Carfentanil in the illicit drug supply: A machine learning approach for trace detection using surface-enhanced Raman...

Carfentanil is an extremely potent synthetic fentanyl analogue often present at trace levels alongside other fentanyl analogues and long-acting sedatives, and has become increasingly prevalent in drug supplies across Canada and the United States. Currently, commonly used drug checking technologies cannot reliably identify carfentanil, limiting the effectiveness of drug checking as an overdose prevention response. This study seeks to respond to these limitations using surface-enhanced Raman spectroscopy (SERS) combined with machine learning to reliably detect carfentanil in samples of unregulated drugs submitted to a drug checking service.

SERS was performed using 50 nm gold nanoparticles in solution, with a handheld Raman spectrometer, to analyze illicit drug samples (n=250) obtained through the Substance drug checking project in Victoria, Canada. A random forest classifier was optimized using SERS data (n=200) to establish a model for detecting carfentanil in complex opioid mixtures (n=50), benchmarked against mass spectrometry results.

The SERS-based model successfully predicted 24 of 25 carfentanil-positive samples with concentrations from 0.06-3.93% w/w, below reported detection limits for Fourier-transform infrared and conventional Raman spectroscopy. It demonstrated high sensitivity (96%), specificity (84%), and strong discriminatory power across thresholds (AUC = 0.96).

SERS paired with machine learning demonstrates strong potential for reliable identification of trace carfentanil in complex illicit mixtures, even with co-occurring fentanyl analogues. Given its portability, low cost, and sensitivity, SERS represents a promising complementary approach for community-based drug checking and has important implications for evidence-informed drug policy and harm reduction practice.