This study uses the most recent data sets available as of October 2007 from the Demographic and Health Surveys (DHS) of the following 11 countries: Burkina Faso, Cameroon, Ethiopia, Ghana, Guinea, Kenya, Lesotho, Malawi, Rwanda, Senegal, and Zimbabwe. The data sets were chosen based on availability of women anthropometric measures and HIV test results to allow comparative analysis. The DHS data sets are from nationally representative surveys of households with at least one woman of reproductive age, usually between 15 to 49 years. The data were collected by various in-country research and statistical agencies with technical assistance from ORC Macro International, Inc (Calverton, Maryland, USA). All eligible women in each selected household were surveyed. The surveys were based on two-stage sample designs. In the first stage, enumeration units or "clusters" were selected from larger regional units within countries. Next households were randomly selected within clusters.

This study is based on an analysis of existing survey data with all identifier information removed. The survey was approved by the Ethics Committee of the ORC Macro at Calverton in the USA and by the National Ethics Committees of the 11 countries. All study participants gave informed consent before participation and all information was collected confidentially. The HIV testing protocol does not allow individual test results to be linked to a specific respondent, to safeguard the confidentiality of the information. Eligible women were informed about the procedures, the confidentiality of the result, and the fact that test results will not be made available to them.

Apparent prevalence estimates were computed using number of HIV positive women with BMI <18.5 and the total number of HIV positive women reported in each country. Because a normal distribution is mandatory for the pooling of data, logit transformation was applied as outlined by Lipsey et al 23 and weighted by inverse variance of logit transformed prevalence. Two techniques were used to calculate the pooled prevalence estimates: the Mantel-Haenszel method 24 assuming a fixed-effects model and the DerSimonian-Laird method 25 assuming a random-effects model.

The logit effect size for prevalence, its standard error, and the inverse variance weight are:

ES l = log e [ p 1 − p ] MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacPC6xNi=xI8qiVKYPFjYdHaVhbbf9v8qqaqFr0xc9vqFj0dXdbba91qpepeI8k8fiI+fsY=rqGqVepae9pg0db9vqaiVgFr0xfr=xfr=xc9adbaqaaeGaciGaaiaabeqaaeqabiWaaaGcbaGaeeyrauKaee4uam1aaSbaaSqaaiabdYgaSbqabaaccaGccqWF9aqpcqqGSbaBcqqGVbWBcqqGNbWzdaWgaaWcbaGaeeyzaugabeaakmaadmaajuaGbaWaaSaaaeaacqWGWbaCaeaacqaIXaqmcqGHsislcqWGWbaCaaaakiaawUfacaGLDbaaaaa@3DEA@

SE l = 1 n p + 1 n ( 1 − p ) MathType@MTEF@5@5@+=feaafiart1ev1aqatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacPC6xNi=xI8qiVKYPFjYdHaVhbbf9v8qqaqFr0xc9vqFj0dXdbba91qpepeI8k8fiI+fsY=rqGqVepae9pg0db9vqaiVgFr0xfr=xfr=xc9adbaqaaeGaciGaaiaabeqaaeqabiWaaaGcbaGaee4uamLaeeyrau0aaSbaaSqaaiabdYgaSbqabaaccaGccqWF9aqpdaGcaaqaaKqbaoaalaaabaGaeGymaedabaGaemOBa4MaemiCaahaaOGaey4kaSscfa4aaSaaaeaacqaIXaqmaeaacqWGUbGBcqGGOaakcqaIXaqmcqGHsislcqWGWbaCcqGGPaqkaaaaleqaaaaa@3E50@

w l = 1 S E l 2 = n p ( 1 − p ) MathType@MTEF@5@5@+=feaafiart1ev1aqatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacPC6xNi=xI8qiVKYPFjYdHaVhbbf9v8qqaqFr0xc9vqFj0dXdbba91qpepeI8k8fiI+fsY=rqGqVepae9pg0db9vqaiVgFr0xfr=xfr=xc9adbaqaaeGaciGaaiaabeqaaeqabiWaaaGcbaGaem4DaC3aaSbaaSqaaiabdYgaSbqabaGccqGH9aqpjuaGdaWcaaqaaiabigdaXaqaaiabdofatjabdweafnaaDaaabaGaemiBaWgabaGaeGOmaidaaaaakiabg2da9iabd6gaUjabdchaWjabcIcaOiabigdaXiabgkHiTiabdchaWjabcMcaPaaa@3F53@

where p is the proportion of women with HIV-related malnutrition and n is the total number of women that are HIV seropositive in the sample. The final pooled logit results and 95% confidence intervals (CIs) were back-transformed to proportion for ease of interpretation using:

p = e logit e logit + 1 MathType@MTEF@5@5@+=feaafiart1ev1aqatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacPC6xNi=xI8qiVKYPFjYdHaVhbbf9v8qqaqFr0xc9vqFj0dXdbba91qpepeI8k8fiI+fsY=rqGqVepae9pg0db9vqaiVgFr0xfr=xfr=xc9adbaqaaeGaciGaaiaabeqaaeqabiWaaaGcbaGaemiCaaNaeyypa0tcfa4aaSaaaeaacqWGLbqzdaahaaqabeaacqqGSbaBcqqGVbWBcqqGNbWzcqqGPbqAcqqG0baDaaaabaGaemyzau2aaWbaaeqabaGaeeiBaWMaee4Ba8Maee4zaCMaeeyAaKMaeeiDaqhaaiabgUcaRiabigdaXaaaaaa@41AF@

To evaluate whether the results of the studies were homogenous, the Cochran's Q test was used 26. The quantity I² that describes the percentage of variation across studies that are heterogeneity rather than chance was calculated 2728. Subgroup analyses were performed according to participants' socioeconomic characteristics: Wealth index, education attainment, present occupation, and type of residence. To evaluate the stability of the results and to test whether one country had an undue influence on the meta-analysis, leave-one-country-out sensitivity analysis was performed 29. The scope of this analysis was to evaluate the influence of individual countries, by estimating prevalence of HIV-related malnutrition in the absence of each country. All tests were two tailed. For all tests, a probability level less than .05 was considered significant. Stata 10 (Stata Corporation, College Station TX) software was used for the statistical analyses.

This study uses data from 11 Demographic and Health Surveys (DHS) conducted between 2003 and 2006 in sub-Saharan Africa (SSA). The countries, years of data collection, and sample sizes are listed in Table 1. Table 1 also shows the proportion of women (both sero-negative and sero-positive) classified as underweight, normal weight, overweight, and obese for countries included in this study. Table 2 illustrates the demographic and economic diversity of the selected countries. Regarding levels of urbanization, the percentage of urban population differs significantly among the selected countries. It varies from 16% – 18% in Ethiopia, Malawi, and Burkina Faso, to more than 40% in Senegal, Ghana, and Cameroon. As for gross domestic product (GDP) per capita, Cameroon, Lesotho, and Senegal emerge as the most affluent countries with values higher than US$ 600, whilst by contrast Ethiopia and Malawi were the most deprived (less than US$ 200). The fastest growths were recorded in Rwanda (5%), Burkina Faso (3%), and Ethiopia (3%). By contrast, Lesotho (1%) and Zimbabwe (1%) witness the slowest growth rates of their urban population. The most marked reductions in GDP per capita from 1990 to 2004 were in Zimbabwe (-2%) and Kenya (-1%), whereas improvements were recorded in Lesotho (5%) and Ghana (2%) to a lesser degree in Cameroon (1%). The selected countries also displayed marked socioeconomic diversities in term of per capital health expenditure, adult literacy, and per capita food production (data not shown).

The prevalence of HIV-related malnutrition among women (HIV-positive women with low body weight) varied widely, from 0.6% in Lesotho to 16.9% in Burkina Faso (crude unweighted mean 7.7%). Figure 1 graph the prevalence estimates and 95% confidence intervals (CIs) from the individual countries and pooled result. Meta-analysis of all 11 countries yielded an overall pooled prevalence of 10.3% (95% CI; 7.4% to 14.1%) with no statistically significant heterogeneity (Cochran's Q test = Q = 4.82 on 10 degree of freedom, P = .903, I² = 0%).

To evaluate the stability of the meta-analysis, leave-one-"country"-out sensitivity analysis was performed (Figure 2). In this analysis, the overall prevalence was calculated, removing one country at a time. This shows the studies conducted in Malawi, Ethiopia, and Burkina Faso had the greatest influence on combined pooled prevalence of HIV-related malnutrition. Omitting studies from Ethiopia and Burkina Faso reduces the pooled prevalence to 9.8% (95% CI; 6.9% to 13.9%) and 8.9% (95% CI; 6.1% to 12.8%) respectively. While omitting study from Malawi increased the pooled prevalence to 11.6% (95% CI; 7.5% to 17.3%). However, the confidence intervals do change materially with exclusion of any of these countries, which remains within the 95% confidence interval of the overall estimate for all countries (Figure 2). This analysis confirmed the stability of the results.

To gain further insight into the prevalence of HIV-associated malnutrition, the prevalence estimates were disaggregated into difference socioeconomic status as shown in Table 3.

As expected, the prevalence of HIV-associated malnutrition decreased monotonically as one move up the wealth index quintiles. Similarly, the pooled prevalence of HIV-related malnutrition decreased with increasing level of education. The prevalence of underweight was higher among HIV-positive women living in the rural areas compared to their urban counterparts (6.8% versus 16.3%). Professionally employed women were less likely to be underweight than skilled or unskilled manual workers (9.0% versus 17.5%) and women not employed (9.0% versus 11.6%).

This meta-analysis of DHS has brought together evidence from 11 countries in sub-Saharan Africa conducted between 2003 and 2006. These surveys have important limitation in term of number of HIV-positive women, thus inferences have been limited by a lack of precision. The study therefore aimed to improve the precision of the estimate of prevalence of HIV-related malnutrition by carrying out a meta-analysis. Pooling the prevalence estimates of 11 countries yielded an overall prevalence of 10.3% (95% CI 7.4% to 14.1%). This study found a strong pattern of decreasing level of proportion of HIV-positive women with low body mass index with increasing level of education. Even more evident is the pattern of decreasing proportion of low body weight among HIV-positive women with increasing wealth index. Type of resident also displays distinctive patterns. HIV-positive women living in rural areas were decidedly thinner than those living in the urban areas.

Greater levels of malnutrition are also evident among women not employed and manual unskilled professionals compared to women in professional field. These findings are consistent with data from other studies that have examined these socioeconomic factors in the general population 30313233343536.

There are a number of caveats to be considered when interpreting these results. Some criticize the aggregation of different study findings, particularly those with different levels of methodological quality 23. Caution is essential when interpreting Cochran Q statistic, the statistical power of tests for heterogeneity are, in most cases, very low due to the small number of combined studies 3738, implying that heterogeneity may be present even if the Q statistics is not statistically significant 283940. One important limitation is that DHS surveys do not collect data on household income or expenditure, the traditional indicators used to measure wealth. The assets-based wealth index used here is only a proxy indicator for household economic status, and it does not always produce results similar to those obtained from direct measurements of income and expenditure where such data are available or can be collected reliably 2122. Another limitation of the present analysis relates to the use of body mass index (BMI) as the only measure of malnutrition. Due to lack of appropriate data on dietary intake and food or calorie consumption, low body mass index among HIV positive women was used as proxy for malnutrition. Cross-sectional data only allow looking at associations; it is impossible to assess directly how the relation between socioeconomic position and BMI may change over time 41.

Despite these limitations, the study strengths are significant. It is a large, population-based study with national coverage from 11 countries, and low heterogeneity among included studies suggesting appropriateness of pooling results from different countries. Meta-analysis has numerous strengths. It allows for synthesis of study findings and comparison of findings across multiple studies 42. Meta-analysis allows researchers to arrive at conclusions that are more accurate and more credible than can be presented in any one primary study or in a non-quantative, narrative review 43. Also, meta-analysis can be used to highlights gaps or limitations in the research literature. The DHS have some important advantages when compared with other surveys. They are often nationally representative, allowing for conclusions that cover the entire nation 44. In addition, same variable is operationalized in the same way and making it possible for numerical values comparable across countries. Overall, the number of included studies and geographic and socioeconomic diversities constitute a good yardstick for the region and help strengthen the findings from the study. In addition, low heterogeneity among studies suggests that the result of the meta-analysis is consistent and allows for generalisability of the results to other countries in sub-Saharan Africa.