Menu
Menu
Menu
Menu
Background: Human immunodeficiency virus (HIV) has continued to be one of the foremost public health problems globally. Even as more people living with the disease can now have access to antiretroviral therapy (ART), there are still some regions in the world with high transmission rates. The objective of this study was to examine the prevalence and individual-, household- and community-level factors associated with HIV infection among women of reproductive age in Mozambique. Methods: We used nationally representative cross-sectional data from the 2015 Survey of Indicators on Immunization, Malaria and HIV or Acquired Immunodeficiency Syndrome (AIDS) in Mozambique. A sample of 4726 women of reproductive age was included in this study. Prevalence was measured in percentage and the factors for HIV infection were examined using a multivariable multilevel logistic regression model. The level of significance was set at P < 0.05. Results: The seroprevalence of HIV among women in Mozambique was 10.3% (95% CI 9.2%, 11.6%). Furthermore, women who had two, three and four or more total lifetime number of sex partners were 2.73, 5.61 and 3.95 times as likely to have HIV infection when compared with women with only one lifetime sex partners, respectively. In addition, women of Islam religion had 60% reduction in HIV infection when compared with Christian women (adjusted odds ratio, AOR = 0.40; 95% CI 0.16, 0.99). The individual-level model (model B) had the best model fitness with the lowest Akaike information criterion (AIC) = 500.87 and Bayesian information criterion (BIC) = 648.88. The variations in the odds of HIV infection across communities (σ2 = 9.61 × 10–8; SE = 0.55) and households (σ2 = 1.02 × 10–4; SE = 1.02) were estimated. Results from the median odds ratio (MOR = 1.00) did not show any evidence of community and household contextual factors shaping HIV infection. MOR equal to unity (1) indicated that there were no community or household variances given the ICC of 0.0%. At both community and household levels, the explained variances were each 100%. This implied total variances in HIV infection has been explained by the individual-level factors. Conclusion: In this study, we found that having multiple total lifetime number of sexual partners and religion were predisposing factors for HIV infection at individual woman level. Female headship and wealth quintiles were associated with HIV infection at household level. Community illiteracy, intimate partner violence, poverty and geographical region were associated with HIV infection at community level. Therefore, multifaceted health intervention by stakeholders in the healthcare system will be useful in addressing the multilevel predisposing factors of HIV infection among Mozambican women.
We used nationally representative cross-sectional data from the 2015 Survey of Indicators on Immunization, Malaria and HIV/AIDS in Mozambique. A sample of 4726 women of reproductive age was included in this study. In the year 2015, under the supervision and sponsorship of the Mozambican Demographic and Health Surveys (DHS) Program, a survey indicator titled “Survey of Indicators on Immunization, Malaria and HIV/AIDS in Mozambique (IMASIDA)” was conducted. This survey served as an update to the previously obtained estimated indicators on the health of the mother and child, malaria and HIV/AIDS. This survey was intended to make available data at the nationwide and regional levels, the participants’ residential areas and in accordance with some of their background characteristics. In collaboration with the Mozambique National Institute of Statistics, the implementation of the project IMASIDA was carried out by the National Institute of Health. The collection of data was done between 8 June and 20 September 2015. Technical support throughout the survey programme was made possible by the Inner City Fund (ICF) with funds from the US Agency for International Development (USAID). The survey’s implementation process was overseen by the Mozambican Government through her health ministry, other national establishments, as well as other international agencies and organizations. The economic and technical backing came from the National Council to Combat HIV and AIDS of Mozambique and the US Centers for Disease Control and Prevention in cooperative agreement with The Global Fund, Health Alliance International/University of Washington (HAI/UW), World Health Organization (WHO), and President’s Emergency Plan for AIDS Relief (PEPFAR), United Nations Population Fund (UNFPA) and United Nations Children's Emergency Fund (UNICEF). The data is publicly available and can be accessed at https://dhsprogram.com/data/available-datasets.cfm. Details of the DHS sampling procedure have been previously reported [23]. The dependent variable was dichotomous with an indication of the seropositivity of the HIV status: a value of 1 and 0 was used to indicate whether a participant was seropositive (1) or seronegative (0). The determination of the serostatus of the participants was done by collecting a blood sample from each participant. Women’s HIV status (positive vs. negative) was explored in this study. The maternal ages were grouped as 15–19, 20–24, 25–29, 30–34, 35–39, 40–44, 45–49; intimate partner violence: yes vs. no; religion: Christianity, Islam and no religion/others; parity: nil, 1–2, 3–4 and 5+; place of delivery: health facility vs. home; antenatal care visit: yes vs. no; total lifetime number of sex partners: 1, 2, 3, 4+; age at sexual debut: not had sex, < 15, 15–17, 18+; marital status: never in marriage, currently married/living with a man, formerly married; health insurance coverage: covered vs. not covered; occupation: not working, professional/managerial, sales/services, agricultural/manual, clerical/household domestic work; educational: no formal education, primary and secondary or higher; neighbourhood socioeconomic disadvantaged status: low, medium and high; exposure to print or electronic media was measured dichotomously (yes vs. no) if a respondent used any of newspaper/magazine, radio or television. The inclusion of these factors was based on the outcome of the examined factors associated with HIV from previous studies [12, 24–26]. Household headship: male vs. female. The wealth index of the household was measured as a cumulative composite of the living standard of each of the surveyed households. It was calculated through the use of easy-to-collect data on each of the selected assets owned by the surveyed household. These selected assets include bicycles and televisions; the materials with which the houses were built or constructed; the type of water to which the households have access as well as their sanitation facilities. The individual households were placed on a continuous relative wealth scale, as the households’ wealth index, generated by the principal component analysis. The interviewed households were separated primarily, by DHS, into five wealth quintiles using principal component analysis (PCA) to compute the household variables. PCA has been proven and validated to be a useful technique for describing how socioeconomic status of a given population is differentiated within that population. It has also been used in the reduction of the number of variables in a given data set [27]. Z scores and factor loadings (factor coefficient) for each household were calculated. The loadings were multiplied by the indicator values of each household and summed, thereby producing the value of each household’s wealth index. The overall assigned scores of the poorest/poorer/middle/richer/richest categories were disentangled with the aid of standardized z score [28]. We used enumeration areas (EAs) to represent communities because the DHS did not collect aggregate-level data at the community level. Hence, community-level variables included in the analysis were based on women’s characteristics, particularly those that have implications for HIV infection. The aggregate community-level variables were constructed by aggregating individual-level characteristics at the community (cluster) level and categorization of the aggregate variables was done as low or high for each community. Residential status: urban vs. rural. Geographical region: Niassa, Cabo Delgado, Nampula, Zambézia, Tete, Manica, Sofala, Inhambane, Gaza, Maputo Provincia, Maputo Cidade. The level of sexual violence within the community [whether half (50%) of the clustered population experience sexual violence or not]. The distribution of uneducated women (illiteracy) within the community [whether half (50%) of them had any form of formal education or not]. The poverty concentration within the community (whether half (50%) of the women fall within the least wealth quintiles or not). The concentration of intimate partner violence within the community (whether half (50%) of the women experience intimate partner violence or not). The exposure to print and electronic media within the community (whether half (50%) of the women within the community use any of the print and electronic media including newspaper/magazine, radio or television or not). This approach is similar to the methods of a previous study [29]. In this study, we utilized population-based secondary data sets available in the public domain/online with all identifier information removed. The authors were granted access to use the data by MEASURE DHS/ICF International. The DHS Program is consistent with the standards for ensuring the protection of respondents’ privacy. ICF International ensures that the survey complies with the US Department of Health and Human Services regulations for the respect of human subjects. No further approval was required for this study. More details about data and ethical standards are available at https://goo.gl/ny8T6X. The Demographic and Health Survey is a de-identified open-source data set. Therefore, the requirement of consent for publication is not applicable. We used the ‘svy’ module to adjust for data strata, clusters and sample weights. A multivariable multilevel logistic regression model was employed in the estimation of the fixed and random effects of the associated factors to HIV infection. Binary response in a three-level model was specified as at level 1 (individual woman), at level 2 (a household) and at level 3 (living in a community). Out of the five models constructed, model A is an unconditional or empty model with no explanatory variables. This first model was employed to specifically decompose the sum of discrepancy that occurred between households and community levels. The null or empty model is important for understanding the community and household variations, and we used it as the point of reference in estimating the extent to which the household and community factors varied. It was also used to justify our usage of the multilevel statistical framework. This is so because in the empty model, if the community variation is not significant, it will be better to use the single-level logistic regression. We determined the level of statistical significance to be at less than 0.05, while data was analysed using Stata version 14 (StataCorp., College Station, TX, USA). The selection criteria in building the last four models: Models B–D require that only individual-, household- or community-level variables which were significant in the univariate analysis were added in the adjusted model. Model E is the full model for all significant variables irrespective of the level. The results obtained from the measures of association (fixed effects) were reported as adjusted odds ratios (AORs) with confidence interval (CI) set at 95%. Intra-class correlation (ICC) and the median odds ratio (MOR) were used for the probable contextual effects. We used ICC to measure any variance in respondents in the same household and within the same community. This tool (ICC) presents the percentage of the overall variation in the probability of HIV infection which is in relation to the household and the community levels. The second or third level (household or community) variance was measured by the MOR, as the odds ratio, and it estimates the probability of HIV infection that can be accredited to household and community context. A MOR value of unity suggests that no household or community was at variance. Contrariwise, higher value suggests that the contextual effects for the understanding of the probability of HIV infection are more important. The Bayesian and Akaike information criteria were used as measurement criteria to determine how well the different models we employed were fitted to the data. When the values on the two criteria are low, it implies a better fit of the model [30].
N/A
