We conducted a retrospective cohort study at the Third People’s Hospital of Shenzhen, the designated hospital for HIV care in Shenzhen, China. The cohort comprises approximately 22,000 PWH who have received ART at this hospital since 2009. ART is provided free-of-charge in accordance with national HIV treatment guidelines. PWH attend routine follow-up visits approximately every 3 months for continuous ART prescriptions and health monitoring. Detailed baseline and follow-up assessments, including demographic characteristics, clinical diagnoses, and laboratory measurements, are collected as part of routine clinical care and recorded in the hospital’s electronic medical record system. Further details on cohort characteristics, including prior studies using this dataset, can be found in the Supplementary materials (Brief Introduction of the Cohort).

Participants were eligible if they had a confirmed HIV diagnosis, initiated ART between January 1, 2009, and June 30, 2016, and had follow-up data available through June 30, 2024. Individuals with missing ART initiation dates or fewer than three recorded diagnostic encounters were excluded to reduce potential bias.

All clinical outcomes were identified using ICD-10 diagnostic codes and clinician-documented diagnoses recorded in the hospital electronic medical records system at the Third People’s Hospital of Shenzhen, with confirmation based on relevant laboratory, imaging, and pathology findings in accordance with standardized diagnostic guidelines. A total of 26 conditions were selected for inclusion in the analysis. These conditions were chosen based on their prevalence in the general population and clinical relevance among PWH. The selected conditions included: (1) Hypertension, (2) Diabetes mellitus, (3) Dyslipidemia, (4) Cardiovascular disease (CVD), (5) Chronic obstructive pulmonary disease (COPD), emphysema, or chronic bronchitis, (6) Esophageal, gastric, or duodenal diseases, (7) Hepatitis B virus (HBV) infection, (8) Hepatitis C virus (HCV) infection, (9) Chronic kidney disease (CKD), (10) Non-AIDS-related cancers, (11) Osteoporosis, (12) Mental health disorders, (13) Sleep disorders, (14) Syphilis, (15) Human papillomavirus (HPV) infection, (16) Metabolic dysfunction–associated steatotic liver disease (MASLD), (17) AIDS-defining cancers, (18) HIV-associated encephalopathy, (19) Pneumocystis jirovecii pneumonia (PJP), (20) Nontuberculous mycobacterial infections, (21) Cytomegalovirus (CMV) infection, (22) Herpes simplex virus (HSV) infection, (23) Varicella-zoster virus (VZV) infection, (24) Toxoplasmosis, (25) Fungal infections, (26) Tuberculosis (TB). Full details on disease definitions and classification are available in the Supplementary materials.

ART initiation was defined as day zero, and a 90-day landmark period was applied to reduce misclassification of pre-existing comorbidities (12). Any predefined comorbid condition diagnosed on or before day 90 relative to ART initiation, including those documented prior to ART initiation, was treated as a baseline condition. The follow-up period was subsequently divided into three distinct phases: baseline (on or before day 90 after ART initiation), Wave1 (day 90 to 4 years), and Wave2 (4 to 8 years).

To characterize multimorbidity dynamics, we operationally classified the 26 predefined conditions into curable and incurable categories. This classification was informed by established disease-specific clinical guidelines and the typical clinical trajectory of each condition, as summarized in Supplementary Tables 1–3. Conditions were classified as curable if they are generally reversible with appropriate therapy and do not require lifelong disease monitoring once resolved (e.g., syphilis, tuberculosis, HCV infection after direct-acting antiviral treatment, acute viral infections, and other transient or episodic infectious conditions). In contrast, conditions were categorized as incurable if they typically reflect chronic or progressive organ dysfunction requiring sustained or lifelong clinical surveillance and/or pharmacological management, even when well controlled (e.g., hypertension, diabetes, dyslipidemia, CKD, chronic HBV infection, osteoporosis, and cancers). This pragmatic, guideline-informed classification reflects the expected long-term disease course rather than short-term clinical fluctuations and is aligned with contemporary clinical guidelines and expert consensus applied in routine HIV care.

Based on this distinction and the number of coexisting conditions, the 26 predefined conditions were pragmatically grouped into six mutually exclusive comorbidity states for the evolution plots: incurable multimorbidity (≥2 incurable conditions), curable multimorbidity (≥2 curable conditions), mixed curable–incurable multimorbidity (coexistence of at least one incurable and one curable condition), single incurable disease, single curable disease, and no disease. Curable conditions were modelled as resolving events and were therefore allowed to appear and disappear across periods, whereas incurable conditions were treated as long-term diagnoses. For each follow-up period, multimorbidity states were defined based on all curable and incurable conditions recorded in that period.

In addition to these six curability-based states, we further defined two etiologically distinct multimorbidity patterns. Metabolic multimorbidity was defined as the presence of at least two metabolic conditions, including hypertension, diabetes mellitus, dyslipidaemia, MASLD, or osteoporosis. Mixed infectious–non-infectious multimorbidity was defined as the coexistence of at least one infectious and one non-infectious condition during the follow-up period. Infectious conditions included chronic viral hepatitis (HBV/HCV), TB, STIs (syphilis/HPV), and major opportunistic infections. A detailed classification of all 26 conditions is provided in Supplementary Table 3.

Loss to follow-up was defined as having no clinical encounter, laboratory test, or recorded death for more than 180 days before the last observed date in the cohort. For the Sankey diagrams, because each participant could occupy only one state per period, we applied a hierarchical assignment rule: within a given period, individuals whose follow-up ended more than 180 days before the end of that period and who did not develop any new diagnoses among the 26 predefined conditions during that period were allocated to a combined “dropout” state, whereas those who developed at least one new diagnosis were classified into the corresponding module.

We extracted key covariates from the hospital records and laboratory results. Demographic variables included age at diagnosis, sex, and marital status. Age at diagnosis was categorized as ≤ 25, 26–35, 36–45, and ≥46 years, approximating quartiles of the distribution and reflecting early, middle, and later adulthood among PWH in this cohort. Anthropometric and laboratory parameters included body mass index (BMI), high-density lipoprotein cholesterol (HDL-C), low-density lipoprotein cholesterol (LDL-C), fasting glucose (FPG), platelet count (PLT), alanine aminotransferase (ALT), and estimated glomerular filtration rate (eGFR), all obtained through standardized clinical protocols. HIV-related factors comprised baseline CD4 count, baseline ART regimen, route of HIV transmission, diagnosis-to-ART initiation interval, and the presence of opportunistic infections at baseline. Behavioral characteristics included smoking and alcohol use.

All analyses were performed using R version 4.4.1 and RStudio version 2024.12.0. Descriptive statistics were used to summarize participant characteristics and the distribution of multimorbidity states across waves.

Sankey diagrams were generated using the ggsankey, ggplot2, and dplyr packages in R to visualize transitions in comorbidity states over time. Retrospective Sankey plots were also constructed to trace prior disease profiles among individuals who developed incurable or mixed curable–incurable multimorbidity patterns.

Stacked bar charts were drawn to compare the distribution of comorbidity states by sex and age groups across waves. For stratified descriptive analyses, age at diagnosis was additionally dichotomised at the cohort median age of 32 years. Differences between subgroups were assessed using Chi-square or Fisher’s exact tests.

Regression analyses focused on two common and clinically important patterns: metabolic multimorbidity and mixed infectious–non-infectious multimorbidity; participants with the corresponding pattern at baseline were excluded. Age was defined as age at HIV diagnosis and treated as a fixed baseline covariate in all regression models. Metabolic multimorbidity was modelled using Cox proportional hazards regression for time to first onset, whereas the mixed pattern was analysed using multivariable logistic regression for ever-occurrence over 8 years, because infectious conditions may resolve and recur, precluding a precise onset time for a sustained mixed state. For the Cox regression models, the proportional hazards assumption was assessed using Schoenfeld residuals. Adjusted hazard ratios (HRs) and odds ratios (ORs) with 95% confidence intervals (CIs) were reported, and p-values < 0.05 were considered statistically significant.

As sensitivity analyses, we repeated the Cox regression for metabolic-related multimorbidity using an extended 6-month ART initiation window and conducted wave-specific analyses restricted to Wave1 (day 90 to 4 years) using the same covariate set. Results are presented in Supplementary Figures 1–3.

A total of 5,950 participants were included in the final analysis. At enrollment, 25.9% of participants had multimorbidity (Table 1). Compared with those without multimorbidity, participants with multimorbidity were older (median age: 35.0 vs. 31.0 years; interquartile range [IQR]: 29–43 vs. 27–38 years; p < 0.05), and more likely to be male (90.5% vs. 85.6%; p < 0.05). Significant differences in BMI were observed between the two groups (p < 0.05), with a higher proportion of underweight individuals among those with multimorbidity. Marital status also showed a significant difference between groups (p < 0.05), with a higher proportion of divorced, separated, or widowed individuals in the multimorbidity group (12.3% vs. 8.0%). Unhealthy lifestyle behaviors, including smoking and alcohol use, were more common in the multimorbidity group (all p < 0.05). The prevalence of hypertension, diabetes, and hypercholesterolemia was significantly higher among participants with multimorbidity (all p < 0.05). Additionally, these participants had lower baseline CD4 cell counts, higher HIV RNA levels, and a greater prevalence of HBV or HCV co-infection (all p < 0.05).

Over the 8-year follow-up period, the distribution of comorbidity states among PWH exhibited notable temporal shifts. At baseline, most participants were classified as having either no comorbidity (24.61%) or a single incurable condition (46.03%) among the 5,950 individuals included. The overall prevalence of multimorbidity, including incurable, curable, and mixed patterns, increased from 25.87% at baseline to 42.49% by Wave2 (Figure 1, Supplementary Table 4). The most substantial increase was observed in the incurable multimorbidity pattern, which rose from 7.66% at baseline to 18.82% in Wave1, and further to 25.50% by Wave2, representing a more than threefold increase. The mixed curable–incurable multimorbidity pattern remained relatively stable over time, consistently affecting approximately 1,000 participants (16–18% of the cohort). In contrast, the curable multimorbidity and single curable disease patterns steadily declined, with their combined prevalence decreasing from 4.27% to just 0.3%.

Among the 1,464 participants who were free of all 26 predefined conditions at baseline, 528 (36.07%) developed multimorbidity during the follow-up period. Of these, 277 transitioned during Wave1, and an additional 251 during Wave2.

In the sex-stratified analysis, females consistently exhibited a lower prevalence of multimorbidity across all waves compared to males. The proportion of participants without any comorbidity declined markedly in both sexes, with a sharper decrease observed in females (from 34.27 to 4.36%) than in males (from 23.15 to 3.06%) (p < 0.05 for both). In contrast, males showed a greater increase in incurable multimorbidity over time, rising from 7.95 to 25.91%, compared with an increase from 5.78 to 22.72% in females (p < 0.05 for both). Meanwhile, the mixed curable–incurable multimorbidity pattern remained relatively stable in males (18.14 to 17.95%) but declined modestly in females (12.71 to 10.14%) (Figure 2A, Supplementary Tables 5–7).

Age-stratified analysis showed that individuals aged >32 years consistently had a higher burden of multimorbidity across all waves (Figure 2B, Supplementary Tables 8–10). At baseline, 31.36% of older participants had multimorbidity compared to 20.70% in those aged ≤32 years (p < 0.05). As follow-up progressed, multimorbidity increased significantly in both groups (p < 0.05), mainly driven by the rise in incurable multimorbidity. By Wave2, the prevalence of incurable multimorbidity reached 32.56% in the older group and 19.90% in the younger group (p < 0.05). While the older group initially had a higher prevalence of mixed multimorbidity (20.29% vs. 14.74%; p < 0.05), this difference diminished over time, and by Wave2 the proportions had converged (16.20% vs. 17.61%; p = 0.16). The proportion of participants with no comorbidity declined in both age groups, with a more substantial reduction among younger participants (from 28.66 to 4.37%; p < 0.05).

In the top three incurable multimorbidity clusters, most individuals maintained the same disease combinations through follow-up, reflecting considerable structural stability. Specifically, individuals classified under the “diabetes–dyslipidemia,” “dyslipidemia–HBV,” or “dyslipidemia–MASLD” at baseline, including those with coexisting curable diseases, largely remained within the same top three incurable disease clusters in Wave1. Within these groups, 18.00, 37.21, and 46.00% of individuals, respectively, had coexisting curable diseases, primarily infections. Among those with curable diseases at baseline, 22.22, 31.25, and 53.97% had two or more curable conditions, respectively. By Wave2, among individuals who remained in the same incurable multimorbidity clusters, the proportions with coexisting curable diseases had declined to 15.07, 20.10, and 27.86%, respectively (Figure 3).

In the mixed curable–incurable pattern, more variation was observed. “Dyslipidemia–syphilis” remained the most common across waves (18.71, 16.51, and 19.17%), while “dyslipidemia–HPV” declined from 11.13% in Wave1 to 4.47% in Wave2. In contrast, “dyslipidemia–sleep disorders” increased to 12.02% in Wave2, becoming the second most frequent combination (Table 2, Figure 4).

In the metabolic multimorbidity model (Figure 5), older age was a strong predictor: individuals aged ≥46 had a 3.42-fold increased risk compared to those ≤25 (HR = 3.42, 95% CI: 2.60–4.49, p < 0.05). Overweight or obesity (BMI ≥ 24) was associated with higher risk (HR = 1.70, 95% CI: 1.46–1.96, p < 0.05), as was male sex (HR = 1.44, 95% CI: 1.16–1.80, p < 0.05). Elevated HDL-C (≥1.0 mmol/L) and FPG (≥6.1 mmol/L) were also significant (HR = 1.21, 95% CI: 1.03–1.41, p < 0.05; HR = 2.70, 95% CI: 2.15–3.40, p < 0.05). ALT ≥1 × ULN and ≥2 × ULN increased the risk by 50 and 40%, respectively (both p < 0.05), while reduced eGFR (<60 mL/min/1.73 m²) was associated with higher risk (HR = 2.24, 95% CI: 1.25–4.05, p < 0.05). Compared to those on non-nucleoside reverse transcriptase inhibitor (NNRTI)-based regimens, PWH on non-NNRTI regimens had a higher risk (HR = 1.44, 95% CI: 1.14–1.83, p < 0.05). Those with IDU also showed increased risk (HR = 1.72, 95% CI: 1.10–2.69, p < 0.05) compared to MSM. Baseline CD4 count, smoking, diagnosis-to-ART interval, and alcohol use were not significantly associated. No meaningful violations of the proportional hazards assumption were detected (Supplementary Table 11). Results were robust in sensitivity analyses, with consistent findings observed using a 6-month ART initiation window and in analyses restricted to Wave1 (Supplementary Figures 1, 2).

In the mixed infectious–non-infectious model (Figure 6), male sex (OR = 1.92, 95% CI: 1.50–2.47, p < 0.05) and IDU (OR = 1.97, 95% CI: 1.10–3.49, p < 0.05) were strong risk factors. In contrast, BMI ≥ 24 was protective (OR = 0.72, 95% CI: 0.58–0.90, p < 0.05), as was being married (OR = 0.76, 95% CI: 0.63–0.90, p < 0.05). CD4 ≤ 100 cells/μL (OR = 1.31, 95% CI: 1.07–1.61, p < 0.05), PLT < 100 × 10⁹/L (OR = 1.62, 95% CI: 1.14–2.30, p < 0.05), and a history of opportunistic infections (OR = 2.16, 95% CI: 1.73–2.69, p < 0.05) were all independently associated. Alcohol consumption was modestly associated (OR = 1.18, p < 0.05), while age, HDL-C, diagnosis-to-ART interval, baseline ART regimen, and eGFR were not significant in this model. Sensitivity analyses restricted to Wave1 yielded consistent results (Supplementary Figure 3).

Several limitations should be acknowledged. First, the study was conducted at a single HIV treatment center, which may limit the generalizability of findings to other regions or healthcare settings. Second, despite efforts to standardize diagnoses using laboratory data, underdiagnosis or underdocumentation of conditions such as mental health disorders and behavioral diseases may persist. Third, age was defined at HIV diagnosis and modelled as a fixed baseline covariate; therefore, age-stratified results reflect baseline differences associated with baseline age rather than changes in attained age during follow-up. Finally, the distinction of curable and incurable was a pragmatic operationalisation and could not account for within-disease variation in severity, control, or treatment response; these dimensions may differentially affect prognosis and should be explored in future work. Future multicenter studies with prospective designs and richer behavioral datasets are needed to validate and expand upon these findings. Because the evolution observed in this study is descriptive, future work will incorporate trajectory-modelling approaches to more rigorously characterise longitudinal patterns.