Joint Applied Mathematics and Statistics Seminar 4.5.

Speaker: Teemu Daniel Laajala

Title: ”Ensemble-based Penalized Cox Regression for mCRPC survival prediction for the DREAM 9.5 Challenge”

Abstract: ”ePCR is a novel ensemble-method developed for the DREAM 9.5 metastatic Castration-Resistant Prostate Cancer (mCRPC) patient survival prediction challenge, in which it significantly outperformed over 50 competing teams from across the world. The ensemble-components were penalized Cox regression models optimized by exploring a parameter space for the both the penalization/regularization parameter (lambda) as well as the L1/L2 norm parameter (alpha). The separately optimized individual ensemble models were constructed in order to block out potential stratification effects, and a consensus prediction was generating over the ensemble members.

The methodology was published along with validation in an independent cohort (1), and network analysis was performed to analyze its underlying Ridge Regression -like structure with pairwise interactions introduced for clinical variables. The resulting ensemble-model was shown to be robust and has since provoked great interest for clinical applications of sophisticated machine learning methods, as well as the use of such models in practical applications (2,3).

Furthermore, I will present results from a follow-up project conducted together with the National Cancer Institute (NCI/NIH, US). The resulting work provides clinicians with a practical graphical user-interface for using the methodology for research purposes, a larger variety of optimized models for different subgroups or research questions, as well as presents extended tools for e.g. time-to-event prediction.

1: Guinney J*, Wang T*, Laajala TD*, et al. Prediction of overall survival for patients with metastatic castration-resistant prostate cancer: development of a prognostic model through a crowdsourced challenge with open clinical trial data. Lancet Oncol. 2017 Jan;18(1):132-142. doi: 10.1016/S1470-2045(16)30560-5.
2: Davis ID. Challenges of data sharing: valuable but costly? Lancet Oncol. 2017 Jan;18(1):15-16. doi: 10.1016/S1470-2045(16)30564-2.
3: Seyednasrollah F, et al. How Reliable are Trial-based Prognostic Models in Real-world Patients with Metastatic Castration-resistant Prostate Cancer? Eur Urol. 2017 May;71(5):838-840. doi: 10.1016/j.eururo.2017.01.043.

*Equal contribution”

Time: 12:15
Place: Lecture Hall XVI (Quantum 1st floor)

All interested are very welcome!

For more information on the seminar and future schedule is here.

Joint Applied Mathematics and Statistics Seminar 4.5.

Speaker: Teemu Daniel Laajala

Title: ”Ensemble-based Penalized Cox Regression for mCRPC survival prediction for the DREAM 9.5 Challenge”

Abstract: ”ePCR is a novel ensemble-method developed for the DREAM 9.5 metastatic Castration-Resistant Prostate Cancer (mCRPC) patient survival prediction challenge, in which it significantly outperformed over 50 competing teams from across the world. The ensemble-components were penalized Cox regression models optimized by exploring a parameter space for the both the penalization/regularization parameter (lambda) as well as the L1/L2 norm parameter (alpha). The separately optimized individual ensemble models were constructed in order to block out potential stratification effects, and a consensus prediction was generating over the ensemble members.

The methodology was published along with validation in an independent cohort (1), and network analysis was performed to analyze its underlying Ridge Regression -like structure with pairwise interactions introduced for clinical variables. The resulting ensemble-model was shown to be robust and has since provoked great interest for clinical applications of sophisticated machine learning methods, as well as the use of such models in practical applications (2,3).

Furthermore, I will present results from a follow-up project conducted together with the National Cancer Institute (NCI/NIH, US). The resulting work provides clinicians with a practical graphical user-interface for using the methodology for research purposes, a larger variety of optimized models for different subgroups or research questions, as well as presents extended tools for e.g. time-to-event prediction.

1: Guinney J*, Wang T*, Laajala TD*, et al. Prediction of overall survival for patients with metastatic castration-resistant prostate cancer: development of a prognostic model through a crowdsourced challenge with open clinical trial data. Lancet Oncol. 2017 Jan;18(1):132-142. doi: 10.1016/S1470-2045(16)30560-5.
2: Davis ID. Challenges of data sharing: valuable but costly? Lancet Oncol. 2017 Jan;18(1):15-16. doi: 10.1016/S1470-2045(16)30564-2.
3: Seyednasrollah F, et al. How Reliable are Trial-based Prognostic Models in Real-world Patients with Metastatic Castration-resistant Prostate Cancer? Eur Urol. 2017 May;71(5):838-840. doi: 10.1016/j.eururo.2017.01.043.

*Equal contribution”

Time: 12:15
Place: Lecture Hall XVI (Quantum 1st floor)

All interested are very welcome!

For more information on the seminar and future schedule is here.