Distinct smell and taste disorder phenotype of post-acute COVID-19 sequelae

The ‘Health after COVID-19 in Tyrol’ online survey (ClinicalTrials.gov: NCT04661462), further referred to as ‘survey study’, encompassed two cohorts of adult ambulatory, non-hospitalized individuals with laboratory-confirmed SARS-CoV-2 infection independently recruited in western Austria (AT, Tyrol province) and northern Italy (IT, South Tyrol province) [1]⁠. The survey study was conducted during outbreaks of the wild-type and alpha variant of SARS-CoV-2 (30th September 2020–5th July 2021). COVID-19 convalescents were invited to participate by a public media call (AT and IT) or by general practitioners (IT). The survey inclusion criteria were PCR- or seropositivity-confirmed SARS-CoV-2 infection, residency in the study regions and adult age (AT: ≥ 16, IT ≥ 18 years). The survey exclusion criterion was COVID-19-related hospitalization [1, 29]⁠. The participants with symptomatic COVID-19 and ≥ 90 days between diagnosis and survey completion were included in the analysis (AT: n = 479, IT: n = 427, Supplementary Fig. S1).

The CovILD observational cohort (NCT04416100), further referred to as ‘CovILD cohort’, included COVID-19 outpatients and inpatients recruited at the Department of Internal Medicine II at the Medical University of Innsbruck, St. Vinzenz Hospital in Zams and Karl Landsteiner Rehabilitation Facility in Muenster (all in Austria) during the wild-type SARS-CoV-2 outbreak (March–June 2020) [2, 5, 28, 29]⁠. The participants who had completed all scheduled follow-ups (60-, 100-, 180- and 360-day after diagnosis) were included in the analysis (n = 108, Supplementary Fig. S1).

The studies were conducted in accordance with the Declaration of Helsinki and European Data Policy. All participants gave written informed consent to participate. The study protocols were approved by the ethics committees of the Medical University of Innsbruck (survey study, AT, approval number: 1257/2020, CovILD cohort: 1103/2020) and of the Autonomous Province of Bolzano—South Tyrol (survey study, IT: 0150701).

In the survey study, participants assigned 42 self-reported symptoms to duration intervals (absent: re-coded as 0 days, 1–3 days: 3 days, up to 1 week: 7 days, up to 2 weeks: 14 days, up to 4 weeks: 28 days, up to or greater than 3 months: 3 months). Acute symptoms were defined as complaints present in the first 14 days after clinical onset. Self-perceived complete recovery, need for rehabilitation and new drugs following COVID-19 were surveyed as single yes/no items. Physical performance loss following COVID-19 was rated with a 0–100% scale [3]⁠. Quality of life and overall mental health impairment (4-item Likert scale each) [3]⁠, anxiety and depression (Patient Health Questionnaire, PHQ-4) [30]⁠ and psychosocial stress (7 item PHQ stress module) [31]⁠ were assessed at the time of survey completion.

In the CovILD cohort, 8 self-reported symptoms (reduced physical performance, OD, dyspnea, sleep problems, cough, fever, night sweating, gastrointestinal symptoms) were recorded at each follow-up. Acute symptoms were assessed retrospectively at the 60-day follow-up [2, 5]⁠. Objective OD was evaluated with the 16-item Sniffin’ Sticks Identification test (Burghart Medizintechnik, Wedel, Germany) at the 100- (n = 95) and 360-day follow-up (n = 63). The nasal chemosensory performance was investigated using pen-like odor-dispensing devices for odor identification of 16 common odorants (multiple forced‐choice from four verbal items per test odorant). Hyposmia was defined for < 13 correctly identified odorants, as per manufacturer criteria [32‐36]⁠.

Study variables are listed in Supplementary Table S1 (survey study) and S2 (CovILD cohort).

The primary analysis endpoint for the ambulatory survey study was the frequency of self-reported OD up to three months after clinical onset of COVID-19. The primary analysis endpoint for the CovILD cohort was the frequency of self-reported OD up to 1 year after after COVID-19 diagnosis.

The secondary analysis endpoint for the CovILD cohort was the comparison of rates of subjective and objective OD. The secondary analysis endpoint for the survey study were co-occurrence of self-reported OD with other symptoms and impact of self-reported OD on self-perceived recovery, physical performance, quality of life and mental health.

Data analysis was conducted with R, version 4.2.3. Numeric variables were presented as medians with interquartile ranges and ranges. Categorical variables are presented as percentages and counts within the complete observation set. Statistical significance was investigated by Mann–Whitney U or paired Wilcoxon test with r effect size statistic, Kruskal–Wallis test with η² effect size statistic, χ² test with Cramer’s V effect size statistic, Cochran Q test with Kendall’s W effect size statistic or McNemar test with Cohen’s g effect size statistic. Subjective and objective OD were compared with Cohen’s κ statistic. Two-dimensional mapping of simple matching distances between symptoms was done by multi-dimensional scaling. Apriori association rule analysis was conducted with arules package [37]. Data sets of symptom-specific recovery times in the AT and IT cohorts of the survey study demonstrated good clustering tendency assessed by Hopkins statistic (H; H = 0: uniform distribution, H = 1 highly clustered data; AT: H = 0.80, IT: H = 0.79). Sizes of the AT (n = 479) and IT cohort (n = 427) were sufficient for reliable clustering analysis as investigated by clustering tendencies of random subsets of the pooled AT/IT recovery time data set (minimal cohort size n = 400, Supplementary Fig. S2). Clustering of the training AT cohort by symptom-specific recovery times was done with the PAM algorithm (partitioning around medoids, Euclidean distance) [38]⁠. Assignment of the test IT cohort participants to the clusters was accomplished with the inverse distance-weighted 7-nearest neighbors classifier [39]⁠. P values were corrected for multiple testing with Benjamini–Hochberg method. For details of the statistical analysis, see Supplementary Methods.

Out of 3140 survey study respondents [1]⁠, 906 non-hospitalized individuals with symptomatic COVID-19 and ≥ 90 days after diagnosis were analyzed. Those individuals were grouped in two cohorts independently recruited in western Austria (AT, n = 479) and northern Italy (IT, n = 427), respectively (Supplementary Fig. S1). The median follow-up time ranged from 140 (IT) to 180 days (AT). The cohorts consisted primarily of working-age individuals (AT: median 43 [IQR: 32–53]; IT: 45 [34–54] years) and females were over-represented (AT: 67%, IT: 70%). Almost half of participants had at least one comorbidity (AT: 49%, IT: 43%), with  hay fever/allergy, arterial hypertension, and obesity (body mass index > 30 kg/m²) as leading conditions (Table 1). AT cohort was characterized by a significantly longer follow-up time. Most AT cohort participants were infected during the spring 2020 outbreak (AT: 59%, IT: 30%), whereas IT cohort individuals were infected predominantly during the summer/fall 2020 SARS-CoV-2 waves (AT: 40%, IT: 69%). The IT collective was also characterized by a significantly lower fraction of obese or overweight participants, lower rates of daily medication, allergies and less participants suffering from frequent bacterial infections prior to COVID-19. Scores of anxiety, depression and impairment of quality of life [3]⁠ were significantly higher in the IT than AT cohort (Table 1, Supplementary Fig. S3).

CovILD study participants with the complete 1-year COVID-19 follow-up (n = 108, Supplementary Fig. S1) were predominantly male (59%). The median age was 56 (IQR: 49–68) years and 75% of the participants had comorbidities such as obesity, cardiovascular or pulmonary disease or type 2 diabetes mellitus. The CovILD study participants were stratified as mild (outpatients, 25%), moderate (inpatients, no ICU, 51%) and severe COVID-19 convalescents (ICU, 24%). The median age and comorbidity rates were significantly higher in moderate or severe than in mild COVID-19 patients (Table 2).

During acute infection of wild-type or alpha variant of SARS-CoV2, subjective OD (AT: 70%, IT: 75%) and self-reported taste disorders (AT: 68%, IT: 74%) along with fatigue, tiredness, diminished appetite, joint pain, tachypnea, cough, and fever were present in most survey participants (Supplementary Fig. S4). Fever, muscle and bone pain, shivering and eye redness, confusion, walk problems and tingling hands in the first 28 days of COVID-19 were significantly more frequent in the IT than AT cohort. In turn, dizziness was more common in the AT collective (Supplementary Fig. S5). Most upper airway and infection symptoms resolved within the first 14 days after clinical onset. Self-reported OD (median recovery: 28 days) and taste disorders (14 days) resolved substantially slower. Accordingly, self-reported OD (AT: 30%, IT: 27%) and taste disorders (AT: 22%, IT: 21%) affected more than one-fifth of convalescents 90 days after COVID-19 onset. Other symptoms with prolonged recovery included memory and concentration problems, tachypnea, tiredness and fatigue (Fig. 1, Supplementary Figs. S4 and S6, Supplementary Table S3).

Self-reported OD was frequent during acute COVID-19 in both inpatients and outpatients of the CovILD cohort (mild: 47%, moderate: 33%, severe COVID-19: 53%). The median recovery time across all disease severity strata was 90 days. Yet, subjective OD affected 16% of mild or moderate CovILD cohort individuals at the 360-day follow-up. By contrast, none of the severe COVID-19 patients reported subjective OD at the 180-day follow-up and beyond (Supplementary Figs. S7, S8, Supplementary Table S4).

Objective OD was assessed in a subset of the CovILD cohort with the 16-item Sniffin’ Stick Test at the 3-month (n = 95) and 1-year follow-up (n = 63). Objective OD was diagnosed for < 13 odorants identified correctly [32‐36]⁠. Objective OD in the entire analyzed CovILD cohort subsets (3-months: 45%, 1 year: 54%) and in their COVID-19 severity strata did not change over time. Furthermore, frequency of objective OD in the entire analyzed CovILD cohort subsets was higher than self-reported OD both at the 3-month (objective: 45%, subjective: 18%) and 1-year follow-up (objective: 54%, subjective: 9.5%). This was also true for each COVID-19 severity strata. Accordingly, the overall concordance between subjective and objective OD was low, as measured by Cohen’s κ (three months: κ = 0.25, 1 year: κ = 0.11). The highest concordance between objective and subjective OD was observed in moderate COVID-19 patients (three months: κ = 0.32, 1 year: κ = 0.29). Strikingly, objective OD affected 71% of severe COVID-19 survivors at the 1-year follow-up, while none of those individuals reported subjective OD (κ = 0) (Fig. 2). The discrepancy between self-reported and objective OD was corroborated by receiver-operating characteristic. For both investigated follow-ups, subjective OD was found to be an insensitive readout of objective OD, regardless of COVID-19 severity (three months: sensitivity 0.1–0.33, one year: sensitivity 0–0.31, Supplementary Figs. S9, S10).

Subsequently, we investigated changes in OD rating by Sniffin’ Stick Test and objective OD in CovILD study participants with complete testing results at both the 3-month and 1-year follow-up (n = 56, Supplementary Table S5). In an analysis of participant-matched testing results, we observed significant differences neither in numbers of correctly identified odorants nor in frequency of objective OD. Interestingly, 26.8% of all longitudinally investigated patients were diagnosed with objective OD at the 1-year follow-up despite normal olfactory function at the 3-month follow-up, suggestive of recurring or COVID-19-independent OD (Supplementary Fig. S11).

During acute COVID-19 in the survey study, self-reported OD was frequently accompanied by taste disorders and multiple non-specific infection symptoms such as diminished appetite, rhinitis and sore throat, as found by multi-dimensional scaling (Supplementary Fig. S12). During recovery, this association disappeared owing to the resolution of most acute infection symptoms. At 28 days and three months after clinical onset, subjective OD and taste disorders were mapped close to each other and far from other leading post-acute symptoms including tachypnea, fatigue, memory and concentration deficits (Fig. 3). This indicates that, post-acute subjective OD and taste disorders are rarely accompanied by other persistent symptoms.

The co-occurrence of OD and taste disorders was supported by association rule mining [40]⁠ in the survey study, where ≥ 82% and ≥ 66% of participants reporting OD at 28 days and three months after clinical onset, respectively, were affected by taste disorders as well. Furthermore, the combination of subjective OD and taste disorders were found in one-third of survey participants at 28 days and one-fifth of participants at three months after COVID-19 onset (Supplementary Fig. S13).

Three subsets of COVID-19 convalescents were identified by PAM clustering [38]⁠ of the training AT survey cohort by resolution times of particular symptoms. As assessed by permutation analysis, tiredness, fatigue, concentration and memory problems, tachypnea, self-reported OD and taste disorders were the most important symptoms for the cluster definition (Supplementary Fig. S15). Subsequently, the IT test cohort individuals were assigned to the clusters with a nearest-neighbor classification procedure [39]⁠. The clustering scheme was highly reproducible as inferred from comparable fractions of explained clustering variances (AT: 0.59, IT: 0.56) and similar distribution of the cluster sizes in the training AT and test IT cohort (Supplementary Fig. S16).

The hallmark of cluster #1 (AT: 21%, IT: 15% of participants) was the slowest resolution of self-reported OD and taste disorders. Most other symptoms including tiredness, fatigue, respiratory, memory and concentration difficulties resolved rapidly in cluster #1. The largest cluster #2 comprised > 50% of participants (AT: 51%, IT: 56%) and was characterized by both the fastest resolution of all surveyed symptoms and the lowest number of acute COVID-19 symptoms. By contrast, cluster #3 (AT: 28%, IT: 29%) displayed the slowest resolution of fatigue, tiredness, tachypnea, memory and concentration deficits and the highest number of acute and post-acute COVID-19 complaints (Fig. 4, Supplementary Fig. S16).

Concerning demographic and clinical background, cluster #3 included the oldest participants with the highest comorbidity and daily medication rates. Clusters #1 and #2 consisted of substantially younger participants with similarly low frequencies of comorbidities. Females were significantly over-represented in clusters #1 (AT: 79%, IT: 84%) and #3 (AT: 76%, IT: 77%) (Supplementary Fig. S17, Supplementary Tables S6, S7).

Weight and physical performance loss as well as rates of new medication and need for rehabilitation were highest in cluster #3. Furthermore, cluster #3 individuals had the worst scoring of anxiety, depression, mental stress, self-reported mental health and quality of life at the time of survey completion. In comparison, those psychometric readouts were significantly better in cluster #1 despite slow recovery from OD and taste deficits and in cluster #2 with the fastest symptom resolution (Fig. 5, Supplementary Tables S8, S9).