Limits...
Olfactory thresholds of the U.S. Population of home-dwelling older adults: development and validation of a short, reliable measure.

Kern DW, Schumm LP, Wroblewski KE, Pinto JM, Hummel T, McClintock MK - PLoS ONE (2015)

Bottom Line: Simulated subjects were assigned n-butanol thresholds drawn from the estimated normosmic distribution and based on these and the model, we simulated administration of both the staircase and constant stimuli methods.Our results replicate both the correlation between the two methods and their reliability as previously reported by studies using human subjects.Thus, testing with a fast, 6-item constant stimuli protocol is informative, and permits olfactory testing in previously inaccessible research settings.

View Article: PubMed Central - PubMed

Affiliation: Institute for Mind and Biology and Department of Comparative Human Development, The University of Chicago, Chicago, IL, United States of America.

ABSTRACT
Current methods of olfactory sensitivity testing are logistically challenging and therefore infeasible for use in in-home surveys and other field settings. We developed a fast, easy and reliable method of assessing olfactory thresholds, and used it in the first study of olfactory sensitivity in a nationally representative sample of U.S. home-dwelling older adults. We validated our method via computer simulation together with a model estimated from 590 normosmics. Simulated subjects were assigned n-butanol thresholds drawn from the estimated normosmic distribution and based on these and the model, we simulated administration of both the staircase and constant stimuli methods. Our results replicate both the correlation between the two methods and their reliability as previously reported by studies using human subjects. Further simulations evaluated the reliability of different constant stimuli protocols, varying both the range of dilutions and number of stimuli (6-16). Six appropriately chosen dilutions were sufficient for good reliability (0.67) in normosmic subjects. Finally, we applied our method to design a 5-minute, in-home assessment of older adults (National Social Life, Health and Aging Project, or NSHAP), which had comparable reliability (0.56), despite many subjects having estimated thresholds above the strongest dilution. Thus, testing with a fast, 6-item constant stimuli protocol is informative, and permits olfactory testing in previously inaccessible research settings.

No MeSH data available.


Related in: MedlinePlus

Scatter and Bland-Altman Plots for Normosmics.A. Scatter plot (plus regression line) of the relationship between the model-estimated thresholds and the true thresholds for a single replication of the 16-dilution constant stimuli design, with thresholds drawn from 590 normosmic distribution (n = 100). B. Corresponding Bland-Altman plot with LOWESS smoother of the 16-dilution constant stimuli design, with thresholds drawn from 590 normosmic distribution (n = 100). C. Scatter plot (plus regression line) of the relationship between the model-estimated thresholds and the true thresholds for a single replication of the 6-dilution constant stimuli design, with thresholds drawn from 590 normosmic distribution (n = 100). The 6 dilutions are evenly distributed across the range of possible thresholds and the administered dilution steps are noted on the lower x-axis. D. Corresponding Bland-Altman plot with LOWESS smoother of the 6-dilution (evenly distributed) constant stimuli design, with thresholds drawn from 590 normosmic distribution (n = 100).
© Copyright Policy
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC4359123&req=5

pone.0118589.g002: Scatter and Bland-Altman Plots for Normosmics.A. Scatter plot (plus regression line) of the relationship between the model-estimated thresholds and the true thresholds for a single replication of the 16-dilution constant stimuli design, with thresholds drawn from 590 normosmic distribution (n = 100). B. Corresponding Bland-Altman plot with LOWESS smoother of the 16-dilution constant stimuli design, with thresholds drawn from 590 normosmic distribution (n = 100). C. Scatter plot (plus regression line) of the relationship between the model-estimated thresholds and the true thresholds for a single replication of the 6-dilution constant stimuli design, with thresholds drawn from 590 normosmic distribution (n = 100). The 6 dilutions are evenly distributed across the range of possible thresholds and the administered dilution steps are noted on the lower x-axis. D. Corresponding Bland-Altman plot with LOWESS smoother of the 6-dilution (evenly distributed) constant stimuli design, with thresholds drawn from 590 normosmic distribution (n = 100).

Mentions: A plot showing the relationship between the model-estimated threshold and the true threshold for a single replication (n = 100) of both the full 16-dilution design and the 6-dilution design (with dilutions evenly distributed over the range of dilutions) is shown in Fig. 2 (Panels A and C). Corresponding Bland-Altman plots (model-estimated threshold minus true threshold plotted versus true threshold) depicting the difference between the model-estimated and true threshold together with a LOWESS smoother are also shown in Fig. 2 (Panels B and D). The correlation between the model-estimated and true thresholds for the 6-dilution design is 0.78 for this replication, only slightly below the average of 0.81 across all replications. The biggest difference between the two designs is the conditional bias (i.e., conditional on the true threshold) for the 6-dilution design, which increases toward the tails of the distribution. This conditional bias is a property of the empirical Bayes method we have used to estimate the individual thresholds, and is due to the individual estimates being shrunk toward the overall mean. The amount of shrinkage decreases as the reliability of the estimates increases; it is, for example, substantially reduced if each of the 6 dilutions is presented twice. Despite this, the empirical Bayes estimates are unconditionally unbiased, thereby permitting estimation of the mean threshold for the population [27].


Olfactory thresholds of the U.S. Population of home-dwelling older adults: development and validation of a short, reliable measure.

Kern DW, Schumm LP, Wroblewski KE, Pinto JM, Hummel T, McClintock MK - PLoS ONE (2015)

Scatter and Bland-Altman Plots for Normosmics.A. Scatter plot (plus regression line) of the relationship between the model-estimated thresholds and the true thresholds for a single replication of the 16-dilution constant stimuli design, with thresholds drawn from 590 normosmic distribution (n = 100). B. Corresponding Bland-Altman plot with LOWESS smoother of the 16-dilution constant stimuli design, with thresholds drawn from 590 normosmic distribution (n = 100). C. Scatter plot (plus regression line) of the relationship between the model-estimated thresholds and the true thresholds for a single replication of the 6-dilution constant stimuli design, with thresholds drawn from 590 normosmic distribution (n = 100). The 6 dilutions are evenly distributed across the range of possible thresholds and the administered dilution steps are noted on the lower x-axis. D. Corresponding Bland-Altman plot with LOWESS smoother of the 6-dilution (evenly distributed) constant stimuli design, with thresholds drawn from 590 normosmic distribution (n = 100).
© Copyright Policy
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC4359123&req=5

pone.0118589.g002: Scatter and Bland-Altman Plots for Normosmics.A. Scatter plot (plus regression line) of the relationship between the model-estimated thresholds and the true thresholds for a single replication of the 16-dilution constant stimuli design, with thresholds drawn from 590 normosmic distribution (n = 100). B. Corresponding Bland-Altman plot with LOWESS smoother of the 16-dilution constant stimuli design, with thresholds drawn from 590 normosmic distribution (n = 100). C. Scatter plot (plus regression line) of the relationship between the model-estimated thresholds and the true thresholds for a single replication of the 6-dilution constant stimuli design, with thresholds drawn from 590 normosmic distribution (n = 100). The 6 dilutions are evenly distributed across the range of possible thresholds and the administered dilution steps are noted on the lower x-axis. D. Corresponding Bland-Altman plot with LOWESS smoother of the 6-dilution (evenly distributed) constant stimuli design, with thresholds drawn from 590 normosmic distribution (n = 100).
Mentions: A plot showing the relationship between the model-estimated threshold and the true threshold for a single replication (n = 100) of both the full 16-dilution design and the 6-dilution design (with dilutions evenly distributed over the range of dilutions) is shown in Fig. 2 (Panels A and C). Corresponding Bland-Altman plots (model-estimated threshold minus true threshold plotted versus true threshold) depicting the difference between the model-estimated and true threshold together with a LOWESS smoother are also shown in Fig. 2 (Panels B and D). The correlation between the model-estimated and true thresholds for the 6-dilution design is 0.78 for this replication, only slightly below the average of 0.81 across all replications. The biggest difference between the two designs is the conditional bias (i.e., conditional on the true threshold) for the 6-dilution design, which increases toward the tails of the distribution. This conditional bias is a property of the empirical Bayes method we have used to estimate the individual thresholds, and is due to the individual estimates being shrunk toward the overall mean. The amount of shrinkage decreases as the reliability of the estimates increases; it is, for example, substantially reduced if each of the 6 dilutions is presented twice. Despite this, the empirical Bayes estimates are unconditionally unbiased, thereby permitting estimation of the mean threshold for the population [27].

Bottom Line: Simulated subjects were assigned n-butanol thresholds drawn from the estimated normosmic distribution and based on these and the model, we simulated administration of both the staircase and constant stimuli methods.Our results replicate both the correlation between the two methods and their reliability as previously reported by studies using human subjects.Thus, testing with a fast, 6-item constant stimuli protocol is informative, and permits olfactory testing in previously inaccessible research settings.

View Article: PubMed Central - PubMed

Affiliation: Institute for Mind and Biology and Department of Comparative Human Development, The University of Chicago, Chicago, IL, United States of America.

ABSTRACT
Current methods of olfactory sensitivity testing are logistically challenging and therefore infeasible for use in in-home surveys and other field settings. We developed a fast, easy and reliable method of assessing olfactory thresholds, and used it in the first study of olfactory sensitivity in a nationally representative sample of U.S. home-dwelling older adults. We validated our method via computer simulation together with a model estimated from 590 normosmics. Simulated subjects were assigned n-butanol thresholds drawn from the estimated normosmic distribution and based on these and the model, we simulated administration of both the staircase and constant stimuli methods. Our results replicate both the correlation between the two methods and their reliability as previously reported by studies using human subjects. Further simulations evaluated the reliability of different constant stimuli protocols, varying both the range of dilutions and number of stimuli (6-16). Six appropriately chosen dilutions were sufficient for good reliability (0.67) in normosmic subjects. Finally, we applied our method to design a 5-minute, in-home assessment of older adults (National Social Life, Health and Aging Project, or NSHAP), which had comparable reliability (0.56), despite many subjects having estimated thresholds above the strongest dilution. Thus, testing with a fast, 6-item constant stimuli protocol is informative, and permits olfactory testing in previously inaccessible research settings.

No MeSH data available.


Related in: MedlinePlus