Distinguishing acute poisoning agents using machine learning models derived from the National Poison Data System (NPDS)

سال انتشار: 1402
نوع سند: مقاله کنفرانسی
زبان: انگلیسی
مشاهده: 196

نسخه کامل این مقاله ارائه نشده است و در دسترس نمی باشد

این مقاله در بخشهای موضوعی زیر دسته بندی شده است:

استخراج به نرم افزارهای پژوهشی:

لینک ثابت به این مقاله:

شناسه ملی سند علمی:

AIMS01_243

تاریخ نمایه سازی: 1 مرداد 1402

چکیده مقاله:

Background and aims: Acute poisoning is known to be a globally major health problem and dueto a patient’s inability to verbally convey the exposure history, obtaining an accurate exposurehistory can be challenging. Artificial intelligence (AI) uses logic and decision trees to replicatehuman decision-making. Machine learning (ML) is a subgroup of AI that benefit from statisticalmodels to execute tasks based on inference but not definite instructions, which allows the modelmore flexibility to learn and improve itself. This study aimed to develop a machine-learning algorithmthat can predict and distinguish eight poisoning agents based on their clinical symptoms.Method: Data were obtained from the National Poison Data System from ۲۰۱۴ to ۲۰۱۸, forpatients with single-agent exposure to ۸ drugs/drug classes (lithium, aspirin, benzodiazepines,bupropion, acetaminophen, calcium channel blockers, diphenhydramine, and sulfonylureas). Thefour classifier prediction models were applied including LightGBM, Logistic regression, XGBoost,and CatBoost. Standard performance measures were also calculated including accuracy,F۱-score, recall, precision, and specificity.Results: In total ۲۰۱۰۳۱ cases were included to develop and test the algorithms. Accuracy amongthese four models ranging from ۷۷-۸۰%, precision, and F۱-scores were ۷۶-۸۰%, and recall was۷۷-۷۸%. Overall specificity was ۹۲% for all four models. Accuracy was highest for distinguishingsulfonylureas, acetaminophen, benzodiazepines, and diphenhydramine poisoning. F۱ scores werereported to be highest for correctly classifying sulfonylureas, acetaminophen, and benzodiazepinepoisonings. Also, recall was highest for sulfonylureas, acetaminophen, and benzodiazepines, andlowest for bupropion. Specificity was >۹۹% for models of sulfonylureas, lithium, calcium channelblockers, and aspirin.Conclusion: LightGBM and CatBoost classifier prediction models had the highest performanceof those which have been tested. The algorithms were most accurate in classifying sulfonylureapoisonings, followed by acetaminophen, benzodiazepines, diphenhydramine, and bupropion poisoning.Although the clinical utility has not been studied, with further development in this subjectthis may serve as a useful diagnostic aid.

نویسندگان

Omid Mehrpour

Data Science Institute, Southern Methodist University, Dallas, TX, USA

Samaneh Nakhaee

Medical Toxicology and Drug Abuse Research Center (MTDRC), Birjand University of Medical Sciences (BUMS), Birjand, Iran

Alireza Kooshki

Student Research Committee, Birjand University of Medical Sciences, Birjand, Iran

Christopher Hoyte

Data Science Institute, Southern Methodist University, Dallas, TX, USA- Department of Emergency Medicine, University of Colorado School of Medicine, Aurora, CO, USA

Heather Delva-Clark

CPC Clinical Research ۲۱۱۵ N Scranton St Suite ۲۰۴۰

Abdullah Al Masud

Data Scientist, Hiperdyne Corporation, Tokyo, Japan