Mackenzie E Hannum, Riley J Koch, Vicente A Ramirez, Sarah S Marks, et. al., Chemical Senses, Volume 48, 2023, bjad043, https://doi.org/10.1093/chemse/bjad043

Abstract

Chemosensory scientists have been skeptical that reports of COVID-19 taste loss are genuine, in part because before COVID-19 taste loss was rare and often confused with smell loss. Therefore, to establish the predicted prevalence rate of taste loss in COVID-19 patients, we conducted a systematic review and meta-analysis of 376 papers published in 2020–2021, with 235 meeting all inclusion criteria. Drawing on previous studies and guided by early meta-analyses, we explored how methodological differences (direct vs. self-report measures) may affect these estimates. We hypothesized that direct measures of taste are at least as sensitive as those obtained by self-report and that the preponderance of evidence confirms taste loss is a symptom of COVID-19. The meta-analysis showed that, among 138,015 COVID-19-positive patients, 36.62% reported taste dysfunction (95% confidence interval: 33.02%–40.39%), and the prevalence estimates were slightly but not significantly higher from studies using direct (n = 15) versus self-report (n = 220) methodologies (Q = 1.73, df = 1, P = 0.1889). Generally, males reported lower rates of taste loss than did females, and taste loss was highest among middle-aged adults. Thus, taste loss is likely a bona fide symptom of COVID-19, meriting further research into the most appropriate direct methods to measure it and its underlying mechanisms.

Introduction

COVID-19, a respiratory infection caused by the novel severe acute respiratory syndrome coronavirus-2 (SARS-CoV-2), was first identified in Wuhan, China, and has since spread throughout the world. When the World Health Organization first declared this a pandemic in March 2020, researchers and clinicians were not yet aware that the virus affected individuals’ senses of smell and taste, but these symptoms soon became apparent via patient reports. As a result of COVID-19, affected people can experience chemosensory dysfunction in a variety of ways, including complete loss of smell or taste (anosmia or ageusia, respectively), partial loss of smell or taste (hyposmia or hypogeusia), and/or a distorted sense of smell or taste (e.g. parosmia, dysgeusia). These chemosensory dysfunctions can be distressing to the affected individuals and can last for extended periods, with some patients experiencing resolution within a few weeks to a month (Lee et al., 2020b; Gerkin et al., 2021) and others with symptoms for 6 months or longer (Blomberg et al., 2021).

Previous meta-analyses have examined smell and taste loss in COVID-19 patients, but often with a focus on onset and duration (Santos et al., 2021) or recovery (Boscutti et al., 2021) of chemosensory symptoms. Many focused only on smell loss (Hannum et al., 2020; Pang et al., 2020; Rocke et al., 2020) or general neurological symptoms (Abdullahi et al., 2020; Favas et al., 2020; Mair et al., 2021; Yassin et al., 2021). Some meta-analyses provide evidence of taste loss related to COVID-19 (Agyeman et al., 2020; Tong et al., 2020; von Bartheld et al., 2020). However, very few continued to evaluate articles published in 2021 and often capped reviewing articles 6–10 months after March 2020, when the pandemic was declared, limiting the number of articles included (ranging from 5 to 59 articles total). Therefore, we decided to conduct a more comprehensive analysis, spanning a year and a half, to ensure fuller coverage of the available research.

Additionally, taste loss is often neglected in research compared to smell loss, as there is a common notion that taste loss is not as “real” as smell loss. Some claim taste loss is indistinguishable from smell loss (Le Bon et al., 2021) or is confused with smell loss (Deems et al., 1991), specifically with retronasal smell perception (Hintschich et al., 2020). For the general population, loss of taste can be difficult to distinguish from smell loss. Therefore, it may be difficult to know, based on self-report measures alone, whether participants truly lost their sense of taste (Hintschich et al., 2020, 2021).

Thus, many chemosensory researchers may attribute the taste loss phenomena seen in the current reports of COVID-19-positive patients to deficiencies of self-report or subjective measures of taste loss. Therefore, we conducted a systematic review and meta-analysis to estimate the true prevalence of taste loss in COVID-19 patients across a wide sample of studies (n = 235) and to evaluate effects of major methodological differences in data collection. In particular, we compared overall findings on taste loss as determined by individual taste tests (herein referred to as direct tests) with those based on self-reports without direct sensory testing. We hypothesized that direct methodologies would support the presence of taste loss as a distinct symptom and that direct measures might produce the same or even higher rates than self-report, despite the possible inflation of self-reported taste loss exacerbated by smell loss.

Currently, scientists are using both direct and self-report measures to examine chemosensory dysfunction, with self-report far more common due to the pandemic restrictions—for example, closure of sensory laboratories where direct testing is often conducted. For taste, direct tests include standardized and nonstandardized tests that contain various sweet, salty, and sometimes bitter and sour stimuli given to participants via solutions, drops, strips, or sprays (Cao et al., 2021; Singer-Cornelius et al., 2021). Nonstandardized direct taste measures created to study COVID-19-related taste dysfunction include solution-based tests, often prepared at home by participants (Vaira et al., 2020f). Self-report measures include interviews with researchers and clinicians, electronic health records, and surveys administered over the phone, online, or in person.

To understand taste loss as a symptom of COVID-19, we conducted a large systematic review and meta-analysis, examining how it has been measured (direct vs. self-report) and how the measurement type can affect prevalence rates. We tested the hypothesis that direct measures are at least as sensitive as self-report measures and would confirm taste loss as a distinct symptom and not merely as misattributed smell loss.

Methods

Article selection

This systematic review and meta-analysis followed the Preferred Reporting Items for Systematic Reviews and Meta-Analyses guidelines (Moher et al., 2009); the annotated PRISMA 2020 Checklist is in Supplementary Materials (see Supplementary Table S1). From 2020 May 15 to 2021 June 1, we used a daily alert tool offered by PubMed and Google Scholar to identify recently published papers containing the keywords “taste” and “COVID-19.” The N = 712 papers arising from these alerts were collected and evaluated as described below.

Initial screening of the articles included reading the titles and abstracts to assess their relevance. Articles with an abstract that reported chemosensory dysfunction in COVID-19-infected individuals were included in the systematic review (n = 376). Next, at least 2 authors read the articles initially deemed relevant, to evaluate whether they fit the inclusion criteria: reporting positive COVID-19 tests, written in the English language, and lack of population bias. COVID-19 must have been confirmed via nasopharyngeal swab, reverse transcription polymerase chain reaction (RT-PCR), or assessment by physician or other medical personnel. The articles were then evaluated on whether they reported taste loss data specifically. In total, 141 articles were excluded based on such criteria as not evaluating taste loss, recruiting participants with chemosensory dysfunction, not testing for COVID-19, and presenting overlapping data (see Figure 1). In total, 235 articles were included in the final meta-analysis (corresponding citations are listed under “Included Articles” at the end of this article).

Figure 1.

CONSORT flow diagram demonstrating the article selection process for this systematic review and meta-analysis.

Open in new tabDownload slide

Data extraction

We extracted from each article either the number or percentage of patients with taste dysfunction due to a SARS-CoV-2 infection. The prevalence of taste loss reported in each article was calculated by dividing the reported number of participants with taste loss as a symptom by the total number of COVID-19-positive participants. Additionally, measures of taste loss were labeled as “self-report” or “direct” to identify the method used to evaluate participants. Self-report measures included reported loss of taste via surveys, interviews, and electronic medical health records. Most articles (n = 220) used self-report methods. Table 1 summarizes the studies that used direct measures (n = 15), comprising actual taste tests administered either at home (Adamczyk et al., 2021; Hintschich et al., 2020; Petrocelli et al., 2020), at a testing facility (Altin et al., 2020; Bidkar et al., 2020; Mazzatenta et al., 2020; Ramteke et al., 2020; Vaira et al., 2020a; Niklassen et al., 2021; Salcan et al., 2021), or both in home and in a hospital environment (Vaira et al., 2020b, 2020c, 2020d, 2020f); Vaira et al. (2020e) had an unknown testing location. Many of the measures consisted of solution-based tests measuring 4 basic taste sensations: sweet, sour, salty, and bitter.

Table 1.

Open in new tab

Overview of direct approaches to assess taste loss in participants

Test	Article	Test method	Test objective	Taste quality measured
				Salty	Sweet	Sour	Bitter	Umami
Four-solution test	Vaira et al. (2020a, 2020b, 2020c, 2020d, 2020e, 2020f); Petrocelli et al. (2020)a	1 mL of each solution, plus deionized water as control, placed on the participant’s tongue via cotton swab. Quarantined patients prepared their own solutions.	Identification	☑	☑	☑	☑	☐
Four-solution test and Taste Strips	Altin et al. (2020); Salcan et al. (2021)	Participants swallowed and identified the solution. Next, paper strips were dipped into each solution and placed on the participant’s tongue.	Identification Duration	☑	☑	☑	☑	☑
Two-solution test (1)	Bidkar et al. (2020)	Two drops (2 mL) of each solution placed on the participant’s tongue via pipette.	Identification	☑	☑	☑	☑	☐
Two-solution test (2)	Mazzatenta et al. (2020)	200 µL of each solution dropped onto participant’s tongue.	Detection	☑	☑	☑	☐	☑
Taste Strips by BMGb	Hintschich et al. (2020); Niklassen et al. (2021)c	Taste Strips (provided by Burghart Messtechnik GmbH) placed on the participant’s tongue.	Identification Threshold concentration	☑	☑	☑	☑	☐
Taste sprays	Niklassen et al. (2021)c	Each solution sprayed onto participant’s tongue.	Identification	☑	☑	☑	☑	☑
Tastant capsules	Adamczyk et al. (2021)	Each of ten 0.33-mL gelatin capsules (one tasteless and nine with tastant) dissolved on participant’s tongue.	Description	☑	☑	☑	☑	☐
Chemosensory test by India protocol	Ramteke et al. (2020)	Authors used coconut oil, chocolates, and flavored milk to test smell and taste function. No further details are provided.	N/A	N/A

^aThese studies used a sweet solution concentration that is double what was used in the other 4-solution tests.

^bValidated test.

^cNiklassen et al. (2021) used both Taste Strips and taste sprays with participants.

Although some authors reported exclusively on smell loss or taste loss, many reported on both senses. Thus, it was necessary to include additional coding options for when authors reported the symptoms in tandem (e.g. “loss of smell or taste”). Therefore, articles were labeled as “taste only,” “smell and/or taste,” “smell and taste,” “smell or taste,” and “smell and/or taste; taste only” (when values for both symptoms were reported, the numbers were summed), depending on how these symptoms were phrased in the article.

We also extracted demographic characteristics of each study, including the population mean and/or median age, sex (expressed as percentage of males in the population), and country of origin (for the geographic distribution of the study populations, see Supplementary Figure S2).

Four authors performed the initial reading of full texts and the data extraction from the studies (R.D.H., A.K.T., S.S.M., R.J.K.). Two authors confirmed this information and resolved any inconsistencies (M.E.H., D.R.R.). Differences were resolved through discussion and renewed consensus on the proposed solution from all authors who read that specific article.

Risk-of-bias assessment

We used a risk-of-bias assessment from Hoy et al. (2012) to examine the articles selected for the meta-analysis. The assessment contained 9 questions, outlined in Supplementary Materials (see Supplementary Table S3), which cover potential areas of concern such as representation of the national and target population, use of random selection, nonresponse rate, data source, case definition, instrument validity, and reported outcomes of interest. Responses were scored as 1 (no) or 0 (yes), with summary scores of low (0–3), moderate (4–6), and high (7–9). Two authors completed the risk-of-bias assessment of each article (S.S.M. and A.K.T.) using the checklist developed by Hoy et al. (2012), as described and adapted by Tong et al. (2020). One author resolved any discrepancies (M.E.H.).

Statistical analysis

The meta-analysis was conducted using the meta package in R (Schwarzer et al., 2019). Generalized linear mixed models were used for the meta-analysis, as recommended for the analysis of binary outcomes and proportions (Bakbergenuly and Kulinskaya, 2018; Schwarzer et al., 2019). Additionally, to confirm the robustness of this model, inverse variance method models were conducted. Heterogeneity (e.g. between-study variances) was assessed using Cochran’s Q, I², and tau squared (τ²). We concluded there was evidence for heterogeneity when the P-value for Cochran’s Q was less than 0.05 and if the I² was greater than 50% (Higgins and Thompson, 2002). Furthermore, τ², which measures the variance of the true effect sizes underlying the data, was estimated using the maximum likelihood method and incorporated in the random-effects model to describe the uncertainty between studies and generate the prediction interval.

The prevalence of taste loss differs from study to study, which could be due to one true prevalence (fixed effect) plus experimental noise, or the true prevalence might differ among studies (random effect). We chose the random-effects model because of the extreme study-to-study variation coupled with the practical observation of genuine study differences, for example, due to host genetics (the genetic background of the study participants) or to differences in experimental protocols (e.g. data collected at different stages of illness) or in collection methodologies. We can identify at least one source of this variation: the stimuli used to measure taste loss, for example, solutions versus strips in the case of direct measures. The presence of excess heterogeneity measured through Cochran’s Q test, I², and τ² validated the inclusion of the random-effects model in our study. Therefore, overall pooled prevalence estimates were computed and reported for a random-effects model with the parameters described for all 235 studies.

Subgroup analysis was performed for studies employing direct methods (n = 15) and self-report methods (n = 220) to assess taste function in COVID-19-positive individuals. Additionally, the average age of participants and their sex (percentage of male subjects in each study) were included as covariates in univariate mixed-regression models (a random-effects model). Subgroup analysis for age was categorized into 5 groupings: adolescents (0–18 years old), young adults (19–35 years old), middle-aged adults (36–50 years old), older adults (51–65 years old), and elderly adults (65+ years old). Finally, studies that used direct tests were separated by type of collection methodology, and subgroup analysis was performed for each type: solution based (N = 11), strip based (N = 2), and other (N = 2). For continuous variables (e.g. % of sex), the transformed beta coefficients are reported. The transformation for the generalized linear mixed models uses a logit transformation for each proportion, so the models are interpreted as the log(odds) or log(p/1 – p), where p is equal to the prevalence for each study. Back-transformation into the prevalence estimates was done for subgroup analysis (e.g. age).

All statistical analyses were performed using R 4.0.5 (R Core Team, 2020) and RStudio 1.4.1106 (RStudio Team, 2020). Visualization of the meta-analysis is displayed as an orchard plot adapted from Nakagawara et al. (2020). The R scripts and compiled data used for this analysis are available without restriction at GitHub (https://github.com/vramirez4/COVID19-TasteLoss). This review was not preregistered.

Results

Risk-of-bias assessment

Each of the 235 articles included in this meta-analysis were reviewed for risk of bias. Among these articles, none had a high risk of bias, 137 studies had a low risk, and 98 studies had a moderate risk (see Supplementary Table S3 for the full assessment). Articles deemed moderate risk often did not meet the sampling standards (e.g. not a true representation of the national population, no random selection). However, this was not specific to only articles classified as self-report, as 7 of the 15 (~47%) direct measure studies were also deemed moderate risk.

Prevalence of taste loss in COVID-19-positive patients

Among the 235 studies, the sample sizes ranged from 12 to over 40,000 patients with COVID-19. The number of cases of taste loss per study ranged from 0 to 4,668, with raw prevalence estimates ranging from 0% to 89.9%. Collectively, the meta-analysis included 138,015 patients who tested positive for COVID-19. Of these, 32,348 patients (23.44%) had some form of taste loss after infection with SARS-CoV-2. Heterogeneity among the prevalence estimates across all studies (n = 235) yielded a significant Cochran’s Q (Q = 29030.73, degrees of freedom [df] = 234, P < 0.001), an I² estimate of 99.2%, and τ² estimate of 1.476. The pooled estimate for taste loss prevalence in COVID-19-positive patients following meta-analysis for the overall cohort was 36.62% (95% confidence interval [CI]: 33.02%–40.37%). The higher random-effect estimate, compared with simply pooling the raw data (32,348 cases among 138,015 patients = 23.44%), is a result of the heterogeneity among studies. This arises due to variance both within and between studies, which is ignored when pooling the raw data as a single sample. Meta-analysis addresses these uncertainties, both within and between studies, before generating the pooled prevalence estimate.

Effect of methodology (direct vs. self-report) on prevalence estimate

We employed a subgroup analysis to determine the effect of direct versus self-report approaches on taste loss prevalence (see Figure 2). Fifteen studies used direct methods to assess taste loss, comprising 2,085 COVID-19 patients, with 1,062 reported cases of taste loss. Per study, the prevalence of taste loss ranged from 0% to 84% among COVID-19 patients. For studies using direct approaches, the pooled estimate of the prevalence for the random-effects model was 45.00% (95% CI: 32.58%–58.08%). Cochran’s Q was significant (Q = 127.72, df = 14, P < 0.001), and an I² of 89.0% and a τ² of 1.22 were obtained, confirming heterogeneity of the data collected via direct measures.

Figure 2.

Orchard plot of taste loss and COVID-19, following the guidelines outlined by Nakagawa et al. (2020). The point estimate of the pooled prevalence (trunk) is represented by the bold turquoise or pink dot. The confidence interval of the pooled prevalence estimate (branch) is represented by the bold black line, and the prediction interval (twig) is represented by the thin black line. Individual prevalence estimates from each study are represented by the scattered colored points (slightly transparent circles, called fruits). Each fruit is scaled by the precision of the point estimate of prevalence for each study (i.e. inverse of the standard error).

Open in new tabDownload slide

A total of 220 studies used self-report methods (e.g. questionnaire, interview), comprising 135,930 COVID-19 patients, with 31,286 cases of taste loss. The reported prevalence of taste loss ranged from 1% to 89.9% per study. The pooled estimate of the prevalence under the random-effects model was 36.11% (95% CI: 32.41%–39.99%). Similar to the direct subgroup, Cochran’s Q was significant (Q = 28795.86, df = 219, P < 0.001), and the I² value was 99.2% and τ² was 1.49, confirming heterogeneity of the data collected via self-report.

Despite the higher prevalence rate of taste loss when directly measured, under the random-effects model the difference between methodologies (direct vs. self-report) was not statistically significant (Cochran’s Q = 1.73, df = 1, P = 0.1889). Thus, while our analysis showed that the prevalence of taste loss was higher when measured directly than by self-report, there was no significant effect of measurement method on the prevalence estimates of taste loss.

Effect of age and sex on taste loss prevalence

Additional analyses were undertaken to assess the effect of age and sex. Univariate mixed models for each covariate revealed that both age and sex had significant effects on the prevalence of taste loss. After categorizing each study by mean age group, we conducted the meta-analysis for those 205 studies that reported the ages of the participants (see Table 2). Among this subset of studies, the pooled prevalence was 36.98%, with significant heterogeneity (Q = 17984.70, df = 204, P < 0.001; I² = 98.9; τ² = 1.49). The prevalence estimates per age category ranged from 11.54% in studies with average ages younger than 18 years to 43.64% in studies with average ages between 36 and 50 years. Heterogeneity in these studies was high (I² = 96.4–98.9%) both within groups (Q = 13802.36, df = 200, P < 0.0001) and between groups (Q = 32.13, df = 4, P < 0.0001). The results demonstrate that both the youngest and oldest age groups report the lowest prevalence of taste loss, while age groups between 18 and 65 years had pooled estimates ranging from 32% to 44%, with the highest in the middle-age (36–50 years) group. Further, subgroup analysis for mean age resulted in reduction of between study variance in age groups where the average age corresponds to adolescent and older age groups, as measured by reduction in τ².

Table 2.

Open in new tab

Random-effects estimate of age group on COVID-19 taste loss prevalence using generalized linear mixed models

Age group	k	Proportion	95% CI	Q	I2	τ2
Adolescent	9	0.12	0.06–0.20	382.59	97.9%	0.8280
Young adult	20	0.32	0.20–0.46	1,482.59	98.7%	1.9610
Middle age	113	0.44	0.38–0.49	10,222.72	98.9%	1.3998
Older	55	0.34	0.29–0.41	1520.83	96.4%	0.9291
Elderly	8	0.18	0.08–0.36	193.64	96.4%	1.6031

The effects of sex were similarly examined. A univariate mixed model using percentage of males in each study as a covariate found an effect of β = −0.0306 (P < 0.001) for each percent increase. Overall, the higher the percentage of males in a study, the lower the prevalence of taste loss.

Effect of type of direct approach on taste loss prevalence

When we compared the types of direct report test (e.g. solution-based test, taste-strip-based test), we found a significant difference in prevalence rates (see Table 3). We classified studies into 3 categories: taste strip testing (n = 2), taste solution testing (n = 11), and “other” (n = 2) for methods that do not employ either solution or strips. Pooled prevalence for solution-based tests was 54.54% (95% CI: 44.56%–64.17%), for taste strips was 23.53% (95% CI: 16.30%–32.71%), and for “other” was 4.66% (95% CI: 0.01–94.12%). There was reduced heterogeneity in the subgroups compared with the overall meta-analysis: solution-based testing, I² = 89.8% and τ² = 0.4178; strip-based testing, I² = 0% and τ² = 0. There was significant heterogeneity between our pooled estimates (Q = 20.02, df = 2, P = 0.0001). Together, our results demonstrate that studies using solution-based taste tests, on average, result in higher prevalence of taste loss in COVID-19 patients than do studies using strips or other methods.

Table 3.

Open in new tab

Random-effects estimate of direct testing type on COVID-19 taste loss prevalence using generalized linear mixed models

Direct testing type	k	Proportion	95% CI	Q	I2	τ2
Solutions	11	0.54	0.4456–0.6417	98.26	89.8	0.4138
Strips	2	0.23	0.1630–0.3271	0.61	0.0%	0.0
Other	2	0.05	0.0001–0.9412	0	0.0%	10.9686

Discussion

Despite the occurrence of true taste loss in a variety of diseases such as cancer (Nolden et al., 2019), as well as in the general population (Rawal et al., 2016), taste loss has often been confused with smell loss (Le Bon et al., 2021). However, the current coronavirus pandemic suggests that taste loss is its own unique feature of the illness. The present meta-analysis found an overall taste loss prevalence of 37% among 138,015 COVID-19-positive participants, which aligns with other meta-analyses of taste loss prevalence, ranging from 38% (Agyeman et al., 2020) to 49% (Hajikhani et al., 2020). This high prevalence is not due to confusion with smell loss because direct taste measures yield similar (or even slightly higher) prevalence than self-report. Therefore, self-reports of taste loss appear to be valid among people with COVID-19 as they are among other groups (Jang et al., 2021), indicating that questions about self-reported taste loss can be incorporated with more confidence into large-scale surveys, such as NIH All of Us (Precision Medicine Initiative [PMI] Working Group, 2015).

The COVID-19 pandemic created an urgent need for direct taste measures suitable for the pandemic research environment (e.g. home testing), and researchers were innovative, but each developed their own method, which makes it difficult to compare results (see Table 1). However, despite the differences in methods, we can draw one general conclusion: the form of the tastant matters—taste solutions are better than taste strips. However, this general conclusion is tentative because the forms of delivery were not compared directly using the same outcome (e.g. thresholds or identification).

The present meta-analysis found that around 4 in every 10 COVID-19 patients experience taste loss. We also found age and sex effects: females experienced higher rates of taste loss than males, aligning with other meta-analyses reporting a similar effect (von Bartheld et al., 2020; Amorim Dos Santos et al., 2021; Saniasiaya et al., 2021). Females may be more susceptible to taste loss because they are in general are more sensitive than males and have more sensory capacity to lose. Additionally, we found that COVID-19-associated taste loss peaks in middle age, aligning with the general consensus across other COVID-19 meta-analyses (Agyeman et al., 2020; von Bartheld et al., 2020). Why the youngest and oldest groups report less taste loss than do middle-age adults is not currently known.

Although COVID-19 has intensified awareness of taste loss and furthered chemosensory research, scientists are still unsure of the biological mechanisms behind this symptom. The amount of SARS-CoV-2 virus in saliva is positively related to loss of taste: the more virus, the more taste loss (Huang et al., 2021; Taziki Balajelini et al., 2021), although this observation is controversial (Jain et al., 2020). Taste cells may be attacked directly by the virus because studies have shown that both ACE2, the receptor protein known to transport the SARS-CoV-2 virus into cells, and TMPRSS2, the protein essential for processing the SARS-CoV-2 spike protein, are expressed in the supporting cells of taste buds (Sakaguchi et al., 2020; Huang et al., 2021), as well as in taste receptor cells themselves, at least in one patient (Doyle et al., 2021). There may also be direct effects on the brain that contribute to taste loss (Douaud et al., 2021).

Limitations and Future Research

In many of the included articles, clinicians and researchers collected self-reported taste loss information in tandem with smell loss (e.g. participants responding yes to “Loss of taste and smell” on a symptom screener), which can confound the results. Therefore, we explored any differences in how taste loss was collected, reflecting how it was reported in the articles (e.g. “smell and taste” vs. “taste only”) and found no significant impact on the prevalence rate (see Supplementary Material S4). This result indicates that taste loss is a common symptom of COVID-19. Additionally, far more articles in this meta-analysis used self-report tests (n = 220) than direct tests (n = 15). This disparity may have prevented us from capturing significant differences between these 2 methods.

Nearly all of the articles included in this meta-analysis were nonspecific to different tastes, instead summarizing scores across multiple stimuli (e.g. sweet and salty) and reporting taste loss as a whole (though in one study participants self-reported taste-specific dysfunction: salt taste loss, 29.3%; sweet taste, 25.9%; general taste, 34.5% [El Kady et al., 2021]). However, specific taste sensitivities can be difficult to assess via self-report. Members of the general population, untrained in chemosensory science, may have difficulties identifying whether or not they truly lost a specific taste. Therefore, it is important to use direct measures to distinguish dysfunctions of specific tastes.

There are clear opportunities for advancements of standardized direct taste tests to measure taste loss. Of the 15 studies that used direct tests in general, only 2 used standardized tests, representing just 0.85% of the 235 studies examined in this systematic review (Hintschich et al., 2020; Niklassen et al., 2021). Six other studies using direct tests were examined during the systematic review but were excluded for not meeting the inclusion criteria: reporting on a case study (Lee and Lee, 2020), recruiting patients with chemosensory dysfunction (Le Bon et al., 2020, 2021; Cao et al., 2021; Singer-Cornelius et al., 2021), and not reporting the required taste loss prevalence data (Huart et al., 2020). Among these studies, Lee and Lee (2020) used a nonstandardized direct test, and the others used the taste strips—although this represents a missed opportunity to analyze more studies that used standardized tests, including these articles would have increased the rate of standardized test use among all 240 studies to just 2.92% (n = 7 of 240 potential studies).

As of September 2021, 220 million individuals have been infected with SARS-CoV-2 virus, with large numbers of recovered individuals having persistent symptoms, including taste loss. It is well documented that disease-related or age-related chemosensory loss has profound effects on an individual’s quality of life. Unlike other disorders, such as those of vision and hearing, for which preventative and screening guidelines exist (United States Preventive Services Taskforce, 2021), taste and smell disorders have no such guidelines available. The COVID-19 pandemic further highlighted this existing gap—the lack of standardized measures and clinical guidelines for screening, assessing, and monitoring the taste system make it difficult for clinicians to identify and track progression. Assessment of taste function in patients with and without confirmed COVID-19 needs to become standard of practice for clinicians. This is particularly important for at least 2 reasons: (i) having baseline measures helps clinicians assess trends over time and (ii) given the interrelatedness between the senses of smell and taste, objective measures of taste collected during clinical assessments may help dissociate smell and taste problems. For patients who report changes in taste function during screening questionnaires, full testing with standardized objective chemosensory tools should be performed. It is critical that clinicians be aware that most patients with chemosensory dysfunction complain of taste alterations; therefore, closer inquiry into a patient’s reports regarding effects on a specific taste quality (e.g. sweet, bitter, sour, salt) is important to further distinguish between taste per se and the perception of flavor, which combines taste and smell.

Conclusion

The COVID-19 pandemic demanded an urgent response from scientists and clinicians, who have been working to understand the novel virus and the symptoms it inflicts. Of the many unique features of this virus, smell and taste dysfunctions are among the most prominent. Yet as taste loss joined smell loss as a more prolific topic in scientific literature, many initially speculated that taste loss rates were overestimated due to confusion between taste and smell in self-reports. In contrast, our meta-analysis found a prevalence rate for taste loss of 36.62% among 138,015 COVID-19-positive individuals (95% CI: 33.02%–40.37%), supported by direct methods, reflecting the validity of this distinct symptom. Dysfunction in the sense of taste was, and still is, a difficult reality for millions of people affected by the virus and merits further research to fully understand the mechanisms behind this phenomenon and how to properly assess and address it. Among the population of 138,015 COVID-19-positive individuals included in this meta-analysis, only 102 of them, across 2 separate studies, were assessed using standardized taste tests. Future research should include the development of fast and accurate taste tests, studies that measure taste and smell function separately to dissociate olfactogustatory interactions, and the use of standardized taste tests in clinical settings to examine taste dysfunction. In addition, clinical trials are needed to elucidate the frequency of screening and the ages at which to start and stop screening for chemosensory disorders. Finally, more mechanistic studies to understand taste and smell disorders associated with COVID-19 are needed to help develop new therapeutic options for patients with long-lasting impairment of their chemical senses.