http://orcid.org/0000-0002-3658-2570
Figures
Abstract
Background
An enabling environment is believed to have significant and critical effects on HIV and AIDS program implementation and desired outcomes. This paper estimates the paths, directionality, and direct and indirect associations between critical enablers with antiretroviral treatment (ART) coverage and to AIDS-related mortality.
Methods
Frameworks that consider the role of enablers in HIV and AIDS programs were systematically reviewed to develop a conceptual model of interaction. Measurements for constructs of the model were pooled from the latest publicly available data. A hypothetical model, including latent/unobserved factors and interaction of enablers, program activities and outcomes, was analyzed cross-sectionally with structural equation modeling. Coefficients of the model were used to estimate the indirect associations of enablers to treatment coverage and the subsequent associated impact on AIDS related mortality.
Findings
The model’s fit was adequate (RMSEA = 0·084, 90% CI [0·062, 0·104]) and the indirect effects of enablers on outcomes were measured. Enablers having significant associations with increased ART coverage were social/financial protection, governance, anti-discrimination, gender equality, domestic AIDS spending, testing service delivery, and logistics.
Citation: Safarnejad A, Izazola-Licea J-A (2017) Direct and indirect effects of enablers on HIV testing, initiation and retention in antiretroviral treatment and AIDS related mortality. PLoS ONE 12(2): e0172569. https://doi.org/10.1371/journal.pone.0172569
Editor: Paul Richard Harrigan, University of British Columbia, CANADA
Received: August 24, 2015; Accepted: February 4, 2017; Published: February 22, 2017
Copyright: © 2017 Safarnejad, Izazola-Licea. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Data Availability: This paper used exclusively publicly available datasets. Complete references and description where databases can be downloaded or consulted are provided in Table 1 of this paper. Data, model specification, and model outputs are available in the supporting information S1-S6 Files.
Funding: The authors received no specific funding for this work.
Competing interests: The authors have declared that no competing interests exist.
Introduction
Substantial scale-up in resources and coverage of services have been observed in low- and middle-income countries in recent years to total an estimated US$19·14 Billion invested including services for 13·6 million people receiving antiretroviral treatment as of June 2014 [1]. More ambitious targets are being proposed by the United Nations AIDS Program (UNAIDS) to achieve by 2030 a 90% decline in number of new HIV infections and AIDS-related deaths compared to 2010, alongside “zero” discrimination [1].
While targets are set and needed investments estimated, debate continues over the individual and meso-level factors that affect the performance of HIV and AIDS programs. Reports suggest these social, political, economic and developmental enablers (heretofore referred to as “enablers”) have a significant effect on program activities in achieving the objectives of the AIDS response [2–4]. Some researchers view the socio-cultural, political and economic factors as drivers of the epidemic and posit the importance of these structural determinants in the success of HIV and AIDS programs [5, 6]. Others have studied the proximate effect of social, political and structural enablers on increased rate of testing, treatment coverage, adherence, or retention [4, 7–11]. There is, however, a gap in research on the combined interaction effect of enablers on the AIDS response, and measurement of their total (direct and indirect) effects on program objectives.
In the present study, we specify a model for the effect of enablers on program activities and outcomes, guided by a combination of theory and exploratory analysis. We test the hypothesized model’s fit to empirical data, and examine its ability to predict the effect of enablers on program activities and outcomes. We also examine the pathway of effect and the size of the total effect of enablers on outcomes and program activities.
To our knowledge, this is the first study to examine the interaction effects of multiple enablers on program activities and outcomes in the AIDS response, using structural equation modeling. It articulates the enablers relevant to an effective AIDS response, and weighs their impact on program activities and outcomes.
Methods
A systematic review identified frameworks of the program response that consider the role of enablers [2, 12–14]. The frameworks were synthesized to formulate a conceptual model of relationship between enablers, program activities, and outcomes (not shown). Table 1 presents all constructs and variables with available data considered in the conceptual model, the conceptual framework which reference the construct, decisions regarding its inclusion into the final hypothesized model, and the year or range of years for the indicators. Efforts were made to include the latest available data, allowing for a time delay between the critical enablers and changes in outcomes of treatment coverage and mortality.
A measurement model was specified for the latent variables of logistics and governance, and estimated by confirmatory factor analysis [15, 16]. Empirical, cross-sectional, national level data was collected for the constructs in the model. The conceptual model was revised based on availability of data for constructs and comparability across countries, and combined with the measurement model of latent constructs to form a Structural Equation Model (SEM). Constructing variables that are proportions of absolute measures and standard denominators creates weighted constructs, which allow for comparability across countries. The latest available data was used, allowing for time delay between the critical enablers and hypothesized changes in the outcomes. The measures of constructs of the SEM are presented in Table 2.
The SEM was tested using a maximum likelihood estimator [17]. The following model fit statistics were evaluated: (1) the overall chi-square test statistics for the null hypothesis that the model is consistent with the data [18], (2) the Bayesian Information Criterion (BIC) for which a negative value indicates the model is preferred over the saturated model allowing for all variables to be inter-correlated [19, 20], (3) the Comparative Fit Index (CFI) and Tucker Lewis Index (TLI) for which values greater than 0.9 are generally taken as indicative of good fit [21, 22], and (4) the Root Mean Square Error of Approximation (RMSEA) for which values less than 0.1 are indicative of good fit in small sample sizes [23, 24]. Model re-specifications were tested and the modifications were retained if they significantly improved the overall model fit as measured by fit statistics. Standardized coefficients of the final model were used to estimate the magnitude of indirect effects from a number of pathways by which enablers and program activities are associated with the treatment coverage outcomes [25]. As AIDS mortality is estimated using antiretroviral treatment (ART) coverage data in epidemiological projection packages [26], critical enablers were correlated only with ART coverage and not to AIDS related deaths. The association between ART and AIDS mortality was then estimated.
Results
The analysis used a cross sectional dataset containing 59 observations (countries) with complete data on all relevant variables. These observations represent 13 countries in Asia and the Pacific, 9 in Latin America and the Caribbean, 7 countries in Eastern Europe and Central Asia, 25 countries in Sub-Saharan Africa, and 5 countries in the Middle East and North Africa. The countries analyzed by income level were 29 in the low-, 26 in the lower-middle- and 4 in the upper-middle-income brackets (Table 3) [27].
Confirmatory factor analysis
Confirmatory factor analysis for the measurement model of the enabler latent constructs indicates an excellent fit of the model with the data. The chi-square test for model fit is non-significant (χ2 = 43·719, df = 34, p-value = 0·1228), the BIC is 97·032, the CFI and TLI are both 0·98, and the RMSEA is 0·07; the standardized parameter estimates for the measurement model are provided in Fig 1.
Diagram Elements: Boxes are manifest (observed) variables; Circles are latent (unobserved) constructs; Straight arrows point from latent constructs to measurement variables; Curved arrows are unexplained covariance among variables; δ (small delta) are residual variances of the observed variables; Values are standardized regression coefficients and number of stars signify their p-value: * p <.1; ** p <.05;*** p <.01.
Structural equation model
The hypothetical model containing all observed and latent constructs, and the relationships between the constructs, representing the hypotheses, was evaluated as a SEM. Following exhaustion of re-specifications, the final standardized model as estimated by Mplus version 7.3 [17], converged normally after 492 bootstrap draws. The final model contains 16 constructs defined by 41 variables, including two latent constructs (Fig 2). Examination of the model fit indices suggests that the structural model demonstrates an adequate fit to the data, with BIC = -183·388, CFI = 0·906, TLI = 0·894, and RMSEA = 0·084 (90% CI 0·062–0·104). The R2 of the model for ART coverage = 0.484, indicating that the model is explaining close to half of the variance in this key outcome. Together these statistics provide evidence for the reliability and convergent validity of the SEM. Data, model specification and model outputs are available in the supporting information S1 to S6 Files.
Diagram Elements: Boxes are manifest (observed) variables; Circles are latent (unobserved) constructs; Arrows point from explanatory variable to dependent variable; Values are standardized regression coefficients and number of stars signify their p-value: * p <.1; ** p <.05; *** p <.01.
The model supports the hypothesis that anti-discrimination, social/financial protection, gender equality, and good governance and logistics are significant enablers of coverage of people living with HIV and AIDS on treatment. The results show that commitment to anti-discrimination had a positive indirect effect on the treatment outcome by way of conventions on anti-discrimination to which States are parties. Domestic AIDS spending with respect to total health spending positively influences treatment coverage through increased social/financial protection. The theoretical model also specifies a reciprocal interaction between domestic AIDS spending and HIV prevalence.
Governance, correlated positively with logistics, had a direct positive influence on creating an enabling environment for treatment coverage and retention. Four indicators best described governance: regulatory quality, control of corruption, rule of law, and government effectiveness. Indicators for political stability and absence of violence/terrorism and voice and accountability were dropped due to weak factoring with other governance indicators and for reducing the overall fit of the model respectively.
HIV prevalence and testing service delivery are positively associated with HIV testing and counseling in the model, while punitive laws and homophobia are negatively associated with HIV testing and counseling. Cluster analysis identified one stratum of four punitive laws and policies, with a significant explanatory power for variations in HIV testing and counseling. Cluster analysis also identified one stratum of nine variables from the gender scorecard, with significant explanatory power for variations in treatment coverage. The stratum, identified as “gender visibility scorecard” in the model, is indirectly related to HIV testing and counseling through punitive laws and homophobia.
Standard effects
The standardized effects of each construct on the outcome variable were derived. These coefficients elucidate the direct and indirect effects of enablers and program activities on treatment coverage, which can in turn be used to estimate changes in AIDS related mortality. Table 4 lists the constructs in the model, rank sorted by the absolute value of the coefficients with ART coverage. The coefficients can be interpreted as a one standard deviation change in a critical enabler being related to one standard deviation change in ART coverage multiplied by the coefficient.
Not all hypothesized relationships between the constructs in the model were significant. The study could not show a significant correlation between TB/HIV co-treatment, health workforce or donor disbursements per person living with HIV and program outcomes. Concerning a number of variables, there was not a sufficiently large sample size (country-wise) to warrant their inclusion without adversely affecting the overall statistical power of the entire model. For example, the intersection of data on intimate partner violence, gender-based violence, unemployment, poverty, education and data from the Stigma Index would have resulted in an inadequate ratio of observations to parameters, and therefore their hypothesized effect on the program outcomes was not collectively tested.
The association between ART coverage and AIDS related mortality was estimated, with a standardized coefficient of -0.665. This coefficient can be interpreted as one standard deviation change in treatment coverage being related to -0.665 of a standard deviation change in AIDS related deaths as a proportion of people living with HIV.
Discussion
In view of the frameworks of the AIDS response, the effect of enablers on program activities to achieve desired outcomes is significant. This study analyzes the variables that might have a mediating effect between HIV diagnosis and treatment attrition with AIDS related deaths [28, 29]. It further elucidates the role of structural factors in creating an enabling environment to achieve maximum effectiveness of interventions by modeling the interaction of enablers, program activities, and outcomes.
Findings of the study show that gender equality, anti-discrimination, social/financial protection, governance, and logistics were significantly related to coverage and retention on antiretroviral treatment. Testing service delivery, homophobia, and punitive laws were significantly related to HIV testing and counseling. A number of other factors were found to have indirect effects on the program response. Domestic AIDS spending with respect to total health spending is positively associated with treatment coverage through lower out-of-pocket expenditures suggesting that social/financial protection may be alleviating the burden that could preclude or be an obstacle for those who test positive to seek treatment.
Gender visibility was found to be indirectly associated with increased rates of HIV testing and counseling, through its relationship with homophobia and punitive laws. This relationship suggests that consideration of gender differences and mainstreaming of gender norms is a preceding stage to reduced stigma and greater human rights for marginalized/discriminated populations. This staging of enablers resembles the structural models of development theory, such as Rostow’s Stages of Growth [30], which follow a linear progression through phases to reach the outcome of interest. Furthermore, building on previous reports that suggest gender equality is critical to the goals of reduced burden of the disease [31, 32], findings of this study show that gender inequality has an indirect and independent effect on HIV prevalence via human development.
Any change in the parameters included in this model does not necessarily translate to changes in outcome measures, in this case in ART coverage or AIDS related mortality, even if this model was causal. In the next section, we are presenting a summary of the association between the different constructs and the outcome, and presenting the metrics of the associated changes that the model predicts.
Stigma, discrimination, and obstacles to enjoying human rights are hypothesized to be intimately connected to the continuum of care by many direct and indirect links [29, 33]. This study tested those linkages with empirical data, to identify where their effect is most significant, and provided a model to predict the magnitude of their effect. This study corroborates the findings of previous research that negatively associates punitive laws and homophobia with access and use of HIV care [34–36].
Testing service delivery is positively associated with treatment coverage via uptake of HIV testing and counseling. Although not tested in this analysis, the country averages of facilities per population need to account for geographical concentration of people living with HIV.
Assumptions and limitations
Although this study offers a number of new insights into effects of enablers on the AIDS response, it does have limitations. One of these limitations is the sample size due to the national level of observation and the number of countries having valid observations for each of the variables in the model. The parameter to sample size ratio of this study is 1:4, whereas a 1:5 ratio is recommended [16, 37]. Given the limitation on the sample size, the number of parameters accepted in the model before saturation is reached was also limited. Therefore, a number of parameters were not considered, due to the strength and significance of their effect at the cost of diminishing predictive power of the model. Although limited degrees of freedom could be eased with use of a panel dataset, due to lack availability of data on enablers at multiple points in time, this study relied on a cross-sectional dataset.
The analysis also used some ordinal categorical measurements such as the homophobia index, punitive laws and policies and anti-discrimination conventions. As customary, these ordinal measurements are considered in the analysis as imprecise observations on continuous normally distributed variables. All three ordinal variables considered in this study met the minimum condition of having five categories when taken as continuous variables [38].
Optimal measures to report goodness of fit using Structural Equation Modeling have been intensively debated. As stated by Barret (2007) “(p)roponents of two kinds of approaches to model fit can be identified: those who adhere strictly to the result from a null hypothesis significance test, and those who ignore this and instead index model fit as an approximation function” [39]. The recommendations of this author are to oppose the strict cut-off values for Approximate Fit Indices. Other authors have suggested alternative measures for editorial policies regarding the reports of Structural Equation Modeling [40–42].
For this analysis, the model was re-specified several times and the one judged as “the best goodness of fit” was reported. Alternative models lead to roughly the same interpretations as the current model, but they had lower R2 statistics (e.g. R2 < 0.4). Removing and adding certain constructs and the associated R2 for that model further validated the constructs and model paths. While the R2 explains only half of the variance in the outcomes, not all the variables in the model were possible to be included due to the limitation in available empirical data on some constructs. More importantly, since “countries” are the unit of analysis (193 maximum), there was a further limitation in the ratio of observations to parameters (i.e. the more parameters introduced, the more observations needed).
Finally, we decided to keep the marginally significant variables at the 90% confidence interval limit where there was conceptual/theoretical basis to keep them, in particular since the objective of this analysis was exploratory and not causal.
Factor analysis allowed inclusion of a number of related variables in the model as a single latent construct. Governance and logistics were introduced in the model as factors defined by a subset of variables considered. The Human Development Index was used as a single composite, and none of the constructs integrated in the index were analyzed individually.
A number of enablers were identified in the reviewed frameworks for which there was no empirical data available: Community Mobilization, Mass Media, Local Response to Change Risk Environment, Community Centered Design and Delivery, Programme Communication, Management and Incentives, Research and Innovation, Community Systems and Employer Practices, Health Information System, Quality of Care, Safety, and Socio-Cultural Factors.
Conclusions
This study quantified the direct and indirect contribution of enablers to improved treatment coverage and the associated reduction in AIDS related deaths. While the observational cross-sectional data and statistical model used do not allow inference of causality from enablers to AIDS related mortality, they do provide input into the paths or mechanisms involved in the theoretical models frequently used. Improvements in monitoring of enablers temporally will allow cross-sectional panel data analysis that addresses further ingredients necessary in search of establishing causality. Increasing the number of observations by using data at sub-national units of analysis may allow more parameters to be introduced into the model, even while some of the variables/constructs apply to the whole country. This statistical analysis of cross-sectional data, however, allows establishment of correlations useful for providing metrics on the potential effect of critical enablers on outcomes.
There is a need for further research considering the effects of enablers on HIV incidence. Recognizing the scarcity of national level measures of enablers, we recommend that the availability of data be amplified to improve the analyses aiming at the understanding of these relationships. In addition, costing of programs that strengthen the enabling environment, together with the modeled effect of enablers on outcomes, can support cost-effectiveness studies of programs that consider a multi-sectorial approach to activity planning.
The use of Structural Equation Modeling can be further explored to help in the understanding of the complex direct and indirect contributions of multiple variables and constructs. However, there is need to also explore the use of other models whose goodness of fit can be reported using more traditional measures.
With increasing evidence highlighting gaps in the response, and particular populations being left behind [1, 43], the AIDS response can no longer afford to offer more of the same. Scale-up of efficacious programs are necessary but not sufficient to end AIDS. Clearly where laws impose capital punishment for same-sex activities, HIV testing and treatment cannot be done in isolation of such hostile environments. The same can be stated for stigma and discrimination that may constitute barriers to the access to services (treatment or preventive; personal or population based). This study paves the way for further evaluation of bundling program activities with strengthening of the enabling environment that will make the end of AIDS (as a public health threat) possible.
Supporting information
S1 Appendix. Detailed description of the development and selection of the final model.
https://doi.org/10.1371/journal.pone.0172569.s001
(PDF)
S1 Dataset. Dataset of variables used in the models.
Database readable by MPlus for analysis of the structural equation model and the measurement model.
https://doi.org/10.1371/journal.pone.0172569.s002
(CSV)
S2 Dataset. Dataset of variables used in the models.
Dataset readable by MPlus for analysis of the structural equation model and the measurement model as spreadsheet.
https://doi.org/10.1371/journal.pone.0172569.s003
(XLSX)
S1 File. MPlus diagram file for drawing the measurement model of the two latent constructs: Logistics and Governance.
https://doi.org/10.1371/journal.pone.0172569.s004
(TXT)
S2 File. Mplus input code file for the measurement model of the two latent constructs: Logistics and governance.
https://doi.org/10.1371/journal.pone.0172569.s005
(TXT)
S3 File. MPlus output results file for the measurement model of the two latent constructs: Logistics and governance.
https://doi.org/10.1371/journal.pone.0172569.s006
(TXT)
S4 File. MPlus diagram file for drawing the structural equation model of all constructs and paths.
https://doi.org/10.1371/journal.pone.0172569.s007
(TXT)
S5 File. MPlus input code file for the structural equation model of all constructs and paths.
https://doi.org/10.1371/journal.pone.0172569.s008
(TXT)
S6 File. MPlus output results file for the structural equation model of all constructs and paths.
https://doi.org/10.1371/journal.pone.0172569.s009
(TXT)
Acknowledgments
The authors’ deep gratitude goes to Mr Patrick Eba, Dr Mariangela Simao, Dr Peter Ghys, and Mrs Alexandrina Iovita for their invaluable feedback and review. The authors also sincerely thank Dr Shawn Bauldry, who as an expert in structural equation models, independently reviewed all models, methods and results, and suggested improvements in execution and reporting.
Author Contributions
- Conceptualization: JAI.
- Data curation: AS.
- Formal analysis: AS.
- Funding acquisition: JAI.
- Investigation: AS.
- Methodology: AS JAI.
- Project administration: JAI.
- Software: AS.
- Supervision: JAI.
- Validation: JAI.
- Visualization: AS JAI.
- Writing – original draft: AS JAI.
- Writing – review & editing: JAI.
References
- 1.
Joint United Nations Programme on HIV/AIDS. Fast-Track: ending the AIDS epidemic by 2030. Geneva: UNAIDS; 2014.
- 2. Schwartländer B, Stover J, Hallett T, Atun R, Avila C, Gouws E, et al. Towards an improved investment approach for an effective response to HIV/AIDS. The Lancet. 2011 Jun 17; 377(9782): p. 2031–41.
- 3. Sumartojo E, Doll L, Holtgrave D, Gayle H, Merson M. Enriching the mix: incorporating structural factors into HIV prevention. AIDS. 2000 Jun; 14(suppl 1): p. S1–S2.
- 4. Gupta GR, Parkhurst JO, Ogden JA, Aggleton P, Mahal A. Structural approaches to HIV prevention. The Lancet. 2008 Sep 5; 372(9640): p. 764–75.
- 5. Dhai A. HIV and AIDS in Africa: social, political, and economic. Theoretical medicine and bioethics. 2008 Sep 1; 29(5): p. 293–6. pmid:19085088
- 6. Kippax S. Understanding and integrating the structural and biomedical determinants of HIV infection: a way forward for prevention. Current Opinion in HIV and AIDS. 2008 Jul 1; 3(4): p. 489–94. pmid:19373010
- 7. Ntsepe Y, Simbayi LC, Shisana O, Rehle T, Mabaso M, Ncitakalo N, et al. Perceptions about the acceptability and prevalence of HIV testing and factors influencing them in different communities in South Africa. SAHARA-J. 2014 Jan 2; 11(1): p. 138–47.
- 8. Underwood C, Hendrickson Z, Van Lith LM, Lengwe Kunda JE, Mallalieu EC. Role of community-level factors across the treatment cascade: a critical review. Journal of Acquired Immune Deficiency Syndromes. 2014 Aug 15; 66: p. S311–8. pmid:25007202
- 9. Mahajan AP, Sayles JN, Patel VA, Remien RH, Ortiz D, Szekeres G, et al. Stigma in the HIV/AIDS epidemic: A review of the literature and recommendations for the way forward. AIDS. 2008 Aug; 22(Suppl 2): p. S67–S79.
- 10. Cameron S, Godwin J. Barriers to legal and human rights in Australia in the era of HIV treatment as prevention. AIDS Education and Prevention. 2014 Jun; 26(3): p. 202–13. pmid:24846483
- 11. Kagee A, Remien R, Berkman A, Hoffman S, Campos L, Swartz L. Structural barriers to ART adherence in Southern Africa: challenges and potential ways forward. Global public health. 2011 Jan 1; 6(1): p. 83–97. pmid:20509066
- 12. Boerma JT, Weir SS. Integrating demographic and epidemiological approaches to research on HIV/AIDS: the proximate-determinants framework. Journal of Infectious Diseases. 2005 Feb 1; 191(Suppl 1): p. S61–7.
- 13.
World Health Organization. Everybody business: strengthening health systems to improve health outcomes: WHO’s framework for action. Geneva: WHO; 2007.
- 14.
Weir S, Sabin K, Abdul-Quader A, Au M, Bok L, Butler J, et al. Operational Guidelines for Monitoring and Evaluation of HIV Programmes for Sex Workers, Men who have Sex with Men, and Transgender People—Volume I National and Sub-National Levels. MEASURE Evaluation. 2013. http://www.cpc.unc.edu/measure/publications/ms-11-49a. Accessed 19 August 2014.
- 15. Jöreskog KG. A general approach to confirmatory maximum likelihood factor analysis. Psychometrika. 1969 Jun; 34(2): p. 183–202.
- 16.
Kline RB. Principles and Practice of Structural Equation Modeling. 3rd ed. New York: Guilford Press; 2011.
- 17.
Muthén LK, Muthén BO. Mplus User’s Guide. 7th ed. Los Angeles, CA: Muthén & Muthén; 1998–2012.
- 18.
Bollen KA. Structural Equations with Latent Variables. 1st ed. New York: Wiley; 1989.
- 19. Raftery A. Bayesian model selection in social research. Sociological Methodology. 1995; 25: p. 111–63.
- 20. Schwarz G. Estimating the dimension of a model. The Annals of Statistics. 1978; 6(2): p. 461–4.
- 21. Bentler PM. Comparative fit indexes in structural models. Psychological Bulletin. 1990 Mar; 107(2): p. 238–46. pmid:2320703
- 22. Tucker LR, Lewis C. A reliability coefficient for maximum likelihood factor analysis. Psychometrika. 1973 Mar; 38(1): p. 1–10.
- 23. Chen F, Curran PJ, Bollen KA, Kirby J, Paxton P. An empirical evaluation of the use of fixed cutoff points in RMSEA test statistic in structural equation models. Sociological Methods & Research. 2008 May; 36(4): p. 462–94.
- 24.
Steiger JH, Lind JC. Statistically-based tests for the number of common factors. In Annual Spring Meeting of the Psychometric Society; 1980; Iowa City.
- 25.
Lleras C. Path Analysis. In Kempf-Leonard K, editor. Encyclopedia of social measurement. New York: Elsevier Academic Press; 2005. p. 25–30.
- 26. Stover J, Johnson P, Zaba B, Zwahlen M, Dabis F, Ekpini RE. The Spectrum projection package: improvements in estimating mortality, ART needs, PMTCT impact and uncertainty bounds. Sex Transm Infect. 2008 Aug; 84(Suppl 1): p. i24–i30.
- 27.
Joint United Nations Programme on HIV/AIDS. Global report: UNAIDS report on the global AIDS epidemic 2013. Geneva: UNAIDS; 2013.
- 28. Lamb MR, El-Sadr WM, Geng E, Nash D. Association of adherence support and outreach services with total attrition, loss to follow-up, and death among ART patients in sub-Saharan Africa. PLoS One. 2012 Jun 7; 7(6): p. e38443. pmid:22685569
- 29. Gardner EM, McLees MP, Steiner JF, Rio Cd, Burman WJ. The spectrum of engagement in HIV care and its relevance to test-and-treat strategies for prevention of HIV infection. Clinical Infectious Diseases. 2011 Mar 15; 52(6): p. 793–800. pmid:21367734
- 30.
Todaro MP, Smith SC. Economic Development. 10th ed. Harlow: Pearson Education; 2009.
- 31. Richardson ET, Collins SE, Kung T, Jones JH, Tram KH, Boggiano VL, et al. Gender inequality and HIV transmission: a global analysis. Journal of the International AIDS Society. 2014 Jun 27; 17: p. 19035. pmid:24976436
- 32. Peacock D, Stemple L, Sawires S, Coates TJ. HIV/AIDS, and human rights. J Acquir Immune Defic Syndr. 2009 Jul 1; 51(Suppl 3): p. S119–S125.
- 33.
Aggleton P, Wood K, Malcolm A, Parker R. HIV-related stigma, discrimination and human rights violations: case studies of successful programmes. Geneva: Joint United Nations Programme on HIV/AIDS (UNAIDS); 2005.
- 34. Ayala G, Makofane K, Santos GM, Arreola S, Hebert P, Thomann M, et al. HIV treatment cascades that leak: correlates of drop-off from the HIV care continuum among men who have sex with men worldwide. J AIDS Clin Res. 2014; 5(8): p. 331.
- 35. Gruskin S, Pierce GW, Ferguson L. Realigning government action with public health evidence: the legal and policy environment affecting sex work and HIV in Asia. Culture, Health & Sexuality. 2014; 16(1): p. 14–29.
- 36.
United Nations Development Programme. Thailand’s Response to HIV/AIDS: Progress and Challenges. Bangkok: UNDP; 2004.
- 37. Bentler PM, Chou CP. Practical issues in structural modeling. Sociological Methods and Research. 1987 Aug 1; 16(1): p. 78–117.
- 38. Hox JJ, Bechger TM. An introduction to structural equation modeling. Family Science Review. 1998 Nov; 11: p. 354–73.
- 39. Barret P. Structural equation modelling: Adjudging model fit. Personality and Individual Differences. 2007 May; 42(5): p. 815–24.
- 40. Goffin RD. Assessing the adequacy of structural equation models: Golden rules and editorial policies. Personality and Individual Differences. 2007 May; 42(5): p. 831–9.
- 41. Bentler PM. On tests and indices for evaluating structural models. Personality and Individual Differences. 2007 May; 42(5): p. 831–9.
- 42. Hayduk L, Cummings G, Boadu K, Pazderka-Robinson H, Boulianne S. Testing! testing! one, two, three—Testing the theory in structural equation models! Personality and Individual Differences. 2007 May; 42(5): p. 841–50.
- 43.
Joint United Nations Programme on HIV/AIDS. The Gap Report. Geneva: UNAIDS; 2014.