A patient's electronic medical record includes diagnosis codes, but may also include information such as notes, test results, and claims data that may suggest unsafe alcohol use.
To find contextual clues, researchers looked at both diagnosis codes and other indicators of hazardous alcohol use, such as weekly alcohol consumption above recommended thresholds or a history of medical problems related to alcohol misuse. We have programmed a natural language processing model to identify.
Alcohol misuse before and after surgery is associated with increased infection rates, longer hospital stays, and other surgical complications. Of the patients studied, 4.8% had medical records that included diagnostic codes related to alcohol consumption. Using situational cues, the model classified three times as many people as at risk, a total of 14.5 percent.
The model showed similar results to a human alcohol use expert panel, agreeing with the classification of a subset of records 87% of the time.
The researchers concluded that the findings show that AI is a potential partner for clinicians looking to identify patients who require intervention or post-operative support.
VG Vinod Vidiswaran, associate professor of learning health sciences at the University of Michigan Medical School and lead author of the paper, said the analysis “lays the foundation for primary care and other risk identification efforts, with appropriate validation. There is a possibility.”news release. “Essentially, this is a way for providers to highlight content that is already included in notes made by other providers without the other provider reading the entire record.”
The researchers say they plan to eventually release the model publicly, but note that it will need to be trained on individual facility medical records.
Automatic detection of hazardous alcohol consumption before surgery using natural language processing
Alcohol: Clinical and experimental studies