SMELL-S olfactory sensitivity and SMELL-R olfactory resolution tests. (A) Schematic of triangle test stimuli for SMELL-S, comprising two glass vials containing solvent (blue) and one containing increasingly diluted mixtures of 30 molecules (red–white mosaic). Olfactory sensitivity of a subject measured with SMELL-S [Subject Expt 1-A023, SMELL-S (v2)]. (B) Schematic of triangle test stimuli for SMELL-R, comprising two jars containing the same mixture of 30 molecules (red–white mosaic) and one containing mixtures of 30 molecules (black–gray mosaic) with an increasing number of molecules shared with the other two. Olfactory resolution of a subject measured with SMELL-R [Subject Expt 1-A016, SMELL-R (v1)]. Circles in A and B indicate starting level for each test.

Test–retest reliability and relationship between SMELL-S and SMELL-R tests. (A) Experiment 1 design, showing one example of the many different presentations of the six smell tests. SMELL-R tests were always administered after SMELL-S or threshold tests in a given visit. (B) Bland–Altman plots of the indicated tests where each dot represents data from one subject. Black dotted lines represent the average of differences between retest and test scores, and gray areas indicate 95% limits of agreement (average difference ±1.96 SD of the difference, n = 74–75). (C) Test and retest scores for SMELL-R (v1) and SMELL-R (v2) where each dot represents data from one subject (ICC, intraclass correlation coefficient; n = 73–75).

Addressing the problem of odor-specific insensitivity. (A) Subject average scores of the indicated tests from experiment 1 between test and retest. Data marked with different letters (a or b) indicate significantly different interindividual variance between groups (P < 0.0001; Conover squared ranks test, n = 74–75). Data from the seven lowest scoring subjects in the phenylethyl alcohol test are connected by lines. (B) Experiment 2 design. (C) The relationship between different etiologies of olfactory dysfunction and UPSIT scores in published studies, as well as of experiment 2 subjects divided by self-reported smell abilities (mean ± 95% confidence interval). References: Korsakoff Syndrome (), traumatic brain injury (), sinonasal disease (), Parkinson’s disease (), postviral olfactory dysfunction (), amnestic mild cognitive impairment (), elderly at higher risk of cognitive decline (), nonamnestic mild cognitive impaired (), and toxic exposure (). (D) UPSIT scores for subjects divided into normal and dysfunctional groups according to phenylethyl alcohol threshold (Sniffin’ Sticks) (cutoff score, 6.5) performance, where each dot represents data from one subject. Subjects scored as normal by the UPSIT but dysfunctional by Sniffin’ Sticks phenylethyl alcohol test are colored in red in D and E. Data from the remaining subjects are colored brown here and in E. Data labeled with different letters (a or b) are significantly different (P = 0.0003, Mann–Whitney test). Medians and interquartile range are represented. (E) Comparison of UPSIT and phenylethyl alcohol (Sniffin’ Sticks) scores. (F) UPSIT scores for subjects divided into normal and dysfunctional groups according SMELL-S (v2) (cutoff score, 7) performance, where each dot represents data from one subject. Subjects scored as normal by the UPSIT but dysfunctional by Sniffin’ Sticks phenylethyl alcohol test in D are colored red in F and G, while the others are in blue. Data labeled with different letters (a or b) are significantly different (P < 0.0001; Mann–Whitney test) and represented as median and interquartile range. (G) Comparison between UPSIT and SMELL-S (v2) scores. Subjects with identical values in E and G are indicated by superimposed open circles and an X and retain the specified color coding.

SMELL-S and SMELL-R diagnostic accuracy. (A) Comparison of UPSIT and SMELL-S (v2) scores for experiment 2 subjects (mean ± SD). Subjects were divided into dysfunctional (n = 7) and normal (n = 26) using an UPSIT cutoff score of 29. Data labeled with different letters (a or b) are significantly different (P = 0.0005, two-sided unpaired t test with Welch’s correction). (B) Area under the ROC curve (AUC) for SMELL-S (v2). The optimal cutoff is indicated by the blue square. (C) Plots of four measures of diagnostic accuracy resulting from different cutoff scores for SMELL-S (v2) (percentage ± 95% confidence interval). The optimal cutoff score for olfactory dysfunction defined by Youden’s Index (Center) is indicated by the blue dot. (D) Comparison of UPSIT and SMELL-R (v2) scores for experiment 2 subjects (mean ± SD). Subjects were divided into dysfunctional (n = 7) and normal (n = 26) using an UPSIT cutoff score of 29. Data labeled with different letters (a or b) are significantly different (P = 0.0035, two-sided unpaired t test with Welch’s correction). Subject Expt 2-A006 had a self-reported normal sense of smell but low scores on the UPSIT as well as SMELL-S (v2) in A and SMELL-R (v2) in D. (E) Area under the ROC curve (AUC) for SMELL-R (v2). The optimal cutoff is indicated by the green square. (F) Plots of four measures of diagnostic accuracy resulting from different cutoff scores for SMELL-R (v2) (rate or percentage ± 95% confidence interval). The optimal cutoff score for olfactory dysfunction defined by Youden’s Index (Center) is indicated by the green dot. Subjects with identical values in A and D are indicated by superimposed open circles and an X and retain the specified color coding.

Addressing the problem of smell testing in different populations. (A) Experiment 3 design. (B) Performance of Taiwanese (n = 36) (Top) and North American (n = 36) (Bottom) subjects for individual UPSIT items; open bar bars indicate correct answers, and solid bars indicate the most frequent incorrect answer. Data are sorted by the size of the difference between the two populations. (C and D) Histogram of North American and Taiwanese subject scores for the UPSIT (C) and SMELL-R (v2) (D). (E) Cross-population comparison of UPSIT and SMELL-R (v2) (mean ± 95% confidence interval) for subjects in C and D.