A model distinguishing clinical sample groups does not automatically define an impairment threshold.
The investigators collected saliva from suspected alprazolam-poisoning patients and comparison groups, selected samples for model training and challenged discrimination with additional samples. Fourier-transform infrared spectral patterns were used to separate groups classified by clinical and laboratory information.
Model performance depends on the reference labels, sample selection and validation design. Co-medication, illness and saliva composition can affect spectra. A classification score is not interchangeable with analyte-specific quantitative mass spectrometry or a measured blood concentration.
Ask what the model actually predicts and how independent its test population was. A positive pattern cannot alone establish a dose, timing, impairment or sole cause of symptoms. Before forensic use, assess external validation, alternative poisonings and the underlying laboratory evidence rather than treating a promising diagnostic study as a universal exposure rule.
Source: PMC12336345, supplied full-text paper. Findings are paraphrased; forensic implications are an interpretation.
Related: Carbamazepine–Alprazolam Interaction Modeling Included Autoinduction.