1. Introduction
Mozambique is among the 10 countries most affected by HIV in the world, with the world’s sixth highest prevalence among adults aged 15–49 [
1]. In 2009 the HIV prevalence in adults was estimated as 11.5%, with a prevalence of 13.1% for women as compared to a prevalence of 9.2% for men [
2]. A more recent AIS reported the HIV prevalence in the adult population as 13.2%, where the HIV prevalence among women was estimated at 15.4% and 10.1% among men within the same age group [
3]. The HIV epidemic in Mozambique is not uniformly distributed throughout the population. Adolescents and youth aged 15–24 are one of the most vulnerable populations, with an estimated prevalence of
in 2009 and
in 2015 [
2,
3]. Moreover, in this age group, women account for a disproportionate number of infections, with a prevalence (11.1%) three times higher than that of their male peers (3.7%). The epidemic also shows substantial variation throughout the country, with the prevalence among this age group ranging from
in Niassa province, in the north, to
in Sofala province, in the center of the country [
2].
There are many potential factors that position adolescents and young people at high risk of HIV. A recent study conducted in Maputo City points to low educational level and early sexual debut as important factors contributing to high HIV prevalence among youth [
4]. Furthermore, Dias et al. [
5] mention numerous biological, socio-economic, and socio-cultural characteristics that are linked to higher HIV prevalence among women in Mozambique, including age, sex of household head, wealth status, and marital status. Although the knowledge of biological, socio-economic, and socio-cultural drivers of the HIV epidemic is quite extensive, there is still a knowledge gap concerning the geographical distribution of HIV/AIDS across the country. Simply put, HIV/AIDS prevalence is not only influenced by socio-demographic, biological, and sexual behavioral factors; it can also differ significantly across districts and regions. Therefore, a thorough understanding of the geographical variability of HIV/AIDS is of paramount importance as it will help health officials formulate targeted interventions and identify where to prioritize the allocation of limited resources [
6].
In this paper, we seek to understand the spatial distribution of HIV/AIDS in Mozambique among adolescents and youth by using generalized geoadditive models. The rationale for the application of this type of model is that spatial information is generally regarded as a surrogate of unobserved risk factor which may appear hard to quantify [
7,
8]. Our approach is an extension of the model introduced by Kammann and Wand [
9]. Vandendijck et al. [
10] proposed to estimate the range parameter in this model, instead of fixing it. The novelty of our approach lies in extending the estimation method of the range parameter in the context of a binary outcome. In addition, our method takes the survey design and the sampling frame into account. The structure of the paper is as follows.
Section 2 gives a description of the data, a brief discussion on traditional methods to analyze geostatistical data, the statistical methods used, and details on the translation of the estimation method proposed by Vandendijck et al. [
10] to the binary setting.
Section 3 summarizes the results, comparing the estimates under different covariance functions and identifying the risk factors that are related to HIV prevalence in young people. We close the manuscript with a discussion including conclusions and opportunities for further research.
4. Discussion
Analyses of the spatial distribution of HIV are very crucial, especially in developing countries where resources are scarce, as the results might direct health officials to turn their attention to locations where more resources are needed and build policies to address the burden of HIV/AIDS in specific locations. In this paper, we applied generalized geoadditive models to analyze the spatial distribution of HIV prevalence among young people, by including the geographical location in the estimation process. The inclusion of the geographical location prompted us to use radial spline basis to study the spatial effect of geographical location on the HIV sero-status of the individuals. We extended the method proposed by Vandendijck et al. [
10] to the binary setting using PQL estimation. The method allowed us to estimate the effect of both the non-spatial covariates and the decay parameter in the radial basis function describing the spatial component in the generalized geoadditive model.
We observed that the two-stage iterative process PQL was robust to the choice of the covariance function. Further, we noted that the proposal by Kammann and Wand [
9] for estimation of the range parameter for some covariance function led to either parameter instability or to a very flat surface (e.g., for Gaussian and Matérn covariance functions no residual variation was observed for the spatial component). Additionally, it is important to note that the Gaussian covariance function resembles a very smooth spatial process. Waller and Gotway [
31] argue that such cases rarely occur in practice. Moreover, Davis and Morris [
32] state that although it is a valid covariance function, it can often lead to singularities in spatial prediction equations. Thus, results under the Gaussian covariance function should be interpreted with caution. Overall, results based on the two-stage iterative PQL-based estimation method were consistent for the spherical, exponential, and Gaussian covariance functions.
The analyses revealed three main regions with high HIV prevalence, namely southern, central, and the north-eastern regions. This is in agreement with what was observed in the Mozambique AIS report [
2]. The high prevalence observed can be explained partly by early marriages and early pregnancies. In the center of Mozambique, the Zambezia and Manica provinces have the highest number of child marriages [
33]. The practice of early marriage subjects children to early sexual debut, increasing their lifetime risk for HIV [
34]. In addition to early sexual debut, they are also exposed to frequent unprotected sex, generally with men older than them who have multiple sexual partners. In turn, this increases their risk of HIV infection [
35,
36]. It is also worth emphasizing that in the northern and central parts of the country, before they are forced into early marriage adolescent girls go through initiation rituals, a socio-cultural ceremony where adults pass to young people attitudes and beliefs about sexuality necessary for transitioning to adulthood, which in turn increases their risk of HIV infection [
37]. The high prevalence in high-burden regions may partly be explained by the presence of the high-risk population in these areas. A study conducted in Maputo, Beira, and Nampula City, Mozambique found that most of female sex workers were at the ages of 15–24 years [
38].
Several socio-demographic, knowledge and attitude-related, behavioral, and biological factors were identified to be associated with HIV infection among individuals aged 15–24 years in Mozambique. In this study, we found that young women were about six times more likely be HIV infected than their male counterparts. Gender inequality has been observed in several studies in sub-Saharan Africa. This can be partly explained by poverty, as poor women are often financially dependent on men, which drives them to early marriage and subsequent low decision-making power. Additionally, oftentimes vulnerable women have little choice but to adopt behaviors that put them at a higher risk of HIV, such as transactional and intergenerational sexual relationships [
39]. Furthermore, the large differences of HIV prevalence between young women and men of the same age may also be explained by the fact than women tend to partner with older men, who are much more likely to be HIV positive [
40,
41]. We found that young individuals aged 20–24 years were more likely to be HIV positive than adolescents aged 15–19 years. This is in line with other studies conducted in Southern Africa [
42,
43].
The findings indicate that higher education has a protective effect against HIV infection. Generally, it is assumed that educated individuals are more knowledgeable about HIV risk behavior and more likely to be empowered to make decisions about sexual debut, condom negotiation, etc. Therefore, the chances of adopting preventive measures are higher. Additionally, young individuals who know more about HIV/AIDS are more likely to know their HIV status, thus contributing to their health-seeking behaviors and consequently helping to lower the risk of HIV transmission [
44]. Our findings are supported by a study done in South Africa showing that adolescent and young women with tertiary education were less likely to be HIV positive [
43]. We also found that higher-income people with low HIV/AIDS awareness were more likely to be HIV positive than poor young individuals with the same level of awareness. In fact, there is evidence that wealthier people tend to have more high-risk behaviors, such as multiple sex partners [
45,
46]. This is an indication of the importance of awareness campaigns on HIV-related matters in the fight against HIV infection.
The analysis further revealed that young individuals in female-headed households in rural areas were more than two times more likely to be HIV positive than young individuals in the male-headed households in rural areas. A different scenario was observed in the urban areas, where the odds of becoming infected for individuals in female-headed households were 32% lower as compared to individuals in male-headed households. In rural areas, many women who become head of household due to the loss of their husband are confronted with many vulnerable experiences, such as loss of property, lack of access to many traditional services, and increased workload to support their basic necessities. Such experiences increase the vulnerability of women, which may result in engaging in practices that increase their risk of HIV. For instance, a study done in Kenya noted that in order to cope with the loss after the death of an adult member, some adolescent daughters engaged in activities that exposed them to higher HIV risk infection, such as early marriage or exchange of sex for money [
47]. Although the literature suggests that female-headed households tend to be poor [
48], different scenarios may be observed in urban areas as poverty is much more pronounced in rural areas.
Potential limitations of our study include: the geographical locations used were at the level of enumeration areas, and all households in the same enumeration area were assumed to have the same geographical location. When there is a lack of information on the sampled units, Bocci [
49] advocates to treat this as a measurement error and assume a distribution on the location inside each enumeration area. The radial basis functions considered in the analysis are only able to describe an isotropic process. In practice, many geostatistical processes are anisotropic. In future, research anisotropy in this type of model should be taken into account. As future research, it would also be interesting to consider the estimation of the decay parameter using a full-likelihood-based estimation approach. Additionally, this was a cross-sectional study, and thus causality of the identified factors cannot be proved. Our approach can also be used for cases where the observed data are counts.
Despite these limitations, the findings are very important and can help health officials and policy makers to design targeted interventions for responding to the HIV epidemic. The results revealed that provinces from the center and the north should be prioritized when allocating resources to fight HIV. Moreover, specific targeted care services and HIV prevention and awareness campaigns are urgently needed among adolescents and youth. To increase awareness, educational campaigns should include messages addressing the specific characteristics of various socio-demographic profiles. Additionally, there is a need to promote education among youth, particularly in rural and low-income areas. Special attention should be given to gender-equity and female empowerment when designing intervention measures.