Kaiku Health has taken its first steps to tackle the big issue related to the challenge of identifying which patients are likely to benefit from immune checkpoint inhibitor (ICI) therapies. The unfortunate fact is that not all eligible patients respond to these therapies at all, and for some of those who have a response, it may in fact appear unpredictably. When making treatment decisions, it would be helpful to know who might actually benefit from the administered therapy. Considering the special immune-related adverse events these treatments may cause, it could improve the quality of life and safety of patients if it were possible to predict the treatment response.
In collaboration with Oulu University Hospital (OYS), Kaiku Health conducted a study which investigated whether machine learning (ML) could be used to predict the treatment response, or more specifically objective response rate (ORR) of ICI treated patients, based on clinical and patient-reported data. ORR was defined as the proportion of patients in whom partial (PR) or complete (CR) responses were seen as the best overall response.
Machine learning based approaches are not new to Kaiku. Previously Kaiku has shown together with OYS, that it is possible to predict both symptom and adverse event onset and continuation in patients receiving ICI therapies.
Data in the study were collected from 31 patients undergoing ICI treatments. The data sources consisted of clinician-assessed treatment responses according to RECIST criteria, immune-related adverse events (irAEs) according to CTCAE, and laboratory measurements. Patients also reported their symptoms (patient-reported outcomes) during the treatment using Kaiku Health module tailored for ICI therapies
The predictive model for ORR was built using a gradient boosting based algorithm, which is commonly used for classification problems. The performance of the predictive model was evaluated using the leave-one-out-cross-validation method. Despite the limited size of the patient cohort, the model performed well in predicting an objective response to treatment.
Results of the study. XGBoost LOOCV performance metrics for predicting ORR.
These promising results indicate that ML-based approaches in treatment response prediction should be investigated and validated further with a larger dataset.
The results of this study were presented at the ESMO IO 2020 Virtual Congress last week and received remarkable attention from the European Society for Medical Oncology. The findings were referred to in both ESMO Oncology news in the ESMO website and in the ESMO IO Congress news which highlighted a few interesting findings from the wide research presented at the Congress.
Want to learn more? Contact us here!
