As medical experts increasingly recognize the dangers of alcohol consumption before surgery, a groundbreaking analysis suggests that Artificial Intelligence (AI) could play a crucial role in identifying risky alcohol use patterns in patients leading up to a procedure.
Published in the journal Alcohol: Clinical & Experimental Research, the study leveraged a natural language processing model to scrutinize the medical records of 53,811 patients who underwent surgery from 2012 to 2019. While diagnostic codes are commonly used in electronic medical records, the study aimed to uncover nuanced information, such as notes, test results, or billing data, that could indicate potential alcohol-related risks.
By programming the natural language processing model to detect both diagnostic codes and contextual indicators like exceeding recommended drink thresholds or a history of alcohol-related medical issues, the researchers sought to enhance the identification of risky alcohol use.
The impact of alcohol misuse around surgery is well-documented, leading to increased infection rates, extended hospital stays, and various surgical complications. Among the patients studied, 4.8 percent had diagnosis codes related to alcohol use. However, with the inclusion of contextual clues, the model identified three times as many, totaling 14.5 percent.
Remarkably, the AI model demonstrated a performance level comparable to a panel of human alcohol-use experts, aligning with their classifications in 87 percent of the cases.
The study’s lead author, V.G. Vinod Vydiswaran, an associate professor of learning health sciences at the University of Michigan Medical School, envisions AI as a potential ally for clinicians seeking to pinpoint patients in need of intervention or postoperative support.
The analysis not only sets the stage for addressing alcohol risks but also lays the foundation for identifying additional health risks in primary care and beyond. Vydiswaran emphasized that AI can effectively highlight pertinent information contained in medical notes without the need for exhaustive record reading.
While the researchers plan to eventually release the model publicly, they emphasize the necessity of training it on medical records from individual facilities to ensure accurate and reliable results. Stay tuned for further developments as AI continues to redefine healthcare interventions.
