User login
ORLANDO – Reviewing patient charts for asthma risk factors using natural language processing can be done 8 times faster than reviewing the charts by hand, and with high levels of accuracy, researchers reported here.
Natural language processing (NLP) is a kind of artificial intelligence in which computers are “trained” through a reiterative process to understand human language.
Researchers at Mayo Clinic previously have shown that a program created in-house can successfully and quickly determine patients’ asthma status. In this study, they turned to assessment of asthma risk factors, Chung-Il Wi, MD, assistant professor of pediatrics at Mayo said in a presentation at the joint congress of the American Academy of Allergy, Asthma and Immunology and the World Asthma Organization.
They used a convenience sample of 177 patient charts to train the NLP system. The system extracted – from key terms and sentences in the electronic health record (EHR) – data such as breastfeeding history and history of atopic conditions such as allergic rhinitis, eczema, and food allergy. From parent charts, the system extracted terms related to family history of asthma and other atopic conditions. The performance of the NLP algorithm was assessed by comparison with results of a manual chart review in a test cohort of 220 patient charts.
Researchers found a high level of agreement between the NLP analysis and the manual review. For breastfeeding, the positive predictive value (PPV) of the NLP was 98% and the negative predictive value (NPV) was 86%. For history of atopic conditions the PPV was at or near 100%, with a NPV of 97% to 99%, depending on the condition.
For family history of atopic conditions, the PPV was 91% to 100%, depending on the condition, and the NPV was 96% to 99%.
“Childhood asthma risk factors identified (an) NLP algorithm using EHR has excellent concordance with chart review,” researchers wrote.
Using an average time per chart, researchers found that it would take 7 hours to complete a manual review for the information presented in the study, compared to 50 minutes for the NLP.
The findings, thought to be the first demonstrating NLP’s value for this purpose, suggest “the huge potential of leveraging NLP for asthma care and research,” researchers said.
Dr. Wi said the system can be applied to any EHR system. He said it only makes sense to put an algorithm to use in this way – it saves both clinical time and time in doing research projects.
“Whenever we do asthma research we need to collect asthma risk factors anyway, but we don’t want to do manual chart review anymore in this EMR era,” he said. “Now, the computer can do it.”
SOURCE: Wi C AAAAI/WAO Joint Congress 2018 abstract 637.
Susan Millard, MD, FCCP, comments: This article brings mixed emotions. On one hand, using artificial intelligence brings a more thorough evaluation regarding asthma risk. On the other hand, our pediatric pulmonary subspecialty has gotten diluted over the last 3 decades. We used to regularly do arterial puncture, thoracentesis, and chest tube placement procedures. Now a computer might replace another aspect of our job, too? The practice of medicine is an art and that art should not be lost.
Susan Millard, MD, FCCP, comments: This article brings mixed emotions. On one hand, using artificial intelligence brings a more thorough evaluation regarding asthma risk. On the other hand, our pediatric pulmonary subspecialty has gotten diluted over the last 3 decades. We used to regularly do arterial puncture, thoracentesis, and chest tube placement procedures. Now a computer might replace another aspect of our job, too? The practice of medicine is an art and that art should not be lost.
Susan Millard, MD, FCCP, comments: This article brings mixed emotions. On one hand, using artificial intelligence brings a more thorough evaluation regarding asthma risk. On the other hand, our pediatric pulmonary subspecialty has gotten diluted over the last 3 decades. We used to regularly do arterial puncture, thoracentesis, and chest tube placement procedures. Now a computer might replace another aspect of our job, too? The practice of medicine is an art and that art should not be lost.
ORLANDO – Reviewing patient charts for asthma risk factors using natural language processing can be done 8 times faster than reviewing the charts by hand, and with high levels of accuracy, researchers reported here.
Natural language processing (NLP) is a kind of artificial intelligence in which computers are “trained” through a reiterative process to understand human language.
Researchers at Mayo Clinic previously have shown that a program created in-house can successfully and quickly determine patients’ asthma status. In this study, they turned to assessment of asthma risk factors, Chung-Il Wi, MD, assistant professor of pediatrics at Mayo said in a presentation at the joint congress of the American Academy of Allergy, Asthma and Immunology and the World Asthma Organization.
They used a convenience sample of 177 patient charts to train the NLP system. The system extracted – from key terms and sentences in the electronic health record (EHR) – data such as breastfeeding history and history of atopic conditions such as allergic rhinitis, eczema, and food allergy. From parent charts, the system extracted terms related to family history of asthma and other atopic conditions. The performance of the NLP algorithm was assessed by comparison with results of a manual chart review in a test cohort of 220 patient charts.
Researchers found a high level of agreement between the NLP analysis and the manual review. For breastfeeding, the positive predictive value (PPV) of the NLP was 98% and the negative predictive value (NPV) was 86%. For history of atopic conditions the PPV was at or near 100%, with a NPV of 97% to 99%, depending on the condition.
For family history of atopic conditions, the PPV was 91% to 100%, depending on the condition, and the NPV was 96% to 99%.
“Childhood asthma risk factors identified (an) NLP algorithm using EHR has excellent concordance with chart review,” researchers wrote.
Using an average time per chart, researchers found that it would take 7 hours to complete a manual review for the information presented in the study, compared to 50 minutes for the NLP.
The findings, thought to be the first demonstrating NLP’s value for this purpose, suggest “the huge potential of leveraging NLP for asthma care and research,” researchers said.
Dr. Wi said the system can be applied to any EHR system. He said it only makes sense to put an algorithm to use in this way – it saves both clinical time and time in doing research projects.
“Whenever we do asthma research we need to collect asthma risk factors anyway, but we don’t want to do manual chart review anymore in this EMR era,” he said. “Now, the computer can do it.”
SOURCE: Wi C AAAAI/WAO Joint Congress 2018 abstract 637.
ORLANDO – Reviewing patient charts for asthma risk factors using natural language processing can be done 8 times faster than reviewing the charts by hand, and with high levels of accuracy, researchers reported here.
Natural language processing (NLP) is a kind of artificial intelligence in which computers are “trained” through a reiterative process to understand human language.
Researchers at Mayo Clinic previously have shown that a program created in-house can successfully and quickly determine patients’ asthma status. In this study, they turned to assessment of asthma risk factors, Chung-Il Wi, MD, assistant professor of pediatrics at Mayo said in a presentation at the joint congress of the American Academy of Allergy, Asthma and Immunology and the World Asthma Organization.
They used a convenience sample of 177 patient charts to train the NLP system. The system extracted – from key terms and sentences in the electronic health record (EHR) – data such as breastfeeding history and history of atopic conditions such as allergic rhinitis, eczema, and food allergy. From parent charts, the system extracted terms related to family history of asthma and other atopic conditions. The performance of the NLP algorithm was assessed by comparison with results of a manual chart review in a test cohort of 220 patient charts.
Researchers found a high level of agreement between the NLP analysis and the manual review. For breastfeeding, the positive predictive value (PPV) of the NLP was 98% and the negative predictive value (NPV) was 86%. For history of atopic conditions the PPV was at or near 100%, with a NPV of 97% to 99%, depending on the condition.
For family history of atopic conditions, the PPV was 91% to 100%, depending on the condition, and the NPV was 96% to 99%.
“Childhood asthma risk factors identified (an) NLP algorithm using EHR has excellent concordance with chart review,” researchers wrote.
Using an average time per chart, researchers found that it would take 7 hours to complete a manual review for the information presented in the study, compared to 50 minutes for the NLP.
The findings, thought to be the first demonstrating NLP’s value for this purpose, suggest “the huge potential of leveraging NLP for asthma care and research,” researchers said.
Dr. Wi said the system can be applied to any EHR system. He said it only makes sense to put an algorithm to use in this way – it saves both clinical time and time in doing research projects.
“Whenever we do asthma research we need to collect asthma risk factors anyway, but we don’t want to do manual chart review anymore in this EMR era,” he said. “Now, the computer can do it.”
SOURCE: Wi C AAAAI/WAO Joint Congress 2018 abstract 637.
REPORTING FROM AAAAI/WAO JOINT CONGRESS 2018
Key clinical point: Using natural language processing, a form of artificial intelligence that trains computers to discern natural human language, accurately and quickly identified asthma risk factors.
Major finding: In 50 minutes, a computer program performed the risk review that took 7 hours for manual chart review, with positive and negative predictive values for most risk factors in the 90% to 100% range.
Study details: A sample of charts for patients in the Olmsted County Birth Cohort, some analyzed with natural language processing and some reviewed manually.
Disclosures: No disclosures.
Source: Wi C AAAAI/WAO Joint Congress 2018 abstract 637.
