Results

Patient characteristics

In the 201 patients, the median time between the last radiotherapy fraction and PD was 6.8 months (IQR: 4.1–11.8 months). An overview of the patient and tumor characteristics is provided in Table 1_._

Recurrence analysis

The median recurrence volume was 14.6 cm3 (IQR: 1.7–45.0 cm3). The median RD80 was 1.2 mm (IQR: 0–6.0 mm). Fig. 2 shows the cumulative distribution of the RD80 in this cohort. The absolute distribution of the RD80 is provided in Supplementary Fig. S1. Based on the 95% isodose line of the clinical radiotherapy plan, recurrences were classified as in-field for 175 patients (87.1%), marginal for 6 patients (3.0%), and distant for 20 patients (10.0%).

Fig. 2: Cumulative distribution of the RD80. The vertical axis represents the proportion of patients whose recurrence is located within a specified RD80. For instance, the RD80 was ≤10.4 mm for 80% of the patients in our cohort (orange dot).

Regression analysis

From the univariable linear regression (Table 2), the variables gender (p = 0.08), contrast-enhancement prior to surgery (p = 0.18), extent of resection (p = 0.05), log-transformed GTV volume (p = 3.1e-5), involvement of the SVZ (p = 0.09), and radiotherapy fractionation scheme (p = 0.06) were identified as potential predictors for the log-transformed RD80. The initial multiple linear regression model included the variables gender, contrast-enhancement prior to surgery, extent of resection, log-transformed GTV volume, involvement of the SVZ, involvement of the SGZ, and radiotherapy fractionation scheme. Backward elimination first removed contrast-enhancement prior to surgery (p = 0.87) from the initial model, followed by radiotherapy fractionation scheme (p = 0.29), extent of resection (p = 0.21), and gender (p = 0.07) in the subsequent models, resulting in a final model in which the log-transformed GTV volume, involvement of the SVZ, and involvement of the SGZ remained as significant predictors. The (exponentiated) coefficients and p-values of the variables in the final multiple linear regression model (F(3, 197) = 10.3, adjusted R__2 = 0.12, overall p-value = 2.5e-6) are shown in Table 3_._ The initial and intermediate multiple linear regression models are provided in Supplementary Table S1-S4. The relationships between the variables retained in the final model and the RD80 are provided in Supplementary Fig. S2, and the relationships among the variables retained in the final model are provided in Supplementary Fig. S3. For the regression analysis with the RD95, the log-transformed GTV volume, involvement of the SVZ, and involvement of the SGZ remained as significant predictors after backward elimination. The (exponentiated) coefficients and p-values of the final RD95 regression model (F(3, 197) = 11.3, adjusted R__2 = 0.13, overall p-value = 6.76e-8) are provided in Supplementary Table S5.

Variable |  | β (coef.) | 95% CI | Overall p-value
Age |  | 0.01 | −0.02–0.03 | 0.59
Gender⁎ | Male (reference) | – | – | 0.078
Female | −0.42 | −0.89–0.05 | 
Hemisphere | Left (reference) | – | – | 0.74
Right | −0.07 | −0.53–0.39 | 
Bilateral | 0.36 | −0.75–1.47 | 
Contrast-enhancement⁎ | Absent (reference) | – | – | 0.18
Present | 0.50 | −0.23–1.22 | 
Extent of resection⁎ | Biopsy (reference) | – | – | 0.053
Partial resection | 0.73 | 0.13–1.34 | 
Gross total resection | 0.76 | −0.05–1.57 | 
MGMT methylation status | Unmethylated (reference) | – | – | 0.35
Methylated | 0.37 | −0.15–0.89 | 
Heterogeneous | 0.01 | −0.95–0.97 | 
GTV volume (log-transformed)⁎ |  | −0.52 | −0.76 to −0.28 | 3.1e-5
ADC_P5 |  | −0.00 | −0.00–0.00 | 0.41
Involvement of the SVZ⁎ | Not involved (reference) | – | – | 0.087
Involved | 0.40 | −0.06–0.85 | 
Involvement of the SGZ | Not involved (reference) | – | – | 0.72
Involved | 0.08 | −0.37–0.53 | 
Radiotherapy fractionation scheme⁎ | 15 × 2.67 Gy (reference) | – | – | 0.055
30 × 2 Gy | 0.47 | −0.01–0.96 | 

Variable |  | β (coef.) | exp(β) | 95% CI | Overall p-value
GTV volume (log-transformed)⁎ |  | −0.70 |  | −0.96 to −0.44 | 3.9e-7
Involvement of the SVZ⁎ | Not involved (reference) | – | – | – | 0.019
 | Involved | 0.53 | 1.70 | 0.09–0.96 | 
Involvement of the SGZ⁎ | Not involved (reference) | – | – | – | 0.032
 | Involved | 0.51 | 1.67 | 0.04–0.98 | 

Sensitivity analysis

In the sensitivity analyses, the final model was not significantly affected by differences in CTV-margin (β = 0.32, p = 0.16) or radiotherapy fractionation scheme (β = 0.32, p = 0.18). In contrast, the number of completed adjuvant chemotherapy cycles (β = 0.13, p = 0.015) and log-transformed time to progression (β = 0.42, p = 0.00075) were independently associated with the RD80 when included as additional covariates. When adding both variables simultaneously as covariates to the final RD80 regression model, time to progression (β = 0.42, p = 0.020) remained significantly associated, whereas number of adjuvant chemotherapy cycles was no longer significant (β = 0.00, p = 0.99). In all sensitivity analyses, the independent variables in the model, i.e. log-transformed GTV volume, involvement of the SVZ and involvement of the SGZ, remained significant predictors. The models for the sensitivity analyses are provided in Supplementary Table S6-S10. The relationships between the number of adjuvant chemotherapy cycles, time to progression and RD80 are provided in Supplementary Fig. S4.