Development of parallel forms of a brief smell identification test useful for longitudinal testing

Area of Science:

• Olfactory Neuroscience
• Machine Learning in Healthcare
• Psychometrics

Background:

• Quantifying olfactory function is crucial, yet existing brief smell identification tests lack parallel forms for longitudinal assessment.
• Knowledge of prior test items can confound longitudinal olfactory testing.
• Optimal test lengths for smell identification have not been empirically established.

Purpose of the Study:

• To develop and validate a set of four brief, parallel smell identification tests.
• To identify the optimal number of odorant items for accurate smell loss differentiation.
• To enable longitudinal olfactory assessment without practice effects.

Main Methods:

• Utilized eight machine learning algorithms to analyze items from the University of Pennsylvania Smell Identification Test (UPSIT).
• Selected items to create four parallel eight-item smell tests differentiating COVID-19 patients from healthy controls.
• Validated test sensitivity in an independent cohort using sequential testing.

Main Results:

• Linear Discriminant Analysis (LDA) demonstrated superior performance among the tested algorithms.
• An eight-item odorant test length was identified as optimal for accurate smell loss differentiation.
• The developed parallel tests achieved high diagnostic accuracy (AUC 0.79-0.83) in distinguishing olfactory dysfunction.

Conclusions:

• The newly developed brief, parallel smell tests are suitable for longitudinal olfactory assessment.
• These tests effectively differentiate olfactory compromise in various conditions, including COVID-19.
• The parallel forms minimize confounding effects, allowing for reliable repeated administration.