A quick practical guide for building learning health systems (LHS). It is focused on providing practical information for developers and clinical teams to get started with developing LHS quickly. The guide is dynamic - updated frequently when new information available.
>>> View LHS Practical Guide >>>
- Provide technical information for developers and health organizations to quickly understand the benefits of LHS and get started with LHS.
- Explain it is more practical to build small ML-LHS units rather than being bogged down by the grand LHS picture.
- Propose a new "synthetic+real" strategy: simulate ML-LHS unit with synthetic patient data first and then apply the process to build ML-LHS unit with real EHR data.
- ML-LHS Performance Hypothesis: Due to its inherent data-centric ML approach, ML-LHS can ultimately achieve high prediction performance (>95%) for most diseases and conditions.
- ML-LHS Equity Hypothesis: ML-LHS over hospital-led clinical research networks can effectively enable small clinics with seamlessly disseminated ML models and thus help reduce health care disparities in underserved populations.
- ML-LHS AI Hypothesis: Because of the intrinsic advantage of having both data-driven and deployment-oriented approaches in ML-LHS, most EMR-based AI tools can be developed and deployed in the form of ML-LHS.
- Open Synthetic Patient Data
- MIT MIMIC medical data repository
- Harvard Dataverse repository
- Mendeley Data repository
- CMS Data
The LHS guide is put together by AJ Chen (Co-Chair of LHS Tech Forum Initiative, Learning Health Community) for encouraging collaborations in the global LHS community. If you are interested in discussing LHS R&D and implementation, you may contact AJ (ajchen(at)web2express.org).