Development of a Prediction Model for Diagnosis of Acute Poisoning in Patients with Altered Mental Status and Absent History of Alcohol/Drug Ingestion.

Background: Diagnosis of acute poisoning in patients with altered mental status and absent history is a challenging diagnostic problem in clinical practice.

Objective: The aims of the study were to develop a simple clinical tool to stratify risk of acute poisoning in patients with altered mental states and no history of alcohol/drug ingestion, and develop a prediction model using initial observations from which a simple risk score could be derived.

Methods: The study was carried out on non-trauma patients aged 15 years and older admitted with altered mental states and no history of alcohol or drug ingestion. Univariate analysis and logistic regression were carried out and a score was derived and validated.

Results: There were 607 patients included, with mean age of 60.3 years and 54% were male. The regression model performed moderately well on both the training and validation sets with areas under the receiver operating characteristic curve of 0.834 and 0.844, respectively. The risk score correlated with the regression model (R = 0.969). At cutoff thresholds of 20% for the model and 2 for the score, sensitivity and specificity of the regression model (67.6% and 85.6%) and the score (67.6% and 85.4%) were moderate, while positive predictive values were low (43.4%) and negative predictive values were high (94.2%) for both the regression model and the score.

Conclusions: A prediction model with a derived risk score was developed with a high negative predictive value and may have potential in assessing risk of poisoning in altered mental status and may have value in a prehospital environment or at triage.