SCENTinel 1.0: Development of a Rapid Test to Screen for Smell Loss

Abstract

Commercially available smell tests are primarily used in research or in-depth clinical evaluations and are too costly and time-consuming for population surveillance in health emergencies like COVID-19. To address this need, we developed the SCENTinel 1.0 test, which rapidly evaluates 3 olfactory functions: detection, intensity, and identification. We tested whether self-administering the SCENTinel 1.0 test discriminates between individuals with self-reported smell loss and those with average smell ability (normosmic individuals) and provides performance comparable to the validated and standardized NIH Toolbox Odor Identification Test in normosmic individuals. Using Bayesian linear models and prognostic classification algorithms, we compared the SCENTinel 1.0 performance of a group of self-reported anosmic individuals (N = 111, 47 ± 13 years old, F = 71%) and normosmic individuals (N = 154, 47 ± 14 years old, F = 74%) as well as individuals reporting other smell disorders (such as hyposmia or parosmia; N = 42, 55 ± 10 years old, F = 67%). Ninety-four percent of normosmic individuals met our SCENTinel 1.0 accuracy criteria compared with only 10% of anosmic individuals and 64% of individuals with other smell disorders. Overall performance on SCENTinel 1.0 predicted belonging to the normosmic group better than identification or detection alone (vs. anosmic: AUC = 0.95, specificity = 0.94). Odor intensity provided the best single-feature predictor to classify normosmic individuals. Among normosmic individuals, 92% met the accuracy criteria at both SCENTinel 1.0 and the NIH Toolbox Odor Identification Test. SCENTinel 1.0 is a practical test able to discriminate individuals with smell loss and will likely be useful in many clinical situations, including COVID-19 symptom screening.

Keywords: anosmia; hyposmia; prediction; smell loss; smell test.

Figure 1.

Sample description by group. Other: participants who self-reported other smell disorders. NIH: Normosmic individuals who completed the NIH Toolbox Odor Identification Test.

Figure 2.

SCENTinel 1.0 response patterns by smell group (anosmic individuals = red; other smell disorders = blue; normosmic individuals = green). Response patterns 1, 3, 5, and 7 met the SCENTinel accuracy criteria.

Figure 3.

Receiver operating characteristic (ROC) curves and statistics on SCENTinel 1.0 scores overall and for single subtests across groups: (A) anosmic individuals versus normosmic individuals; (B) other smell disorders versus normosmic individuals; (C) anosmic individuals versus other smell disorders) based on the linear discriminant analysis algorithm. AUC, area under the curve; P, P value; D, DeLong’s test for 2 ROC curves; df, degrees of freedom.

Figure 4.

Concordance between SCENTinel 1.0 and the NIH Toolbox Odor Identification Test in normosmic individuals: (A) concordance based on flower odor identification performance; (B) concordance based on full completion of both smell tests.