AI Case Study
NHS investigates whether inflammation plays a role in mental illnesses by extracting blood test results from unstructured medical records with advanced natural language processing
The NHS wants to identify whether inflammation plays a role in psychiatric disorders such as bipolar. They have access to hundreds of thousands of anonymised records that include results of blood tests and inflammation. But this information is unstructured and hard to extract. The NHS is working with EvolutionAI and University College London to extract this blood test information using advanced natural language processing.
Industry
Healthcare
Healthcare Providers And Services
Project Overview
"By pre-reading large numbers of example records, Evolution AI’s system has been trained, like a doctor, to rapidly recognise the ‘language of blood tests’ in the 200, 000 anonymised records that will be used in this research study. An example of a record entry might be: ‘Xxxx presents with a history of a manic mood type. His associated symptoms include irritable mood. He reports the duration has been for improving for last 2 weeks. Blood tests showed Cholesterol, Total 210 mg/dl, CRP of 12mg/L and HbA1c level 6.0%’.
Having extracted the data using Evolution AI products, the research team will map how inflammatory markers, such as CRP, change at times of relapse and recovery, over the eight years the psychiatric records cover."
Reported Results
Results are undisclosed. However EvolutionAI claims "by pre-reading large numbers of example records, Evolution AI’s system has been trained, like a doctor, to rapidly recognise the ‘language of blood tests’ in the 200, 000 anonymised records that will be used in this research study."
Technology
Natural language processing
Function
Strategy
Data Science
Background
"Faced with pages of notes describing several episodes of an illness, how do doctors find small details such as blood test results in the text? They use their medical knowledge to identify likely blood test ‘numbers’ e.g. 12mg/L, and contextual clues such as clusters of associated words like ‘blood’, ‘test’, ‘CRP’, and ‘cholesterol’.
Because psychiatric records are so long and unstructured, medical research studies have rarely looked at more than a few hundred i.e. numbers that a small team of human researchers can read."
Benefits
Data
"200, 000 anonymised records that will be used in this research study."