Liesbeth Hondelink

LUMC

In short, Liesbeth Hondelinks' fellowship entails the following: 

Multi-input prognostic models for time to resistance prediction

'Non-small cell lung adenocarcinoma is one of the deadliest and most common cancers globally. In the past decades, treatment with the targeted tyrosine kinase inhibitor Osimertinib has substantially improved disease-free survival in selected patients whose cancers harbor a mutation in the epithelial growth factor receptor (EGFR). However, acquired resistance to Osimertinib inevitably occurs, resulting in disease progression. In the management of Osimertinib-treated disease, swift detection of acquired resistance is paramount, in order to keep treating patients effectively. However, there is substantial variance between patients with regard to time to resistance, which complicates treatment and monitoring. An accurate prediction of time to resistance would greatly benefit Osimertinib-treated patients.

Over the course of the Hanarth Fellowship, Liesbeth Hondelink will develop a convolutional deep learning model, which combines Pathology images, Radiology data, clinical parameters and molecular sequencing data into a comprehensive prediction tool, using the newest data science technology. The ultimate aim is to predict time to resistance in Osimertinib-treated lung cancer patients, which will eventually lead to better patient management.'

Short summary of progress / results March 2023

Lung cancer is one of the most dangerous and most common cancers in the world, impacting almost 2 million people annually. The treatment of lung cancer is – in the last decade – increasingly based on newly discovered ‘targeted’ therapies, especially for lung cancer in patients who have never smoked cigarettes. These targeted therapies are targeted against specific molecules, which are present in lung cancer cells only. Targeted therapy therefore has fewer side-effects than conventional treatments, such as chemotherapy. Unfortunately, response to these targeted drugs varies greatly between patients. Some patients benefit only for 2 or 3 months, while others maintain progression-free for several years. This wide variance makes choosing the best cancer treatment difficult and unpredictable.

Artificial intelligence is able to give additional insight into which patient will respond to targeted therapy. By combining multiple patient-factors into one predictive model, we have started to discover which factors lead to good and bad response. This is an important step in better understanding the effects and limitations of targeted therapy, which could potentially help in selecting the best treatment for each lung cancer patient.