Secondary Logo

Journal Logo


Accuracy of self-reported HIV-testing history and awareness of HIV-positive status in four sub-Saharan African countries

Xia, Yiqinga; Milwid, Rachael M.a; Godin, Arnauda; Boily, Marie-Claudeb; Johnson, Leigh F.c; Marsh, Kimberlyd; Eaton, Jeffrey W.b; Maheu-Giroux, Mathieua

Author Information
doi: 10.1097/QAD.0000000000002759



Monitoring the HIV treatment and care cascade is central to the Joint United Nations Programme on HIV/AIDS (UNAIDS) objective of ending the AIDS epidemic as a public health threat by 2030 [1]. Routine tracking of population-level progress towards the UNAIDS’ 2030 95–95–95 diagnostic, treatment, and viral load suppression targets can guide public health initiatives and improve programmatic efficiencies [2]. However, estimating progress towards the first pillar of the targets – the percentage of people living with HIV (PLHIV) who know their HIV status – is challenging. In sub-Saharan Africa (SSA), where 67% of the 38 million PLHIV were estimated to reside in 2019 [3], measures of awareness are typically constructed from model synthesized data using self-reported HIV testing history (i.e. ever tested for HIV) [4,5] or reported directly from nationally representative household surveys (i.e. reporting a positive HIV test) [6–9].

Consideration of the potential for measurement bias is needed when interpreting self-reported survey data. Studies have shown that self-reports concerning sensitive information, for instance, an individual's HIV-testing history or status, could be affected by nondisclosure [8,10,11]. For example, inconsistencies have been documented in Kenya and Malawi between self-reported data and biomarkers for metabolites of antiretrovirals (ARVs) and viral load suppression [7,11]. Although previous studies have sought to validate the accuracy of self-reported HIV status [11–13], analyzing recent data on both nondisclosure of self-reported HIV-testing history and HIV status among PLHIV is key to improving the validity of these estimates.

Surveys that collect both self-reported information and ARV metabolite biomarkers can be used to assess the accuracy of those self-reporters. In this study, Bayesian latent class models are used to estimate the sensitivity of (i) self-reported HIV-testing history and (ii) awareness of HIV status among PLHIV based on the presence of detectable ARVs [14].


Study population

The Population-based HIV Impact Assessment (PHIA) surveys are nationally representative multistage household-based surveys designed to provide population-level information on the burden of HIV disease and to document the progress of HIV programs [15–18]. All four PHIA surveys with available microdata on PLHIV aged 15+ years were included in our analysis: Swaziland (Eswatini) HIV Incidence Measurement Survey 2 (2015–2016), Malawi PHIA (2015–2016), Tanzania HIV Impact Survey (2016–2017), and Zambia PHIA (2016).

Self-reports and antiretroviral status

Participants who reported having received the results of any HIV test were classified as ever tested and received results (hereafter referred to as ‘ever tested’; see Table S1, Participants who reported having ever received a positive test result after any HIV test, were classified as aware of HIV-positive status. A detailed description of the PHIA and ARV laboratory algorithms can be found Text S2,

Bayesian latent class models

Bayesian latent class models are useful when there are no gold standards [16] and can efficiently account for parameter uncertainty [19]. We used this type of model to quantify the sensitivity of two self-reported outcomes among PLHIV: HIV-testing history and HIV-positive status awareness. Cross-tabulations of self-reports with ARV biomarkers provide empirical information on their sensitivity among those with detectable antiretrovirals (ARVs) (Fig. 1).

Fig. 1:
Observed and unobserved data structure of self-reported ever tested and received results, and antiretroviral metabolites status.

For participants with detectable ARVs, it assumed that: they had been tested for HIV, received their results, and were aware of their status (as they were receiving care); and there were no false positives in the detection of ARV metabolites. In the four PHIA surveys, an overall 99.9% of HIV-negative participants report having never received a positive HIV result, which aligns with other findings [12], we assumed no false positive for self-reported HIV-testing history, or awareness of HIV-positive status.

As ARV biomarker data only provide information about the sensitivity of self-reports among participants receiving treatment, the ratio of nondisclosure for PLHIV without versus with detectable ARVs was given a log-normal prior distribution with a mean of log(1.48) (standard error: 4) to estimate the sensitivity among people without detectable ARVs. This prior was elicited by reviewing available studies. The pooling of two studies conducted in rural Mozambique and Malawi [20,21] suggests that people not receiving ARVs are 1.48 more likely to not disclose their diagnosis. Additional analyses were conducted to investigate the influence of this prior on our results. Equations and prior distributions are presented in Table S2, and Text S1,

Given known biases in self-reported estimates of HIV status awareness, analysts often reclassify individuals not aware of their status but with detectable ARVs as being aware of their HIV-positive status, as in published PHIA reports. To examine the impact of this partial adjustment, we compared the unadjusted, ARV-reclassified (as in PHIA reports), and Bayesian-adjusted (ARV status and nondisclosure) estimates of PLHIV aware of their status.

Models were calibrated separately for each country and for subgroup analyses (i.e. age, sex, urban/rural, and socio-economic status). Bayesian hierarchical models using Markov Chain Monte Carlo, implemented through the JAGS software [22] and the rjags packages, were used to approximate the posterior densities [23,24].


Ethics approval for secondary data analyses was obtained from McGill University's Faculty of Medicine's Institutional Review Board (A10-E72-17B).


Overall, 3003 PLHIV from Eswatini, 2227 from Malawi, 1831 from Tanzania, and 2467 from Zambia were included in the analyses, with 76.0%, 68.1%, 61.5%, and 53.9% of them with detectable ARVs. In all countries, a high fraction of PLHIV reported having ever been tested and the proportion of PLHIV reporting being aware of their status ranged from 58.6% in Tanzania to 86.5% in Eswatini (Table S3,

Sensitivity of self-reports

Self-reported testing history

Among participants with detectable ARVs, the estimated sensitivity was highest in Eswatini (99.5%; 95% credible interval [95% Crl]: 99.2–99.8%), followed by Malawi (98.2%; 97.5–98.8%), Zambia (97.4%; 96.5–98.1%), and Tanzania (96.6%; 95.3–97.6%) (Fig. 2 ). For people without detectable ARVs, the estimated sensitivity was 2.4% points (0.1–11.4%) lower than those with detectable ARVs in Tanzania. The differences were smaller elsewhere (Table S4,

Fig. 2:
Posterior medians and 95% credible intervals for selected outcomes.
Fig. 2 (Continued):
Posterior medians and 95% credible intervals for selected outcomes.

Self-reported awareness of HIV status

The sensitivity of self-reported awareness of HIV-positive status among participants with ARV metabolites was 97.4% (96.7–98.0%) in Eswatini, 94.2% (93.0–95.4%) in Malawi, 92.3% (90.5–93.8%) in Tanzania, and 91.6% (90.1–92.9%) in Zambia (Fig. 2 a). The estimated differences in sensitivity between PLHIV with and without detectable ARVs ranged from 1.8% points (0.1–8.5%) in Eswatini to 6.2% points (0.3–29.0%) in Zambia.

Differences by sex, age, rural/urban, and socioeconomic status

Among participants with detectable ARVs, women had higher sensitivities of self-reported HIV-testing history (0.9–2.4% points) and HIV status awareness (0.8–4.5% points) than men (Fig. 2 b). The estimated sensitivities were the lowest at age 15–24 years (94.7–97.2% for HIV-testing history and 83.9–91.9% for HIV status awareness) in all countries (Fig. 2 c). Participants residing in urban and rural areas had similar sensitivities and variations by socio-economic status were also small (Figure S1, Similar results for PLHIV without detectable ARVs were observed (Figure S2,

Adjusted proportion of people living with HIV ever tested and people living with HIV aware of their status

Adjusting for ARV status and nondisclosure influenced the estimates of the proportion ever tested for HIV less (largest difference between the adjusted and the self-reports was 3.9% points in Tanzania) than the estimated proportions of PLHIV aware of their status (largest difference was 7.2% points in Zambia) (Fig. 2 d). Results were robust to the assumed nondisclosure ratio for PLHIV without detectable ARVs when antiretroviral therapy (ART) coverage is high (Figure S3,


Self-reported information on HIV-testing and diagnosis are primary data sources used to monitor trends in the HIV treatment and care continuum [2,10]. Such data have also been proposed to estimate cross-sectional HIV incidence [25]. In this study, we leveraged ARV biomarkers from representative surveys in Eswatini, Malawi, Tanzania, and Zambia to estimate the sensitivity of self-reported HIV-testing history and awareness of HIV-positive status among PLHIV. We found that, among PLHIV with detectable ARVs, self-reports of HIV-testing history have a high sensitivity (>96%) and self-reported awareness of HIV status had a marginally lower sensitivity (>91%) across these settings. These findings support the use of self-reported testing history to estimate testing trends and to model diagnosis coverage [5].

Social desirability bias may partly explain the nonnegligible lower sensitivities among male PLHIV and those aged 15–24 years [26]. However, differences in survey instruments could also result in higher sensitivities for specific groups. For example, in the PHIA survey, women were asked about HIV-testing up to four times (before pregnancy, during pregnancy, during labor, and at their last HIV test), whereas men were asked only once. The classification of women based on the positive response to any of the four questions could increase the probability of women disclosing their true status.

The availability of ARV biomarkers enables the reclassification of PLHIV who do not disclose their status but for which ARV metabolites are detected as ‘aware’. We have found that this partial adjustment may be insufficient, especially if ART coverage is low in the surveyed population and the ratio of nondisclosure among those not on ART is high [27–30]. To accurately estimate awareness of status, results must also be adjusted for nondisclosure among PLHIV not receiving ART.

Our results need to be interpreted considering certain limitations. First, none of the PHIA surveys with publicly available microdata, are located in the West and Central African regions, where nondisclosure could be higher [31]. Indeed, a recent meta-analysis suggest that, globally, one in five PLHIV may not report their status [32]. The PHIA included here had some of the lowest levels of nondisclosure of the reviewed studies, suggesting that other settings could have lower sensitivities. Second, our study design limited our assessment of the sensitivity of testing history to PLHIV, and findings should not be extrapolated to people not living with HIV. Finally, it is not possible to empirically validate the sensitivity of self-reports among PLHIV without ARV metabolites. As such, we used information from two previous studies that used medical records to inform the nondisclosure ratio. Results could be sensitive to this nondisclosure ratio but the high ART coverage in the four countries mitigates this influence (Figure S3,

Strengths of this study include the use of standardized survey and laboratory data (i.e. detection of ARV metabolites). Second, the Bayesian latent class models propagate uncertainty to our results by assuming prior distributions and generating posterior credible intervals. Finally, we examined differences in the sensitivity of self-reports within different subgroups.

In conclusion, self-reported HIV-testing histories have high sensitivities in the countries examined but self-reported awareness of HIV status is lower. Whenever available, ARV biomarker data can be used to adjust self-reports but such adjustments may still underestimate diagnosis coverage, especially if ART coverage is low in that population. Our results can be used to produce more accurate estimates of the UNAIDS targets. Future research should extend this work to other regions and populations.


We acknowledge funding from the Steinberg Fund for Interdisciplinary Global Health Research (McGill University). MMG's research program is funded through a Canada Research Chair (Tier 2) in Population Health Modeling. J.W.E. acknowledges funding from the Bill and Melinda Gates Foundation and UNAIDS. M.C.B. acknowledge funding from MRC Centre for Global Infectious Disease Analysis (MRC GIDA, MR/R015600/1). This award is jointly funded by the UK Medical Research Council (MRC) and the UK Department for International Development (DFID) under the MRC/DFID Concordat agreement and is also part of the EDCTP2 programme supported by the European Union. L.F.J. acknowledges funding from UNAIDS.

Authors’ statement: Y.X., A.G., R.M.M., and M.M.G. were responsible for study design. Y.X. was responsible for literature search, drafting the article, and statistical analysis. R.M.M., A.G., M.C.B., L.F.J., K.M., J.W.E., and M.M.G. were responsible for reviewing and editing the article.

Conflicts of interest

There are no conflicts of interest.


1. UNAIDS. 90–90–90: an ambitious treatment target to help end the AIDS epidemic. Geneva: Joint United Nations Programme on HIV/AIDS (UNAIDS); 2014.
2. Rentsch CT, Reniers G, Machemba R, Slaymaker E, Marston M, Wringe A, et al. Nondisclosure of HIV testing history in population-based surveys: implications for estimating a UNAIDS 90-90-90 target. Global Health Action 2018; 11:1553470.
3. UNAIDS. UNAIDS data 2019.
4. Maheu-Giroux M, Marsh K, Doyle CM, Godin A, Laniece Delaunay C, Johnson LF, et al. National HIV testing and diagnosis coverage in sub-Saharan Africa: a new modeling tool for estimating the ’first 90’ from program and survey data. AIDS 2019; 33: (Suppl 3): S255–S269.
5. Giguère K, Eaton JW, Marsh K, Johnson LF, Johnson CC, Ehui E, et al. Trends in knowledge of HIV status and efficiency of HIV testing services in sub-Saharan Africa (2000–2020): a modelling study of survey and HIV testing program data. medRxiv 2020.
6. Staveteig S, Wang S, Head SK, Bradley SEK, Nybro E. Demographic patterns of HIV testing uptake in sub-Saharan Africa: DHS comparative reports 30. ICF Macro, Calverton. 2013.
7. Kim AA, Mukui I, Young PW, Mirjahangir J, Mwanyumba S, Wamicwe J, et al. KAIS Study Group. Undisclosed HIV infection and antiretroviral therapy use in the Kenya AIDS indicator survey 2012: relevance to national targets for HIV diagnosis and treatment. AIDS 2016; 30:2685–2695.
8. Anand A, Shiraishi RW, Bunnell RE, Jacobs K, Solehdin N, Abdul-Quader AS, et al. Knowledge of HIV status, sexual risk behaviors and contraceptive need among people living with HIV in Kenya and Malawi. AIDS 2009; 23:1565–1573.
9. Cherutich P, Kaiser R, Galbraith J, Williamson J, Shiraishi RW, Ngare C, et al. KAIS Study Group. Lack of knowledge of HIV status a major barrier to HIV prevention, care and treatment efforts in Kenya: results from a nationally representative study. PLoS One 2012; 7:e36797.
10. WHO. Consolidated strategic information guidelines for HIV in the health sector. In. Geneva: World Health Organization; 2015.
11. Fishel J, Barrere B, Kishor S. Validity of data on self- reported HIV status and implications for measurement of ARV coverage in Malawi. DHS Working Paper No. 81. ICF International, Calverton, Maryland, USA. 2012.
12. Fisher DG, Reynolds GL, Jaffe A, Johnson ME. Reliability, sensitivity and specificity of self-report of HIV test results. AIDS Care 2007; 19:692–696.
13. Rohr JK, Xavier Gomez-Olive F, Rosenberg M, Manne-Goehler J, Geldsetzer P, Wagner RG, et al. Performance of self-reported HIV status in determining true HIV status among older adults in rural South Africa: a validation study. J Int AIDS Soc 2017; 20:21691.
14. Goncalves L, Subtil A, de Oliveira MR, do Rosario V, Lee PW, Shaio MF. Bayesian Latent Class Models in malaria diagnosis. PLoS One 2012; 7:e40633.
15. Malawi population-based HIV impact assessment (MPHIA) 2015–2016 data use manual supplement. New York, NY. December 2018.
16. Swaziland HIV incidence measurement survey 2 (SHIMS2) 2016–2017 data use manual supplement. New York, NY. April 2019.
17. Tanzania HIV impact survey (THIS) 2016–2017 data use manual supplement. New York, NY, December 2018.
18. Zambia population-based HIV impact assessment (ZAMPHIA) 2016–2017 data use manual supplement. New York, NY, February 2019.
19. Joseph L, Gyorkos TW, Coupal L. Bayesian estimation of disease prevalence and the parameters of diagnostic tests in the absence of a gold standard. Am J Epidemiol 1995; 141:263–272.
20. Fuente-Soro L, Lopez-Varela E, Augusto O, Sacoor C, Nhacolo A, Honwana N, et al. Monitoring progress towards the first UNAIDS target: understanding the impact of people living with HIV who re-test during HIV-testing campaigns in rural Mozambique. J Int AIDS Soc 2018; 21:e25095.
21. Chasimpha SJD, McLean EM, Dube A, McCormack V, Dos-Santos-Silva I, Glynn JR. Assessing the validity of and factors that influence accurate self-reporting of HIV status after testing: a population-based study. AIDS 2020; 34:931–941.
22. Plummer M. JAGS: A program for analysis of Bayesian graphical models using Gibbs sampling. Proceedings of the 3rd International Workshop on Distributed Statistical Computing (DSC 2003). Vienna, Austria, 2003.
23. Gelfand AE, Smith AFM. Sampling-based approaches to calculating marginal densities. J Am Stat Assoc 1990; 85:398–409.
24. Gelfand AE, Hills SE, Racine-Poon A, Smith AFM. Illustration of Bayesian inference in normal data models using Gibbs sampling. J Am Stat Assoc 1990; 85:972–985.
25. Fellows IE, Shiraishi RW, Cherutich P, Achia T, Young PW, Kim AA. A new method for estimating HIV incidence from a single cross-sectional survey. PLoS One 2020; 15:e0237221.
26. Mooney AC, Campbell CK, Ratlhagana MJ, Grignon JS, Mazibuko S, Agnew E, et al. Beyond social desirability bias: investigating inconsistencies in self-reported HIV testing and treatment behaviors among HIV-positive adults in North West province, South Africa. AIDS Behav 2018; 22:2368–2379.
27. Eswatini GotKo. Swaziland HIV incidence measurement survey 2 (SHIMS2) 2016–2017. Final report. Government of the Kingdom of Eswatini, Mbabane. April 2019.
28. Ministry of Health M. Malawi population-based HIV impact assessment (MPHIA) 2015–2016: Final report. Ministry of Health, Lilongwe. October 2018.
29. Tanzania Commission for AIDS (TACAIDS), (ZAC) ZAC. Tanzania HIV impact survey (THIS) 2016–2017: Final report. Tanzania, Dar es Salaam. December 2018.
30. Ministry of Health Z. Zambia population-based HIV impact assessment (ZAMPHIA) 2016: Final report. Ministry of Health, Lusaka. February 2019.
31. Patrick EM. HIV-specific legislation in sub-Saharan Africa: a comprehensive human rights analysis. Afr Hum Rights Law J 2015; 15:224–262.
32. Soni N, Giguère K, Boily M, Fogel MJ, Maheu-Giroux M, Dimitrov D, et al. Under-reporting of known HIV-positive status among people living with HIV: a systematic review and meta-analysis. Available at SSRN: or

Bayesian latent class; HIV disclosure; HIV/AIDS; self-report; sensitivity; testing behaviors

Supplemental Digital Content

Copyright © 2020 Wolters Kluwer Health, Inc. All rights reserved.