Determining whether a vaccine’s protective effect decreases over time — vaccine waning — matters for clinical decisions and public-health policy. The authors argue that standard methods to quantify waning frequently rely on assumptions that are difficult to justify, which undermines the interpretability of estimated declines in efficacy. To address this, they propose a different objective: rather than attempting to estimate the full time-varying efficacy curve under strong assumptions, formulate a formal hypothesis test for whether the individual-level treatment effect is constant over time. This reframing focuses on detecting the presence of waning with higher statistical power while maintaining interpretable assumptions appropriate for randomized vaccine trials.
Classical analytical approaches to vaccine waning typically require assumptions that the authors describe as unreasonable if one wants to interpret results as reflecting individual-level declines in protection. In response, recent methodological work has introduced causal estimands specifically designed to quantify waning effects under a formal causal framework. Those estimands can be bounded under weaker, more plausible assumptions; however, the resulting bounds often remain wide, which limits their utility for making definitive claims about whether waning occurs in a given dataset.
The core contribution is a formal test that evaluates whether the treatment effect is constant over time for individuals. The null hypothesis is that the effect does not vary with time since vaccination; rejection of that null indicates evidence of waning. The test is constructed to be valid under assumptions the authors consider interpretable in the context of vaccine trials. By directly targeting the hypothesis of constancy versus time-varying effect at the individual level, the test avoids some of the inferential difficulties that complicate direct estimation of waning curves.
The paper presents three distinct approaches to compute the test statistic. Two of these approaches are notable because they require only summary data that are typically available from completed clinical trials, which increases the test’s practicality and accessibility for secondary analyses. The third approach uses more detailed data and can be applied when individual-level information is available. The authors compare these computational approaches and describe their use in both simulated and real-data contexts.
Compared with existing approaches — including methods based on bounding causal estimands — the proposed test provides a considerable gain in statistical power to detect departure from constant effect over time. The authors illustrate this power advantage using simulations and real examples, showing that the test can identify waning in settings where other methods may lack sensitivity. The increased power stems from the test’s focus on the hypothesis of individual-level nonconstancy and its construction under assumptions tailored to randomized vaccine trial designs.
As an illustrative real-world application, the authors reanalyze data from a randomized controlled trial of the BNT162b2 COVID-19 vaccine. According to the abstract, a prior analysis of these trial data did not establish evidence of waning. Using the new test, the authors report rejection of the null hypothesis of no waning, indicating that their method detected time-varying vaccine effect where the earlier analysis did not. The abstract does not provide numerical test statistics, p-values or effect-size estimates; those specifics are reported in the full article.
In addition to the hypothesis test, the authors provide new analytical results that place bounds on the possible magnitude of the waning effect. These bounds are intended to supplement hypothesis testing by offering constrained estimates of how large waning could be under stated assumptions. The abstract notes that bounds derived from causal estimand approaches can be wide; the authors’ additional results aim to improve interpretability and inference regarding effect size.
A practical strength of the proposed methodology is that two of the three computation strategies rely only on summary trial data, enabling retrospective application to existing trial reports. When individual-level data are accessible, the third approach can be used and may provide additional precision. The validity of the test depends on assumptions that the authors consider interpretable for randomized vaccine trials; the abstract emphasizes that the test’s validity and increased power are achieved under these reasonable assumptions. Specific details about required assumptions, implementation steps, and example code are not provided in the abstract and are available in the full text.
The authors conclude that their formal hypothesis test offers a simple and powerful tool to assess individual-level vaccine waning. It provides considerably greater power than some existing methods, is valid under interpretable assumptions for vaccine trials, and can be computed using either summary or individual-level data depending on availability. Applied to a trial of BNT162b2, the test detected waning where a prior analysis had not. These features make the test a useful addition to the methodological toolbox for analyzing time-varying vaccine effects in randomized trials and for informing decisions on booster policies and vaccine strategies.