Skip to main content
  • Loading metrics

Estimating HIV incidence over a decade in Zimbabwe: A comparison of the catalytic and Farrington models

  • Rutendo Beauty Birri Makota ,

    Roles Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Project administration, Software, Validation, Visualization, Writing – original draft, Writing – review & editing

    Affiliation Division of Epidemiology and Biostatistics, School of Public Health, Faculty of Health Sciences, University of the Witwatersrand, Johannesburg, South Africa

  • Eustasius Musenge

    Roles Supervision, Writing – review & editing

    Affiliation Division of Epidemiology and Biostatistics, School of Public Health, Faculty of Health Sciences, University of the Witwatersrand, Johannesburg, South Africa


Over the years, numerous modelling studies have been proposed to estimate HIV incidence. As a result, this study aimed to evaluate two alternative methods for predicting HIV incidence in Zimbabwe between 2005 and 2015. We estimated HIV incidence from seroprevalence data using the catalytic and Farrington-2-parameter models. Data were obtained from 2005–06, 2010–11, and 2015 Zimbabwe Demographic Health Survey (ZDHS). These models were validated at the micro and macro-level using community-based cohort incidence and empirical estimates from UNAIDS EPP/SPECTRUM, respectively. The HIV incidence for the catalytic model was 0.32% (CI: 0.28%, 0.36%), 0.36% (CI: 0.33%, 0.39%), and 0.28% (CI: 0.26%, 0.30%), for the years 2005–06, 2010–11, and 2015, respectively. The HIV incidence for the Farrington model was 0.21% (CI: 0.16%, 0.26%), 0.22% (CI: 0.20%, 0.25%), and 0.19% (CI: 0.16%, 0.22%), for the years 2005–06, 2010–11, and 2015, respectively. According to these findings, the catalytic model estimated a higher HIV incidence rate than the Farrington model. Compared to cohort estimates, the estimates were within the observed 95% confidence interval, with 88% and 75% agreement for the catalytic and Farrington models, respectively. The limits of agreement observed in the Bland-Altman plot were narrow for all plots, indicating that our model estimates were comparable to cohort estimates. Compared to UNAIDS estimates, the catalytic model predicted a progressive increase in HIV incidence for males throughout all survey years. Without a doubt, HIV incidence declined with each subsequent survey year for all models. To improve programmatic and policy decisions in the national HIV response, we recommend the triangulation of multiple methods for incidence estimation and interpretation of results. Multiple estimating approaches should be considered to reduce uncertainty in the estimations from various models.


Understanding the dynamics of HIV epidemics is essential for determining and evaluating strategies for combating HIV [1]. The availability of antiretroviral therapy (ART) for people living with HIV (PLHIV) has significantly increased their lifespan, resulting in a larger population of HIV-positive individuals [2]. This shift from HIV as an acute condition to a chronic one highlights the importance of accurate estimates of HIV incidence and prevalence for effective public health responses [3]. Quantifying the dynamics of HIV epidemics is crucial for evaluating the efficacy of interventions and guiding policy to enhance HIV/AIDS control [1]. HIV prevalence and incidence are two primary indicators used to characterise the dynamics of the HIV epidemic. Obtaining accurate estimates of HIV incidence and prevalence has always been a significant health concern before the introduction of antiretroviral treatment (ART). Additionally, utilising HIV prevalence as an indicator of the epidemic has proved difficult due to differences in the time interval between infection and mortality due to increased ART availability [4]. To get a real-time evaluation of the HIV epidemic, HIV incidence estimation can provide greater insights [5]. Unfortunately, since HIV infection shares similar signs and symptoms with other diseases, individuals who have been recently infected with HIV may remain undiagnosed for significant periods, resulting in failing to detect the HIV incidence [1].

A significant number of modelling studies have been developed and proposed to predict HIV incidence [1, 4, 6, 7]. A few mathematical or statistical modelling studies have been developed in Zimbabwe to estimate HIV incidence. Among the few HIV incidence models done in Zimbabwe, Hargrove et al. used a simple Susceptible-Infected (SI) model to estimate the HIV incidence among perinatal women in Zimbabwe between 1990 and 2009 [8]. Gregson et al. used mortality data to estimate the HIV incidence in Zimbabwe using back-calculation methods [9]. Using population data from a survey done in 2015 under the Zimbabwe Population-Based HIV Impact Assessment Survey (ZIMPHIA 2015–2016), Gonese et al. compared the HIV incidence estimates obtained through recency testing to UNAIDS Spectrum incidence estimates [10]. Lastly, Ali et al. estimated HIV incidence among female sex workers (FSW) using HIV prevalence by age and number of years since started selling sex (YSSS) between 2011–2017 [11]. These methods have revolutionised how HIV incidence is estimated in various contexts and datasets worldwide [4]. However, few comparison studies to improve practical estimating settings have been reported [4, 1214]. Therefore, the primary objective of this study was to employ estimation models that have been used for incidence estimation to predict HIV incidence using seroprevalence data.

Serial prevalence surveys have been proposed as a less expensive and invasive alternative to collecting longitudinal data on a cohort of people to determine disease incidence [15]. However, there are two significant disadvantages to this strategy. The first limitation is that any estimate based on serial prevalence data implicitly assumes disease irreversibility due to a lack of exact data on the duration of the condition. Because HIV is an incurable illness, the first limitation has no bearing on our study. The second limitation will be that there will be differential mortality between the infected and uninfected individuals.

Differential selection occurs when the likelihood of being selected varies between infected and uninfected individuals in a cross-sectional sample, resulting in a skewed incidence estimate. For example, when infected people have a higher mortality rate than uninfected people, differential selection occurs; as a result, diseased individuals are less likely to be included in a cross-sectional sample of the living population, resulting in an underestimate of disease incidence [16]. Given the possibility that the rate of mortality or survival after infection may be a factor in predicting incidence, our primary objective was to investigate two models in this study: the catalytic model, which assumes differential mortality, and the Farrington model, which assumes non-differential mortality, to determine if there are differences in the estimates of these models. The catalytic model has been used to estimate the incidence of HIV using seroprevalence data. In contrast, the Farrington model has been used to estimate the incidence of infectious diseases other than HIV. Hence, our secondary objective was to determine if the Farrington model behaved similarly to the catalytic model previously used to estimate HIV incidence. Lastly, we wanted to validate the estimates obtained from these models at micro and macro-level using community-based cohort incidence and empirical estimates from UNAIDS EPP/SPECTRUM, respectively.

The goal of this strategy was to triangulate results such that these two models may be utilised among the existing models that can be used to estimate HIV incidence using seroprevalence data from the Demographic and Health Survey (DHS). Data from three major household surveys in Zimbabwe from 2005 to 2015 was used to apply the models to ascertain the HIV incidence in the country.


Survey data

Incidence estimation methods in this paper used national household survey data from the Zimbabwe Demographic Health Survey (ZDHS) conducted in 2005–06 [17], 2010–11 [18] and 2015 [19]. The three surveys applied a two-stage random sampling design. The first stage involved selecting enumeration areas (EAs), which allowed specific weights to be assigned in the design. The second stage involved a random sampling of a fixed number of households [1719]. The survey collected data from consenting individuals with age, sex, marital status, and HIV status. Data were collected in all households on females and males aged 0–49 and 0–54 years, respectively. In addition, blood samples were collected from consenting individuals. The finger prick method collected blot spots on filter paper and transported them to a laboratory for testing. Two ELISA tests were performed, with the second being a retest for 5–10% of the negative and all positive tests. A new ELISA or Western Blot test was performed if there were discordant results on the two ELISA tests [1719].

Estimation models

The catalytic and Farrington models were used to predict the HIV incidence in Zimbabwe. The HIV incidence was estimated using the ZDHS seroprevalence data in both models. The catalytic model simultaneously accounted for differential mortality, whereas the Farrington model assumed non-differential mortality, which was of interest in this analysis. The models are described in the sections that follow:

Catalytic model

This paper adapted Mossong et al.’s [20] catalytic model to estimate age-dependent HIV incidence. The catalytic model is described fully elsewhere [9, 20]. A lognormal function was used to fit the incidence function (Eq 1), with one survival parameter and three incidence parameters [20] to age-specific HIV seroprevalence data from the 2005–06, 2010–11 and 2015 ZDHS of individuals aged 15–49 years. In fitting the lognormal function, an offset corresponding to the age of sexual debut was set at 12 years [20]. Homogeneity across time was assumed within the three-time periods considered in this paper, and the chance of survival following HIV infection was assumed to follow a Weibull distribution. The model assumed differential mortality, which was estimated simultaneously with the HIV incidence. 1 Where: λ(a) is the incidence function with parameters mean μ and standard deviation σ, the scale of θ and an offset corresponding to the age of sexual debut fixed at 12 years.

Farrington model

The Farrington model is a non-linear model that has been used to estimate the force-of-infection for many infectious diseases, including measles, mumps, and rubella [15]. The Farrington model is fully described elsewhere [21], but briefly, the model assumes that the force-of-infection increases to a peak in a linear trend and then declines exponentially [22]. Additionally, the model assumes that the Force-of-infection (FOI) begins at zero at birth and reaches a maximum at an age corresponding to the maximum transmission rate among susceptible and infected people. This results in the Farrington-3-parameter model represented in Eq 2: 2 where; the age-specific Force-of-infection (FOI) is represented by λ(a), a is the age in years, and α, β and γ are model constants to be estimated with γ as the long-term residual value of the FOI.

If the residual value is set to zero (i.e. γ = 0, the Force-of-infection declines to zero as age tends to infinity (a ⟶ ∞), resulting in Eq 3, which is a Farrington-2-parameter model: 3

The Farrington model assumed non-differential mortality, which indicates that disease-related mortality did not affect how the observed serological profile was interpreted.

The age-specific incidence function, λ(a), in Eq 3 was estimated using age-specific seroprevalence data from the 2005–06, 2010–11, and 2015 ZDHS of persons aged 15–49 years.

Validation of the Farrington and catalytic models

The Farrington and catalytic models were validated at micro and macro levels. The objective was to determine if our models could estimate HIV incidence at the provincial and national levels. At the micro-level, ZDHS data for Manicaland province was used to estimate the HIV incidence using the Farrington and catalytic models. Estimates obtained from these models were then compared to the observed incidence from a Manicaland Community-Based Cohort. Empirical estimates from the UNAIDS EPP/SPECTRUM were used to validate the Farrington and catalytic models at the national level.


UNAIDS created the Estimation Projection Package (EPP)/SPECTRUM [2325] to estimate HIV prevalence and predictions [4]. The HIV prevalence estimated by EPP will then be used to estimate the HIV incidence using another software called SPECTRUM, which uses the AIDS Impact Model (AIM) module. A summary of the SPECTRUM methodology is as follows:

Let the number of HIV infections, the total number of adults in the population and the HIV prevalence of individuals aged a at time t be denoted by Ia,t, N a,t and Pa,t, respectively. Therefore; 4 and; 5 where;

λa,t is the number of new HIV infections in the year t minus the number of survived infections from the year t-1;

is the number of deaths caused by AIDS in year t-1; and

is the number of deaths caused by other reasons other than AIDS in year t-1.

EPP was used to calibrate HIV prevalence data from household surveys using the maximum likelihood method, fitting a simple epidemiological model to surveillance data from prenatal clinics in Zimbabwe. Curves expressing the uncertainty in the prevalence data were created using Bayesian melding [26]. Finally, using assumptions about the average survival of HIV-infected people and the impact of ART on survival, adult HIV incidence was computed from HIV prevalence through time.

Community-based cohort study–Manicaland

HIV incidence rates from community-based cohort studies in Zimbabwe from 2005 to 2015 were abstracted from published papers and conferences. Data from a community-based cohort study in Manicaland HIV/STD Prevention Project in Zimbabwe under the Alpha networks were used [27]. In addition, the Manicaland cohort HIV incidence data were taken from the supplemental material of research conducted by Hallet et al. [28]. The Manicaland cohort was a “closed cohort” in which those who joined within a certain time period, were not seen at a follow-up, or were not confirmed to have died were excluded.

Ethical approval and consent to participate

This work was granted ethical clearance by the University of Witwatersrand’s Human Research Ethics Committee (Medical) (No. M151154). Application to the Measure DHS program, granted on May 16, 2017, yielded the dataset utilized in this investigation. The DHS Program is permitted to provide free survey data files to qualified academic researchers. Registration was required for access to data [1719].


The ZDHS prevalence data from 2005–06, 2010–11, and 2015 served as the primary input data to inform indirect model-based incidence estimates. Fig 1 displays the changes in HIV prevalence by age group, sex, and survey year. Generally, the HIV prevalence was higher in females than in males for both youth and adults. In addition, the HIV prevalence was higher for adults (25–49 years) than for youth (15–24 years) for both sexes. The HIV prevalence also declines with each successive survey year for both sexes.

Fig 1.

Prevalence trends by age group and survey year for females (left) and males (right).

HIV incidence for the two models

Table 1 shows the annual HIV incidence for each modelling method stratified by sex. The HIV incidence for the catalytic model was 0.32% (CI: 0.28%, 0.36%), 0.35% (CI: 0.33%, 0.38%) and 0.28% (CI: 0.25%, 0.30%) for the year 2005–06, 2010–11, and 2015, respectively. The HIV incidence for the Farrington-2-parameter model was 0.21% (CI: 0.16%, 0.26%), 0.22% (CI: 0.20%, 0.25%) and 0.19% (CI: 0.16%, 0.22%) for the year 2005–06, 2010–11, and 2015, respectively. Based on these results, HIV incidence estimates were higher for the catalytic model than for the Farrington model. Table 1 also shows that HIV incidence increased from 2005–06 to 2010–11 and decreased in 2015 for both the catalytic and the Farrington-2-parameter models.

Table 1. Overall annual HIV incidence for the two models by year of survey.

As seen in Fig 2, the Farrington model underestimates HIV incidence compared to the catalytic model. Rather than the typical pattern observed in HIV incidence estimates for females, in which the incidence increases to a peak and then declines, HIV incidence estimates for males in 2010–11 and 2015 increased gradually without declining for the catalytic model. According to the Akaike Information Criteria (AIC), the catalytic model was the best fit for the survey data.

Fig 2. HIV incidence for males and females based on the catalytic and Farrington models.

Models’ validation using a cohort study

We obtained Manicaland cohort HIV measured incidence for the years 2005 and 2010. Therefore, comparisons were made based on these two available cohort years, as shown in Fig 3.

Fig 3. HIV incidence estimates based on the cohort, catalytic and Farrington models.

Table 2 summarises Manicaland’s estimated and observed incidence in various age groups. Almost all estimations fall within the observed measurements’ 95% confidence range. Both models performed well overall since the pattern and incidence are almost identical to the observed values, as shown in Table 2 and Fig 3. Table 2 shows that the estimations are within the 95% confidence interval of the observed, with 88% and 75% agreement scores for the catalytic and Farrington models.

Table 2. Measurements and estimates of incidence using Manicaland province data.

The agreement between observed (cohort) and estimated (Farrington/catalytic) incidence was assessed using Bland-Altman plots, as illustrated in Fig 4. The plots give two critical pieces of information: the average of all differences determined using the t-test and referred to as bias and the 95% limit of agreement. The limits of agreement shown in Fig 4 are narrow for all plots, indicating that our model estimates were comparable to cohort estimates. However, a few outlier values were observed in the 2005–06 Farrington females model, the 2010–11 catalytic females model, and the 2005–06 Farrington males model.

Fig 4. Bland-Altman plots demonstrating agreement between estimated incidence (Farrington and catalytic models) and observed incidence.

Models validation using UNAIDS EPP estimates

Fig 5 illustrates the estimates of HIV incidence using three models: catalytic, Farrington, and EPP/SPECTRUM. From around 15 to 25 years, females had a higher HIV incidence than men; subsequently, the pattern reversed, with males having a higher HIV incidence. It is worth noting how the catalytic model predicts a progressive increase in HIV incidence for males throughout all survey years. HIV incidence declined with each subsequent survey year for all models. According to Fig 5, the Farrington model appeared more comparable to the UNAIDS EPP/SPECTRUM estimations.

Fig 5. HIV incidence for males and females based on the catalytic, Farrington and EPP/SPECTRUM estimates.


Given the high expense of directly estimating HIV incidence using the “gold standard,” which requires long-term follow-up [29], indirect approaches based on mathematical or statistical methods have become cost-effective alternatives [4]. Therefore, we examined two models in this research that estimated age-specific HIV incidence using age-specific seroprevalence data. The objectives of this study were to estimate HIV incidence using the catalytic and Farrington models and to test the validity of these estimates at the national and sub-national levels.

The results indicate that HIV incidence estimates from the two models are substantially comparable in estimates and trends; nevertheless, caution should be exercised when interpreting the results due to the models’ different assumptions. The catalytic model best suited the data between the two models employed to predict HIV incidence, evidenced by its low AIC values. The Bland-Altman plots gave an approach to investigate the extent of agreement between the cohort HIV incidence estimates and the two models (catalytic and Farrington). Comparing the catalytic and Farrington models to the cohort observed estimates provides evidence that the two models can be used as part of the models that can be relied on in reporting the subnational HIV estimates in Zimbabwe. Overall, there was reasonable agreement between the two models and the cohort estimates, however, a few discrepancies were seen when some of the data points were not included in the 95% limits of agreement.

Although at the national level comparison, the trends of the two models seemed to be similar to the EPP/SPECTRUM trends, it should be noted that actual HIV estimates fluctuated over time. For example, we noticed that in 2005, the HIV estimates for the two models were underestimated as compared to the EPP/SPECTRUM model however, as we move with time, the inverse becomes true as the two models now overestimate the HIV incidence as compared to the EPP/SPECTRUM estimates. This may be accounted for by the changes that have been made to the EPP/SPECTRUM model over the years as a way to improve its estimates.

Despite considerable model assumption disparities, two mathematical models produced similar HIV incidence levels and trend estimates. Additionally, the models differed in their assumptions about death, with the catalytic model assuming differential mortality and the Farrington model assuming uniform mortality. As seen in Fig 2, the Farrington model generates lower HIV incidence estimates, possibly due to the differences in mortality assumptions; however, the differences seem minimal.

Estimating HIV incidence from serial age-specific cross-sectional prevalence data may be subject to several potential biases, including the quality of the age data, the comparability of the various survey populations, and changes over time in factors affecting HIV prevalence, such as socioeconomic status, migration, and fertility [30]. We believe that these possible biases are negligible in our investigation. This is also corroborated by research conducted by Batter et al. [30], who reported that the effect on mortality in their study was negligible. Other researchers, such as Bucyendore et al. [31] and Miotti et al. [32], utilized HIV prevalence data from surveys of childbearing women without adjusting for mortality or fertility impacts. Although mortality impacts of HIV infection are difficult to measure, they should ideally be included in HIV monitoring systems.

While estimating HIV incidence across age groups is challenging in the era of ART programs, it is critical for tracking changes in the age-specific pattern of HIV prevalence. Our findings imply that combining various approaches for estimating HIV incidence in the same population can aid in generating solid estimates of national HIV incidence levels and trends. This has been practised in many countries, for example, a study by Kim et al. [33] estimated HIV incidence among adults in Kenya and Uganda by comparing multiple methods, and they recommended that triangulation of methods was needed to determine the best supported estimates of incidence to guide programs. In Zimbabwe, Gonese et al. [10] validated the HIV incidence obtained through recency testing with the EPP/SPECTRUM estimates which are used in Zimbabwe by policymakers. In this study, unlike Gonese et al. [10] who used recency testing to estimate HIV incidence, we used statistical models to estimate HIV incidence.

To put the results of the two models into context, we note that young women between 2005 and 2015 had the highest risk of HIV incidence. However, our findings reveal a trend toward a decline in the HIV incidence of young women over time. Similarly, men over 30 years old are at an increased risk of HIV infection despite a gradual decline in recent years. Our findings corroborate the issue of age mixing in Zimbabwe [34, 35] and other sub-Saharan countries [3639].


Our study provides valuable insights into the trends in HIV prevalence and incidence among different age groups in Zimbabwe. The catalytic model outperformed the Farrington model in estimating HIV incidence, particularly in capturing the progressive increase in incidence observed in males. The observed agreement between the estimated and observed incidence values, as depicted in the Bland-Altman plots, further validates the reliability of our model-based estimates. However, a few outlier values suggest the need for caution when interpreting the results.

In conclusion, our findings contribute to the understanding of HIV dynamics in Zimbabwe and highlight the importance of utilising robust statistical models to estimate incidence and prevalence accurately. Further research is warranted to explore the underlying factors driving the observed differences in HIV incidence patterns between males and females.

Lastly, to improve programmatic and policy decisions in the national HIV response, we recommend the triangulation of multiple methods for incidence estimation and interpretation of results. Multiple estimating approaches should be considered to reduce uncertainty in the estimations from various models.


  1. 1. 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(8):e0237221. pmid:32785257
  2. 2. Busza J, Dauya E, Bandason T, Simms V, Chikwari CD, Makamba M, et al. The role of community health workers in improving HIV treatment outcomes in children: lessons learned from the ZENITH trial in Zimbabwe. Health policy and planning. 2018;33(3):328–34. pmid:29309578
  3. 3. Mahomva A, Greby S, Dube S, Mugurungi O, Hargrove J, Rosen D, et al. HIV prevalence and trends from data in Zimbabwe, 1997–2004. Sexually transmitted infections. 2006;82(suppl 1):i42–i7. pmid:16581759
  4. 4. Sun X, Nishiura H, Xiao Y. Modeling methods for estimating HIV incidence: a mathematical review. Theoretical Biology and Medical Modelling. 2020;17(1):1–14. pmid:31964392
  5. 5. Rehle T, Johnson L, Hallett T, Mahy M, Kim A, Odido H, et al. A comparison of South African national HIV incidence estimates: a critical appraisal of different methods. PloS one. 2015;10(7):e0133255. pmid:26230949
  6. 6. Artzrouni M, Brown T, Feeney G, Garnett G, Ghys P, Grassly N, et al. Improved methods and assumptions for estimation of the HIV/AIDS epidemic and its impact: Recommendations of the UNAIDS Reference Group on Estimates, Modelling and Projections. AIDs. 2002;16(9).
  7. 7. Rajan SS, Sokal D. A simpler tool for estimation of HIV incidence from cross-sectional, age-specific prevalence data. Journal of Epidemiology & Community Health. 2011;65(2):111–5. pmid:20558420
  8. 8. Hargrove JW, Humphrey JH, Mahomva A, Williams BG, Chidawanyika H, Mutasa K, et al. Declining HIV prevalence and incidence in perinatal women in Harare, Zimbabwe. Epidemics. 2011;3(2):88–94. pmid:21624779
  9. 9. Gregson S, Donnelly CA, Parker CG, Anderson RM. Demographic approaches to the estimation of incidence of HIV-1 infection among adults from age-specific prevalence data in stable endemic conditions. AIDS (London, England). 1996;10(14):1689–97. pmid:8970690
  10. 10. Gonese E, Musuka G, Ruangtragool L, Hakim A, Parekh B, Dobbs T, et al. Comparison of HIV Incidence in the Zimbabwe Population-Based HIV Impact Assessment Survey (2015–2016) with Modeled Estimates: Progress Toward Epidemic Control. AIDS research and human retroviruses. 2020;36(8):656–62. pmid:32498542
  11. 11. Ali MS, Wit MD, Chabata ST, Magutshwa S, Musemburi S, Dirawo J, et al. Estimation of HIV incidence from analysis of HIV prevalence patterns among female sex workers in Zimbabwe. AIDS. 2022;36(8):1141–50. pmid:35170527
  12. 12. Hallett TB. Estimating the HIV incidence rate–recent and future developments. Current Opinion in HIV and AIDS. 2011;6(2):102. pmid:21505383
  13. 13. Brookmeyer R. Measuring the HIV/AIDS epidemic: approaches and challenges. Epidemiologic reviews. 2010;32(1):26–37.
  14. 14. Europe WGoEoHPi. HIV in hiding: methods and data requirements for the estimation of the number of people living with undiagnosed HIV. Aids. 2011;25(8):1017–23.
  15. 15. Ades AE, Nokes DJ. Modeling Age- and Time-Specific Incidence from Seroprevalence: Toxoplasmosis. American Journal of Epidemiology. 1993;137(9):1022–34. pmid:8317447
  16. 16. Marschner IC. A method for assessing age-time disease incidence using serial prevalence data. Biometrics. 1997:1384–98. pmid:9423255
  17. 17. Zimbabwe Central Statistical Office, Macro International. Zimbabwe Demographic and Health Survey 2005–06. Calverton, Maryland, USA: Central Statistical Office/Zimbabwe and Macro International, 2007.
  18. 18. Zimbabwe National Statistics Agency—ZIMSTAT, ICF International. Zimbabwe Demographic and Health Survey 2010–11. Calverton, Maryland, USA: ZIMSTAT and ICF International, 2012.
  19. 19. Zimbabwe National Statistics Agency, ICF International. Zimbabwe Demographic and Health Survey 2015: Final Report. Rockville, Maryland, USA: Zimbabwe National Statistics Agency (ZIMSTAT) and ICF International, 2016.
  20. 20. Mossong J, Grapsa E, Tanser F, Bärnighausen T, Newell M-L. Modelling HIV incidence and survival from age-specific seroprevalence after antiretroviral treatment scale-up in rural South Africa. AIDS (London, England). 2013;27(15):2471. pmid:23842131
  21. 21. Hens N, Shkedy Z, Aerts M, Faes C, Van Damme P, Beutels P. Modeling infectious disease parameters based on serological and social contact data: a modern statistical perspective: Springer Science & Business Media; 2012.
  22. 22. Hens N, Aerts M, Shkedy Z, Theeten H, Van Damme P, Beutels P. Modelling multisera data: the estimation of new joint and conditional epidemiological parameters. Statistics in medicine. 2008;27(14):2651–64. pmid:17972342
  23. 23. Case KK, Johnson LF, Mahy M, Marsh K, Supervie V, Eaton JW. Summarizing the results and methods of the 2019 Joint United Nations Programme on HIV/AIDS HIV estimates. AIDS (London, England). 2019;33(Suppl 3):S197. pmid:31725433
  24. 24. Mahy M, Brown T, Stover J, Walker N, Stanecki K, Kirungi W, et al. Producing HIV estimates: from global advocacy to country planning and impact measurement. Global health action. 2017;10(sup1):1291169. pmid:28532304
  25. 25. Silhol R, Gregson S, Nyamukapa C, Mhangara M, Dzangare J, Gonese E, et al. Empirical validation of the UNAIDS Spectrum model for subnational HIV estimates: case-study of children and adults in Manicaland, Zimbabwe. Aids. 2017;31(1):S41–S50. pmid:28296799
  26. 26. Brown T, Le Bao JWE, Hogan DR, Mahy M, Marsh K, Mathers BM, et al. Improvements in prevalence trend fitting and incidence estimation in EPP 2013. AIDS (London, England). 2014;28(4):S415. pmid:25406747
  27. 27. Gregson S, Garnett GP, Nyamukapa CA, Hallett TB, Lewis JJ, Mason PR, et al. HIV decline associated with behavior change in eastern Zimbabwe. Science. 2006;311(5761):664–6. pmid:16456081
  28. 28. Hallett TB, Zaba B, Todd J, Lopman B, Mwita W, Biraro S, et al. Estimating incidence from prevalence in generalised HIV epidemics: methods and validation. PLoS Med. 2008;5(4):e80. pmid:18590346
  29. 29. Padian NS, McLOY SI, Balkus JE, Wasserheit JN. Weighing the gold in the gold standard: challenges in HIV prevention research. AIDS (London, England). 2010;24(5):621. pmid:20179575
  30. 30. Batter V, Matela B, Nsuami M, Manzila T, Kamenga M, Behets F, et al. High HIV-1 incidence in young women masked by stable overall seroprevalence among childbearing women in Kinshasa, Zaïre: estimating incidence from serial seroprevalence data. AIDS (London, England). 1994;8(6):811–7.
  31. 31. Bucyendore A, Van de Perre P, Karita E, Nziyumvira A, Sow I, Fox E. Estimating the seroincidence of HIV-1 in the general adult population in Kigali, Rwanda. AIDS (London, England). 1993;7(2):275–7. pmid:8466692
  32. 32. Miotti PG, Dallabetta G, Ndovi E, Liomba G, Saah AJ, Chiphangwi J. HIV-1 and pregnant women: associated factors, prevalence, estimate of incidence and role in fetal wastage in central Africa. AIDS (London, England). 1990;4(8):733–6. pmid:2261131
  33. 33. Kim AA, Hallett T, Stover J, Gouws E, Musinguzi J, Mureithi PK, et al. Estimating HIV incidence among adults in Kenya and Uganda: a systematic comparison of multiple methods. PloS one. 2011;6(3):e17535. pmid:21408182
  34. 34. Schaefer R, Gregson S, Eaton JW, Mugurungi O, Rhead R, Takaruza A, et al. Age-disparate relationships and HIV incidence in adolescent girls and young women: evidence from Zimbabwe. AIDS (London, England). 2017;31(10):1461. pmid:28426534
  35. 35. Mapingure MP, Msuya S, Kurewa NE, Munjoma MW, Sam N, Chirenje MZ, et al. Sexual behaviour does not reflect HIV-1 prevalence differences: a comparison study of Zimbabwe and Tanzania. Journal of the International AIDS Society. 2010;13(1):1–9.
  36. 36. Tanser F, Bärnighausen T, Grapsa E, Zaidi J, Newell M-L. High coverage of ART associated with decline in risk of HIV acquisition in rural KwaZulu-Natal, South Africa. Science. 2013;339(6122):966–71.
  37. 37. Bärnighausen T, Tanser F, Gqwede Z, Mbizana C, Herbst K, Newell M-L. High HIV incidence in a community with high HIV prevalence in rural South Africa: findings from a prospective population-based study. Aids. 2008;22(1):139–44. pmid:18090402
  38. 38. Ott MQ, Bärnighausen T, Tanser F, Lurie MN, Newell M-L. Age-gaps in sexual partnerships: seeing beyond ‘sugar daddies’. AIDS (London, England). 2011;25(6):861.
  39. 39. Chapman R, White R, Shafer L, Pettifor A, Mugurungi O, Ross D, et al. Do behavioural differences help to explain variations in HIV prevalence in adolescents in sub‐Saharan Africa? Tropical Medicine & International Health. 2010;15(5):554–66.