ABSTRACT
We introduce MediSage, an AI decision support assistant for medical professionals and caregivers that simplifies the way in which they interact with different modalities of electronic health records (EHRs) through a conversational interface. It provides step-by-step reasoning support to an end-user to summarize patient health, predict patient outcomes and provide comprehensive and personalized healthcare recommendations. MediSage provides these reasoning capabilities by using a knowledge graph that combines general purpose clinical knowledge resources with recent-most information from the EHR data. By combining the structured representation of knowledge with the predictive power of neural models trained over both EHR and knowledge graph data, MediSage brings explainability by construction and represents a stepping stone into the future through further integration with biomedical language models.
Supplemental Material
- Khushbu et al. Agarwal. 2019. Snomed2Vec: Random Walk and Poincar\ ’e Embeddings of a Clinical Knowledge Base for Healthcare Analytics. arXiv preprint arXiv:1907.08650 (2019).Google Scholar
- Khushbu et al. Agarwal. 2022. Preparing for the next pandemic via transfer learning from existing diseases with hierarchical multi-modal BERT: a study on COVID-19 outcome prediction. Scientific Reports 12, 1 (2022), 10748.Google ScholarCross Ref
- Payal Chandak, Kexin Huang, and Marinka Zitnik. 2022. Building a knowledge graph to enable precision medicine. BioRxiv (2022), 2022–05.Google ScholarCross Ref
- Sutanay Choudhury, Sumit Purohit, Peng Lin, Yinghui Wu, Lawrence Holder, and Khushbu Agarwal. 2018. Percolator: Scalable pattern discovery in dynamic graphs. In WSDM.Google ScholarDigital Library
- Ian C Covert, Scott Lundberg, and Su-In Lee. 2021. Explaining by removing: A unified framework for model explanation. The Journal of Machine Learning Research 22, 1 (2021), 9477–9566.Google ScholarDigital Library
- Somalee Datta et al.2020. A new paradigm for accelerating clinical data science at Stanford Medicine. (2020). https://arxiv.org/abs/2003.10534Google Scholar
- Jiří et al. Gallo. 2022. Odds-ratio network for postoperative factors revealing differences in the 2-year longitudinal pattern of satisfaction between women and men after total knee arthroplasty. Scientific Reports 12, 1 (2022), 17470.Google ScholarCross Ref
- Scott M Lundberg and Su-In Lee. 2017. A unified approach to interpreting model predictions. Advances in neural information processing systems 30 (2017).Google Scholar
- Renqian et al. Luo. 2022. BioGPT: generative pre-trained transformer for biomedical text generation and mining. Briefings in Bioinformatics 23, 6 (2022).Google Scholar
- Long et al. Ouyang. 2022. Training language models to follow instructions with human feedback. arXiv preprint arXiv:2203.02155 (2022).Google Scholar
- Karan et al. Singhal. 2022. Large Language Models Encode Clinical Knowledge. arXiv preprint arXiv:2212.13138 (2022).Google Scholar
- Ping Wang, Khushbu Agarwal, Colby Ham, Sutanay Choudhury, and Chandan K Reddy. 2021. Self-supervised learning of contextual embeddings for link prediction in heterogeneous networks. In Proceedings of the Web Conference 2021.Google ScholarDigital Library
- Michihiro Yasunaga, Jure Leskovec, and Percy Liang. 2022. Linkbert: Pretraining language models with document links. arXiv preprint arXiv:2203.15827 (2022).Google Scholar
Recommendations
Quality and Cost Improvement of Healthcare via Complementary Measurement and Diagnosis of Patient General Health Outcome Using Electronic Health Record Data: Research Rationale and Design
In this evolving `third era of health', one of the US Health Care Reform Act's goals is to effectively facilitate the primary care physician's ability to better diagnose and manage the health outcome of the outpatient. That goal must include research on ...
Comments