Clarifying questions about "risk factors": predictors versus explanation

Abstract

Background: In biomedical research much effort is thought to be wasted. Recommendations for improvement have largely focused on processes and procedures. Here, we additionally suggest less ambiguity concerning the questions addressed.

Methods: We clarify the distinction between two conflated concepts, prediction and explanation, both encompassed by the term "risk factor", and give methods and presentation appropriate for each.

Results: Risk prediction studies use statistical techniques to generate contextually specific data-driven models requiring a representative sample that identify people at risk of health conditions efficiently (target populations for interventions). Risk prediction studies do not necessarily include causes (targets of intervention), but may include cheap and easy to measure surrogates or biomarkers of causes. Explanatory studies, ideally embedded within an informative model of reality, assess the role of causal factors which if targeted for interventions, are likely to improve outcomes. Predictive models allow identification of people or populations at elevated disease risk enabling targeting of proven interventions acting on causal factors. Explanatory models allow identification of causal factors to target across populations to prevent disease.

Conclusion: Ensuring a clear match of question to methods and interpretation will reduce research waste due to misinterpretation.

Keywords: Cause; Confounding; Predictor; Risk factor; Scientific inference; Selection bias; Statistical inference.