The MARA ("Mapping Malaria Risk in Africa") climate suitability index was used for determining the country endemicity, i.e. high ≥ 0.75 and low <0.75 10. However, Botswana, Namibia, South Africa, Swaziland, and Zimbabwe, because of their successful vector control programme, were not classified according to this criterion and despite a climate suitability index >0.75 were considered as low rather than high transmission areas 10.

Incidence rates of symptomatic malaria (uncomplicated and severe) by transmission intensity and age group were collected from the available literature 6. In high endemic rural areas the average incidence was estimated at 1.4 malaria episodes/person-year (py) among children <5 years of age, 0.59/py in the 5–14 year old and 0.11/py in adults (≥ 15 year old)10. In low endemic rural areas, incidence was estimated at 0.18/py in children <15 years old and 0.09/py in those aged 15 years and above (Table 1). In high endemic urban areas, the malaria incidence was estimated to be similar to that in low endemic rural areas while in low endemic urban areas, the malaria incidence was estimated at half of the low endemic rural areas 11 (Table 1). In Botswana, Namibia, South Africa, Swaziland, and Zimbabwe, the average incidence was estimated at 0.03/py 10, with the age pattern distribution similar to that of low endemic areas 6. The proportion of urban dwellers was assumed not to vary by malaria endemicity. Urban and rural populations were based on countries' definitions, with no standardization between countries 12.

The best evidence for an association between HIV infection and incidence of symptomatic malaria comes from four longitudinal population-based studies in rural Uganda and Malawi, where malaria is highly endemic. In Uganda, odds ratios of symptomatic malaria in HIV-infected compared to uninfected adults were 6.0, 3.4 and 1.2 for CD4 counts <200/μl, 200–499/μl and ≥ 500/μl, respectively 1. In another study in Uganda, the incidence of symptomatic malaria in HIV-infected individuals varied according to CD4 count and was 0.14/py, 0.093/py and 0.057/py for CD4 count of <200/μL, 200–499/μL and >500/μL, respectively (0.09/py, 0.053/py, and 0.022/py when considering only episodes with more than 2,800 parasites/μL)21. In Malawi, incidence rates of symptomatic malaria in HIV-1 infected adults varied with CD4 count; compared to individuals with CD4 count of ≥ 500, the malaria incidence was 3-fold higher with a CD4 count of 200–499/μL and a 4.4-fold higher with a CD4 count <200/μL 22. Another study from Malawi reports malaria incidence of 3.2/py, 4.9/py and 5.4/py in HIV-1 infected adults with CD4 count of ≥ 400/μL, 200–399/μL and <200/μL, respectively 23. In most of these studies malaria infection was not assessed at the time that CD4 count was done. This is a limitation as malaria itself causes a temporarily and reversible decrease in peripheral CD4 count (by one third) 24. Therefore, the risk for clinical malaria may increase at a higher-than-expected CD4 count. We assumed that, compared to HIV-1 negative individuals, the risk of clinical malaria in HIV-1 infected individuals with a CD4 count <200/μl and 200–499/μL is 5 and 3-fold higher, regardless of endemicity and age. HIV-1 infected with a CD4 count ≥ 500 were expected to have no additional risk.

HIV-1 infection considerably increases the risk of severe malaria in adults 162526. In South Africa, in an area of low (unstable) transmission, malaria-attributable mortality was nearly 7-fold higher in HIV-infected malaria patients. Similar observations have been collected in an ongoing study in a meso-endemic region of Zambia, with the highest risk in HIV-1 infected with low CD4 count (Chalwe, unpublished data). We assumed that, compared to HIV-1 negative individuals, the risk of severe malaria in HIV-1 infected individuals with a CD4 count <200/μl and 200–499/μL is 15 and 6-fold higher, regardless of endemicity and age. HIV-1 infected with a CD4 count ≥ 500 were expected to have no additional risk.

HIV-1 infection tends to increase the mean parasite density, although this effect is largely limited to patients with advanced immune suppression (Table 2). In HIV-1 infected individuals, the geometric mean parasite density in uncomplicated malaria was assumed to be three-fold higher at CD4 count <200/μl and 50% higher at CD4 count 200–499/μL, irrespective of malaria endemicity and patient's age. For severe malaria, because parasite density is already high and few data are available, it was conservatively assumed that parasite density does not vary according to HIV infection.

Point estimates of national HIV-1 prevalence for 2005 are available from the Joint United Nations Programme on HIV/AIDS (UNAIDS) 27. The impact of HIV infection was evaluated separately for urban and rural areas, because of varying intensity of malaria transmission. The urban-to-rural ratios in HIV-1 prevalence were estimated from national household surveys or antenatal clinic surveillance data 2728. For all countries, except Madagascar, it was assumed that 37% and 25% of HIV-infected people have a CD4 count <200/μL and 200–499/μL respectively, a good approximation for the African populations where HIV prevalence has stabilized 6. For Madagascar, where adult HIV prevalence is low but rising, 37% of HIV-infected people were assumed to have a CD4 count <500 μL 6.

Annual parasite biomass was defined as the incidence rate (per person-year) of symptomatic (uncomplicated and severe) episodes multiplied by the total number of parasites circulating in the patient's peripheral blood at the time of the clinical episode (i.e. when parasite density is measured at a health facility). The parasite biomass was estimated for each country by age group (<5, 5–15 and above 15 years), by urban/rural and by low and high endemicity taking into account the HIV-1 prevalence, the incidence rates of uncomplicated and severe (hyperparasitaemic) malaria and their corresponding parasite density. The overall country parasite biomass was computed by summing up the estimations for each category listed above.

To assess the sensitivity of the results, both univariate and multivariate sensitivity analyses were performed. In the univariate sensitivity analysis, the impact of HIV was recalculated for two alternative scenarios, corresponding to the lower and upper bound of the likely range. In the multivariate sensitivity analysis, 95% confidence intervals were calculated using the Monte Carlo simulation in which 1,000 samples were taken from triangular distributions for all parameters 29 and the resulting proportional increase in parasite was calculated. All statistical analyses were performed using Microsoft Excel and the statistical software R version 2.2.1 30.

The HIV-1 epidemic may have increased malaria parasite biomass in sub-Saharan Africa by 17.4%. The highest absolute increases were found in countries with both high malaria and HIV burden (Zambia, Malawi, Mozambique and the Central African Republic). A recently published model 7 found that the largest epidemiological impact occurs when one baseline measure is very high while the other is very low and near its endemic threshold. When both prevalences were very high, the direct epidemiological impact of the interaction seemed to be minimal. It is indeed conceivable that, when both diseases reach an epidemic equilibrium, the epidemiological impact of the interaction will be minimal while the disease interaction should be maximal. Nevertheless, when both diseases are at epidemic equilibrium, there may be an indirect but important epidemiological impact such as the increased parasite biomass, higher biodiversity and higher probability of de novo resistance mutations emerging.

The model presented has several limitations. Pregnant women were not considered in the model as the parasite biomass in this high-risk group and its changes according to the HIV-1 infection could not be estimated, partly because of the sequestration of parasites in the placenta. Considering that pregnant women are particularly vulnerable to malaria infection and that such vulnerability is increased by HIV-1 infection, their exclusion may have led to an underestimation of the actual impact of HIV-1 infection at population level.

Moreover, the possible impact of HIV-1 infection on parasite density in asymptomatic individuals was not considered. This simplification may be justified because, due to the logarithmic distribution of malaria parasites in human populations, asymptomatic carriers, despite outnumbering the clinical cases, carry only a small proportion (probably less then 1/1000) of the world's potentially transmissible parasites 3132. In addition, asymptomatic infections most often are not treated, reducing the risk of selecting resistant parasites. In absence of drug pressure, mutant resistant parasites may be "less fit" and possibly less transmitted. However, once resistance has emerged and spread above a certain threshold, asymptomatic infections could play a substantial role for its further spread, especially in high endemic areas, and this effect could be enhanced by a higher prevalence of asymptomatic infection in HIV co-infected individuals. The spread of these resistant strains should be calculated by a stochastic model where asymptomatic carriers most probably play an important role.

An additional limitation of the model presented is the assumption that the effect of the HIV-1 infection on parasite density is constant throughout the clinical episode. This may not be true as the actual parasite density may vary throughout the clinical episode.

Several factors determine the probability of selecting the de novo mutations linked to antimalarial drug resistance 33. Amongst them, the frequency at which the resistance mutation occurs and the number of parasites in the human host exposed to the drug are considered as independent risk factors. If the probability of de novo resistance mutations is distributed evenly among all parasites then the biomass increment due to HIV-1 directly increases the rate at which these mutations occur.

Mathematical models of malaria epidemiology and drug resistance represent important tools in guiding malaria control strategies. They consistently predict that once drug resistance arises, it spreads rapidly, with the level of host immunity (both aspecific and specific) considered as an additional important independent factor. Although the dynamics underlying the spreads of antimalarial resistance may be much more complex than previously realized 34, the cycle of the emergence and subsequent spread of antimalarial drug resistance 35, and the influence of HIV-1-related immune-suppression can be schematically represented (Figure 3). Drug resistance starts with one or several genetic mutations in parasites which, if exposed to the drug, are selected and may spread. HIV-1 infection contributes to the emergence and spread of drug resistance by increasing drug exposure and drug pressure. HIV-1 infection increases the probability of a malaria infection progressing to symptomatic illness and to a higher parasite densities, increasing the probabilities of treatment and contact between the parasites and the drug 1212223. In addition, immune suppressed HIV-infected adults suffer frequently of non-malaria-attributable acute fevers that may be misdiagnosed as malaria and treated as such 22. Once resistance has emerged and spread above a certain threshold, malaria treatment of non-malaria fever in HIV immune-suppressed individuals might contribute to the further spread of drug resistance by increasing the drug pressure on asymptomatic malaria infections. After treatment, a delayed cure rate 36 and a higher rate of recrudescence 3738, a phenomenon already described in HIV-1 infected individuals, accelerate the spread of resistant parasites and increase the parasite biomass in symptomatic patients and asymptomatic carriers alike.

Besides contributing to the emergence and spread of antimalarial drug resistance, HIV-1 infection may influence and modify its expected geographical pattern. It has been postulated that antimalarial drug resistance first emerges and spreads in low-endemic areas, where the population acquired immunity is so low that most infections are symptomatic and thus treated, exposing a large numbers of parasites to antimalarial drugs 39. In highly endemic areas, less malaria infections become symptomatic and are treated (especially in adults). In sub-Saharan Africa, when considering the geographical distribution of both the HIV-1 infection and the spread of drug resistance (to chloroquine and sulphadoxine-pyrimethamine) 3340, the latter having been faster in eastern-southern (high HIV-1 prevalence) than in western Africa (relatively low HIV-1 prevalence) 41, the question on the role of HIV-1 pandemic as an additional contributing factor could be raised. It is impossible to provide a definitive answer though the results of the model presented in this paper are compatible with an influence of HIV-1 infection on the epidemiology of antimalarial drug resistance.

Malaria parasite biomass is influenced through various malaria control interventions. Easy access, prompt diagnosis and efficacious treatment with combination therapy are essential to halt or delay the emergence and spread of antimalarial drug resistance 9. The artemisinin derivatives, commercialized in the last decade, are the most potent antimalarial drugs with parasite reduction levels of 10,000 per cycle. Furthermore, if used in combination, the artemisinin component is protected by the partner drug and the parasite reduction rate is positively affected. Unfortunately, it is still unclear how treatment combinations for both malaria and HIV-1 interact. More data on the biological effects of therapy combinations would be of great utility to complement recent 'in vitro' evidence of a synergistic effect between protease inhibitors and antimalarials 32. Preventive interventions are equally important as they diminish the parasite load, transmission and drug pressure. Indeed, the incidence of symptomatic malaria in HIV-infected adults can decrease by 95% by giving cotrimoxazole prophylaxis and antiretroviral treatment combined with insecticide-treated bed nets 42. Furthermore, next to individual protection, preventive activities as indoor residual spraying, use of bed nets and cotrimoxazole prophylaxis reduce the parasite biomass at community level. For indoor residual spraying this is quite obvious as it is a community intervention but it has been shown that cotrimoxazole prophylaxis taken by HIV-infected people was associated with decreased morbidity and mortality among family members 43. Recent modeling has demonstrated that insecticide-treated nets protection has a substantial impact on the circulating malaria parasite biomass and has community-wide benefits that are often underestimated 44. Therefore, during the ongoing scale-up of HIV/AIDS and malaria control measures, HIV-1 infected individuals should be considered a priority group for the distribution of insecticide-treated nets, co-trimoxazole prevention and artemisinin-based combination therapy.