### Study design and study population

A cross-sectional study was conducted among PLHIV over the age of 16 in the 10 municipalities of Yunnan province, China, from October 2019 to May 2020. A convincing sampling method was used. The sample size was calculated by the Yaro Yamane approach for a finite population using the formula of ({ text {n}} = N / left ({1 + N (e) ^ {2}} right) ) [20]. In our study, n represents the expected sample size. *NOT* represents the finite population from which the sample is derived. We set a number for the total estimate of PLWHA in the ten selected areas. e represents the significance level and we define 0.05. Finally, the estimated sample size was 354. Based on the reported number of PLHIV from each municipality, a practical sample of 150-200 was included for each selected area. We also excluded respondents with cognitive disabilities who were unwilling to complete the survey. Our study total included 1997 participants. All investigators from the local Center for Disease Control and Prevention (CDC) and social organizations have been strictly trained to implement the face-to-face survey.

### Data gathering

#### Health-related quality of life

Individual-level HRQoL data was measured using the SF-12 and EQ-5D-5L. The 12-item Short Health Survey (SF-12), which is the shortened version of the 36-item Short Health Survey (SF-36), could explain at least 90% of the accuracy of the SF -36 [21]. The SF-12 consists of eight domains to generate two distinct summary scores, Physical Functional Scores (PCS) and Mental Functional Scores (MCS) ranging from 0 to 100. Higher scores indicate better HRQH. de Cronbach = 0.89. We also used EQ-5D-5L to simultaneously measure HRQoL. The EQ-5D-5L could define the 3125 possible health states by the different combinations. We adopted the Chinese population-based preference trade-off time (TTO) to transform the measures into a utility index (UI, Table 1), thus producing a single preference-based index ranging from -0.391 to 1000, where 0 was equal to death. and – 0.391 meant worse than death. For example, when we calculated a combination of “21145”, the UI was equal to 1 – 0.066 – 0 – 0 – 0.252 – 0.258 = 0.424 [22]. de Cronbach = 0.79.

#### HIV demographic and diagnostic variables

All demographic data, including age, race / ethnicity, level of education, marital status, household income per year, if infection status was known to others, and HIV diagnostic variables including baseline infection status, pattern of transmission, duration of ART, and most recent CD4 counts were obtained using self-designed questionnaires.

#### Social support, depression and anxiety

We used the Social Support Rating Scale (SSRS) established by Xiao Shuiyuan in 1986 primarily for the Chinese population. [23]. It consisted of ten items and three dimensions. A respondent’s social support was measured on three scales: objective social support, subjective social support, and use of support. Final social support was obtained by averaging all three-dimensional item scores. A higher total score demonstrated a higher level of perceived social support. de Cronbach = 0.68. Anxiety and depression were measured using the Chinese version of the Hospital Anxiety and Depression Scale (HADS) [24], which is a short scale with 14 items designed for the diagnosis of anxiety and depression in non-psychiatric patients. Anxiety and depression were assessed using seven items respectively. Higher scores demonstrated more severe symptoms of depression or anxiety. de Cronbach = 0.85.

#### Area level data collection

We gathered socio-economic data at the municipal level from the Yunnan Statistical Yearbook in 2020 produced by the Statistical Bureau of Yunnan Province. [25]. We used the gross domestic product (GDP) per capita, employment rate, birth rate, death rate and natural growth rate to calculate the social economic effect at the municipal level, which was encouraged to measure the economic and social status of the regions. Other municipal data on the prevention strategy came from the Implementation Strategy Quality Assessment System, which was designed by the Yunnan CDC, which included epidemic surveillance, risk behavior intervention, care for PLWHA, monitoring and experimental management to build the prevention strategy. The strategy could be formed of different models including good quality strategy (strategy 1), traditional strategy with advantage (strategy 2), advanced strategy (strategy 3) and general strategy (strategy 4).

### Data analysis

For the statistical description, we used the mean (standard deviation) and median (interquartile range) to describe the total HRQoL measured using EQ-5D-5L and PCS-12 and MSC-12, respectively.

For the statistical analysis, our study first used five indicators (GDP per capita, employment rate, birth rate, death rate and natural growth rate) to demonstrate the socio-economic effect. The six indicators (the epidemic surveillance score, the global intervention score of sex workers, the global intervention score for men who have sex with men, the global score for the care and monitoring of PLWHIV and the quality score of HIV laboratory tests) demonstrated the strategy implemented in each area. We used principal component analysis to construct the socio-economic and strategic effects of each domain. Given the sensitivity to dimensions for principal component analysis, all calculated indicators were adjusted between 0 and 1 using min-max normalization to eliminate the influence of dimensional inconsistency [12]. The normalization equation is represented as fluid.

$$ S_ {ij} = frac {{x_ {ij} – x_ {ij ( min)}}} {{x_ {ij ( max)} – x_ {ij ( min)}}} $$

*S*_{I} demonstrated the transfer *I* area indicator *j*, *X*_{I} demonstrated the original *I* area indicator *j*, and *X*_{ij (min)} and *X*_{ij (max)} demonstrated the max and the min *I* indicators in all areas.

We defined the principal components with reference to the variation greater than 80%, as well as explanatory variables according to the practice for the socio-economic effect and the strategy effect. In our study, for the socio-economic and strategic effect, the scores of the first and second components were calculated as follows:

The score of the first component = – 0.170 × GDP per capita – 0.229 × employment rate + 0.387 × birth rate + 0.222 × death rate + 0.362 × natural growth rate

The score of the second component = 0.488 × GDP per capita + 0.434 × employment rate + 0.183 × birth rate + 0.322 × death rate + 0.112 × natural growth rate

For the practical effect of the strategy, the scores of the first and second components were calculated as follows:

The score of the first component = 0.127 × the epidemic surveillance score + 0.343 × the overall score of the intervention with sex workers + 0.372 × the overall score of the intervention men has sex with men + 0.379 × the overall score for the care and monitoring of PLWHIV + 0.062 × the score for the quality of HIV laboratory tests

The score of the second component = 0.428 × the epidemic surveillance score – 0.324 × the overall score of the intervention of sex workers + 0.033 × the overall score of the intervention of men who have sex with men + 0.011 × the overall score for the care and monitoring of PLWHA + 0.648 × the score for the quality of HIV laboratory tests

Second, a one-way ANOVA was used to perform univariate analysis to identify predictors with significant differences. Candidates for multivariate analysis included the following variables: (1) professionals associated with HRHL among PLHIV; (2) social support, anxiety and depression; and (3) variables at the level of *P* less than 0.1 in one-way ANOVA.

Our study used a multi-level model (MLM) to explore the effects of personality, social economy, and strategy on health-related quality of life in PLHIV. [13, 26]. We define the individual level at level 1 and the regional level at level 2. Based on the socio-economic models and models of strategic practice by component analyzes, we mainly examined the effect of the strategy as variables at the level. domain level to predict HRQL, with age, race / ethnicity, marital status, education level, occupation, household income per year, others know HIV status, baseline infectious status, transmission pattern, duration of ART, most recent CD4 counts, social support score, anxiety score, and depression score as individual-level variables to predict HRQoL . We adapted the random coefficient model to fit it. Let *Yes*_{I} be the HRQoL score for the individual *I* of the area *j*. We used an individual-level predictor and an area-level predictor to keep the scoring simple and without loss of generality, indicated by *X*_{I} and *z*_{j}. Respectively. We have listed the traditional model at a level like

$$ y_ {ij} = beta_ {0} + beta_ {1} x_ {ij} + beta_ {2} z_ {j} + varepsilon_ {ij} $$

Occasional heterogeneity was expressed by adding an interaction term, ( beta_ {3} x_ {ij} z_ {j} ), to the model.

For MLM, the individual-level model only includes individual-level predictors and its regression coefficients were not fixed but varied from domain to domain and integrated into a domain-level model.

Model at the individual level: (y_ {ij} = beta_ {0j} + beta_ {1j} x_ {ij} + varepsilon_ {ij} ).

Zone-level interception model: ( beta_ {0j} = beta_ {00} + beta_ {01} z_ {j} + mu_ {0j} ).

Slope model at zone level: ( beta_ {1j} = beta_ {10} + beta_ {11} z_ {j} + mu_ {1j} ).

Zone-level errors ( left ({ mu_ {0j} mu_ {1j}} right) sim N left ({0, sum {= left[ begin{gathered} tau_{00} tau_{01} hfill tau_{01} tau_{11} hfill end{gathered} right]} } right)) and were assumed to be independent of errors at the individual level ( varepsilon_ {ij} sim N left ({0, sigma ^ {2}} right) ). The intersection and slope of the model at the individual level were determined by the domain-level variable. The main effect of domain-level variables and causal heterogeneity were determined by examining the intercept and slope of the model at the individual level, respectively.

The MLM divided the total variance of HRQoL between countries (i.e. Σ) and within country (i.e. σ^{2}) difference.

We could also include several independent variables in the complete MLM, such as multivariate models. In our study, the independent variables at the individual level were age, race / ethnicity, level of education, household income per year, recent CD4 + T count, pattern of transmission, duration of ART, social support, anxiety and depression. The domain-level predictors in the study were the socio-economic effect and the strategy effect.

We used STATA version 14.0 (StataCorp LLC, College Station, TX) to perform all statistical analysis.