The current study was a retrospective, single-center cross-sectional data analysis including healthy controls and patients who presented to the Department of Neurology of the Medical University of Vienna between March 2000 and November 2023, either through referral by a physician or self-referral. All patients underwent a neurological examination, blood testing, neuroimaging, and neuropsychological assessment. Corresponding patient data were obtained from the Allgemeines Krankenhaus Information Management (AKIM) database and the Research Documentation and Analysis (RDA) system. The study was carried out with approval from the ethics committee of the Medical University of Vienna.

All study participants underwent both objective olfactory testing using the SS-OIT and subjective assessment using the ASOF inventory. In addition, an extensive neuropsychological assessment was performed, including the Mini-Mental State Examination (MMSE), the Vienna Visuoconstructive and Visual Memory Test (VVVMT), the Wortschatztest (WST-IQ), the Beck Depression Inventory-II (BDI-II), and the Neuropsychological Test Battery Vienna (NTBV). Sociodemographic data including age, sex, and years of formal education were also collected. Exclusion criteria comprised a history of severe brain injuries, radiological or clinically confirmed stroke, psychiatric diagnoses except (sub)depressive symptoms, any form of dementia other than AD, and medical conditions likely to compromise cognition (e.g., cardiac, pulmonary, renal, or hepatic diseases). Furthermore, only participants aged 50 years or older were included in the study. In total, data from 958 participants (56.8% female, excluding HC) were analyzed. The patient cohort comprised three diagnostic subgroups—SCD, MCI, and AD—with participants classified according to neurological and neuropsychological assessments as well as predefined criteria. Individuals with SCD were identified according to the diagnostic criteria by Jessen et al., requiring subjectively perceived, persistent cognitive decline, despite intact neuropsychological test results, adjusted for age, gender, and education [6]. The diagnosis of MCI was based on Petersen’s 2004 criteria, including subjective and objectively confirmed memory impairment, preserved general cognition, daily functioning, and the absence of dementia [7]. Patients with AD were diagnosed according to the NINCDS-ADRDA guidelines [23]. Using these criteria, the patient cohort comprised 131 SCD, 337 naMCI, 247 aMCI, 53 AD, and 190 HC participants.

The neuropsychological assessment included the MMSE to assess overall cognitive functioning [24], the VVVMT to assess visuo-constructive abilities [25], the WST-IQ to assess verbal intelligence [26], and the BDI-II for depression screening [27]. Furthermore, the NTBV was conducted to assess multiple cognitive domains, including attention, memory, language, and executive functioning [28]. Olfactory assessment consisted of the SS-OIT and the ASOF inventory. The SS-OIT objectively assesses olfactory function by presenting 16 well-known odorants that must be correctly identified in a multiple-choice format. The corresponding cut-off score is set at 10 points, with scores of 11 points or higher indicating normosmia, whereas lower scores indicate hyposmia or anosmia [29]. By contrast, the 12-item ASOF questionnaire evaluates odor performance subjectively and is subdivided into three categories: the single-item Subjective Olfactory Capability (SOC), the five-item Self-Reported Capability of Perceiving Specific Odors (SRP), and the six-item Olfactory-Related Quality of Life (ORQ) subscales. Proposed cut-off scores are ≤ 3.0 for ASOF-SOC, ≤ 2.9 for ASOF-SRP, and ≤ 3.7 for ASOF-ORQ [30, 31].

Statistical calculations were conducted using IBM SPSS® Version 29.0.2 for Mac®OSX with a significance level set at α = 5%, and p ≤ 0.05 considered statistically significant. Standardized effect sizes according to Cohen’s classification were assessed to interpret the relevance of the results. In this context, Cohen’s d values of ≥ 0.20, ≥ 0.50, and ≥ 0.80 indicate small, medium, and large effects [32]. Furthermore, effect sizes (Pearson’s r) derived from Mann–Whitney U tests were interpreted as small (≥ 0.10), medium (≥ 0.30), and large (≥ 0.50) effects. Additionally, odds ratios were calculated to evaluate the strength of associations, with values of ≥ 2, ≥ 3, ≥ 7 indicating weak, moderate, and strong effects, respectively [33].

Descriptive statistics included the mean (M), standard deviation (SD), and range (minimum [min] and maximum [max]) for metric parameters. The median (Mdn) and interquartile range (IQR) were calculated for skewed distributions. Categorical, nominally scaled variables are reported as absolute frequencies (n) and percentages (%). For the estimation of expected probabilities in a population, corresponding confidence intervals [95%-CI: lower bound, LB; upper bound, UB] were implemented. In the case of a 5% type‑I error probability, the two-sided z value of 1.96 was applied [34].

Inferential statistics were conducted as follows: Subgroups differences in educational level, MMSE, VVVMT, and BDI-II scores were examined using the non-parametric Kruskal–Wallis test due to skewed data distribution. Post hoc pairwise comparisons were conducted using Mann–Whitney’s U test with Bonferroni correction. Differences in WST-IQ performance were examined using a one-way Welch analysis of variance (ANOVA). For comparisons of the SS-OIT and ASOF test results across subgroups, one-way Welch ANOVAs were applied to account for heterogeneity of variance. Furthermore, principal component analysis (PCA) was performed for dimensionality reduction of the NTBV-24 subtests to identify a smaller number of independent, weighted domains. Adequacy was evaluated using the Kaiser–Meyer–Olkin (KMO) criterion, and higher interpretability was achieved using orthogonal varimax rotation. To assess the discriminative ability of the SS-OIT and ASOF subtests in distinguishing AD from non-AD patients, receiver operating characteristic (ROC-AUC) analyses were performed. Optimal cut-off scores were determined using the Youden index. Positive predictive value (PPV), negative predictive value (NPV), and likelihood ratios (LR+ and LR−) were calculated to assess diagnostic utility. Sensitivity and specificity, the primary measures of diagnostic accuracy for dichotomous outcomes, are inversely related; increasing specificity reduces sensitivity and vice versa. Furthermore, PPV and NPV describe the predictive performance of a test, whereas accuracy reflects the overall proportion of correctly identified cases. Likelihood ratios (LR+ and LR−) were calculated and used as indicators for clinical applicability. An LR+ > 3 and LR− < 0.33 were considered indicative of acceptable diagnostic accuracy [35]. Finally, hierarchical binary logistic regression analysis was performed to evaluate the explanatory value of sociodemographic, olfactory, and neuropsychological covariates in predicting group membership (non-AD vs. AD). Odds ratios were used to interpret effect sizes, and model fit was assessed using Nagelkerke’s R2.