Diagnostic Data in Clinical Toxicology—Should We Use a Bayesian Approach?^*

Abstract

A number of toxidromes (toxicology syndromes) have been described for various poisonings and are promoted as a means of reaching a diagnosis in patients presenting with unknown poisonings. Many are based entirely on deductive reasoning from the known pharmacological effects of these drugs rather than on documented clinical experience. In this paper, we used our database, where we have recorded clinical signs on presentation in unselected poisonings to explore how clinical signs actually alter the odds of ingestion of different poisons. Many signs substantially altered the list of drugs likely to have been ingested. We found that the most important factor determining whether an unconscious patient had ingested a particular drug was how frequently that drug was taken generally (i.e., the a priori probability), rather than the presence of any particular physical sign. It also follows that our (or anyone else's) intuitive or deductive approach to diagnosis, derived from experience, will not necessarily be very useful at another place where predominantly different drugs are involved in poisoning. Our data were used to derive odds ratios as a measure of the strength of association of physical signs or investigations with ingested poisons. These can be used to develop simple diagnostic algorithms or flow charts to identify the most likely drugs ingested, or using more complicated programming, could also be used to calculate the precise probability of different drug-ingestion using Bayes' Theorem. The usefulness (i.e., external validity) of clinical research from other centers can also be determined.

Keywords :

• Toxidrome
• Diagnosis
• Poisoning
• Likelihood
• Data collection methodology