Ambient Air Pollutants and Olfaction among Women 50–79 Years of Age from the Sister Study

Abstract

Background:

Poor olfaction is common in older adults and may have profound adverse implications on their health. However, little is known about the potential environmental contributors to poor olfaction.

Objective:

We investigated ambient fine particulate matter [PM $\le 2.5\mathrm{\mu }\mathrm{m}$ in aerodynamic diameter (${\mathrm{PM}}_{2.5}$)] and nitrogen dioxide (${\mathrm{NO}}_2$) in relation to poor olfaction in middle-aged to older women.

Methods:

The Sister Study is a nationwide cohort of 50,884 women in the United States with annual average air pollutant exposures estimated based on participants’ residences from enrollment (2003–2009) through 2017. This analysis was limited to 3,345 women, 50–79 years of age as of January 2018, who completed the Brief Smell Identification Test (B-SIT) in 2018–2019. Poor olfaction was defined as a B-SIT score of $\le 9$ in the primary analysis. We conducted multivariable logistic regressions, accounting for covariates and study sampling design.

Results:

Overall, we found little evidence for associations of air pollutants with poor olfaction. The odds ratio (OR) and 95% confidence interval (CI) of poor olfaction for each interquartile range (IQR) increment of air pollutants in 2006 were 1.03 (95% CI: 0.91, 1.17) for ${\mathrm{PM}}_{2.5}$ (per $3.3{\mathrm{\mu }\mathrm{g}/\mathrm{m}}^3$) and 1.08 (95% CI: 0.96, 1.22) for ${\mathrm{NO}}_2$ (per $5.7\text{\hspace{0.17em}ppb}$). Results were similar in the analyses using the most recent (2017) or the cumulative average (2006–2017) air pollutant exposure data. Secondary analyses suggested potential association in certain subgroups. The OR per IQR was 1.35 (95% CI: 1.11, 1.65) for ${\mathrm{PM}}_{2.5}$ among younger participants ($<54.2$ years of age) and 1.87 (95% CI: 1.29, 2.71) for ${\mathrm{NO}}_2$ among current smokers.

Discussion:

This study did not find convincing evidence that air pollutants have lasting detrimental effects on the sense of smell of women 50–79 years of age. The subgroup analyses are exploratory, and the findings need independent confirmation. https://doi.org/10.1289/EHP12066

Introduction

Poor olfaction or impaired sense of smell affects up to 25% of older adults in the United States.^1,2 The prevalence is age dependent and is higher in men than in women and higher in Black people compared with White people. Unlike hearing or vision impairments, poor olfaction often goes unnoticed,^2,3 yet it may have significant public health ramifications. In older adults, poor olfaction may gradually diminish their appetite or degrade food choices, adversely affecting diet and nutrition, eventually leading to weight loss and poor health.^4,5 Poor olfaction may also impair detection of environmental hazard cues (e.g., smoke, spoiling foods, toxic chemicals), resulting in accidents.^6,7 More importantly, poor olfaction is one of the earliest sensory and most important prodromal symptoms of neurodegenerative diseases, such as Parkinson’s and Alzheimer’s,^8–10 and robustly predicts total mortality in older adults.^11–14 Further, recent studies suggest that poor olfaction may also relate to a higher risk of pneumonia,¹⁵ cardiovascular diseases,¹⁶ diabetes,¹⁷ depression,¹⁸ physical functional decline,¹⁹ and frailty.²⁰ Given the high prevalence of poor olfaction in older adults and its potential health implications, it is important to identify the causes of olfactory loss. However, available literature has focused mainly on olfactory loss during or following medical conditions, such as traumatic brain injury,²¹ chronic rhinosinusitis,²² and, most recently, COVID-19.²³

Airborne environmental exposures, air pollutants in particular, are of special interest to the research of poor olfaction because the upper respiratory structure is right at the interface between the environment and the host.²⁴ Air pollutants may damage the olfactory epithelium, impairing nerve function, and inducing local acute or chronic inflammation, which may gradually lead to olfactory loss.^25,26 These damages may be exacerbated in older adults because the integrity and health of nasal mucosa may deteriorate owing to aging and cumulative insults from a lifetime of airborne pollutants and pathogen invasions. Further, experimental studies suggest that exposure to fine particulate matter [PM $\le 2.5\mathrm{\mu }\mathrm{m}$ in aerodynamic diameter (${\mathrm{PM}}_{2.5}$)] may lead to neurotoxic damages to the olfactory bulb^27,28 and might accelerate the development of neurodegenerative pathologies in rodents.²⁹ Finally, postmortem studies of young residents of air-polluted Mexico City reported pathological and neuroinflammatory changes in their olfactory bulb that are indicative of potential neurodegeneration.^30,31 Several epidemiological studies have examined the link between ambient air pollutants of ${\mathrm{PM}}_{2.5}$ and nitrogen dioxide (${\mathrm{NO}}_2$) to olfactory impairment, but the evidence is preliminary and inconsistent.^32–36 We therefore examined associations of exposures to ${\mathrm{PM}}_{2.5}$ and ${\mathrm{NO}}_2$ with objectively assessed olfactory impairment in a sample of a large nationwide cohort of U.S. middle-aged and older women. Further, we conducted exploratory interaction analyses with age, race, and smoking because the olfaction impairment is age dependent,^1,2 substantially more common in Blacks than in Whites,^1,37 and may be associated with smoking, as shown in a recent meta-analysis.³⁸

Methods

Study Population and Sampling

The Sister Study is an ongoing nationwide cohort established by investigators from the National Institute of Environmental Health Sciences/National Institutes of Health (NIEHS/NIH) to investigate environmental and genetic risk factors for breast cancer and other chronic diseases.^39,40 In 2003–2009, the study enrolled a total of 50,884 women, 35–74 years of age, who had a sister with breast cancer and were from all 50 U.S. states and Puerto Rico and completed a comprehensive computer assisted telephone interview, multiple mailed comprehensive questionnaire surveys, and a home visit. Since enrollment, the cohort has been followed with annual contact updates and detailed follow-up surveys (DFUs) every 2–3 y. At the DFU-3 in 2014–2016, we asked participants: “Do you suffer from a decrease in or loss of your sense of smell?” Based on answers to this question, we selected a case–control sample of Sister Study participants for an olfaction substudy in 2018–2019. Of the 36,491 women who met the eligibility criteria for the substudy (presumed alive and 50–79 years of age on 1 January 2018), we selected all participants who reported a poor sense of smell ($n=\mathrm{2,819}$) and a random sample of 1,200 participants who reported a normal sense of smell. In this olfaction substudy, we invited all selected participants to take the self-administered Brief Smell Identification Test (B-SIT), distributed efficiently by mail. A total of 2,353 (83%) self-reported cases and 1,053 (88%) self-reported controls returned the B-SIT booklets with valid smell test results. Details on olfaction substudy design, sampling process, and participation rate have been reported previously.³ The study was approved by the institutional review boards at Michigan State University, the NIEHS/NIH, and the U.S. Department of Defense.

Outcome Assessment

The B-SIT is an abbreviated version of the 40-item Pennsylvania Smell Identification Test⁴¹ and has been widely used to screen for olfactory impairment in epidemiological studies.^42–45 The B-SIT was designed to be self-administered, either by mail or in person.⁴¹ The test is a small booklet that contains 12 commonly experienced odorants,⁴⁶ one embedded on each page. Participants are instructed to scratch and smell each of these odorants, one at a time, and to identify the correct odorant from four possible answers in a forced multiple-choice format.⁴¹ Every correct answer is awarded one point, summing to a final score ranging from 0 to 12, with a higher score indicating a better sense of smell.⁴⁶ In addition to nonparticipation (445 self-reported cases/143 self-reported controls), we considered the B-SIT score as invalid if it had $>3$ items missing (21/4). For those with 1–3 missing items (104/46), we prorated the score based on completed items. Given that White women generally have better olfactory abilities than men or Black women and that our study participants were predominantly non-Hispanic White women (89.5%) and were relatively young at 50–79 years of age,^1,3,37 we defined poor olfaction as a B-SIT score of $\le 9$ in the primary analyses. With this definition, $\sim 13\%$ of the age-eligible Sister Study population had poor olfaction.³ As expected, owing to the low accuracy in self-reported olfactory impairment, 1,449 (61.6%) of the 2,353 self-reported poor olfaction were reclassified as normal upon B-SIT testing, and 116 (11.0%) of the 1,053 self-reported normal were reclassified as poor.³ After further excluding 61 women who were missing air pollutant exposures, the primary analytic sample included 1,001 cases with B-SIT–tested poor olfaction and 2,344 subjects with tested normal olfaction.

Exposures of Interest

Air pollutant concentration was estimated at each participant’s residence using a synthesized suite of regional, spatiotemporal models.⁴⁷ As detailed elsewhere, this fine-scale air pollution modeling was developed for large-scale epidemiologic research.⁴⁷ Briefly, the models used pollutant concentration data from over 900 research monitors—including residential monitoring campaigns, fixed sites, and (for the ${\mathrm{NO}}_2$ model) gradient monitoring near roadways, along with 1,500 regulatory agency monitors across the United States, and a large ($>200$) suite of geographic covariates calculated at each monitoring location and each residential location.^48,49 This modeling approach accounted for complex spatiotemporal dependencies in a land use and spatial smoothing framework. It produced 2-wk average pollutant concentrations at the geospatial point for each participant’s residential address. In the present study, the concentrations were further averaged to the annual exposure level from 2006 to 2017, accounting for moving and updated addresses during the follow-up. Model performance was evaluated using the cross-validation method, and the cross-validated $R^2$ value was 0.89 for ${\mathrm{PM}}_{2.5}$ and 0.87 for ${\mathrm{NO}}_2$.⁴⁷ In our primary analysis, we considered average ${\mathrm{PM}}_{2.5}$ and ${\mathrm{NO}}_2$ levels in 2006 (baseline of the Sister Study) as the exposures of interest because they were evaluated in the mid–time point of Sister Study enrollment with the least missing values in exposures. In sensitivity analyses, we also used the latest available exposure levels from 2017 for ${\mathrm{PM}}_{2.5}$ and ${\mathrm{NO}}_2$, as well as the average levels during the measurement period (2006–2017).

Covariate Assessments

The Sister Study has collected extensive data on demographics, lifestyle, environmental exposures, and health status at enrollment and its follow-ups. In the analysis, we considered covariates that might affect potential confounding, missingness in the outcome data, and the accuracy of baseline exposure assessment. They were selected primarily based upon the available literature.

We considered the following baseline variables as potential confounders: age,^2,50 race and ethnicity (non-Hispanic White, non-Hispanic Black, and other, including American Indian or Alaska Native, Asian, Native Hawaiian or Other Pacific Islander, and all those of Hispanic ethnicity regardless of race),^1,51 educational attainment (high school or less, some college or bachelor’s degree, and graduate work),^52–54 household annual income ($<\$\mathrm{50,000}$, $50,000–$100,000, and $>\$\mathrm{100,000}$),^53,55 body mass index (BMI; in kilograms per meter squared),^56,57 smoking status (never smoker, former smoker, and current smoker),^58,59 U.S. Census region (Northeast, Midwest, South, and West),⁶⁰ residential area type (rural, small town, suburban, urban, and others),⁶¹ and self-reported health status (excellent, very good or good, and fair or poor). We further considered two additional variables given that they might have affected the accuracy of the baseline exposure assessment: time spent living at the enrollment address ($<10$ y and $\ge 10$ y prior to Sister Study enrollment) and whether they had ever moved since enrollment. At each of the follow-up, study participants were asked whether they had moved since the previous follow-up survey, and we defined movers as those who ever moved from study enrollment to 15 March 2018.

To account for missing outcome data that was primarily due to nonparticipation in the substudy,⁶² we further considered the following variables from DFU-3 because they might have affected participation in the olfaction substudy: marital status (married or living as married vs. others),⁶² depression (score of $\ge 3$ in Patient Health Questionnaire-2⁶³ or use of antidepressant),⁶⁴ physical activity level (minutes per week of moderate/vigorous/walking time, in quartiles), and self-reported health status.

Statistical Analyses

Because study participants were sampled based on their self-reported olfaction status at the third follow-up survey and the B-SIT–tested olfaction was the outcome of interest in the substudy, we examined how the study design might affect our analytical plan. A valid statistical approach would be able to use the substudy sample to estimate the association of air pollutant with B-SIT–tested olfaction in the target population. First, under the assumption that the association of air pollutant with B-SIT–tested olfaction is not differential by the DFU-3 self-reported olfaction status (i.e., the sample selection stratum), the outcome regression model conditioning on self-reported olfaction generates a valid estimate for the conditional association of air pollutant with B-SIT–tested olfaction.^65–67 Further, with odds ratio (OR) as our measure of association, this conditional model is also valid as long as the B-SIT–tested olfaction and sample selection are mutually independent conditioning on air pollutant, self-reported olfaction status, and covariates.^68,69 Finally, if exposure to air pollutant is associated with self-reported olfaction, then the conditional model above may attenuate the association of exposure with B-SIT–tested olfaction. We therefore examined whether exposure to an air pollutant was related to the self-reported olfaction at DFU-3 (Table S1), and whether they interacted with each another in the regression on B-SIT–tested olfaction (Table S2) using multivariable logistic regression, adjusting for baseline age, race and ethnicity, education, smoking status, BMI, health status, census region, residential area type, and household income. In Table S2, we further adjusted for moving since enrollment and marital status to account for nonparticipation of the olfaction substudy. These analyses suggested that we could obtain valid results on the association of air pollutant with B-SIT–tested olfaction conditioning on self-reported olfaction.

For the exposures of interest, namely, ${\mathrm{PM}}_{2.5}$ and ${\mathrm{NO}}_2$, we assumed missing values to be missing completely at random. For the outcome of interest, we assumed missing at random and performed a logistic regression for the missingness of B-SIT–tested olfaction on covariates. Assuming one of the candidate missingness models is true, we adopted the more parsimonious method, that is, the best subset method based on the Bayesian information criterion, for covariate selection.⁷⁰ After the model selection, depression and physical activity were removed. Because few participants had missing values on the covariates, we simply combined them with the reference group in the analysis.

In the descriptive analysis, we presented weighted mean and percentage to extrapolate the data to all eligible participants of the third follow-up, using each individual’s weight calculated as the product of the sampling weight and inverse of predicted probability of not being missing on outcome. In the primary analyses, we examined air pollutant levels in 2006 in relation to B-SIT–tested poor olfaction in 2018–2019 using multivariable logistic regression adjusted for three different sets of covariates. Model A adjusted for baseline age, race and ethnicity, and educational attainment. Model B additionally adjusted for smoking status, BMI, self-reported health status, census region, residential area type, household income, moving status since enrollment, and marital status. Model C further added self-reported sense of smell to model B. We primarily used model C results to interpret study findings. In addition, we conducted multiple sensitivity analyses. First, we examined whether residency stability might have affected study results by limiting analyses to study participants who had lived at their baseline residence for at least 10 y before study enrollment or to those who did not move since enrollment. Second, we repeated the primary analyses using the latest exposure estimates available (2017) and the averaged exposure over the assessment period (2006–2017). Third, as the B-SIT cutoff was somewhat arbitrary, we repeated the primary analyses using another commonly used cutoff of 8^71,72 and further analyzed the B-SIT test score as an ordinal variable ($\ge 10$, 8–9, or $\le 7$). Finally, we conducted interaction analysis using the CONTRAST statement with the LOGISTIC procedure in SAS, adjusting for age, race, education, census region, residential area type, BMI, health status, smoking status, household income, self-reported sense of smell, moving status, and marital status. We present interaction analysis results by baseline median age (54.2 y), race, and smoking status. All analyses were based on Sister Study data release 9.1 and performed using SAS (version 9.4; SAS Institute Inc.) with two-sided tests with a significance level of 0.05.

Results

Table 1 presents features of the substudy samples with means and percentages weighed back to all eligible participants of the Sister Study. Compared with women with a normal B-SIT test score, those who failed the test were older and they were more likely to report non-Hispanic Black race, lower educational level, residency in the South census region, fair-to-poor health, and lower household income. Although the levels of both ${\mathrm{PM}}_{2.5}$ and ${\mathrm{NO}}_2$ decreased substantially from 2006 to 2017, at both time points, the exposure levels were comparable between women with and without poor olfaction.

Table 1.

Participant characteristics of the Sister Study’s olfaction substudy, by objectively tested olfaction status.

Covariatesa	Olfaction status by the Brief Smell Identification Test (B-SIT)
	Normal (B-SIT $\text{score}>9$, $n=\mathrm{2,344}$)b		Poor (B-SIT $\text{score}\le 9$, $n=\mathrm{1,001}$)b
	$N$ in substudy	Weighted statisticsc	$N$ in substudy	Weighted statisticsc
Age in 2006 [y ($\text{mean}\pm \text{SD}$)]	—	$54.3\pm 7.4$	—	$58.4\pm 6.9$
Race and ethnicity [$N$ (%)]
Non-Hispanic White	2,110	88.3	884	81.7
Non-Hispanic Black	108	5.8	76	13.8
Others	126	5.9	40	4.5
Missing	0	/	1	/
Education [$N$ (%)]
High school or below	258	10.2	156	16.3
Some college or bachelor’s degree	1,472	61.3	558	59.7
Graduate work	613	28.4	286	23.9
Missing	1	/	1	/
Census Region [$N$ (%)]
South	741	32.9	331	38.0
Midwest	666	26.5	268	21.5
West	527	22.5	235	22.8
Northeast	407	18.0	165	17.6
Missing	3	/	2	/
Residential Area Type [$N$ (%)]
Suburban	901	39.5	377	34.0
Small town	507	21.9	213	21.3
Rural	515	20.5	224	23.0
Urban	413	17.9	182	21.6
Other	3	0.1	2	0.1
Missing	5	/	3	/
BMI [${\mathrm{kg}/\mathrm{m}}^2$ ($\text{mean}\pm \text{SD}$)]	—	$27.1\pm 5.8$	—	$27.9\pm 6.5$
Health status [$N$ (%)]
Excellent, very good or good	2,223	95.8	921	91.5
Fair or poor	120	4.1	79	8.5
Missing	1	/	1	/
Smoking status [$N$ (%)]
Never smoker	1,323	56.3	534	54.2
Former smoker	854	35.7	384	35.9
Current smoker	167	8.0	82	9.9
Missing	0	/	1	/
Household income [$N$ (%)]
$<\$\mathrm{50,000}$	464	19.0	237	25.5
$50,000–$100,000	955	41.7	455	47.9
$>\$\mathrm{100,000}$	851	36.3	281	22.6
Missing	74	/	28	/
${\mathrm{PM}}_{2.5}$ (${\mathrm{\mu }\mathrm{g}/\mathrm{m}}^3$) [$\text{mean}\pm \text{SD}$ (IQR)]
2006	—	$10.1\pm 2.5$ (3.5)	—	$10.4\pm 2.7$ (3.5)
2017	—	$6.2\pm 1.5$ (1.9)	—	$6.3\pm 1.5$ (1.8)
2006–2017 average	—	$8.2\pm 1.8$ (2.4)	—	$8.4\pm 1.9$ (2.7)
${\mathrm{NO}}_2$ (ppb) [$\text{mean}\pm \text{SD}$ (IQR)]
2006	—	$9.3\pm 4.9$ (6.0)	—	$9.2\pm 5.5$ (5.9)
2017	—	$6.1\pm 3.3$ (3.4)	—	$6.1\pm 3.7$ (3.8)
2006–2017 average	—	$7.5\pm 3.8$ (4.5)	—	$7.5\pm 4.4$ (4.3)

Note: —, not applicable; /, calculation not needed; BMI, body mass index; IQR, interquartile range; ${\mathrm{NO}}_2$, nitrogen dioxide; ${\mathrm{PM}}_{2.5}$, fine particulate matter (PM $\le 2.5\mathrm{\mu }\mathrm{m}$ in aerodynamic diameter); ppb, parts per billion; SD, standard deviation.

^a Covariates were assessed at Sister Study enrollment (2003–2009) unless otherwise specified, and the olfaction test was conducted in 2018–2019.

^b Missing in covariates are only presented with frequency count with no weighted statistics.

^c Weighted percentage for categorical variables and weighted $\text{mean}\pm \text{SD}$ for continuous variables are presented.

In the primary analysis, neither baseline exposure to ${\mathrm{PM}}_{2.5}$ or ${\mathrm{NO}}_2$ was significantly associated with B-SIT–tested poor olfaction (Table 2). The adjustment of different sets of covariates barely changed the results. In the fully adjusted model, compared with the lowest quartile, the ORs and 95% confidence intervals (CIs) for the highest quartile were 1.07 (95% CI: 0.82, 1.40) for ${\mathrm{PM}}_{2.5}$ and 1.18 (95% CI: 0.88, 1.58) for ${\mathrm{NO}}_2$. For each IQR increment, the estimates were 1.03 (95% CI: 0.91, 1.17) and 1.08 (95% CI: 0.96, 1.22), respectively. Sensitivity analyses limited to participants with a stable residency at enrollment or during the follow-up showed similar results (Table 3). Further, we did not find significant associations of poor olfaction with the most recent exposure levels in 2017 or the cumulative averages between 2006 and 2017 (Table 4). Similar results were observed in the analyses using a different B-SIT cutoff (Table S3) or B-SIT as an ordinal outcome (Table S4).

Table 2.

Baseline exposures (2006) to air pollutants and objectively tested olfaction (2018–2019) in the Sister Study’s olfaction substudy.

Exposures	Normal/poor olfactiona	OR (95% CI)b
		Model A	Model B	Model C
${\mathrm{PM}}_{2.5}$ [${\mathrm{\mu }\mathrm{g}/\mathrm{m}}^3$ (min, max)]
Quartile
Q 1 (2.60, 8.52)	604/248	Ref	Ref	Ref
Q 2 (8.53, 10.47)	602/258	1.02 (0.82, 1.26)	1.04 (0.82, 1.32)	0.99 (0.77, 1.26)
Q 3 (10.48, 11.84)	576/242	1.06 (0.85, 1.32)	1.10 (0.85, 1.43)	1.02 (0.78, 1.33)
Q 4 (11.84, 19.91)	551/249	1.06 (0.85, 1.32)	1.09 (0.84, 1.40)	1.07 (0.82, 1.40)
Per IQR ($3.3{\mathrm{\mu }\mathrm{g}/\mathrm{m}}^3$)	—	1.03 (0.93, 1.14)	1.04 (0.92, 1.18)	1.03 (0.91, 1.17)
${\mathrm{NO}}_2$ [ppb (min, max)]
Quartile
Q 1 (1.18, 5.64)	612/250	Ref	Ref	Ref
Q 2 (5.64, 8.14)	559/259	1.15 (0.92, 1.42)	1.17 (0.93, 1.48)	1.18 (0.93, 1.50)
Q 3 (8.14, 11.38)	604/241	0.98 (0.79, 1.22)	1.01 (0.78, 1.31)	1.02 (0.78, 1.34)
Q 4 (11.38, 41.32)	558/247	1.07 (0.86, 1.33)	1.10 (0.83, 1.46)	1.18 (0.88, 1.58)
Per IQR (5.7 ppb)	—	1.02 (0.94, 1.12)	1.05 (0.94, 1.17)	1.08 (0.96, 1.22)

Note: CI, confidence interval; IQR, interquartile range; ${\mathrm{NO}}_2$, nitrogen dioxide; OR, odds ratio; ${\mathrm{PM}}_{2.5}$, fine particulate matter (PM $\le 2.5\mathrm{\mu }\mathrm{m}$ in aerodynamic diameter); ppb, parts per billion; Q, quartile; Ref, reference.

^a Analysis was based on 3,330 participants with valid Brief Smell Identification Test score and non-missing baseline ${\mathrm{PM}}_{2.5}$ or ${\mathrm{NO}}_2$ measures.

^b Model A adjusted for baseline age, race and ethnicity, and education; model B additionally adjusted for smoking status, body mass index, health status, census region, residential area type, household income, moving status since enrollment, and marital status; model C additionally adjusted for self-reported sense of smell.

Table 3.

Baseline (2006) exposures to air pollutants and objectively tested olfaction (2018–2019) among participants with a stable residential history.

Exposures	Living $\ge 10$ y at enrollment addressa		Not moving since enrollmentb
	Normal/poor olfaction	OR (95% CI)c	Normal/poor olfaction	OR (95% CI)c
${\mathrm{PM}}_{2.5}$ [${\mathrm{\mu }\mathrm{g}/\mathrm{m}}^3$ (min, max)]
Quartile
Q 1 (2.60, 8.52)	317/122	Ref	355/144	Ref
Q 2 (8.53, 10.47)	341/134	0.95 (0.67, 1.34)	355/165	1.12 (0.81, 1.55)
Q 3 (10.48, 11.84)	342/154	1.14 (0.79, 1.65)	366/158	1.09 (0.77, 1.54)
Q 4 (11.84, 19.91)	303/149	1.23 (0.84, 1.78)	321/157	1.19 (0.83, 1.69)
Per IQR ($3.3{\mathrm{\mu }\mathrm{g}/\mathrm{m}}^3$)	—	1.07 (0.90, 1.28)	—	1.02 (0.86, 1.21)
${\mathrm{NO}}_2$, [ppb (min, max)]
Quartile
Q 1 (1.18, 5.64)	348/149	Ref	375/156	Ref
Q 2 (5.64, 8.14)	297/135	1.18 (0.85, 1.64)	333/165	1.29 (0.95, 1.75)
Q 3 (8.14, 11.38)	344/136	1.06 (0.73, 1.53)	362/152	1.11 (0.78, 1.58)
Q 4 (11.38, 41.32)	314/139	1.19 (0.80, 1.77)	327/151	1.24 (0.84, 1.82)
Per IQR (5.7 ppb)	—	1.16 (0.99, 1.37)	—	1.10 (0.94, 1.28)

Note: —, not applicable; CI, confidence interval; IQR, interquartile range; ${\mathrm{NO}}_2$, nitrogen dioxide; OR, odds ratio; ${\mathrm{PM}}_{2.5}$, fine particulate matter (PM $\le 2.5\mathrm{\mu }\mathrm{m}$ in aerodynamic diameter); ppb, parts per billion; Q, quartile; Ref, reference.

^a Limited to participants who had lived at the enrollment address for at least 10 y ($n=\mathrm{1,862}$).

^b Limited to participants who did not move since enrollment ($n=\mathrm{2,021}$).

^c Adjusted for baseline age, race and ethnicity, education, smoking status, body mass index, health status, census region, residential area type, household income, marital status, and self-reported sense of smell.

Table 4.

Most recent (2017) and averaged air pollutant levels (2006–2017) in relation to objectively tested olfaction (2018–2019) in the substudy.

Exposures	2017 exposure estimatesa		2006–2017 averaged exposure estimatesb
	Normal/poor olfaction	OR (95% CI)c	Normal/poor olfaction	OR (95% CI)c
${\mathrm{PM}}_{2.5}$
Quartile
Q 1	623/228	Ref	580/239	Ref
Q 2	581/269	1.22 (0.96, 1.56)	612/251	0.96 (0.75, 1.22)
Q 3	562/253	1.24 (0.96, 1.60)	613/242	0.83 (0.64, 1.08)
Q 4	563/239	1.08 (0.83, 1.40)	539/269	1.24 (0.95, 1.62)
Per IQR	—	1.00 (0.89, 1.13)	—	1.09 (0.96, 1.25)
${\mathrm{NO}}_2$
Quartile
Q 1	577/238	Ref	605/255	Ref
Q 2	623/256	1.04 (0.83, 1.31)	592/247	0.99 (0.78, 1.26)
Q 3	577/258	1.13 (0.88, 1.45)	600/260	1.12 (0.87, 1.46)
Q 4	552/237	0.99 (0.76, 1.30)	547/239	1.10 (0.83, 1.46)
Per IQR	—	1.03 (0.93, 1.13)	—	1.07 (0.96, 1.19)

Note: For the most recent (2017) exposure: IQR for ${\mathrm{PM}}_{2.5}=1.8{\mathrm{\mu }\mathrm{g}/\mathrm{m}}^3$, range for ${\mathrm{PM}}_{2.5}$: 1.51–5.43 for Q1, 5.44–6.44 for Q2, 6.44–7.19 for Q3, and 7.19–12.91 for Q4 and IQR for ${\mathrm{NO}}_2=3.6\text{\hspace{0.17em}ppb}$, range for ${\mathrm{NO}}_2$: 0.95–3.82 for Q1, 3.83–5.34 for Q2, 5.34–7.43 for Q3, and 7.45–37.66 for Q4; for the averaged exposure 2006–2017: IQR for ${\mathrm{PM}}_{2.5}=2.4{\mathrm{\mu }\mathrm{g}/\mathrm{m}}^3$, range for ${\mathrm{PM}}_{2.5}$: 2.49–7.13 for Q1, 7.14–8.57 for Q2, 8.57–9.53 for Q3, and 9.53–17.48 for Q4 and IQR for ${\mathrm{NO}}_2=4.3\text{\hspace{0.17em}ppb}$, range for ${\mathrm{NO}}_2$: 1.22–4.75 for Q1, 4.75–6.56 for Q2, 6.57–9.06 for Q3, and 9.07–37.37 for Q4. —, not applicable; CI, confidence interval; IQR, interquartile range; ${\mathrm{NO}}_2$, nitrogen dioxide; OR, odds ratio; ${\mathrm{PM}}_{2.5}$, fine particulate matter (PM $\le 2.5\mathrm{\mu }\mathrm{m}$ in aerodynamic diameter); Q, quartile; Ref, reference.

^a Limited to participants without missing values in outcome and exposures at 2017 ($n=\mathrm{3,318}$).

^b Limited to participants without missing values in outcome and with at least one non-missing exposure from 2006 to 2017 ($n=\mathrm{3,345}$).

^c Adjusted for baseline age, race and ethnicity, education, smoking status, body mass index, health status, census region, residential area type, household income, moving status since enrollment, marital status, and self-reported sense of smell.

In the exploratory interaction analyses (Figures 1 and 2), data suggested positive associations of baseline air pollutants with poor olfaction among younger women ($<54.2$ years of age), and the finding was statistically significant for ${\mathrm{PM}}_{2.5}$ (${\text{OR}}_{\text{per}\text{\hspace{0.17em}IQR}}=1.35$; 95% CI: 1.11, 1.65). In addition, baseline ${\mathrm{NO}}_2$ was statistically associated with poor olfaction among current smokers (${\text{OR}}_{\text{per}\text{\hspace{0.17em}IQR}}=1.87$; 95% CI: 1.29, 2.71). Similar results were obtained in the analyses using cumulative exposure between 2006 and 2017 (Figure S1) or using $\le 8$ as the cutoff for B-SIT (Figure S2).

Figure 1.

Baseline ${\mathrm{PM}}_{2.5}$ exposure in 2006 and objectively tested poor olfaction in 2018–2019, interaction analyses ($N=\mathrm{3,330}$). ORs and 95% CIs were derived from logistic regression models, adjusting for baseline age, race and ethnicity, education, smoking status, body mass index, health status, census region, residential area type, household income, moving status since enrollment, marital status, and self-reported sense of smell. Poor olfaction was defined as a B-SIT score of $\le 9$. OR and 95% CI estimates were for each interquartile range increment in ${\mathrm{PM}}_{2.5}$ level ($3.3{\mathrm{\mu }\mathrm{g}/\mathrm{m}}^3$). Note: B-SIT, Brief Smell Identification Test; CI, confidence interval; LCL, lower confidence limit for each OR estimate; ${\mathrm{PM}}_{2.5}$, fine particulate matter (PM $\le 2.5\mathrm{\mu }\mathrm{m}$ in aerodynamic diameter); OR, odds ratio; UCL, upper confidence limit for each OR estimate.

Figure 2.

Baseline ${\mathrm{NO}}_2$ exposure in 2006 and objectively tested poor olfaction in 2018–2019, interaction analyses ($N=\mathrm{3,330}$). ORs and 95% CIs were derived from logistic regression models, adjusting for baseline age, race and ethnicity, education, smoking status, body mass index, health status, census region, residential area type, household income, moving status since enrollment, marital status, and self-reported sense of smell. Poor olfaction was defined as a B-SIT score of $\le 9$. OR and 95% CI estimates were for each interquartile range increment in ${\mathrm{NO}}_2$ level (5.7 ppb). Note: B-SIT, Brief Smell Identification Test; CI, confidence interval; LCL, lower confidence limit for each OR estimate; ${\mathrm{NO}}_2$, nitrogen dioxide; OR, odds ratio; UCL, upper confidence limit for each OR estimate.

Discussion

In this nationwide cohort, we did not find convincing evidence that ambient air pollutants of ${\mathrm{PM}}_{2.5}$ or ${\mathrm{NO}}_2$ had a lasting detrimental effect on the olfaction of middle-aged to older women. However, we could not exclude the possibility that ${\mathrm{PM}}_{2.5}$ or ${\mathrm{NO}}_2$ may be associated with poor olfaction among relatively younger women and active smokers. These preliminary findings warrant further investigations. Further, our study findings may not be readily generalizable to participants with other sex or racial and ethnic backgrounds because our study population was composed predominantly of non-Hispanic White women, who are less likely to have poor olfaction than men or Black women.^1,2

As justified in the introduction, poor olfaction may have profound implications on the health of older adults,^15–20 and there is a need to understand the potential environmental contributors to poor olfaction, especially air pollutants.^25,26 However, the empirical evidence on air pollutants and olfactory loss is limited and preliminary. Earlier human studies compared olfactory performance of residents in highly polluted cities, such as Mexico City, with that of residents in surrounding areas of cleaner air.^{26,30,73–76} However, these studies are ecological in design, small in sample size, and often enrolled children and young or middle-aged adults. Because the prevalence of olfactory impairment is age dependent, it is much more interesting to investigate the roles of air pollutants in the development of poor olfaction in older adults.

This topic has increasingly gained attention with five analytical epidemiological studies published since 2016 (Table S5). These studies included older adults, but they had different designs, population characteristics, definitions of exposures and outcomes, and study findings. The National Social Life, Health, and Aging Project (NSHAP, $n\approx \mathrm{2,000}$) published two cross-sectional analyses in 2016. Ajmani et al.³² reported that each ${\mathrm{PM}}_{2.5}$ IQR increment in the past 6 months was significantly associated with 28% higher odds of poor olfaction, defined as a score of $\le 3$ of the five-item Sniffin Stick odor identification test (SS-OIT), and the association was stronger in younger participants (57–64 years of age at test, compared with those 65–74 or 75–85 years of age), women, Black participants, and noncurrent smokers. In the NSHAP analysis of ${\mathrm{NO}}_2$, Adams et al.³³ reported 35% higher odds of having poor olfaction for each ${\mathrm{NO}}_2$ IQR increment in the past 365-d period and reported no significant interactions between ${\mathrm{NO}}_2$ exposure and age, sex, race or smoking on poor olfaction.

Two more studies were published in 2021. In a Swedish study,³⁵ short-term exposure to ${\mathrm{PM}}_{2.5}$ was not associated with either odor identification or the threshold test results in their cross-sectional analyses. However, they found that long-term exposure to ${\mathrm{PM}}_{2.5}$ was associated with better odor identification performance. The other study was a hospital-based case–control study with 538 clinically diagnosed anosmia patients and 2,152 matched controls. Zhang et al.³⁴ reported a significant 73% higher odds of anosmia per $5{\text{-}\mathrm{\mu }\mathrm{g}/\mathrm{m}}^3$ increase in 12-month averaged ${\mathrm{PM}}_{2.5}$ exposure. The most recent study was published in 2022, with longitudinal findings from the Swedish National Study on Aging and Care in Kungsholmen.³⁶ The authors reported that both higher ${\mathrm{PM}}_{2.5}$ and ${\mathrm{NO}}_{\mathrm{x}}$ exposures were associated with a faster declining rate of olfaction as assessed using the 16-item SS-OIT. Taken together, only a few epidemiological studies have investigated the potential role of the air pollutants of ${\mathrm{PM}}_{2.5}$ and ${\mathrm{NO}}_2$ in olfactory impairment in older adults.

Compared with prior studies, our study is modestly larger but included only women. Like the NSHAP, our study population is nationwide, enabling examinations of ambient air pollution beyond limited geographic regions. The olfaction test we used was comparable with that used in most others and was more intensive than the five-item test in NSHAP. Our exposure assessment was rigorous, based on state-of-the-science exposure assessment modeling.⁴⁷ Unlike prior studies that assessed current or recent exposures to air pollutants, our exposure assessments covered a period of more than a decade, befitting our aim to investigate the role of long-term air pollution in poor olfaction in the context of aging. Further, our findings were mostly consistent in the analyses of using air pollutant exposures at baseline (2006), the most recent exposures (2017), or the cumulative average exposures (2006–2017). We further accounted for moving status and residential stability in the analysis, which enabled more accurate evaluations of long-term exposures of air pollutants in relation to health outcomes. Finally, we analyzed the outcome using different cutoffs and as an ordinal variable, and we conducted multiple sensitivity analyses to assess finding robustness.

Our primary analysis did not reveal statistically significant associations of these air pollutants with poor olfaction in middle-aged to older women. The estimates were in the same direction as most of other studies but appeared to be weaker. One potential reason is that our study mostly comprised non-Hispanic White women, who were much less likely to have poor olfaction than their male counterparts or Black women,^1,2 raising the possibility that they were either less exposed to risk factors or were more resistance to their potential adverse effects. Further, our study focused on the potential long-term effects of air pollution vs. most prior studies, which evaluated cross-sectional associations or short-term effects. Finally, the ambient air pollutant levels in our study were also lower than that of the NSHAP, but they were comparable or higher than some of the other studies.^34–36

Our subgroup analyses are exploratory, but some findings may be interesting. The associations with ${\mathrm{PM}}_{2.5}$ and ${\mathrm{NO}}_2$ both appear to be stronger in younger women and in active smokers. Similar findings were observed when we used different outcome definitions or cumulative average of air pollutants over the observational period. These findings, although preliminary, are consistent with some of the NSHAP findings that the association of ${\mathrm{PM}}_{2.5}$ and poor olfaction was found only among younger participants. This more evident association among relatively young participants may relate to the possibility that the mild effect of air pollution on olfaction might be overwhelmed by the strong adverse effects of aging in older adults.³² Our subgroup findings on air pollutants and poor olfaction by smoking was inconsistent with that of NSHAP, but potential synergetic effects of ${\mathrm{PM}}_{2.5}$ and smoking have been previously reported for other outcomes, such as lung cancer mortality⁷⁷ and cardiovascular mortality.⁷⁸ However, it must be noted that our subgroup analyses were largely exploratory and based on smaller samples, and findings should be interpreted with caution.

A limitation of our study is that participants are predominantly well-educated White women volunteers. Compared with men or Black women, they are least likely to have poor olfaction,¹ and therefore our findings may not be generalizable to men or Black women. Second, like most of the other studies, poor olfaction was defined based on a single odor identification test. This approach, although highly feasible in large epidemiological studies, is not ideal for assessing an outcome that has multiple dimensions (e.g., identification, threshold, and discrimination). Further, olfaction was assessed only once $\sim 12$ y after the exposure assessment, making it impossible to assess the potential role of air pollutants in olfactory loss and to explore their temporal relationships in the context of aging. More comprehensive and repeated olfaction assessments will be needed to evaluate the role of air pollutants or any other risk factors in olfactory loss in the context of aging. Third, the B-SIT test was conducted $\sim 3$ y after the self-reported sense of smell in DFU-3, based on which our samples were drawn. During these 3 y, participants’ olfactory status might have changed. Finally, our study aimed to investigate the potential long-term effect of air pollutants in the context of aging. We did not have concurrent air pollutant data to assess potential short-term effects of air pollutants on olfaction.

In conclusion, this study did not find convincing evidence that air pollutants have lasting detrimental effects on the sense of smell of middle-aged to older women. The subgroup findings were exploratory and need independent confirmation. Future studies should include men and women of more diverse racial and ethnic backgrounds and assess olfaction longitudinally at multiple time points over the years.