OBJECTIVES: Our objective was to measure the performance of an explainable artificial intelligence (AI), the Literature Review Network (LRN), to replicate health economics and outcomes research findings through systematic literature review (SLR) and meta-analysis. Vaccines are regarded as affordable and effective interventions, yet they collectively are under-evaluated for their economic value from U.S. payer and societal perspectives. We applied LRN to address an important, unanswered question, “What is the Cost-effectiveness of Vaccines for Infectious Diseases in the United States?”

METHODS: Two independent investigators developed search strategies and employed LRN version 1.5 to query studies indexed in PubMed. LRN utilized proprietary generative and discriminative algorithms for data extraction and classification and underwent 3 iterations of reinforcement learning. The effectiveness of LRN was evaluated using explainability and performance metrics. Parameters utilized by LRN were quantified as correlations. LRN then extracted and tabularized quality-adjusted life years (QALYs) gained and cost per QALY for the meta-analysis using researchers’ prompts. LRN calculated composite incremental cost-effectiveness ratios (ICERs), overall and by therapeutic area, for a forest plot.

RESULTS: In 240 minutes, the explainable AI achieved high performance for the inclusion criteria (recall = 89.19%, precision = 63.46%). LRN processed 850 studies, selecting 154 for the SLR and meta-analysis. Significant interactions (FDR-adjusted p-value < 0.05) were identified by LRN between concepts like QALYs and vaccines. LRN determined most vaccines were cost-effective and calculated composite ICERs from the identified studies, visualized in a forest plot.

CONCLUSIONS: Vaccines were found to be generally cost-effective health technologies in the U.S. at conventional willingness-to-pay thresholds. Our findings suggest that a preferential status of vaccines on formularies may generate health benefits as well as cost savings. Meanwhile, explainable AI such as LRN can be accurate and efficient tools for conducting SLRs and evaluating the effectiveness of health technologies to advance population health objectives.