ESTRO 2024 - Abstract Book

S4996

Physics - Radiomics, functional and biological imaging and outcome prediction

ESTRO 2024

for inter-patient variation through the inclusion of random effect terms. Parameter values and 95% confidence intervals were calculated, as well as performance metrics in the form of the area under the receiver operating characteristic curve (AUC) and the F1-score to evaluate the rates of correctly labelled events and non-events, the models’ sensitivity and specificity, respectively.

Results:

The bivariate mixed-effect model accounting for random effects in the intercept and parameters (formula: Events ~ Dose+LETd+(1 │ Patient ID)+(Dose │ Patient ID)+(LETd|Patient ID)) showed a positive slope for the fixed effects of both dose and LETd (Table 1, Figure 2a-b). The standard deviations of the model’s random effects were 11.19 for dose and 4.22 for LETd. The standard bivariate logistic regression model (formula: Events ~ Dose+LETd) had lower parameter values and narrower confidence intervals than the mixed effect model (Table 1, Figure 2a-b). The univariate logistic regression models had a positive slope for the dose but a shallow negative slope for the LETd (Table 1, Figure 2b-c). Only the mixed-effect model reported a defined F1-score since the others had a sensitivity of zero. The mixed-effect model had a low F1-score with solid specificity but low sensitivity (Table 1).

Table 1: Dose and LETd coefficients with 95% confidence intervals and AUC for all four models, and F1-score for the mixed effect model.

Model

Dose Coefficient

LETd coefficient

AUC

F1-Score

Mixed Effect

8.24 [6.36, 10.12] 3.80 [3.68, 3.91] 2.89 [2.79, 2.98]

2.29 [1.41, 3.17] 0.74 [0.69, 0.78]

0.90 0.76 0.76 0.59

0.12

Bivariate Logit

Undefined Undefined Undefined

Dose Logit LETd Logit

-0.56 [-0.58, -0.53]