Oops, you're using an old version of your browser so some of the features on this page may not be displaying properly.

MINIMAL Requirements: Google Chrome 24+Mozilla Firefox 20+Internet Explorer 11Opera 15–18Apple Safari 7SeaMonkey 2.15-2.23

e-Poster Display Session

38P - Predicting objective response rate (ORR) in immune checkpoint inhibitor (ICI) therapies with machine learning (ML) by combining clinical and patient-reported data


09 Dec 2020


e-Poster Display Session


Sanna Iivanainen


Annals of Oncology (2020) 31 (suppl_7): S1428-S1440. 10.1016/annonc/annonc391


S. Iivanainen1, J. Esktrom2, H. Virtanen2, L. Lang2, V. Kataja2

Author affiliations

  • 1 Oncology And Radiotherapy, Oulu University Hospital, 90220 - Oulu/FI
  • 2 Kaiku Health Oy, Helsinki/FI

Abstract 38P


ICIs have been approved as standard of care in several malignancies. However, only a subset of eligible patients benefits from ICIs according to overall response rate (ORR) which predicts long-term benefit from therapy. Predicting ORR could enable more rational use of ICIs. The aim of this study was to investigate if ML could be used to predict ORR of patients undergoing ICIs using clinical and patient-reported data.


ML-based prediction model was built by using data from 31 patients receiving ICI therapies for various advanced cancers. Several data sources were used as inputs for the model. Clinician-assessed treatment responses (n=63), immune-related adverse events (irAEs) according to CTCAE, and laboratory measurements (9 values) were collected prospectively. Patient-reported symptom data including 18 monitored symptoms was collected using the Kaiku Health digital platform. In addition, time from treatment initiation, age and sex were included. Preceding lab values and patient-reported symptoms, as changes from the baseline, were linked to the treatment responses. The prediction model for ORR was built using gradient boosting technique (XGBoost). Prediction performance for unseen samples was evaluated using leave-one-out cross-validation (LOOCV), which trained and tested 63 models, each time iteratively leaving one sample out as a test set. The LOOCV prediction performance was evaluated with accuracy, AUC (Area Under Curve), F1 score and MCC (Matthew’s correlation coefficient).


Table: 38P

Accuracy AUC F1 score MCC
XGBoost LOOCV performance 75 % 0,71 0,58 0,40


This study demonstrated that it is possible to predict ORR for patients undergoing ICI therapies with ML model combining clinical, routine laboratory, and patient-reported data even with a limited size cohort. These promising results indicate that ML-based approaches in treatment response prediction should be investigated and validated further with a larger dataset.

Clinical trial identification


Legal entity responsible for the study

Jussi Koivunen, Sanna Iivanainen; OYS, MRC Oulu.


Has not received any funding.


J. Esktrom: Full/Part-time employment: Kaiku Health Oy. H. Virtanen: Shareholder/Stockholder/Stock options, Full/Part-time employment: Kaiku Health Oy. L. Lang: Full/Part-time employment: Kaiku Health Oy. V. Kataja: Full/Part-time employment: Kaiku Health Oy. J. Koivunen: Advisory/Consultancy: Kaiku Health Oy. All other authors have declared no conflicts of interest.

This site uses cookies. Some of these cookies are essential, while others help us improve your experience by providing insights into how the site is being used.

For more detailed information on the cookies we use, please check our Privacy Policy.

Customise settings