Presenter of 1 Presentation
O044 - USING LASSO REGRESSION TO ESTIMATE THE POPULATION-LEVEL IMPACT OF PNEUMOCOCCAL CONJUGATE VACCINES (ID 572)
Abstract
Background
The pneumococcal conjugate vaccines (PCVs) protect against diseases caused by Streptococcus pneumoniae, such as meningitis, bacteremia, and pneumonia. Estimating the population-level impact – including direct and indirect effects – of PCVs is challenging because of the lack of a perfect control population and the subtleness of signals when the outcome – like all-cause pneumonia – is attributable to a wide range of pathogens. Here we present a new approach to estimate PCVs’ impacts – using LASSO regression to predict the counterfactual outcome in different age groups for vaccine impact inference.
Methods
First, we designed a simulation study to test the performance of LASSO regression and three established methods – interrupted time series (ITS), synthetic control (SC), and seasonal-trend decomposition plus PCA (STL+PCA) – by comparing their ability to estimate the pre-specified vaccine impact. Then we applied LASSO to published pneumonia hospitalization data from Chile, Ecuador, Mexico, and the US.
Results
In the simulations, we found that both LASSO and SC achieved accurate and precise estimation, with high coverage (LASSO: 96–100%; SC: 69–89%); while estimates from ITS and STL+PCA were sometimes biased, with variable coverage (ITS: 0–100%; STL+PCA: 0–100%). The performance of LASSO and SC remained robust in complex simulation scenarios where the association between outcome and all control variables was non-causal. When applied to real data, we found that LASSO yielded similar estimates of vaccine impact to SC.
Conclusions
The LASSO method is accurate, easily implementable and interpretable. In complement to existing methods like SC, LASSO can be used to study the population-level impact of PCVs and other vaccines.