Association of Olfactory Impairment With All-Cause Mortality

Key Points

Question

Is olfactory impairment associated with mortality?

Findings

In this multiadjusted observational systematic review and meta-analysis of 11 studies including 21 601 participants, with moderate-quality evidence overall, olfactory impairment was associated with a 52% higher risk of all-cause mortality.

Meaning

This study suggests that olfactory impairment may be an important risk marker of general health and aging; further research is required to uncover the potential underlying causal or confounding relationships.

Abstract

Importance

Olfactory impairment is highly prevalent and associated with multiple comorbidities, including neurodegenerative, cardiovascular, nutritional, and immune disorders. However, epidemiologic associations between olfactory impairment and mortality are discordant.

Objective

To systematically clarify the epidemiologic associations between olfactory impairment and mortality.

Data Sources

The PubMed, Embase, and Cochrane Library databases were searched from inception to August 13, 2021.

Study Selection

Two blinded reviewers selected observational studies published as full-length, English-language articles in peer-reviewed journals that reported the presence or severity of chronic olfactory impairment, whether objectively measured or self-reported, in association with any mortality estimate, among adults aged 18 years or older.

Data Extraction and Synthesis

Two reviewers independently extracted data, evaluated study bias using the Newcastle-Ottawa Scale, and appraised the quality of the evidence using the Grading of Recommendations Assessment, Development and Evaluation framework, following Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) and Meta-analysis of Observational Studies in Epidemiology (MOOSE) guidelines and a PROSPERO-registered protocol. Maximally adjusted estimates were pooled using mixed-effects models, heterogeneity was measured using I² statistics, sources of heterogeneity were investigated using meta-regression and subgroup meta-analyses, and publication bias was qualitatively and quantitatively assessed.

Main Outcomes and Measures

Hazard ratios for all-cause mortality.

Results

One retrospective cohort study and 10 prospective cohort studies (with a total of 21 601 participants) from 1088 nonduplicated records were included. Ten studies had a low risk of bias, whereas 1 study had a moderate risk; exclusion of the latter did not alter conclusions. Nine studies were included in the meta-analysis. Olfactory loss was associated with a significantly higher pooled hazard of all-cause mortality (hazard ratio, 1.52; 95% CI, 1.28-1.80; I² = 82%). Meta-regression sufficiently explained heterogeneity, with longer mean follow-up duration weakening the pooled association, accounting for 91.3% of heterogeneity. Self-reported and objective effect sizes were similar. Associations were robust to trim-and-fill adjustment and the Egger test for publication bias. The overall quality of evidence was moderate.

Conclusions and Relevance

The findings of this systematic review and meta-analysis suggest that olfactory impairment is associated with all-cause mortality and may be a marker of general health and biological aging. Further research is required to establish the underlying mechanisms and the scope for interventions.

This systematic review and meta-analysis investigates the epidemiologic associations between olfactory impairment and mortality.

Introduction

Population-based studies of olfactory impairment (OI) indicate an overall prevalence ranging from 19% to 24%,^1,2,3 with a prevalence exceeding 50% for individuals aged 65 to 80 years, and reaching 80% among those older than 80 years.⁴ Olfactory impairment is also associated with multiple comorbidities: neurogenerative,^5,6,7 cardiovascular,⁸ nutritional,⁹ and immune disorders.^10,11 Although OI is common, it frequently goes unnoticed compared with other senses, such as hearing or vision.^3,12 Because olfaction is used for a variety of purposes and is a critical component of human physiology,¹³ its impairment may result in appetite suppression and malnutrition,⁹ decreased safety¹⁴ owing to increased risk of hazardous events (eg, cooking-related incidents, ingestion of spoiled food, and accidental gas poisoning), and overall lower quality of life.¹⁵

Furthermore, there has been an increasing body of evidence suggesting that OI is an early sign of neurodegenerative disorders, such as Alzheimer disease,^5,6,7 Parkinson disease,^16,17,18 and cognitive decline,^7,19,20 all of which are associated with greater mortality risk. Despite this evidence, 1 study suggests that only one-third of poor olfaction–associated mortality could be explained by dementia or Parkinson disease,²¹ which suggests that this impairment may be associated with a broader range of clinical diseases,²² diminishing cell regeneration,²³ and age-related accumulation of environmental exposures.²⁴ Olfactory impairment is also associated with other comorbidities, including diabetes,²⁵ kidney disease,²⁶ and epilepsy.²⁷

Emerging evidence suggests that OI is independently associated with excess all-cause mortality among older adults.^28,29,30,31 However, there is inconsistency in the literature, with at least 2 studies reporting no association between OI and all-cause mortality among older adults.^32,33 Hence, we aim to systematically clarify this association and the role of potential confounders, such as demographic characteristics, socioeconomic status, and medical comorbidities. Given the growing burden of OI globally, we believe that this review is both timely and relevant and should yield public health implications and benefit patient care.

Methods

This study’s protocol is registered on PROSPERO (CRD42021279531) and reported in accordance with the Meta-analysis of Observational Studies in Epidemiology (MOOSE) reporting guideline and the Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) reporting guideline.^34,35 The MOOSE checklist is included in eTable 1 in the Supplement.

Search Strategy

We searched PubMed, Embase, and the Cochrane Library databases from inception to August 13, 2021, using search terms for OI and mortality (eMethods in the Supplement). We also hand-searched the bibliographies of included articles and relevant reviews but identified no additional relevant records.

Study Selection, Data Extraction, and Risk-of-Bias Grading

Two of us (N.Y.-L.P. and A.S.R.C.) independently selected eligible studies (based on title and abstract, followed by full-text articles), extracted relevant data, and evaluated the risk of bias in a blinded manner, with conflicts resolved by a third author (J.X.T.). We included observational studies, published as full-length articles in peer-reviewed journals, that reported the associations between OI and mortality, with any estimate of mortality (eg, all-cause mortality, cardiovascular mortality, or mortality due to injury) among adults aged 18 years or older compared with participants without OI. We accepted objective (eg, the 12-item Brief Smell Identification Test [B-SIT]) and subjective self-reported measures of OI. To avoid duplication, the selected studies focused on the burden of chronic OI before the COVID-19 pandemic, instead of COVID-19–related OI, which occurs in a more acute setting.³⁶ We excluded case reports, reviews, letters, conference abstracts, other non–full-length articles, and non–English-language publications. We extracted key data (eMethods in the Supplement) from each included article and assessed the risk of bias using the Newcastle-Ottawa Scale.³⁷ Per the grading in the past, studies were assessed as having either a high (<5 stars), moderate (5-7 stars), or low (≥8 stars) risk of bias (eTable 2 in the Supplement).

Statistical Analysis

We used mixed-effects models (eMethods in the Supplement) to pool maximally covariate-adjusted hazard ratios (HRs) from each study. Given that HRs, odds ratios, and risk ratios numerically approximate one another when follow-up duration, mean rate of event, and magnitude of risk are low,³⁸ we pooled these maximally adjusted ratios as an overall HR if the aforementioned conditions were met. We assessed and considered between-study heterogeneity as significant if the P value of the Q test was less than .10 or if the I² statistic was 50% or greater. If a study used an analytical method that is incompatible for synthesis, we converted the effect estimate to an appropriate ratio for synthesis or excluded the study from meta-analysis.

To investigate potential sources of heterogeneity, we performed subgroup, sensitivity, and meta-regression analyses of the following prespecified study-level characteristics: method of measuring OI (eg, objective vs subjective or self-reported, or by the specific smell test), mortality types (eg, all cause and cardiovascular), age, sex, duration of follow-up, prevalence of OI, and adjustment for covariates (eg, presence of cognitive impairment). To investigate small-study effects, we assessed funnel plot asymmetry both visually and using the Egger bias test, imputing potentially missing studies using the trim-and-fill method (eMethods in the Supplement) if publication bias was suspected.

We conducted all analyses using R studio, version 1.3 (R Group for Statistical Computing) (eMethods in the Supplement). Unless otherwise specified, we considered a 2-sided P < .05 as statistically significant.

Certainty of Evidence

We evaluated the quality of pooled evidence at the outcome level (eTable 3 in the Supplement) using the Grading of Recommendations Assessment, Development and Evaluation (GRADE) framework.³⁹

Results

The study selection is summarized in Figure 1. We included 11 articles from 1088 nonduplicated records after initial selection based on title and abstract and subsequent selection based on full texts.^{21,22,28,29,30,31,32,33,40,41,42}

Figure 1. PRISMA Flow Diagram of Study Selection Process.

CVD indicates cardiovascular disease; IHD, ischemic heart disease; and TBI, traumatic brain injury.

Study Characteristics

Of the 11 included studies (Table),^{21,22,28,29,30,31,32,33,40,41,42} all were observational studies, with 1 retrospective cohort study and 10 prospective cohort studies. Risk of bias when assessed using the Newcastle-Ottawa Scale ranged from low (10 studies)^{21,22,28,29,30,31,32,33,40,42} to moderate (1 study).⁴¹ Sensitivity analyses (excluding the moderate risk-of-bias study) did not substantially alter our results. A total of 7 studies were conducted in North America,^{21,22,28,30,31,33,41} 2 in Europe,^29,40 1 in Australia,³² and 1 in China.⁴² The mean age of participants ranged from 59 to 81 years. We included 9 studies in our meta-analyses.^{21,22,28,29,30,32,33,40,42}

Table. Summary of Included Studies.

Study	Country	Study design	Sample size, No.	Mean age, y	Male patients, %	Median follow-up duration, y	Impairment definition	Impairment prevalence, %	Covariates	NOS (maximum 9)
Choi et al,33 2021 NHANES	US	Retrospective cohort study	3503	59.0	47.7	5	NHANES Pocket Smell Test (Sensonics International) OI defined as score of ≤5	17.9	Age, sex, race and ethnicity, income, educational level, hypertension, cardiovascular diseases, diabetes, stroke, smoking, recent cold symptoms, previous sinus infection, previous head injury, nasal or facial fracture, cognition, major depressive disorder	9
Devanand et al,28 2015 Washington Heights–Inwoods Columbia Aging Project	US	Prospective cohort study	1169	81.0	30.7	4.1	University of Pennsylvania Smell Identification Test OI stratified by quartiles and analyzed as a continuous variable	NA	Age, education, race and ethnicity, sex, language, comorbidity index, dementia, depression, head injury, alcohol abuse, BMI, smoking, hearing impairment, vision impairment	9
Ekström et al,29 2017 Betula Study	Sweden	Prospective cohort study	1774	63.5	45.3	9.4	Scandinavian Odor-Identification Test OI defined as score ≤4	NA	Age, sex, years of education, history of heart disease, stroke, high blood pressure, diabetes, depression, cognitive performance, dementia, apolipoprotein E4 gene	9
Leschak and Eisenberger,41 2018 NSHAP	US	Prospective cohort study	2264	69.1	NA	5	Validated field olfaction test	NA	NA	6
Gopinath et al,32 2012 Blue Mountains Eye Study	Australia	Prospective cohort study	1636	73.5	41.9	5	San Diego Odor Identification Test Mild OI: score <6 and >3 Moderate OI: score ≤3	27	Age, sex, BMI, systolic blood pressure, current smoking status, alcohol consumption, poor self-rated health, visual impairment, presence of hypertension and/or diabetes, history of cancer, angina, stroke, and/or acute myocardial infarction, serum total cholesterol, cognitive impairment	8
Laudisio et al,40 2019 Invecchiare in Chianti	Italy	Prospective cohort study	1035	74.7	44.3	9	Self-reported recognition of mint, air, and coffee OI defined as:\ ≥2 smells not recognized	57	Age, sex, educational level, glomerular filtration rate, hemoglobin level, CES-D, Mini-Mental State Examination, ADL, diagnosis of malignant neoplasm, peripheral arterial disease, use of ACE inhibitors and benzodiazepines, frailty	8
Liu et al,21 2019 Health, Aging, and Body Composition Study	US	Prospective cohort study	2289	75.6	48.1	13	Brief Smell Identification Test Poor: score 0-8 Moderate: score 9-10	66.1	Age, sex, race and ethnicity, educational level, BMI, alcohol drinking, physical activity, smoking, self-reported health status, coronary heart disease, CHF, cerebrovascular disease, peripheral vascular disease, cancer, diabetes, hypertension, depressive symptoms, chronic kidney disease	9
Pinto et al,22 2014 NSHAP	US	Prospective cohort study	2918	68.0	48.4	5	Validated odor identification test presented using felt-tipped pens (Burghardt) Anosmia: 4-5 errors Hyposmia: 2-3 errors	26.5	Age, anosmia, hyposmia, sex, race and ethnicity, educational level, comorbidity index, myocardial infarction, heart failure, stroke, diabetes, hypertension, emphysema or COPD, liver damage, cancer	9
Schubert et al,30 2017 Epidemiology of Hearing Loss Study	US	Prospective cohort study	2418	68.9	42.2	12.8	San Diego Odor Identification Test OI defined as score of ≤6 (range, 0-12); 10th percentile	23.8	Age, sex, educational level, hypertension, diabetes, cardiovascular disease, cancer, cognitive impairment, frailty, smoking, exercise, BMI, alcohol, intima-media thickness, C-reactive protein, interleukin 6	9
Wilson et al,31 2011 Rush Memory and Aging Project	US	Prospective cohort study	1162	79.7	25.5	4.2	Brief Smell Identification Test OI defined as score of ≤6 (range, 0-12); 10th percentile	NA	Age, educational level, sex, Boston Naming Test (visual confrontation naming), Katz Scale for ADL, vascular risk factors (smoking, hypertension and diabetes), vascular conditions (myocardial infarction, CHF, stroke, and claudication), BMI, cognitive activity, social activity, physical activity, depressive symptoms, apolipoprotein E gene	8
Xiao et al,42 2021 Shanghai Aging Study	China	Prospective cohort study	1443	68.9	45.8	8.6	Sniffin’ Sticks Screening Test	53.6	Age, sex, failure to identify rose odor, ADL score, chronic kidney disease, serum LDL cholesterol, failure to identify fish odor, anemia, failure to identify coffee odor	9

Abbreviations: ACE, angiotensin-converting enzyme; ADL, activities of daily living; BMI, body mass index; CES-D, Center for Epidemiologic Studies–Depression; CHF, congestive heart failure; COPD, chronic obstructive pulmonary disorder; LDL, low-density lipoprotein; NA, not applicable; NHANES, National Health and Nutrition Examination Survey; NOS, Newcastle-Ottawa Scale; NSHAP, National Social Life, Health, and Aging Project; OI, olfactory impairment.

Association of OI With Mortality

Definition of OI

Ten of the included studies^{21,22,28,29,30,31,32,33,41,42} (Table) measured olfaction objectively via standardized tests, and 1 study⁴⁰ assessed olfaction via self-report. Of the 10 studies that assessed olfaction objectively, OI was defined as the inability to recognize 2 of 3 odors (mint, coffee, and air),⁴⁰ the inability to correctly identify more than 2 of 8 odors (2 studies),^30,32 the inability to correctly identify 9 or more of 13 odors (1 study),²⁹ and the inability to correctly identify 2 of 12 odors (2 studies).^21,31 Three studies further stratified their population based on the severity of the impairment.^22,28,42

Meta-analysis for All-Cause Mortality

Meta-analysis included 9 studies (Figure 2).^{21,22,28,29,30,32,33,40,42} Participants with OI had a significantly higher pooled hazard of all-cause mortality (HR, 1.52; 95% CI, 1.28-1.80) compared with participants without OI. Between-study heterogeneity was considerable (I² = 82%) and statistically significant. Many of the studies adjusted for covariates: namely, age and sex (9 of 9), body mass index (2 of 9), educational level (7 of 9), smoking (5 of 9), alcohol use (4 of 9), hypertension (4 of 9), diabetes (6 of 9), cardiovascular disease (7 of 9), cognitive impairment (5 of 9), and head injury (2 of 9). Two studies were excluded from meta-analysis; Leschak and Eisenberger⁴¹ used mediation analysis and had no compatible effect size to pool, and Wilson et al³¹ calculated the HR with olfactory function as a continuous variable (B-SIT score) instead of dichotomizing participants into those with OI and those without OI. Wilson et al³¹ found that, for each additional accurate choice on an odor identification test, the risk of death within a mean time span of 5 years decreased by approximately 6%. Leschak and Eisenberger⁴¹ examined whether social deficits mediated the association between OI and mortality and found that physical intimacy significantly mediated the association between OI and mortality in women even after correcting for social network size, whereas this association was not mediated by emotional intimacy.

Figure 2. Forest Plot Showing the Longitudinal Association Between Olfactory Impairment and All-Cause Mortality.

HR indicates hazard ratio; seTE, standard error treatment effect; and TE, treatment effect.

Meta-regression and Subgroup Meta-analyses (by OI Measurement)

The meta-analysis of OI and all-cause mortality contained sufficient studies for further analyses. Meta-regression (eTable 4 in the Supplement) identified mean follow-up duration as a significant effect moderator (β [SE], –0.0504 [0.0120]; P = .02), accounting for 91.3% of heterogeneity. The pooled HR decreased by a factor of 0.95 for every additional yearly increase in follow-up. All other characteristics (sample size, country [non-US vs US], mean age, method of OI measurement [self-report vs objective], percentage of male participants, number of covariates, and prevalence of OI) were not significant effect moderators.

The pooled association remained significant and similar when subgroup meta-analyses were conducted (eTable 5 in the Supplement). When stratified by the method of OI measurement, heterogeneity for objective OI measurement subgroup analysis remained significant (I² = 86%), whereas that for self-reported OI measurement was eliminated (I² = 0%) (eFigure 1 in the Supplement). Further subgroup analyses including studies that used the same olfactory tests eliminated between-study statistical heterogeneity (I² = 0%). When stratified by country, heterogeneity in non-US countries was eliminated, whereas that for US studies remained high (I² = 90%) (eFigure 2 in the Supplement).

Sensitivity Analysis

Sensitivity analysis was performed, and 4 studies were excluded.^21,22,28,42 Because Devanand et al²⁸ and Liu et al²¹ stratified their population based on severity of OI and did not report an overall study estimate, we computed an overall study estimate using a fixed-effects model for our overall meta-analysis (Figure 2). Xiao et al⁴² used a single-smell test, unlike the other studies in which olfactory function was determined by a score derived from the identification of multiple odors. Pinto et al²² used odds ratios instead of HRs as reported by the other studies. Hence, owing to the differences in methods and statistical analyses of these 4 studies, we excluded them in a sensitivity analysis. The pooled HR remained statistically significant, and its 95% CI became narrower (HR, 1.35; 95% CI, 1.20-1.50). Heterogeneity was also eliminated (I² = 0%).

Publication Bias

Publication bias was assessed qualitatively by visual inspection for funnel plot asymmetry (Figure 3), as well as quantitatively using the Egger test and trim-and-fill imputation. The funnel plot suggested possible asymmetry about the midpoint. However, results of the Egger test suggest that funnel plot asymmetry was not statistically significant.

Figure 3. Contour-Enhanced Funnel Plot for the Longitudinal Association of Olfactory Impairment With All-Cause Mortality.

The shaded circles indicate the original study estimates, and the unshaded circles indicate the missing estimate imputed via the trim-and-fill method. Dark gray, gray, and light gray contour lines indicate conventional milestones in levels of statistical significance (P < .10, P < .05, and P < .01, respectively).

To err on the side of caution, trim-and-fill adjustement was performed, which imputed 5 additional studies. The recomputed overall HR did not change substantially (HR, 1.21; 95% CI, 1.04-1.40).

Quality of Evidence

Using the GRADE framework (eTable 3 in the Supplement), the certainty of evidence for all-cause mortality was determined to be moderate. The certainty was upgraded by 1 level for studies reporting a dose-response relationship. Downgrading was not necessary because the initial detected heterogeneity was sufficiently explained by meta-regression (follow-up duration), with mean follow-up duration accounting for 91.3% of heterogeneity. The Egger test also demonstrated that bias was not significant, and trim-and-fill analysis showed that the pooled association remained significant.

Discussion

In this systematic review and meta-analysis of 11 observational studies comprising a combined cohort of 21 601 participants, we found that participants with OI had 52% increased hazards of all-cause mortality (HR, 1.52; 95% CI, 1.28-1.80) compared with participants with normal olfaction. This association was adjusted for potential confounders, such as age, sex, cognitive impairment, smoking, and alcohol use, and remained robust to quantitative and qualitative assessments of publication bias. Heterogeneity was sufficiently explained via subgroup analyses, sensitivity analyses, and meta-regression.

To our knowledge, this is the first systematic review and meta-analysis of the epidemiologic association between OI and mortality. Although we have shown the existence of this association, our study is unable to definitively deduce a causal relationship between OI and mortality. This evidence represents the burden of OI that is not due to COVID-19. Although COVID-19–related anosmia is associated with lower mortality, based on an observational study of 576 patients,³⁶ it is unknown whether patients with COVID-19 and long-term anosmia are at higher subsequent risk of mortality compared with patients with COVID-19 who regain their sense of smell.

To date, the exact pathophysiologic characteristics linking OI and mortality are yet to be fully understood. Confounding is likely (Figure 4) and may include factors such as neurodegenerative disorders, systemic diseases, depression, advanced biological aging, and environmental pollution, among others. First, OI is a known early biomarker prior to full emergence of various neurodegenerative diseases, including Alzheimer disease and Parkinson disease.^{43,44,45,46,47} Neurodegenerative markers such as α-synuclein, β-amyloid, and tau have been implicated in the olfactory tract based on previous pathologic studies.^45,46,48,49 Second, systemic diseases such as diabetes, autoimmune diseases, or iron deficiency may cause central and peripheral olfactory dysfunction^10,11 and may be associated with mortality. Third, depression can generate endogenous inflammatory cytokines and glucocorticoids, which may impair olfactory neurogenesis and may also increase the risk of mortality.^50,51,52 Fourth, OI may reflect underlying advanced physiological aging that predisposes an individual to mortality because cellular regeneration is an essential process in maintaining olfaction.^23,53,54,55 Fifth, the olfactory nerve is the only cranial nerve in direct contact with the environment. Airborne toxicants may injure olfactory neurons⁵⁵ as well as other organs, including the central nervous, pulmonary, and cardiovascular systems, which may be associated with increased mortality.³³ Although not all these factors can be adjusted for in epidemiologic studies, many of the included studies had accounted for some of these factors as covariates. For instance, dementia or Parkinson disease explained only part of the association between OI and mortality,²¹ whereas adjustment for diabetes and depression did not weaken the association.^29,30 Hence, mechanisms other than confounding should be considered.

Figure 4. Schematic Representation of Some Possible but Unconfirmed Confounders and Causal Pathways Between Olfactory Impairment and Mortality.

One potential mediator on the pathway between OI and mortality is physical frailty (Figure 4).⁵⁶ Frailty is a medical syndrome characterized by decreased strength, endurance, and physiological function, increasing an individual’s vulnerability to external stressors,^57,58 which may lead to increased mortality. There is some evidence that OI may be associated with frailty, and thus mortality, by means of malnutrition, depression, and cognitive decline. First, because taste is largely a function of retronasal olfaction, OI may impair the enjoyment of food, thus causing a loss of appetite, loss of weight, and malnutrition,⁴⁷ which are factors associated with frailty. Second, OI may reduce food satisfaction, create concerns about personal hygiene, lessen social participation, and thus increase the risk of depression.¹⁵ Biologically, OI likely has an association with brain function, particularly emotional control,¹⁵ because olfactory bulb ablation elicits chemical and behavioral changes similar to those of depression.⁵⁹ Third, smell may be involved in cognition because endoscopic sinus surgery that is associated with improved olfaction has been shown to enhance cognition.⁶⁰ Together, malnutrition, depression, and cognitive impairment may be associated with frailty and mortality. Alternatively, OI may be directly associated with all-cause mortality owing to the inability to detect fires, gas leaks, spoiled food, and toxic substances.¹⁴ Nonetheless, our study cannot draw causal conclusions owing to the observational nature of the included studies.

Strengths and Limitations

There are several strengths of this review. The adequate number of systematically included studies from a varied range of situations enhances the generalizability of our findings. A strict prespecified method of systematically searching for bias and evaluating the quality of grading according to international criteria was followed. No studies were excluded based on high risk of bias. Of 11 included studies, only 1 had moderate risk of bias, whereas the rest had low risk of bias. To effectively account for relevant confounders in the extant literature, we extracted and pooled the maximally adjusted effect estimates. Meta-regression could effectively explain heterogeneity, with extended follow-up periods diminishing the association between OI and all-cause mortality. In addition, most of the included studies (8 of 9) used objective methods to quantify OI. These objective methods are frequently undertaken in controlled environments with defined processes that better measure OI and rely on genuine perception of stimulus when offered, reducing response and measurement bias.⁶¹

Nevertheless, several limitations must be considered. First, within the objective categories, different measurement methods were used by different studies, which may have resulted in measurement bias and may have reduced accuracy in quantifying OI. One of the 9 included studies in the meta-analysis used self-report, which may have resulted in inaccuracy becaused subjective measurements are prone to variation, inconsistency, and recall bias. Second, owing to the observational nature of the included studies, causal conclusions could not be made because residual confounding could not be ruled out, although the longitudinal nature of the studies showed a clear temporal sequence. It is possible that some unmeasured demographic characteristics, environmental factors, or preexisting comorbidities were not included as covariates. Third, the wide variation in the quantity and types of covariates among the included studies could have resulted in biased findings. Despite this, meta-regression failed to identify the number of covariates as a major effect moderator because most studies controlled for sociodemographic factors, lifestyle factors, and critical comorbidities, such as cardiovascular and neurodegenerative disease. Some studies may have established a negative bias by overadjusting their statistical models by incorporating variables that are mediators of the association between OI and mortality. Fourth, the considerable statistical heterogeneity of the included studies (I² = 82%) may have limited the generalizability of our meta-analysis. This statistical heterogeneity may be due to clinical (eg, self-report and objective measurements) or methodological differences, or both. Fifth, the exclusion of articles not written in English may have restricted the external validity of our study and introduced a selection bias in the population. Sixth, given that vision and hearing impairments are also associated with mortality,⁶² further work should explore whether the excess mortality associated with OI may be compounded by the presence of concomitant sensory impairments.

Conclusions

In this multiadjusted observational systematic review and meta-analysis of 21 601 participants in 11 studies with moderate-quality evidence overall, OI was associated with a 52% higher risk of all-cause mortality. Olfactory impairment is a prevalent problem, especially among elderly individuals. Hence, physicians should consider the relevance of OI to general health and aging. Patients should be educated on the importance of OI and how to manage it and its underlying medical comorbidities. Moving forward, more research is required to determine the exact mechanism underlying the association between OI and mortality and the scope for interventions. These results will aid in understanding the causal relationship between OI and mortality and will be of great utility in deciding whether interventions (eg, medication or olfactory training) and testing for OI should be offered to reduce mortality and improve patients’ quality of life.

Supplement.

eMethods.

eReferences.

eFigure 1. Forest Plot Showing the Longitudinal Association Between Olfactory Impairment and All-Cause Mortality, Stratified by Method of Measurement of Olfactory Impairment

eFigure 2. Forest Plot Showing the Longitudinal Association Between Olfactory Impairment and All-Cause Mortality, Stratified by Country

eTable 1. Meta-analysis of Observational Studies in Epidemiology (MOOSE) Checklist

eTable 2. Evaluation of Risk of Bias Using the Newcastle-Ottawa Scale (NOS) for Cohort Studies

eTable 3. Evaluation of Quality of Pooled Evidence Using the Grading of Recommendations Assessment, Development and Evaluation (GRADE) Framework

eTable 4. Random-Effects Meta-Regression of Log(HRs) Against Potential Effect Moderators (Continuous and Categorical Study-Level Characteristics) for the Longitudinal Association of Olfactory Loss With All-Cause Mortality

eTable 5. Meta-Analyses in Subgroups, Stratified by Categorical Study-Level Characteristics for the Longitudinal Associations of Olfactory Impairment With All-Cause Mortality