Supervised Machine Learning Analysis of Food‐Related Quality of Life Impact in Olfactory Dysfunction

Objective The aim of this study was to assess the effect of olfactory dysfunction (OD) on quality of life (QoL) using supervised machine learning to identify potential key factors. Methods This retrospective, cross‐sectional study included 260 (63.5% women) patients with a mean age of 43.05 (SD = 17.06) years who reported OD of various causes. Olfactory function was assessed using the TDI‐Sniffin' Sticks test (mean TDI scores of 22.51 (SD = 8.48)) and OD‐related QoL was measured using the Questionnaire of Olfactory Dysfunction Negative Statements (QOD‐NS). A random forest model was used to determine the most significant factors for the prediction of overall OD‐related QoL impairment. Results Most influential for the prediction of QoL were food‐related QOD‐NS statements with Statement 7 (i.e., “The changes in my sense of smell annoy me when I eat”) being most influential (percentage increase in mean squared error = %IncMSE 16.02) and Statement 14 (i.e., “Because of the difficulties with smelling, I am scared of getting exposed to certain dangers (e.g., gas, rotten food)”) being least influential (%IncMSE 0.09). We also found significantly better TDI scores ( H (9) = 45.31, p < 0.001) in patients with COVID‐19‐associated OD compared to posttraumatic, congenital, and iatrogenic OD. Conclusion OD‐related QoL seems to be predominantly affected by food‐related difficulties rather than other factors, highlighting the importance of food for patients with OD and the important role of patients' subjective perception of the sense of smell when assessing QoL. Level of Evidence 4.