Identification of subtypes of HIV/AIDS-related symptoms in China using latent profile analysis and symptom networks

study design

This study used data from the HIV-related Symptoms Monitoring Survey (HSMS). The HSMS is a cross-sectional dataset collected by our team covering PWH from 11 cities in the East (Shanghai City), Central (Changsha in Hunan Province) and Southwest (Ruili, Tengchong, Kunming, Longxing, Changning, Baoshan, Linchang, and Longchuan in Yunnan Province and Nanning in Guangxi Province) China from 2017 to 2019. More information on HSMS can be found elsewhere5. Ethical approval was obtained from the Institutional Review Board, School of Nursing, Fudan University (IRB#TYSA2016-3-1). This research was conducted in accordance with the Declaration of Helsinki. Written informed consent was obtained from the participants prior to data collection.


Participants were included in the study if they were (1) HIV positive and (2) ≥ 18 years of age and older. PWH who did not complete a self-reported symptom checklist or (3) who were diagnosed with severe neurocognitive disorders were excluded from the study. From 2017 to 2019, we recruited 3017 participants through a simple sample from 11 hospitals in 11 cities, as specified in the study design, that are responsible for HIV/AIDS-related treatment and care in those areas. Ninety participants were excluded due to missing data. As a result, a total of 2927 suitable PWH were included in the analysis.


Sociodemographic and clinical data

Demographic, socioeconomic, and clinical data were collected from a self-completed questionnaire. Demographic variables included age (continuous), gender (male and female), and ethnicity (Han and minority). Socioeconomic variables included marital status (married, single, and otherwise), employment status (employed and otherwise), educational attainment (middle school or lower, high school or equivalent, bachelor’s degree or equivalent, and master’s degree or higher), and primary caregiver (myself, family members and others). Clinical variables included years since HIV diagnosis (years, continuous), ART use (yes or no), duration of ART use (years, continuous), last CD4+ T-cell count (continuous), and comorbidities (yes or no) . All sociodemographic and clinical data were confirmed by medical records.

Self-Reported Symptoms

The HIV/AIDS Self-reported Symptom Checklist (HSSC) was used to assess the severity of 27 common HIV/AIDS-related symptoms16. The 27 symptoms included in the HSSC were categorized into 5 symptom clusters (wasting syndrome, dizziness/headache, cognitive function, skin-muscle-joint disorder and mood disorder) and 7 individual symptoms (fatigue, sleep disturbances, cough, hair loss, blurred vision, low sex drive and lipodystrophy). Responses ranged from not at all (0) to difficult (3). The total score was determined by summing the scores of these 27 items (ranging from 0 to 81). The total scores for the wasting syndrome cluster, dizziness/headache cluster, cognitive functioning cluster, skin, muscle and joint disorders cluster, and mood disorder cluster were 15, 6, 15, 12, and 12, respectively The HSSC had good expert validity (content validity index = 0.918) and internal consistency (Cronbach’s α = 0.961).

Basic activities of daily living

The ability to perform basic activities of daily living was assessed using the Barthel Index (BI).30. The BI is a widely used measure to assess the ability to perform basic activities of daily living such as bathing, dressing and personal hygiene. The BI contains 10 items and the total score ranges from 0 to 100. A higher score indicates a higher ability to perform basic activities of daily living. The measure showed good internal consistency in our sample (Cronbach’s α = 0.941).

drug liability

Patient-reported medication intake was measured using one question, “In the last 7 days, how often have you forgotten to take your medication?” Responses ranged from never (1) to always (5).

Discrimination Perceived by PWH

Perceived discrimination by PWH was assessed using the Expanded Everyday Discrimination Scale31. This measure consists of 10 items and describes different scenarios in everyday life in which PWH can perceive discrimination. The overall score ranges from 10 to 40. A higher score indicates a lower level of perceived discrimination. The measure showed good internal consistency in our sample (Cronbach’s α = 0.913).

Self-reported health status, quality of life, and self-management skills

Self-reported health status, quality of life, and self-management skills were measured using the questions “How do you rate your general health?”, “How do you rate your quality of life?”, and “How do you rate your quality of life? Your ability to self-manage for HIV/AIDS?” The Responses for these variables ranged from very good (1) to very bad (5).

Statistical analysis

Mplus 8.1 was used to perform LPA to identify person-centered subtypes of HIV/AIDS-related symptoms. LPA was performed based on the severity of symptom clusters (wasting syndrome, dizziness/headaches, cognitive function, skin-muscle-joint disorder, and mood disorder) and individual symptoms (fatigue, trouble sleeping, cough, hair loss, blurred vision, low sex drive and lipodystrophy). We calculated the overall severity for each cluster (dizziness/headache, cognitive dysfunction, skin-muscle-joint disorder, wasting syndrome, mood disorder). For 7 symptoms that could not be categorized into clusters, we used Likert items to rate the severity of each symptom. We used LPA because the severity scores were continuous response variables, while latent class analysis (LCA) was used for categorical variables. The number of classes was determined by comparing the Bayesian Information Criterion (BIC), Akaike Information Criterion (AIC), Sample Size-Adjusted BIC (ABIC), Lo Mendell Rubin Likelihood Ratio Test (LMR), Bootstrapped Likelihood ratio tests ( BLRT) and entropy of each model. Smaller AIC, BIC, and ABIC values ​​indicate better model fit. P-values ​​greater than 0.05 for LMR and BLRT indicated that the k-1 model was rejected and the k-model was supported. The theoretical interpretability was also taken into account. We plotted the conditional probabilities of symptom severity for each of the classes. The code for performing LPA in Mplus is shown in Supplementary File 1.

After identifying the classes, we analyzed differences in sociodemographic and clinical data, basic activities of daily living, medication adherence, perceived discrimination, self-reported health status, quality of life, and self-management capacity using the chi-square test and one-way analysis of variance (ANOVA) with post hoc tests (Fisher’s least significant difference). In the case of heterogeneous variances, Tamhane’s T2 multiple comparison test was used. Furthermore, a multinomial logistic regression analysis of the five profiles was performed. The Nagelkerke R2 and χ2 were used as indicators of model fitness. We considered a two-tailed one P< 0.05 to indicate statistical significance in all analyses.

R 4.0.2 and the Qgraph module were used to perform the network analysis. We used Spearman correlations to assess the relationships (edges) between pairs of symptoms (nodes) in the full sample and the subgroups. The Fruchterman-Reingold (FR) algorithm and spring layout were used to generate symptom networks26. In the FR algorithm, the node (symptom) with the strongest centrality was placed in the middle of the network, and nodes with similar properties were placed closer. Covariates were selected from bivariate analysis results and included age (continuous), sex (male = 1, female = 2), race (Han = 1, minority = 2), educational level (middle school or below = 1, high school or above). = 2), employment (employed = 1, else = 2), marital status (married = 1, else = 2), primary caregiver (myself = 1, else = 2), have ART (yes = 1, no = 2) , years with ART (continuous), comorbidities (yes=1, no=2), lgCD4 (continuous), medication adherence (continuous), self-management ability (continuous), and perceived discrimination (continuous).

We used three centrality indices (strength, intermediate, and proximity) to identify the most central symptoms32.33. Strength is a measure of network connectivity. The greater the severity, the greater the likelihood that the symptom will appear along with other symptoms. Betweenness quantifies the frequency with which a node acts as a bridge on the shortest path between two nodes. A node with higher Betweenness Centrality has more influence on the network. Proximity represents the average distance (inverse distance) from a symptom to all other nodes. The larger the value of proximity, the shorter the path. For a concurrent network, strength is used as the main indicator among the three indices. We used the absolute value of all Spearman’s coefficients (∑s) to indicate the density of symptom network connections.

Ethical Approval

Ethical approval was obtained from the Institutional Review Board, School of Nursing, Fudan University (IRB#TYSA2016-3-1).

Add a Comment

Your email address will not be published.