The MaxART project is funded by the Dutch Postcode Lottery’s Dream Fund. The project pursues the dream of reaching all people in Swaziland who are in need of treatment with an ultimate goal of preparing the country for the possibility of ending the HIV epidemic. MaxART comprises of a unique package of interventions aimed at addressing the remaining barriers to HIV testing, care and treatment and further strengthening the collective efforts of the many involved programmes and partners in the country. Examples of these interventions are:  health days that focus on hard to reach groups like men and teenagers; strengthening rural laboratory services by rolling out point-of-care CD4 testing; mobilizing communities by increasing involvement, leadership, knowledge and awareness; and text messages to remind patients about clinic appointments.  The project also supports a number of existing interventions and systems strengthening activities aimed at improving the health of the people of Swaziland.

SACEMA’s mandate within the MaxART consortium concerns various modelling and analysis activities. We have developed a dynamic, mathematical model to simulate the expected impact of a nationwide Treatment as Prevention (TasP) programme on HIV incidence. The model was calibrated to epidemiological and programmatic data for Swaziland. In addition to the dynamic model, an excel spreadsheet was created to provide insight into the minimally required sample size of an implementation study that will pilot treatment access to all those who test HIV-positive in selected locations. The implementation study will assess the feasibility, acceptability, and scalability of using treatment not only to benefit the health of those in need, but also to prevent new HIV infections.

The implementation study will use Swaziland’s currently used monitoring and evaluation systems as far as possible. However, as additional data will need to be captured, modifications to these systems may be necessary. With the statistical analysis of the data generated during the implementation study in the back of our mind, we are working with the monitoring and evaluation specialists at the Swaziland MoH to ensure that all variables necessary for the analysis will be recorded in the national data capturing and management systems.

Furthermore, we are providing technical support in the statistical analysis of the PHDP survey. This survey was conducted last year by SWANNEPHA and GNP+, and documents the experiences of people living with HIV in Swaziland in terms of Positive Health, Dignity, and Prevention in the context of national scale-up of HIV testing, care and treatment.

Lastly, SACEMA is involved in revisiting the currently used methods to estimate the coverage of HIV treatment in Swaziland, together with our MaxART partners. ART coverage crucially depends on the number of people in need of HIV treatment and the subset of people that are actually receiving it. ART coverage estimation requires aggregating data from all facilities that provide ART in Swaziland and running epidemiological models of HIV transmission and HIV treatment in Swaziland to infer the fraction of HIV-positive people that are in need of HIV treatment. The data management process may introduce bias during the steps of capturing, transfer, cleaning and merging of data. Furthermore, some of the demographic, immunological and programmatic assumptions made by the epidemiological model may be at odds with reality, and hence the model-based estimate of the fraction of HIV-positive people that is in need of ART, may be biased as well. By carefully scrutinising the data and model input, we aim to arrive at updated ART coverage estimates based on a transparent, validated method.

It is of note, that a considerable share of South Africa’s tuberculosis burden affects those people who have previously been treated for tuberculosis – many of them successfully. In 2011, almost one out of every eight notified tuberculosis cases had a history of previous treatment (1). The number of re-treatment cases after treatment success was more than seven times higher than the number of cases after failure of treatment and almost four times higher than the number of cases that had defaulted, i.e. that interrupted treatment for at least two consecutive months. Re-treatment tuberculosis was recently shown to contribute significantly to both HIV-associated and non-HIV associated tuberculosis in the City of Cape Town (2).

Studying the rate of re-treatment for smear positive tuberculosis

In this context, we recently published the results of a retrospective cohort study that was conducted using tuberculosis treatment register data from two communities in suburban Cape Town (3). The purpose of this study was to investigate, among tuberculosis patients who had either successfully completed or defaulted from their treatment, the rate of subsequent re-treatment for smear-positive tuberculosis. We further aimed to investigate, among patients who did not complete their treatment, whether sputum conversion prior to defaulting and the duration of treatment were associated with smear-positive tuberculosis re-treatment.

In this treatment register-based cohort study, we included all (index) treatment episodes of sputum smear-positive tuberculosis cases who had been treated for smear-positive tuberculosis in the communities between 1996 and 2008 and who had either completed or defaulted from this treatment episode. A probability record linkage algorithm was used to link subsequent treatment episodes recorded in the treatment register to individual persons. The rate of subsequent re-treatment for sputum smear-positive tuberculosis was measured for both groups of patients.

High risk of smear-positive tuberculosis in defaulters

We show in the published study that the hazard rate of re-treatment for smear-positive tuberculosis was between 3- and 5.26-times higher in tuberculosis cases who had defaulted from treatment compared to successfully treated cases. By the end of the second year after the date when the patient had defaulted, 27.9% had already experienced a re-treatment episode for documented sputum smear-positive tuberculosis compared to 5.8% after treatment success.

Among treatment defaulters, the rate of smear-positive tuberculosis re-treatment was depended on whether the patient had converted to smear-negative at the end of the second month of treatment. Furthermore, there was an inverse linear relationship between the duration of the first treatment episode until the date of default and the rate of subsequent re-treatment for smear-positive tuberculosis. Both age at the time of the index treatment episode and the smear-grade, i.e. the density of acid-fast bacilli found in the diagnostic sputum, were positively associated with subsequent re-treatment, independent of what the later treatment outcome had been.     

To our knowledge, this the first study demonstrating a high risk of smear-positive tuberculosis in patients who had been non-adherent to a full course of treatment before. Our results suggest that tuberculosis was initially contained in the majority of defaulters but worsened again soon after stopping the treatment. Although the rate of re-treatment was substantially higher among defaulters, cases after treatment success account for the vast majority of smear-positive re-treatment cases, due to the fact that far more tuberculosis cases were successfully treated than had defaulted, and the rate of re-treatment after success was lower in the first two years but remained constantly high in the following years.

True rate of smear-positive tuberculosis might even be higher

We believe that our rates of re-treatment underestimate the true rate of smear-positive tuberculosis among previously treated tuberculosis cases in the communities, because we were unable to account for tuberculosis cases who experienced a subsequent episode of smear-positive tuberculosis after treatment success or default without being diagnosed or being diagnosed but without being treated. Furthermore, the denominator in our rate estimate included person-years at risk for previously treated tuberculosis cases who might have moved away from the area or who might have died. Therefore we expect that the true rate of smear-positive tuberculosis among previously treated tuberculosis cases is even higher. Our study is also limited by the fact that we were unable to control for other known risk factors of disease recurrence such as residual lung cavitation at the end of the first treatment episode, greater area of involved lung tissue and positive sputum culture two month after the start of the first treatment episode (4). In addition, we were unable to determine the effect of HIV co-infection, a known risk factors for tuberculosis recurrence (5), on the risk of smear-positive tuberculosis among previously treated cases. The proportion of tuberculosis cases who were co-infected with HIV in our setting was relatively small compared to other areas in the Western Cape Province.

Implications for tuberculosis control

The results of our study have implications for tuberculosis control as they emphasize the need to ensure treatment adherence in order to prevent subsequent smear-positive tuberculosis and, possibly, related adverse health effects such as chronic pulmonary impairment (6), death (7), and acquisition of drug-resistance (8). High rates of smear-positive tuberculosis after previous treatment raises questions about the extent to which previously treated tuberculosis cases contribute to on-going tuberculosis transmission in the communities. Den Boon et al. had published the results of a tuberculosis prevalence survey conducted in the same communities (9). During this survey, 18 smear-positive tuberculosis cases were detected in a randomly selected household sample, 10 (56%) of whom had had a history of previous treatment. If these cases were representative of all infectious source cases in the communities, than it is reasonable to conclude that previously treated tuberculosis cases contribute significantly to transmission. This study and our results suggest that disease recurrence might play an important role for the burden of tuberculosis in the communities.

The study presented here forms part of a larger on-going research project in which we investigate the burden and underlying causes of recurrent tuberculosis and the risk of progression to multidrug-resistant (MDR) tuberculosis among patients with multiple episodes of tuberculosis (treatment).

In spite of the combined efforts of national governments and their global partners, such as the President’s Emergency Plan for AIDS Relief (PEPFAR), The Global Fund, the Clinton Health Access Initiative (CHAI), and international non-governmental organizations (NGOs) such as Médecine sans Frontières (MSF) and many others, the incidence of HIV and TB remain very high. Recent modelling studies  (1,2) have shown that modest reductions in HIV incidence are possible with high coverage of anti-retroviral therapy for people whose CD4+ cell count is below 350 cells/µL, but substantial reductions are only possible if people are tested at intervals of about 1 year and started on treatment as soon as they are found to be infected with HIV. This approach, termed “treatment as prevention” (TasP) or “test and treat” (TnT), in which people are started on ART as soon as they are found to be HIV-positive, irrespective of their CD4+ cell counts, is now recommended by the International AIDS Society (IAS) as well as the Department of Health and Social Services (DHSS) in the United States and is likely to become the treatment protocol of choice in all countries.

The recent Declaration and Code of Conduct on TB in the Mining sector signed by the SADC Ministers of Health are a recognition of the close link between mining and TB in sub-Saharan Africa (3).  Not only does silicosis increase the risk of TB by about 3 times on average, but the working conditions and the living conditions in the mine hostels, especially around gold mines, are such as to greatly enhance the spread of TB (4). 

HIV infection drives the epidemic of TB partly because people living with HIV are more susceptible to being infected with TB and also because those with a latent TB infection, which in some places may be up to 80% of the adult population, are more likely to break down to active disease.

For most of the past century the system of oscillating (going to and from) labour migration, especially to the gold mines in South Africa, has helped to spread TB throughout southern Africa and it now helps to spread HIV (3,5). However, there is also evidence that other forms of migrant and seasonal labour have contributed to the spread of both infections. A behavioural study carried out on commercial agricultural farms in Mpumalanga and Limpopo  (6) where 31% of the workers were migrants, found a prevalence of HIV of 40%, a case-notification rate of TB of 6% per annum with 59% of TB cases being HIV positive (2.3, 9 and 1.5 times the respective national averages).

The impact of transport corridors and migrant labour on the prevalence of HIV

In a study on the impact of migrant labour in the mines in South Africa on the burden of HIV and TB in Mozambique (7) a close correlation was found between the geographical distribution of HIV and TB and patterns of migrant labour. Mozambique can be divided into 3 regions, each of which has a different burden of disease. (National data). The northern region (HIV prevalence 6%, TB incidence of 99 cases per 100,000 people per year) is relatively isolated, does not have major transport corridors passing through it and is not used for recruiting agricultural or mine workers. The central region (HIV prevalence 12%, TB incidence of 178 cases per 100,000 people per year) has the main transport corridors from Zimbabwe and Zambia to the coastal ports of Beira and Maputo. The southern region (HIV prevalence 18%, TB incidence of 327 cases per 100,000 people per year) has the transport corridors into South Africa and the transport corridor to Zimbabwe and Zambia. Records of The Employment Bureau for Africa (TEBA) over the last decade (Personal communication) show 96% of miners recruited in Mozambique for South African mines come from the southern region.

In Figures 1 and 2 we compare the HIV prevalence and the TB notification rates respectively in each of the three geographical regions. If we assume, as a first approximation, that the prevalence in the northern region is the expected prevalence without the presence of major transport corridors, then the difference between the prevalence in the central region and the northern region gives an approximate estimate of the impact of these transport corridors on the prevalence of HIV. Similarly, the difference between the prevalence in the southern region and the central region gives an approximate estimate of the impact of migrant labour on the prevalence of HIV. This is a very conservative estimate, since the transport to Maputo passes through the Central region, and so the burden of infection through the transport routes is likely to fall more heavily on the central region than it is on the Southern region.

Figure 1. HIV Prevalence (%) by region showing contributions of corridors and migrant labour (14) Blue: Baseline without corridors or migrant labour. Red: Transport corridors. Purple: Migrant labour.

Figure 2. TB Incidence rate (per 100,000 per year) by region showing contributions of corridors and migrant labour. Blue: Baseline without corridors or migrant labour. Red: Transport corridors. Purple: Migrant miners.

The patterns of the circulatory migration in the three groups (miners, agricultural workers and truck drivers) differ in the frequency of migration and in the time the migrants spend away from their homes. But in all three cases they spend a lot of time away from home, are relatively well paid, have little to spend their money on when away from home, and have little with which to occupy themselves, do not have the social support and intimacy that they would have when at home and visiting sex workers becomes more attractive (8). Furthermore, many sex workers are also migrants seeking work and money  (9). Finally, the wives or partners of migrant workers who remain in the labour sending areas are themselves often living in poverty, are isolated and without male support and may also engage in risky sexual behaviour (10).

Migration key issue in the regional HIV and TB epidemics

Modelling studies have shown that even if we maintain the same patterns of sexual behaviour the presence or absence of migration can lead to dramatically different outcomes.

HIV can only be transmitted by direct, intimate contact, and sexual transmission of HIV is very inefficient; each person with HIV only infects one other person every one to two years, on average. For HIV to spread over wide areas it must be carried by people who move from one place to another. Without migration the prevalence may reach high levels locally but those infected will die and the epidemic will not spread. Even if the proportion of migrants is quite modest the epidemic can spread over large geographical areas provide the migrants move sufficiently far. Because the gold mines of South Africa have traditionally recruited men from almost all of the countries in southern Africa the countries in the region become epidemiologically one and this may be the main reason why the nine worst affected countries in the world are all in southern Africa.

Although there is a strong link between mining and TB and HIV the spread of both diseases in sub-Saharan Africa depends critically on migration, especially of agricultural workers, truck drivers and mine workers. The problem is not simply that men work in the mines but rather the migrant nature of their employment, their living and social conditions, and in particular the breakdown of family life that is associated with the particular form of oscillating migration in southern Africa (13).  While we have focused on three of the key groups, migrant workers are also employed extensively in the construction industry and as domestic workers and for other casual labour. Although they may not necessarily live in similar confined social conditions, they are isolated from their families and often live in shared accommodation in informal settlements.

Unless a comprehensive and fully coordinated multi-country and multi-sectoral programme is implemented and followed through, we may find that the HIV and TB epidemics are far more resilient than consideration of the epidemics in each country suggests.

As resources are invested in developing more appropriate tests for recent infection, the methods for evaluating and optimising the performance of candidate tests need to be further refined. We have recently explored a range of statistical methods for estimating the mean duration of intection (MDRI) by investigating an unusually good data set from Harare, Zimbabwe where the BED Capture Enzyme Immuno-Assay (BED-CEIA or simply BED assay) was applied (2). In this case, ‘recent infection’ is taken to be an optical density less than a chosen cut-off – indicating non-matured levels of proportion of HIV specific IgG (an important antibody class). An estimate of the MDRI of the assay is thus obtained by estimating the time spent below the optical density cut-off C, for patients who have been HIV positive for at most T , for some pre-defined time T.(1,3) We investigated whether there is an optimum way of estimating the mean recency duration or whether several estimating procedures provide similar answers and in that case whether simple approaches provide adequate answers.  We also ask how estimates of the mean recency duration and incidence are affected by our choice of cut-off and whether these effects differ with our choice of estimation method.

Example MDRI estimation

The MDRI was estimated using BED data from the Zimbabwe Vitamin A for Mothers and Babies (ZVITAMBO) Trial. 14 110 women were tested for HIV postpartum, at 6 weeks, and at 3 monthly intervals from 3 to 24 months. MDRI was estimated using subsets of BED optical density (OD) data from 353 seroconverted women of the original 9 562 HIV negative women.

MDRI (ω_t) was estimated using four different estimation methods:

1. Proportion of recent infections among seroconverters (r/s). Assuming uniformly distributed seroconversion events over the time period [0,T], the MDRI is estimated from the proportion of all seroconverters testing recent according to some OD cut-off.
2. Linear Mixed Model (LMM). A LMM with fixed and random effects is used to model a linear relationship between the OD and time since infection (transformed according to the recommendation of the developer). This model yields a straight line for each woman, from which the estimated time spent in the recent state is obtained by using an inverse prediction technique (4), with the upper limit restricted to T.  Bootstrap techniques were applied to these individual estimates to obtain the final estimate of the MDRI as well as the associated confidence interval.
3. Non-linear Mixed Model (NLMM). A more biologically plausible NLMM is used to model the relationship between the assay OD and the time since infection. Unlike the LMM, this function approaches finite asymptotes for short and long times since infection. For this model, time of seroconversion is assumed to be uniformly distributed between the dates of last negative and first positive HIV tests, and individual recency durations were obtained by using an inverse prediction technique (4), with the maximum set as T.  Markov Chain Monte Carlo (MCMC) methods are applied to obtain a distribution for the MDRI.
4. Survival Analysis (SA). For this method, the time of seroconversion for each individual is approximated by the mid-point between the time of last negative and time of first positive HIV test. An interval for the time to reach the pre-defined OD cut-off is obtained from the data by using the last time point with OD below the cut-off and the first time point with OD above the cut-off. The data are then considered  to be interval censored and Turnbull’s modification of the Product-Limit Estimator is used to obtain  a survival function which, when integrated over [0, T], provides an estimate of MDRI and its corresponding confidence intervals.

The estimates of the MDRI produced by the above four estimation methods are provided in Table 1. The optical density cut-off was fixed at the ‘package insert’ value of 0.8 for all methods, a minimum of two samples per case were required and the maximum allowable time between the last negative and first positive HIV tests was 120 days. This resulted in a sample size of 100 women.

Table 1 Results of MDRI estimation using different methods.

There are no significant differences between the four estimates when the OD cut-off C lies between 0.6 and 1.0. Estimates are fairly insensitive to varying the minimum allowable samples per women between 2 and 4 and varying the maximum number of days between last negative and first positive HIV test between 60 and 180. In all cases, the coefficient of variation (the ratio of the standard error to the mean) is the lowest for the NLMM.

The 12 months post-partum follow-up incidence of 3.46% (95% CI: 3.05%–3.87%) was compared with BED incidence calculated using only women who seroconverted during the year. For all values of C tested, the NLMM estimates of incidence showed the smallest deviation from the follow-up estimate, varying only between 3.23% and 3.50%. For values of C between 0.8 and 1, the estimates of incidence for the NLMM, LMM and SA methods did not differ significantly from each other. The r/s method was not used, since r/s is a constant multiple of the follow-up incidence.

Which estimator and cut-off point to choose?

For a cut-off C = 0.8, there was no significant difference between MDRI estimates, but coefficients of variation (CoVs) were higher for the r/s and SA estimators and lower for regression estimators, LMM and NLMM, which use information on time-dependent increases in OD. NLMM, additionally, uses a biologically plausible function that best fits the data approaching finite asymptotes for small and large times since infection. NLMM estimates of MDRI changed most regularly with C and produced HIV incidence estimates most closely approximating observed follow-up incidence. Regression works less well when only 2-3 BED results are available for all cases: the r/s and SA methods are then preferable.

With respect to the cut-off point: As C (and thus MDRI estimate) decreases, the number of observed recent infections declines, and the CoV of the MDRI estimate increases: as C increases so does the false-recent rate, and there is an increasing risk of violating assumptions of constant incidence. For NLMM, CoVs were 2.8% at C = 0.4 and 1.8% at C = 1.0: the conventional value of C = 0.8 combines acceptable CoV (2.0%), MDRI (~0.5 years) and false recent rate (~5%) estimates.

Way forward

The ZVITAMBO dataset was an extraordinary data set, but usually developers of assays and analysis methods need to work with less ideal data sets. New studies, usually following seroconverters for other reasons, will continue to be occasionally used as sources for additional specimens for analysis of this type. BED per se was once widely used, but because of its limitations is by no means the expected dominant assay for future investigations. Crucially, the current trend is to use multiple biomarkers, each of which has some time evolution within newly infected individuals, and to combine these into a recent/non recent categorisation.

So – moving forward, we need to know how variations in type and quality and quantity of data impact on analyses of this type , so that we can adjudicate the difference in performance of the various statistical approaches. There is indeed a variety of techniques, including systematic benchmarking of methods on simulated data where we can be sure what the ‘real’ answer is. SACEMA is also involved in a number of key collaborations where such methods are being applied to numerous assays on well characterised specimen panels, for example the Consortium for the Evaluation and Performance of HIV Incidence Assays (CEPHIA).

This study predicted the role of dosR-regulon genes as central in the regulation of latency. These genes were predicted to cluster with adaptation, detoxification and virulence genes and several other  genes of unknown or putative functions (Rv3131, Rv0569, Rv2032, Rv2530c, and Rv2694c), some of which were predicted both in the regulation of the stationary and non-replicating phase of the bacteria.

This study furthermore predicted that other genes (Rv1133c, Rv2890c, Rv1177, Rv2710, Rv2532c and Rv0982) also regulate latency. These gene were shown in other studies to be essential for H37Rv growth and in vivo bacterial survival. Through sensitivity analysis of the predicted gene regulatory networks and computational gene deletion experiments, the genes Rv2031, Rv3133c, Rv2032, Rv3131, Rv2530c, Rv2527, Rv1909c, and Rv0569 were identified most potent in disrupting the bacteria latency and dormancy program. Hypothetically, these genes are possible drug targets. Interestingly, Rv3131, a putative NAD(P)H nitroreductase, which is quite central in regulation of the stationary phase was shown in other studies to be essential for bacterial growth, making it an attractive drug target. The predicted genes have functions that are linked to human immune response mechanisms, such as induction of interferon-c (IFN-c) and interleukin-2 (IL-2) (Rv2032, Rv3133c, Rv3131, and Rv2031c), oxidative stress (Rv1909c) and nitrosative stress (Rv3131, Rv3133c, Rv2031c). The latter are associated with nitrogen and oxygen reactive intermediate effector mechanisms of the immune response and were shown in other studies to have the potential to prime the human immune response.

Although these genes are predicted  in this and other studies to be involved in the regulation of TB latency, it must be highlighted that: (i) they are not fully annotated and characterised, (ii) they have not been assigned to any functional class, (iii) no assay information is available on them and (iv) they have not been mapped to any specific metabolic KEGG pathways. Nevertheless, it remains imperative that they should be tested through biological experiments for verification and validation to determine if they can be used for possible latency TB drug design.

This study predicted the latency regulatory mechanism to be dynamic with respect to the stress and time spent under the stress. This suggests that experiments simulating latency have a good chance of closely predicting and explaining what takes place in latency if they are carried out over a long time period.

Models of processing around the transmission event itself may be useful to shed light on questions such as:

• Are risk factors altered moderately or substantially by sexual violence, relative to consensual sex?
• Are men participating in ongoing sexual violence at particularly high risk of acquisition due to penile trauma, leading in turn to a particularly high risk in the subsequent exposure of women during acute infection of these men?
• What is the plausible range of benefit from topical or systemic post exposure prophylaxis for rape survivors, and over what time scale?
• What specific parameters (such as diffusion length scales, time scales, intermediate event counts, cell densities, etc.) might be measurable in live animal models, explants, or culture experiments, which would correspond to factors impacting the above questions?

To demonstrate these ideas, we outline a simple population dynamics-type model (1,2) which was originally developed by Welte and Walwyn (3) in order to support thinking about HIV vaccines, and then discuss how the model’s mathematical form can be reinterpreted in  the context of sexual violence.

Model equations cannot distinguish between any external factors which have the same effect on the model parameters. Hence, a pre-exposure prophylaxis regimen which inhibits reverse transcription could also be seen as reducing viral ‘fertility’; and other active ingredients, or pharmacological actions, could induce changes (perhaps in either direction) in general inflammation of, or immune cell trafficking to, the mucosal interface. Besides sexual violence, a mature transmission model should shed some light on topical and systemic pre-exposure and post-exposure prophylaxis (PreP/PEP), vaccine development (4), hormonal contraception(5,6), treatment as prevention (TasP) (7), cervical ectopy (8), differences in viral replication due to the site of infection (9), and the biological impacts of sexual violence on the risk of HIV transmission and acquisition.

Recent work has shown that the risk of HIV transmission appears to scale non-linearly with viral load (10, 11). At levels below 100,000 RNA copies per ml, the risk of transmission can be predicted using a power law between the two variables (i.e. a linear correlation based on the logarithmically transformed data).  However, at higher viral loads, further increases in RNA copies per ml are not associated with a greater risk of transmission (12, 13, 14).  It has been pointed out that it is more biologically plausible, and a good fit to the data, to view this as a linear increase for smaller viral loads, and asaturation for higher viral loads.

This interpretation is consistent with the view that sexual contacts at low viral loads carry minimal risk, but extra viral load does indeed increase risk in direct proportion to viral load up to the saturation point, beyond which this effect is insignificant relative to other factors.  It is plausible, and disconcerting, that violence-associated risk factors may be a substantial contributor to the risk of infection, if not epidemiologically, then potentially still at the level of the individual.

Our approach, which is rather conventional in the applied mathematical analysis of dynamical systems, and is routinely applied within the growing field of biophysics, has not, as far as we are aware, been systematically applied to the problem of HIV transmission – probably partly because movement of infectious material from person to person during intercourse is not a system readily amenable to controlled manipulation and observation. However, even working within the data-poor regime in which this field finds itself, this paradigm has the potential to assist with hypothesis generation and study design: using these tools, many regimes of diverse systems can be summarised into a far smaller number of formally distinct models, the use of which reduces technical difficulty and facilitates the transfer of intuition. We consider the development of appropriate population-dynamic models of initial infection to be a long term process, requiring new inputs and insights into the very early stages of infection (15).

Rising volume of evidence suggests that HIV viral load (VL) among HIV infected individuals rises if they acquire an infection such as malaria as a consequence of immune system interactions (1). Higher levels of HIV VL indicate an increase in HIV infectiousness in the population, and consequently larger epidemics. We assessed whether the massiveness of the HIV epidemics in SSA can be explained by a higher HIV VL among HIV infected individuals in SSA, potentially as a result of their co-infection with other pathogens. We examined this hypothesis by addressing two research questions (2): 1) Are mean VL levels higher in SSA compared to other regions? 2) Can these higher VLs explain, at least in part, the fulminant HIV transmission dynamics observed in SSA, in contrast to other regions?

We addressed the first question by gathering and analyzing a database of 71,668 VL measurements from 44 cohorts of HIV infected antiretroviral therapy naïve individuals from seven geographic regions. Our findings revealed a striking heterogeneity in HIV VL globally, with SSA, including East and Southern Africa, having the highest HIV VL levels (Figure 1). Compared with North America, adjusted for CD4 count, gender and pregnancy, the estimated mean log10 VL was 0.29 higher (95% confidence interval (CI): 0.11 to 0.47) in West Africa, 0.71 higher (95% CI: 0.48 to 0.93) in East Africa, and 0.74 higher (95% CI: 0.55 to 0.92) in Southern Africa. The estimated mean log10 VL was modestly, but significantly, higher in Asia than in North America [0.14 higher (95% CI: 0.03 to 0.26)], but no difference was seen in mean log10 VL between Europe or South America and North America.

Figure 1: Differences in mean regional HIV-1 plasma RNA viral load (VL) compared to North America. The black marker for South Africa shows the results including bDNA assays, and the red marker shows the results excluding bDNA assays.

We addressed the second question by developing a mathematical model to describe the impact of the regional VL differences on the HIV epidemic trajectory. The model calculates the proportion of incident HIV infections that are directly attributable to the higher VL levels and was applied to a setting representative of a hyper-endemic HIV epidemic in SSA (Kisumu, Kenya). The analysis showed that:

• The proportion of incident HIV infections directly attributable to the VL effect increased with time, reaching 14.4% (95% CI: 7.7% to 20.5%) at the epidemic peak in the mid to late 1990s. The proportion of cumulative incident HIV infections directly attributable to the VL effect was 13.9% by 2010.
• We estimated that in Kisumu (adult population of about 200,000), the VL effect has contributed , including the onward transmission, over 30,000 excess HIV infections from 1980 to 2010 out of a total of approximately 135,000 HIV infections.

In conclusion, community HIV VL appears to be higher in SSA than in other regions, and this could be a central driver of the massive HIV epidemics in this region. The elevated HIV VL in SSA possibly reflects the high burden of co-infections in SSA.

This two-week modelling clinic, mounted in collaboration with the International Clinics on Infectious Disease Dynamics and Data (ICI3D) Program, and the African Institute for Mathematical Sciences (AIMS), will emphasize the use of data in understanding infectious disease dynamics. The Clinic will bring together graduate students, post-doctoral students and researchers from Africa and North America, with the goal of engaging the participants in epidemiological modelling projects that use real data to grapple with practical questions in a meaningful way.

Download additional information and/or an application form. 

The author of the article, Mr Damian Kajunguri, hails from Uganda: he came to South Africa to do the Graduate Diploma in Mathematics at the African Institute of Mathematical Science at Muizenberg.  While there he did a project on estimating extinction probabilities of tsetse populations subjected to various control strategies.  This led to him being offered a SACEMA bursary to do a Masters, which he completed in 2009.  By the time this issue of the SACEMA Quarterly goes to press Damian will have submitted the present work for his PhD thesis with the University of Stellenbosch.  This will mark the first PhD thesis completed in the area of tsetse and trypanosomiasis biology.  We hope, however, that it will not be the last and that the work will be a happy harbinger of things to come.  Thus, at the time of writing we have just been informed that SACEMA has been short-listed for WHO/TDR funding of a project in the area of health and climate change.  The proposed project will focus on modelling the way in which various climate change scenarios might affect the population dynamics of tsetse flies and the trypanosomes that they transmit.  

Increasing populations in Africa drive agricultural development of remote areas, leading to increased contact between marginalised people (particularly women, children, and the elderly) and tsetse-transmitted trypanosomes.  Understanding the likely development of this situation, how it will be affected by various climate change scenarios, and how best to mitigate the resultant disease situation, is a multi-disciplinary problem involving work in entomology, parasitology, climatology, statistics and mathematical modelling.  Such studies are often hampered by the absence of detailed biological data that can be used to assess the dependence of vector and disease on climatic parameters and, therefore, how changes in climate are liable to affect the situation.

The SACEMA-led Consortium for this project is in the happy position of being an exception to the data-poor situation referred to above: we have access to large, long-term, unique archives of data of the type required to address these questions.  These data will be augmented during the study through field studies in Zimbabwe and Tanzania, aimed at understanding the spatiotemporal variability of disease threat and how this is likely to change at different locations and altitudes in the context of climate change.  Field studies will address particularly the problem of the interface between humans and tsetse, and suggest optimal methods of disease control – an area in which the Consortium has a proven track record.  The Consortium will prioritise development of local expertise, particularly through training at the Masters and PhD level.  The Consortium includes workers with unparalleled experience in tsetse and trypanosomiasis field research and control, Africa’s leading climatologists, and mathematicians who have taken a leading role in modelling various aspects of the biology of vector and disease alike.