Our novel and most important findings are illustrated in the contrast between and . We observed a pronounced change in the odd ratios in Lusaka and Copperbelt for the 2001/2 ZDHS and Lusaka and Central for the 2007 ZDHS following adjustment of the geographical location (spatial auto-correlation) arising from the population mobility (migration), and the younger age structure of the urban population‥ Also, Western province, which was among the lowest prevalence areas in 2001/2002, changed to one of the highest prevalence areas in 2007. And Southern province, which was among the highest prevalence areas in 2001/2002, changed to one of the lowest prevalence areas in 2007. Copperbelt was no longer among the provinces with the highest marginal odds ratios in both surveys.
We also observed a rapid increase in sero-prevalence among men aged 15. This finding is unlikely due to earlier sexual debut: respondents aged 15–19 reporting sex before age 15 declined from 27% to 16% between the 2001/2 and 2007 ZDHS (Central Statistical Office, 2004 and 2009
). A more likely explanation is riskier intercourse among males. The percentage of adolescent men aged 15–19 reporting using a condom with their most recent sexual partner declined from 35% in 2000 to 31% in 2009, and only 38% reported that they could get condoms on their own (Central Statistical Office, 2010
That the highest probability of being HIV-infected was observed among respondents residing in Western and Central provinces could be explained in part by the mobility of the Zambian population along transportation routes, acknowledged by the World Bank and other leading institutions as a major vector in the spread of HIV (World Bank, 2008
; World Bank, 2009
). Truck drivers, seasonal workers and commercial sex workers all comingle along these routes, often moving from less developed/urbanized areas to more developed regions with greater economic opportunities. With an unemployment rate exceeding 50% and 68% of Zambians living in poverty, many young women have little choice but to become commercial sex-workers, with HIV rates documented at 65% (UNAIDS/WHO, 2008; Biennial Report to UNGASS, 2010
). Often they are denied the kinds of prevention services available to non-sex workers (Graham, 2009
). By contrast, the rapid decrease in prevalence rates in Southern province could possibly reflect the heavy concentration of USAID Strategic Objective 9 (SO9) funding and activities in Southern province, particularly along the major transport route between Livingstone and Lusaka (USAID Zambia S09 Activities, 2010
) ( bottom
As a President’s Emergency Plan for AIDS Relief (PEPFAR) focus country, Zambia has benefited from high levels of external funding, rising from US$6 in 2003 to US$10 per capita in 2006. During this period, the proportion of external funding rose from 70% to 74%, with PEPFAR the largest funder (50%), followed by the Global Fund to Fight AIDS, Tuberculosis and Malaria (GFATM) (16%) and the World Bank MAP program (Multi-Country HIV/AIDS Program for Africa), which ended in 2008 (Global HIV/AIDS Initiatives Network, 2009
). GFATM has allocated Zambia USD 42.5 million to counter HIV/AIDS. This major influx of donor funding has supported a variety of programmes during this period, but it is difficult if not impossible to precisely inventory these activities. The USAID Zambia Mission publically details the specific HIV/AIDS Multi-sector Response activities funded through PEPFAR listed in (USAID Zambia Mission, 2010
). Funding for these combined activities was roughly USD 100 million, or nearly one sixth of the US government total of USD 620 million for the period 2003 to 2008, which constituted approximately 60% of all donors funding to Zambia (USAID Zambia S09 Activities, 2010
). Of the USD 270 million approved by PEPFAR for Zambia in fiscal year 2009, approximately 28% was earmarked for prevention activities (U.S. State Department, 2010
Table 2 Major USAID HIV/AIDS Multisector Response Activities, 2004–2010* (source reference 22)
The USAID program most relevant to the spread of HIV along transport corridors, the 3-year (2006–2009), USD 11 million Corridors of Hope II (COH II) program, has neglected the major east-west corridor between Lusaka and Mongu at the western terminus of the major road into Western province (see map, ). This program became fully implemented in 2007 in seven Zambian high-prevalence border and transport districts: Livingstone, Kazungula, Chipata, Kapiri Mposhi, Nakonde, Solwezi, and Siavonga (Chirundu) (COH II Program, 2008). In 2009 the Corridors of Hope III (COH III) program was proposed to continue and potentially expand the activities of CHO II, with three basic elements focusing on prevention of sexual transmission—condoms and other prevention, abstinence-based (AB) activities, and counselling and testing (CT) services. Funded at close to USD 25 million over five years and potentially adding additional sites in years 3–5, COH III will continue to provide services in the same seven districts as COH II (COH III Program, 2009
). Our findings suggest that this program should be expanded into transport towns along the east-west corridor into Western province.
Border migration from Angola into Western province is another possible explanation for the increase in Western province rates, but Angola has the lowest HIV prevalence rate (2.5%) in continental southern Africa, while countries to the south (Namibia, 15%; Botswana, 24%; and Zimbabwe, 15%) have higher rates (UNAIDS/WHO, 2010
). The Trans-Caprivi highway links landlocked Botswana, Zambia and Zimbabwe with the deepwater port of Walvis Bay in Namibia, and is thus a very active transport zone/border crossing for truckers. The Caprivi region of Namibia along this route has the highest prevalence of HIV in Namibia, estimated at over 40% (Afrol News, 2006
). Also, declining prevalence in Southern province could be attributed to increased AIDS-related mortality, but provincial-level aids-specific mortality data are not available to test this hypothesis.
Also, the primary drivers of the Zambian epidemic are not necessarily the high-risk populations noted above, although they certainly contribute a disproportionate share of the cases via multiple, concurrent, geographically-dispersed sexual partnerships. By far, the largest source of new infections (over 70% of the total) is casual heterosexual sex (National HIV/AIDS/STI/TB Council, 2010
). The greatest prevention impact would therefore be expected for programmes broadly targeting this much larger risk group.
A major limitation of the findings was discussed previously (Kandala et al., 2008
; Garcia-Calleja et. al., 2006
). Residual spatial distributions of risk of HIV infection might be influenced by model variable selection. Urban/rural location and province are both chosen as explanatory factors in the geo-additive regression model. However, it is possible that place of residence (urban/rural location) shared the effect of geographic distribution of the regions (provinces), especially in Lusaka and Copperbelt, as these areas are highly urbanized and almost one-half of the country’s population is concentrated in these few urban zones. It is worth investigating this issue further if one has several factors included in HIV/AIDS studies. It was not possible to confirm or disprove the above statement in our study since we had only a limited set of variables (province, urban/rural residence and gender).
It is worth mentioning some of the advantages of the approach described above over more conventional approaches like regression models with fixed or random province effects; or standard 2-level multilevel modelling with unstructured spatial effects. Most studies of HIV/AIDS prevalence astonishingly neglect the geographic location, spatial autocorrelation, and nonlinear effects of covariates, which in our view is likely to result in misleading conclusions regarding the prevalence of the disease. Additionally, and more importantly, the impact of this neglect is an underestimation of standard errors of the fixed effects which in turn inflates the apparent significance of the estimates. Our analysis included this correlation structure and accounted for the dependence of neighbouring provinces in the model. Since ZDHS data are based on a random sample within communities, the structured component introduced here allows us to ‘borrow strength’ from neighbouring clusters in order to cope with the sample variation of province effect and obtain estimates for areas that may have had inadequate sample sizes or be unsampled. This gives more reliable estimates of the fixed effect standard errors. A failure to take into account the posterior uncertainty in the spatial location (province) would overestimate the precision of the prediction of HIV prevalence in regions with inadequate sample size. Controlling for important risk factors such as geographical location (spatial auto-correlation) arising from the population mobility (migration), age structure of the population, and gender gave estimates of prevalence that are statistically robust.