Protecting against researcher bias in secondary data analysis: Challenges and solutions

Protecting against researcher biases – both conscious and unconscious – can help to ensure robust findings and correct inferences in epidemiology. While pre-registration can be an effective way to achieve this, it brings several challenges for researchers analysing existing datasets. Here we describe these challenges, and propose solutions and alternatives. For each solution, we provide guidance, and highlight practical considerations for researchers. The adoption of these practices will allow researchers to effectively pre-register secondary data analysis studies, or use an alternative approach, in order to protect themselves against common human biases. In turn, this will increase the robustness and credibility of epidemiological research based on secondary data.