Clinical note-scanning technology could help decipher between acute and chronic low back pain

Last updated: 04-05-2020

Read original article here

Clinical note-scanning technology could help decipher between acute and chronic low back pain

Clinical note-scanning technology could help decipher between acute and chronic low back pain
March 27, 2020
Add Topic To Email Alerts
Please provide your email address to receive an email when new articles are posted on this topic.
Submit
Receive an email when new articles are posted on this topic.
CLICK HERE TO SUBSCRIBE
Scanning health records with artificial intelligence technology may create a distinction between acute and chronic lower back pain and their respective treatments, according to a recently published study.
Ismail Nabeel
The study by Ismail Nabeel, MD, MPH, and colleagues at the Icahn School of Medicine at Mount Sinai, came out of a necessity for a more efficient way to decipher between low back pain (LBP) severities. For example, while acute and chronic LBP are different conditions with different treatments, these use the same International ICD-10 code (M54.5), and “can be differentiated only by retrospective chart reviews. This prevents an efficient definition of data-driven guidelines for billing and therapy recommendations, such as return-to-work options,” the researchers wrote in their study.
“It was very difficult to understand – just by IC-10 codes – what exactly the level of acuity of a condition is,” Nabeel told Healio Orthopedics. “A medical condition does not behave similar to a code. In practice, we don’t see a definition of pain. The only way to really understand or define the acuity of pain or the characteristics of pain is to read the physician notes.”
For the study, Nabeel and colleagues contrasted “supervised and unsupervised strategies for automated identification: keyword search, topic modeling, logistic regression with bag of n-grams and manual features, and deep learning (a convolutional neural network-based architecture [ConvNet]),” they wrote.
They used manual annotation and a dataset of 17,409 clinical notes from various practices to train the convolutional neural network-based architecture.
“The convolutional network does this in a very similar way as we do as humans. We are giving them resources to train, and that’s helping them understand a condition,” Nabeel said. Overall, researchers found 891 documents were manually noted as acute LBP and 2,973 were linked with LBP according to ICD-10 codes.
Of the tested models, ConvNet received the best results with a precision of 0.65, a recall of 0.73 and a F score of 0.70. The area under the receiver operating characteristic curve was 0.98, according to the study. According to the researchers, the ability to autonomously identify and separate LBP cases would help create a more efficient system of therapeutic strategies, billing guidelines and management options for acute LBP.
“The other thing we have demonstrated is the scalability of the approach, in terms of moving from a 100- people or a chart review process, which is manually done at this point, to a more machine based-way of modeling,” Nabeel said.
He stressed the significant part of this work is application at the point of care.
“If you determine that the pain condition is acute and certain management options can be given to these individuals, the turnaround time to implement these into practice gets significantly reduced, because we are doing it in an automated fashion,” Nabeel said.
“I think it is sort of a foundational work, I call it. We also want to replicate a similar understanding in other musculoskeletal conditions like knee or shoulder,” Nabeel said. “The field is constantly evolving.” – by Max R. Wursta
 
Disclosures: Nabeel reports no relevant financial disclosures. This research was funded by Pilot Projects Research Training Program of the NY and NJ Education and Research Center (ERC), National Institute for Occupational Safety and Health grant # T42 OH 008422, Hasso Plattner Foundation and NVIDIA.
Scanning health records with artificial intelligence technology may create a distinction between acute and chronic lower back pain and their respective treatments, according to a recently published study.
Ismail Nabeel
The study by Ismail Nabeel, MD, MPH, and colleagues at the Icahn School of Medicine at Mount Sinai, came out of a necessity for a more efficient way to decipher between low back pain (LBP) severities. For example, while acute and chronic LBP are different conditions with different treatments, these use the same International ICD-10 code (M54.5), and “can be differentiated only by retrospective chart reviews. This prevents an efficient definition of data-driven guidelines for billing and therapy recommendations, such as return-to-work options,” the researchers wrote in their study.
“It was very difficult to understand – just by IC-10 codes – what exactly the level of acuity of a condition is,” Nabeel told Healio Orthopedics. “A medical condition does not behave similar to a code. In practice, we don’t see a definition of pain. The only way to really understand or define the acuity of pain or the characteristics of pain is to read the physician notes.”
For the study, Nabeel and colleagues contrasted “supervised and unsupervised strategies for automated identification: keyword search, topic modeling, logistic regression with bag of n-grams and manual features, and deep learning (a convolutional neural network-based architecture [ConvNet]),” they wrote.
They used manual annotation and a dataset of 17,409 clinical notes from various practices to train the convolutional neural network-based architecture.
“The convolutional network does this in a very similar way as we do as humans. We are giving them resources to train, and that’s helping them understand a condition,” Nabeel said. Overall, researchers found 891 documents were manually noted as acute LBP and 2,973 were linked with LBP according to ICD-10 codes.
Of the tested models, ConvNet received the best results with a precision of 0.65, a recall of 0.73 and a F score of 0.70. The area under the receiver operating characteristic curve was 0.98, according to the study. According to the researchers, the ability to autonomously identify and separate LBP cases would help create a more efficient system of therapeutic strategies, billing guidelines and management options for acute LBP.
“The other thing we have demonstrated is the scalability of the approach, in terms of moving from a 100- people or a chart review process, which is manually done at this point, to a more machine based-way of modeling,” Nabeel said.
PAGE BREAK
He stressed the significant part of this work is application at the point of care.
“If you determine that the pain condition is acute and certain management options can be given to these individuals, the turnaround time to implement these into practice gets significantly reduced, because we are doing it in an automated fashion,” Nabeel said.
“I think it is sort of a foundational work, I call it. We also want to replicate a similar understanding in other musculoskeletal conditions like knee or shoulder,” Nabeel said. “The field is constantly evolving.” – by Max R. Wursta
 
Disclosures: Nabeel reports no relevant financial disclosures. This research was funded by Pilot Projects Research Training Program of the NY and NJ Education and Research Center (ERC), National Institute for Occupational Safety and Health grant # T42 OH 008422, Hasso Plattner Foundation and NVIDIA.
See Also


Read the rest of this article here