Long COVID Longitudinal Symptoms Burden Clusters Within A National Community-Based Cohort

Long COVID Longitudinal Symptoms Burden Clusters Within A National Community-Based Cohort | medRxiv /* */ /* */ <!-- <!-- /*! * yepnope1.5.4 * (c) WTFPL, GPLv2 */ (function(a,b,c){function d(a){return"[object Function]"==o.call(a)}function e(a){return"string"==typeof a}function f(){}function g(a){return!a||"loaded"==a||"complete"==a||"uninitialized"==a}function h(){var a=p.shift();q=1,a?a.t?m(function(){("c"==a.t?B.injectCss:B.injectJs)(a.s,0,a.a,a.x,a.e,1)},0):(a(),h()):q=0}function i(a,c,d,e,f,i,j){function k(b){if(!o&&g(l.readyState)&&(u.r=o=1,!q&&h(),l.onload=l.onreadystatechange=null,b)){"img"!=a&&m(function(){t.removeChild(l)},50);for(var d in y[c])y[c].hasOwnProperty(d)&&y[c][d].onload()}}var j=j||B.errorTimeout,l=b.createElement(a),o=0,r=0,u={t:d,s:c,e:f,a:i,x:j};1===y[c]&&(r=1,y[c]=[]),"object"==a?l.data=c:(l.src=c,l.type=a),l.width=l.height="0",l.onerror=l.onload=l.onreadystatechange=function(){k.call(this,r)},p.splice(e,0,u),"img"!=a&&(r||2===y[c]?(t.insertBefore(l,s?null:n),m(k,j)):y[c].push(l))}function j(a,b,c,d,f){return q=0,b=b||"j",e(a)?i("c"==b?v:u,a,b,this.i++,c,d,f):(p.splice(this.i++,0,a),1==p.length&&h()),this}function k(){var a=B;return a.loader={load:j,i:0},a}var l=b.documentElement,m=a.setTimeout,n=b.getElementsByTagName("script")[0],o={}.toString,p=[],q=0,r="MozAppearance"in l.style,s=r&&!!b.createRange().compareNode,t=s?l:n.parentNode,l=a.opera&&"[object Opera]"==o.call(a.opera),l=!!b.attachEvent&&!l,u=r?"object":l?"script":"img",v=l?"script":u,w=Array.isArray||function(a){return"[object Array]"==o.call(a)},x=[],y={},z={timeout:function(a,b){return b.length&&(a.timeout=b[0]),a}},A,B;B=function(a){function b(a){var a=a.split("!"),b=x.length,c=a.pop(),d=a.length,c={url:c,origUrl:c,prefixes:a},e,f,g;for(f=0;f<d;f++)g=a[f].split("="),(e=z[g.shift()])&&(c=e(c,g));for(f=0;f<b;f++)c=x[f](c);return c}function g(a,e,f,g,h){var i=b(a),j=i.autoCallback;i.url.split(".").pop().split("?").shift(),i.bypass||(e&&(e=d(e)?e:e[a]||e[g]||e[a.split("/").pop().split("?")[0]]),i.instead?i.instead(a,e,f,g,h):(y[i.url]?i.noexec=!0:y[i.url]=1,f.load(i.url,i.forceCSS||!i.forceJS&&"css"==i.url.split(".").pop().split("?").shift()?"c":c,i.noexec,i.attrs,i.timeout),(d(e)||d(j))&&f.load(function(){k(),e&&e(i.origUrl,h,g),j&&j(i.origUrl,h,g),y[i.url]=2})))}function h(a,b){function c(a,c){if(a){if(e(a))c||(j=function(){var a=[].slice.call(arguments);k.apply(this,a),l()}),g(a,j,b,0,h);else if(Object(a)===a)for(n in m=function(){var b=0,c;for(c in a)a.hasOwnProperty(c)&&b++;return b}(),a)a.hasOwnProperty(n)&&(!c&&!--m&&(d(j)?j=function(){var a=[].slice.call(arguments);k.apply(this,a),l()}:j[n]=function(a){return function(){var b=[].slice.call(arguments);a&&a.apply(this,b),l()}}(k[n])),g(a[n],j,b,n,h))}else!c&&l()}var h=!!a.test,i=a.load||a.both,j=a.callback||f,k=j,l=a.complete||f,m,n;c(h?a.yep:a.nope,!!i),i&&c(i)}var i,j,l=this.yepnope.loader;if(e(a))g(a,0,l,0);else if(w(a))for(i=0;i (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0];var j=d.createElement(s);var dl=l!='dataLayer'?'&l='+l:'';j.src='//www.googletagmanager.com/gtm.js?id='+i+dl;j.type='text/javascript';j.async=true;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-P4HH5NV'); Skip to main content Home About Submit ALERTS / RSS Search for this keyword Advanced Search Long COVID Longitudinal Symptoms Burden Clusters Within A National Community-Based Cohort View ORCID Profile Yanhan Shen , Zach Shahn , View ORCID Profile McKaylee M. Robertson , Kelly Gebo , Denis Nash , the CHASING COVID Cohort Study Team doi: https://doi.org/10.1101/2025.11.01.25339283 Yanhan Shen 1 Institute for Implementation Science in Population Health (ISPH), City University of New York (CUNY) ; New York, NY, USA 2 Department of Epidemiology and Biostatistics, Graduate School of Public Health and Health Policy, City University of New York (CUNY) ; New York, NY, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Yanhan Shen For correspondence: jenny.shen{at}sph.cuny.edu Zach Shahn 1 Institute for Implementation Science in Population Health (ISPH), City University of New York (CUNY) ; New York, NY, USA 2 Department of Epidemiology and Biostatistics, Graduate School of Public Health and Health Policy, City University of New York (CUNY) ; New York, NY, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site McKaylee M. Robertson 1 Institute for Implementation Science in Population Health (ISPH), City University of New York (CUNY) ; New York, NY, USA 2 Department of Epidemiology and Biostatistics, Graduate School of Public Health and Health Policy, City University of New York (CUNY) ; New York, NY, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for McKaylee M. Robertson Kelly Gebo 3 Department of Medicine, Johns Hopkins University ; Baltimore, MD, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site Denis Nash 1 Institute for Implementation Science in Population Health (ISPH), City University of New York (CUNY) ; New York, NY, USA 2 Department of Epidemiology and Biostatistics, Graduate School of Public Health and Health Policy, City University of New York (CUNY) ; New York, NY, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site 1 Institute for Implementation Science in Population Health (ISPH), City University of New York (CUNY) ; New York, NY, USA Abstract Full Text Info/History Metrics Supplementary material Data/Code Preview PDF Abstract Background Long COVID is clinically heterogeneous, with evolving symptom trajectories that complicate classification. Prior clustering studies often rely on Electronic Health Records data, risking underreporting. We used longitudinal, community-based data to characterize symptom clusters and risk factors. Methods We analyzed CHASING COVID Cohort participants with confirmed SARS-CoV-2 infection between December 2020 and December 2022, ≥12 months of follow-up, and long COVID (≥1 new symptom and concurrent activity limitation 3–12 months post-infection, both absent pre-infection). The infection in this window was the index infection; those with pre-index long COVID were excluded. Symptoms were self-reported pre-infection and at ∼3, 6, 9, and 12 months. Missing data were handled via multiple imputation by chained equations (30 datasets). Longitudinal K-means clustering was performed within each imputed dataset, with hierarchical aggregation to derive final assignments. Multivariable logistic regression (adjusting for age, sex) assessed associations of demographic, clinical, and social factors with cluster membership. Within the highest-burden cluster, hierarchical clustering identified symptom phenotypes. Results Of 1,787 infected participants, 511 met criteria (22% ≥50 years; 55% female; 60% White non-Hispanic; 54% mental health disorder; 7.2% immunodeficiency; 21% ≥2 comorbidities; 30% prior infection; 24% never vaccinated pre-index). Three clusters emerged: highest, moderate, and lowest burden. The highest-burden cluster (median 6 symptoms at 6 and 9 months) was characterized by fatigue (57%–74%), concentration difficulty, post-exertional malaise, myalgia, sleep disturbance, gastrointestinal symptoms, headache, irritability, and mobility limitations (each ∼46%–53%). The moderate cluster peaked at 3 symptoms (fatigue 42%); the lowest remained ∼1 symptom (fatigue 24% at 12 months). Older age (aOR 2.68, 95% CI 1.59–4.53), female sex (2.36, 1.48–3.76), mental health disorder (2.65, 1.65–4.25), and immunodeficiency (4.23, 1.71–10.51) were associated with highest vs lowest burden. Within the highest-burden cluster, three phenotypes emerged: neurological/multisystemic, psychiatric/neurological, and physical/respiratory. Conclusions Distinct long-COVID clusters and phenotypes underscore heterogeneity and support tailored management and risk stratification. INTRODUCTION Long COVID is a complex condition characterized by substantial breadth, with diverse symptoms—including fatigue, respiratory, neurological, cardiovascular, and gastrointestinal manifestations—spanning multiple organ systems 1 , 2 . Its substantial depth arises from complex underlying mechanisms such as viral persistence, immune dysregulation, endothelial dysfunction, autonomic disturbances, and mitochondrial impairment, presenting significant clinical and research challenges 1 , 3 , 4 . Clustering analysis offers a valuable tool to uncover symptom heterogeneity and enhance our understanding of disease pathophysiology 5 – 13 . Clustering algorithms analyze data distribution and patterns to group patients with similar characteristics 5 , 14 , 15 . Longitudinal studies utilizing electronic health records (EHR) data collected at multiple time points have identified distinct phenotypes of long COVID, characterized by varying symptom clusters and trajectories across different stages of recovery. For instance, a two-year prospective study observed that long COVID symptoms often peak between 6 to 12 months post-infection, significantly impacting the quality of life during this period 16 . Another longitudinal study employing latent class analysis on patient-reported outcomes across multiple time points identified distinct symptom profiles, highlighting the heterogeneous nature of symptom presentation over time 8 . Additionally, a study clustering participants based on symptoms documented at both 9 and 12 months post-infection identified four clusters: no or minor symptoms; multi-symptoms; joint pain; and neurocognitive-related symptoms, which were similar at both 9 and 12 months post-infection 17 . Collectively, these findings emphasized that symptom clustering in long COVID is dynamic and changes over time and underscored the importance of continuous monitoring and personalized management strategies for individuals experiencing long COVID. There are limitations to using EHR data to cluster long COVID patients’ symptoms due to under-reporting or under-ascertainment of milder or transient symptoms and the inability to adequately capture subjective experiences such as disabilities and limitations in daily life 6 , 8 – 12 , 18 . Furthermore, since EHR data rely on individuals actively engaging in healthcare, they often fail to capture longitudinal long COVID symptoms 19 . In contrast to EHR-based studies, which are limited to individuals actively engaging with the healthcare system and often miss milder or subjective symptoms, community-based cohort studies include individuals both within and outside of the healthcare system and can capture longitudinal, self-reported symptoms, providing a more comprehensive picture of long COVID’s progression and impact over time. We aimed to leverage the community-based data to identify symptom clusters of participants with long COVID based on systematically ascertained self-reported symptoms. We also aimed to identify demographic, socioeconomic, and comorbidity characteristics of those experiencing the highest symptom burden. METHODS Participants The Communities, Households, and SARS-CoV-2 Epidemiology (CHASING) COVID Cohort study was a national prospective cohort, which recruited participants between March 28, 2020 and August 21, 2020 20 . The community-based sample included adults residing in 50 U.S. states, the District of Columbia, Puerto Rico, and Guam. Study participants were recruited through social media ads (Facebook, Instagram, Scruff), Qualtrics Panel, and referrals from personal networks. Advertisements were written in English and Spanish, targeting people in the U.S. and U.S. territories aged 18 and older. As part of the study design, participants completed approximately quarterly surveys on COVID-19 exposures, symptoms, and testing, and also completed repeat serologic testing for SARS-CoV-2 infection approximately annually. Details of the study design, recruitment procedures, and serology-based incidence in this cohort are described elsewhere 21 – 23 . The study protocol was approved by the Institutional Review Board of the City University Of New York (CUNY) (New York, NY, USA). Participant consent was obtained online at baseline and periodically during follow-up assessments. Pre- and Post-infection Symptoms and Activity Limitations In each follow-up assessment since November 2020 approximately every three months, participants were asked whether they were currently experiencing any of the following symptoms: shortness of breath, difficulty walking more than 15 minutes, difficulty running/exercising, fatigue, post-exertional malaise, headache, trouble concentrating/brain fog, dizziness, irritability, erratic heartbeat, gastrointestinal issues, low-grade fever, myalgia, loss or alteration of taste/smell, and difficulty sleeping. We categorized the longitudinal symptom assessments into pre-infection, 3-, 6-, 9-, and 12-month post-infection periods based on the number of days between the assessment and the infection dates (determination of SARS-CoV-2 infection status and dates is outlined in Appendix 1 ). Specifically, assessments conducted 45-365 days before infection were classified as pre-infection data, while those within 45-135, 136-225, 226-315, and 316-405 days post-infection were categorized as 3-, 6-, 9- and 12-month post-infection, respectively. If participants completed more than one symptom assessment within a given interval, only the response closest to the target time point (i.e., 90, 180, 270, or 360 days post-infection) was retained to avoid potential overcounting of symptoms. To address missing symptoms data, we applied multiple imputation by chained equation (MICE) with the fully conditional specification (FCS), generating 30 imputed datasets 24 , 25 . Given the non-monotone missing data pattern and the assumption of missing at random, we accounted for occasional missed follow-ups and timing constraints that prevented appropriate interval assignment. We assumed that missing symptom data resulted either from participants occasionally missing follow-up assessments or from the timing of the follow-up assessment fielding, which sometimes prevented us from assigning the appropriate follow-up intervals. Variables selected for the imputation model included those correlated with the missing symptom data and predictive of missingness in follow-ups 26 . A logistic regression model was used for each outcome, given the outcomes at other time points and the covariates 27 . Specifically, we included the following variables as the dependent variables in the logistic regression model: symptom data from prior time points, age, gender, race/ethnicity, education, number of comorbidities, employment category, geographic region of residence, and material hardship. Details on demographics, socioeconomic factors, comorbidities, prior infection history, and COVID-19 vaccination status are provided in Appendix 2 . Participants were classified as having activity limitations if they reported difficulty in daily activities or household responsibilities 28 . At each follow-up, participants who reported “Some difficulty” or “A lot of difficulty” to a question assessing challenges in engaging in daily activities or household responsibilities due to physical, mental, or emotional problems—one of the two-item disability measurement from the BRFSS survey—were considered to have activity limitations 29 . Long COVID Case Definition We defined long COVID cases as participants who experienced at least one symptom once between 3 and 12 months post-infection that was not present in the year preceding the index infection date, in combination with concurrent activity limitations which was present at the same time of symptom and absent prior to the index infection. The long COVID case definition was informed by guidance from the World Health organization (WHO) 30 , the Centers for Disease Control and Prevention (CDC) 31 , and the National Academies of Science, Engineering, and Medicine (NASEM) 32 , all of which emphasize new or ongoing symptoms and functional impairment occurring beyond the acute phase of infection. Long COVID case status was determined solely from observed survey data and imputed values were not used for case ascertainment. Cohort Definition and Eligibility criteria The cohort comprised participants who met the following criteria: 1) had evidence of SARS-CoV-2 infection, based on a positive PCR, antigen or serology test between December 01, 2020, and December 31, 2022. This infection date was defined as the index infection date in this study, ensuring for a pre-infection symptom assessment and a minimum of 12 months of post-infection follow-up; 2) had at least one follow-up assessment in the year before and after the index infection; 3) had an exact infection date or the imputation interval of no more than 90 days; 4) were identified as long COVID cases according to the primary long COVID definition; 5) were not previously identified as having long COVID prior to the index infection. Centroid-Based Partition Clustering Analysis of Longitudinal Symptoms We applied K-means clustering for longitudinal data analysis, utilizing a centroid-based partitioning method. This approach assigns groups by minimizing the distance between each participant’s symptom trajectory within each cluster while maximizing the distance between cluster centroids 33 . The method proceeded in two steps, as follows: We first conducted hierarchical cluster analysis to estimate the number of optimal clusters by the Ward linkage method using McClain, Dunn, and Silhouette criteria 14 . The McClain’s index, Dunn index, and Silhouette score assessed clustering quality by optimizing two key concepts: cluster compactness (how close points within a cluster are to each other) and cluster separation (how distinct clusters are from one another) 34 . We applied the K-means for longitudinal data clustering algorithm using the optimal cluster number determined in the step described above 35 . Based on the selected optimal number of clusters, we performed longitudinal K-means consensus clustering with dynamic time warping distance to identify symptom clusters by counting each symptom equally 36 , 37 . In this algorithm, each observation is reassigned to the group with the closest mean and the operation is repeated until convergence. Statistical Analysis K-means clustering was performed separately on each of the 30 imputed datasets to assign participants into clusters based on longitudinal symptoms. To consolidate potentially inconsistent cluster assignments across imputations, we implemented a hierarchical aggregation procedure (the details are provided in Appendix 3 ) 38 – 40 . This approach overcomes the limitations posed by arbitrary and potentially non-aligned cluster labels across imputations by relying on patterns of co-assignment rather than direct label matching. After finalizing the cluster assignments, we visualized symptom trajectories using radar plots and presented the proportions and corresponding 95% confidence intervals, stratified by clusters, symptom and time points (pre-infection, 3-, 6-, 9-, and 12-month post-infection). For each cluster and time point, pooled symptom proportions were calculated by averaging estimates across the 30 imputations, with the corresponding variances combined using Rubin’s rules to integrate both within- and between-imputation variability 41 , 42 . After participants were classified into clusters, we characterized the long COVID symptom burden by first computing, for each imputed dataset, the total number of symptoms at each time point (calculated as the sum of 15 binary symptoms). For each participant, the median symptom count across the 30 imputations was then calculated. For each cluster and time point, the pooled symptom burden was defined as the median of these participant-level medians, and the variability across imputations was calculated by computing the sample standard deviation of the medians divided by the square root of the number of participants, yielding approximate 95% confidence intervals as the pooled median ± 1.96 times this standard error 41 , 42 . We presented the prevalence of demographic, socioeconomic and comorbidities characteristics, prior infection status at index date, post index date reinfection status, and vaccination status, stratified by cluster. To examine factors associated with cluster assignment, we used multivariable logistic regression models to estimate adjusted odds ratios (aORs) and 95% CI for two comparisons: highest verse lowest symptom burden and moderate versus lowest symptom burden, with each model adjusted only for age and gender. To further explore symptom variations within the highest symptom burden cluster, we conducted hierarchical clustering to identify distinct symptom phenotypes. The heatmap represents the prevalence of each symptom reported for at least six months post-infection across the identified phenotypic subgroups. Color intensity indicates symptom prevalence, with darker shades representing higher prevalence. Details of sensitivity analyses are provided in Appendix 4 . We used SAS version 9.4 (SAS Institute, Cary, NC, USA) for data cleaning and R version 2024.09.0+375 (R Foundation for Statistical Computing, Vienna, Austria) for data analyses. RESULTS Long COVID Cases Characteristics A total of 511 participants out of 1,787 infected individuals who met the eligibility criteria were included in the analysis ( eFigure 1 ). Among these 511 long COVID cases, 22% were aged ≥50, 55% were female, 60% were White Non-Hispanic, 45% did not have a college degree, and 35% reported an annual household income of 35,000 or less ( Table 1 ). In terms of comorbidities, 21% had 2 or more underlying conditions. The prevalence of specific conditions included mental health disorders (54%), type 2 diabetes (8%), heart disease (5.3%), HIV (5.1%), lung disease (4.7%), cancer (4.7%), kidney disease (2.0%), and immunodeficiency (7.2%). Regarding prior COVID-19 history, 30% of long COVID cases had a SARS-CoV-2 infection prior to the index infection date, and 18% experienced a subsequent SARS-CoV-2 infection during the 3-12 months following the index infection. Vaccination status preceding the index infection was as follows: 24% were unvaccinated, 5% were vaccinated but had not completed the primary series, 22% had completed the primary series, and 49% had completed the primary series and received at least one additional dose. View this table: View inline View popup Download powerpoint Table 1: Median Number of Long COVID Symptoms and 95% Confidence Intervals by Symptom Burden Cluster and Timepoint, the CHASING COVID Cohort Study (N = 511) Symptoms Cluster Assignment As shown in Figure 1 and eTable 1 , distinct symptom profiles were observed across clusters. Among the individuals in Cluster 2, all of whom were identified as having long COVID, fatigue had the highest prevalence at 12 months post-infection (57%, 95% CI: 44-70%). Several other symptoms--difficulty running, difficulty walking, difficulty sleeping, gastrointestinal issues, headache, irritability, post-exertional malaise, myalgia, and trouble concentrating--each had a prevalence 46-53%, and collectively represented the most commonly reported symptoms within this cluster. For fatigue, the 12 months prevalence remained elevated compared to the pre-infection level of 51% (95% CI: 43-60%), though it was lower than the peak prevalence of 74% (95% CI: 63-85%) at 3-month post-infection. Download figure Open in new tab Figure 1. Radar plots representing the pooled prevalence estimates of Long COVID symptoms across three clusters at five time points, the CHASING COVID Cohort study, December 2020 – December 2023 (N=511) Abbreviations: PEM, post-exertional malaise; Anosmia/Ageusia, loss of smell or taste; DiffSleep, difficulty sleeping; LGF, low-grade fever; Irritabilt, irritability; DiffRun, difficulty running; DiffWalk, difficulty walking; SOB, shortness of breath; GI, gastrointestinal symptoms; EHB, erratic heartbeats; TC, trouble concentrating. Three side-by-side radar plots (Clusters 1–3) show the prevalence (0–100%) of 15 long-COVID symptoms—fatigue, trouble concentrating, post-exertional malaise, myalgia, difficulty sleeping, gastrointestinal symptoms, headache, irritability, dizziness, erratic heartbeats, shortness of breath, difficulty walking, difficulty running, loss/alteration of smell or taste, and low-grade fever—measured at five time points (pre-infection and 3, 6, 9, and 12 months post-infection). Cluster 2 has the highest and most persistent burden: multiple symptoms (especially trouble concentrating, fatigue, post-exertional malaise, sleep difficulty, GI symptoms, myalgia, and mobility limitations) reach roughly 40– 60% at 6–9 months and remain elevated at 12 months. Cluster 1 shows a moderate burden, with most symptoms around 20–35% and modest decline by 12 months. Cluster 3 shows the lowest burden, generally ∼10–25% across time. Pre-infection levels are lower than post-infection in all clusters. In the Cluster 1, fatigue was the most prevalent symptom at 12-months post-infection (prevalence: 42%, 95% CI: 33-52%), with prevalence slightly declined compared to earlier post-infection time points ( Figure 1 and e Table 1 ). In addition, post-exertional malaise, and trouble concentrating were prevalent close to 30% at 12-month post-infection, which remained elevated compared to their pre-infection level. In Cluster 3, although fatigue was the most prevalent at 12 months (24%, 95% CI: 16-31%), which was higher than pre-infection level and slightly declined compared to 3-, 6-, and 9-months post-infection levels. The prevalence of difficulty sleeping was slightly declined at month 12 compared to the 3-, 6-, and 9-months post-infection. As shown in Table 1 , the median number of long COVID symptoms across time points was used to characterize the Cluster 2, 1, 3 as highest, moderate, and lowest symptom burden clusters, respectively. The highest symptom burden cluster exhibited a consistently elevated symptom profile, with median symptom counts of 5–6 throughout the post-infection period. The cluster featured by moderate symptom burden had a median of 3 symptoms at 3-, 6- and 12-months post-infection, with a slight decline to 2 symptoms at 9 months. In contrast, the lowest symptom burden cluster maintained a stable median of 1 symptom across all post-infection time points. Risk Factors Associated with Cluster Assignment Adjusted odds ratios (aOR) for classification into highest vs. lowest symptom burden clusters were shown in Table 2 . Participants aged over 50 had more than a two-fold adjusted odds of being classified into the highest (vs. lowest) symptom burden cluster compared to those under 50 (aOR 2.68, 95% CI: 1.59-4.53). Female participants had significantly higher adjusted odds of being classified into the highest (vs. lowest) symptom burden cluster compared to males (aOR 2.36, 95% CI: 1.48-3.76). Additionally, individuals with two or more pre-infection comorbidities had over three times the adjusted odds of being classified into the highest (vs. lowest) symptom burden cluster compared to those with no comorbidities (aOR 5.91, 95% CI: 3.08-11.37). Participants with a history of mental health disorders had significantly elevated odds of being classified into the highest (vs. lowest) symptom burden cluster (aOR: 2.65, 95% CI: 1.65–4.25), and those with immunodeficiency had more than four-fold increased odds (aOR: 4.23, 95% CI: 1.71–10.51). View this table: View inline View popup Table 2: Characteristics and Adjusted Odds of Classification into Long COVID Symptom Burden Clusters Identified by K-Means Clustering, the CHASING COVID Cohort Study, December 2020 – December 2023 Symptom Phenotype for those Classified in the Highest Symptom Burden Cluster Figure 2 . illustrates the distribution of symptom phenotypes among participants in the highest symptom burden cluster, with symptoms categorized into three phenotypes (A, B, and C), each representing approximately one-third of the cluster (32%, 36%, and 32%, respectively). The heatmap displays the prevalence of specific symptoms that were reported at multiple time points spanning six or more months post-infection, with darker shades indicating higher prevalence. Phenotype A is predominantly characterized by multi-systemic dysfunction, with a higher prevalence of fatigue, post-exertional malaise, and brain fog/trouble concentrating. Phenotype B is defined by a combination of psychiatric and neurological symptoms, including irritability and difficulty sleeping. Phenotype C is primarily associated with physical and respiratory limitations, including fatigue, difficulty walking and running, and shortness of breath. Download figure Open in new tab Figure 2. Symptom Phenotypes Among Individuals with Highest Symptom Burden Figure Description: Symptom phenotypes among individuals classified in the high symptom burden cluster, showing the prevalence of symptoms persisting for ≥6 months. •Phenotype A: Predominantly neurological and multi-systematic dysfunction, including a higher prevalence of fatigue, post-exertional malaise, and brain fog/trouble concentrating. •Phenotype B: Defined by a mix of psychiatric and neurological symptoms, including irritability and difficulty sleeping. •Phenotype C: Characterized by physical and respiratory limitations, including fatigue, difficulty walking and running, and shortness of breath. DISCUSSION We identified distinct long COVID symptom burden clusters up to 12 months post-infection in a community-based cohort, demonstrating heterogeneity in symptom counts, profile-specific recovery trajectories, and associated risk factors. Unlike EHR/clinic-based studies, systematic, repeated community self-reporting captured individuals who did not seek care, improving case ascertainment and generalizability. These findings clarify the spectrum of long COVID presentations and can inform diagnosis, risk stratification, and targeted interventions. Using a data-driven approach, we delineated three long COVID subgroups—highest, moderate, and lowest symptom burden—defined by both total symptom counts and longitudinal prevalence profiles, capturing differences in overall intensity across time points. This structure accords with prior reports of heterogeneity 43 and, consistent with RECOVER 7 , includes a subset with persistently elevated burden. In the high-burden cluster, fatigue, concentration difficulties, post-exertional malaise, and gastrointestinal symptoms predominated, with prevalence peaking at ∼6 months. Across the 12-month follow-up we observed a descriptive decline in fatigue in all clusters, which may indicate partial improvement; however, because clusters were formed on elevated early symptoms, some of this trend could reflect regression to the mean rather than true biological recovery and should be interpreted cautiously. Overall, these findings reinforce the diversity of long-COVID symptomatology and trajectories and support tailored management, risk stratification, and targeted intervention development 44 . Consistent with prior reports, we found that older age, female sex, and the presence of underlying conditions such as mental health disorders and immunodeficiency were associated with an increased risk of being in the highest versus the lowest symptom burden cluster 45 , 46 . Recent studies have identified factors associated with varying symptom burdens among individuals with long COVID. For instance, Evans et al. indicated that female sex and older age were linked to more severe long COVID symptom profiles 46 . In a large cohort study, Xie et al. found that individuals with baseline mental health conditions, such as anxiety and depression, had a significantly higher risk of experiencing persistent long COVID symptoms 47 . Similarly, Al-Aly et al. demonstrated that individuals with compromised immune function were at increased risk for prolonged and severe symptomatology following acute SARS-CoV-2 infection, likely due to impaired viral clearance and ongoing immune activation 48 . These studies support the hypothesis that underlying psychosocial and immunological vulnerability may contribute to more severe or persistent long COVID phenotypes. These findings suggest that targeted interventions may be necessary for these subgroups to address their specific needs. To be noted, this study incorporated key social determinants of health—including education and household income—which are often underrepresented in long COVID research. Although these variables were not significantly associated with symptom burden in our primary models, their inclusion provides important context and may help explain disparities in long COVID outcomes in future studies. In the highest symptom burden cluster, the identification of distinct phenotypes—including multi-systematic symptoms, neurological and psychiatric symptoms, and physical and respiratory limitations—underscored the heterogeneity of long COVID manifestations. This variability had been corroborated by multiple studies, emphasizing the necessity for personalized care strategies. A multicenter, prospective cohort study identified four distinct phenotypes of post-COVID conditions at 3 and 6 months, including minimal symptoms, tiredness/headache/musculoskeletal symptoms, loss of smell/taste, and multisystem symptoms 49 . Blankestijn et al. delineated three distinct long COVID clusters based on patient characteristics, lung function, and symptom severity, further illustrating the complexity of the condition 50 . These findings collectively suggested that the observed heterogeneity in long COVID symptomatology might reflect different mechanisms and pathophysiologic pathways. Understanding these distinct symptom profiles could potentially inform more precise diagnostic criteria, improve risk stratification, and ultimately guide the development of more effective, phenotype-specific therapeutic strategies. This study had several important limitations. Symptoms were self-reported rather than clinician-elicited, which may overestimate prevalence; however, uniform ascertainment across participants and time points limits differential bias and preserves comparative validity. Symptom burden was quantified by counts rather than severity, constraining inferences about intensity; future work should capture severity. Follow-up requirements may have excluded those with the most debilitating illness, potentially underrepresenting the highest-burden cluster. Finally, post–time-zero SARS-CoV-2 reinfections could have contributed symptoms not attributable to the index infection, though reinfection rates were balanced across clusters, making differential effects unlikely. This study highlighted the heterogeneity of long COVID symptom trajectories and its associated risk factors. The identification of distinct symptom sub-phenotypes underscored the complexity of long COVID and the need for personalized approaches to management and care. Recognizing the diverse trajectories of recovery and symptom patterns is essential for diagnosis, improving outcomes and addressing the long-term impact of this condition. Data Availability The data underlying this study contain potentially identifiable health information and are not publicly available due to ethical and legal restrictions. De-identified data may be shared on reasonable request and subject to institutional approvals and a Data Use Agreement. Requests will be reviewed case-by-case by the study investigators and the CUNY ISPH data governance/IRB. Please contact the corresponding author. https://cunyisph.org/chasing-covid/ FUNDING Funding for this project is provided by The National Institute of Allergy and Infectious Diseases (NIAID), award number UH3AI133675 (MPIs: D Nash and C Grov), Pfizer Inc, the CUNY Institute for Implementation Science in Population Health ( cunyisph.org ) and the COVID-19 Grant Program of the CUNY Graduate School of Public Health and Health Policy. The NIH played no role in the production of this manuscript nor necessarily endorses the findings. The study design was developed by CUNY without input from Pfizer. CONFLICT of INTERESTS Author DN receives consulting fees from Gilead Sciences and AbbVie. All other authors have no conflicts of interest, financial or otherwise. PATIENT CONSENT STATEMENT Informed consent forms were completed in a web browser on participants’ computer or mobile device at baseline, each round of serological testing, and at periodic follow-up assessments. The study was approved by the Institutional Review Boards of the City University Of New York (CUNY) Graduate School of Public Health and Health Policy (New York, NY, U.S.) (protocol 2020-0256). Reference 1. ↵ Davis HE , McCorkell L , Vogel JM , Topol EJ . Long COVID: major findings, mechanisms and recommendations . Nat Rev Microbiol . 2023 ; 21 ( 3 ): 133 – 146 . OpenUrl CrossRef PubMed 2. ↵ Nalbandian A , Sehgal K , Gupta A , et al. Post-acute COVID-19 syndrome . Nat Med . 2021 ; 27 ( 4 ): 601 – 615 . OpenUrl CrossRef PubMed 3. ↵ Proal AD , VanElzakker MB . Long COVID or post-acute sequelae of COVID-19 (PASC): An overview of biological factors that may contribute to persistent symptoms . Front Microbiol . 2021 ; 12 : 698169 . OpenUrl CrossRef PubMed 4. ↵ Turner S , Khan MA , Putrino D , Woodcock A , Kell DB , Pretorius E. Long COVID: pathophysiological factors and abnormalities of coagulation . Trends Endocrinol Metab . 2023 ; 34 ( 6 ): 321 – 344 . OpenUrl CrossRef PubMed 5. ↵ Loftus TJ , Shickel B , Balch JA , et al. Phenotype clustering in health care: A narrative review for clinicians . Front Artif Intell . 2022 ; 5 : 842306 . OpenUrl PubMed 6. ↵ Deer RR , Rock MA , Vasilevsky N , et al. Characterizing long COVID: Deep phenotype of a complex condition . EBioMedicine . 2021 ; 74 ( 103722 ): 103722 . OpenUrl PubMed 7. ↵ Thaweethai T , Jolley SE , Karlson EW , et al. Development of a definition of postacute sequelae of SARS-CoV-2 infection . JAMA . 2023 ; 329 ( 22 ): 1934 – 1946 . OpenUrl CrossRef PubMed 8. ↵ Wong AW , Tran KC , Binka M , et al. Use of latent class analysis and patient reported outcome measures to identify distinct long COVID phenotypes: A longitudinal cohort study . PLoS One . 2023 ; 18 ( 6 ): e0286588 . OpenUrl CrossRef PubMed 9. Chen C , Parthasarathy S , Leung JM , et al. Distinct temporal trajectories and risk factors for Post-acute sequelae of SARS-CoV-2 infection . Front Med (Lausanne) . 2023 ; 10 . doi: 10.3389/fmed.2023.1227883 OpenUrl CrossRef PubMed 10. Frontera JA , Thorpe LE , Simon NM , et al. Post-acute sequelae of COVID-19 symptom phenotypes and therapeutic strategies: A prospective, observational study . PLoS One . 2022 ; 17 ( 9 ): e0275274 . OpenUrl CrossRef PubMed 11. Ozonoff A , Schaenman J , Jayavelu ND , et al. Phenotypes of disease severity in a cohort of hospitalized COVID-19 patients: Results from the IMPACC study . EBioMedicine . 2022 ; 83 ( 104208 ): 104208 . OpenUrl PubMed 12. ↵ Kisiel MA , Lee S , Malmquist S , et al. Clustering analysis identified three long COVID phenotypes and their association with general health status and working ability . J Clin Med . 2023 ; 12 ( 11 ): 3617 . OpenUrl PubMed 13. ↵ Global Burden of Disease Long COVID Collaborators , Wulf Hanson S , Abbafati C , et al. Estimated global proportions of individuals with persistent fatigue, cognitive, and respiratory symptom clusters following symptomatic COVID-19 in 2020 and 2021 . JAMA . 2022 ; 328 ( 16 ): 1604 – 1615 . OpenUrl CrossRef PubMed 14. ↵ Everitt BS , Landau S , Leese M , Stahl D. Cluster Analysis . 5th ed. Wiley-Blackwell ; 2011 . 15. ↵ Ivanov OA . Easy as π? 1999th ed. Springer ; 2013 . 16. ↵ Demko ZO , Yu T , Mullapudi SK , et al. Two-year longitudinal study reveals that long COVID symptoms peak and quality of life nadirs at 6-12 months postinfection . Open Forum Infect Dis . 2024 ; 11 ( 3 ): ofae027 . OpenUrl CrossRef PubMed 17. ↵ Moniz M , Ruivinho C , Goes AR , Soares P , Leite A. Long COVID is not the same for everyone: a hierarchical cluster analysis of Long COVID symptoms 9 and 12 months after SARS-CoV-2 test . BMC Infect Dis . 2024 ; 24 ( 1 ): 1001 . OpenUrl PubMed 18. ↵ Ozonoff A , Jayavelu ND , Liu S , et al. Features of acute COVID-19 associated with post-acute sequelae of SARS-CoV-2 phenotypes: results from the IMPACC study . Nat Commun . 2024 ; 15 ( 1 ). doi: 10.1038/s41467-023-44090-5 OpenUrl CrossRef PubMed 19. ↵ Young JC , Conover MM , Funk MJ . Measurement error and misclassification in electronic medical records: methods to mitigate bias . Curr Epidemiol Rep . 2018 ; 5 ( 4 ): 343 – 356 . OpenUrl PubMed 20. ↵ CHASING COVID cohort study . CUNY ISPH . Accessed September 6, 2024 . https://cunyisph.org/chasing-covid/ 21. ↵ Robertson MM , Kulkarni SG , Rane M , et al. Cohort profile: a national, community-based prospective cohort study of SARS-CoV-2 pandemic outcomes in the USA—the CHASING COVID Cohort study . BMJ Open . 2021 ; 11 ( 9 ): e048778 . OpenUrl Abstract / FREE Full Text 22. Nash D , Srivastava A , Shen Y , et al. Seroincidence of SARS-CoV-2 infection prior to and during the rollout of vaccines in a community-based prospective cohort of U.S. adults . Sci Rep . 2024 ; 14 ( 1 ). doi: 10.1038/s41598-023-51029-9 OpenUrl CrossRef 23. ↵ Nash D , Rane MS , Robertson MM , et al. Severe acute respiratory syndrome Coronavirus 2 incidence and risk factors in a national, community-based prospective cohort of US adults . Clin Infect Dis . 2023 ; 76 ( 3 ): e375 – e384 . OpenUrl PubMed 24. ↵ van Buuren S , Boshuizen HC , Knook DL . Multiple imputation of missing blood pressure covariates in survival analysis . Stat Med . 1999 ; 18 ( 6 ): 681 – 694 . OpenUrl CrossRef PubMed Web of Science 25. ↵ Raghunathan TE , Lepkowski JM , Van Hoewyk J. A Multivariate Technique for Multiply . 26. ↵ Nguyen CD , Carlin JB , Lee KJ . Model checking in multiple imputation: an overview and case study . Emerg Themes Epidemiol . 2017 ; 14 ( 1 ): 8 . OpenUrl PubMed 27. ↵ Horton NJ , Lipsitz SR , Parzen M. A potential for bias when rounding in multiple imputation . Am Stat . 2003 ; 57 ( 4 ): 229 – 232 . OpenUrl CrossRef Web of Science 28. ↵ Stevens AC , Courtney-Long EA , Okoro CA , Carroll DD . Comparison of 2 disability measures, behavioral risk factor surveillance system, 2013 . Prev Chronic Dis . 2016 ; 13 ( 160080 ). doi: 10.5888/pcd13.160080 OpenUrl CrossRef 29. ↵ Froehlich-Grobe K , Jones D , Businelle MS , Kendzor DE , Balasubramanian BA . Impact of disability and chronic conditions on health . Disabil Health J . 2016 ; 9 ( 4 ): 600 – 608 . OpenUrl PubMed 30. ↵ Soriano JB , Murthy S , Marshall JC , Relan P , Diaz JV . WHO Clinical Case Definition Working Group on Post-COVID-19 Condition. A clinical case definition of post-COVID-19 condition by a Delphi consensus . Lancet Infect Dis . 2022 ; 22 ( 4 ): e102 – e107 . OpenUrl CrossRef PubMed 31. ↵ CDC. Long COVID basics . COVID-19. July 29 , 2024 . Accessed August 26, 2024 . https://www.cdc.gov/covid/long-term-effects/index.html 32. ↵ Ely EW , Brown LM , Fineberg HV , National Academies of Sciences, Engineering, and Medicine Committee on Examining the Working Definition for Long Covid. Long covid defined . N Engl J Med . Published online July 31, 2024. doi: 10.1056/NEJMsb2408466 OpenUrl CrossRef PubMed 33. ↵ Mullin S , Zola J , Lee R , et al. Longitudinal K-means approaches to clustering and analyzing EHR opioid use trajectories for clinical subtypes . J Biomed Inform . 2021 ; 122 ( 103889 ): 103889 . OpenUrl PubMed 34. ↵ Jain AK . Data clustering: 50 years beyond K-means . Pattern Recognit Lett . 2010 ; 31 ( 8 ): 651 – 666 . OpenUrl CrossRef Web of Science 35. ↵ M CC . Pattern Recognition and Machine Learning . Springer ; 2006 . 36. ↵ Petitjean F , Ketterlin A , Gançarski P. A global averaging method for dynamic time warping, with applications to clustering . Pattern Recognit . 2011 ; 44 ( 3 ): 678 – 693 . OpenUrl CrossRef 37. ↵ Huang X , Ye Y , Xiong L , Lau RYK , Jiang N , Wang S. Time series k -means: A new k - means type smooth subspace clustering for time series data . Inf Sci (Ny) . 2016 ; 367-368 : 1 – 13 . OpenUrl 38. ↵ Aschenbruck R , Szepannek G , Wilhelm AFX . Imputation strategies for clustering mixed-type data with missing values . J Classif . 2023 ; 40 ( 1 ): 2 – 24 . OpenUrl CrossRef 39. Ding Q , Lu Y , Herrin J , Zhang T , Marrero DG . Uncovering heterogeneous cardiometabolic risk profiles in US adults: the role of social and behavioral determinants of health . BMJ Open Diabetes Res Care . 2023 ; 11 ( 5 ): e003558 . OpenUrl Abstract / FREE Full Text 40. ↵ Harder AA , Olbricht GR , Ekuma G , Hier DB , Obafemi-Ajayi T. Multiple imputation for robust cluster analysis to address missingness in medical data . IEEE Access . 2024 ; 12 : 42974 – 42991 . OpenUrl 41. ↵ Rubin DB . Multiple Imputation for Nonresponse in Surveys . 99th ed. John Wiley & Sons ; 2009 . 42. ↵ White IR , Royston P , Wood AM . Multiple imputation using chained equations: Issues and guidance for practice . Stat Med . 2011 ; 30 ( 4 ): 377 – 399 . OpenUrl CrossRef PubMed 43. ↵ Ballouz T , Menges D , Anagnostopoulos A , et al. Recovery and symptom trajectories up to two years after SARS-CoV-2 infection: population based, longitudinal cohort study . BMJ . 2023 ; 381 : e074425 . OpenUrl Abstract / FREE Full Text 44. ↵ O’Mahoney L , Khunti K. Long covid: risk factors, outcomes, and future directions for research . BMJ Med . 2022 ; 1 ( 1 ): e000257 . OpenUrl CrossRef 45. ↵ Sahanic S , Tymoszuk P , Luger AK , et al. COVID-19 and its continuing burden after 12 months: a longitudinal observational prospective multicentre trial . ERJ Open Res . 2023 ; 9 ( 2 ). doi: 10.1183/23120541.00317-2022 OpenUrl Abstract / FREE Full Text 46. ↵ Evans RA , McAuley H , Harrison EM , et al. Physical, cognitive, and mental health impacts of COVID-19 after hospitalisation (PHOSP-COVID): a UK multicentre, prospective cohort study . Lancet Respir Med . 2021 ; 9 ( 11 ): 1275 – 1287 . OpenUrl PubMed 47. ↵ Xie Y , Xu E , Al-Aly Z. Risks of mental health outcomes in people with covid-19: cohort study . BMJ . 2022 ; 376 : e068993 . OpenUrl Abstract / FREE Full Text 48. ↵ Mehandru S , Merad M. Pathological sequelae of long-haul COVID . Nat Immunol . 2022 ; 23 ( 2 ): 194 – 202 . OpenUrl CrossRef PubMed 49. ↵ Gottlieb M , Spatz ES , Yu H , et al. Long COVID clinical phenotypes up to 6 months after infection identified by latent class analysis of self-reported symptoms . Open Forum Infect Dis . 2023 ; 10 ( 7 ): ofad277 . OpenUrl PubMed 50. ↵ Blankestijn JM , Abdel-Aziz MI , Baalbaki N , et al. Long COVID exhibits clinically distinct phenotypes at 3-6 months post-SARS-CoV-2 infection: results from the P4O2 consortium . BMJ Open Respir Res . 2024 ; 11 ( 1 ): e001907 . OpenUrl Abstract / FREE Full Text View the discussion thread. Back to top Previous Next Posted November 04, 2025. Download PDF Supplementary Material Data/Code Email Thank you for your interest in spreading the word about medRxiv. NOTE: Your email address is requested solely to identify you as the sender of this article. Your Email * Your Name * Send To * Enter multiple addresses on separate lines or separate them with commas. You are going to email the following Long COVID Longitudinal Symptoms Burden Clusters Within A National Community-Based Cohort Message Subject (Your Name) has forwarded a page to you from medRxiv Message Body (Your Name) thought you would like to see this page from the medRxiv website. Your Personal Message CAPTCHA This question is for testing whether or not you are a human visitor and to prevent automated spam submissions. Share Long COVID Longitudinal Symptoms Burden Clusters Within A National Community-Based Cohort Yanhan Shen , Zach Shahn , McKaylee M. Robertson , Kelly Gebo , Denis Nash , the CHASING COVID Cohort Study Team medRxiv 2025.11.01.25339283; doi: https://doi.org/10.1101/2025.11.01.25339283 Share This Article: Copy Citation Tools Long COVID Longitudinal Symptoms Burden Clusters Within A National Community-Based Cohort Yanhan Shen , Zach Shahn , McKaylee M. Robertson , Kelly Gebo , Denis Nash , the CHASING COVID Cohort Study Team medRxiv 2025.11.01.25339283; doi: https://doi.org/10.1101/2025.11.01.25339283 Citation Manager Formats BibTeX Bookends EasyBib EndNote (tagged) EndNote 8 (xml) Medlars Mendeley Papers RefWorks Tagged Ref Manager RIS Zotero Tweet Widget Facebook Like Google Plus One Subject Area Infectious Diseases (except HIV/AIDS) Subject Areas All Articles Addiction Medicine (568) Allergy and Immunology (863) Anesthesia (299) Cardiovascular Medicine (4425) Dentistry and Oral Medicine (443) Dermatology (382) Emergency Medicine (607) Endocrinology (including Diabetes Mellitus and Metabolic Disease) (1507) Epidemiology (15221) Forensic Medicine (30) Gastroenterology (1123) Genetic and Genomic Medicine (6588) Geriatric Medicine (667) Health Economics (997) Health Informatics (4524) Health Policy (1368) Health Systems and Quality Improvement (1612) Hematology (540) HIV/AIDS (1264) Infectious Diseases (except HIV/AIDS) (15910) Intensive Care and Critical Care Medicine (1103) Medical Education (623) Medical Ethics (145) Nephrology (667) Neurology (6588) Nursing (346) Nutrition (998) Obstetrics and Gynecology (1143) Occupational and Environmental Health (956) Oncology (3331) Ophthalmology (970) Orthopedics (369) Otolaryngology (420) Pain Medicine (435) Palliative Medicine (129) Pathology (663) Pediatrics (1690) Pharmacology and Therapeutics (691) Primary Care Research (710) Psychiatry and Clinical Psychology (5440) Public and Global Health (9219) Radiology and Imaging (2195) Rehabilitation Medicine and Physical Therapy (1369) Respiratory Medicine (1196) Rheumatology (593) Sexual and Reproductive Health (710) Sports Medicine (529) Surgery (710) Toxicology (99) Transplantation (289) Urology (265) (function(){function c(){var b=a.contentDocument||a.contentWindow.document;if(b){var d=b.createElement('script');d.innerHTML="window.__CF$cv$params={r:'9ffa8faa69a30704',t:'MTc3OTQzOTY5OQ=='};var a=document.createElement('script');a.src='/cdn-cgi/challenge-platform/scripts/jsd/main.js';document.getElementsByTagName('head')[0].appendChild(a);";b.getElementsByTagName('head')[0].appendChild(d)}}if(document.body){var a=document.createElement('iframe');a.height=1;a.width=1;a.style.position='absolute';a.style.top=0;a.style.left=0;a.style.border='none';a.style.visibility='hidden';document.body.appendChild(a);if('loading'!==document.readyState)c();else if(window.addEventListener)document.addEventListener('DOMContentLoaded',c);else{var e=document.onreadystatechange||function(){};document.onreadystatechange=function(b){e(b);'loading'!==document.readyState&&(document.onreadystatechange=e,c())}}}})();