This Week's Discoveries | 23 June 2020
- Tuesday 23 June 2020
- This Week's Discoveries
- Kaltura Live Room (see link below)
So, you think you need AI?
Join the lecture in Kaltura Live Room
Marta Fiocco leads the Leiden DASPO (Data Analysis and Survival for Personalised Oncology) group since 2015. She works half-time for the Prinses Maxima Center for Child Oncology (PMC), and is faculty member at the European Organisation for Research and Treatment of Cancer (EORTC).
The DASPO group, which is based at MI but has extensive LUMC involvement, focuses its research on prognostic models for oncology data. These models form an important tool in personalised medicine, where a tailored treatment of cancer patients avoids undertreatment of high-risk patients and overtreatment in low-risk patients.
Over the last decade, interest and publications on machine learning approaches in cancer research have grown, giving rise to an ongoing debate on the value of machine learning (ML) approaches versus more traditional statistical modelling approaches. A statistical model exists within a mathematical framework and makes probabilistic assumptions about the data generation process. Generally, it allows for easy interpretation.
In contrast, ML approaches make no assumptions about the data generating process, and no a priori specified relationship is imposed on the predictors and outcome. They rely on algorithms and learn relationships from the data itself. While the absence of preconceived structures allows for complicated non-linear effects and high-order interactions between variables, the relationships cannot be accurately quantified, and at best, only a general indication of predictor importance can be obtained.
In this talk a comparison is made between ML methods and a statistical model commonly used for survival analysis- the Cox proportional hazards model.
Data on osteosarcoma patients from the EURAMOS-1 clinical trial are used to compare the performance of ML methods and a statistical model in the prediction of survival. Osteosarcoma is the second most common primary bone tumour, and is primarily diagnosed in adolescents and young adults.