Test-negative designs with various reasons for testing: statistical bias and solution

• DOI: 10.48550/arxiv.2312.03967
• Publication Date: 2023-01-01
• Authors: Mengxin Yu; Kendrick Qijun Li; Nicholas P. Jewell; Eric Tchetgen Tch...; Dylan S. Small; Xu Shi; Bingkai Wang
• Keywords: Statistical power

Test-negative designs are widely used for post-market evaluation of vaccine effectiveness. Different from classical test-negative where only healthcare-seekers with symptoms included, recent have involved individuals various reasons testing, especially in an outbreak setting. While including these data can increase sample size and hence improve precision, concerns been raised about whether they will introduce bias into the current framework designs, thereby demanding a formal statistical examination this modified design. In article, using derivations, causal graphs, numerical simulations, we show that standard odds ratio estimator may be biased if testing not accounted for. To eliminate bias, identify three categories symptoms, disease-unrelated reasons, case contact tracing, characterize associated properties estimands. Based on our characterization, propose stratified estimators incorporate multiple to achieve consistent estimation precision by maximizing use data. The performance proposed method is demonstrated through simulation studies.