Differential Toxicological Diagnoses Using a Computerized Knowledge-Based Model

Background: Utilizing poison center data, a prototype knowledge-based system for developing a differential diagnosis list for toxic exposures was created. The goal of the system is to generate differential diagnoses for unknown exposure cases based on the clinical effects observed in patients. Serving as a medical decision support system, the system seeks to provide pertinent case-based summary data that is normally unavailable to medical practitioners. Methods: The computerized application was automatically generated by applying data mining techniques to a database supplied by the Florida Poison Information Center Network. For diagnosis, the system makes use of pre-test probabilities and likelihood ratios. To overcome the limitations of traditional likelihood ratios, the equation employed by the system is adjusted to account for every possible outcome. Using adjusted likelihood ratios facilitates system stability while closely modeling the calculations of traditional likelihood ratios. Accuracies are calculated as the percentage of correct diagnoses in the top 10% of all possible diagnoses. Results: Trained and tested on single exposure data from 2002-2005, the system achieved accuracies as high as 81.0% on cases involving at least three clinical effects. Adding exposure data from 2006, the system was trained on a combination of single exposures as well as the primary contributors in multiple exposure cases. With this training combination, the system achieved accuracies as high as 86.9% when diagnosing other primary contributors in multiple exposure cases. Discussion: The results of this research are modest, yet promising. The current system design assumes no prior knowledge in the field of toxicology. System performance should improve by the addition of certain knowledge, such as removing the "unknown toxin" diagnosis, combining various formulations of the same generic substance, and grouping substances by intelligently based on similar clinical effects. Conclusion: Many improvements to increase system utility and accuracy are readily apparent. With time, it is hoped that these studies will yield an effective consultant for the diagnosis of primary contributors in toxic exposure cases.