Challenges in Estimating the Impact of Vaccination with Sparse Data

Area of Science:

• Epidemiology
• Biostatistics
• Public Health

Background:

• Synthetic control models are used to assess vaccine impact by adjusting for unrelated disease trends.
• Sparse, subnational time series data pose challenges for traditional synthetic control models.
• A novel approach, STL+PCA (seasonal-trend decomposition plus principal component analysis), was developed to address data sparsity.

Purpose of the Study:

• To evaluate the performance of synthetic control models with sparse time series data.
• To compare the accuracy of synthetic control models versus the STL+PCA method for estimating vaccine impact.
• To assess the impact of the 10-valent pneumococcal conjugate vaccine on pneumonia hospitalizations in Brazil.

Main Methods:

• Employed both synthetic control and STL+PCA models to estimate vaccine impact.
• Utilized subnational data from Brazil (2004-2014) for pneumonia hospitalizations in infants and the elderly.
• Conducted simulation analyses to compare model performance under varying data sparsity.

Main Results:

• Synthetic control models showed bias and low accuracy in smaller states with sparse data.
• STL+PCA analysis demonstrated significantly reduced bias (90% lower) and improved accuracy (95% credible intervals covering true estimates).
• The 10-valent pneumococcal conjugate vaccine's impact was estimated using both methods, with STL+PCA providing more reliable results.

Conclusions:

• Synthetic control models may yield biased estimates with sparse epidemiological data.
• The STL+PCA method provides a more robust and accurate approach for evaluating vaccine effectiveness in smaller populations.
• STL+PCA enhances the reliability of public health surveillance and intervention impact assessments.