Psychometric Evaluation of the Importance of Olfaction Questionnaire Using Item Response Theory

Key Points

• The IOQ measures the value patients place on olfaction, a construct distinct from disease burden.

• The IOQ is most precise across the central trait range typical of routine ENT patients.

• Association items carry most measurement information, enabling a brief affective screen.

1. Introduction

Patient‐reported outcome measures (PROMs) are a cornerstone of modern rhinology, used to monitor disease burden and treatment response. Olfaction is vital for nutrition, hazard detection, and social communication, and its loss impairs quality of life [1]. However, individuals differ in how much importance they place on smell. The Importance of Olfaction Questionnaire (IOQ) was developed to capture this subjective valuation [2]. It quantifies the importance patients place on olfaction in everyday life rather than the burden imposed by its loss, setting it apart from disorder‐specific instruments such as the QOD and its short form [3]. Perceived importance is highest in younger, normosmic individuals and declines as smell loss persists [4]. Crucially, this importance shapes how severely smell loss is experienced, beyond what objective testing captures. Yet how well the IOQ measures importance has never been formally evaluated. Traditional validation relies on Classical Test Theory, which gives limited insight into item‐level performance. Item response theory (IRT) instead identifies which items best separate patients and where the scale measures most precisely [5]. We applied IRT to evaluate the IOQ's measurement structure in olfactory dysfunction.

2. Methods

This psychometric evaluation analyzed IOQ data collected at the Smell and Taste Clinic, Department of Otolaryngology, Head and Neck Surgery, Medical University of Vienna, between December 2020 and January 2025. The study was approved by the local ethics committee (ethics approval number: 2030/2025). The cohort comprised 357 adults (231 female; mean age 44.2 years, range 6–86) spanning the full range of function (16.1% normosmic, 56.2% hyposmic, 27.6% anosmic; TDI 1.0–44.0) and multiple etiologies, most commonly COVID‐19‐associated (49.0%) and idiopathic (19.2%). All participants providing valid responses to at least one IOQ item were included. The IOQ rates 20 statements on the value of smell on a 4‐point scale (0–3). Items span three domains, Association (emotion and memory), Application (everyday use of smell), and Consequence (decisions from smell), plus two Aggravation items intended as a response‐bias filter [2]. Because the Aggravation items showed poor category functioning and discrimination, analyses used the refined 18‐item version. After exploratory factor analysis on polychoric correlations supported a dominant single factor, we fitted a unidimensional graded response model (GRM), estimating item discrimination and thresholds. Item fit used S‐χ² with RMSEA. Analyses ran in R using the mirt package.

3. Results

The final sample comprised 357 adults, with IOQ total scores spanning the full importance range (32.7 ± 12.7). Diagnostic calibration confirmed that the two Aggravation items performed poorly (α = 0.90 and 0.40, with misfit) and were removed, leaving an 18‐item scale (Figure S1).

The 18‐item IOQ was unidimensional: all items loaded moderately to strongly on a single factor (0.46–0.87), with Item 14 (“Certain smells immediately activate strong feelings”) the strongest marker (Table S1). Model fit was good, with 17 of 18 items showing nonsignificant misfit statistics and low error (RMSEA ≤ 0.038; Table S2).

Item discrimination varied by domain (Table 1; Figures S2–S4). The Association items were the sharpest at separating patients, led by Item 14 (α = 3.59) and Item 10 (α = 3.02). Consequence items such as Item 4 and Item 20 discriminated weakly (α  =  0.94). Thresholds covered a broad trait range (β from −1.91 to 2.90), and the scale was most precise across the central range (θ ≈ −1.5 to 1.5), where most patients fall (Figure S2).

TABLE 1.

Item discrimination (α) and threshold (β ₁–β ₃) parameter estimates for the 18‐item IOQ derived from a unidimensional graded response model.

Item	α	β1	β2	β3
Item 1: Social liking	1.22	−1.49	−0.69	0.71
Item 2: Spoiled food	1.88	−1.72	−1.3	−0.52
Item 3: Sniffing food	2.33	−1.57	−0.89	−0.03
Item 4: Museum offer	0.94	−1.17	−0.35	0.89
Item 5: Shampoo purchase	1.69	−1.79	−1.36	−0.71
Item 6: Hunger trigger	1.75	−1.73	−1.12	−0.11
Item 8: Locating odor	1.81	−1.91	−1.27	0.04
Item 9: Odor disturbance	2.04	−1.25	−0.37	0.8
Item 10: Memory activation	3.02	−1.45	−0.99	−0.16
Item 11: Beverage aroma	2.22	−1.26	−0.67	0.05
Item 12: Tomato odor	1.66	−0.8	0.15	0.79
Item 13: Avoiding kissing	1.49	−1.04	−0.34	0.72
Item 14: Strong feelings	3.59	−1.04	−0.54	0.18
Item 15: Clothes hygiene	1.63	−1.53	−0.85	0.07
Item 16: Leaving room	1.57	−1.38	−0.26	1.36
Item 17: Fantasy stimulation	2.51	−1.14	−0.44	0.44
Item 19: Alcohol/smoke	1.07	−0.22	0.87	2.07
Item 20: Candle purchase	0.94	0.2	1.6	2.9

Item information confirmed this concentration (Figure 1; Table S3). Within the Association domain, three items (14, 10, and 17) supplied roughly two‐thirds (66.7%) of the domain measurement information. Information was more evenly distributed across the Application and Consequence domains.

FIGURE 1.

Item information curves for the 18 IOQ items, grouped by domain (Application, Consequence, Association). The latent trait (θ) represents the importance a person places on olfaction, from low to high; item information I(θ) represents how precisely an item measures at each level of that trait.

4. Discussion

In this analysis, IRT confirmed that the 18‐item IOQ is a psychometrically robust, unidimensional measure of olfactory importance. Its single‐trait structure supports interpretation of the total score as one coherent quantity rather than a composite of unrelated constructs. Precision was highest in the central trait range, where most routine ENT patients fall, suiting the scale to longitudinal monitoring. Importantly, a validated instrument is not required to be unidimensional overall. The SNOT‐22 illustrates this: although it is multidimensional overall, it has been examined with IRT by analyzing its dimensions separately [6, 7].

Item‐level analysis showed that the Association items, which capture odor‐evoked emotion and memory, carried the most measurement information and the highest discrimination (α > 3.0 for Items 14 and 10). This reflects the limbic basis of olfaction, which lends odor‐evoked memories their intensity [8]. Such high discrimination may partly reflect conceptual proximity and warrants caution. Clinically, their concentration of information suggests that a brief subset, namely Items 10, 14, and 17, could serve as a rapid screen of a patient's affective investment in olfaction (Figure 1). The two Aggravation items, by contrast, were removed for weak discrimination and poor category functioning, yielding a cleaner and more interpretable total score.

The IOQ retained precision across the full range of function, including in normosmic individuals. Burden‐focused instruments such as the QOD cannot capture this, tied to the severity of dysfunction. Perceived importance can diverge from severity [9], and helps explain why comparable objective deficits produce different subjective burden [10]. Establishing its measurement properties is a prerequisite for use alongside such tools. A direct comparison within a single cohort is warranted. As a single‐center study, generalizability is limited and longitudinal responsiveness is unestablished. Within these limits, the 18‐item IOQ is a reliable, clinically interpretable measure and a basis for a shorter, importance‐focused screen. Beyond abridgement, IRT can also supports adaptive measurement and other clinically targeted applications (see Table S3).

Funding

The authors have nothing to report.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Supporting Figure 1: ICC for the two Aggravation items (Items 7 and 18), conditioned on the latent trait of the 20‐item IOQ. Diagnostic calibration refers to the psychometric quality check carried out before the main analysis: for each item, the model estimates how well the item separates patients along the trait (discrimination) and how closely the observed responses follow the pattern the model predicts (fit). Items that discriminate poorly or misfit the model contribute little information and can be flagged for removal. Here, the two Aggravation items showed weak discrimination (α = 0.90 and 0.40) and were therefore excluded, leaving the 18‐item scale used in all subsequent analyses.

Supporting Figure 2: ICC for the Application domain items.

Supporting Figure 3: ICC for the Consequence domain items.

Supporting Figure 4: ICC for the Association domain items.

Supporting Table 1: Standardized factor loadings for the 18 IOQ items from exploratory factor analysis on polychoric correlations.

Supporting Table 2: Item fit statistics (S–χ² with RMSEA) for the 18 IOQ items.

Supporting Table 3: Total item information and percentage contribution to overall information within the respective subdomain of the IOQ.