What shapes the quality of life among older persons living with HIV? A multidimensional assessment from The Aids Support Organisation clinics in Uganda Authors: Christine Atuhairwe, Hillary Muhanguzi, Titus Ochieng Published: 2026 ABSTRACT IntroductionThe Quality of Life (QoL) Composite Index provides a multidimensional, person-centred measure integrating functioning, symptom burden, general health perception, psychological, social, and living environmental, offering a critical metric for assessing successful ageing with HIV. However, evidence on the determinants of this composite measure among older persons living with HIV (OPLHIV) in Uganda remains limited.ObjectiveTo estimate the Quality-of-Life Composite Index and identify its key determinants among OPLHIV receiving care from The AIDS Support Organisation (TASO) in Uganda.MethodsA cross-sectional study was conducted among OPLHIV aged ≥ 50 years attending TASO clinics across Uganda. The primary outcome was the QoL Composite Index (range: 0–1), derived from a validated multidimensional instrument capturing functioning, symptom burden, general health perception, and living environment. Data were collected through face-to-face interviews. Multivariable associations were assessed using a Beta regression with log link. Conversely, a random forest regression as a machine learning model was run to reveal the important variables shaping the QoL as assessed through the Index.ResultsAmong 439 participants, the median QoL Composite Index was 0.51 (IQR: 0.41–0.61). After model regularization, two socioeconomic factors were independently associated with a higher QoL Composite Index: attainment of tertiary/university education (β = 0.226, 95% CI 0.105–0.346, p < 0.001) and being employed (β = 0.078, 95% CI 0.018–0.138, p = 0.011). The presence of a non-communicable disease (co-morbidity) was strongly associated with a lower QoL Composite Index. HIV-specific clinical indicators—including age, duration of HIV infection, viral load, and ART adherence—were not independently associated with the composite measure.ConclusionThe QoL Composite Index effectively elucidates the multidimensional phenomenon of well-being among older Ugandans living with HIV and demonstrates that this is shaped predominantly by socioeconomic conditions and multi-morbidity rather than HIV-specific clinical markers. Improving the quality of life in later life with HIV therefore requires care models that extend beyond viral suppression to address socioeconomic security and integrated geriatric health needs. Introduction The Quality of Life (QoL) Composite Index for people living with HIV is a paradigm shift in the outcome measurement, moving beyond isolated biomedical indicators such as viral load suppression, WHO Clinical stage, antiretroviral therapy (ART) adherence, and CD4 cell counts towards a multidimensional assessment of holistic well-being [1, 2]. This index integrates functioning, symptoms, general physical health, lifestyle, and living environment to quantify co-occurring poor QoL aspects, offering a person-centred understanding of what it means to live well with a chronic condition such as HIV [3, 4]. Unlike traditional tools like MOS-HIV or WHOQOL-BREF that rely on single summary metrics, this composite measure incorporates structural and psychosocial elements to identify at-risk groups, particularly older persons living with HIV (OPLHIV) in Ugandan context. Evidence from studies conducted by The AIDS Support Organisation (TASO) demonstrate the value of this index in capturing accelerated ageing effects, comorbidities, social isolation, and post-ART improvements in QoL, with the prevalence of poor QoL declining from 52% before ART initiation to 35% after treatment initiation [5–7]. National HIV surveillance data from the United States of America National HIV/AIDS Strategy data (2018–2021) derived from Centers for Disease Control (CDC) Medical monitoring project surveys, PLHIV had at least one poor QoL indicator [8]. A dose–response relationship was observed between increasing numbers of poor QoL indicators and poorer HIV control, a pattern mirrored globally, advocating routine clinical adoption of composite QoL measures [2, 5, 7, 9]. Within HIV epidemiology among older Ugandan populations, the QoL composite index strengthens questionnaire design and survival analyses by integrating psychosocial metrics and linking elevated scores to risks such as disengagement from care or hospitalization, thereby prioritizing holistic interventions [9–11]. It is an approach that saliently captures the interaction between ageing-related challenges and long-term HIV management, supporting tailored interventions [4]. By associating higher composite scores with adverse outcomes like care non-retention or hospitalizations, the index advocates prioritizing psychosocial support alongside biomedical treatment [9–11]. Advances in ART have transformed HIV into a manageable chronic condition, enabling many individuals in sub-Saharan Africa to survive into older age [11]. In Uganda, which bears a substantial HIV burden, this demographic shift challenges health systems historically oriented towards younger populations [11–14]. Although viral suppression remains essential, it offers only a partial account of successful ageing with HIV [12–14]. Older adults living with HIV frequently experience multi-morbidity, functional decline, social isolation, and economic insecurity—factors that are not reflected in viral load measurements but are central to overall quality of life [12–15]. Existing research in low-resource settings largely prioritizes clinical outcomes and survival, with limited attention to the determinants of QoL composite index among OPLHIV [2, 12, 15]. Evidence from broader HIV populations suggests that education, employment, and social circumstances influence quality of life, yet their role in later life-amid declining socioeconomic stability and social support remain underexplored [2, 12–15]. The influence of HIV-specific markers on QoL within community-based care models such as those provided by The AIDS Support Organisation (TASO) warrant further scrutiny [6, 16, 17]. This study therefore constructed and examined determinants of the Quality-of-Life Composite Index among OPLHIV attending TASO clinics across Uganda. By analyzing clinical, demographic, and socioeconomic predictors, it seeks to identify the principal drivers of holistic well-being in this growing population and informs the evolution of HIV care towards a more integrated, person-centred model that supports not only longevity, but also dignity and fulfilment in later life. Methods Study design and setting A cross-sectional study was conducted in 2024 at eleven stand-alone TASO clinics across Uganda (Entebbe, Gulu, Jinja, Masaka, Masindi, Mbale, Mbarara, Mulago, Rukungiri, Soroti, Tororo). TASO is Uganda's oldest and largest indigenous HIV care and support organisation, providing a representative network of service delivery points in both urban and rural settings across the country. Study population and sampling The study population comprised OPLHIV aged 50 years and above, who were registered clients of TASO, had been on ART for at least six months, and were mentally competent to provide informed consent. Individuals with severe cognitive impairment or acute illness at the time of data collection were excluded. A multi-stage sampling technique was employed. First, all eleven TASO clinics were included. Second, a proportionate-to-size random sample of eligible clients was drawn from each clinic’s register. The minimum sample size of 384 was calculated using Cochran's formula, assuming a 50% prevalence of suboptimal QoL, a 5% margin of error, and 95% confidence level. A final sample of 439 participants was achieved. Data collection and variables Data were collected through face-to-face interviews using structured questionnaires and abstraction from participants' medical records. The primary outcome variable was the Quality of Life (QoL) Composite Index. This index was derived from an aggregation of variables grouped under three dimensions: Functioning, Symptoms and General Health [2, 5]. Domain scores were standardized and aggregated to form a single composite index ranging from 0 (worst imaginable QoL) to 1 (best imaginable QoL). The conceptual framework for the index is shown in Fig. 1.Figure 1: Framework for the Construction of the Quality of Life (QoL) Composite Index Figure 1: Framework for the Construction of the Quality of Life (QoL) Composite Index A total of twenty-three variables were identified and assigned to the 10-dimension pillars shown in the Fig. 1. To ensure a positive relationship with the index, some of the variables were inverted so that a higher value indicates a better outcome, as shown in Table 3. The variable raw score values were transformed in accordance with the HIV-MOS and Scoring instructions, using the formula one.1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$Transformed \,score=\frac{raw \,score-possible \,minimum}{variable \,raw \,score \,range}$$\end{document} To obtain the dimension scores, the transformed variable scores were equally weighted and additively aggregated at the first stage. At the second stage, the dimension scores were as well equally weighted and aggregated arithmetically to obtain the QoL composite index scores. Considering that weights imply value judgements [5, 9, 10], equal weighting across dimensions was adopted in line with recent composite indicator methodological guidance and applied statistical modelling [9, 17]. To assess the differentials, independent variables (mostly categorical) were grouped into:Clinical Factors: duration of HIV infection, current viral load (copies/mL), missed drug refills, NCD, history of opportunistic infection, and self-reported ART adherence (Poor: < 85%, Fair: 85–95%, Good: > 95%).Demographic & Social Factors: Gender, age, education level (None, Primary, Secondary, Tertiary/University), religious affiliation, marital status, and number of sexual partners in the last year.Socioeconomic Factors: Employment status (Yes/No) and livelihood activities. Clinical Factors: duration of HIV infection, current viral load (copies/mL), missed drug refills, NCD, history of opportunistic infection, and self-reported ART adherence (Poor: < 85%, Fair: 85–95%, Good: > 95%). Clinical Factors: duration of HIV infection, current viral load (copies/mL), missed drug refills, NCD, history of opportunistic infection, and self-reported ART adherence (Poor: < 85%, Fair: 85–95%, Good: > 95%). Demographic & Social Factors: Gender, age, education level (None, Primary, Secondary, Tertiary/University), religious affiliation, marital status, and number of sexual partners in the last year. Demographic & Social Factors: Gender, age, education level (None, Primary, Secondary, Tertiary/University), religious affiliation, marital status, and number of sexual partners in the last year. Socioeconomic Factors: Employment status (Yes/No) and livelihood activities. Socioeconomic Factors: Employment status (Yes/No) and livelihood activities. Data analysis Data were analysed using R-software. Descriptive statistics (means, medians, frequencies) were calculated for selected variables. The values of the QoL Composite Index are continuous and range between (0–1) thus beta regression was the appropriate model as it generated predicted values that fitted within that range [9, 10]. This model estimated the adjusted association (coefficient, β) between each predictor and the QoL Index, controlling for all other variables in the model. Statistical significance was set at p < 0.05. Model diagnostics for the regularized model, were performed to ascertain suitability. For parsimonious representation, a regularized beta regression model was fitted. Diagnostic assessments confirmed good model fit as shown in the Supplementary Material. Considering that recent advancements in statistical analysis entail incorporation of machine learning and artificial intelligence-regression algorithms to offer alternative regression modelling [9], a random forest (RF) regression was run to reveal feature importance that can be readily communicated to stakeholders. A correlation analysis of the fitted Random Forest values and beta regression was performed to indicate the convergence of the approaches. Ethical considerations The study was approved by The AIDS Support Organization (TASO) Research Ethics Committee (TASOREC/108/2022-UG-REC-009), the AIDS support organization administration clearance (TASO REC/ADMC010/2022-UGREC-009) and the Uganda National Council for Science and Technology (UNCST: SS1396ES). All guidelines and regulations as stipulated in the TASO handbook were followed. The study was executed in alignment with the principles of the Declaration of Helsinki, safeguarding the rights, safety, and welfare of all participants. We informed participants about the study's purpose, procedures, potential risks and benefits, and the option to withdraw at any time. Prior to data collection, each participant provided written or thumb-printed informed consent, affirming that they understood the study's purpose and agreed to participate voluntarily. During data collection, participants were interviewed in a quiet place known as ''counselling huts,'' where each participant was interviewed separately from others, enhancing their privacy. In addition, the data was anonymised, and participants were issued unique codes. Results Participant characteristics and overall quality of life A total of 439 older persons living with HIV (OPLHIV) aged ≥ 50 years were included in the analysis. Most participants were female (62%). Approximately forty-one% were aged 50–55 years, while 32% were 60 years or older, with a mean age of 58 years. Regarding marital status, 44% were widowed, and 33% reported having one sexual partner. More than half of the participants (52%) had completed primary education, and 68% were unemployed. The most common livelihood activity was crop cultivation (59%) (Table 1).Table 1Participants socio-demographic characteristicsNumber (n = 439)Percent (%)Socio-demographic characteristicsGender Male16738.0 Female27262.0Age (Mean 58.3 years ± 6.8 SD) 50–55 years18241.2 56–60 years11726.7 > 60 Years14132.1Level of education None9922.6 Primary22751.7 Secondary8218.7 Tertiary317.1Marital status Currently married/cohabiting16838.3 Separated/divorced7717.5 Widow/widower19444.2Number of current sexual partners None25257.4 One14533.0 Two or more429.6Currently employed Yes13931.7 No30068.3Livelihood activities Crop cultivation25858.8 Business/trade8018.2 Formal job/salaried419.3 Retired61.4 Others5412.3Clinical characteristicsDuration on ART < 10 Years12127.6 10–15 Years16437.4 > 15 Years15435.1Place of ART initiation Clinic initiated40291.6 Transfer-in378.4WHO HIV clinical stage Stage I14533.0 Stage II25658.3 Stage III and IV286.4Adherence status Poor < 85%92.1 Fair 85–95%317.1 Good > 95%39990.9History of Opportunistic Infections (genital ulcer, oral candida, herpes zoster, tuberculosis) Yes8920.3 No35079.7Tuberculosis status Coinfection317.1 No TB symptoms92.9Has NCD (diabetes, hypertension) Yes10223.2 No33776.8Quality of life Good18642.4 Poor25357.6Total439100Table adopted from Atuhairwe et al. 2025 Doi: https://doi.org/10.37432/jieph-d-25-00126 Table 1Participants socio-demographic characteristicsNumber (n = 439)Percent (%)Socio-demographic characteristicsGender Male16738.0 Female27262.0Age (Mean 58.3 years ± 6.8 SD) 50–55 years18241.2 56–60 years11726.7 > 60 Years14132.1Level of education None9922.6 Primary22751.7 Secondary8218.7 Tertiary317.1Marital status Currently married/cohabiting16838.3 Separated/divorced7717.5 Widow/widower19444.2Number of current sexual partners None25257.4 One14533.0 Two or more429.6Currently employed Yes13931.7 No30068.3Livelihood activities Crop cultivation25858.8 Business/trade8018.2 Formal job/salaried419.3 Retired61.4 Others5412.3Clinical characteristicsDuration on ART < 10 Years12127.6 10–15 Years16437.4 > 15 Years15435.1Place of ART initiation Clinic initiated40291.6 Transfer-in378.4WHO HIV clinical stage Stage I14533.0 Stage II25658.3 Stage III and IV286.4Adherence status Poor < 85%92.1 Fair 85–95%317.1 Good > 95%39990.9History of Opportunistic Infections (genital ulcer, oral candida, herpes zoster, tuberculosis) Yes8920.3 No35079.7Tuberculosis status Coinfection317.1 No TB symptoms92.9Has NCD (diabetes, hypertension) Yes10223.2 No33776.8Quality of life Good18642.4 Poor25357.6Total439100Table adopted from Atuhairwe et al. 2025 Doi: https://doi.org/10.37432/jieph-d-25-00126 Participants socio-demographic characteristics Table adopted from Atuhairwe et al. 2025 Doi: https://doi.org/10.37432/jieph-d-25-00126 Clinically, 35% of participants were not virally suppressed. Most had been on antiretroviral therapy (ART) for 10–15 years (40%) and were classified under WHO clinical stage II (62%). Self-reported adherence was high, with 89% reporting ≥ 95% adherence. Recent opportunistic infections were reported by 15% of the participants while 23% reported non-communicable diseases (NCDs). The Quality of Life (QoL) Composite Index ranged from 0 to 1, with higher values indicating better perceived well-being. The mean national QoL Index was 0.51 (SD ± 0.18), and the median was 0.51 (IQR: 0.41–0.61), indicating moderate level of quality of life among the cohort. Geographic variation in the quality-of-life composite index The QoL Composite Index varied substantially across TASO clinics, with median site-level scores ranging from 0.48 in Mbale to 0.56 in Soroti. Considerable within-site variability was observed, particularly in Mulago and Gulu, indicating heterogeneous client outcomes within the same facilities (Table 2).Table 2Heat map showing Site-level QoL Composite Index for the top 10 and bottom 10 clients Table 2Heat map showing Site-level QoL Composite Index for the top 10 and bottom 10 clients Heat map showing Site-level QoL Composite Index for the top 10 and bottom 10 clients At the national level, the functioning dimension had the lowest score (0.33), while symptom-related scores were highest (0.63). Participants with the highest QoL composite index scores (0.67–0.72), generally reported low symptom burden (mostly > 0.85), and moderate general health (0.65–0.81). In contrast, lower QoL composite clients score (0.28–0.36) were associated with higher symptom burden and poorer functioning, suggesting that symptom control plays a central role in determining overall QoL. The key patterns are that performance on symptoms and functions are low and inconsistent, while the performance on general health ranges were between 0.33 to 0.54. These results indicate that severe symptoms can substantially reduce QoL, even when functional ability is relatively maintained. Site-level extremes in quality-of-life composite index performance The highest observed QoL Composite Index score was 0.72, while the lowest was 0.28. Site-level patterns indicated that Soroti clients site frequently appeared among top performers, often with consistently high symptom scores, suggesting strong clinical management or adherence support. In contrast, Mulago site displayed wide variability, with clients appearing in both high-and low-performing groups, indicating heterogeneous outcomes within the same facility. Mbarara appears more frequently among lower scoring clients, suggesting potential area for program review or modification. Overall, symptom burden emerged as the dominant driver of the QoL composite index, followed by general health while functioning was the least influential. This is reflected by cases where clients with lower functioning but good symptoms control still achieved relatively high QoL scores, whereas those with poor symptom control scored low despite adequate functioning. These findings are consistent with evidence from chronic disease and HIV care literature, which shows that patients may tolerate functional limitations if symptoms are managed (Table 3).Table 3Some summary statistics for the QoL Composite index values by TASO Service SiteTASO service sitesMeanMedianRangeEntebbe0.520.540.26Gulu0.490.500.37Jinja0.510.500.27Masaka0.530.530.19Masindi0.500.500.17Mbale0.480.480.26Mbarara0.520.520.35Mulago0.500.490.38Rukungiri0.530.530.28Soroti0.550.560.32Tororo0.490.500.21Total0.510.510.44QoL Index: The Quality of Life Index is a composite score derived from the study-specific instrument, with values standardized on a 0 to 1 scale, where higher values indicate better perceived well-being. The range is the difference between the maximum and minimum observed scores at each site Table 3Some summary statistics for the QoL Composite index values by TASO Service SiteTASO service sitesMeanMedianRangeEntebbe0.520.540.26Gulu0.490.500.37Jinja0.510.500.27Masaka0.530.530.19Masindi0.500.500.17Mbale0.480.480.26Mbarara0.520.520.35Mulago0.500.490.38Rukungiri0.530.530.28Soroti0.550.560.32Tororo0.490.500.21Total0.510.510.44QoL Index: The Quality of Life Index is a composite score derived from the study-specific instrument, with values standardized on a 0 to 1 scale, where higher values indicate better perceived well-being. The range is the difference between the maximum and minimum observed scores at each site Some summary statistics for the QoL Composite index values by TASO Service Site QoL Index: The Quality of Life Index is a composite score derived from the study-specific instrument, with values standardized on a 0 to 1 scale, where higher values indicate better perceived well-being. The range is the difference between the maximum and minimum observed scores at each site The mean and median QoL scores across sites were consistent with the national averages. However, variation differed across sites, with Mulago showing the highest variability (0.38), likely reflecting its diverse urban catchment HIV population and Masindi exhibiting the lowest variation, suggesting more homogeneous outcomes. Variation in quality of life by socio-demographic characteristics Higher median QoL composite scores were observed among men, participants with higher educational attainment, and those who were employed. A clear positive gradient was observed with increasing education, while religious affiliation showed minimal differences (Table 4).Table 4Median QoL Composite Index score by socio-demographic characteristicsSocio-demographic characteristicsCategoryMedian indexGenderMale0.53Female0.50Education LevelNone0.50Primary school0.50Secondary school0.53Tertiary/University0.57Religion affiliationRoman Catholic0.52Protestant0.51Others0.50Marital statusMarried/cohabiting0.53Separated/divorce0.50Widow/widower/single0.51Employment statusYes0.53No0.50Total0.51Data presented as the median Composite Quality of Life (QoL) Index score. The QoL Index is scaled from 0 to 1, with higher scores indicating better perceived well-being Table 4Median QoL Composite Index score by socio-demographic characteristicsSocio-demographic characteristicsCategoryMedian indexGenderMale0.53Female0.50Education LevelNone0.50Primary school0.50Secondary school0.53Tertiary/University0.57Religion affiliationRoman Catholic0.52Protestant0.51Others0.50Marital statusMarried/cohabiting0.53Separated/divorce0.50Widow/widower/single0.51Employment statusYes0.53No0.50Total0.51Data presented as the median Composite Quality of Life (QoL) Index score. The QoL Index is scaled from 0 to 1, with higher scores indicating better perceived well-being Median QoL Composite Index score by socio-demographic characteristics Data presented as the median Composite Quality of Life (QoL) Index score. The QoL Index is scaled from 0 to 1, with higher scores indicating better perceived well-being Multivariable determinants of the quality-of-life composite index Results from the generalized linear model (gamma distribution with log link) are presented in Table 5. After adjustment for all covariates, education level and employment were the only two socioeconomic factors independently associated with a higher QoL composite Index.Table 5Multivariable generalized linear model resultsVariablesCoefficientStd. errzP > zIntercept0.04570.11320.40300.6866Gender Male0.05910.03201.8450.0650Age 50–55 years− 0.00790.0327− 0.2430.8076 56–60 years0.00180.03550.0510.9595Education level Primary− 0.00680.0346− 0.1970.8439 Secondary0.07280.04411.6510.0987 Tertiary0.21100.06233.3830.0007Marital status Widowed/single− 0.03650.0346− 1.0550.2914Employment status Yes0.08770.03702.3720.0176Livelihood activities Crop cultivation− 0.00710.0421− 0.1700.8651 Formal job/salarised0.00870.05580.1560.8760 Others0.01150.05120.2250.8223Place of ART initiation Transferred -in0.00700.04870.1440.8858WHO clinical Stage Stage I− 0.02570.0914− 0.2810.7785 Stage II− 0.01530.0899− 0.1710.8642 Stage III− 0.06760.1044− 0.6480.5172Duration on ART 11–15 years− 0.03580.0338− 1.0600.2891 > 15 years− 0.04540.0344− 1.3180.1873Viral load (copies/mL 0–200 copies /mL0.05770.04411.3080.1909History of opportunistic infections Yes− 0.05290.0357− 1.4810.1384Presence of a chronic disease Yes− 0.12510.0318− 3.9308.49e-05Missed ART drug refills Yes0.00650.04850.1350.8929 Sometimes− 0.07730.0746− 1.0370.2996Significance codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1 Table 5Multivariable generalized linear model resultsVariablesCoefficientStd. errzP > zIntercept0.04570.11320.40300.6866Gender Male0.05910.03201.8450.0650Age 50–55 years− 0.00790.0327− 0.2430.8076 56–60 years0.00180.03550.0510.9595Education level Primary− 0.00680.0346− 0.1970.8439 Secondary0.07280.04411.6510.0987 Tertiary0.21100.06233.3830.0007Marital status Widowed/single− 0.03650.0346− 1.0550.2914Employment status Yes0.08770.03702.3720.0176Livelihood activities Crop cultivation− 0.00710.0421− 0.1700.8651 Formal job/salarised0.00870.05580.1560.8760 Others0.01150.05120.2250.8223Place of ART initiation Transferred -in0.00700.04870.1440.8858WHO clinical Stage Stage I− 0.02570.0914− 0.2810.7785 Stage II− 0.01530.0899− 0.1710.8642 Stage III− 0.06760.1044− 0.6480.5172Duration on ART 11–15 years− 0.03580.0338− 1.0600.2891 > 15 years− 0.04540.0344− 1.3180.1873Viral load (copies/mL 0–200 copies /mL0.05770.04411.3080.1909History of opportunistic infections Yes− 0.05290.0357− 1.4810.1384Presence of a chronic disease Yes− 0.12510.0318− 3.9308.49e-05Missed ART drug refills Yes0.00650.04850.1350.8929 Sometimes− 0.07730.0746− 1.0370.2996Significance codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1 Multivariable generalized linear model results Significance codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1 Attainment of tertiary or university education was strongly associated with improved quality of life scores (β = 0.211, p < 0.001), while employment was also a significant positive predictor (β = 0.088, p = 0.018). Secondary education showed a marginal positive association, whereas primary education was not significantly associated with the QoL composite Index. Among clinical variables, the presence of a chronic non-communicable disease was the strongest negative predictor of QoL index (β = − 0.125, p < 0.001). Age group, duration on ART, viral load suppression, WHO clinical stage, ART adherence, and history of opportunistic infections were not independently associated with the Composite QoL Index in the adjusted model. Gender and marital status were also non-significant after adjustment. The model explained approximately 16.7% of the variation in the Composite QoL Index (pseudo R2 = 0.166). Parsimonious model of key predictors A regularized beta regression model was fitted to identify the most robust predictors of QoL composite index (Table 6). This parsimonious model retained six variables and explained a comparable proportion of variance (pseudo R2 = 0.149). Under beta regression, the R2 is rarely considered as a metric of regression models as it is sensitive to model complexity (overfitting and under-fitting) and ignores residual analyses.Table 6Regularised/Parsimonious beta regression modelVariablesCoefficientStd. errZP > zIntercept0.00480.03000.1600.8726Gender Male0.08350.02792.9900.0027**Education level Primary− 0.00620.0338− 0.1860.8526 Secondary0.07550.04261.7720.0763 Tertiary0.23180.06023.8470.0001**Employment status Yes0.08010.02962.6980.0069**History of opportunistic infections Yes− 0.06030.0330− 1.8280.0675Presence of a NCD Yes− 0.12510.0314− 3.9770.0001**Significance codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1 Table 6Regularised/Parsimonious beta regression modelVariablesCoefficientStd. errZP > zIntercept0.00480.03000.1600.8726Gender Male0.08350.02792.9900.0027**Education level Primary− 0.00620.0338− 0.1860.8526 Secondary0.07550.04261.7720.0763 Tertiary0.23180.06023.8470.0001**Employment status Yes0.08010.02962.6980.0069**History of opportunistic infections Yes− 0.06030.0330− 1.8280.0675Presence of a NCD Yes− 0.12510.0314− 3.9770.0001**Significance codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1 Regularised/Parsimonious beta regression model Significance codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1 Male gender, tertiary education, and employment remained significant positive predictors of the QoL Index, while the presence of a non-communicable disease remained the strongest negative predictor. A history of opportunistic infections showed a borderline negative association, while lower levels of education were not significantly associated with QoL index after regularization. In the beta regression model, coefficients were estimated on the logit link scale, where positive coefficients indicate higher expected QoL index values, and a negative coefficient indicated lower expected index value, holding other variables constant. The male gender (β = 0.083, p = 0.0028) was associated with significantly higher QoL composite index compared to females, while tertiary education (β = 0.231, p = 0.0026) was strongly associated with improved QoL index. Overall, the findings indicate that QoL composite index among OPLHIV improves with socio-economic advantage-particularly education and employment and declines in the presence comorbidity health conditions especially non-communicable diseases. Model diagnostics indicated good fit, low dispersion, and stable coefficient estimates. Model diagnostics Diagnostic evaluation confirmed the appropriateness and robustness of the beta regression model for analysing the bounded QoL composite index for OPLHIV in Uganda. The model's assumptions were evaluated by comparing both the logit and complementary log–log links functions with the logit link demonstrating superior fit and interpretability for the proportional QoL index scores. In Fig. 2, residual plots, including plots of standardized weighted residuals versus fitted values and leverage measures, indicated no major deviations from the expected distribution, indicating a well-specified mean structure and constant precision parameter. An analysis of Cook's distance and leverage statistics revealed no influential observations, confirming the stability of the model. The pseudo R2 suggested a satisfactory proportion of explained deviance, and model predictions showed a strong correlation with those generated by a random forest model, supporting the robustness of the results.Fig. 2Diagnostic Plots for beta regression model of QoL in older persons living with HIV in Uganda Diagnostic Plots for beta regression model of QoL in older persons living with HIV in Uganda Random forest regression analysis The random forest analysis was used to assess the relative importance of predictors QoL composite index, providing a complementary non-parametric perspective to the earlier regression models (Table 7).Table 7RF regression outputVariable%IncMSEIncNodePurityAge group0.60900.1811ART duration1.02370.1850Viral Load4.02160.0926Gender/sex8.95510.1007Highest level of education completed14.43150.2825Marital status6.74780.1540Employment status7.00570.0975Activity for livelihood6.27080.2089Place of ART initiation− 3.55850.0583HIV clinical stage2.08780.1647History of opportunistic infections2.82760.0797Presence of chronic NCD7.23490.1010Missed drug refill2.47740.1303 Table 7RF regression outputVariable%IncMSEIncNodePurityAge group0.60900.1811ART duration1.02370.1850Viral Load4.02160.0926Gender/sex8.95510.1007Highest level of education completed14.43150.2825Marital status6.74780.1540Employment status7.00570.0975Activity for livelihood6.27080.2089Place of ART initiation− 3.55850.0583HIV clinical stage2.08780.1647History of opportunistic infections2.82760.0797Presence of chronic NCD7.23490.1010Missed drug refill2.47740.1303 RF regression output Education level emerged as the strongest predictor of QoL composite index, with the highest percentage increase in Mean Squared Error (%IncMSE = 14.43) and substantial node purity. This finding reinforces the importance educational attainment as a key socioeconomic determinant of well-being in later life in OPLHIV, likely through improved health literacy, access to resources, and adaptive coping strategies. Other important predictors were gender, presence of a chronic non-communicable disease (NCD), employment status and marital status. These results are consistent with the findings from the beta regression analysis, further supporting the role of socioeconomic advantage and the presence comorbidities in shaping QoL outcomes. Comparison of the random forest (predicted values) and beta regression (fitted values) The exceptionally strong and statistically significant correlation (r ≈ 0.755, p < 0.001) between the predicted values from the random forest model and the fitted values from the beta regression model indicates an elevated level of agreement between these two distinct analytical approaches. This convergence suggests that, despite their different methodological foundations—one a machine learning ensemble and the other a parametric model for bounded data—both are capturing the same essential patterns in the dataset. The narrow 95% confidence interval (0.712 to 0.793) confirms the precision and reliability of this strong positive relationship. Consequently, this alignment reinforces the robustness and credibility of the findings, validating that the key drivers of the outcome variable are consistently identified and providing greater confidence for stakeholders in using the model outputs for prediction or interpretation. In Fig. 3, the comparison of predicted values from the random forest model with fitted values from the beta regression model showed a strong positive correlation (r = 0.755, p < 0.001; Fig. 3), indicating high concordance between parametric and machine-learning approaches and reinforcing the robustness of the identified determinants of QoL.Fig. 3Predictive performance correlation: random forest vs. beta regression Predictive performance correlation: random forest vs. beta regression Discussion This study demonstrates that the determinants of the Quality-of-Life Composite Index for older persons living with HIV in Uganda are decisively anchored in socio-economic resources and geriatric health rather than solely clinical metrics alone. The consistent finding across multiple analytical models—that tertiary education and employment are the primary independent predictors of a higher composite Index, while the presence of non-communicable diseases (NCDs) is its strongest negative correlate—fundamentally redefines the parameters of "successful aging" in this population [1, 2, 5, 12, 14]. The composite Index, by design, captures the multidimensional essence of well-being, and our results reveal that this totality is shaped more by life-course advantages and the burdens of aging than by HIV viral control alone. The powerful role of tertiary education suggests it operates as a form of lifelong capital that enhances the composite Index through multiple pathways. It likely fosters greater health literacy for navigating complex care needs, provides cognitive resilience against stressors, and facilitates access to social and economic networks that buffer the challenges of aging with a chronic illness [2, 5, 7]. The fact that only tertiary—not primary or secondary—education was significant indicates a threshold effect; advanced education equips individuals with a unique repertoire of skills and resources that directly enhance perceived well-being across physical, psychological, and social domains [7]. Similarly, employment’s significant association with the composite Index underscores that work provides far more than income for older adults. It offers purpose, social integration, and structured daily activity—components directly measured within the social and psychological domains of the composite Index—which are crucial for mitigating the 'role-lessness' and dependency that can undermine well-being in later life [2, 5, 18]. The most compelling clinical insight is the dominant negative influence of NCDs on the composite Index. This finding underscores the central challenge of multi-morbidity for OPLHIV. Conditions like hypertension, diabetes, and arthritis directly impair physical function and general health perceptions (core domains of the Index) while also increasing treatment burden and psychological distress. Their effect on the composite well-being score appears to outweigh that of well-managed HIV, highlighting a critical mismatch between a historically infection-focused care system and the present geriatric reality of its clients. The non-significance of HIV-specific markers like viral load and adherence in our models is therefore interpretable not as a dismissal of their importance, but as an indication that they are necessary but insufficient for achieving a high Quality of Life Composite Index. Once basic disease control is attained, disparities in this holistic measure are driven by socioeconomic and geriatric factors. The significant geographic variation in the composite Index across TASO sites further illustrates how local context—including clinic culture, community support structures, and regional economic opportunities [19, 20]—acts as a key social determinant of health, moderating the expression of individual-level factors. The internal disparities within sites also stress that composite well-being is unevenly distributed, necessitating targeted support for the most vulnerable. Implications for policy and practice HIV programmes should integrate routine geriatric assessment and non-communicable disease management, mainstream socioeconomic support interventions, promote lifelong learning and health literacy, and adopt site-sensitive, person-centred service delivery models that use quality of life composite as a core outcome. Limitations and future research The cross-sectional design limits causal inference, and some measures relied on self-report. Future longitudinal and intervention studies are needed to evaluate strategies that address socioeconomic and geriatric determinants of quality of life among OPLHIV. Conclusion A high Quality of Life Composite Index among older persons living with HIV in Uganda is driven less by HIV-specific clinical indicators and more by accumulated socioeconomic advantage and effective management of ageing-related health conditions. As the population of OPLHIV grows, HIV care must expand beyond viral suppression to actively support the social and structural foundations of holistic well-being across the life course.