Introduction
Dementia is a clinical syndrome of cognitive decline which severely interferes with social and individual function. Progression is gradual; individuals with cognitively normal (CN) function may develop mild cognitive impairment (MCI) which advances to dementia over years to decades (Dallora et al., Reference Dallora, Minku, Mendes, Rennemark, Anderberg and Sanmartin Berglund2020). Dementia affects over 55 million individuals globally, with Alzheimer’s disease (AD) being the most common subtype (Nichols, Reference Nichols2022). With an absence of curative pharmacological treatments for dementia, attention has shifted toward early detection and modifiable risk factors, including tobacco use, physical inactivity, metabolic risks, and possibly depression (Almeida et al., Reference Almeida, Hankey, Yeap, Golledge and Flicker2017).
Late-life depression (LLD) is a potential contributor to dementia; it refers to depression in someone aged above 60 years, regardless of initial onset. Longitudinal studies show that older adults with depression have a 1.71- to 6.75-fold higher risk of developing dementia compared to nondepressed, age-matched controls (Mirza et al., Reference Mirza2016). Effective treatment for LLD exists, and depression treatment could potentially reduce dementia incidence (Dafsari and Jessen, Reference Dafsari and Jessen2020).
In the other direction, older individuals with existing dementia experience significantly higher incidence of depression (Huang et al., Reference Huang, Wang, Li, Xie and Liu2011). Moreover, incident dementia correlates strongly with new-onset LLD, but not with a remote history of depression, suggesting that LLD may be a prodrome of dementia, rather than a precipitating factor (Mirza et al., Reference Mirza2016).
Olfaction often deteriorates as part of the normal aging process, affecting around 15–25% of older adults (Choi et al., Reference Choi, Hur, Chow, Shen and Wrobel2018). The major domains of olfaction include odor identification (OI), threshold, memory, and discrimination (Kotecha et al., Reference Kotecha, Corrêa, Fisher and Rushworth2018). Studies suggest a link between OD and depression (Athanassi et al., Reference Athanassi, Dorado Doncel, Bath and Mandairon2021; Croy and Hummel, Reference Croy and Hummel2017), with evidence of reduced olfactory bulb volume and olfactory function scores in depressed patients (Rottstädt et al., Reference Rottstädt2018; Taalman et al., Reference Taalman, Wallace and Milev2017; Zucco and Bollini, Reference Zucco and Bollini2011). There is also evidence that this relationship is reciprocal (Kohli et al., Reference Kohli, Soler, Nguyen, Muus and Schlosser2016), as some olfactory functions normalize after antidepressant therapy (Rochet et al., Reference Rochet, El-Hage, Richa, Kazour and Atanasova2018). However, only one study has explored the directionality of this relationship longitudinally, finding that older adults with baseline OD were more likely to develop frequent depressive symptoms within 5 or 10 years, but not the reverse (Eliyan et al., Reference Eliyan, Wroblewski, McClintock and Pinto2021).
OD is also a common early symptom of dementia, presenting in 90–100% of individuals with the most common dementia subtypes, namely AD, Parkinson’s disease dementia, and frontotemporal dementia (Zou et al., Reference Zou, Lu, Liu, Zhang and Zhou2016). However, despite the high prevalence of OI impairment in individuals with dementia, assessment of OI is little used in clinical settings (Zou et al., Reference Zou, Lu, Liu, Zhang and Zhou2016). A review of the literature in 2019 concluded that, among other sensory biomarkers including hearing loss and visual changes, OI impairment most closely predicted conversion to MCI in healthy individuals, proving promising for disease identification in the preclinical stage (Murphy, Reference Murphy2019). Other studies investigating the utility of OI as a potential early marker for dementia pathology found low OI scores could accurately distinguish dementia from CN older participants (Rottstädt et al., Reference Rottstädt2018) and predict conversion from MCI to dementia (Stanciu et al., Reference Stanciu, Larsson, Nordin, Adolfsson, Nilsson and Olofsson2014).
Research looking at inter-relationships between olfactory function, depression, and dementia is scarce. Chen et al. (Reference Chen2021b) found LLD and impaired OI in older adults had an additive effect on symptoms associated with dementia. Individuals with both LLD and impaired OI had more severe structural and functional brain abnormalities compared to those with LLD and intact OI (Chen et al., Reference Chen2018). However, due to the cross-sectional nature of these studies, they were unable to determine whether the combination of OD and LLD predicted conversion to dementia in CN individuals. Reinforcing this, Murphy (Reference Murphy2019) found that OI dysfunction closely paralleled increased reductions in hippocampal and entorhinal cortex volumes, pathologies common to LLD and AD (Byers and Yaffe, Reference Byers and Yaffe2011; O’Shea et al., Reference O’Shea2018).
However, there are studies that found a positive association between impaired OI and dementia without finding a link between OI and LLD (Cha et al., Reference Cha, Kim and Son2022; Marine and Boriana, Reference Marine and Boriana2014). It is possible that these incongruent findings are due to heterogeneity in methodology, particularly in the OI measures used (Marine and Boriana, Reference Marine and Boriana2014).
Overall, limitations of current studies addressing olfaction, depression, and dementia include small sample sizes and relatively short follow-up periods, resulting in limited statistical power and wide confidence intervals (Pentzek et al., Reference Pentzek, Grass-Kapanke and Ihl2007; Zucco and Bollini, Reference Zucco and Bollini2011). Several studies utilized screening measures of global cognition – such as the Mini-Mental State Exam (MMSE) (Folstein et al., Reference Folstein, Folstein and McHugh1975) – to categorize participants into dementia and non-dementia subgroups, rather than expert clinical diagnosis based on comprehensive neuropsychological batteries (Cha et al., Reference Cha, Kim and Son2022; Duff et al., Reference Duff, McCaffrey and Solomon2002). Some studies employed convenience sampling (Cha et al., Reference Cha, Kim and Son2022; Chen et al., Reference Chen2018; Chen et al., Reference Chen2021b), selecting their participants from university hospitals or medical centeres, introducing possible selection bias. Most significantly, the cross-sectional nature of these studies precluded observations regarding dementia incidence in those with OD and/or LLD, limiting investigation of the temporal relationships between these three factors.
While recent cross-sectional research suggests that LLD and impaired OI may have additive contributions to dementia (Chen et al., Reference Chen2021b), to date, no longitudinal studies investigating this have been published. Furthermore, inconsistent findings in the relationship between OD and LLD make it difficult to ascertain the unique explanatory abilities of OD and depression when predicting incident dementia risk (Stanciu et al., Reference Stanciu, Larsson, Nordin, Adolfsson, Nilsson and Olofsson2014).
To address this gap, this study used data from the Sydney Memory and Ageing Study (MAS). The MAS was a longitudinal study of aging and cognition in a large, well-characterized community-dwelling cohort with a long follow-up time. It also employed validated measures of olfactory function and depressive symptomology; clinical dementia diagnoses were made by expert consensus.
The specific objectives of the current study were as follows:
-
1. To identify the cross-sectional relationship between olfactory function and depression.
-
2. To assess the individual and combined effects of baseline olfactory dysfunction and depression on the risk of incident dementia over 12 years of follow-up.
We hypothesized that olfactory function would show an association with depression, and that the ability of these factors to predict dementia would be additive.
Methodology
Participants
Participants were 780 older adults from the MAS, a longitudinal study that followed 1037 community-dwelling older adults (70–90 years) without dementia at baseline, who were recruited between 2005 and 2007 (Wave 1) (Sachdev et al., Reference Sachdev2010). Follow-up assessments (Waves 2 onward) were conducted by trained research assistants in 2-year intervals. Wave 1 assessments comprised a medical history interview, questionnaires, a medical examination, and a comprehensive neuropsychological examination designed to assess cognitive domains important for the diagnosis of dementia. Each subsequent wave included neuropsychological testing, a medical history interview, and a medical exam. The final wave of data was collected from the remaining 258 participants at 12-year follow-up in 2018–2020 (Wave 7).
At baseline, participants were required to speak and write English at a level proficient enough to complete a psychometric assessment and consent to participate in the study. Participants were excluded at baseline if they had a previous diagnosis of dementia, psychotic symptoms or a diagnosis of schizophrenia or bipolar disorder, developmental disability, progressive malignancies, or any medical or psychological conditions that might have prevented them from completing assessments (Sachdev et al., Reference Sachdev2010). If a participant received a diagnosis of dementia from the study team, after assessment according to The Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition (DSM-IV) (American Psychiatric Association, 1994), they were also excluded. Further exclusion criteria included an MMSE (Folstein et al., Reference Folstein, Folstein and McHugh1975) score of <24 after adjustment for age, education, and non-English-speaking background (NESB) (Anderson et al., Reference Anderson, Sachdev, Brodaty, Trollor and Andrews2007).
For the current study, 164 NESB participants were excluded. A further 93 participants were excluded according to specific criteria, including current smokers, participants missing Brief Smell Identification Test (BSIT) data, those with nasopharyngeal cancer or those with reduced sense of smell following surgery. Current smokers were excluded due to their increased risk of OD (Ajmani et al., Reference Ajmani, Suh, Wroblewski and Pinto2017). NESB participants were excluded as a lack of normative data for this group may have affected the accuracy of their neuropsychological profiles (Anderson et al., Reference Anderson, Sachdev, Brodaty, Trollor and Andrews2007; Sachdev et al., Reference Sachdev2010).
Measures
Olfaction
Olfactory function, specifically OI, was assessed using the BSIT (Doty et al., Reference Doty, Marcus and Lee1996). The BSIT is a noninvasive, 12-item test that asks participants to smell an odorant strip and, for each odorant, to identify the corresponding scent from a four-category multiple-choice questionnaire. The participant must choose one of the four options for each odorant and receives a point for every item identified correctly out of the 12 presented. A higher BSIT score relates to better olfactory performance. The BSIT has been found to have good internal reliability and validity (Menon et al., Reference Menon, Westervelt, Jahn, Dressel and O'Bryant2013).
Depression
Depressive symptoms were assessed using the 15-item Geriatric Depression Scale (GDS) (Sheikh and Yesavage, Reference Sheikh and Yesavage1986), a self-report questionnaire shown to be a valid and reliable measure of depressive symptoms in older adults, including those with MCI (Mitchell et al., Reference Mitchell, Bird, Rizzo and Meader2010). It consists of 15 dichotomous (yes/no) questions and is scored on a 15-point scale, with a greater score indicating greater depressive symptoms.
Global cognition
Global cognitive function was assessed using the MMSE (Folstein et al., Reference Folstein, Folstein and McHugh1975), a well-validated test of global cognitive function used to screen for dementia. The MMSE is an 11-question measure that tests five areas of cognitive function, including orientation, attention, recall, and language. It is scored on a 30-point scale with a lower score suggesting a greater degree of cognitive impairment (Anderson et al., Reference Anderson, Sachdev, Brodaty, Trollor and Andrews2007).
Framingham cardiovascular risk
Cardiovascular disease (CVD) risk scores were computed according to the Framingham stroke study protocol (D'Agostino et al., Reference D'Agostino2008). Participants were scored on several known risk factors for cardiovascular incidents, including sex, cholesterol, smoking status, diabetic status, systolic blood pressure, lipoprotein levels, and medication status. These scores were tallied to produce a total CVD risk score.
APOE4
The apolipoprotein E (APOE) genotype is a known risk factor for several diseases. Polymorphic alleles of APOE have been shown to carry different levels of risk of dementia, with the highest risk associated with the e4 allele (APOE4). Each individual carries two copies of the APOE gene (Rasmussen et al., Reference Rasmussen, Tybjærg-Hansen, Nordestgaard and Frikke-Schmidt2018). In this study, those carrying one or two copies of the e4 allele were classified as APOE4 carriers and given a score of 1. Noncarriers received a score of 0.
Clinical diagnosis of dementia
Clinical diagnoses of dementia were made at each wave by a clinical panel comprised of at least three experts, including a neuropsychiatrist, psychogeriatrician, and a neuropsychologist, who discussed all available clinical, neuropsychological, laboratory, and imaging data to reach a diagnosis for participants sent to consensus review. Participants were reviewed if they met the following criteria: MMSE ≤24; drop in MMSE ≥3 points; impaired activities of daily living not due to physical impairments; elevated scores on an informant-reported scale of instrumental activities of daily living (IADL); or a prior interim diagnosis of dementia.
A diagnosis of dementia was based on DSM-IV criteria (American Psychiatric Association, 1994), which includes the development of one or more cognitive deficit(s) that represent a decline from a previous level of performance that are sufficiently severe as to cause impairment in daily functioning (Bayer IADL scale score ≥3.0) (Hindmarch et al., Reference Hindmarch, Lehfeld, de Jongh and Erzigkeit1998). Participants who had complete neuropsychological test data and did not meet criteria for a dementia diagnosis were classified as “not having dementia” at each wave. Clinical diagnoses were available for Waves 1–7 (12-year follow-up).
Statistical analysis
Baseline demographics
Baseline differences in demographics and test scores for participant groups, such as those with and without dementia, were examined using independent samples t-tests for non-skewed continuous variables, independent samples Mann–Whitney U tests for skewed continuous variables, and chi-square tests for categorical variables.
Correlations among BSIT and depression
Constructing a categorical depression score
Due to the highly skewed, nonnormal distribution of GDS scores in the mostly healthy, nondepressed sample at Wave 1, as well as the possible influence of antidepressant use on these scores, a categorical depression variable was constructed according to the following criteria:
-
0 = GDS <5 and not using antidepressants
-
1 = using antidepressants OR GDS ≥5 but not both
-
2 = using antidepressants AND GDS ≥5
An initial one-way analysis of variance (ANOVA) was performed to examine the association between this tiered depression variable and continuous BSIT scores (see Results). However, due to the small size of the group using antidepressants AND GDS ≥5 (n = 16), and the nonsignificant differences between the second and third groups, it was condensed to a binary depression variable according to the following:
-
0 = neither depressed nor using antidepressants
-
1 = using antidepressants OR GDS ≥5
For the purposes of this study, individuals using antidepressants were classified as depressed regardless of their GDS score. The values “0” and “1” for this variable will correspond to the labels “not depressed” and “depressed.”
Correlation analysis
The relationship between BSIT and depression was investigated by computing a simple point-biserial correlation and a partial point-biserial correlation controlling for several covariates, including age, sex, education, CVD risk, MMSE score, and APOE4 carrier status.
Olfaction, depression, and incident dementia
Cox regressions were performed to assess the effects of baseline BSIT scores and baseline depression status on incident dementia across the 12-years of follow-up. Scaled Schoenfeld residuals plots were generated to visually inspect the violation of the proportional hazards assumption. Moreover, interaction effects between survival time and each predictor were examined to further test for the violation of this assumption.
A hierarchical Cox regression model was performed to assess whether the predictive value of BSIT for dementia was above-and-beyond that provided by binary depression. A second model was performed, which adjusted for several relevant covariates including age, sex, education, MMSE, cardiovascular disease risk, and APOE4 carrier status. The first block contained all the covariates, the second block contained the binary depression score, and the third block contained the BSIT scores. Additional analysis exploring whether there was any interaction effect between BSIT scores and depression status was conducted. Furthermore, we examined a cause-specific hazard model for dementia diagnosis accounting for death. In this model, censoring was specified on the date of death or at the end of follow-up/participant drop-out.
To compare the discrimination ability between depression and BSIT, concordance statistics (c-statistics) (Harrell et al., Reference Harrell, Califf, Pryor, Lee and Rosati1982) were generated to assess how well a model used risk scores to predict time-to-event, acting as a measure of goodness-of-fit (Uno et al., Reference Uno, Cai, Pencina, D'Agostino and Wei2011). In particular, we compared a model containing covariates and depression, with another model containing covariates and BSIT. Furthermore, positive redictive value (PPV) and negative predictive value (NPV) were calculated for the binary depression variable, and for olfactory impairment, using the literature standard cutoff of ≤8/12 on the BSIT (El Rassi et al., Reference El Rassi2016).
All statistical analyses were performed using IBM SPSS Statistics 26 for Windows (IBM Corporation, 2021). A two-sided p-value <.05 was considered statistically significant.
Results
Sample characteristics and baseline comparisons
Table 1 displays group comparisons between English-speaking background participants included and excluded from the study. CVD risk scores were lower and BSIT higher in the included compared to the excluded samples.
Baseline differences between English-speaking background participants included and excluded from the study
MMSE, Mini-Mental State Exam; IQR, interquartile range; CVD, cardiovascular disease; APOE4, apolipoprotein e4 allele carrier status; BSIT, Brief Smell Identification Test; GDS, Geriatric Depression Scale.
aDepression operationalized as GDS ≥5 OR current antidepressant use.
By Wave 7, of the 527 participants no longer in the study, 288 (54.6%) passed away, 189 (35.9%) withdrew, and 50 (9.5%) were not assessed for reasons including poor health and geographical relocation. At baseline, participants who remained in the study at the 12-year follow-up were younger, were more educated, and had lower GDS and higher BSIT scores compared to participants who left the study, with a greater proportion of females compared to males. Baseline demographics of participants who remained in the study at Wave 7 versus those that did not are presented in Table 2.
Baseline differences between Wave 7 completers and noncompleters
SD, standard deviation; MMSE, Mini-Mental State Exam; IQR, interquartile range; CVD, cardiovascular disease; APOE4, apolipoprotein e4 allele; BSIT, Brief Smell Identification Test; GDS, Geriatric Depression Scale.
aDepression operationalized as GDS ≥5 OR current antidepressant use.
For the current study, we examined group differences between baseline demographics for participants who did (n = 195) and did not (n = 585) progress to dementia by Wave 7, which are presented in Table 3. At baseline, participants who progressed to dementia were significantly older than those who did not, were more likely to carry APOE4 alleles, and had lower scores on the BSIT. However, there were no group differences for sex, years of education, MMSE scores, GDS scores, or antidepressant use.
Baseline differences between dementia and nondementia groups
SD, standard deviation; MMSE, Mini-Mental State Exam; IQR, interquartile range; CVD, cardiovascular disease; APOE4, apolipoprotein e4 allele; BSIT, Brief Smell Identification Test; GDS, Geriatric Depression Scale.
aDepression operationalized as GDS ≥5 OR current antidepressant use.
Relationships between B-SIT and depression at baseline
The one-way ANOVA between the three-tier depression variable and BSIT scores revealed no statistically significant difference in BSIT scores between any groups (F(2, 738) = 1.608, p = .201). However, a downward trend in BSIT scores with each tier of increased depression “severity” was observed (M = 9.35, 9.04, 8.75 for the respective depression categories).
A simple correlation between BSIT scores and the binary depression variable at Wave 1 revealed a weakly and negatively associated, but nonsignificant, association (r = −.060, p = .094). When controlling for age, sex, education, MMSE, CVD risk, and APOE4 carrier status on the relationship between depression and BSIT scores, the association remained nonsignificant (r = −.051 p = .173).
Prediction of incident dementia
Table 4 displays the results of a hierarchical Cox regression examining the associations between baseline BSIT scores and depression, and risk of dementia over 12 years of follow-up. An unadjusted Cox regression showed that participants classified as depressed at Wave 1 were at a 1.7-fold increased risk of progression to dementia compared to nondepressed participants (HR = 1.742, 95%CI 1.216–2.495, p = .002). Adding BSIT scores to the model increased the model fit (χ2 = 37.668, p < .001). Moreover, a lower BSIT score was associated with a higher risk of dementia after adjusting for depression, and a one-unit increase in BSIT score was associated with a 19.6% decrease in the risk of developing dementia (HR = 0.804, 95%CI .754–.857, p < .001). Depression remained a significant predictor even after BSIT scores were added to the model (HR = 1.668, 95%CI 1.164–2.389, p = .005). No significant interaction effect between depression and BSIT scores was found.
Hierarchical Cox regression results
MMSE, Mini-Mental State Exam; CVD, cardiovascular disease; APOE4, apolipoprotein e4 allele; BSIT, Brief Smell Identification Test; GDS, Geriatric Depression Scale.
adf = 1 for all variables.
bDepression operationalized as GDS ≥5 OR current antidepressant use.
Bolded to denote statistical significance at p < 0.05.
Controlling for covariates did not change the associations. Participants classified as depressed at Wave 1 remained at a 1.7-fold increased risk of progression to dementia compared to nondepressed participants (HR = 1.706, 95%CI 1.185–2.456, p = .004). Adding BSIT scores to the model increased the model fit (χ2 = 20.882, p < .001). A lower BSIT score was again associated with a higher risk of dementia, with a one-unit increase in BSIT score associated with a 15.5% reduction in risk of developing dementia (HR = .845, 95%CI .789–.905, p < .001). Depression remained a significant predictor even after BSIT scores were added to the model (HR = 1.649, 95%CI 1.145–2.374, p = .007). Scaled Schoenfeld residual plots did not indicate any violation of the proportional hazards assumption (see Supplementary Materials Figure 1), and there was no significant interaction between survival time and each predictor observed.
The cause-specific hazard model accounting for the competing risk of death revealed results similar to those in the main analysis (Supplementary Materials Table 1). Incident dementia was significantly associated with depression (cause-specific HR = 1.612, 95%CI 1.119–2.322, p = .010) and lower BSIT scores (csHR = .864, 95%CI .808–.924, p < .001) after controlling for covariates.
Concordance statistics were calculated for the two models which used depression or BSIT scores as a predictor of incident dementia. The c-statistic for the model containing covariates and depression was 0.699 (95%CI 0.663–0.734), compared to .723 (95%CI 0.688–0.758) for the model containing the same covariates and BSIT scores. Both c-statistics indicated acceptable model fit and comparable ability in the classification of incident dementia cases. Olfactory impairment had a PPV of 33.0% and NPV of 77.9% for dementia. Depression had a PPV of 28.5% and NPV of 75.6% for dementia.
Discussion
This study first aimed to determine whether olfaction and depression were associated cross-sectionally in a large, community-dwelling sample of older adults without dementia. Next, the study aimed to determine whether olfaction or depression were individually stronger predictors of incident dementia (Almeida et al., Reference Almeida, Hankey, Yeap, Golledge and Flicker2017; Murphy, Reference Murphy2019; Pacyna et al., Reference Pacyna, Han, Wroblewski, McClintock and Pinto2023), and whether the predictive value improved when both were included in the same model.
Although this study hypothesized that olfactory performance and depression would be correlated cross-sectionally, our results did not support this. While this finding runs contrary to a number of other studies (Athanassi et al., Reference Athanassi, Dorado Doncel, Bath and Mandairon2021; Chen et al., Reference Chen2021a; Kohli et al., Reference Kohli, Soler, Nguyen, Muus and Schlosser2016; Taalman et al., Reference Taalman, Wallace and Milev2017), it is not entirely unusual (Rochet et al., Reference Rochet, El-Hage, Richa, Kazour and Atanasova2018), with multiple studies finding that the depressed individuals maintain similar olfactory function to a healthy cohort (Marine and Boriana, Reference Marine and Boriana2014; Pentzek et al., Reference Pentzek, Grass-Kapanke and Ihl2007). Scinska et al. (Reference Scinska2008) were unable to find a correlation between GDS and olfactory identification scores in non-demented older adults, whereas Economou (Reference Economou2003) found no association between BSIT scores and scores on a depression inventory. Moreover, Rossi et al. (Reference Rossi2015) observed no difference in OI scores between dementia patients with and without depression. Heterogeneity in definitions for “depression” and “olfactory dysfunction” may explain this discrepancy. Chen et al. (Reference Chen2021a) demonstrated that depression scores were higher than controls in anosmic, but not hyposmic patients. Meanwhile, Khil et al. (Reference Khil, Rahe, Wellmann, Baune, Wersching and Berger2016) found OI impairment only in patients with major depressive disorder with high symptom severity. Other studies using a clinical diagnosis of depression rather than depression inventory scores produce similar results (Croy and Hummel, Reference Croy and Hummel2017), suggesting that a strong association may only emerge when using more severe disease measures. In contrast, participants in the current study were relatively healthy and nondepressed, with a median GDS score of 2/15 and a median BSIT score of 10/12. The absence of a cross-sectional correlation also points away from LLD and OD co-occurring as a prodrome in the early stages of dementia (Singh-Manoux et al., Reference Singh-Manoux2017). Future studies following BSIT performance and depression across several time points might observe a more marked correlation closer to dementia diagnosis, or as participants develop more severe olfactory or depressive symptoms.
Furthermore, this study found that olfactory function at baseline was significantly different in the group that progressed to dementia (median BSIT score of 9/12) compared to the group which did not progress to dementia (10/12) (Table 3). Our findings align with the literature demonstrating that those with lower scores on olfactory function tests such as the BSIT have a higher risk of progressing to dementia (Adams et al., Reference Adams, Kern, Wroblewski, McClintock, Dale and Pinto2018; Pacyna et al. Reference Pacyna, Han, Wroblewski, McClintock and Pinto2023). This accelerated olfactory deterioration may be due to involvement of the olfactory bulb even in the preclinical stage of dementia pathology (Alves et al., Reference Alves, Petrosyan and Magalhães2014). Detecting subtle olfactory impairment may provide clinicians with early insight into disease risk, although the small median difference of 1 point in our findings may not be clinically practicable. Adams et al. (Reference Adams, Kern, Wroblewski, McClintock, Dale and Pinto2018) used an olfactory function test to predict dementia in CN, community-dwelling older adults after 5 years and established a PPV of 9%. Adams suggested that this value was due to low disease prevalence and would improve with a longer follow-up period. With 12 years of follow-up, the current study demonstrated that impaired olfaction (using a cutoff of BSIT ≤ 8) had a PPV of 33% and an NPV of 77.9%, with a median time of 7 years to dementia conversion (Table 1). Similarly, depression had a PPV of 28.5% and an NPV of 75.6%. In context, better results are achieved using blood/cerebrospinal fluid biomarkers such as p-tau181 (PPV = 50%, NPV = 87%). However, it must be noted that these fluid biomarkers are more invasive and only effective in predicting Alzheimer’s dementia (Chatterjee et al., Reference Chatterjee2022). Thus, as binary variables, olfactory dysfunction and depression may best be used as broad, noninvasive screening tools to assist in predicting dementia many years in advance.
Importantly, this study found that lower BSIT scores and having depression were significant independent predictors of progression to dementia longitudinally, in line with other longitudinal findings (Adams et al., Reference Adams, Kern, Wroblewski, McClintock, Dale and Pinto2018; Chen et al., Reference Chen2021b; Pacyna et al. Reference Pacyna, Han, Wroblewski, McClintock and Pinto2023). This study adds to the literature by demonstrating that their predictive abilities are unique and provide additional predictive value when used in combination, which no other longitudinal study has investigated (Chen et al., Reference Chen2021b). Stanciu et al. (Reference Stanciu, Larsson, Nordin, Adolfsson, Nilsson and Olofsson2014) suggested that the association between OD and dementia may be due to shared variance with depressive symptoms. However, our study demonstrates that both variables remained unique significant predictors in the combined model (p < .001). Moreover, given the number of participant deaths throughout the duration of the study, the possibility of death as a competing risk was considered. A cause-specific competing risk model was tested, which revealed similar patterns to the primary analysis. Thus, mortality as a competing risk does not significantly alter the above conclusions.
Furthermore, no longitudinal studies have compared the predictive abilities of depression and OD for dementia (Bergmann et al., Reference Bergmann, Stögmann and Lehrner2021), and an aim of this study was to determine whether one of these had better diagnostic ability. The concordance statistics for the final two-block models were similar, and the overlap in their confidence intervals suggests that their diagnostic abilities do not differ significantly. Thus, there is no clear evidence for either olfactory dysfunction or depression (as measured in this study) as the superior predictor of dementia. Clinically, these findings are important, suggesting that olfaction and depression screening may best be used in conjunction for discriminating individuals at a greater risk of developing dementia early in the disease course.
The BSIT has limitations such as being single use (Doty et al., Reference Doty, Marcus and Lee1996; El Rassi et al., Reference El Rassi2016) and potentially costly to implement in the growing over-60 population. Targeting olfactory screening toward higher risk older adults using medical history factors may improve the PPV and reduce costs (Adams et al., Reference Adams, Kern, Wroblewski, McClintock, Dale and Pinto2018). Despite these costs, the BSIT is time-efficient, can be sent by mail, and is significantly cheaper than current dementia biomarkers, including cerebrospinal fluid sampling and neuroimaging (Pacyna et al. Reference Pacyna, Han, Wroblewski, McClintock and Pinto2023; Wittenberg et al., Reference Wittenberg, Knapp, Karagiannidou, Dickson and Schott2019). Our descriptive results show that a single administration of the BSIT at Wave 1 was sufficient to distinguish conversion to dementia, which occurred on average 7 years later (Table 3). Thus, the value provided by potential early detection of dementia by even infrequent review of olfactory function would likely outweigh the costs of administration. Sensitivity and specificity analyses in an older adult population, using multiple age-normed BSIT cutoff scores to predict dementia, would lay the groundwork for the use of olfactory function testing for dementia screening clinically.
Our findings also show the value of assessing depressive symptoms in older adults. The 15-item GDS is resource-efficient; capturing GDS scores and recording antidepressant medication is relatively simple to perform regularly, allowing for monitoring of depressive symptoms over time and more targeted early intervention. Beyond its predictive value, future research may again focus on the directionality of the relationship between dementia and depression, and whether treatment of depression could slow or even reverse cognitive decline.
This study has limitations, including a baseline cohort that was generally healthy, well-educated, and largely of White European ancestry (94%), limiting generalizability. Moreover, while the literature broadly supports depression as a predictor or driver of dementia (Byers and Yaffe, Reference Byers and Yaffe2011), our depression measure consisted of antidepressant use and a single GDS score, rather than a clinical diagnosis of depression. Furthermore, the depression measure was condensed to a binary variable due to the high skew of the GDS scores in the relatively nondepressed baseline cohort. Moreover, all participants using antidepressants were categorized as depressed in order to account for potential depression-in-remission. However, it is possible that some participants were using antidepressants for off-label uses (Wong et al., Reference Wong, Motulsky, Eguale, Buckeridge, Abrahamowicz and Tamblyn2016). Another limitation of measuring BSIT performance and depression at one time point is the inability to examine their change. Repeated assessments of OI and depression would allow analysis of how the rate and severity of deterioration in these factors influences the risk of dementia (Kim et al., Reference Kim2019). Moreover, a more tiered outcome variable – classifying individuals as CN, MCI, or dementia – could be used to observe the trajectory of cognitive decline rather than just dementia incidence.
Strengths of this study include the large, well-characterized cohort of community-dwelling older adults, with comprehensive records of their physiological health markers and neuropsychological data. The study controlled for highly influential covariates such as APOE4 gene status, which has been strongly linked to dementia incidence (Rasmussen et al., Reference Rasmussen, Tybjærg-Hansen, Nordestgaard and Frikke-Schmidt2018) and cardiovascular risk scores. This improves confidence in our findings as dementia is a multifactorial condition (Dallora et al., Reference Dallora, Minku, Mendes, Rennemark, Anderberg and Sanmartin Berglund2020). As dementia is a slowly progressing disease, (Dallora et al., Reference Dallora, Minku, Mendes, Rennemark, Anderberg and Sanmartin Berglund2020) the long 12-year study period allowed us to follow many participants who developed dementia, providing more power to our analyses. Frequent follow-up allowed us to obtain clinical data from participants who were censored before Wave 7, enabling us to conduct survival analyses. Additionally, between the English-speaking background participants included and excluded from the study, there appeared to be no unintended group differences. Those excluded for smoking, nasopharyngeal cancer, and post-surgical anosmia were expected to have higher CVD risk and lower BSIT scores. None of the other variables differed significantly between groups.
Furthermore, the measures used in this study, including the BSIT and GDS-15, were selected for their high validity and ease of administration (Menon et al., Reference Menon, Westervelt, Jahn, Dressel and O'Bryant2013; Mitchell et al., Reference Mitchell, Bird, Rizzo and Meader2010). Dichotomizing the depression variable allowed the study to account for antidepressant use, a factor expected to affect GDS scores (Almeida et al., Reference Almeida, Hankey, Yeap, Golledge and Flicker2017). Point-biserial partial correlations allowed for examining associations between depression and olfactory function without assuming the directionality of the relationship (Demirtas and Hedeker, Reference Demirtas and Hedeker2016). Additionally, Cox regression survival analysis was used to investigate the impact of depression and olfactory function on dementia risk as it allowed for a more granular analysis of incidence across all waves and accounted for those who had been censored before Wave 7 (Goerdten et al., Reference Goerdten, Carrière and Muniz-Terrera2020). Another strength is the robust method of determining dementia – through a combination of multiple well-validated neuropsychological assessments and consensus diagnosis from an expert panel. Future studies may consider a similarly rigorous approach, particularly using clinical diagnosis, to assess depression.
Conclusion
In conclusion, the present study has shed light on the potential for depression and OI to improve early prediction of dementia. While these associations have been examined previously, the current study addresses the absence of a rigorously designed longitudinal cohort study in which direct evidence of neurodegeneration can be observed in individuals with OI and/or depression (Chen et al., Reference Chen2018; Chen et al., Reference Chen2021b; Petersen, et al., Reference Petersen, Bresolin and Monteiro2021). These findings have important clinical implications. In particular, that the use of an olfactory function measure in conjunction with depressive symptomatology can predict progression to dementia over 12 years better than either depression or olfactory impairment alone.
Conflict of interest
The authors declare none.
Acknowledgments
First and foremost, we thank the MAS participants for their enthusiastic support. We also thank the CHeBA Co-Directors, Henry Brodaty and Perminder Sachdev, the MAS research assistants, the CHeBA Data Manager, the Centre Manager, and the CHeBA Communications team. I would like to extend my thanks to my supervisors Dr Simone Reppermund, Dr Katya Numbers, and Dr Darren M. Lipnicki for their support and guidance throughout this project; it goes without saying that I could not have done this without their help. I’m also tremendously grateful to Dr Ben Lam and Saly Mahalingam for their time, generosity, and expertise in all things statistical.
Ethics statement
Written informed consent from participants was obtained. Approval for the current study was obtained from the Human Research Ethics Committee of the University of New South Wales (HC: 05037, 09382, 14327, 190962).
Supplementary material
To view supplementary material for this article, please visit https://doi.org/10.1017/S1041610223000480.
