A Phase II Study of Stereotactic Body Radiotherapy (SBRT) for Prostate Cancer Using Simultaneous Integrated Boost and Normal Structure-Sparing IMRT Planning

Author

Lu Mao

Published

August 13, 2025

Statistical analysis

Dosimetric variables are extracted from the dose-volume histogram (DVH) data for each subject. The association of these variables with acute and late toxicity and quality of life (QoL) is assessed. The dosimetric variables are selected based on their predictive performance for acute toxicity using the receiver operating characteristic (ROC) curve analysis. The association of the selected variables with acute toxicity, late toxicity, and QoL is evaluated using boxplots and Wilcoxon rank-sum tests. Clinical variables (prostate size, symptom score, AUASS) are also considered in the analysis. Logistic regression models are used to assess the predictive performance of the dosimetric variables with and without clinical variables. A 5-fold cross-validation (CV) is employed to estimate the area under the ROC curve (AUC) for the logistic regression models. P values < 0.05 are considered statistically significant. All analyses are performed using R version 4.1.1.

Results

Dose-volume histogram

The dose-volume histogram for each subject is plotted in Figure 1.

Figure 1: Dose-volume histogram for each subject

Selection of dosimetric variables

To select the dosimetric variables, we assess their associations with acute toxicity. We perform ROC analysis for each variable against the binary variable acute_max (max grade \(\ge 2\)). The area under the curve (AUC) is used as a measure of predictive performance. The ROC curves are shown below.

Figure 2: ROC curves for dosimetric variables against acute toxicity

So Dmax, D5%, D10% are the most predictive dosimetric variables for acute toxicity.

Association with acute toxicity

The boxplot in Figure 3 shows the distribution of Dmax, D5%, and D10% by acute toxicity grade. Table 1 shows the summary statistics of these variables by acute toxicity grade.

Figure 3: Boxplot of Dmax, D5%, D10% by acute toxicity grade and overall
  • Note: The low outlier is subject 54; see Figure 1.
Table 1: Dmax, D5%, D10% by acute toxicity grade
Characteristic <2, N = 651 2+, N = 501 p-value2
Dmax 38.80 (38.20, 40.10) 39.80 (39.00, 40.70) 0.004
D5 37.80 (37.40, 38.50) 38.30 (37.63, 38.80) 0.029
D10 37.50 (37.20, 38.20) 38.05 (37.40, 38.48) 0.028
1 Median (IQR)
2 Wilcoxon rank sum test

Utility of clinical variables

Adding clinical variables (prostate size, symptom score, AUASS) to Dmax (AUC 0.667) does not meaningfully improve the predictive performance of Dmax alone (0.655) for acute toxicity.

Association with late toxicity

We also assess the association of Dmax, D5%, and D10% with late toxicity. The boxplot in Figure 4 shows the distribution of these variables by late toxicity grade. Table 2 shows the summary statistics of these variables by late toxicity grade.

Figure 4: Boxplot of Dmax, D5%, D10% by late toxicity grade and overall

  • There are subjects without late toxicity data.
Table 2: Dmax, D5%, D10% by late toxicity grade
Characteristic <2, N = 891 2+, N = 241 p-value2
Dmax 39.20 (38.50, 40.20) 40.05 (38.83, 40.93) 0.15
D5 38.00 (37.50, 38.60) 38.30 (37.48, 39.15) 0.5
D10 37.90 (37.30, 38.30) 38.00 (37.18, 38.83) 0.7
Dmin 2 (1, 13) 1 (0, 3) 0.020
D90 4 (1, 32) 1 (1, 5) 0.023
D80 9 (2, 37) 1 (1, 10) 0.011
V5 84 (66, 100) 64 (46, 88) 0.007
1 Median (IQR)
2 Wilcoxon rank sum test
  • Dmin, D90, D80, and V5 are also included in the table as they are most predictive of late toxicity.

Other variables vs late toxicity (Figure 5).

Figure 5: ROC curves for dosimetric variables against late toxicity

Utility of clinical variables

Adding clinical variables (prostate size, symptom score, AUASS) to D80 (AUC 0.612) does not improve the predictive performance of D80 alone (0.669) for late toxicity.

Association with quality of life

We also assess the association of Dmax, D5%, and D10% with quality of life (QoL). We focus on the change in Urinay Irritative/Obstructive Domain Score (UI) from baseline to 24 months. The boxplot in Figure 6 shows the distribution of these variables by QoL change. Table 3 shows the summary statistics of these variables by QoL change.

Figure 6: Boxplot of Dmax, D5%, D10% by UI score change status (Baseline to 24 months) and overall

  • A number of subjects do not have baseline score.
Table 3: Dmax, D5%, D10% by QoL change
Characteristic Negative, N = 581 Positive, N = 381 p-value2
Dmax 39.20 (38.50, 40.13) 39.90 (38.83, 40.65) 0.064
D5 37.90 (37.60, 38.50) 38.25 (37.55, 39.08) 0.14
D10 37.65 (37.40, 38.20) 38.00 (37.40, 38.88) 0.2
Dmin 1 (1, 5) 2 (1, 7) 0.4
D90 2 (1, 11) 3 (1, 30) 0.3
D80 4 (1, 29) 14 (2, 36) 0.3
V5 77 (60, 100) 86 (67, 100) 0.4
1 Median (IQR)
2 Wilcoxon rank sum test

Utility of clinical variables

Adding clinical variables (prostate size, symptom score, AUASS) to Dmax (AUC 0.649) only mildly improves the predictive performance of Dmax alone for QoL change (AUC 0.613).