Psychological Validation of a Standardized Odor Set for Olfactory Training Free

Currently, only a few long-term, effective treatments for olfactory disorders exist [1]. However, common interventions with the potential to recover function include systemic steroids, topical therapy, nonsteroidal oral medications, olfactory training, and acupuncture [2]. Olfactory training improves olfactory function at both peripheral and central levels, which is reflected in structural and functional changes in brain regions that process olfactory information [3]. Its protocol was developed in Germany [4], spread to Europe and eventually the world [3].

To increase the concentration and effectiveness of olfactory training, a common practice is to instruct patients to sniff an odor while recalling its quality; this allows patients to rely on memories before they lose their olfactory ability [5]. The more familiar the odor, the more vivid the image of odor quality [6]. The concentration of odors is also considered to be an important factor affecting the effectiveness of training [7]. One previous study used “intense odors” for olfactory training [4], and another previous study reported that more patients saw improved olfactory function when training with a high-concentration odor set than with a low-concentration one [8]. It should be noted, however, that low-concentrated odors accelerated the recovery of olfactory function than high-concentrated odors in the pediatric population [9], and that the improvement in olfactory function was not concentration-dependent [5]. In addition, some researchers emphasize pleasantness when selecting odors for olfactory training [10‒12]. Reasons for this include enhanced compliance [13], which is defined as the degree to which a patient’s behavior follows the physician’s therapeutic recommendations [14], and breathing patterns in which a more unpleasant odor decreases the amount and duration of sniffing [15, 16].

Using the same odors for every patient (henceforth referred to as a “standardized odor set”) is a common treatment in olfactory training [4], but its psychological appropriateness is open to investigation because of individual differences in psychological responses to odors [17‒19]. This study constructed two models to psychologically validate the appropriateness of a standardized odor set for olfactory training. First, standardization of odor set is desirable for commercial dissemination of olfactory training kits, so a model incorporating three psychological dimensions (intensity, familiarity, and pleasantness) with a proven track record was constructed using mean evaluation values rather than individual ones (henceforth referred to as the “average model”). Second, given individual differences in psychological responses to odors, it is theoretically possible to prescribe a different odor set for each patient; therefore, a model was constructed for each person (hereafter referred to as the “individual model”). In recent years, the influence of cultural familiarity on olfactory perception has been questioned [19‒22]. However, considering previous studies that reflected cultural familiarity in odor selection for olfactory training [23, 24], Japanese participants evaluated the psychological dimensions of their everyday odors. Principal component analysis was conducted for both the average and individual models, with the corresponding psychological dimensions as observed variables, to rank the odors using a composite indicator. A strong correlation in odor rankings between the two models suggests that the average model reflects individual odor perception, supporting the psychological appropriateness of the standardized odor set.

This study was conducted in accordance with the revised version of the Declaration of Helsinki. All procedures were approved by the appropriate ethics review committees for research involving human subjects of Ritsumeikan University (Approval No.: Kinugasa-human-2019-53) and Sony Group Corporation (Approval No.: 19-F-0031). The experimental approach was explained to the participants both during recruitment and at the start of the experiment. All participants signed an informed consent statement prior to participation in the study.

Participants were recruited using a research firm’s registration database and screened in two steps. The first screening step included three sub-steps: (1) access the URL link provided in the invitation email; (2) review an outline of this experiment and the recruitment requirements; (3) complete a questionnaire on demographic characteristics and olfactory ability.

Based on their responses, the research firm identified volunteers who satisfied the following criteria: (1) 20–64 years old; (2) non-smokers or those who quit more than 6 months ago; (3) no experience in business (marketing, development, or quality control) related to odors such as foods and cosmetics; (4) not consciously experiencing an olfactory disorder; (5) no serious nasal or sinus disease now or in the past; (6) not suffering from multiple chemical sensitivities; (7) not uncomfortable with sniffing or smelling odors; (8) can use a keyboard of a personal computer without stress; and (9) scored 70% or more in the self-administered odor questionnaire [25]. The questionnaire, which lists 20 odor names, was devised by the Japan Rhinologic Society committee on olfaction tests to provide a simple and subjective evaluation of olfactory ability based on odor perception.

In the second screening step, the research firm narrowed the potential participants to mitigate age and gender biases. Specifically, the research firm called volunteers who passed the initial screening and inquired about their willingness to participate in this experiment. Finally, the research firm selected volunteers who could participate on the specified date and time. The day before the experiment, each potential participant received a telephone call asking about their physical condition (i.e., whether they could detect odors sufficiently). Ultimately, 173 healthy volunteers aged 20–64 (88 women and 85 men, mean age ± standard deviation [SD] = 42.50 ± 12.48 years) participated.

From the 118 available everyday Japanese odors (Takasago International, Tokyo, Japan), we used 35 odors that satisfied the following criteria: (1) not remarkably unpleasant, (2) not an alcoholic beverage odor, (3) not a potentially allergenic odor, (4) matching the odor name and odor quality, and (5) exhibiting as distinctive odor quality as possible (see online suppl. Table S1 for details; for all online suppl. material, see https://doi.org/10.1159/000545368). All odorants were diluted to 10% (volume/volume, v/v) using propylene glycol (product code 166-07256; Fujifilm Wako Pure Chemical Industries, Osaka, Japan) and adjusted to an easily detectable intensity.

The flow of odor evaluation for each trial is shown in Figure 1. After the first sniff, intensity was evaluated using a scale from “not detectable: 0” to “very strongly: 1.” Only when the odor intensity was 0.2 or higher, familiarity and pleasantness evaluated after the second sniff. This sequence was repeated 3 and 32 times in the practice and test sessions, respectively. The presentation order was fixed (i.e., orange followed by milk and then grilled fish) for simplicity of instruction in the practice session, but was randomized among participants in the test session. A commercially available diffuser (AROMASTIC; model number OE-AS01(W); Sony, Tokyo, Japan) was used to present the odor.

Table 1 shows the amount of data used for analysis. Evaluation values for the three psychological dimensions (intensity, familiarity, and pleasantness) were acquired in 5,332 trials (acquisition rate 96.3%) of 5,536 trials (32 trials × 173 participants). The mean number of data per participant was 30.82 ± 1.90 trials (mean ± SD; range 16–32 trials). The evaluation value was determined by considering the left edge of the scale as 0 and the right edge as 1.

A composite indicator (hereafter referred to as the “good smell index”; GSI) was developed to rank the 32 odors. To identify a principal component for the calculation of GSI for each odor, we conducted a principal component analysis using the variance-covariance matrix of the mean evaluation values (Table 1), with three psychological dimensions as the observed variables.

This study adopted two decision criteria (i.e., scree plot and cumulative contribution ratio) to comprehensively determine the number of principal components that should be retained. The scree plot shows principal component numbers on the x-axis and eigenvalues on the y-axis. In most cases, the principal components prior to the point where the line connecting the plots forms the “elbow” (i.e., the point at the “elbow” is excluded) are retained [26]. The cumulative contribution ratio is an empirical criterion using a predetermined cutoff value. Often the cutoff is set to 70% [27].

In principal component analysis, loadings are calculated for each principal component and observed variable. The loadings ranged from −1 to 1 [28], indicating what the principal components represent [29]. Specifically, a positive (or negative) loading indicates that the observed variable is positively (or negatively) correlated with the principal component; a larger loading implies a stronger effect of the observed variable on the principal component [30]. We defined GSI as the score of the principal component that satisfies the following two conditions. (1) Psychological dimensions other than intensity have positive loadings because the higher the evaluation value, the more appropriate the odor is for olfactory training; the sign of the intensity loading was ignored because a stronger odor does not necessarily mean a good smell. (2) If multiple principal components satisfy the first condition, the principal component with the highest contribution ratio is adopted.

The analysis method was the same as that for the average model, except that principal component analysis was conducted for each participant using the individual evaluation values of the three psychological dimensions for each odor (see database in online suppl. Table S2).

For odors with an intensity below 0.2, GSI was not calculated in each individual model because familiarity and pleasantness were evaluated. For convenience, such odors were placed under the odor with the lowest GSI. For example, if one odor did not have a calculated GSI in an individual model, it was ranked 32nd.

To examine whether the average model reflects odor perception of each participant, odor rankings were compared between the average and individual models. We computed the mean ranking of each odor in the individual models, calculated Spearman’s rank correlation coefficient for odor rankings between the average and individual models, and performed test of no correlation. We used BellCurve for Excel 4.08 (Social Survey Research Information, Tokyo, Japan) for statistical analysis with the significance level set to 5%.

Table 2 shows the eigenvalues, contribution ratios, and cumulative contribution ratios in the three principal components extracted by principal component analysis. The first principal component was retained in the overall judgment using the two decision criteria. Table 3 shows the eigenvectors and loadings of the first principal component. Because the first principal component showed positive loadings for two observed variables (familiarity and pleasantness), we defined the score of the first principal component as GSI. Table 1 shows GSI and ranking of each odor.

Currently, a multicenter clinical study is being conducted in Japan to establish an olfactory training method for Japanese patients, using two odor sets [31, 32]. The first odor set is the pioneering one proposed by Hummel and colleagues [4] and includes four components selected on the basis of an odor prism with six corners (i.e., categories) [33]: rose (phenyl ethyl alcohol, representing flowery odor), eucalyptus (eucalyptol, resinous), lemon (citronellal, fruity), and clove (eugenol, spicy). The second odor set is a new one for Japanese patients: coconut (γ-octalactone, flowery), poultice (methyl salicylate, resinous), pineapple (ethyl caproate, fruity), and vanilla (ethyl vanillin, spicy). Although the present study did not use coconut odor, the GSIs of spearmint (a cool, poultice-like odor), pineapple, and vanilla were all in the top half of the 32 odors, ranking 3rd, 11th, and 14th, respectively. The GSIs of lemon and rose in the pioneering odor set also ranked 4th and 6th, suggesting that they are suitable for application to Japanese patients.

The GSI in the average model revealed that grapefruit, strawberry, spearmint, and lemon were the top four odors. Three of these odors belong to the fruit category: besides, grapefruit and lemon are citrus fruits. This odor set does not follow the criteria of the pioneering odor set by Hummel and colleagues [4] who selected components from different categories. Therefore, based on the odor ranking using GSI in the average model and the odor prism, four candidate components are listed: grapefruit (ranked 1st, fruity), spearmint (3rd, resinous), peanut butter (5th, burnt), and rose (6th, flowery) (see Castro and colleagues [34] for odor classification). Previous studies examined the number of odors used in olfactory training [10, 35‒37], but the effect of category bias on treatment efficacy remains an issue for future research.

Online supplementary Table S3a–c shows the eigenvalues, contribution ratios, and cumulative contribution ratios for the three principal components extracted by principal component analysis. The overall judgment using the two decision criteria retained the first principal component for 82 participants and the first and second principal components for 91 participants. For 171 participants, the first principal component had the highest contribution ratio and positive loadings on two observed variables (familiarity and pleasantness). For the remaining two participants, the first principal component had a negative loading for pleasantness only, while the second principal component showed positive loadings for familiarity and pleasantness. Therefore, we defined scores of the identified principal component as GSI for each participant. Online supplementary Table S3d, e show the eigenvectors and loadings of the principal component used to calculate the GSI for each participant. In addition, online supplementary Table S3f, g list GSIs and rankings calculated by participant and odor.

Figure 2 shows the relationship between ranking in the average model and mean ranking in the individual models for each odor (see also Table 1). Odor rankings were significantly and strongly correlated between the average and individual models (ρ = 0.98; t(30) = 25.25, p < 0.001). The result suggests that the average model reflects the odor perception of each participant, supporting the psychological appropriateness of the standardized odor set.

Olfactory training is a medical treatment that is primarily intended for patients with olfactory disorders. Therefore, we note as a current limitation that we have validated average and individual models based on odor evaluations by healthy participants. For example, the reduced sensitivity in patients with olfactory disorder (i.e., hyposmia/anosmia, parosmia, and phantosmia) may depend on the odor molecules rather than occurring to the same degree for all odor molecules [38]. Trigger molecules have also been identified that are likely to cause symptoms of parosmia [39]. In addition, a database on the interaction between odorant molecules and olfactory receptors is also provided [40]. These previous studies suggest that the selection of odors for olfactory training should focus not only on psychological dimensions but also on chemical properties. The development of more effective rehabilitation tools (e.g., selection based on GSI only vs. selection based on GSI and category) may be achieved by considering different properties, such as molecular structure, molecular weight, and olfactory receptor responsiveness.

The psychological appropriateness of using the same odors for every patient in olfactory training was supported by the significantly strong correlation between the average and individual models for odor ranking. We need to note, however, that the present study was conducted on healthy participants, not on those with olfactory disorder. Ranking odors using GSI in the average model may also be useful in selecting new candidate components for a standardized odor set, but should be considered a tool to reinforce findings in clinical studies.

This study was conducted in accordance with the revised version of the Declaration of Helsinki. All procedures were approved by the appropriate ethics review committees for research involving human subjects of Ritsumeikan University (Approval No.: Kinugasa-human-2019-53) and Sony Group Corporation (Approval No.: 19-F-0031). Written informed consent was obtained from participants to participate in this study.

The authors have no conflicts of interest to declare.

This research was partly supported by the Japan Science and Technology Agency (JST-Mirai Program Grant No. JPMJMI21D1) and the Japan Society for the Promotion of Science (JSPS KAKENHI Grant No. 23H00079). The funder had no role in the design, data collection, data analysis, and reporting of this study.

Naomi Gotow: data curation, formal analysis, investigation, validation, visualization, writing – original draft, and writing – review and editing. Takanobu Omata: conceptualization, investigation, methodology, project administration, resources, and supervision. Satomi Kunieda: conceptualization, investigation, methodology, and resources. Tatsu Kobayakawa: conceptualization, funding acquisition, investigation, methodology, software, supervision, validation, and writing – review and editing.

The mean evaluation values for the three psychological dimensions (intensity, familiarity, and pleasantness) for 32 odors are listed in Table 1; each participant’s evaluation values are provided as a database within the online supplementary information file (i.e., online suppl. Table S2).