Clinical Strategies for Managing Anticholinergic Toxicity in Overdose vs. Neuroleptic Malignant Syndrome: Diagnostic Challenges and Therapeutic Interventions

Neuroleptic Malignant Syndrome (NMS) and anticholinergic toxicity are two different but clinically similar disorders that are commonly seen in neuropsychiatric and emergency toxicological settings. Rapid differential diagnosis is made more difficult by the symptoms of both disorders, which frequently include delirium, heat, and autonomic instability. Interventions that result from misidentification may be detrimental or ineffectual. Using a machine learning-driven methodology, this work attempts to systematically compare these conditions in order to facilitate prompt and precise clinical decision-making. Random Forest, XGBoost, and Support Vector Machine (SVM) were the three supervised classifiers we used to categorize 450 patient cases into three diagnostic categories: anticholinergic toxicity (n = 189), NMS (n = 170), and other conditions (n = 91). Feature inputs included clinical signs, symptoms, and laboratory biomarkers. All models demonstrated high accuracy (96% - 97%), with the Random Forest classifier slightly outperforming others in F1-scores. Feature importance analysis and SHAP explainability techniques revealed creatine kinase, white blood cell (WBC) count, mydriasis, and dry mucous membranes as the most discriminative features. Specifically, creatine kinase and WBC count were significantly elevated in NMS cases, while anticholinergic toxicity was marked by mydriasis and dry mucous membranes. Treatment protocol comparison further highlighted the clinical need for precise diagnosis. Anticholinergic toxicity often requires supportive care with benzodiazepines and, in some cases, physostigmine, whereas NMS mandates aggressive cooling, dopamine agonist therapy, and intensive monitoring. This study demonstrates the utility of machine learning models in enhancing diagnostic accuracy for toxic syndromes with overlapping presentations. Integrating algorithmic predictions with clinical expertise can substantially improve patient outcomes and guide personalized treatment interventions.