Predicting olfactory loss in chronic rhinosinusitis using machine learning

Area of Science:

• Otolaryngology
• Data Science
• Medical Informatics

Background:

• Olfactory dysfunction (OD) is a common complication of chronic rhinosinusitis (CRS).
• Predicting and understanding the risk factors for CRS-OD remains a clinical challenge.

Purpose of the Study:

• To compare machine learning (ML) predictive analytics with logistic regression for classifying CRS-OD.
• To identify predictors of olfactory dysfunction in a large cohort of refractory CRS patients.

Main Methods:

• A prospective, multi-institutional observational study of adult CRS patients.
• Classification of normosmia versus CRS-OD using smell identification tests.
• Comparison of four ML methods against traditional logistic regression.

Main Results:

• 34% of CRS patients exhibited olfactory loss.
• ML methods demonstrated favorable predictive ability compared to logistic regression.
• Identified predictors included objective disease measures, demographics, and socioeconomic factors.

Conclusions:

• ML methods are effective for classifying CRS-OD and incorporating numerous risk factors.
• Identified actionable features can enhance understanding and future study of sinonasal disease-induced hyposmia.
• ML holds promise for studying olfactory loss, the most common cause of persistent smell loss.