Development and validation of a Partial Least Squares-Discriminant Analysis (PLS-DA) model based on the determination of ethyl glucuronide (EtG) and fatty acid ethyl esters (FAEEs) in hair for the diagnosis of chronic alcohol abuse

• Published in: Forensic science international 2018-01, Vol.282, p.221-230
• Main Authors: Alladio, E., Giacomelli, L., Biosa, G., Corcia, D.Di, Gerace, E., Salomone, A., Vincenti, M.
• Format: Article
• Language: English
• Subjects: Abuse; Alcohol; Alcohol abuse; Alcohol biomarkers; Alcoholism; Alcohols; Biomarkers; Classification; Consumption; Cut-off; Discriminant analysis; Drug abuse; Esters; EtG; Ethyl esters; FAEE; Fatty acids; Forensic sciences; Hair; Medical diagnosis; Metabolism; Principal components analysis
• ISSN: 0379-0738; 1872-6283
• DOI: 10.1016/j.forsciint.2017.11.010

[Display omitted] •Superior reliability of FAEEs over indirect biomarkers was demonstrated once more.•Chemometrics turns effective for the identification of chronic alcohol drinkers.•PLS-DA combined the predictive capabilities of both EtG and FAEEs parameters.•This model yielded a classification decision based on probabilistic foundation.•PLS-DA overcomes most of the drawbacks related to the use of single cut-off values. The chronic intake of an excessive amount of alcohol is currently ascertained by determining the concentration of direct alcohol metabolites in the hair samples of the alleged abusers, including ethyl glucuronide (EtG) and, less frequently, fatty acid ethyl esters (FAEEs). Indirect blood biomarkers of alcohol abuse are still determined to support hair EtG results and diagnose a consequent liver impairment. In the present study, the supporting role of hair FAEEs is compared with indirect blood biomarkers with respect to the contexts in which hair EtG interpretation is uncertain. Receiver Operating Characteristics (ROC) curves and multivariate Principal Component Analysis (PCA) demonstrated much stronger correlation of EtG results with FAEEs than with any single indirect biomarker or their combinations. Partial Least Squares Discriminant Analysis (PLS-DA) models based on hair EtG and FAEEs were developed to maximize the biomarkers information content on a multivariate background. The final PLS-DA model yielded 100% correct classification on a training/evaluation dataset of 155 subjects, including both chronic alcohol abusers and social drinkers. Then, the PLS-DA model was validated on an external dataset of 81 individual providing optimal discrimination ability between chronic alcohol abusers and social drinkers, in terms of specificity and sensitivity. The PLS-DA scores obtained for each subject, with respect to the PLS-DA model threshold that separates the probabilistic distributions for the two classes, furnished a likelihood ratio value, which in turn conveys the strength of the experimental data support to the classification decision, within a Bayesian logic. Typical boundary real cases from daily work are discussed, too.