Predictive in silico modeling of emetic potency of liquid cleaning products using an historical in vivo database

The induction of vomiting by activation of mechanisms protecting the body against ingested toxins is not confined to natural products but can occur in response to manmade medicinal and non-medicinal products such as liquid cleaning products where it is a commonly reported adverse effect of accidental ingestion. The present study examined the utility of an historic database (>30 years old) reporting emetic effects of 98 orally administered liquid cleaning formulations studied in vivo (canine model) to objectively identify the main pro-emetic constituents and to derive a predictive model. Data were analysed by categorizing the formulation constituents into 10 main groups followed by using multivariate correlation, partial least squares and recursive partitioning analysis. Using the ED₅₀ we objectively identified high ionic strength, non-ionic surfactants (alcohol ethoxylate) and alkaline pH as the main pro-emetic factors. Additionally, a mathematical model was developed which allows prediction of the ED₅₀ based on formulation. The limitations of the use of historic data and the model are discussed. The results have practical applications in new product formulation and safety but additionally the principles underpinning this in silico study have wider applicability in demonstrating the potential utility of such archival data in current research contributing to animal replacement.