Gerharz, Carl (2024) Can Machine Learning Help us in Prediction of Survival for Patients with Extremity Metastases? - A Retrospective External Validation Study of a Bayesian Belief Network(PathFx v. 3.0) for Survival Prediction in a cohort of 292 Dutch Patients treated Surgically and Conservatively. thesis, Medicine.
Full text available on request.Abstract
Metastatic bone cancer significantly impacts patients' quality of life, underscoring the necessity for personalized treatment strategies. Surgical interventions rely upon anticipated patient survival, a parameter that remains challenging to estimate accurately for physicians. Machine learning algorithms, like PathFx version 3.0, aim to address this challenge. This study focuses on the external validation of PathFx 3.0 within a Dutch cohort with both surgically treated(n=190) and conservatively treated patients(n=102). We retrospectively included 292 patients treated for long-bone metastasis in the Netherlands between 2010 and 2022 with primary outcomes being 90-day and 1-year survival. To assess overall model performance, we analysed discrimination, calibration, overall performance(Brier score), and performed decision curve analysis. Unfortunately, PathFx 3.0 demonstrated suboptimal performance, exhibiting AUC values below 0.7 for both 90-day and 1-year survival predictions across all patient groups. Calibration analysis showed a systematic underestimation of survival. The model's limitations, such as missing modern treatments as predictors and potential population-specific disparities, likely contributed to its diminished performance. In conclusion, PathFx 3.0's predictive accuracy for 90-day and 1-year survival in long-bone metastasis patients, whether treated surgically or non-surgically, falls below the threshold for clinical utility. This study underscores the significance of tailored, internally developed algorithms to enhance predictive precision across specific patient populations.
| Item Type: | Thesis (UNSPECIFIED) |
|---|---|
| Supervisor name: | Prof. Dr. Doornberg, Job and de Groot, Tom |
| Date Deposited: | 23 Sep 2024 11:54 |
| Last Modified: | 23 Sep 2024 11:54 |
| URI: | https://umcg.studenttheses.ub.rug.nl/id/eprint/3760 |
Actions (login required)
 |
View Item |