Triangulation of epidemiological evidence and risk of bias evaluation: A proposed framework and applied example using formaldehyde exposure and risk of myeloid leukemias

Evidence triangulation may help identify the impact of study design elements on study findings and to tease out biased results when evaluating potential causal relationships; however, methods for triangulating epidemiologic evidence are evolving and have not been standardized. Building upon key principles of epidemiologic evidence triangulation and risk of bias assessment, and responding to the National Academies of Sciences, Engineering, and Medicine (NASEM) call for applied triangulation examples, the objective of this manuscript is to propose a triangulation framework and to apply it as an illustrative example to epidemiologic studies examining the possible relationship between occupational formaldehyde exposure and risk of myeloid leukemias (ML) including acute (AML) and chronic (CML) types.

A nine-component triangulation framework for epidemiological evidence was developed incorporating study quality and ROB guidance from various federal health agencies (i.e., US EPA TSCA and NTP OHAT). Several components of the triangulation framework also drew from widely used epidemiological analytic tools such as stratified meta-analysis and sensitivity analysis. Regarding the applied example, fourteen studies were identified and assessed using the following primary study quality domains to explore potential key sources of bias: 1) study design and analysis; 2) study participation; 3) exposure assessment; 4) outcome assessment; and 5) potential confounding. Across studies, methodological limitations possibly contributing to biased results were observed within most domains. Interestingly, results from one study – often providing the largest and least-precise relative risk estimates, likely reflecting study biases, deviated from most primary study findings indicating no such associations. Triangulation of epidemiological evidence appears to be helpful in exploring inconsistent results for the identification of study results possibly reflecting various biases. Nonetheless, triangulation methodologies require additional development and application to real-world examples to enhance objectivity and reproducibility.