Introduction
Globally, there have been more than 645 million confirmed cases of COVID-19 registered since the start of the pandemic [1]. An estimated 4–12% of patients experience long-term symptoms after infection with SARS-CoV-2; this is referred to as long COVID or post-COVID-19 condition (PCC) [Reference Ballering, van Zon, olde Hartman and Rosmalen2, 3]. PCC is defined by the World Health Organization (WHO) as ‘the continuation or development of new symptoms 3 months after the initial SARS-CoV-2 infection, with these symptoms lasting for at least 2 months with no other explanation’ [4]. PCC can be prevalent after both symptomatic and asymptomatic COVID-19 infections [Reference Adler, Gazit, Pinto, Perez, Mizrahi Reuveni, Yehoshua, Hoffman, Azuri and Patalon5, Reference Doykov, Hällqvist, Gilmour, Grandjean, Mills and Heywood6]. The global health burden of PCC is growing [Reference Faghy, Owen, Thomas, Yates, Ferraro, Skipper, Barley-McMullen, Brown, Arena and Ashton7], as well as its associated costs, and is estimated to have a prevalence of over 144 million individuals affected to date [Reference Wulf Hanson, Abbafati, Aerts, Al-Aly, Ashbaugh, Ballouz, Blyuss, Bobkova, Bonsel, Borzakova, Buonsenso, Butnaru, Carter, Chu, De Rose, Diab, Ekbom, El Tantawi, Fomin, Frithiof, Gamirova, Glybochko, Haagsma, Javanmard, Hamilton, Harris, Heijenbrok-Kal, Helbok, Hellemons, Hillus, Huijts, Hultström, Jassat, Kurth, Larsson, Lipcsey, Liu, Loflin, Malinovschi, Mao, Mazankova, McCulloch, Menges, Mohammadifard, Munblit, Nekliudov, Ogbuoji, Osmanov, Peñalvo, Petersen, Puhan, Rahman, Rass, Reinig, Ribbers, Ricchiuto, Rubertsson, Samitova, Sarrafzadegan, Shikhaleva, Simpson, Sinatti, Soriano, Spiridonova, Steinbeis, Svistunov, Valentini, van de Water, van den Berg-Emons, Wallin, Witzenrath, Wu, Xu, Zoller, Adolph, Albright, Amlag, Aravkin, Bang-Jensen, Bisignano, Castellano, Castro, Chakrabarti, Collins, Dai, Daoud, Dapper, Deen, Duncan, Erickson, Ewald, Ferrari, Flaxman, Fullman, Gamkrelidze, Giles, Guo, Hay, He, Helak, Hulland, Kereselidze, Krohn, Lazzar-Atwood, Lindstrom, Lozano, Magistro, Malta, Månsson, Mantilla Herrera, Mokdad, Monasta, Nomura, Pasovic, Pigott, Reiner, Reinke, Ribeiro, Santomauro, Sholokhov, Spurlock, Walcott, Walker, Wiysonge, Zheng, Bettger, Murray and Vos8].
Those with PCC generally experience a wide variety of symptoms [Reference Ani, Amar and Elaine9], with the clinical manifestation differing among patients [Reference Deer, Rock, Vasilevsky, Carmody, Rando, Anzalone, Basson, Bennett, Bergquist, Boudreau, Bramante, Byrd, Callahan, Chan, Chu, Chute, Coleman, Davis, Gagnier, Greene, Hillegass, Kavuluru, Kimble, Koraishy, Köhler, Liang, Liu, Liu, Madhira, Madlock-Brown, Matentzoglu, Mazzotti, McMurry, McNair, Moffitt, Teshamae, Monteith, Parker, Perry, Pfaff, Reese, Saltz, Schuff, Solomonides, Solway, Spratt, Stein, Sule, Topaloglu, Vavougios, Wang, Haendel and Robinson10]. Frequently reported symptoms include shortness of breath, fatigue, increased heart rate, and cognitive problems [Reference Ani, Amar and Elaine9]; however, patients can experience many other symptoms that vary in severity [Reference Nasserie, Hittle and Goodman11, Reference Salamanna, Veronesi, Martini, Landini and Fini12]. The multitude of symptoms reported by patients involve different organ systems [Reference del Rio, Collins and Malani13], and the underlying mechanisms as to why patients experience long-term symptoms after COVID-19 are still unknown [Reference Nehme, Braillard, Chappuis, Courvoisier and Guessous14]. However, several hypotheses are made including immune dysregulation, microbiota disruption, autoimmunity and immune priming, blood clotting and endothelial abnormalities, and dysfunctional neurological signalling [Reference Davis, McCorkell, Vogel and Topol15]. PCC symptoms can last for months or years, and it is unclear whether they disappear [Reference Han, Zheng, Daines and Sheikh16]. In addition to physical symptoms, PCC affects mental well-being, with higher rates of mood disorders and anxiety disorders being reported in patients with PCC than in patients with other respiratory tract infections or influenza [Reference Taquet, Geddes, Husain, Luciano and Harrison17]. The impact on mental well-being can only partially be explained by the disruptiveness and unpredictability of the pandemic and preventive and protective measures against the spread of COVID-19, such as quarantine and social distancing [Reference Penninx, Benros, Klein and Vinkers18]. The acute and post-acute neuropsychiatric sequelae seen in patients with COVID-19 also profoundly impact mental well-being [Reference Penninx, Benros, Klein and Vinkers18, Reference Crook, Raza, Nowell, Young and Edison19]. Furthermore, long-term symptoms of COVID-19 negatively impact the health-related quality of life (HRQoL) [Reference Malik, Patel, Pinto, Jaiswal, Tirupathi, Pillai and Patel20, Reference Fischer, Zhang, Elbéji, Wilmes, Oustric, Staub, Nazarov, Ollert and Fagherazzi21]. Meta-analysis of PCC in relation to the HRQoL shows that 58% of patients report a poor HRQoL on the EQ-5D-5L [Reference Malik, Patel, Pinto, Jaiswal, Tirupathi, Pillai and Patel20]. This systematic review however does not distinguish between subgroups, differences in symptoms, or severity.
Management and care for PCC is posing a substantial burden on healthcare systems [Reference Menges, Ballouz, Anagnostopoulos, Aschmann, Domenghino, Fehr and Puhan22, Reference McNaughton, Austin, Sivaswamy, Fang, Abdel-Qadir, Daneman, Udell, Wodchis, Mostarac, Lee and Atzema23]. The knowledge of long-term effects of COVID-19 and treatment options for PCC are still evolving [Reference Raveendran, Jayadevan and Sashidharan24]. Due to the nature of PCC, treatment often requires a multi-disciplinary approach that includes a thorough evaluation, treatment of symptoms, treatment of underlying problems, physical therapy, occupational therapy, and psychological support [Reference Raveendran, Jayadevan and Sashidharan24, Reference Greenhalgh, Knight, A’Court, Buxton and Husain25]. Identifying subtypes of PCC might help tailor treatment approaches by allowing treatment plans to be specified to the symptoms experienced in each subtype, and including only relevant healthcare providers for those specific symptoms in the treatment plan. It can help healthcare systems, healthcare providers, and patients to be better prepared and informed in managing PCC. The existing literature on PCC symptom clusters or subtypes mainly focuses on symptom patterns, frequently reported symptoms, demographic and medical background characteristics, and potential risk factors [Reference Deer, Rock, Vasilevsky, Carmody, Rando, Anzalone, Basson, Bennett, Bergquist, Boudreau, Bramante, Byrd, Callahan, Chan, Chu, Chute, Coleman, Davis, Gagnier, Greene, Hillegass, Kavuluru, Kimble, Koraishy, Köhler, Liang, Liu, Liu, Madhira, Madlock-Brown, Matentzoglu, Mazzotti, McMurry, McNair, Moffitt, Teshamae, Monteith, Parker, Perry, Pfaff, Reese, Saltz, Schuff, Solomonides, Solway, Spratt, Stein, Sule, Topaloglu, Vavougios, Wang, Haendel and Robinson10, Reference Yelin, Margalit, Nehme, Bordas-Martínez, Pistelli, Yahav, Guessous, Durà-Miralles, Carrozzi, Shapira-Lichter, Vetter, Peleato-Catalan, Tiseo, Wirtheim, Kaiser, Gudiol, Falcone and Leibovici26–Reference Whitaker, Elliott, Chadeau-Hyam, Riley, Darzi, Cooke, Ward and Elliott29]. Furthermore, Deep Phenotyping by Human Phenotype Ontology (HPO) has been applied to PCC to map the phenotypic profile of PCC [Reference Deer, Rock, Vasilevsky, Carmody, Rando, Anzalone, Basson, Bennett, Bergquist, Boudreau, Bramante, Byrd, Callahan, Chan, Chu, Chute, Coleman, Davis, Gagnier, Greene, Hillegass, Kavuluru, Kimble, Koraishy, Köhler, Liang, Liu, Liu, Madhira, Madlock-Brown, Matentzoglu, Mazzotti, McMurry, McNair, Moffitt, Teshamae, Monteith, Parker, Perry, Pfaff, Reese, Saltz, Schuff, Solomonides, Solway, Spratt, Stein, Sule, Topaloglu, Vavougios, Wang, Haendel and Robinson10]. This study aimed to add to the existing literature by not only identifying PCC phenotypes but also investigating healthcare use and the HRQoL per phenotype identified.
Methods
Study design and population
In this cross-sectional study, questionnaire data from patients with PCC were used. This study was conducted in collaboration with C-support, a Dutch organisation that informs, advices, supports, and provides care for patients experiencing long-term symptoms after SARS-CoV-2 infection. Patients can ask for support and register at C-support if they have long-term symptoms three months or longer after SARS-CoV-2 infection. This is in line with the clinical case definition of PCC established by the WHO [4]. From February 2022 onwards, patients registered at C-support long COVID were invited via email to complete an online questionnaire. The questionnaire was only available in the Dutch language. Data for this study were collected until September 2022.
Online informed consent was obtained from all participants for the usage of data in scientific research. The Medical Ethics Review Board of the Erasmus University Medical Centre approved the study protocol (MEC-2021-0751).
Measures
Socio-demographic and medical characteristics
The questionnaire included the socio-demographic variables gender (male, female, unidentified, prefer not to say), age (in years), educational level (highest attained level of formal education), and ethnicity (Dutch, Turkish, Moroccan, Antillean, Surinamese, Indonesian, prefer not to say). Seven age categories were formed: 18–24 years, 25–34 years, 35–44 years, 45–54 years, 55–64 years, 65–74 years, and 75 years and older. Education level was categorised into three categories according to the International Standard Classification of Education (ISCED-97): low (ISCED 0, 1, and 2), middle (ISCED 3 and 4), and high (ISCED 5 and 6) [30]. Ethnicity was categorised as Dutch and other.
Additionally, the questionnaire included items on medical characteristics: self-reported pre-existing chronic conditions (e.g., asthma, chronic obstructive pulmonary disease (COPD), inflammatory bowel disease (IBD), consequences of a stroke, depression/anxiety, severe heart disease, arthrosis, rheumatism, severe back complaints, hypertension, cancer, diabetes, and thyroid abnormalities), height in centimetres, weight in kilogram, month of first SARS-CoV-2 infection, hospital admission for COVID-19, and vaccination status. Time since SARS-CoV-2 infection in months was calculated based on month of the first SARS-CoV-2 infection and date of completing the questionnaire. Body mass index (BMI) was derived from height and weight and consisted of the following categories: underweight (<18.5 kg/m2), normal weight (18.5–25 kg/m2), overweight (25–30 kg/m2), and obese (>30 kg/m2) [31]. Month of the first SARS-CoV-2 infection was used as a proxy for the dominant SARS-CoV-2 strain based on the registry of the Dutch National Institute for Public Health and the Environment [32].
Primary outcome measure
The questionnaire included a list of 34 health symptoms (Supplementary Material 1). The list was based on the literature (Supplementary Material 1) and checked by and supplemented with input from patients and healthcare professionals. For every health symptom, the participants were asked to indicate whether or not they were currently experiencing it or had previously experienced the specific symptom since their infection with COVID-19 by checking a box when the symptom was experienced.
Secondary outcome measures
Secondary outcome measures were the HRQoL and healthcare use. The HRQoL was measured with the EQ-5D-5L, a generic instrument consisting of five items: mobility, self-care, usual activities, pain/discomfort, and anxiety/depression [Reference Herdman, Gudex, Lloyd, Janssen, Kind, Parkin, Parkin, Bonsel and Badia33]. For each item, there are five response categories: no problems, slight problems, moderate problems, severe problems, and extreme problems/unable to. Participants were asked to complete the EQ-5D-5L items for their health today (e.g., health status on the day of filling out the questionnaire). The EQ-5D utility score was calculated using a Dutch value set [Reference Versteegh, Vermeulen, Evers, de Wit, Prenger and Stolk34]. This score represents how good or bad someone’s health state is according to health state preferences of the general population. The utility score is anchored from 0 (a state as bad as being dead) to 1 (full health) [Reference Foundation35]. The EQ-5D-5L also includes a standardised visual analogue scale: EQ-VAS. The EQ-VAS ranges from 0 (‘worst imaginable health’) to 100 (‘best imaginable health’) [Reference Herdman, Gudex, Lloyd, Janssen, Kind, Parkin, Parkin, Bonsel and Badia33]. Participants were asked to score their health today on the EQ-VAS.
Additionally, the questionnaire included a list of 26 healthcare providers (supplementary material 2). The list was based on a list of healthcare providers previously used in research on Q-fever and checked and supplemented with input from C-support, patients, and healthcare professionals. Participants were asked to check a box for each healthcare provider they had a consultation with since the onset of COVID-19. Furthermore, participants were asked to indicate the number of consultations they had with each healthcare provider.
Statistical analysis
Descriptive statistics were performed for socio-demographic and medical characteristics of the population. Frequencies and percentages were obtained for gender, age categories, education level, nationality, other chronic conditions, BMI, hospital admission, and vaccination status. Mean score, standard deviation, median, and interquartile range were obtained for age (continuous) and time since SARS-CoV-2 infection.
A principal component analysis (PCA) of the 34 reported PCC symptoms was performed to identify symptom groups. The Oblimin rotation method was used in the PCA. The number of principal components was fixed at 8 based on the results of a Monte Carlo stimulation that was conducted to determine the statistically correct number of principal components. Loading equal to or higher than 0.25 was considered significant. The principal components were saved as weighted average of the symptoms that loaded on each principal component.
The principal components were included in a K-means clustering algorithm. The clustering algorithm was used to find the optimal grouping of participants based on the principal components. By K-means clustering, each participant is assigned to the cluster or phenotype that has the nearest mean. K was set to 3 based on the iteration history and Bonferroni testing.
Descriptive statistics were performed for each phenotype. Frequencies and percentages were obtained for gender, age categories, education level, ethnicity, other chronic conditions, BMI, hospital admission, vaccination status, symptoms, and healthcare use. Mean score, standard deviation, median, and interquartile range were obtained for age (continuous), time since SARS-CoV-2 infection, number of symptoms, and number of healthcare providers. The median was reported because of non-normality of the data. The Kruskal–Wallis test was used to compare age (continuous), time since SARS-CoV-2 infection, EQ-VAS score, EQ-5D utility score, healthcare use, and number of symptoms among the three phenotypes. Chi-square tests were used to compare gender, age (categorical), education level, nationality, other chronic conditions, BMI, hospital admission, vaccination, and EQ-5D-5L dimension scores among the three phenotypes.
Data analysis was performed by IBM SPSS statistics version 28.0.1.0. A p-value of 0.05 was considered significant.
Results
Characteristics of participants
In total, 8630 of the 14,791 (58.3%) invited patients completed the questionnaire. Table 1 describes the characteristics of participants. The majority of participants was female (76.8%), and the median age was 48 (IQR = 16) years. Most participants received a high level of education (53.6%), followed by a middle level of education (34.5%) and a low level of education (11.7%). Over half of the participants did not have other chronic conditions (53.3%). Of all participants, 8.6% was admitted to the hospital; of those, the majority was not admitted to the intensive care unit (ICU). The vast majority of the participants was vaccinated against COVID-19 (92.2%), mostly after their initial SARS-CoV-2 infection. Most participants (60.1%) were first infected with SARS-CoV-2 before 15 February 2021, when no primary dominant strain was registered; therefore, the dominant strain is ‘other’ (Supplementary Table 1). Fewer participants were infected when Alpha (19.9%), Beta (11.0%), or Omicron (8.7%) was dominant.
Socio-demographic and medical characteristics of participants (N = 8630)
Principal component analysis
The median number of symptoms reported by the total population is 18 (IQR = 9), with the top 3 most reported symptoms being fatigue, concentration problems, and decreased physical condition. Using principal component analysis, 8 patterns of the 34 patient-reported symptoms were identified (for loading data see Supplementary Table 2).
Cluster analysis
Three phenotypes were identified. All participants were grouped into ‘phenotype 1’ (n = 1116), ‘phenotype 2’ (n = 4437), or ‘phenotype 3’ (n = 3077). Cluster allocation was based on the loading of each principal component. Phenotype 1 has participants with a lower-than-average number of symptoms allocated (Figure 1). Phenotype 2 has participants with an average number of symptoms allocated. Phenotype 3 has participants with a higher-than-average number of symptoms allocated.
Principal component loading per phenotype.
The number of symptoms reported by participants significantly differed between phenotypes 1, 2, and 3 (p = 0.000) (Table 2). The highest median number of symptoms, 24 (IQR = 4), was reported in phenotype 3, while the lowest median number of symptoms, 9 (IQR = 5), was reported in phenotype 1. In phenotype 2, a median number of 16 symptoms was reported. The most frequently reported symptoms differ per phenotype (Supplementary Figure 1). The percentage of participants who reported each symptom is the highest for phenotype 3 on all reported symptoms.
Number of reported symptoms by phenotype
Compared to the other phenotypes, a significantly higher percentage of females was allocated to phenotype 3 (85.3%), and a significantly lower percentage of females was allocated to phenotype 1 (65.0%) (p < 0.001) (Table 1). Phenotype 1 has a significantly higher percentage of participants in age categories 65–74 (9.1%) and 75+ (2.2%) (p < 0.001). In phenotype 3, the percentage of participants who do not have other chronic conditions is significantly lower (49.1%) than that in phenotypes 1 and 2 (58.7 and 54.9%, respectively) (p < 0.001). The median duration of PCC was 14 months in phenotypes 1 and 2, and 15 months in phenotype 3; the difference between the phenotypes was significant (p < 0.001). The results showed no significant differences in the hospital admission rate between the phenotypes (9.5%, 8.2%, and 8.8%, respectively). In all phenotypes, the majority of participants was vaccinated against COVID-19; however, in phenotype 3, the percentage of participants who was vaccinated was significantly lower than that in the other phenotypes (p < 0.001). For each phenotype, most participants were infected before a single dominant strain was registered; therefore, ‘other’ is the dominant strain (62.4%, 57.2%, and 63.5%, respectively) (Supplementary Table 1). For each phenotype, this is followed by Alpha (16.4%, 21.8%, 18.4%), Beta (12.6%, 11.0%, 11.2%), and Omicron (8.7%, 10.0%, 6.9%), with the phenotypes differing significantly from each other (p < 0.001).
Differences in HRQoL between phenotypes
Figure 2 shows the distribution of the level of problems experienced on the EQ-5D-5L dimensions by phenotype. For each of the five dimensions, participants in phenotype 3 reported significantly more problems than participants in phenotypes 1 and 2 (Supplementary Table 3). Most problems were experienced for usual activities. The largest differences between phenotypes were seen for self-care between phenotype 1 and phenotype 3, with 28.0% and 50.1% of the participants experiencing no problems, respectively. The median EQ-5D utility (Figure 3) and EQ-VAS score (Figure 4) by phenotype differed significantly (Supplementary Table 4), with the lowest scores for participants classified into phenotype 3. The median EQ-5D utility score was 0.74 for phenotype 1, 0.66 for phenotype 2, and 0.50 for phenotype 3. The median EQ-VAS score was 60 for phenotype 1, 51 for phenotype 2, and 40 for phenotype 3.
EQ-5D-5L dimensions by phenotype.
Descriptive statistics of the EQ-5D utility score by phenotype.
X denotes mean, the line in the box denotes median, the box is the interquartile range, and the whiskers are the minimum and maximum points with outliers removed and depicted as dots.
Descriptive statistics of the EQ-VAS score by phenotype.
X denotes mean, the line in the box denotes median, the box is the interquartile range, and the whiskers are the minimum and maximum points with outliers removed and depicted as dots.
Differences in healthcare use between phenotypes
The median number of different healthcare providers consulted by participants was the highest in phenotype 3, 7 (IQR = 4), and the lowest in phenotype 1, 4 (IQR = 3) (Supplementary Table 5). The percentage of participants who mentioned consulting any healthcare provider is the highest in phenotype 3 and the lowest in phenotype 1 for each of the listed healthcare providers (Figure 5). The average number of consultations per healthcare provider differs among the phenotypes (Supplementary Figure 2).
Percentage of participants who consulted each healthcare provider by phenotype.
Discussion
In this cross-sectional study of individuals with PCC, eight symptom patterns and three phenotypes were identified. These three phenotypes were driven by number of symptoms, rather than symptom pattern. As a result, each symptom pattern was present in each of the phenotypes identified. The main difference between the three phenotypes was the number of reported symptoms. In the phenotype with the highest number of symptoms, HRQoL was the poorest and healthcare use was the highest.
This finding is consistent with the results of a study by Bowyer et al. (2023) in which data from 9 longitudinal studies were used in a latent class analysis. Based on an extensive set of symptoms, they identified two patterns, one with a low prevalence of symptoms and the other with a higher prevalence of symptoms [Reference Bowyer, Huggins, Toms, Shaw, Hou, Thompson, Kwong, Williams, Kibble, Ploubidis, Timpson, Sterne, Chaturvedi, Steves, Tilling and Silverwood27]. Other studies that aimed to identify symptom clusters of patterns yielded different results. Deep et al. (2021) found that clinical manifestation of PCC varies and has a wide range of symptoms. Yelin et al. (2022) found six symptom patterns based on the type of symptoms [Reference Yelin, Margalit, Nehme, Bordas-Martínez, Pistelli, Yahav, Guessous, Durà-Miralles, Carrozzi, Shapira-Lichter, Vetter, Peleato-Catalan, Tiseo, Wirtheim, Kaiser, Gudiol, Falcone and Leibovici26]. These symptom patterns are similar to the symptom patterns that we identified in our study through PCA. The current study used a principal component analysis followed by a K-means cluster analysis, yielding different final phenotypes. Furthermore, a Bayesian meta-regression of 54 studies and 2 medial record databases also found patterns based on the type of symptoms [3]. The Global Burden of Disease Long COVID Collaborators (2022) pre-defined clusters (persistent fatigue with bodily pain or mood swings, cognitive problems, or ongoing respiratory problems) and calculated the proportion of patients fitting in at least one of those clusters [Reference Strain, Sherwood, Banerjee, Van der Togt, Hishmeh and Rossman36]. Another reason for the discrepancy in results may be the time between initial COVID-19 infection and data collection. The studies by Yelin et al. (2022) and the Global Burden of Disease Long COVID Collaborators (2022) reported a shorter time duration between infection and data collection than our study, with a median of 14 months. Furthermore, both studies used a shorter list of symptoms than the current study.
Additionally, this study compared socio-demographic and medical background characteristics, HRQoL, and healthcare use across the three phenotypes. The duration of PCC was slightly longer in the phenotype with a higher number of symptoms. This in line with the findings of the study by Bowyer et al. (2023), where the cluster with a higher burden appeared to be more common in individuals who had a COVID-19 infection more than 12 weeks before the study. There was no difference in the hospital admission rates between the phenotypes. The phenotype with the highest median number of symptoms had a higher percentage of female participants than the other two phenotypes and the study population, whereas in the phenotype with the lowest median number of symptoms, the percentage of males was higher than that in the study population. Bowyer et al. (2023) reported that the symptom pattern with a higher symptom burden is more common in females; this is coherent with our findings.
Results showed that participants in the phenotype with the highest number of symptoms have a lower vaccination rate than the other phenotypes. A study by Strain et al. (2022) showed a reduction of symptoms in patients with PCC after vaccination. The majority of the participants was vaccinated after their initial SARS-CoV-2 infection, while others were vaccinated before their initial SARS-CoV-2 infection. The vaccination rate in all three phenotypes is high compared to the vaccination rate in the Netherlands as a whole (82.1%) [37]. Even with a high vaccination rate in each of the phenotypes, it could be possible that the lower vaccination rate in this phenotype contributed to the high number of symptoms.
For the majority of participants in each of the phenotypes, no single dominant strain can be marked. When a dominant strain was found, Alpha was most found for each of the phenotypes. Results of a meta-analysis of PCC caused by different strains of SARS-CoV-2 including 51 studies show that the wild-type and different strains could all cause PCC [Reference Du, Ma, Deng, Liu and Liu38]. In addition, generally, no differences in presentation of PCC were found between the different strains [Reference Du, Ma, Deng, Liu and Liu38].
As expected, participants in the phenotype with the highest median number of symptoms experienced the worst HRQoL. The prevalence of problems on the EQ-5D-5L dimensions in the study population and the three phenotypes was higher than the prevalence of problems described in a meta-analysis of 12 studies on the HRQoL of PCC patients [Reference Malik, Patel, Pinto, Jaiswal, Tirupathi, Pillai and Patel20], indicating that the HRQoL is lower in the current study.
With the data from the current study, it is not known whether symptoms are solely caused by PCC or whether symptoms are, at least partly, caused by the indirect impact of the COVID pandemic, that is governmental measures against the spread of COVID-19, such as lockdown and quarantine measures, on health and well-being of the population. Wang et al. (2021) used a chain mediation model to show that the perceived impact of the COVID pandemic is a sequential mediator between symptoms and mental health outcome [Reference Wang, Chudzicka-Czupała, Tee, Núñez, Tripp, Fardin, Habib, Tran, Adamus, Anlacan, Aparicio García, Grabowski, Hussain, Hoang, Hetnał, Le, Ma, Pham, Reyes, Shirazi, Tan, Tee, Xu, Xu, Vu, Zhou, Chan, Kuruchittham, McIntyre, Ho, Ho and Sears39]. This could also be the case for the current study with HRQoL as the outcome.
Additionally, the mean score on the EQ-VAS in the study population and all three phenotypes was lower than the pooled EQ-VAS score reported in the meta-analysis [Reference Malik, Patel, Pinto, Jaiswal, Tirupathi, Pillai and Patel20]. These differences in findings may be due to a difference in the definition of PCC that was used. In this study, symptoms needed to be occurring for 3 months or longer after initial COVID-19 infection, whereas the study by Malik et al. did not include a restriction on the duration of symptoms. Moreover, the differences in findings may be due to differences in the study population. The studies included in the meta-analysis reported on participants who have been hospitalised with a COVID-19 infection, whereas only a small proportion of the participants in our study population has been hospitalised with COVID-19 infection, but they were registered in a long COVID-19 registry. The patients registered in the C-support long COVID-19 registry are generally patients with a healthcare demand. Compared to the results of a cross-sectional study among non-hospitalised PCC patients, the current study found similar HRQoL outcomes [Reference Meys, Delbressine, Goërtz, Vaes, Machado, Van Herck, Burtin, Posthuma, Spaetgens, Franssen, Spies, Vijlbrief, Van’t Hul, Janssen, Spruit and Houben-Wilke40]. The phenotype with fewer complaints scored slightly higher on the EQ-VAS in our study than that in Meys et al. (2020), whereas the phenotypes with an average or a higher number of complaints scored slightly lower on the EQ-VAS. EQ-VAS scores for each of the phenotypes, as well as EQ-5D utility scores, were lower than the scores for the general population with COVID-19 [Reference Tran, Nguyen, Le, Latkin, Pham, Vu, Le, Nguyen, Pham, Ta, Nguyen, Ho and Ho41]. Furthermore, when comparing EQ-VAS and EQ-5D utility scores of persons with post-COVID conditions to those of persons living with other medical conditions, it can be seen that each phenotype scores were lower on the EQ-VAS or EQ-5D utility in the study population than in patients suffering from diabetes [Reference Nguyen, Moir, Nguyen, Vu, Luong, Nguyen, Nguyen, Tran, Tran, Latkin, Zhang, Ho and Vu42], human immunodeficiency virus (HIV) [Reference Tran, Dang, Truong, Ha, Nguyen, Do, Nguyen, Latkin, Ho and Ho43], respiratory disease [Reference Ngo, Phan, Vu, Pham, Nguyen, Vu, Tran, Nguyen, Tran, Latkin, Ho and Ho44], dengue fever [Reference Tran, Thu Vu, Hoang Nguyen, Tuan Le Nguyen, Thanh Tran, Thanh Nguyen, Thi Thai, Latkin, Ho and Ho45], and skin disease [Reference Nguyen, Nguyen, Vu, Nguyen, Le, Tran, Latkin, Ho and Ho46].
This study found that participants with PCC consulted between 0 and 21 healthcare providers, with a median of 6 healthcare providers. These results were in line with other studies advising a multi-disciplinary approach for PCC [Reference Raveendran, Jayadevan and Sashidharan24, Reference Greenhalgh, Knight, A’Court, Buxton and Husain25]. A relatively high healthcare use was found in this study, which is supported by a study by Hedberg et al. (2022) which showed an increase in healthcare use by PCC patients in the 12 months after acute infection compared to that before COVID-19 infection [Reference Hedberg, Granath, Bruchfeld, Askling, Sjöholm, Fored, Färnert and Naucler47]. Participants in the phenotype with fewer symptoms consulted fewer healthcare providers than those with more symptoms.
For future studies, it is recommended that a wide variety of symptoms experienced by PCC patients should be taken into account when investigating the long-term effects of COVID-19 and different phenotypic presentations of PCC should be further investigated. This could include taking into account differences in time since SARS-CoV-2 infection or gender. In addition, a clear definition of PCC is needed for coherent research into the causes, impact, and treatment of PCC. This study yet again highlights the complexity of PCC and the variety and amount of symptoms experienced by those suffering from PCC. Even though this study cannot provide a clear definition of what PCC entails, we hope the results of this study add to understanding PCC and its different phenotypes.
Strengths and limitations
The most important strengths of this study include the large sample size of patients with PCC and the wide range of symptoms. Furthermore, using the combination of PCA and K-means clustering is a strength because both symptom patterns and PCC phenotypes were identified. Additionally, the link between symptom phenotypes and the outcome measures HRQoL and healthcare use adds to the existing literature on PCC clusters and subtypes. Insights into HRQoL and healthcare use of patients in different PCC phenotypes can help get a better understanding of PCC. The knowledge of PCC and different phenotypes or subgroups can be a basis for further research into healthcare and support needs of PCC patients. This could possibly help tailor care and support for patients with PCC, more to the specific needs of the phenotypes.
This study has several limitations. First, the study population might not be representative of the entire population of patients with PCC. The study population seems to underrepresent mostly or fully recovered patients, and patients with low numbers of PCC symptoms. PPC patients need to register at C-support themselves; therefore, mostly patients who are actively looking for help are included in this study. These are likely patients with most symptoms and also patients who are capable of finding C-support. Furthermore, although this study identified phenotypes based on the presence of symptoms, the severity of symptoms was not taken into account. A second limitation is the non-response bias and selection bias, which were possibly introduced in this study, especially in those with a longer time since COVID-19 infection. PCC patients who are mostly or fully recovered might be less inclined to fill out the questionnaire or they were not part of the C-support PCC registry. Furthermore, patients might be less likely to fill out the extensive questionnaire when they experience a lot of cognitive symptoms. This may have resulted in a very specific groups of PCC patients with severe symptoms having been included in this study. A non-response analysis was not performed because the data needed for this were not available; therefore, characteristics of patients who were invited to participate in this study but did not is not available. Another limitation is recall bias, especially in participants with a longer time between COVID-19 infection and data collection. It could be difficult to recollect all the symptoms they experienced over the entire period that they have been living with PCC, and therefore, it could be difficult to answer questions about the entire period of PCC. Furthermore, the way in which the questions about having symptoms were posed could pose a limitation. Respondents were asked which symptoms they had had since COVID-19 infection. Therefore, it is unknown whether all reported symptoms were present simultaneously and at the moment of filling out the questionnaire. Another limitation is not taking sex-specific symptoms into account. Menstrual complaints were included in the analyses of this study that included both men and women, even though menstrual complaints are a sex-specific symptom experienced by 15.7% of the total study population. Additionally, all patients were included in analysis, regardless of the time between SARS-CoV-2 infection and data collection. Therefore, it is unclear whether phenotype deferrers based on duration of PCC.
Conclusion
In conclusion, patients with PCC experience a large variety of symptoms to a greater or lesser degree. Three specific phenotypes were identified based on symptom groups and differ in the number of symptoms patients experience. In the phenotype with participants who experience a higher number of symptoms, a lower HRQoL and higher healthcare use are found. This suggest there is a relationship between the number of symptoms and HRQoL and healthcare use. With these findings in mind, additional studies into specific healthcare needs and treatment strategies are needed to tailor treatment plans to patient’s needs.
Supplementary material
The supplementary material for this article can be found at http://doi.org/10.1017/S0950268823001139.
Data availability statement
The dataset generated and analysed during the current study is available from the corresponding author on reasonable request.
Author contribution
I.G., I.B., I.S., S.B.R., S.P., and J.H. designed the study. I.B., I.S., and S.P. collected the data. I.G., S.P., and J.H. performed the analyses and interpretation of the data. I.G. drafted the manuscript. I.G., I.B., I.S., S.B.R., S.P., and J.H. worked together to revise it critically. All authors read and approved the final manuscript.
Financial support
The funding for this study was provided by C-support, a Dutch organization, commissioned by the Ministry of Health, that informs, advises and supports patients who experience long-term complaints after the initial COVID-19 infection. C-support assisted in the design and execution of this study.
Competing interest
S.B.R. is working for C-support in the capacity of medical advisor. The participation of S.B.R. in the realization of this study and the revision of the manuscript was conducted in a responsible and ethical manner. The remaining authors have no conflicts of interest to declare.
Open access
