[3Xin4] Poster session 1

[3Xin4-43] Inference of Tumor Marker Value Fluctuation of Colorectal Cancer Patients and Feature Importance Analysis Using Large-Scale Electronic Medical Records

〇Shumpei Muramatsu¹, Yuika Watanabe¹, Satoshi Hirakawa², Keisuke Yamamoto², Michishi Kashiwabara², Shohei Sawa², Takuya Fukumoto¹, Hiromasa Horiguchi³, Masanao Igarashi³, Takashi Okumura⁴, Yoshinobu Kano¹ (1.Shizuoka University, 2.Seirei Hamamatsu General Hospital, 3.National Hospital Organization, 4.Kitami Institute of Technology)

Keywords:Cancer, Electronic Health Record, NLP

We aim to predict the effectiveness of cancer treatment of colorectal cancer patients, which numerical data are extracted from a large set of electronic medical records.
We use the tumor marker values to predict by hospital examination results and drug therapy implementation information, analyze features that contribute to the prediction.
We were able to make predictions which scores were better than our baselines.
The feature importance analysis suggested that, in addition to the features that are considered as important in the colorectal cancer, some other features contributed to the predictions that were thought to have little relationship with colorectal cancer.