Prevalence of Chemosensory Dysfunction in COVID-19 Patients: A Systematic Review and Meta-analysis Reveals Significant Ethnic Differences

Abstract

A significant proportion of people who test positive for COVID-19 have chemosensory deficits. However, the reported prevalence of these deficits in smell and taste varies widely, and the reason for the differences between studies is unclear. We determined the pooled prevalence of such chemosensory deficits in a systematic review and meta-analysis. We searched the COVID-19 portfolio of the National Institutes of Health for studies that reported the prevalence of smell or taste deficits or both in patients diagnosed with COVID-19. One-hundred-four studies reporting on 38 198 patients qualified and were subjected to a systematic review and meta-analysis. Estimated random prevalence of olfactory dysfunction was 43.0%, that of taste dysfunction was 44.6%, and that of overall chemosensory dysfunction was 47.4%. We examined the effects of age, gender, disease severity, and ethnicity on chemosensory dysfunction. Prevalence of smell or taste dysfunction or both decreased with older age, male gender, and disease severity. Ethnicity was highly significant: Caucasians had a three times higher prevalence of chemosensory dysfunctions (54.8%) than Asians (17.7%). The finding of geographic differences points to two causes that are not mutually exclusive. A virus mutation (D614G) may cause differing infectivity, while at the host level genetic, ethnicity-specific variants of the virus-binding entry proteins may facilitate virus entry in the olfactory epithelium and taste buds. Both explanations have major implications for infectivity, diagnosis, and management of the COVID-19 pandemic.

Keywords: SARS-CoV-2; anosmia; ethnicity; prevalence; smell; taste.

Figure 1.

Flowchart of the search strategy, article selection, application of inclusion and exclusion criteria, and removal of duplicates according to the PRISMA guidelines.

Figure 2a-c.

Funnel plots of the prevalence of dysfunction of smell (a), taste (b), and smell and/or taste (c) in COVID-19 patients. Each dot represents a single study with the x-axis showing the logit transformed proportion of people in each study that lost their sense of (a) smell, (b) taste, and (c) smell and/or taste; the y-axis shows the standard error (SE) as a measure of precision. Most studies are large/precise (higher on y-axis), but medium- and small-size studies are also present (gaps in the y-axis are minor). Many studies fall outside of the 95% CI limit (dotted triangle), suggesting that study heterogeneity is high. There is some evidence that small-study effects are contributing to heterogeneity, as studies with larger SE show slightly smaller logit transformed proportions.

Figure 3a-c.

Forest plots of the prevalence of smell dysfunction (a), taste dysfunction (b), and smell and/or taste dysfunction (c) in COVID-19 patients. Estimated random proportions are shown by red boxes with 95% confidence intervals (95% CI) extending as whiskers, the overall estimated random proportion of subgroups is shown in gray, and the results for all studies combined are shown in black. Note the difference between Asians and Caucasians.

Figure 4a, b.

Prevalence of any chemosensory deficit (smell and/or taste) in COVID-19 patients. a. World map as a heat map showing the size and approximate location of cohorts. Three studies (from Germany, India, and Somalia) were inadvertently missed and not included in the meta-analysis, but their cohorts were added to the world map.^– b. Estimated random prevalence of chemosensory dysfunction - comparison of Caucasians and Asians. Error bars indicate 95% confidence intervals. Note the significant 3-fold difference in prevalence between Caucasian and Asian populations.

Figure 5a-b.

Comparison of the prevalence of chemosensory dysfunction between cohorts from Asia (a), and from Western countries (b). The trendline for studies from February through August 2020 is very slightly increasing, with a similar slope in Asia as in Western countries. The prevalence data based on subjective self-reporting is shown with blue dots, the data based on objective tests is shown with orange dots.