Prediction model for the selection of patients with glioma to proton therapy
Background and purpose: The selection of patients with low-grade gliomas for proton therapy (PT) is often based on the comparison of photon and PT plans and demonstrating meaningful dose reduction to the healthy brain or critical structures. The aim of this retrospective study was to identify clinic...
Saved in:
| Main Authors: | , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Medical Journals Sweden
2025-07-01
|
| Series: | Acta Oncologica |
| Subjects: | |
| Online Access: | https://medicaljournalssweden.se/actaoncologica/article/view/43883 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Background and purpose: The selection of patients with low-grade gliomas for proton therapy (PT) is often based on the comparison of photon and PT plans and demonstrating meaningful dose reduction to the healthy brain or critical structures. The aim of this retrospective study was to identify clinical parameters associated with referral to PT and build a prediction model.
Patients and methods: The dataset consisted of patients with isocitrate dehydrogenase (IDH)-mutant grades 2 and 3 glioma and candidates for PT at the Aarhus University Hospital. Clinical (age, diagnosis, clinical target volume [CTV], and treatment) and dosimetric (prescribed dose and mean dose (Dmean) to healthy brain) parameters were collected. Univariate and multivariate logistic regression were used to assess the association with selection for PT. The dataset was split into training (n = 37, period 2019–2022) and test (n = 12, period 2023) cohorts. Prediction models were built using logistic regression algorithms and support vector machines (SVMs) and evaluated using the area under the precision-recall curve (AUC-PR).
Results: Age (p = 0.03) and CTV (p = 0.01) were significantly associated with the selection for PT and were used for model prediction. The logistic regression demonstrated AUC-PR at 0.999 (CI 0.999–1.000) and 0.998 (0.996–1.000) for training and test cohorts, respectively. SVM showed similar results with AUC-PR at 0.993 (0.993–0.994) for training and 0.999 (0.998–1.000) for test cohorts.
Interpretation: Logistic regression and SVM using age and CTV performed equally well and achieved a very high positive predictive value. With the pending external validation in a larger dataset, the prospects of this work suggest more consistent and efficient patient referral for PT.
|
|---|---|
| ISSN: | 1651-226X |