Improving Local HIV Estimates Through Data Sharing

Dynamic Models are useful tools for understanding the spread of diseases like HIV. These models help estimate how many people are infected, how many new infections occur, and how many people die from the disease. They have worked well at national levels, but there is a growing need for more detailed Data at local or sub-national levels to support better planning and resource allocation for HIV interventions.

As we look into HIV epidemics, we realize that not all areas are the same. Some places have a much higher rate of HIV than others. For instance, in Zambia, the HIV rate was much higher in some regions compared to others in 2013. This variation means that simply using national data might not be enough to guide local actions. Accurate data at the local level can help ensure resources are allocated where they are most needed.

However, getting enough data at local levels can be tough. Many areas do not have enough information to make reliable Estimates, which creates challenges for understanding the real situation. One possible way to overcome this issue is to share information between different areas within the same country. Unfortunately, incorporating this information into existing models can be complex and slow.

To address this problem, we propose a straightforward approach to include data from different regions. Our method generates additional data points based on existing trends, allowing local areas to benefit from information available in other similar regions without adding extra complexity to the modeling process. This way, we can produce better estimates while using the combined information effectively.

#Importance of Local Data for HIV Estimation

Recent estimates show that global efforts to combat HIV have made some progress. New infections have declined significantly, and deaths related to AIDS have decreased as well. However, not all regions are experiencing these benefits equally. In high-burden countries, variations in HIV rates can be quite large, indicating that local data is crucial for understanding the true picture.

Many current methods for estimating HIV rates at local levels use only data from within that specific area. When data is scarce, these models can yield unreliable results. This leads us to consider that the parameters used in estimating these epidemics might be related across different regions. By using a hierarchical approach, we can borrow information from areas with similar epidemic patterns, thus improving our estimates.

The problem is that fitting these dynamic models to multiple areas can take a lot of time and computational resources. Therefore, we aim to find a more efficient way to incorporate this information without complicating the model too much.

#Proposed Method for Sharing Information

Our method aims to enhance local HIV estimates by combining various data sets without significantly increasing computational demands. Here’s how the process works:

1. Data Collection: For each region, we initially gather existing local data on HIV prevalence.

2. Generating Auxiliary Data: We then create additional data points, which we call auxiliary data, based on the trends observed in nearby regions. This auxiliary data helps to represent what might be expected in a given area based on what we observe in similar regions.

3. Model Fitting: With the combined data-local and auxiliary-we fit a local dynamic model. This model allows us to better estimate HIV rates, drawing on the trends of neighboring areas.

This process ensures that the data from similar areas can influence local estimates while keeping the model simple and computationally feasible.

#Understanding HIV Surveillance Data

HIV sentinel surveillance is a method used to collect information about HIV prevalence from specific groups in selected locations. This data helps track changes in HIV rates over time. Typically, areas are chosen based on the levels of HIV, the presence of key populations, and accessibility.

Data collection occurs at these sites annually or every few years, focusing on certain populations such as pregnant women, key populations, or others at risk. From this data, we can glean vital information about how HIV is spreading in different regions and among various groups.

HIV epidemics can generally be categorized into two types: generalized epidemics, where a significant portion of the general population is affected, and concentrated epidemics, where the disease primarily affects specific high-risk groups. Understanding these differences is crucial for targeting interventions appropriately.

#Estimating HIV Epidemics

One of the major models used for estimating HIV rates is the Estimation and Projection Package (EPP), which has been utilized globally to understand HIV's impact on populations. The EPP model looks at various data points and aims to predict future trends in infection and death rates.

The model primarily focuses on adults aged 15 to 49 and divides this group into those who are susceptible to infection and those who are already infected. Various parameters are estimated within the model, such as the rate at which new infections occur and how many people die from HIV-related causes.

This model has been effective in providing national estimates, but when it comes to local data, the application becomes more challenging. Different regions might have differing behaviors, risks, and interaction patterns, which can complicate estimates.

#Adding Hierarchical Structures into Estimation Models

Currently, the common method to extend the EPP model for local areas is to apply it independently, using only local data. This can lead to issues where areas with limited data provide unreliable estimates.

Our goal is to improve accuracy in places where data is sparse while retaining the model's simplicity. The key concept here is to pool data from all regions to create a broader picture. By applying a generalized linear mixed model (GLMM), we can incorporate time trends and area-specific effects into our estimates.

#Steps in Our Approach

1. Pooling Data: We gather data from all regions and fit a GLMM that captures the overall trend while allowing for variations among different areas.

2. Using Predictions: From this model, we derive estimates of HIV prevalence for each area and utilize them in the EPP model as auxiliary data.

3. Adjusting the EPP Model: The EPP model for each area is then fitted using this auxiliary data, allowing for better estimates without overwhelming computational demands.

This method allows each area to benefit from shared information while providing an effective way to estimate HIV rates using available data.

#Model Validation and Performance

To ensure our model works effectively, we conduct validation through a cross-validation process. This involves splitting data into training and test sets, applying our model to the training data, and then evaluating its accuracy against the test data.

For regions with richer data, we assess how well our model predicts actual observed rates. By tracking metrics such as mean absolute error (MAE) and the coverage of prediction intervals, we can understand the performance of our model.

A particular focus is on how much the introduction of auxiliary data contributes to improved predictions. In some cases, we have observed significant reductions in error rates, suggesting that sharing information across regions yields better estimates than relying solely on local data.

#Examples: Nigeria and Thailand

To illustrate our approach, we examined HIV surveillance data from Nigeria and Thailand, representing two different types of epidemics.

#Nigeria

In Nigeria, where the HIV epidemic is generalized, the estimates showed improvement after adjusting our model to include auxiliary data. The results indicated a lower mean absolute error in the predictions, signifying a closer match to observed data.

#Thailand

Conversely, in Thailand, where the epidemic is concentrated among specific high-risk groups, the use of auxiliary data similarly improved estimates. The adjustments helped capture fluctuations and trends more accurately than when using only localized data.

Both examples effectively demonstrated how our approach can enhance understanding and estimation of HIV prevalence at local levels.

#Challenges in HIV Surveillance Data

While collecting HIV data is crucial, challenges remain. The frequency of surveillance can vary, with some regions conducting annual checks and others doing so less frequently. This inconsistency can lead to gaps in data and challenges in modeling trends accurately.

Additionally, some regions may have site selection biases or missing data due to various factors. Over time, addressing these biases will be necessary to ensure that HIV estimates are truly representative of the entire population.

#Future Directions

The method we introduced is just a starting point. There are many ways this framework can be expanded. For example:

• Other Modeling Techniques: We used penalized splines, but other models might provide further insights into the data.

• Incorporating Additional Factors: Including socio-economic factors or geographic information can enhance models.

• Addressing Missing Data: More sophisticated methods for dealing with missing information can improve overall accuracy.

Overall, our proposed method aims to link different areas' data effectively while simplifying the modeling process. By continuing to enhance data-sharing techniques and methods for estimating HIV rates, we can better support local decision-making and resource allocation.

#Conclusion

Estimating HIV rates accurately at local levels is critical for effective public health responses. By sharing information across different regions while maintaining a simple modeling approach, we can provide better estimates that truly reflect the status of the epidemic. This is essential for guiding interventions and ensuring that resources are allocated where they can have the most significant impact.

The techniques discussed not only contribute to the understanding of HIV but can also provide principles applicable to other areas of public health, such as tracking other infectious diseases or health outcomes. As we strive to refine these models and methods, the overarching goal remains clear: to enhance outcomes and ultimately reduce the burden of HIV worldwide.