Abstract 1841P
Background
ICI therapies have introduced novel irAEs, which can arise from various organ systems, and at any time during or after the discontinuation of the therapy. Early detection of irAEs could result in an improved safety profile of the treatment and better quality of life for patients. ML models utilizing ePROs and other clinical data could enable earlier detection of irAEs.
Methods
The utilized dataset consisted of three data sources. The first dataset of 18 symptoms monitored with Kaiku Health digital platform consisted of 16 540 reported symptoms from 33 ICI treated cancer patients. The second dataset included laboratory measurement data (8 different values) from the same 33 patients during ICI therapy. The third dataset included prospectively collected irAE data, including the onset and end dates, and the severity of 26 irAEs based on CTCAE. The ML model was built using extreme gradient boosting. The ePRO data and lab measurement data was used to train the model to detect the onset (0-21 days prior to diagnosis) of irAEs. The dataset was split into training (70 % of the data) and test sets (30 % of the data) by random allocation. The test set was left out from the model training and tuning, and was used only to evaluate the model performance.
Results
The ML model trained to predict the onset of irAEs had an excellent performance with the test dataset with all considered performance metrics. The used performance metrics and their respective scores were the following: accuracy of the model was 0.94, AUC (Area under the curve) 0.97, F1 score 0.75 and Matthew’s correlation coefficient 0.73.
Conclusions
This study indicates that ML based prediction models, trained with a dataset combined from two sources, ePRO data and lab measurements, can predict the onset of irAEs with high accuracy. Thus, ML models utilizing digital symptom monitoring data combined with other data sources could enable early detection of irAEs in ICI treated cancer patients. The results should be validated with a larger dataset from prospective clinical trials.
Clinical trial identification
NCT03928938.
Editorial acknowledgement
Legal entity responsible for the study
The authors/ the study investigators.
Funding
Has not received any funding.
Disclosure
J. Ekstrom; V. Kataja: Full/Part-time employment: Kaiku Health Oy. H. Virtanen: Shareholder/Stockholder/Stock options, Full/Part-time employment: Kaiku Health Oy. J. Koivunen: Advisory/Consultancy: Kaiku Health Oy. All other authors have declared no conflicts of interest.