Ethiopia is Africa’s second most populous country, with an estimation about 126.5 million people (2023), divided into twelve 12 regional states and two federal-level administrative cities. The area of country was 112,127 sq. km, of which, 7,444 sq. km is water (17). The topography of Ethiopia is highly diverse, with elevation ranging from 125 m at the Denakil Depression to 4,620 m at Ras Dejen. More than 45% of the country is dominated by a high plateau with a chain of mountain ranges that is divided by the East African Rift Valley. This region with elevations greater than 1,500 m is known as the highlands where almost 90% of the nation’s population resides, perhaps to take advantage of its relatively disease-free environment (28). Surrounding the highlands are regions known as the lowlands (<1,500 m), where most of the remaining population (mostly pastoralists) lives (28). Ethiopia’s varied topography has created three climatic zones, which have been known since antiquity as the dega, the weina dega, and the kolla (18, 19). According to the 2016 Demographic Health Survey, 22% of reproductive-age women were underweight (20). In Ethiopia, about 610,000 people were estimated to be living with HIV in 2023 (21).

This was a cross-sectional study design using secondary data from the EDHS 2005, 2011, and 2016, a nationally representative household survey conducted every 5 years in Ethiopia. In this analysis, we extracted a wide range data of demographic, socioeconomic and HIV prevalence and anemia status from DHIS2. The EDHS employs a multistage, stratified sampling design to ensure the survey is representative. In the first stage, clusters (enumeration areas) are selected from the country’s administrative regions. In the second stage, households within these clusters are randomly chosen for inclusion in the survey. All women of reproductive age (15–49 years) residing in the selected households are eligible to participate in the EDHS. The survey collects information on various aspects of the participants’ health, including their anemia status, which is determined through the measurement of hemoglobin levels. Additionally, the survey gathered data on the participants’ HIV status, which was ascertained through blood sample testing.

This study employed a cross-sectional design, utilizing data from the 2005, 2011, and 2016 EDHS. The target population consisted of all women of reproductive age (15–49 years) residing in Ethiopia at the time of the survey. From this larger population, the researchers focused specifically on the subsample of HIV-positive women, as identified through the HIV testing component of the EDHS.

The EDHS program measured anemia in women by pricking their fingers and using the HemoCue hemoglobin device. The results were sorted into two categories: “yes” and “no.” Further categorization was done to reclassify the data: “non-anemic” stayed the same, while “mild,” “moderate,” and “severe” anemia were combined into “anemic.” The chosen explanatory factors cover a wide range of socio-demographic, economic, and health-related aspects, providing a comprehensive understanding of the various factors that influence anemia prevalence. The independent variables were divided into individual and community-level factors.

Age was categorized into 15–24, 25–29, 30–34, and 35–49. Religion was categorized as Orthodox, Muslim, protestant, and others. Media exposure was categorized as yes and not at all. Marital status was categorized as married and unmarried. Educational level was categorized as no education, primary, and secondary and above. HIV status was categorized as positive and negative. Pregnancy or breastfeeding status was categorized as yes or no. The occupation was categorized as had work and had no work. BMI category was categorized as underweight, normal, and overweight. Distance to the health facility was categorized as a big problem and not a big problem. Family size was categorized as 1–5 and greater than 5. The number of children was categorized as no child, 1–2, and more than 2. The sex of the household was categorized as male and female. Toilet type was categorized as improved and unimproved. The source of drinking water was categorized as improved and unimproved. The use of modern contraceptives was categorized as yes, and no. The history of abortion was categorized as yes and no.

The poverty level of the community was determined by analyzing the percentage of women in the two lowest income quintiles within each cluster. Similarly, community-level media exposure was defined as the percentage of women who had access to at least one form of media (television, radio, or newspaper). These community-level factors were categorized as low or high based on the national median value (22).

To assess the distribution of the proportion values for each community, a histogram was utilized. If the composite variable demonstrated a normal distribution, the mean value was used for categorization. However, if the variable did not show a normal distribution, the median value was used for categorization. As the data did not exhibit a normal distribution, median values were used to categorize communities as “high” or “low.” The explanatory variables at the aggregate community level, like community poverty level and community illiteracy level, were derived by consolidating individual-level characteristics at the country level. These variables were then classified as either high or low based on the proportion values calculated for each community (cluster). Furthermore, based on their developmental stage and the governmental support, Ethiopia’s 11 regions are grouped into three major categories; ‘three Metropolis’ (Addis Ababa, Harari, and Diredewa), the large central regions (Tigray, Amhara, Oromia and Southern Nations, Nationalities and Peoples region (SNNPR)), and the ‘small peripherals’ (Afar, Benshangul-Gumuz, Gambela, and Somali) based on their development status and the need for governmental support (23).

The data were recorded and analyzed using STATA 17, employing both descriptive and multilevel logistic regression analyses. To ensure the data’s representativeness and obtain reliable estimates, the data were weighted prior to analysis. Given the hierarchical nature of the DHS data, a multilevel binary logistic regression analysis was conducted. Metrics such as the Interclass Correlation Coefficient (ICC), Proportional Change in Variance (PCV), and Median Odds Ratio (MOR) were calculated to determine the presence of clustering. Four models were developed: the null model (without explanatory variables), Model I (including individual-level factors), Model II (including community-level factors), and Model III (incorporating both individual and community-level factors). Model comparison was based on deviance statistics, with the lowest deviance value considered as the best-fitted model. Both bivariable and multivariable multilevel logistic regression analyses were performed. In the bivariable analysis, variables with a p-value ≤ 0.25 were considered for the multivariable analysis. Ultimately, variables with a p-value ≤ 0.05 in the multivariable analysis were identified as significant factors associated with anemia prevalence among HIV-positive women of reproductive age in Ethiopia.

The DHS data, which have data on demographic, socioeconomic, HIV prevalence and anemia status were included. However, this analysis did not include survey data sets that did not have HIV prevalence and anemia status dataset. Furthermore, analysis only incorporates surveys completed in 2005, 2011, and 2016. After checking the type and percentage, missing values for the dependent and independent variables were very small and deleting the missing records.

Ethical approval or participant consent was not necessary for this study as it involved the analysis of secondary data that is publicly accessible through the MEASURE DHS program¹. The authors received authorization to access and utilize the data from the program’s website (see text footnote 1). The datasets do not include any personal identifying information, such as names or addresses of individuals or households.

Most women were aged 35–49 (40.8%), followed by those in the 30–34 age group (23.3%). Regarding education, (34.5%) had no formal education. More than half (55.5%) were unmarried, and a large majority identified as Orthodox Christian (72.7%). Regarding employment status, (64.7%) of women were employed. The majority of women were not pregnant (96.5%), while about (16.6%) were currently breastfeeding. Only (14%) reported ever having a terminated pregnancy. BMI revealed that (27.4%) were underweight, whereas (15.1%) were overweight. Family size data indicated that (28%) had more than 5 members. In terms of children, (44.2%) of women had more than two children. Access to health facilities posed a significant challenge for (38.7%) of women. About two-thirds (66.5%) of households used unimproved toilet facilities. Modern contraceptive utilization was low, with only (26.5%) of women using modern contraceptives. More than half of the households were headed by females (54.6%). Media exposure was prevalent, with (78.4%) having access to media, particularly in communities with high literacy levels (75.4%). Additionally, (79.5%) resided in communities with high socio-economic levels. Most women were from urban areas (79.5%), and (76.4%) lived in large central regions, compared to smaller peripheral areas and metropolitan zones (Table 1).

The study revealed that the overall prevalence of anemia in the surveyed population was 23.9% (95% CI: 19.24–29.24). However, this rate exhibited significant regional variation, with anemia affecting 31.2% of individuals in small peripheral regions, 23.8% in large central regions, and 22.4% in the three metropolitan areas (Figure 1).

Over the survey years, mean hemoglobin values fluctuated, recorded at 12.4 g/dL (11.8–12.9) in 2005, 13.1 g/dL (12.8–13.3) in 2011, and 12.6 g/dL (12.3–12.8) in 2016. When analyzed by age, the mean hemoglobin levels were 13.1 g/dL for individuals aged 15–24, 12.7 g/dL for those aged 25–29, 13.1 g/dL for the 30–34 age group, and 12.5 g/dL for those aged 35–49 (Figure 2).

The prevalence of anemia also varied by survey year, with rates of 32.9% in 2005, 20.2% in 2011, and 24.5% in 2016. Further analysis highlighted disparities in anemia prevalence across various demographic and socioeconomic factors. Women aged 35–49 had the highest anemia prevalence at 28.3%, while those aged 30–34 had the lowest at 16.5%. Marital status also played a role, with unmarried women experiencing a higher prevalence (25.2%) than their married counterparts (22.4%). Employment status showed that working women had a slightly higher prevalence of anemia (24.2%) compared to those who were unemployed (22.4%). Additionally, the prevalence of anemia was highest among underweight women (34%) and lowest among overweight women (15.1%). Pregnant women also had a higher prevalence of anemia (30%) compared to non-pregnant women. Other factors, such as family size, literacy levels, and residence type, influenced anemia rates, with rural women experiencing a significantly higher prevalence (30.7%) compared to urban women (Table 2).

The variability in anemia prevalence at the community level was evaluated using a random effect model. The analysis revealed significant clustering, as evidenced by the increase in the ICC from 0.06 in the null model (Model I) to 0.56 in Model II, 0.47 in Model III, and 0.51 in Model IV. This indicates a substantial influence of community factors on anemia prevalence. The MOR further highlighted the impact of community-level factors, increasing from 1.53 in the null model to 6.86 in Model II, suggesting heightened heterogeneity in anemia prevalence across communities. The PCV provided additional insights, showing a − 19.31% change in variance for Model II compared to the null model, with slight increases of 0.30% and −15.74% for Model III and Model IV, respectively. Model fitness was assessed using the Log-Likelihood Ratio (LLR) and Deviance (−2LLR), with improvements noted in the subsequent models. Model IV exhibited the best fit, with the lowest deviance (160.17), followed by Model II (168.17), both significantly better than the null model (937.83). These findings underscore the pivotal role of community-level factors in influencing anemia prevalence among HIV-positive women in Ethiopia (Table 3). Furthermore, the model’s predictive performance was evaluated using the Area Under the Curve (AUC) and the Receiver Operating Characteristic (ROC) curve metrics, which are plotted based on sensitivity and 1-specificity probabilities. The final model achieved an AUC of 86.1%, demonstrating a strong capability to predict anemia prevalence (Figure 3).

In the binary multilevel logistic regression analysis, variables such as age category, marital status, occupation, number of children, family size, and history of abortion had a p-value greater than 0.25. The remaining variables, which had marginal or lower p-values, were included in the multivariable multilevel regression analysis. The study found several significant predictors of anemia among HIV-positive reproductive-age women in Ethiopia. Women with low BMI had 3.9 times (AOR = 3.9, 95% CI: 1.21–9.70) odds of anemia compared to those who were overweight. Additionally, women living in female-headed households were about 4.5 times more likely to develop anemia (AOR = 4.5, 95% CI: 1.14–11.25) as compared with male-headed households. Moreover, pregnant mothers took iron during their last pregnancy the odd of having anemia was 2.9 (AOR = 2.9; 95% CI: 1.48–9.32) times lower than as compared to their counterparts. In addition to this, women using unimproved toilet facilities were about 1.6 times more likely to develop anemia (AOR = 1.6; 95% CI: 1.18–6.87) as compared with women using improved toilet facilities. Other factors such as media exposure, and current pregnancy status were not significantly associated with anemia (Table 4).