2.4. Mendelian randomization analysis

In this study, circulating proteins were selected as exposure, and aging proxy indicators (TL, FA, FI, and 90th/99th survival percentile) were selected as the outcome. MR was performed with “TwoSampleMR.” In instances where a single pQTL was available for a protein, we utilized the Wald ratio. However, when multiple genetic instruments were at our disposal, we applied the inverse variance‐weighted MR (MR‐IVW) (Deng et al., 2022), subsequently followed by a heterogeneity analysis. The odds ratio (OR) for the augmented risk of aging‐related traits is denoted by the increase in the standard deviation (SD) of plasma protein levels.

To deem an instrumental variable valid, it must satisfy three fundamental assumptions, as illustrated in eFigure 1 in Appendix S2.

The genetic variation must have a strong association with the exposure. This association was verified by calculating the instrument's F‐statistic, employing the formula: squared beta divided by the squared standard error (Burgess & Thompson, 2011). To mitigate weak instrument bias, we sought an F‐statistic exceeding 10. The MR‐power online tool (Burgess, 2014) was instrumental in determining the robustness of our MR estimates.

The selected genetic variants should function independently of confounders that potentially influence the exposure‐outcome nexus. Leveraging the PhenoScanner database (Kamat et al., 2019), we identified traits—excluding the exposure—with significant ties to outcomes (TL, FA, and FI) at thresholds of p < 5 × 10⁻⁸. Literature review pinpointed traits such as depression, type 2 diabetes mellitus (T2DM), and ischemic stroke as potential risk factors for outcomes. To minimize horizontal pleiotropy, we precluded variants associated with these particular traits.

The genetic variants should impact the outcome strictly via the determined exposure. To counteract confounding from inherent population stratifications, we maintained methodological consistency by centering our analysis on a homogenous ancestry group. Additionally, linkage disequilibrium poses confounding risks. This motivated our adoption of a Bayesian co‐localization analysis, designed to ascertain the likelihood of genetic confounding (Giambartolomei et al., 2014).