Introduction

Olfactory dysfunction has been gaining increasing recognition in the fight against COVID-19 [1, 2]. What began as anecdotal reports of patients presenting with anosmia as the sole symptom has evolved into changes in clinical case definitions for suspect cases internationally.

In the context of COVID-19 infections, acute olfactory dysfunction (OD) is defined as decreased or altered sense of smell of a duration of 14 days or less, in the absence of chronic rhinosinusitis, a history of head trauma or neurotoxic medications. OD can be associated with flavour (smell + taste) dysfunction. However, COVID-19 may also affect real taste (sweet, salty, bitter, acidic, umami).

OD is estimated to afflict 3–20% of the population [3, 4]. Post-viral anosmia accounts for up to 40% cases of anosmia or which coronaviruses are thought to account for 10–15% of these cases [5, 6]. As such, it is plausible that COVID-19 may cause OD.

Though the exact pathogenesis is unclear, the high rate of recovery of olfactory function within 1–3 weeks after the onset of OD [7,8,9,10] may provide clues on the mechanism and extent of injury to olfactory epithelium and/or neurones. There are two proposed mechanisms by which COVID-19 causes anosmia. Coronaviruses are known to infect olfactory epithelium [11, 12]. Human angiotensin-converting enzyme 2 (ACE-2) receptor, which is a SARS-CoV-2 receptor, is expressed in the olfactory epithelial cells within the olfactory cleft, specifically the sustentacular cells [13, 14]. Inflammation of the olfactory cleft mucosa can cause conductive OD by reducing airflow and hence odorant presentation to the olfactory cleft [15].

This symptom may hence represent a potential clinical screening tool to facilitate testing of asymptomatic individuals. However, it remains unclear if these findings are causally and uniquely related to COVID-19 infection, or due to increased recognition of OD as a symptom [16]. Amongst patients afflicted with COVID-19, decreased awareness of olfactory dysfunction may be overshadowed by more severe symptoms such as respiratory distress. Furthermore, data in the literature suggests that self-reporting of the sense of smell is specific but not sensitive [17, 18]. Amongst those with measured olfactory dysfunction, 74.2% did not recognise it [18]. This is so amongst patients afflicted with COVID-19 as well [19•].

As such, we set out to conduct a systematic review and meta-analysis on OD in COVID-19 to quantify the clinical utility of identifying OD in the diagnosis of COVID-19 and determine an estimate of the frequency of OD amongst these patients. We also aimed to look separately at survey-reported and smell test-reported OD given the reported variance between the two.

Methods

The Preferred Reporting Items for Systematic reviews and Meta-analyses (PRISMA) Statement [20] was referenced to structure the study. A study protocol was not registered, and no ethics approval was required.

Information Sources and Search Strategy

Studies were eligible if they were indexed on PubMed. The search was performed on 9 May 2020, and the strategy used was “(anosmia OR smell OR hypos* OR olfact*) AND (COVID* OR SARS-CoV-2 OR 2019-nCoV OR coronavirus).” The search was not limited by publication date and there was no language filter applied. The search was updated on 1 August 2020.

Study Selection and Data Collection

Screening of titles and abstracts was performed by 2 independent researchers to determine if the studies met the inclusion criteria. If abstracts were not available, the full text was retrieved and analysed. Any disagreements between the 2 researchers were resolved by discussion and by consulting a third, senior researcher. Data extracted from eligible studies included the author, year of publication, study design, country of origin, OD testing method, COVID-19 testing method and number of cases reporting OD amongst COVID-19 positive and negative patients. Data was entered into Excel sheets independently by the 2 researchers and then compared. Methodological quality was rated independently by two reviewers using the risk of bias tool for prevalence studies by Hoy et al. [21].

Inclusion and Exclusion Criteria

To quantify the clinical utility of identifying OD in the diagnosis of COVID-19, we compared the frequency of OD in patients stratified by COVID-19 test results using the reverse transcription polymerase chain reaction (RT-PCR). This was performed in Meta-analysis A. Studies were included if they compared the frequency of smell disturbance in COVID-19 positive patients (proven by RT-PCR) to COVID-19 negative controls in case-control studies. Appropriate controls were defined as patients who were suspected of having COVID-19 infection or fulfilled local guidelines for COVID-19 testing but were COVID-19 negative on RT-PCR testing. The data items were the number of COVID-19 positive and negative patients with OD and total number of patients tested. Principal summary measures were pooled sensitivity, specificity, positive likelihood ratio (LR), negative LR and diagnostic odd ratios (DOR).

To investigate the estimated frequency of OD amongst COVID-19 patients, meta-analysis B included studies if they described the frequency of OD in COVID-19 positive patients and if smell tests were performed or if OD symptoms were explicitly asked in questionnaires or interviews. The latter criterion was chosen as OD symptoms were not routinely asked in early studies, which might explain the low frequency of OD reported in China. The data items were the number of COVID-19 positive patients with OD. The principal summary measure was the frequency of OD. Subgroup analyses was performed to investigate if the frequency differed between survey/questionnaire-reported OD and smell test-reported OD.

Statistical Analysis

R Studio version 1.2.5042 [22] and R version 4.0.0 [23] were used for all statistical analyses. The packages meta [24], mada [25] and dmetar [26] were used in the analyses. All data are presented as effect estimates with 95% confidence intervals in parenthesis. Heterogeneity amongst studies was tested using the Cochran’s Q test and I2. A random effects model was used if I2 > 50%. Forest plots were generated to summarise the results. Funnel plots and Egger tests were used to detect any publication bias.

Results

Meta-analysis A: the Clinical Significance OD in the Diagnosis of COVID-19

A total of 498 studies were retrieved from PubMed. A total of 422 articles were excluded based on their titles and abstracts, and 57 of the remaining 76 articles were excluded for reasons as described in Fig. 1. The remaining 19 articles were included in the meta-analysis.

Fig. 1
figure 1

Flow diagram for meta-analysis A showing the clinical significance of OD in the diagnosis of COVID-19. aFifty-seven full-text articles were excluded: 49 did not include controls, 4 utilised inappropriate controls who were not swabbed for COVID-19 (3 studies used healthy asymptomatic individuals as controls and 1 study used historical influenza patients as controls), 2 utilised self-reported COVID testing results, 1 added in OD symptoms to their data collection sheet midway through the study and 1 did not explicitly ask for OD symptoms

Full size image

Study Characteristics

A total of 1861 COVID-19 positive patients and 15,556 COVID-19 negative patients were included across the 19 studies as seen in Table 1. The patients were from Canada, France, Germany, Hungary, Italy, Netherlands, Singapore, Spain, Turkey and the USA. All studies utilised RT-PCT as the COVID-19 diagnostic testing method. All studies described survey/questionnaire-reported OD.

Table 1 Characteristics of full-text articles assessed for eligibility
Full size table

Clinical Utility of Identifying OD in the Diagnosis of COVID-19

With reference to Fig. 2, patients with OD were more likely to test positive for COVID-19 (DOR 11.5 (8.01 to 16.5), positive LR 6.10 (4.47 to 8.32) and negative LR 0.58 (0.52 to 0.64)). The pooled sensitivity was 0.48 (0.40 to 0.56), and the pooled specificity was 0.93 (0.90 to 0.96) in using OD to predict COVID-19 infection. There was significant heterogeneity amongst the 6 studies (I2 = 76.4%, p < 0.0001). The Funnel plot is shown in Fig. 5a. Egger’s test suggested the presence of publication bias (p < 0.001).

Fig. 2
figure 2

Meta-analysis A showing the clinical significance OD in the diagnosis of COVID-19. a Diagnostic odds ratio. b Pooled sensitivity. c Pooled specificity of OD in predicting COVID-19 infection

Full size image

Meta-analysis B: Estimating the Frequency of OD Amongst COVID-19 Patients

A total of 498 studies were retrieved from PubMed. A total of 422 articles were excluded based on their titles and abstracts, and 16 of the remaining 76 articles were excluded for reasons as described in Fig. 3. The remaining 60 articles were included in the meta-analysis.

Fig. 3
figure 3

Flow diagram for meta-analysis B estimating the frequency of OD amongst COVID-19 patients. aSixteen full-text articles were excluded: 9 did not specify if OD symptoms were explicitly asked, 5 likely used overlapping data and 2 utilised self-reported COVID testing results

Full size image

Study Characteristics

A total of 17,401 COVID-19 positive patients across 60 studies were included in Meta-analysis B, of which 8606 reported OD. The patients were from all major continents. All utilised RT-PCT as the COVID-19 diagnostic testing method. All used questionnaire-based, symptom-based reporting of OD except for 6 studies (2 used Sniffin’ Sticks, 1 used UPSIT, 1 used the Connecticut Chemosensory Clinical Research Test (CCCRT), 1 used ethyl alcohol and 1 used a combination of CCCRT and ethyl alcohol).

Estimating the Frequency of OD Amongst COVID-19 Patients

With reference to Fig. 4, the overall pooled frequency of OD amongst COVID-19 patients was 0.56 (0.47 to 0.64). There was significant heterogeneity amongst the 60 studies (I2 = 98.8%, p < 0.001). Funnel plot is shown in Fig. 5b. Egger’s test did not suggest the presence of publication bias (p = 0.204).

Fig. 4
figure 4

Meta-analysis B estimating the frequency of OD amongst COVID-19 patients. Pooled prevalence of olfactory dysfunction (OD) amongst COVID-19 patients with subgroup analysis by OD testing method

Full size image
Fig. 5
figure 5

Funnel plots for a meta-analysis A showing the clinical significance OD in the diagnosis of COVID-19 and b meta-analysis B estimating the frequency of OD amongst COVID-19 patients

Full size image

In subgroup analysis in Fig. 4, the frequency of smell test detected OD amongst COVID-19 patients differs between detection via smell testing (0.76 [0.51–0.91]) vs survey/questionnaire report (0.53 [0.45–0.62]), although not reaching statistical significance (p = 0.089).

Risk of Bias

Table 2 summarises the risk of bias of all studies included in both meta-analyses A and B. Overall, the studies were of moderate to high risk of bias due to the lack of smell testing except for 6 studies, the presence of non-response bias using the questionnaire methodology or the inclusion of only particular groups of patients (e.g. only hospitalised patients, or only outpatients, or only those with mild-moderate disease).

Table 2 Risk bias assessment of included studies
Full size table

Discussion

The pooled frequency of OD in COVID-19 positive patients (17,401 patients, 60 studies) was 0.56 but differed between detection via validated smell testing (0.76) vs survey/questionnaire reports (0.53). This inconsistency of olfactory dysfunction between survey/questionnaire reports and validated smell tests has also been recognised in the literature [17, 18]. Moein et al. [19•] reported that 29% of their patients reported self-reported OD. However, validated smell tests on this same group of patients showed 58% to have anosmia or severe microsmia, with only 2% with normal olfactory function. Similarly, Vaira [9] reported 28.3% patients having s OD, while 98% had OD on validated smell tests. A significant number of patients with olfactory dysfunction do not report symptoms. Even within the realm of administered smell tests, cultural differences may result in inaccurate identification of smell dysfunction [96]. This might suggest that at least some of the variation in frequency rates of OD in COVID-19 may be attributed to differences in data collection methods.

Notwithstanding this, patient-reported OD as a symptom was highly specific (93%) but not sensitive (48%), for COVID-19 infection. The results of this meta-analysis further suggest that patients with reported OD were more likely to test positive for COVID-19 (diagnostic OR 11.5), with positive (6.10) and negative (0.58) LR. The presence of patient-reported OD can hence be used as an additional screening question to triage patients in determining the need for COVID-19 testing regardless of the presence of other concomitant upper respiratory symptoms. Whether smell test detected OD may serve as a more accurate screening tool remains to be investigated.

It is increasingly recognised that the COVID-19 infection can manifest as mild, moderate, severe or critical illness [97]. Yan et al. [15] reported that patients with OD may be associated with a milder clinical course. Izquierdo-Domínguez also reported that patients with more severe OD were less likely to be hospitalised and had a lower level of C-reactive protein [34]. However, patients who were intubated or deceased at the time of data collection could not be included in their study. If this were indeed true, the presence of OD might assist in deciding the disposition of patients i.e. admission vs outpatient care. However, Moein et al. [19•] reported that there was no statistically significant difference in the mean UPSIT score between patients with mild, moderate or severe COVID-19. As such, this may be purely be due to recall bias, where patients with severe COVID-19 may be less cognizant of OD due to the presence of more bothersome symptoms such as dyspnoea. The prognostic value of OD in COVID-19 patients remains to be elucidated but is unlikely to override traditional, objective and actionable clinical measurements such as oxygen saturation, pulse rate and respiratory rate.

Various Otolaryngologic societies have issued statements addressing OD in COVID-19. On 21 March 2020, a press release was issued by ENT UK and the British Rhinological Society on Twitter, recommending that anosmia be added to the current symptom criteria used to trigger quarantine and that individuals with new-onset anosmia should self-isolate to reduce the risk of further transmission of COVID-19 [5]. This was largely based on anecdotal physician and media reports [98]. A similar statement was released by the American Academy of Otolaryngology-Head and Neck Surgery (AAO-HNS) on 22 March 2020 [99], and a joint statement was released by the Chapter of Otorhinolaryngologists, College of Surgeons, Singapore, and the Society of Otolaryngology-Head and Neck Surgery, Singapore, on 17 April 2020 [100]. The US Centers for Disease Control and Prevention added “new loss of taste or smell” to the list of COVID-19 symptoms on 17 April 2020, while the World Health Organisation (WHO) has added the above as of 9 May 2020 [101], albeit as a “less common symptom”.

The major limitation of the meta-analysis was the significant heterogeneity amongst included studies. Sources of heterogeneity include different inclusion criteria across studies (e.g. only hospitalised patients or only outpatients included, only mild-moderate illness included), different ways in which the OD questions were phrased and possibly the different RT-PCR sensitivities across different institutions around the world for detection of SARS-CoV-2 RNA. We were unable to perform a meta-analysis of the onset, duration and severity of OD due to the varied data collection protocols. As questionnaires were used in most of the studies, there might have been a strong recall bias in which patients who knew they were COVID-19 positive were more likely to report anosmia. Furthermore, it is impossible to survey intubated or deceased patients so findings may not be generalisable to the most severe of patients. Nevertheless, the clinical utility of patient-reported OD in identifying COVID-19 infection amongst patients with mild-moderate symptoms remains important to facilitate cohorting and isolation, to minimise transmission.

Future research should utilise validated instruments for both survey/questionnaire (i.e. visual analogue scale [VAS]) and smell testing of OD across various time points to quantify the onset and severity of OD and track its recovery. However, we recognise the inherent difficulties in conducting these tests amongst COVID-19 positive patients as it puts researchers at risk of infection. While it is important to correctly diagnose and classify the severity OD in order to study of the characteristics of hyposmia/microsmia or anosmia amongst COVID-19 positive, from a public health perspective, it can be argued that the detection of self-reported OD via surveys of questionnaires is equally important in curbing the COVID-19 pandemic by assisting in identifying COVID-19 positive patients.

Conclusion

Patient-reported OD is a highly specific symptom of COVID-19 which should be included as part of the pre-test screening of suspect patients.