Enhancing safety for infusion treatment in oncology through a data-driven multisensory approach based on optical methods

Incorrect medication poses a significant risk to patient safety. Cancer patients are particularly vulnerable to healthcare errors, given the complexity and toxicity of their treatments and their frequent personal interactions with multiple providers across various care settings. Effective prevention could be achieved by early detection and management of incorrect medication since most errors occur during administration. Our approach combines three different analytical optical techniques suitable for point-of-care diagnosis: Raman spectroscopy, UV/Vis spectroscopy, and refractometry. In collaboration with the Department of Pharmacy of the University Medical Center in Mainz, representative cytostatics administered in oncology wards were selected and spectroscopically investigated regarding their qualitative and quantitative differences. The experimental data was analyzed using classical methods and applied machine learning algorithms. Here, we compare the effect of four data analysis methods on the recognition rates of the investigated cytostatics: linear models, ensemble methods, decision trees, and established neural networks.