Improving prevalence estimation through data fusion: methods and validation

Area of Science:

• Epidemiology
• Biostatistics
• Health Services Research

Background:

• Traditional health prevalence estimation relies on single surveys.
• Integrating multiple data sources via data fusion offers enhanced insights.
• This study validates data fusion for improving cardiovascular disease and diabetes prevalence estimates.

Purpose of the Study:

• To validate data fusion for improving health prevalence estimation.
• To compare different imputation methodologies for data fusion.
• To assess the impact of single vs. multiple imputation frames.

Main Methods:

• Utilized a data fusion operation on population health data.
• Enriched a large-scale survey with imputed health risk factors.
• Compared k-nearest neighbor, probabilistic modeling, and regression imputation techniques using validation statistics.

Main Results:

• Multiple imputation yielded more stable and accurate estimates than single imputation.
• Probabilistic methods best preserved multivariate data structure.
• Sequential regression methods provided higher accuracy, while nearest neighbor methods offered more realistic data distributions.

Conclusions:

• Data fusion effectively integrates diverse data sources for comprehensive analysis.
• Multiple imputed sequential regression models offer superior interpretability for health policy.
• Fused data can yield more accurate prevalence estimates than observed data alone.