Expansion of electronic health data
Electronic health data has exploded since 2008 in the USA: 9.4% of hospitals had made their data digital in 2008 compared to 83.8% in 2015.
In France, the Health Data Hub was launched at the end of November 2019. This Hub will bring together all the data from French public healthcare institutions : the Health Insurance and hospitals.
This data will be used to drive new algorithms that will enable better diagnostics and treatment, as well as optimizing patient care in the emergency department. French researchers and practitioners will have to take examples from pragmatic projects such as those led by David Sontag and his teams from the Clinical Machine Learning Group at MIT.
Health Data Hub: French healthcare data platform
Health Data Hub : 6 questions sur la plateforme de données de santé et sa polémique
Le Health Data Hub est une nouvelle plateforme française destinée à croiser les bases de données de santé existantes, à…
AI used to make a diagnosis and propose protocols in the Emergency Department.
“The goal is to lay the foundation for the next generation of intelligent electronic health records, in which artificial intelligence is embedded to facilitate medical diagnosis, customize treatment suggestions, make documentation faster and better, and predict adverse events before they occur. “ David Sontag
Using the following data, David Sontag has developed algorithms capable of diagnosing the most common pathologies in the Emergency Department. This algorithm fed a web application of diagnoses made available to emergency practitioners.
- Patient age and gender data (yellow and light orange)
- Patient medication data (dark orange)
- Emergency treatment data (gray)
- Notes taken at the triage by emergency nurses (in blue)
The data are both structured and unstructured (the blue notes), hence the advantage of using Machine Learning which can automate the processing of unstructured data.
Via a web interface, the diagnostic prediction is used to guide care in the Emergency Department. Practitioners are free to choose a care protocol other than the one induced by the algorithm.
Increase Emergency Triage Productivity
David Sontag’s team is using Artificial Intelligence to develop self-completion systems that allow triage nurses to work quickly and accurately. An algorithm makes predictions based on the constants entered upon arrival in the ER and the patient’s medical record.
These predictions are proposed in an “auto-completed” mode to enter the reason for the visit to the Emergency Room. Caregivers are given the possibility to choose the suggested prediction or to enter the reasons for admission to the Emergency Department themselves.
Knowledge graphs to refine diagnostics
Health records are also used by researchers to create graphs of causal knowledge learning the relationships between diseases and symptoms. These knowledge graphs, once established, will be used to propose diagnoses and treatments.
This is how digital health records are transformed into knowledge :
1/ Extraction of data about diseases and associated symptoms from millions of health medical records.
2/ Use of the Machine Learning to establish statistic models between diseases and symptoms (% of having a brain tumor when the patient has a headache).
3/ Transformation of statistical models into knowledge graphs.
Knowledge graphs for prevention
These knowledge graphs can also be used in the context of disease control. Researchers have chosen to use the power of Machine Learning to help evaluate a total of 7,962 biologically active compounds found in food sources.
These molecules were compiled in a database and entered into an algorithm, which determined that among these compounds, 110 molecules appeared to have anticancer properties.
Based on their findings, the researchers then created a knowledge graph that shows the value of different foods based on their anticancer potential. On the knowledge graph, each circle node represents a particular food; the larger the circle, the more anti-cancer molecules the food contains. The interconnecting lines between the nodes show when the connected foods contain a similar range of molecules.
Article written from this video: Lightning Talk: AI-Powered Electronic Medical Records (David Sontag)
Knowledge Graphs & Deep Learning at YouTube
Knowledge graphs and medicine
Welcome to the NCBO BioPortal | NCBO BioPortal
We are integrating the UMLS 2020AA release into BioPortal which will be completed by the end of Wednesday Sept 30, All…