APOE plays a central role in research on lipids and cardiovascular disease, neurodegenerative disorders such as Alzheimer’s disease, and aging and survival more broadly (Corder et al., 1993; Deelen et al., 2011). Despite this extensive literature, studies of APOE and mortality have produced heterogeneous results across studies, with reports of protective, harmful, or null associations depending on study design and population (Wolters et al., 2019). An important contributor to this heterogeneity is the strongly age-dependent nature of APOE effects. For several key outcomes, including cognition, cognitive decline, and Alzheimer’s disease risk, associations with APOE at younger ages are often weak, inconsistent, or directionally heterogeneous, with evidence suggesting antagonist pleiotropy or context-dependent effects rather than uniform risk (Weissberger et al., 2018; Remer et al., 2020; Reynolds et al., 2019; Chang et al., 2016). Associations tend to increase markedly in later adulthood and may attenuate again at very old ages. Differences in age at study entry, follow-up windows, and choice of time scale can therefore generate divergent estimates even under otherwise similar analytical strategies.

This heterogeneity is often attributed to biological complexity, pleiotropy, or context-dependent effects. While these explanations are plausible, they obscure an equally important source of inconsistency: differences in analytical choices. We argue that associations between APOE and mortality are highly sensitive to modeling decisions, making reproducibility, rather than biological uncertainty, the central challenge.

This issue has important implications for meta-analyses, interpretation of genetic risk factors, and the generalizability of findings in genetic epidemiology. Without transparent and comparable analytical frameworks, it becomes difficult to distinguish genuine biological differences from artifacts of study design and modeling (Ioannidis, 2008).

These issues mirror broader concerns about reproducibility in biomedical research, where analytic flexibility and heterogeneous reporting practices have been identified as key contributors to irreproducible findings. Studies of APOE and mortality thus exemplify how the replication crisis in genetic epidemiology is driven not only by limited power or biological complexity, but by avoidable differences in modeling and reporting.

A substantial source of irreproducibility in studies of APOE, morbidity, and mortality arises not from biology, but from inconsistent variable definitions and reporting practices. Even when similar cohorts and outcomes are analyzed, differences in how APOE genotypes are handled, which covariates are included, and which models are reported can lead to results that are not meaningfully comparable across studies (Hernán and Robins, 2016).

Although this Perspective is motivated by the extensive literature on APOE, morbidity, and mortality, the issues raised are not specific to this locus. Rather, APOE illustrates a broader challenge in genetic epidemiology: how modeling choices, variable definitions, and reporting practices shape reproducibility and cumulative inference.

A central implication is that irreproducibility often reflects differences in estimands rather than contradictory evidence. Studies that differ in genotype inclusion, covariate adjustment, time scales, or outcome definitions may address fundamentally different questions, even when nominally investigating the same association (Greenland et al., 1999; Westreich and Greenland, 2013). Although much of the contemporary APOE literature now converges on broadly similar baseline specifications, the persistence of heterogeneous modeling choices even in this comparatively mature field illustrates how sensitive genetic associations remain to analytical decisions.

These considerations are increasingly relevant as genetic epidemiology moves toward larger consortia, meta-analyses, and polygenic approaches. Polygenic scores aggregate effects across many loci, many of which are pleiotropic and subject to similar modeling sensitivities as APOE (Wray et al., 2014; Boyle et al., 2017; Mostafavi et al., 2020). If reproducibility is compromised at the level of single, well-characterized loci, it is unlikely to improve when complexity is scaled up.

More broadly, concerns about reproducibility extend beyond genetic epidemiology and reflect structural challenges in modern biomedical research (Ioannidis, 2005; Peng, 2011; Munafò et al., 2017). In this sense, the challenges illustrated by APOE–mortality studies are not exceptional, but representative of the broader replication crisis in genetic epidemiology, where flexible modeling choices and inconsistent reporting continue to undermine cumulative inference.

In conclusion, APOE serves as a stress test for genetic epidemiology. Its well-established biology, strong effects on multiple disease pathways, and sensitivity to selection and modeling decisions expose weaknesses in current analytical practice. Addressing these weaknesses requires that baseline models prioritize reproducibility, transparency, and comparability over statistical sophistication within individual studies. Concretely, improving reproducibility in studies of APOE and mortality requires that all observed APOE genotypes be included in primary analyses and that simple baseline models, adjusted only for age, sex, and ancestry-related variables used to account for population structure, be routinely reported alongside more complex specifications.