Study design and population

This study utilized a community-based cross-sectional design using Demographic and Health Survey (DHS) data collected between 2016 and 2023 from selected sub-Saharan African (SSA) countries. The source population comprised all women aged 15–24 years living in SSA. However, due to data availability, the study population was restricted to women aged 15–24 years included in the most recent DHS datasets from 24 SSA countries conducted within this timeframe. Thus, the analysis is based on a subset of SSA countries with available and comparable DHS data, and the findings should be interpreted in the context of these included countries rather than the entire SSA region.

Data source

The data used in this study were obtained from the most recent DHS dataset for SSA countries. The DHS program ensures the use of standardized data collection methods across all countries [30]. These surveys are designed to gather comparable data. They employ large sample sizes and are population-based, aiming to provide nationally representative information for each country [30]. Each country’s survey included separate datasets for men, women, infants, births, and households. For this analysis, the women’s datasets (IR files) were utilized. The focus of the study was on young women, as defined by the United Nations (UN) as individuals between the ages of 15 and 24. The most recent DHS conducted in the 24 SSA countries were appended for the analysis. In total, a weighted sample of 140,054 young women was used in this study.

Sample size determination, and methods

On average, every five years, a systematic compilation of DHSs is conducted over 90 low- and middle-income countries. The DHS samples are typically divided between urban and rural areas within each administrative geographic region. The DHSs used stratified two-stage cluster sampling. This means that, in the first stage, primary sampling units (PSUs) are selected with a probability that is proportional to their size. In the second stage, a fixed number of households (or residential dwellings) is selected from a list of households obtained from the selected PSUs during an updating operation. A PSU refers to a specific area or part of an area known as an Enumeration Area (EA) [31]. Following the listing of households, equal probability systematic sampling is used to select a specific number of households from inside the defined cluster [30].

Variables of study

Operational definitions for some factors

Mass media exposure: Participants in the study who reported exposure to television, radio, or magazines/books were classified as having “media exposure.” This classification was applied if they had exposure to at least one of the three media sources. If participants do not have exposure to any of these media sources, they are classified as not having mass media exposure.

HIV knowledge: HIV knowledge was assessed by generating scores based on six questions related to HIV (know that a healthy-looking person can have HIV, know that HIV cannot be transmitted by mosquito bites, know that HIV cannot be transmitted by witchcraft or supernatural means, know that a person cannot become infected by sharing food with a person who has HIV, know that people can reduce the risk of getting HIV by using a condom every time they have sex, and know that people can reduce the risk of getting HIV by having just one uninfected sex partner who has no other sex partners [32,33]. Participants were classified into three categories: low knowledge (if a woman answered three or fewer questions correctly), moderate knowledge (if a woman answered four to five questions correctly), or comprehensive knowledge (if a woman answered all six questions correctly).

Risky sexual behavior: Risky sexual behaviors were evaluated using five questions: whether the participant had an STI in the last 12 months; had sexual experience before 18 years old; experienced genital sores/ulcers/discharge in the last 12 months; had at least one sexual partner other than their husband in the last 12 months; and engaged in multiple lifetime sexual partnerships [32,33]. These questions were combined to create an index of risky sexual behavior classified into two categories: “low/no risk” (if the response was “no” for all questions) and “risky” (if the response was “yes” for one of the five questions)

Attitude to HIV patients: The HIV Stigma indicator was assessed using six questions (would be ashamed if someone in the family had HIV; would buy vegetables from a vendor with HIV; children with HIV should be allowed to attend school with children without HIV; people hesitate to take an HIV test because of the reaction of other people if positive; people talk badly about people with or believed to have HIV; and people with or believed to have HIV lose respect from other people) that measured negative attitudes towards individuals living with HIV/AIDS [32,33]. The variable was categorized as follows: “no stigma (good attitude)” (if a score of 6 was obtained) and “stigma (poor attitude)” (if a score of less than 6 was obtained).

County income: The variable "country income" was calculated using the World Bank classification based on the DHS time period, which divides countries into low-income (≤ $1,135), lower-middle-income ($1,136−4,465), and upper-middle-income ($4,466−13,845) categories [34].

Country literacy rate: Evidence from the World Bank and World Population Review indicates that developed nations have an average literacy rate of more than 90%, while least developed nations have an average literacy rate of only 65% [35,36]. Then, based on the DHS periods of the countries and the World Bank evidence as a baseline, the researchers divided the countries’ literacy rates into the following categories: high literacy: countries with a literacy rate greater than or equal to 65%; and low literacy: countries with a literacy rate less than 65%.

Subregion: Based on the classification of the WHO, DHS, and other international health indicators [37,38], the countries were grouped according to the following classification: East Africa, West Africa, and South/Central Africa. This was based on their geographic location within the continent.

Variable selection approaches

Variable selection for this study was guided by both theoretical and empirical considerations. Independent variables were chosen based on prior research and informed by behavioral and socio-ecological theories that explain HIV testing uptake among young women in SSA. Variables representing individual, household, and community-level factors were included due to their established relevance in influencing health-seeking behaviors. In addition, the Empirical Bayesian Lasso algorithm (EBglmnet package) was applied to empirically refining the set of predictors by shrinking less important coefficients to zero, retaining only those with meaningful contributions to the model. This combination of theory-driven and data-driven selection enhances model validity and interpretability

Data quality, collection tools and procedures

The quality of DHS data was ensured through various measures. These included providing training for data collectors, supervisors, and field editors; conducting ongoing supervision; using standardized and translated questionnaires in international, national, and country-specific local languages; and employing data processing specialists for data entry and management. To ensure data accuracy, a pre-test was conducted before data collection, and a debriefing session with pre-test fieldworkers was held. Questionnaires modified as needed based on the feedback received. More detailed guidance on the data collection process can be found in the DHS documentation [32].

Data management processes

Prior to conducting any additional statistical analysis, the data underwent checks to ensure completeness and were appropriately weighted. The process of data extraction was using STATA V. 17. Other processes like cleaning, recording, and analysis were carried out using R software. Country or administrative boundary data used for map visualization were obtained from the open-source Database of Global Administrative Areas (GADM) (https://gadm.org/) AND imported into R for spatial analysis. Once the data is obtained from the DHS, proper data management practices are implemented. This includes addressing missing observations using appropriate techniques and conducting recording and variable recategorization as necessary. These steps ensure the integrity and reliability of the collected data.

Missing data management

We first looked at missing data using charts and graphs to spot any patterns. For variables like antenatal care visits and place of delivery, which were missing because women were not pregnant, these data were not truly missing but just not applicable, so we did not impute these values. Instead, we did descriptive and bivariate analyses for these variables and removed them from the final model because they significantly affected the results. For other missing data, we used Little’s MCAR test (p < 0.05) to check if the data were missing completely at random, which showed that they were not. Therefore, we treated the data as missing at random (MAR). We handled this by using Multivariate Imputation by Chained Equations (MICE) with Bayesian logistic regression in R to create 10 imputed datasets. This method included variables like alcohol consumption, distance to the health facility, health insurance coverage, and smoking status and ensured that the imputed data did not differ more than 6 percent from the complete cases.

Statistical analyses processes

A sampling weight was applied to account for the non-proportional allocation of the sample across different regions and potential variations in selection and response rates. This will ensure that the survey results accurately represent both national and regional levels. Descriptive statistics was presented using frequencies, texts, and percentages.

Spatial autocorrelations

The Global Moran’s I statistic was employed to determine whether the distribution of uptake of HIV testing in SSA shows a dispersed, clustered, or random pattern [39,40]. In this study, we evaluated the global spatial autocorrelation of HIV testing uptake among young women in SSA by applying permutation inference to Moran’s I statistic. Through 9,999 random permutations, we determined the rank of the observed Moran’s I value. The Moran’s I statistic produces a single value ranging from −1 to +1, where values close to −1 indicate a dispersed pattern, values close to +1 indicate a clustered pattern, and a value of 0 suggests complete randomness. The results indicated a statistically significant clustered spatial pattern of HIV testing uptake across the included SSA countries.

Cluster and outlier analysis (Anselin local Moran’s I)

The Anselin Local Moran’s I, also known as the Local Indicator of Spatial Association (LISA), is a valuable tool for identifying spatial clusters and outliers within the data. The local Moran’s I analyzes spatial autocorrelation at the local level, allowing for the identification of specific areas with statistically significant clustering or outliers. After confirming global spatial autocorrelation with Moran’s,I we employed a Monte Carlo simulation with 9999 permutations to assess local spatial autocorrelation. To manage multiple comparisons and control the false discovery rate, we utilized the false discovery rate (FDR) p-value adjustment method.

Hot spot analysis (Getis-Ord Gi*)

Using a conditional permutation approach with 999 iterations, hotspot analysis was conducted using the Getis-Ord Gi* statistic, which provides information on the intensity of core hotspots (areas with a high prevalence of uptake of HIV test) or cold spots (areas with a low prevalence of uptake of HIV test) [41]. This method helps evaluate the strength of clustering across different areas.

SaTScan analysis

SaTScan analysis, a spatial scan statistical analysis, was performed to identify significant primary and secondary clusters of uptakes of HIV testing based on the Bernoulli distribution using a maximum spatial cluster size of 50% of population at risk [42]. The scanning window with the maximum likelihood represents the most likely performing cluster [42]. The distribution and statistical significance of clusters were explored through Monte Carlo replication of data with more than 999 replications, ensuring sufficient power in the cluster identification process. The likelihood ratio (LR) test statistic and its corresponding p-value were utilized to determine if the observed number of women who have undergone HIV testing within a potential cluster is significantly higher or lower than expected.

Spatial Kriging interpolation analysis

Since collecting data from every region in each country is often impractical, spatial interpolation methods are employed to estimate values for areas where data is not available. In this study, we utilized the ordinary Kriging geostatistical interpolation method from the gstat package in R to predict HIV testing uptake among young women in unobserved regions across SSA. Prior to applying Kriging interpolation, we conducted an exploratory data analysis (EDA) to examine the distribution of HIV testing uptake, test for normality, and apply necessary transformations. We performed variogram analysis to identify the best-fitting model (exponential, spherical, or Gaussian) based on the sum of squared errors. The model with the lowest sum of squared errors was chosen for our predictions. Assumptions of stationarity and isotropy were tested through trend analysis and spatial autocorrelation assessments (e.g., Moran’s I). To validate the model, we applied cross-validation by predicting held-out data and calculating the mean squared error (MSE), ensuring the data met the necessary statistical requirements for accurate Kriging interpolation [43].

Model diagnosis and selection

To assess the best-fitted model, we ran both a Bayesian spatial multilevel model, and a Bayesian multilevel model. Using model diagnosis criteria applied to the full model, we selected the best-fitted model based on its performance. In this case, the Bayesian multilevel model found as the best-fitted model due to its lower DIC, and WAIC

Bayesian hierarchical model, and convergence of algorithm

We employed a mixed-effects logistic regression analysis using the Bayesian approach. In this study, Bayesian methods were chosen for their flexibility and ability to incorporate prior knowledge, which is useful in complex data scenarios like HIV testing uptake across diverse regions of SSA. Unlike frequentist approaches, Bayesian analysis provides credible intervals that offer a more intuitive understanding of uncertainty, which is crucial in public health decision-making. Additionally, these methods are effective in handling complex survey data with stratified sampling and weights, allowing for more robust and precise estimates. Overall, Bayesian methods helped provide reliable and actionable insights into HIV testing uptake among young women in SSA. To determine the individual- and community-level factors for HIV testing uptake among young women, two-stage Bayesian multilevel logistic regression models were employed. Four models were used in this study. In Model I, there were no explanatory factors (null model); Model II; only individual-level variables; Model III, only community-level variables; and Model IV; both individual- and community-level variables. Adjusted odds ratios (AORs) with 95% credible intervals (CrIs) were used to present the results. Unlike classical methods, Bayesian models incorporate prior beliefs, offering plausible parameter values. By utilizing Bayesian methods, we effectively address uncertainty in estimating factors influencing HIV testing uptake, ensuring a comprehensive understanding of the data [44]. Our analysis addressed variation in HIV testing uptake across enumeration areas (EAs) using random variables.

The likelihood function.

The key ingredients to a Bayesian analysis are the likelihood function, which reflects information about the parameters contained in the data, and the prior distribution, which quantifies what is known about the parameters before observing data. The prior distribution and likelihood can be easily combined to form the posterior distribution, which represents total knowledge about the parameters after the data has been observed. Bayesian multilevel logistic analysis specifies a dichotomous dependent variable as a function of a set of explanatory variables [45,46].

Likelihood contribution from the subject in the group is Bernoulli.

Where, represents the probability of the event for subject i in group j that has covariate vector and indicates the presence () or absence () of the event for that subject. In multilevel logistic regression, we know that

Where, is fixed part of the model and is random part of the model and . is the probability of the woman in the group being have HIV testing uptake, so that the likelihood contribution for the subject in the group is

Since individual subjects in the group are assumed to be independent of each other, the likelihood function over a data set of subjects in the group is then

Posterior distribution.

The full conditional distribution for parameter is given by [47]:

For parameter , the full conditional distribution is [47]:

Where is the total number of observations.

In our analysis, we utilized the Brms-R package to estimate the parameters of the variable of interest and assess the extent of random variations between clusters. The Brms-R package offers advanced statistical modeling capabilities, particularly using Hamiltonian Monte Carlo (HMC) and the No-U-Turn Sampler (NUTS). Through the utilization of the Brms-R package, specifically its implementation of HMC and NUTS, we were able to effectively estimate the parameters of the variable and quantify the random variations between clusters, enhancing the efficiency and reliability of our analysis [47,48]. The flexibility of Brms package in modeling categorical and ordinal data allows for a more thorough exploration and understanding of the factors influencing the outcome variable [47,49].

The regression coefficients for each parameter and variance were estimated using a flat prior with a normal distribution (0, 1000) and a half-Cauchy distribution (0, 25), respectively. Additionally, we employed the following settings for the estimation process: iteration = 6,000, warm-up (number of discarded iterations) = 2,000, cores = 4 (specifying the number of cores used for the algorithm), chains = 4, adapt delta (controlling divergent transitions) = 0.95, and initials (starting values for the iterations) = 0. These settings allowed us to estimate the posterior distribution.

The widely applicable information criterion (WAIC) was utilized to identify the best-fitting model. This criterion is preferred over the commonly used deviance information criterion (DIC) for model selection [50,51]. The model with the lowest WAIC was considered the best-fitting model. Fortunately, HMC algorithms often converge much more quickly than alternative methods [47,52,53]. Therefore, to evaluate the convergence diagnostics criteria, we utilized the following indicators: Rhat = 1, effective sample sizes (both Bulk_ESS and Tail_ESS) surpassing 1000. Furthermore, we inspected time series plots to ensure well-mixed chains, observed smooth density plots, and verified that the autocorrelation plots revealed minimal correlation with zero for most parameters.

Measure of variations between levels

To estimate the EA effects on HIV testing uptake outcome and to quantify the variation in the HIV testing uptake outcome between EAs (i.e., clusters), we applied the variance partition coefficient and the median odds ratio (MOR), respectively.

Intra-class correlation (ICC): measures the proportion of the total observed individual variation in the outcome that is explained by the between-cluster variation. It requires an estimate of the variance at the individual level (level 1), and for the standard logistic distribution, this variance is equal to [54]. Therefore, in the multilevel logistic regression models M1 and M2 with a logit link the ICC is approximated by

Where σ²μ is the variance of the random parameter at the cluster-level representing the amount of unobserved heterogeneity between clusters; and is parameter representing the amount of unobserved heterogeneity between women (individual-level variance).

Median Odds Ratio (MOR): The MOR measures how much variability in the uptake of HIV exists between EAs (i.e., clusters) by comparing two persons from two randomly chosen, different EAs[55]. Consider two people chosen randomly from two different EAs but with the same values of covariates in the model. The MOR is the median odds ratio between the person at EA with a higher probability of receiving HIV test and the person at survey EA with a lower probability of receiving HIV testing. It is a function of the estimated EAs, i.e., cluster variance,  and is given by

Where σ² is the variance of each model and Φ⁻¹ is the inverse of the standard normal cumulative distribution function is the 75th percentile is the exponential function [55,56].

Proportional change in variance (PCV) was calculated as; , where V₀ is the variance of the null model and V_x is the variance of each model at each level with variables [54].

Ethical consideration and consent to participate

The study was carried out with authorization obtained from www.dhsprogram.com through an online request to access SSA DHS data. The survey descriptions were submitted to the DHS MEASURE program for review prior to data access. The datasets to be used in this study were handled with the utmost confidentiality and treated as highly reliable. Stringent protocols were followed in DHSs to protect the rights and privacy of participants. These protocols included obtaining informed consent, maintaining confidentiality, and ensuring the anonymity of collected data. The surveys were designed to uphold high ethical standards and prioritize the well-being and safety of participants. Ethical clearance was not required because there was no involvement of patients or the public in the study.

Sociodemographic-related characteristics of young women

This study targeted young women aged 15–24 years in SSA, with a total weighted sample size of 140,054. Of these, 53.64% (75,130) were late adolescent girls. Half of the participants, 50.14% (70,228), had completed secondary school or higher. Over half, 57.93% (81,129), were not employed. Nearly one in four, 24.27% (33,989), belonged to the wealthiest category. Regarding marital status, more than half, 60.85% (85,229), had never been married. Additionally, 70.29% (98,448) of households were headed by men (Table 1).

Download:

• PNG
  larger image
• TIFF
  original image

Table 1. Socio-demographic characteristics of young women and HIV testing uptake in SSA (weighted sample: n = 140,054).

https://doi.org/10.1371/journal.pone.0353117.t001

Healthcare utilization

Of the participants, 82.83% (116,006) lacked health insurance. A majority, 92.83% (47,140), reported antenatal care visits. Only 5.86% (8,212) of respondents made healthcare decisions independently. Most 80.75% (113,092) visited a health facility once in the past year. Over half 64.66% (90,557) did not consider distance from a health facility a significant barrier (Table 2)

Download:

• PNG
  larger image
• TIFF
  original image

Table 2. Healthcare utilization and maternal reproductive-related characteristics of HIV testing uptake among young women in SSA (weighted n = 140,054).

https://doi.org/10.1371/journal.pone.0353117.t002

Behavioral and knowledge-related characteristics

Regarding mass media exposure, more than half, 52.29% (73,240), reported having access to mass media. Knowledge of HIV was found to be comprehensive among 41.42% (58,014) of the participants. About attitudes toward HIV patients, 60.40% (84,593) showed a low level of negative attitudes. About contraceptive utilization, 80.03% (112,090) of respondents reported not using contraceptives. Risky sexual behavior was noted in 54.25% (75,978) of respondents. Lastly, smoking was quite rare, with 97.38% (136,379) of respondents indicating that they did not smoke (Table 3).

Download:

• PNG
  larger image
• TIFF
  original image

Table 3. Behavioral and knowledge-related characteristics of HIV testing uptake among young women in SSA (weighted n = 140,054).

https://doi.org/10.1371/journal.pone.0353117.t003

Country and community-level characteristics

Over half, 57.95% (81,156), lived in rural areas. Regarding country income classification, over half, 58.22% (81,543) of respondents were from low-income countries. More than half, 66.41% (93,010) of respondents were from countries with higher literacy rates. Nearly half, 48.86% (68,433) of respondents were from East Africa (Table 4).

Download:

• PNG
  larger image
• TIFF
  original image

Table 4. Country and community-level characteristics of HIV testing uptake among young women in SSA (weighted n = 140,054).

https://doi.org/10.1371/journal.pone.0353117.t004

Pooled prevalence HIV testing uptake among young women in SSA

The pooled prevalence of HIV testing uptake in SSA was about 38.9% [95% CrI: 29.0,50.4]. There is significant variation among countries. The variability across countries is estimated at 1.13 [95% CrI:8.5,15.2]. Zambia showed the highest uptake, with a rate of 76.1% [74.9, 77.2], followed by Uganda at 73.7% [74.9,74.2] and Malawi at 69.5% [68.6,70.4]. Conversely, the countries with the lowest uptake are Mauritania, at 4.3% [3.8,4.8], Madagascar, at 11.3% [10.6,12.0], and Benin, at 12.1% [11.3,12.0] (Fig 1)

Download:

• PNG
  larger image
• TIFF
  original image

Fig 1. Forest plot of pooled prevalence of HIV testing uptake among young women in sub-Saharan Africa using recent DHS data (2016-2023).

https://doi.org/10.1371/journal.pone.0353117.g001

Spatial autocorrelation (Global Moran’s I statistics)

Spatial autocorrelation analysis showed that HIV testing uptake among young women aged 15–24 years in SSA revealed a nonrandom and clustered pattern. The Moran’s I statistic for this study were 0.53, with a p-value of < 0.0001 which showed spatial clustering of HIV test uptake across countries in SSA. The density plot of simulated Moran’s I values from permutation testing shows that most values are clustered around zero. This distribution suggests that, under the null hypothesis of a random distribution, Moran’s I value is typically close to zero. The observed clustering in HIV testing uptake indicates that similar values are spatially grouped more than would be expected by chance (Fig 2).

Download:

• PNG
  larger image
• TIFF
  original image

Fig 2. Global Moran’s I of HIV testing uptake among young women in sub-Saharan Africa using recent DHS data (2016-2023).

https://doi.org/10.1371/journal.pone.0353117.g002

Cluster and outlier analysis using Anselin’s Local Moran’s I

To detect spatial outliers and assess local clusters of HIV testing uptake among young women in SSA, we employed Asselin’s Local Moran’s I. The analysis revealed distinct cluster types on the map. Low-Low Clusters (red points) are found in regions where both the young women and their neighbors have low rates of HIV testing uptake, with notable concentrations in West Africa (around Ghana, Nigeria, and neighboring countries) and parts of Southern Africa (including Mozambique and Madagascar). High-low clusters (purple points) represent areas where young women show high uptake but are surrounded by regions with lower uptake rates, particularly dispersed across Eastern and Southern Africa. Low-high clusters are characterized by low uptake among young women, with neighboring areas exhibiting higher uptake; these are less common and observed in parts of Kenya and Tanzania, suggesting that targeted interventions could help align these areas with their higher-uptake neighbors. Lastly, high-high clusters (blue points) indicate regions where both the young women and their neighbors have high uptake rates, prominently found in East Africa, especially in Kenya, Uganda, and parts of Tanzania (Fig 3).

Download:

• PNG
  larger image
• TIFF
  original image

Fig 3. Cluster and outlier analysis of HIV testing uptake among young women in SSA using recent DHS data (2016−2023) (https://data.humdata.org/dataset/5e8d83a5-1210-49be-b7d9-cf286dbc15df).

https://doi.org/10.1371/journal.pone.0353117.g003

Hot spot analysis Getis-ord G*

To identify hotspots for HIV testing uptake among young women in sub-Saharan Africa, we applied the Getis-Ord G* statistic. The map classified regions into high hotspots (blue points) and low hotspots (red points), reflecting areas of significant clustering of high or low HIV testing uptake. Red points denote regions with consistently low HIV testing uptake, notably in West Africa (including Mauritania, Mali, Gambia, Guinea, Sierra Leone, Liberia, Ghana, Burkina Faso, and Benin), as well as in Madagascar and parts of eastern and central Africa like Ethiopia and Angola. Blue points indicate regions with high HIV testing uptake, primarily found in eastern and southern Africa, including Kenya, Uganda, Tanzania, Rwanda, Burundi, Zambia, Malawi, Mozambique, and South Africa (Fig 4)

Download:

• PNG
  larger image
• TIFF
  original image

Fig 4. Hotspot and cold spot analysis of HIV testing uptake among young women in SSA using recent DHS data (2016−2023 (https://data.humdata.org/dataset/5e8d83a5-1210-49be-b7d9-cf286dbc15df).

https://doi.org/10.1371/journal.pone.0353117.g004

Spatial interpolation

To predict prevalence of HIV test uptake in unobserved areas across Sub-Sharan Africa ordinary Kriging geostatistical interpolation method. We compared several empirical variogram models, exponential, spherical and Gaussian. Exponential model was selected for our predictions due to its lowest sum of squared errors as shown in the figure below

The ordinary Kriging interpolation predicted values of HIV testing uptake was assessed using categorized into four ranges, each represented by different colors. Countries colored in red (predicted value: 0–20%) showed the lowest predicted HIV testing uptake among young women, indicating significant challenges in accessing or utilizing HIV testing services. This category included Mauritania, Mali, Ghana, Angola, Mozambique, and Eastern Ethiopia. Secondly, in the orange range (predicted value: 20% – 40%), countries showed moderate-low HIV testing uptake. This group consists of Burkina Faso, Benin, Sierra Leone, Cameroon, half part of Ethiopia, Burundi, Malawi, and Madagascar. Countries in the blue range (predicted value: 40% − 60%) demonstrated moderate predicted HIV testing uptake in northwest and central Ethiopia, Côte d’Ivoire, Tanzania, and Zambia. Finally, countries colored green (predicted value: > 60%) represented those with the highest predicted HIV testing uptake among young women (Fig 5) (https://data.humdata.org/dataset/5e8d83a5-1210-49be-b7d9-cf286dbc15df).

Download:

• PNG
  larger image
• TIFF
  original image

Fig 5. Ordinary Kriging interpolation of HIV testing uptake among young women in SSA using recent DHS data (2016-2023).

https://doi.org/10.1371/journal.pone.0353117.g005

SaTScan analysis

This is the largest and most significant low-uptake cluster consisted of (LLR: 6318.48, p < 0.0001) in West Africa countries including Mauritania, Mali, Ghana, Sierra Leone, Liberia, Burkina Faso, Benin, and Côte d’Ivoire). This largest window spans 5,271 enumeration areas and is centered at coordinates (25.422785 N, 10.759942 W) with a radius of 3053.55 km, covering much of Western Africa. Within this cluster, the relative risk (RR) of low HIV testing uptake is 1.64 (RR = 1.64), indicating that the likelihood of low uptake (did not take HIV testing) is 1.64 times greater than in the surrounding areas. This cluster has a log-likelihood ratio (LLR) of 6318.48 and a p-value of <0.0001. The second, and third clusters highlights Angola, and Ethiopia as a region with significantly low HIV testing uptake. This second window spans 256 enumeration areas and is centered at coordinates (18.790704 S, 46.666002 E) with a radius of 753.42 km, covering much of Western Africa. Within this cluster, the relative risk (RR) of low HIV testing uptake is 1.59 (RR = 1.59), indicating that the likelihood of low uptake (did not take HIV testing) is 1.59 times greater than in the surrounding areas. This cluster has a log-likelihood ratio (LLR) of 9,21.1 and a p-value of <0.0001 (Table 5, and Fig 6)

Download:

• PNG
  larger image
• TIFF
  original image

Table 5. Shows significant clusters detected in spatial SaTScan analysis for HIV testing uptake among young women in SSA.

https://doi.org/10.1371/journal.pone.0353117.t005

Download:

• PNG
  larger image
• TIFF
  original image

Fig 6. SaTScan Analysis of HIV Testing Uptake Among Young Women in SSA Using Recent DHS Data (2016−2023) (https://data.humdata.org/dataset/5e8d83a5-1210-49be-b7d9-cf286dbc15df).

https://doi.org/10.1371/journal.pone.0353117.g006

Measurement of cluster variation (random effects) and model comparison

The null model (model I), which serves as the baseline, shows a community-level variance of 2.88. This indicated a significant amount of variation in women’s HIV testing uptake at the community level. The ICC for the null model was 46.61%, meaning that 46.61% of the total variance in HIV testing uptake was attributable to differences between communities. Model II showed a reduction in the community-level variance to 1.97, a 31.59% decrease from the null model. This suggests that the inclusion of individual-level variables in this model was able to explain some of the between-community variation in HIV testing uptake. The ICC for Model II was 37.45%, indicating that 37.45% of the total variance is still explained by community-level factors. In Model III, the community-level variance was found to be 2.19, and the ICC is 39.96%. This implies that the addition of community-level variables in this model did significantly improve the explanation of between-community variation compared to Model I. Finally, Model IV showed the lowest community-level variance of 1.51, a 47.61% reduction from the null model. The ICC for this model IV was 31.46%, suggesting that 31.46% of the total variance in HIV testing was still attributable to community-level factors. This model, with the inclusion of both individual and community-level variables, appears to be the best-fitting model, as indicated by the lower values for the DIC, and WAIC compared to the other models. Overall, the results indicate that both individual- and community-level factors play a significant role in determining young women’s HIV testing in SSA, with the community-level factors accounting for a substantial portion of the total variance (Table 6).

Download:

• PNG
  larger image
• TIFF
  original image

Table 6. Random parameters and model comparison among young women in HIV testing uptake in SSA.

https://doi.org/10.1371/journal.pone.0353117.t006

Determinants of HIV testing uptake among young women in Sub-Saharan Africa

In analyzing HIV testing uptake among young women aged 15–24 years in SSA countries from 2016 to 2023, several factors significantly impact testing rates, with corresponding adjusted odds ratios (AOR) and 95% Credible Intervals (CrI). The significant factors influencing HIV testing uptake among young women in SSA include primary education, secondary or higher education, employment, exposure to mass media, marital status (married, widowed, or divorced), use of modern contraceptives, age (20–24 years), household wealth (middle-income and rich), engagement in risky sexual behaviors, attitude towards HIV patients, alcohol consumption, pregnancy status, female household head, rural versus urban residence, and perception of distance to health facilities. Accordingly, women with primary education have 1.52 times higher odds of being tested for HIV compared to those with no education (AOR 1.52, 95% CrI: 1.42–1.63). Study participants with secondary or higher education were 3.11 times more likely to be tested for HIV (AOR 3.11, 95% CrI: 2.89–3.34) than their counterparts. Employed women had 1.17 times higher odds of undergoing testing compared to unemployed women (AOR 1.17, 95% CrI: 1.13–1.22). Exposure to mass media increased testing odds by 1.39 times (AOR 1.39, 95% CrI: 1.33–1.45) compared with their counterparts.

Young married women were 2.02 times more likely to be tested than single women (AOR 2.02, 95% CrI: 1.88–2.16), and widowed or divorced women had 2.22 times higher odds (AOR 2.22, 95% CrI: 2.05–2.39) of HIV testing uptake. Those using modern contraceptives had 1.39 times higher odds of testing of HIV (AOR 1.39, 95% CrI: 1.33–1.47). Young women aged 20–24 had 2.45 times higher odds of HIV testing compared to those aged 15–19 (AOR 2.45, 95% CrI: 2.35–2.55). Richer women show increased testing odds, with middle-income women having 1.09 times (AOR 1.09, 95% CrI: 1.03–1.15) and rich women having 1.25 times higher odds of HIV testing (AOR 1.25, 95% CrI: 1.18–1.32) compared with poor households.

Women engaging in risky sexual behaviors had 2.62 times higher odds of being tested (AOR 2.62, 95% CrI: 2.49–2.73). Those with a poor attitude towards HIV patients were 6.42 times more likely to test (AOR 6.42, 95% CrI: 6.09–6.77). Young women with alcohol consumption had associated with 0.32 times lower odds of HIV testing uptake (AOR 0.68, 95% CrI: 0.63–0.73). Young women with pregnancies had 1.56 times higher odds of getting HIV testing (AOR 1.56, 95% CrI: 1.46–1.66) compared with their counterparts. Women with a female household head have 1.21 times higher odds of being tested (AOR 1.21, 95% CrI: 1.16–1.26). Young women of rural residents had 0.28 times lower odds of testing compared to urban residents (AOR 0.72, 95% CrI: 0.67–0.77). Lastly, women who do not view distance to a health facility as a major issue had 1.05 times higher odds of being tested for HIV (AOR 1.05, 95% CrI: 1.01–1.09) compared with their counterparts. (Table 7).

Download:

• PNG
  larger image
• TIFF
  original image

Table 7. Individual and community-level factors associated with HIV testing uptake among young women in SSA countries from 2016-2023.

https://doi.org/10.1371/journal.pone.0353117.t007

Strengths and limitations of the study

This study had several strengths but also notable limitations. One of its key strengths is the use of large, weighted, and nationally representative samples from 24 SSA countries, allowing for generalizable findings across the region. Additionally, by examining the most recent DHS data and focusing on a highly affected population—young women—this study provides timely and relevant insights. The use of national survey data also enhances the potential to influence policymaking and intervention planning at both the national and regional levels. The application of Bayesian hierarchical modeling and spatial analysis further strengthens these findings. However, this study has several limitations that must be considered. First, the study did not account for time as an explanatory variable, which may have influenced changes in HIV testing uptake. Second, because DHS data rely on self-reports from respondents, there is a risk of recall bias, particularly in reporting HIV testing behaviors. Additionally, as the study is based on secondary data, certain cultural and health service-related variables may not have been captured, which could limit the full understanding of the local contextual factors affecting HIV testing. Another limitation arises from the displacement of geographical coordinates during data collection (5 km in rural areas and 10 km in urban areas) to maintain respondent confidentiality. This may have affected the accuracy of the spatial analyses, particularly in identifying localized patterns of HIV testing uptake. Despite these limitations, the findings offer valuable insights, but they should be interpreted with caution, especially in relation to potential biases and geographic data constraints

Recommendations and the way forward

To improve HIV testing and care in Sub-Saharan Africa (SSA), governments and NGOs should focus on infrastructure development and targeted outreach. The ministries of health must prioritize enhancing healthcare access in low-uptake regions, particularly in West Africa, by investing in both fixed healthcare facilities and mobile clinics to reach remote communities. Additionally, ministries of education should incorporate comprehensive sexual and reproductive health education into school curricula and promote youth-friendly outreach programs that provide knowledge about HIV and available testing services. Governments should increase funding for healthcare services in underserved areas, strengthen partnerships with local community organizations, and ensure that HIV testing programs are both culturally appropriate and accessible. Regional and international organizations can adapt successful HIV testing models from Southern and Eastern Africa, tailoring them to local contexts to improve coverage and equitable resource distribution in the region. Moreover, investing in education and employment programs can address socioeconomic barriers to testing uptake. Finally, continued research is crucial for understanding local barriers to testing, evaluating the effectiveness of existing interventions, and conducting longitudinal studies that track changes in factors affecting testing behavior. Geospatial analysis should also be incorporated to identify areas that require tailored, context-specific interventions to reduce disparities in HIV testing rates.