved | 112 (55.7%)

CTV-margin
15 mm | 133 (66.2%)
20 mm | 68 (33.8%)

Radiotherapy fractionation scheme
30 × 2 Gy | 141 (70.1%)
15 × 2.67 Gy | 60 (29.9%)

Adjuvant chemotherapy cycles
Median (IQR) | 6 (3–6)

Sensitivity analysis

From May 2018 onwards, the standard CTV-margin that was used at the Department of Radiotherapy of the Erasmus MC Cancer Institute was reduced from 20 mm to 15 mm. Hence, a number of patients in this cohort received radiation treatment with a radiotherapy plan that employed a 20 mm CTV-margin instead of 15 mm. Additionally, variations in radiotherapy fractionation scheme and number of completed adjuvant chemotherapy cycles existed in our cohort. These variations may have had an impact on the location of tumor recurrence. Therefore, sensitivity analyses were performed by adding the CTV-margin, radiotherapy fractionation scheme, and number of completed adjuvant chemotherapy cycles prior to PD as covariates to the final RD80 regression model. Additionally, the log-transformed time to progression was added as an additional covariate to the final RD80 regression model.