What evidence triangulation means
No observational method is bias-free. Triangulation is the discipline of combining approaches with different, largely unrelated sources of bias — then systematically comparing where they agree and interrogating where they don't.
Strengthening causal inference by integrating results from several different approaches, each with different and largely unrelated sources of potential bias.
— After Lawlor, Tilling & Davey Smith, Int. J. Epidemiology, 2016
Doubly-robust g-methods
Combine a model of the exposure (propensity score) with a model of the outcome — as in emulated target trials and structural nested models. Valid if either model is correct, and flexible for time-varying exposures.
Genetic & policy instrumental variables
Use extraneous variation in an exposure — genetic variants or policy differences — that is otherwise unrelated to the outcome. Remains valid even when not all confounders are measured.
Bias-detecting (reverse) Mendelian randomization
Use variants known to cause dementia to predict candidate risk factors, quantifying when reverse causation begins and how large a bias it introduces across adulthood — which in turn sharpens forward genetic-IV studies.
Quantitative bias analysis
Use known associations with confounders, selection, and measurement error to place bounds on effect estimates and show how conclusions shift under alternative, explicitly stated assumptions.