Multivariate Data Analysis for Drug Identification Using Energy-Dispersive X-Ray Diffraction

Preliminary studies have shown the effectiveness of multivariate analysis (MVA) for drug identification from energy-dispersive X-ray diffraction patterns. A statistical model to predict drug content from the diffraction profile of a sample of mixed composition was developed by applying MVA to both experimental and simulated data. Separate data-sets were used for building and testing the models. Both experimental and simulated data were used and the MVA predictions compared. Experimental data included diffraction patterns from small (5 mm diameter) drug samples with various cutting agents, acquired with a HPGe detector; simulated data included diffraction patterns of samples including materials simulating drugs (i.e., materials featuring sharp diffraction peaks in the relevant momentum transfer range) and typical packaging materials. Both a HPGe detector (energy resolution 0.7 keV at 59.5 keV) and a CZT detector (energy resolution 4 keV at all energies) were simulated. MVA was used to predict the drug content. In all cases different statistics were applied to assess the detection limits of the models. Multivariate analysis has proved effective in both identifying the presence of a drug and its concentration. Due to the large contribution to peak broadening given by angular resolution, no significant decrease in accuracy has been found when using CZT with respect to HPGe data.