Predicting Outcomes for Cancer Patients with Transformer-Based Multi-task Learning
- Publisher:
- Springer International Publishing
- Publication Type:
- Conference Proceeding
- Citation:
- Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), 2022, 13151 LNAI, pp. 381-392
- Issue Date:
- 2022-01-01
In Progress
Filename | Description | Size | |||
---|---|---|---|---|---|
AJCAI21_Transformer_based_multi_task_for_cancer.pdf | Accepted version | 394.28 kB |
Copyright Clearance Process
- Recently Added
- In Progress
- Open Access
This item is being processed and is not currently available.
Cancer patients often experience numerous hospital admissions as a result of their cancer and treatment, which can negatively impact treatment progress and quality of life. Accurately predicting outcomes for cancer patients is therefore crucial in providing personalised care and improving patient outcomes. Existing models leveraging deep learning with Electronic Health Record (EHR) data to predict outcomes for cancer patients are limited, despite the demonstrated success of these approaches with cancer imaging data and non-cancer EHR applications. Additionally, current methods focus on single-task predictions, and increasing evidence suggests jointly training a model on two related tasks can improve predictive performance. To address these limitations, we propose a Transformer-based Multi-Task (TransMT) model that captures relationships between diagnosis codes and sequential hospital visits to simultaneously predict related outcomes for hospitalised cancer patients. Experiments conducted on two public datasets show the proposed model outperforms both single-task and recurrent neural network approaches in predicting future diagnosis and hospital readmission, and demonstrates the benefits of using deep learning with EHR data for cancer-related research.
Please use this identifier to cite or link to this item: