Five cluster classifications of long COVID and their background factors: A cross-sectional study in Japan

Abstract

Purpose

The long-term symptoms of coronavirus disease 2019 (COVID-19), i.e., long COVID, have drawn research attention. Evaluating its subjective symptoms is difficult, and no established pathophysiology or treatment exists. Although there are several reports of long COVID classifications, there are no reports comparing classifications that include patient characteristics, such as autonomic dysfunction and work status. We aimed to classify patients into clusters based on their subjective symptoms during their first outpatient visit and evaluate their background for these clusters.

Methods

Included patients visited our outpatient clinic between January 18, 2021, and May 30, 2022. They were aged ≥ 15 years and confirmed to have SARS-CoV-2 infection and residual symptoms lasting at least 2 months post-infection. Patients were evaluated using a 3-point scale for 23 symptoms and classified into five clusters (1. fatigue only; 2. fatigue, dyspnea, chest pain, palpitations, and forgetfulness; 3. fatigue, headache, insomnia, anxiety, motivation loss, low mood, and forgetfulness; 4. hair loss; and 5. taste and smell disorders) using CLUSTER. For continuous variables, each cluster was compared using the Kruskal–Wallis test. Multiple comparison tests were performed using the Dunn’s test for significant results. For nominal variables, a Chi-square test was performed; for significant results, a residual analysis was conducted with the adjusted residuals.

Results

Compared to patients in other cluster categories, those in cluster categories 2 and 3 had higher proportions of autonomic nervous system disorders and leaves of absence, respectively.

Conclusions

Long COVID cluster classification provided an overall assessment of COVID-19. Different treatment strategies must be used based on physical and psychiatric symptoms and employment factors.

Introduction

The coronavirus disease 2019 (COVID-19) pandemic has drawn research attention not only to the acute phase of the disease but also to its sequelae. Prolonged symptoms following severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) infection are collectively known as long COVID [1]. A cohort study in the Netherlands reported that 12.7% of patients develop sequelae following infection [2]. The most common symptoms are fatigue and dyspnea, followed by smell and taste disorders, headache, chest pain, brain fog and memory loss, and sleep disturbances [3]. Long COVID can also be a social problem because of its long-term impact on daily life. However, owing to the difficulty in evaluating long COVID subjective symptoms, which may be due to complex pathogeneses, it is difficult to conduct clinical trials [4], and there is no established pathophysiology or treatment for long COVID. Long COVID is multisystemic and presents different clusters of symptoms. Therefore, precise management and intervention strategies should be administered for different sequelae phenotypes in community COVID-19 survivors [5, 6].

Several studies have classified symptoms of long COVID into clusters. Kenny et al. reported a cluster analysis of long COVID; the clusters were divided into three categories: pain, cardiovascular symptoms, and the lowest median number of symptoms [7]. Wong-Chew et al. identified 45 symptoms across eight symptom clusters (neurological; mood disorders; systemic; respiratory; musculoskeletal; ear, nose, and throat; dermatological; and gastrointestinal). Although most symptoms decreased in frequency between days 30 and 90, the number of patients in the alopecia and the dermatological symptom cluster significantly increased [8]. Whitaker et al. classified long COVID into two clusters from a large survey in the UK. The first cluster had a high prevalence of persistent fatigue, which coincided with myalgia, difficulty in sleeping, and shortness of breath. The second cluster had a high prevalence of respiratory symptoms, such as shortness of breath, chest tightness, and chest pain [9]. Fernández-de-las-Peñas et al. showed that patients who had greater number of medical comorbidities had more COVID-19 symptoms at the acute phase, greater number of post-COVID symptoms, and more limitations with activities of daily living [10]. Fischer et al. classified long COVID symptoms by severity that persists for 12 months [11]. A report from China classified the symptoms into four clusters according to severity. The moderate group was classified into two categories: physical and mental symptoms [6].

Although there have been several reports of cluster analysis symptoms of long COVID, to our knowledge, there have been no studies that actually examined all patients in the outpatient clinic, evaluated their fatigue status in the acute phase and at the time of presentation, assessed their autonomic disorders and employment status, and investigated the background of their clusters. Therefore, in the present study, we aimed to perform a cluster classification of patients with long COVID and identify background factors for each classification obtained from face-to-face outpatients.

St. Marianna University Hospital started an outpatient clinic specializing in COVID-19 sequelae on January 18, 2021, and had treated 500 patients by May 30, 2022. Patients presented with various symptoms, such as fatigue, hair loss, mood depression, memory impairment, and taste and smell impairment.

This study classified patients into clusters based on their subjective symptoms at their first outpatient visit at this hospital and described the treatment course for each cluster, including patient quality of life during the 1st year following presentation at the hospital. We hypothesized that the clusters would have patients with different prognoses and treatment approaches according to this classification. We present a long COVID cluster classification system based on symptoms obtained from an initial interview and patient background evaluation for each cluster.

Materials and methods

Study design and setting

This descriptive study was conducted between January 18,2021, and May 30, 2022, at the outpatient clinic of St. Marianna University Hospital (hereafter referred to as “our hospital”). The long COVID outpatient clinic treats patients suspected of having COVID-19 sequelae and is referred from elsewhere.

Participants

The included patients were aged ≥ 15 years, had a SARS-CoV-2 infection confirmed using an antigen or polymerase chain reaction (PCR) test, and had residual symptoms lasting at least 2 months post-infection.

Patients with negative antigen and PCR tests in the acute phase were excluded.

Data collection

Twenty-three symptom categories from onset to the first visit were evaluated using a questionnaire: low-grade fever, fatigue, dyspnea, cough, taste disorder, smell disorder, hair loss, sore throat, joint pain, numbness in limbs, muscle pain, headache, dizziness, chest pain, palpitations, nausea, diarrhea, lack of motivation, insomnia, anxiety, depressed mood, forgetfulness, and skin symptoms. Patients were asked to mark “〇” for symptoms they were aware of and “◎” for major symptoms. For a more detailed evaluation of fatigue, patients were given a performance status (PS) table (Table 1), indicating the degree of symptoms of chronic fatigue syndrome, and were asked to select the degree of fatigue in the early stages of infection (approximately 1 week following symptom onset) and on the day of the first visit to the clinic. If the patient complained of fatigue, autonomic dysfunction was evaluated using the Schellong test. The patient was considered to have postural orthostatic tachycardia syndrome (POTS) when there was no decrease in blood pressure, the pulse rate increased by more than 30 beats/min, and the patient showed intolerance during the standing test [12]. We categorized patients as having clinical POTS when the pulse rate increased between 15 and 30 beats/min, and intolerance was observed during the orthostatic test.

Table 1.

PS criteria indicating the degree of symptoms of chronic fatigue syndrome

PS	Criteria
0	You can live normally without fatigue. You can act without restrictions
1	You can live and work normally, but may sometimes feel tired
2	You can live and work normally, but may sometimes need rest due to general fatigue
3	You cannot work for several days/month due to general fatigue
4	You cannot work for several days/week due to general fatigue
5	You cannot live and work normally. Light work is possible, but rest at home is required for several days/week
6	You can do light work on days when your condition is good, but you need to rest at home 3–4 days/week
7	You cannot live normally or do light work, but you can take care of yourself
8	You can take care of yourself at the minimum level, but often need help and lie down for more than 50% of the day
9	You cannot take care of yourself, always needing help and lying down all day long

PS Performance status

Furthermore, the following data were collected from all patients: body mass index (BMI), pneumonia complications in the acute phase, smoking status, comorbidities, and current employment status. Employment status was classified into four categories: continuing, change in job description, leave of absence, and retired. If the patient was a housewife or student, the status was treated as “continuing” if the patient experienced no difficulty in performing housework or commuting to school; “change in job description” if the patient was performing housework but taking regular rest breaks or commuting to school a few times a week; and “leave of absence” if the patient was unable to do housework or commute to school.

Statistical analyses

Continuous variables are summarized using median and interquartile range and categorical variables as frequency and percentage. Cluster classification of post-SARS-CoV-2 infection symptoms was performed based on symptoms described in the questionnaire at the time of the hospital visit. Input variables for clustering were 23 patient symptom variables rated at three levels: 0, no symptoms; 1 (“〇”) for mild symptoms; and 2 (“◎”) for major symptoms. Cluster analysis was performed using CLUSTER (SAS Ver 9.4, SAS Institute Inc., Cary, NC, USA), and cluster classification was performed. This classification and symptoms were used in accordance with actual clinical practice. Differences in patient characteristics according to cluster classifications were assessed as follows. Binary variables were subjected to the Chi-square test followed by residual analysis; continuous variables (i.e., 0, 1, and 2) were subjected to the Kruskal–Wallis test followed by multiple comparisons using the Dunn’s test. A p value < 0.05 was considered statistically significant. All statistical analyses were performed using STATA version 15 software (Stata Corporation, College Station, TX, USA).

Results

Patient characteristics

In total, 500 patients were included. Among them, 497 patients were analyzed, excluding two patients for whom antigen and PCR tests were not performed at the time of the interview and one who was an outlier in the cluster analysis.

Some patients had mild anemia or mild abnormality of thyroid function on blood tests at the initial visit, and some had suspected recurrent depression on interview. However, it was difficult to determine whether these abnormalities were associated with infection; therefore, these patients were not excluded. The included patients comprised 214 male (43.1%) and 283 female (56.9%) patients (average age, 41.6 years). Of them, 113 cases (23.7%) were complicated by pneumonia. Figure 1 shows all symptoms (〇 + ◎) declared at the initial visit (multiple answers allowed), and Fig. 2 shows the major symptoms.

Fig. 1.

Symptoms declared at the time of initial examination. Fatigue (59.8%) was the most common symptom, followed by anxiety (42.3%), smell disorders (41.9%), depressed mood (40.2%), headache (38.6%), lack of motivation (38.4%), and forgetfulness (37.8%)

Fig. 2.

Strong subjective symptoms at the time of initial examination. Fatigue (40.2%) was the most common symptom, followed by smell disorders (26.6%), taste disorders (18.1%), hair loss (14.9%), dyspnea (13.7%), and headache (11.1%)

Cluster analysis and patient characteristics between clusters

Table 2 shows the cluster analysis results. Five clusters were classified according to actual clinical practice, and the symptoms of each cluster were defined as those with a value of ≥ 1.8. The patients’ symptoms are presented in Table 3. Patients in clusters 1–3 had “fatigue” in common, and those in clusters 2 and 3 had “forgetfulness” in common. Patients in cluster 4 experienced only hair loss, and those in cluster 5 only had smell and taste disorders.

Table 2.

Cluster analysis results

Cluster	1	2	3	4	5
Low-grade fever	1.296	1.227	1.184	1.061	1.035
Fatigue	2.778	2.587	2.505	1.082	1.195
Dyspnea	1.519	2.440	1.233	1.367	1.053
Cough	1.287	1.280	1.262	1.204	1.018
Taste disorder	1.204	1.307	1.388	1.051	2.398
Smell disorder	1.278	1.373	1.534	1.082	2.929
Hair loss	1.185	1.413	1.408	1.878	1.372
Sore throat	1.074	1.213	1.165	1.041	1.027
Joint pain	1.139	1.547	1.282	1.061	1.009
Numbness in limbs	1.056	1.320	1.379	1.061	1.027
Muscle pain	1.204	1.400	1.262	1.092	1.009
Headache	1.491	1.707	1.990	1.265	1.106
Dizziness	1.185	1.453	1.641	1.092	1.044
Chest pain	1.148	1.880	1.291	1.204	1.053
Palpitations	1.167	2.200	1.388	1.102	1.035
Nausea	1.056	1.280	1.184	1.041	1.044
Diarrhea	1.065	1.187	1.252	1.020	1.018
Lack of motivation	1.361	1.547	1.990	1.102	1.133
Insomnia	1.185	1.400	1.903	1.112	1.080
Anxiety	1.296	1.747	2.243	1.204	1.168
Depressed mood	1.315	1.560	2.117	1.133	1.150
Forgetfulness	1.324	1.827	1.864	1.337	1.177
Skin symptoms	1.037	1.013	1.039	1.051	1.000

Table 3.

Patient symptoms in clusters

Cluster	No.	Symptoms
1	108	Fatigue
2	74	Fatigue	Dyspnea	Chest pain	Palpitations	Forgetfulness
3	103	Fatigue	Headache	Lack of motivation	Insomnia	Anxiety	Depressed mood	Forgetfulness
4	98	Hair loss
5	113	Taste disorder	Smell disorder

Patient backgrounds for each cluster

Table 4 shows the patient backgrounds in all clusters. The median age in cluster 4 was higher than that in the other clusters, with a higher proportion of female patients in clusters 2 and 4; BMI was not significantly different between the clusters. Pneumonia complications in the acute phase were more common among patients in cluster 2 and less common among those in cluster 5. Comorbidities were less common in cluster 5. Regarding employment status, patients in clusters 2 and 3 were more likely to take leaves of absence. Patients in clusters 4 and 5 were more likely to continue in their jobs. Patients in cluster 5 had significantly more days from symptom onset to hospital visit. PS at onset was the highest in patients in cluster 2, followed by those in clusters 1 and 3; it was the lowest among those in clusters 4 and 5. PS at initial presentation was higher among patients in clusters 1–3 than among those in clusters 4 and 5. Cases in cluster 2 were more frequently complicated with clinical POTS.

Table 4.

Patient backgrounds in all clusters

		Cluster 1		Cluster 2		Cluster 3		Cluster 4		Cluster 5		p value
		N = 108	R	N = 74	R	N = 103	R	N = 98	R	N = 113	R
	Age, years [median (IQR)]	40 (29–50)		43 (33–51)		44 (29–50)		46 (35–55)		40 (26–52)		0.006
	Sex (male) [n (%)]	53 (49.1%)	1.41	24 (32.4%)	− 2.02	52 (50.5%)	1.69	46 (46.9%)	0.85	39 (34.5%)	− 2.11	0.020
	BMI [median (IQR)]	22 (18–25)		22 (20–32)		22 (18–23)		24 (20–29)		21 (21–22)		0.74
	Pneumonia complication of acute phase [n (%)]		− 1.88		2.28		0.65		1.99		− 2.47	0.012
	Absence	55 (72.4%)		19 (46.3%)		41 (58.6%)		34 (54.0%)		54 (77.1%)
	Presence	20 (26.3%)		21 (51.2%)		27 (38.6%)		29 (46.0%)		16 (22.9%)
	Unknown	1 (1.3%)		1 (2.4%)		2 (2.9%)		0 (0.0%)		0 (0.0%)
	Comorbidity [n (%)]	65 (61.3%)	0.58	48 (64.9%)	1.14	68 (66.7%)	1.8	62 (64.6%)	1.27	46 (40.7%)	− 4.47	< 0.001
	Status of employment [n (%)]											< 0.001
	continuing	54 (50.0%)	− 2.57	31 (41.9%)	− 3.59	45 (43.7%)	− 3.97	76 (77.6%)	3.82	95 (84.1%)	5.79
	Change in job description	26 (24.1%)	4.75	10 (13.5%)	0.66	10 (9.7%)	-0.57	2 (2.0%)	− 3.23	8 (7.1%)	− 1.61
	Leave of absence	22 (20.4%)	− 0.73	27 (36.5%)	2.99	41 (39.8%)	4.56	16 (16.3%)	− 1.75	8 (7.1%)	− 4.57
	Retirement	6 (5.6%)	0.28	6 (8.1%)	1.31	7 (6.8%)	0.92	4 (4.1%)	− 0.48	2 (1.8%)	− 1.81
	Smoking status [n (%)]											0.015
	Never	74 (68.5%)	− 0.42	46 (62.2%)	− 1.63	71 (68.9%)	− 0.31	69 (70.4%)	0.06	88 (77.9%)	2.04
	Former smoker	29 (26.9%)	1.14	24 (32.4%)	2.15	17 (16.5%)	− 1.71	23 (23.5%)	0.18	20 (17.7%)	− 1.47
	Current smoker	5 (4.6%)	− 1.11	4 (5.4%)	− 0.6	15 (14.6%)	3.34	6 (6.1%)	− 0.4	5 (4.4%)	− 1.24
	Days from symptoms onset to hospital visit [median (IQR)]	108 (69–178)		110 (78–168)		116 (79–184)		112 (82–134)		134 (104–194)		0.004
	PS at the time of infection [median (IQR)]	7 (5–8)		7 (6–8)		7 (3–8)		6 (1–8)		3 (1–7)		< 0.001
	PS at the time of visit [median (IQR)]	2 (1–5)		4 (2–6)		3 (2–5)		0 (0–1)		0 (0–1)		< 0.001
	POTS [n (%)]											< 0.001
0	Absence	78 (72.9%)	− 1.5	40 (54.1%)	− 5.46	74 (71.8%)	− 1.76	84 (90.3%)	3.15	104 (95.4%)	4.94
1	Clinical judgment	16 (15.0%)	0.69	25 (33.8%)	5.79	17 (16.5%)	1.21	3 (3.2%)	-3.11	2 (1.8%)	− 3.93
2	Presence	13 (12.1%)	1.36	9 (12.2%)	1.09	12 (11.7%)	1.23	6 (6.5%)	− 0.91	3 (2.8%)	− 2.54

BMI Body mass index; PS Performance status; POTS Postural orthostatic tachycardia syndrome; R Residents

Discussion

In this study, long COVID symptoms were classified into five clusters. Patients with fatigue were classified into clusters 1, 2, and 3, which consisted of patients with fatigue alone; fatigue with strong physical symptoms of dyspnea, chest pain, and palpitations; and fatigue with strong psychological symptoms of insomnia, anxiety, loss of motivation, and low mood, respectively. Cluster 4 included patients with mainly hair loss, and cluster 5 included those with taste and smell disorders.

Fatigue has been reported to occur in 23–42% of patients evaluated from 4 to 8 weeks following the onset of COVID-19 [13] and is the most common complaint among the sequelae [3]. Herein, fatigue was the most frequently reported symptom generally and among the strong symptoms at the initial visit, with a median PS (median) of 2, 4, and 3 for clusters 1, 2, and 3, respectively, and was significantly stronger than that for clusters 4 and 5.

In the present study, 11–12.2% of the patients in clusters 1–3 met the definition of POTS; cluster 2 included 33% of patients considered to have clinical POTS. Autonomic dysfunction has been reported to play a role in long COVID [14], and POTS evaluation in patients with long COVID has been recommended [15, 16]. The results of the present study mirror the results of the previous studies; because POTS is therapeutic [17], it is important to perform orthostatic testing in patients with long COVID who complain of fatigue.

Cluster 3 included many psychological symptoms, which have been reported to occur in 62.3% of COVID-19 cases. It is possible that underlying mental illness may be exacerbated by COVID-19 [18]. Notably, cluster 3 had the highest percentage of leaves of absence in our study. Financial concerns owing to symptoms have been reported to be associated with an increase in both anxiety and depression [19]. In addition, in a previous study, which investigated “post-COVID-19 status, unspecified” using the International Classification of Diseases, 10th revision, long COVID patients were skewed toward those living in areas of low poverty and unemployment [20]. These results are consistent with those of our study in which a large number of leaves of absence occurred in patients in cluster 3.

Brain fog is a lay term often used to describe this cognitive dysfunction and may incorporate symptoms, such as concentration issues, word-finding difficulties, memory impairment, or disorientation [21]. Brain fog has been reported to occur in 7.2% of long COVID cases and is also associated with severe fatigue [21, 22]. This result is consistent with our study, in which cluster classifications 2 and 3 resulted in a combination of brain fog and fatigue. It is possible that the reason patients took leaves of absence was not only because of fatigue but also because of brain fog. Clusters 2 and 3 are dominated by symptoms of physical and psychological mental health impairment. These findings are similar to those of a previous study in China that found four sequelae clusters [6].

Ili et al. [23] showed that fatigue caused by COVID-19 may be related to autonomic dysfunction, impaired cognition, and decreased mood. Similarly, in the present study, clusters 1, 2, and 3 were characterized by fatigue, which may be related to POTS (autonomic dysfunction) and brain fog (impaired cognition), as well as psychological symptoms, including depressed mood. It is assumed that all clusters have a similar background of medical conditions and that the cluster classification was based on the symptom duration and severity and employment status. Clusters with fatigue as a major symptom were classified based on the severity of other symptoms, suggesting that social factors, such as work restrictions, may also be involved, especially in cluster 3. Sandler et al. [12] stated that assessing long COVID fatigue requires a response using the Bio-Psycho-Social model. The results of our study suggest that a multifaceted approach is necessary when evaluating patients who complain of fatigue as a result of long COVID, including psychological and social evaluation and care.

Patients in clusters 4 and 5 were classified as having no noticeable fatigue. In cluster 4, hair loss was the main symptom; although the median value of PS was as high as 6 in the early stages of infection, it improved to a median value of 0 at the time of the first visit. Various sequelae symptoms tended to improve 50 days following the acute phase. However, hair loss tends to start 2 months following the onset of symptoms because hair loss in long COVID is caused by telogen effluvium [24, 25]. This cluster could be considered to be a group of patients who were referred to the clinic because of strong symptoms other than hair loss, including fatigue, during the acute phase, but whose other symptoms improved and only hair loss remained. Cluster 5 had symptoms of taste and smell disorders. A large proportion of patients in this cluster had no complications of pneumonia in the acute phase or comorbidity and a significant amount of time from symptom onset to hospital visit. In addition, the median PS at onset was 2 and 0 at the time of the hospital visit, indicating that fatigue did not greatly interfere with daily life from the acute phase. Patients in the other categories felt more fatigue during the acute phase; therefore, the pathogenic mechanism of taste and smell disorders as sequelae may be different from that of the other symptom categories. This study revealed a relationship between the classification of symptoms and patient background. The most common symptom, fatigue, was found to be associated with autonomic dysfunction from POTS, brain fog, and psychological symptoms that force patients to take leave of absence. In contrast, there were cases such as those in cluster 4, in which symptoms improved spontaneously, and cluster 5, in which taste and smell disorders persisted only in isolation, indicating that sequelae cannot be explained in a uniform manner.

Our study results provide several implications regarding long COVID not only for healthcare workers but also for the general population. First, these findings indicate that a certain number of patients with long COVID complaining of fatigue also have POTS and psychiatric symptoms. Although the symptoms of long COVID vary, the findings of this study may lead to appropriate management. Second, the results of this study will allow us to provide more accurate long-term information for the general population. Providing accurate information to the public and patients could help alleviate their anxiety. Finally, the results of this study may be useful for a wide range of healthcare workers and others not involved in healthcare.

Study limitations

The limitations of this study must be acknowledged. First, this was a cross-sectional, single-center study conducted at a university hospital. In Japan, the number of university hospitals, such as where the study took place, is small, and few patients with long COVID are referred to university hospitals; thus, patients treated at our hospital may have represented a selection bias. Second, our outpatient clinic requires face-to-face consultations; thus, we were unable to evaluate a group of patients with severe fatigue who could not leave their homes. A study with a larger number of patients is warranted to ensure the generalizability of our findings. Finally, the variant of COVID-19 differs according to the time of infection. Smell disorders are less common in patients infected with the omicron variant [26]; therefore, it is possible that performing cluster classification for each infection period would lead to different results.

Conclusions

Long COVID symptoms were classified into five categories: (1) fatigue alone, (2) fatigue, dyspnea, chest pain, palpitations, and forgetfulness; (3) fatigue, headache, insomnia, anxiety, loss of motivation, depressed mood, and forgetfulness; (4) hair loss; and (5) taste and smell disorders. When fatigue is present, autonomic dysfunction and psychiatric symptoms should be evaluated, and social issues, such as employment status, should also be addressed.

Authors’ contributions

TT and TM helped in conceptualization, NY, MH, and HS helped in methodology, NY, MH, and HS worked in formal analysis and investigation, TT helped in writing—original draft preparation, KI, KK, YN, and YK helped in writing—review and editing, TT, KI, KK, YI, and MH worked in resources, and YO worked in supervision.

Funding

The authors declare that no funds, grants, or other support were received during the preparation of this manuscript.

Data availability

The data that support the findings of this study are available upon request from the corresponding author. The data are not publicly available due to privacy reasons.

Declarations

Conflict of interest

The authors have no relevant financial or non-financial interests to disclose.

Ethical approval

The questionnaire and methodology for this study was approved by the ethics committee of Osaka Metropolitan University Graduate School of Medicine (Approval Number: 2020–003).

Consent to participate

Adult patients (≥ 18 years) and minor patients and their parents provided written informed consent for initial and follow-up data collection as part of routine treatment. Informed consent was obtained from patients who visited our hospital before the study was approved by the ethics committee using an online opt-out form.