•  
  •  
 

Saudi Toxicology Journal

Keywords

Potentially Inappropriate Prescribing, Older Patients, Emergency Departments, Machine Learning, STOPP/START Criteria, Predictive Modeling

Document Type

Research Article

Abstract

Potentially inappropriate prescribing (PIP) is an avoidable factor leading to adverse drug events in older adults, particularly in the emergency department (ED). This study aimed to train and validate a machine learning (ML) model to predict the probability of PIP in elderly patients in the ED. This retrospective analysis encompassed patients aged 65 years and above who presented to the Al-Qatif Central Hospital. PIPs were identified using a validated screening tool designed for older persons to ensure appropriate treatment. Variables have been detected by least absolute shrinkage and selection operator regression. An extreme gradient boosting model was trained and assessed across training, internal validation, and external validation cohorts. The evaluation of model performance was performed using receiver operating characteristic areas under the curve (ROC-AUC) and decision curve analysis. Among 4942 patients, 80% exhibited at least one PIP. The final model exhibited robust discrimination in the external validation cohort (ROC-AUC = 0.7, accuracy = 0.84, Brier score = 0.13). Important factors contributing to PIP risk were cardiovascular disease, the number of medications prescribed in the ED, arthritis-related diseases, and musculoskeletal pain. The model exhibited effective calibration and offered clinical value across a broad spectrum of thresholds (20 – 60%). This model demonstrated robust predictive efficacy in detecting PIP risk among older ED patients. Its implementation may augment prescribing safety and promote clinical outcomes in geriatric care.

Publisher

Saudi Toxicology Society

DOI

https://doi.org/10.70957/uqu.edu.sa/s.toxicology.s/stj.2025.1.3.4

Download

Included in

Pharmacology Commons

Share

COinS