Symptom clusters and symptom networks of symptom experiences in patients with SARS-CoV-2 infection Authors: Hongmin Ye, Xiuni Gan, Wen Zhou, Yan Gao, Zhechuan Mei, Qiulan Zheng, Xiaoqing Luo, Chunlan Yuan, Yan Wu Published: 2024 ABSTRACT AbstractContextSymptom clusters and symptom networks can potentially enhance the precision in managing symptoms. However, limited research has been conducted on the symptom network experienced by patients infected with Severe Acute Respiratory Syndrome Coronavirus 2.ObjectivesTo identify the composition of symptom clusters in SARS-CoV-2 infected patients, establish a symptom network to explore the centrality indices, and investigate independent risk factors influencing the occurrence of symptom clusters.MethodsBetween February 2022 and June 2023, a total of 418 patients diagnosed with SARS-CoV-2 infection were recruited in the Second Affiliated Hospital of Chongqing Medical University. A symptom questionnaire was utilized to assess three dimensions encompassing a comprehensive range of 40 symptoms. Principal component analysis was employed to identify distinct symptom clusters, while network analysis elucidated the interconnections among these symptoms. Univariate analysis and multiple linear regression analysis were conducted to investigate the factors influencing the manifestation of these symptom clusters.ResultsEight symptom clusters were identified, namely the nasopharyngeal-related symptom cluster, the circulatory-related symptom cluster, the neural-related symptom cluster, the physical-related symptom cluster, the digestive-related symptom cluster, the respiratory-related symptom cluster, the fever-related symptom cluster, and the sensory-related symptom cluster. The three centrality indices with the highest values were chest tightness (rs = 7.84, rc = 0.013, rb = 6.99), muscle aches (rs = 7.32, rc = 0.013, rb = 2.72), and smell abnormality (rs = 6.56, rc = 0.011, rb = 4.58). Variables including age, gender, income, education, hyperlipidemia, chronic bronchitis, and tumor were associated with the occurrence of these eight symptom clusters.ConclusionThe findings of this study highlight the necessity to explore symptom clusters and symptom networks in order to enhance the effectiveness of symptom management in patients with SARS-CoV-2 infection. Particularly crucial is the evaluation of centrality indices as an integral component of caring for such patients. Early detection of high-risk individuals within each symptom cluster can provide a scientific foundation for developing interventions that will optimize patient prognosis.Keywords: Symptom clusters, Network analysis, Symptom management, SARS-CoV-2 infection, Nursing Introduction A newly discovered zoonotic virus known as coronavirus (Cov) is responsible for illnesses like the common cold, Middle East Respiratory Syndrome (MERS), and Severe Acute Respiratory Syndrome (SARS) [1]. In December 2019, Corona Virus Disease 2019 (COVID-19) was discovered in Wuhan, China, emerging as one of the most impactful infectious diseases of this century. The virus causing it was designated as Severe Acute Respiratory Syndrome Coronavirus 2 (SARS-CoV-2) [2]. On March 11, 2023, WHO declared the global COVID-19 pandemic to have reached its third anniversary. Since December 2019, there have been over 760 million reported cases of illness and approximately 6.9 million documented fatalities worldwide, yet the true figure is thought to be higher [3]. In the current context of open policies, SARS-CoV-2 continues to adapt to mutation and evolution. Following initial infection, patients may become reinfected with either the same or different mutant strains. Despite a decrease in the incidence rate, hundreds of thousands of individuals are still being infected each week [3]. Angiotensin-converting enzyme 2 (ACE2), which is expressed in several human tissues, is bound by SARS-CoV-2 when it enters human tissue cells through its surface S protein [4,5]. This process produces a range of signs and symptoms, including fever, cough, fatigue, dyspnea, diarrhea, and so on. The potential synergistic and reinforcing nature of these symptoms implies that they can exacerbate other manifestations, add to the disease's burden, and potentially lead to long-term consequences [6]. Effective symptom management measures can not only improve the illness experience of patients with SARS-CoV-2 infection during the infection period but also contribute to their recovery process. Symptom cluster management focuses on the internal relationship between several associated symptoms. Compared to managing individual symptoms, addressing multiple symptoms as clusters can enhance the efficiency of symptom management. A symptom cluster is composed of two or more stable, accompanied, and interrelated symptoms, which are relatively independent of other symptom clusters [7]. However, due to different survey tools and statistical methods, the number and composition of symptom clusters are different, and the mechanism of interaction between symptoms cannot be discovered [8]. The use of factor analysis is prevalent in symptom grouping, as it explores the interrelationships between variables based on a matrix of correlation coefficients. Additionally, principal component analysis (PCA) generates principal component scores for parametric analyses such as tests of differences between groups, which can accurately explain the correlation between variables and analyze the structure of the data [9]. Contemporaneous symptom network is a new research paradigm, which refers to the symptom network composed of the symptoms of the same patient group at a certain measurement time point [10]. It serves as an aggregate depiction of disease-related symptoms and facilitates quantitative analysis of interconnections among diverse symptoms through complex network analysis. It can provide more specific indicators, such as centrality and density [11]. The precision and efficacy of clinical therapies can be improved by using these clinically meaningful data to help medical practitioners identify the primary symptoms and take into account how different symptoms interact. Research on symptom clusters and symptom networks remains limited, with earlier studies primarily focusing on the characteristics of symptoms in patients infected with SARS-CoV-2. Therefore, the purpose of this study is to explore the composition of symptom clusters in patients with SARS-CoV-2 infection, establish a symptom network, and explore the centrality indices to guide the clinical development of symptom management strategies. In addition, patient demographics were collected and analyzed to explore independent risk factors associated with the formation of symptom clusters, which can inform the precise implementation of symptom management. Methods Study design and participants A cross-sectional design was conducted to answer the research questions. Patients with SARS-CoV-2 infection were recruited from February 2022 to June 2023 at the Second Affiliated Hospital of Chongqing Medical University using convenience sampling. Eligible participants met the following criteria: 1) positive results from a novel coronavirus nucleic acid test or antigen kit; 2) age greater than 18 years old; 3) voluntary participation in the study with informed consent; 4) basic reading, comprehension, and objective expression of inner thoughts; and 5) stable vital signs. Exclusion criteria included pregnancy, significant heart, brain, kidney, or other organ malfunction diagnosis, and co-occurring major mental diseases. The ratio of each item to the sample size for factor analysis is typically between 1: 5 to 1: 10. With an estimated loss rate of visits at 20 %, a sample size ranging from 324 to 648 cases was calculated based on the inclusion of all 54 items in the questionnaire. The symptom scale Demographic and disease characteristics Demographic characteristics that could potentially influence the individual's symptom experiences were assessed. Based on a comprehensive analysis of existing literature and clinical expertise, these factors primarily included gender, age, educational level, per capita monthly income, underlying medical conditions, and smoking history. COVID-19 symptoms Using the Symptom Management Theory (SMT) as the theoretical framework [6], the questionnaire formed an item pool through literature analysis, expert meetings, and up-to-date domestic and international treatment guidelines. The questionnaire comprised a total of 40 symptoms, each assessed based on three dimensions: frequency, severity, and distress. Asymptomatic patients scored 0, while symptomatic patients scored 1. For symptomatic individuals, both symptom severity and distress level were evaluated using a Likert scale ranging from 1 to 5 points representing not serious at all, mild, moderate, severe, and extremely serious. The degree of distress ranges from 1 to 5 points, with none, little, some, considerable, and very large. Validity and rigour Our research group developed a questionnaire to assess symptoms of SARS-CoV-2 infection, which was utilized in this investigation. Ten experts specializing in infectious disease, critical care medicine, and respiratory medicine evaluated the items and confirmed their content validity. Each item's content validity ranged from 0.8 to 1.0, and the questionnaire's overall content validity was 0.83. Furthermore, based on a modified questionnaire and a pre-survey involving 30 individuals, the Cronbach's α coefficient was calculated as 0.946. Procedures The questionnaires were distributed through WeChat, an online application (https://weixin.qq.com/). Participants who met the study criteria accessed a dedicated website (https://www.wenjuan.com/) to complete the survey. Submission was only possible after all study subject options had been completed. In addition, investigators underwent training before the survey to ensure that they were familiar with data collection requirements and methods. The questionnaire was attached with the purpose of the survey and filling instructions, while also offering notes in Mandarin or any other appropriate dialect if necessary. For participants who were unable to complete the questionnaire on their own, their family members answered the questionnaire on their behalf. By using these techniques, participants are guaranteed to receive the same guidance, and the approach is kept consistent throughout the data collection procedure. Ultimately, 425 questionnaires were distributed, out of which 418 valid responses were obtained, resulting in an effective recovery rate of 98.3 %. The surveys with less than 3 min to answer will be excluded. All questionnaires were fully complete. Statistical analysis Exploratory factor analysis for symptom cluster identification Excel 2021 was used for data processing, and two people checked the input to ensure the accuracy of the data. SPSS26.0 for Windows (IBM, Armonk, NY) was used for statistical analysis. Mean and standard deviation (M±SD) were used for measurement data, and frequency and percentage (%) were used for counting data. Independent sample T-test, one-way analysis of variance, and multiple linear regression analysis were employed ed to analyze the independent risk factors of symptom clusters. A significance level of P < 0.05 was considered statistically significant. Firstly, the correlation matrix between symptoms in the severity dimension is examined to ensure an adequate level of correlation for factor analysis. If the Kaiser-Meyer-Olkin (KMO) values are greater than 0.7 and the significance of Bartlett's sphericity test is less than 0.05, it indicates that the data is suitable for factor analysis [12]. Then, an exploratory factor analysis (EFA) was performed using principal component analysis, with a preference for employing the maximum variance method for factor rotation. For the sake of expressing the symptoms accurately, symptoms with an incidence of ≤30 % were excluded [9,13]. The number of factors is determined by Keyser's rule and lithotripsy diagram, and the factors with eigenvalues greater than 1 are retained. The selection of common factors was guided by the following principles: the load value of symptoms under each factor was ≥0.5, symptoms were removed if they loaded at least 0.5 on multiple factors, and each factor contains at least 2 entries. After reviewing their apparent validity and clinical plausibility, the final clusters were determined. Each cluster was assigned a name to reflect its possible underlying cause. Network analysis to determine symptom associations R version 4.3.2 was used to perform all statistical analyses. We constructed a contemporaneous network consisting of 24 symptoms, where each node represents a symptom and an edge signifies the correlation between two symptoms. The thicker the edge, the stronger the correlation. Initially, we employed the qgraph package to visualize the network, utilizing a spring-based algorithm for layout generation. This algorithm placed nodes with stronger connections closer together in the center of the network, resulting in an undirected relational network. Then, we performed a centrality analysis using the following three centrality indices: strength, betweenness, and closeness. Node centrality serves as an indicator for identifying core symptoms within the network, with nodes exhibiting high centrality indices being considered as core nodes. Strength is determined by summing up the absolute edge weights connected to a specific node, thereby indicating which node possesses the strongest overall connection. Strength provides a metric for identifying the most connected nodes within the network. Betweenness measures the number of times a node lies on the shortest path between two other nodes, suggesting which nodes can act as bridges between other nodes in the network. Closeness sums up the average distance of a node from all other nodes in the network, which allows one to identify nodes that significantly affect other nodes in the network. Besides, we used the R package bootnet to perform the Bootstrapping method to assess the accuracy and stability of the network. Based on 1000 bootstrap iterations, we tested the significance between the edge weights and the accuracy of the edge using the bootstrap confidence region. To assess the stability of centrality exponential order, we employed a combination of case and node deletion subset bootstrap technique along with correlated stability coefficient (CS coefficient). An ideal CS coefficient should be at least 0.25 but preferably greater than 0.5 [14]. Lastly, we used the R package bootnet to run a difference test to examine potential differences in the edge weights or centrality of certain nodes. This method calculates the difference in node centrality or edge weights and constructs confidence intervals from their repeated sampling. According to the null hypothesis test, if zero lies within the confidence interval it means that the difference in node centrality or edge weights is not significant. Results Characteristics of the participants This study included 418 participants, ranging in age from 19 to 86 years, with an average age of (37.75 ± 15.40) years. Most participants were female(n = 335,80.10 %). Additional demographic information is presented in Table 1. Table 1Characteristics of participants (n = 418).Table 1Characteristicsn(%)Cluster 1 (x‾ ±s)Cluster 2 (x‾ ±s)Cluster 3 (x‾ ±s)Cluster 4 (x‾ ±s)Cluster 5 (x‾ ±s)Cluster 6 (x‾ ±s)Cluster 7 (x‾ ±s)Cluster 8 (x‾ ±s)GenderMale83(19.9)2.00 ± 0.821.81 ± 1.011.65 ± 0.781.94 ± 1.141.49 ± 0.722.92 ± 1.253.45 ± 1.221.93 ± 1.24Female335(80.1)2.67 ± 1.011.85 ± 0.922.16 ± 1.012.01 ± 0.951.61 ± 0.812.94 ± 1.092.77 ± 1.061.90 ± 1.13t value−6.271−0.345−5.039−0.547−1.243−0.1454.6920.186P valueP < 0.001∗0.730P < 0.001∗0.5850.2150.885P < 0.001∗0.853Agemean37.75 ± 15.40range19–8618–44319(76.3)2.69 ± 0.991.82 ± 0.922.14 ± 1.002.04 ± 0.951.62 ± 0.822.92 ± 1.082.72 ± 1.041.86 ± 1.1145–5945(10.8)2.25 ± 1.021.88 ± 1.021.98 ± 0.971.76 ± 0.971.48 ± 0.672.86 ± 1.193.10 ± 1.292.00 ± 1.2460–7439(9.3)1.91 ± 0.782.01 ± 1.021.72 ± 0.831.76 ± 1.071.56 ± 0.772.96 ± 1.253.88 ± 1.001.88 ± 1.19≥7515(3.6)1.80 ± 0.811.90 ± 0.831.48 ± 0.642.31 ± 1.451.40 ± 0.573.33 ± 1.443.57 ± 1.182.77 ± 1.43F value12.2050.5184.1622.3720.7410.72216.4953.154P valueP < 0.001∗0.6700.006∗0.0700.5280.539P < 0.001∗0.025∗EducationSenior Secondary Schools or below58(13.9)1.80 ± 0.842.13 ± 0.991.73 ± 0.822.03 ± 1.281.50 ± 0.772.99 ± 1.393.75 ± 1.092.07 ± 1.32Undergraduate271(64.8)2.61 ± 0.991.80 ± 0.942.13 ± 1.031.99 ± 0.941.62 ± 0.802.94 ± 1.092.74 ± 1.101.88 ± 1.14Master's Degree or above89(21.3)2.78 ± 0.981.78 ± 0.882.05 ± 0.891.99 ± 0.931.55 ± 0.802.87 ± 1.042.85 ± 0.991.88 ± 1.07F value20.3433.1434.0350.0600.7240.24921.3710.670P valueP < 0.001∗0.044∗0.018∗0.9420.4860.779P < 0.001∗0.512Monthly personal income¥2000 or less85(20.3)2.62 ± 1.061.94 ± 0.872.04 ± 0.882.09 ± 1.041.68 ± 0.843.02 ± 1.083.10 ± 0.991.92 ± 1.13¥2000 through ¥500054(12.9)2.46 ± 1.042.08 ± 1.072.15 ± 1.032.22 ± 1.111.79 ± 1.052.88 ± 1.073.09 ± 1.151.97 ± 1.31¥5000 and higher279(66.7)2.53 ± 0.991.77 ± 0.922.05 ± 1.011.92 ± 0.941.53 ± 0.712.92 ± 1.152.80 ± 1.151.89 ± 1.13F value0.4373.0330.2492.6393.2010.3273.1700.130P value0.6460.049∗0.7790.0730.042∗0.7220.043∗0.878SmokingYes14(3.3)1.71 ± 0.371.71 ± 0.831.79 ± 0.891.93 ± 1.411.54 ± 0.572.32 ± 0.872.96 ± 1.372.11 ± 1.35Quit smoking32(7.7)2.06 ± 0.841.95 ± 1.041.63 ± 0.741.98 ± 1.311.63 ± 0.833.47 ± 1.343.73 ± 1.171.97 ± 1.43Never372(89)2.61 ± 1.021.84 ± 0.932.11 ± 1.002.00 ± 0.941.59 ± 0.802.91 ± 1.102.83 ± 1.081.89 ± 1.12F value9.5600.3214.0340.0350.0635.91310.0060.281P valueP < 0.001∗0.7260.018∗0.9660.9390.003∗P < 0.001∗0.755COVID-19 vaccinationUnvaccinated21(5.0)2.16 ± 0.881.93 ± 0.831.80 ± 0.881.89 ± 1.311.45 ± 0.593.17 ± 1.203.71 ± 1.011.90 ± 1.25One53(12.7)2.54 ± 1.081.78 ± 0.992.10 ± 1.051.98 ± 0.991.65 ± 0.842.78 ± 1.172.77 ± 1.111.81 ± 1.13Two270(64.6)2.59 ± 1.011.85 ± 0.922.07 ± 0.991.98 ± 0.981.57 ± 0.812.99 ± 1.132.85 ± 1.111.87 ± 1.15Three67(16.0)2.5 ± 1.011.91 ± 1.052.05 ± 0.972.08 ± 0.901.67 ± 0.782.79 ± 1.022.91 ± 1.142.09 ± 1.16Four7(1.7)1.86 ± 0.831.25 ± 0.252.25 ± 1.031.95 ± 1.011.57 ± 0.732.71 ± 1.223.36 ± 1.072.14 ± 0.94F value1.7480.8890.4660.2150.4520.9403.4210.640P value0.1390.4700.7610.9300.7710.4410.009∗0.634ComplicationsYes98(23.4)1.60 ± 0.651.70 ± 0.971.82 ± 0.791.87 ± 1.191.47 ± 0.483.00 ± 1.243.70 ± 1.311.33 ± 0.62t value5.5800.6100.9750.5040.988−0.235−2.8223.508P valueP < 0.0010.5420.3300.6140.3370.8150.0050.003CHDYes9(2.1)2.09 ± 0.802.36 ± 0.911.89 ± 0.891.59 ± 0.891.39 ± 0.492.56 ± 1.043.00 ± 0.711.94 ± 1.36t value1.344−1.6730.5271.2310.7711.018−0.414−0.099P value0.1800.0950.5980.2190.4410.3090.6890.921HypertensionYes39(9.3)2.02 ± 0.922.06 ± 1.041.84 ± 0.881.90 ± 1.111.67 ± 0.712.95 ± 1.103.51 ± 1.142.22 ± 1.30t value3.388−1.5351.4700.634−0.625−0.091−3.614−1.779P value0.001∗0.1260.1420.5270.5320.927P < 0.001∗0.076DiabetesYes16(3.8)2.03 ± 0.882.14 ± 1.142.13 ± 1.001.63 ± 0.741.81 ± 0.753.06 ± 1.263.63 ± 1.241.72 ± 0.80t value2.071−1.287−0.2671.524−1.138−0.469−2.6410.666P value0.039∗0.1990.7900.1280.2560.6390.009∗0.506HyperlipidemiaYes6(1.4)2.53 ± 0.902.33 ± 1.083.08 ± 0.382.22 ± 0.752.08 ± 0.663.08 ± 1.283.58 ± 0.801.5 ± 0.77t value0.009−1.286−6.441−0.573−1.532−0.329−1.4970.872P value0.9930.1990.001∗0.5670.1260.7420.1350.384Chronic bronchitisYes12(2.8)1.73 ± 0.522.17 ± 0.961.67 ± 0.911.61 ± 0.761.33 ± 0.493.83 ± 1.213.79 ± 1.251.96 ± 1.57t value5.216−1.2061.4051.3621.140−2.837−2.803−0.158P valueP < 0.001∗0.2280.1610.1740.2550.005∗0.005∗0.875COPDYes10(2.3)2.04 ± 0.722.10 ± 0.851.70 ± 0.801.97 ± 1.361.45 ± 0.803.25 ± 1.483.35 ± 1.161.75 ± 1.36t value1.574−0.8701.1700.0850.567−0.902−1.2760.436P value0.1160.3850.2430.9320.5710.3680.2030.663TumorYes15(3.5)1.60 ± 0.651.70 ± 0.971.82 ± 0.791.87 ± 1.191.47 ± 0.483.00 ± 1.243.70 ± 1.311.33 ± 0.62t value5.5800.6100.9750.5040.988−0.235−2.8223.508P valueP < 0.001∗0.5420.3300.6140.3370.8150.005∗0.003∗CHD, Coronary heart disease; COPD, Chronic obstructive pulmonary disease.∗ Statistically significant (P < 0.05) Characteristics of participants (n = 418). Symptoms and symptom clusters Table 2 shows the symptom frequency, severity, and distress of the participants. The five most frequently reported symptoms were cough (91.4 %), fever (85.6 %), expectoration (81.1 %), dry throat (79.4 %), and sore throat (76.8 %). Cough, fever, fatigue, sore throat, and dry throat exhibited the highest level of severity and distress. Table 2Symptom prevalence, severity, and distress of participants (n = 418).Table 2SymptomPrevalence(%)Severity(M ± SD)Distress(M ± SD)Fever85.63.00 ± 1.213.18 ± 1.33Chill65.32.37 ± 1.282.38 ± 1.32Cold limbs46.21.85 ± 1.111.85 ± 1.14Shiver40.91.74 ± 1.081.77 ± 1.13Sweating71.82.35 ± 1.252.32 ± 1.26Cough91.43.21 ± 1.273.27 ± 1.34Expectoration81.12.65 ± 1.232.71 ± 1.33Nasal congestion67.52.48 ± 1.352.52 ± 1.40Runny nose68.72.38 ± 1.252.40 ± 1.30Dry throat79.42.77 ± 1.342.75 ± 1.36Sore throat76.82.78 ± 1.392.77 ± 1.41Hoarse voice64.12.25 ± 1.252.25 ± 1.28Loss of appetite69.92.50 ± 1.402.47 ± 1.41Nausea37.61.62 ± 1.001.63 ± 1.02Vomiting23.91.41 ± 0.841.45 ± 0.94Gastric distention28.01.47 ± 0.901.47 ± 0.92Stomachache20.31.32 ± 0.761.33 ± 0.78Diarrhea36.81.55 ± 0.891.57 ± 0.92Constipation23.21.39 ± 0.841.40 ± 0.89Increase in fart volume29.41.47 ± 0.871.45 ± 0.85Dyspnea41.91.72 ± 1.081.74 ± 1.14Wheezing56.02.12 ± 1.312.14 ± 1.36Chest tightness46.91.80 ± 1.111.82 ± 1.13Chest pain23.91.41 ± 0.851.41 ± 0.87Palpitation41.41.72 ± 1.071.72 ± 1.11Dizziness58.12.10 ± 1.222.08 ± 1.24Headaches58.62.17 ± 1.292.19 ± 1.32Insomnia46.91.90 ± 1.191.91 ± 1.24Sleepiness50.71.93 ± 1.211.91 ± 1.22Memory decline49.01.87 ± 1.121.87 ± 1.13Abnormal urination19.61.31 ± 0.751.31 ± 0.75Lumbago38.81.75 ± 1.141.77 ± 1.18Taste abnormality48.12.06 ± 1.362.06 ± 1.39Smell abnormality36.61.74 ± 1.181.71 ± 1.16Tinnitus27.31.44 ± 0.901.42 ± 0.87Eye pain29.91.49 ± 0.901.48 ± 0.88Rash13.21.19 ± 0.611.21 ± 0.65Fatigue73.42.79 ± 1.512.81 ± 1.53Muscle aches63.62.42 ± 1.402.43 ± 1.43Bone and joint pain41.91.88 ± 1.281.89 ± 1.30 Symptom prevalence, severity, and distress of participants (n = 418). Table 3 shows the results of EFA. The KMO value of 0.892 and Bartlett's spherical test of P < 0.01 indicated that the data are accessible for EFA. Finally, eight symptom clusters were defined, which accounted for 65.129 % of the overall variation. These eight components exhibited variance contribution rates ranging from 5.908 % to 10.085 %. Each factor's load levels varied from 0.576 to 0.823. Every symptom remained stable. Table 3Symptom clusters in COVID-19 patients based on severity of 24 symptoms.Table 3SymptomLoadingCluster 1Cluster 2Cluster 3Cluster 4Cluster 5Cluster6Cluster 7Cluster 8Sore throat0.8230.0930.1490.0220.0830.1130.1290.035Hoarse voice0.7270.1060.0750.1340.2290.0080.0970.013Dry throat0.6760.1560.0120.0220.2390.1830.242−0.038Nasal congestion0.640−0.0360.3330.213−0.0100.228−0.2560.171Runny nose0.5900.0420.2070.224−0.0020.370−0.2310.156Chest tightness0.1420.7870.2610.1330.1820.0720.0510.078Wheezing0.0250.7360.1190.0920.1640.2460.2180.036Dyspnea0.0470.7230.1390.189−0.0490.0570.1080.137Palpitation0.1580.7020.1340.1230.331−0.0270.0400.131Bone and joint pain−0.0030.1830.7260.1430.0490.1460.0630.114Muscle aches0.1590.1970.7000.0570.0170.1940.2640.076Lumbago0.1610.1780.6440.1180.2950.026−0.1000.111Headaches0.3170.1230.5950.0160.306−0.0700.1800.051Shiver0.1110.0730.1470.8070.1530.0530.0550.002Chill0.0600.2200.0290.7650.0690.1560.2350.070Cold limbs0.1200.1500.1120.7340.2010.0620.0310.119Nausea0.1360.0830.1080.2720.6600.1190.0640.119Diarrhea0.1370.2040.1410.1130.6500.154−0.1130.011Expectoration0.2180.1060.1170.1630.1960.783−0.0240.042Cough0.2730.1150.1400.0590.0540.7310.1970.042Fever0.1430.1320.0400.379−0.0370.0760.6210.135Fatigue−0.0640.3080.336−0.0440.1570.3160.5760.127Taste abnormality−0.0220.2150.0940.0510.1630.1490.2160.820Smell abnormality0.1840.1120.2950.1610.108−0.0570.0060.794Eigenvalues9.1922.3461.8141.5081.3301.1861.0971.065Variation (%)10.0859.8689.8478.6428.2026.2926.2855.908Cumulative variation (%)10.08519.95329.80038.44146.64352.93559.22065.129 Symptom clusters in COVID-19 patients based on severity of 24 symptoms. Risk factors for symptom clusters The results of the univariate analysis are shown in Table 1. Variables that were statistically significant in the univariate analysis were used as independent variables, while the scores of each symptom cluster served as dependent variables in the multiple linear regression analysis, and the results are shown in Table 4. Gender was a risk factor for Cluster 1(adjust R2 = 0.111, F = 7.529, P < 0.001). Income and education were risk factors for Cluster 2 (adjust R2 = 0.016, F = 4.493, P = 0.012). Gender, age, and hyperlipidemia were risk factors for Cluster 3 (adjust R2 = 0.064, F = 6.734, P < 0.001). Income was a risk factor for Cluster 5 (adjust R2 = 0.007, F = 3.981, P = 0.047). Comorbid chronic bronchitis was a risk factor for Cluster 6(adjust R2 = 0.015, F = 4.141, P = 0.017). Gender and age were risk factors for Cluster 7 (adjust R2 = 0.106, F = 5.966, P < 0.001). Age and comorbid tumor were risk factors for Cluster 8 (adjust R2 = 0.023, F = 5.809, P = 0.003). Table 4Multiple linear regression analysis of symptom clusters of COVID-19 patients (n = 418).Table 4BStd.ErrorBetatP value95 % CICluster 1Constant1.4420.4473.2270.0010.564–2.320Gender0.3810.1510.1502.5180.012∗0.084–0.678Education0.1990.1010.1161.9680.0500.000–0.397Cluster 2Constant2.5440.23810.6990.0002.077–3.011Education−0.1830.079−0.115−2.3340.020∗−0.338∼-0.029Income−0.1290.057−0.112−2.2660.024∗−0.242∼-0.017Cluster 3Constant1.8050.4394.1110.0000.942–2.669Gender0.4250.1480.1722.8630.004∗0.133–0.717Age−0.1930.077−0.157−2.5180.012∗−0.344∼-0.042Hyperlipemia1.2180.3950.1473.0840.002∗0.442–1.995Cluster 5Constant1.8260.12414.7240.0001.582–2.070Income−0.0950.048−0.097−1.9950.047∗−0.189∼-0.001Cluster 6Constant2.7250.3697.3790.0001.999–3.451Chronic bronchitis0.9560.3320.1422.8780.004∗0.303–1.608Cluster 7Constant3.4590.5875.8880.0002.304–4.613Gender−0.4910.169−0.174−2.9000.004∗−0.824∼-0.158Age0.2680.1060.1912.5270.012∗0.059–0.476Cluster 8Constant1.6580.11314.6920.0001.436–1.88Age0.1980.0720.1382.7680.006∗0.057–0.339Tumor−0.8000.308−0.129−2.5950.010∗−1.405~-0.194∗ Statistically significant (P < 0.05) Multiple linear regression analysis of symptom clusters of COVID-19 patients (n = 418). Symptom network and centrality indices Fig. 1A shows the symptom network of patients with SARS-CoV-2 infection. Fig. 1B illustrates three centrality indices: strength, closeness, and betweenness. Among these centrality indices, chest tightness (rs = 7.84, rc = 0.013, rb = 6.99), muscle aches (rs = 7.32, rc = 0.013, rb = 2.72), and smell abnormality (rs = 6.56, rc = 0.011, rb = 4.58) exhibit the highest values across the network. Fever(rs = 5.00, rc = 0.009, rb = 0.00) demonstrates lower centrality compared to other symptoms. Symptom network and centrality measures in the full sample network (N = 418). (A) Symptom network of 24 symptoms; (B) strength, betweenness, and closeness of 24 symptoms. Accuracy, stability, and difference test of symptom network Fig. 2A shows the results of the edge weight accuracy test. Given the narrow bootstrap confidence interval, the network appears to have acceptable accuracy. Fig. 2B shows the results of the stability test for centrality, with a CS coefficient of 0.593 for the expected influence and 0.517 for the strength, indicating that the network remains stable. Accuracy and stability of the symptom network. (A) Bootstrap analysis results of the edge weights; (B) correlation stability coefficient for strength, expected influence, and closeness. Fig. 3A shows the results of the bootstrapped edge difference test. “Nasal congestion and Runny nose”, “Taste abnormality and Smell abnormality” and “Expectoration and Cough” exhibited the strongest edge weights. Fig. 3B shows the results of the bootstrapped node difference test. Chest tightness stood out as significantly distinct compared to other nodes. Results of difference tests. (A) Bootstrapped difference test for edges; (B) bootstrapped difference test for nodes. Discussion This study explored the symptom clusters and symptom networks of patients with SARS-CoV-2 infection. We identified eight symptom clusters: nasopharyngeal-related symptom cluster, circulatory-related symptom cluster, neural-related symptom cluster, physical-related symptom cluster, digestive-related symptom cluster, respiratory-related symptom cluster, fever-related symptom cluster, sensory-related symptom cluster. Contemporaneous symptom networks can assist researchers in identifying the most significant symptoms within the network structure. According to this study, cough, fever, expectoration, dry throat, and sore throat were the most common and severe symptoms in patients with SARS-CoV-2 infection. However, these symptoms demonstrated relatively lower centrality compared to others. The core symptoms were chest tightness, muscle aches, and smell abnormality. Cluster 1 is the nasopharyngeal-related symptom cluster, including sore throat, hoarse voice, dry throat, nasal congestion, and runny nose. The mechanism may be related to the binding of SARS-CoV-2 to the ACE2 receptor expressed in the nasopharynx, triggering an interleukin-mediated immune response [15]. Leakage of cellular components and tissue fluid can stimulate nerve endings, exacerbating pain. We found a high incidence, severity, and distress of pharyngeal discomfort in patients with SARS-CoV-2 infection, which was similar to the findings of previous research [16]. Previous studies suggested that traditional Chinese medicine has a therapeutic effect in treating pharyngeal symptoms caused by SARS-CoV-2 infection [17]. Nevertheless, more research on its application value can be further explored in the future. Cluster 2 is the circulatory-related symptom cluster, including chest tightness, wheezing, dyspnea, and palpitation. Previous studies also demonstrated that shortness of breath and chest pain were part of a particular COVID-19 symptom cluster [18]. Some patients reported that the circulatory-related symptoms as a significant factor leading to hospital admission. In addition, this symptom cluster may have a long-term impact [19]. The direct infection of cardiac cells by SARS-CoV-2 can lead to symptoms such as fatigue, dyspnea, and chest tightness [20]. We discovered that chest tightness was the core symptom of the three central indices, which suggests a focus on intervention for this symptom. Once the patient presents with comparable symptoms, prompt collection of medical history, basic laboratory tests, and timely consultation with cardiovascular specialists are essential. Cluster 3 is the neural-related symptom cluster, including bone and joint pain, muscle aches, lumbago, and headache. The discovery aligns with the findings of an earlier investigation that showed the common pain symptoms experienced by patients [21]. The potential of SARS-CoV-2 to invade nerves can be spread by hematogenous peripheral nerve pathways or retrograde axonal transport [22]. Moreover, SARS-CoV-2 binds to ACE2 receptors, which are expressed in neural and muscular tissue cells, causing associated tissue involvement [23]. Besides, we found that the higher coefficient of the three central indices of muscle soreness among all included symptoms. Pain seriously affects patients’ daily lives and brings miserable illness experience. Medical personnel should pay more attention to these patients by giving appropriate medication, physical therapy, and psychotherapy to effectively meet their analgesic needs. Hyperlipidemia also plays a role in this symptom cluster. Previous studies have demonstrated poorer outcomes in COVID-19 patients with comorbid hyperlipidemia [24]. While SARS-CoV-2 viral infection also causes fluctuations in lipid levels, this may be exacerbating the occurrence of this symptom cluster [25]. It also requires special attention in clinical practice. Cluster 4 is the physical-related symptom cluster, including shivering, chill, and cold limbs. We confirmed that SARS-CoV-2 infection would induce systemic physical discomfort in the early stage. It is worth noting that the early stage of SARS-CoV-2 infection has the highest viral titer. The viral titer has dropped when there are severe clinical symptoms [26]. Thus, identifying this symptom cluster is helpful to screen out patients with SARS-CoV-2 infection as early as possible to prevent the progression of the disease. Cluster 5 is the digestive-related symptom cluster, including nausea and diarrhea. SARS-CoV-2 can change the activity of ACE2, lead to intestinal flora imbalance, and then cause gastrointestinal discomfort [27]. Additionally, an earlier study also discovered that early stages of COVID-19 infection are characterized by intestinal complications like anorexia, diarrhea, nausea, and vomiting [28]. Relevant studies have provided evidence supporting that probiotics can alleviate intestinal symptoms in patients with SARS-CoV-2 infection [29], but further theoretical and experimental data is necessary for a comprehensive investigation. Cluster 6 is the respiratory-related symptom cluster, including cough and expectoration. Previous studies explored that this symptom cluster was combined with obvious abdominal pain, diarrhea, and confusion [30]. COVID-19 is a viral infectious disease mainly involving the lungs [31], and its inflammatory response may exacerbate the symptoms of patients with chronic bronchitis and even affect the patient in the long term [32]. Therefore, it is particularly important to develop an individualized pulmonary rehabilitation functional training program for patients. More clinical studies are needed to analyze the respiratory impairment that occurs with COVID-19, so that better treatment modalities can be found to improve the prognosis and reduce the disease burden and economic burden of patients. Cough has been extensively researched as the most common and severe symptom in patients with SARS-CoV-2 infection in earlier studies. This study is no exception, however, our study found that the three central indices of cough were lower than other symptoms, which was consistent with the findings of the previous study [33]. Cluster 7 is the fever-related symptom cluster, including fever and fatigue. Fever is a prevalent clinical manifestation in patients with SARS-CoV-2 infection [21], mainly due to the acute inflammatory response caused by SARS-CoV-2 [34]. The heightened body temperature will consume more energy, while the increased interleukins will aggravate fatigue. Relevant studies have discovered that the majority of patients with SARS-CoV-2 infection experienced persistent problems with housekeeping [35,36]. Therefore, corresponding rehabilitation training should be taken for this group to improve their prognosis. Cluster 8 is the sensory-related symptom cluster, including taste abnormality and smell abnormality. The direct damage and neuroinvasiveness of SARS-CoV-2 to gustatory cells may be the primary cause of gustatory disorders [37]. The indirect etiology may also be related to tumor radiotherapy. This contributed to the high percentage of tumor patients in this study who developed this symptom cluster [38]. Notably, compared with parainfluenza virus, rhinovirus, and other coronaviruses, SARS-CoV-2 causes more olfactory disorders [39,40]. This symptom cluster may persist even after recovery from SARS-CoV-2 infection. A self-reported meta-analysis pointed out that 5 % of patients still had smell or taste disorders half a year after infection [41], which may also affect patients’ cognitive function and mental health [42]. Furthermore, we found that smell abnormality has the second-highest betweenness centrality. Based on this finding, we hypothesize that this symptom may play a role as a bridge, but further research is required to verify it. Therefore, intervention in this symptom may potentially alleviate other symptoms. Relevant evidence demonstrated that symptoms can be improved by oral care, olfactory training, and traditional Chinese medicine [23]; nevertheless, the effect of related research remains to be tested. We also found that demographic data such as gender, age, income, and education level significantly affect the occurrence of certain symptom clusters. This study showed that compared with men, women accounted for a lower proportion of Cluster 7 and a higher proportion of Cluster 1 and Cluster 3, which was similar to the results of a previous study [43]. On the contrary, most studies concluded that males and females were equally susceptible to SARS-CoV-2 infection, whereas male patients even had a worse prognosis [44]. It is evident that the association between gender and SARS-CoV-2 infection is controversial, and the fact that most of the subjects included in this study were female may have created a bias. Therefore, further studies are needed to investigate the relationship in the future. Besides, this study showed differences in the clinical presentation of COVID-19 patients in different age groups, which is similar to previous studies. However, it is important to acknowledge that diagnosing COVID-19 in elderly patients is extremely challenging because of the atypical clinical manifestations such as delirium and hypoxia [45]. And such patients are at greater risk of developing severe cases and mortality. Therefore, attention should be focused on older COVID-19 patients, while identifying symptom clusters can help enhance the accuracy of disease diagnosis. In addition, our study also found an association between income and symptom clusters, which may be related to its direct impact on patient health-seeking behavior. Another interesting finding of this study is the inverse correlation observed between the high level of education and Cluster 2. Knowledge and attitudes related to SARS-CoV-2 infection can influence prevention behavior, but only explain 26 % of the variance [46]. This suggests that despite the impact of knowledge and attitude on behavior, there are additional influential factors at play, as supported by the present study. Therefore, health education on the prevention and control of SARS-CoV-2 infection should be comprehensively implemented to improve behavioral interventions for patients. Limitations Although our research has many advantages, it also has some limitations. There are a variety of options available for identifying symptom clusters, but the best practices have not been established. For example, Symptom Graph is a symptom mining system, that generates a symptom graph based on clinical notes to discover the symptom clusters and the clinical parameters associated with the symptom clusters. However, our study did not use recent symptom clustering techniques, instead opting for EFA and network analysis based on the fit of the study purpose. This leaves the validity of other clustering methods unknown and more research is needed in the future to determine the optimal clustering method. Besides, the data of the cross-sectional survey does not allow for causal inference between symptoms. In the future, it is necessary to develop trajectories of dynamic symptom networks and centrality indices in longitudinal data to determine which symptoms adversely affect others. Additionally, although we collected all possible symptoms of SARS-CoV-2 infection, reporting bias still exists. Patients could refrain from reporting symptoms that could embarrass them or be stigmatizing. Therefore, qualitative research can be conducted, for example, ethnographic methods emphasizing comprehension of the patient's experience will add an important perspective to further analysis. Lastly, our study was conducted online by recruiting COVID-19 patients on the ward who met the eligibility criteria. Despite providing clear instructions for questionnaire completion, there may still exist information bias. Therefore, in the future, we should try to collect information face-to-face whenever possible. Conclusions This study has contributed novel insights into symptom clusters among patients with SARS-CoV-2 infection by identifying eight distinct symptom clusters. Chest tightness was the most central symptom. Gender, age, education, income, and some chronic diseases were associated with specific symptom clusters. These findings underscore the importance of evaluating symptom clusters and their interconnections to enhance symptom management in individuals affected by SARS-CoV-2 infection. In addition, the development of symptoms over time is unclear, and further research is urgently needed to explore. CRediT authorship contribution statement Hongmin Ye: Writing – original draft, Software, Investigation, Formal analysis, Data curation. Xiuni Gan: Supervision, Project administration, Methodology, Funding acquisition, Conceptualization. Wen Zhou: Writing – review & editing, Methodology. Yan Gao: Writing – review & editing. Zhechuan Mei: Conceptualization. Qiulan Zheng: Methodology, Conceptualization. Xiaoqing Luo: Investigation. Chunlan Yuan: Investigation. Yan Wu: Investigation. Ethical approval statement The work described has not been published previously except in the form of an abstract. The article is not under consideration for publication elsewhere. The article's publication is approved by all authors and the Ethics Committee of the Second Affiliated Hospital of Chongqing Medical University (No. 2023041). If accepted, the article will not be published elsewhere in the same form, in English, or any other language, including electronically without the written consent of the copyright-holder. Funding source This work was supported by the Second Affiliated Hospital of Chongqing Medical University [grant numbers 2023IITXG03]. Declaration of competing interest The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper. References