Statistically significant association does not imply improvement in prediction of clinical outcomes.

In the current landscape of clinical studies, the concept of statistically significant association is often mixed up with the expectation of improved prediction performance. We discuss the two concepts, association and prediction, and present the epidemiologic principles and statistical constructs that underlie the discrepancy between statistically significant associations and the rationale for their lack of impact on improving prediction in terms of discrimination. This issue is illustrated using an existing breast cancer dataset. The concept of statistically significant association should not be mixed up with the expectation of improved discrimination performance. While some markers may not markedly improve discrimination, they can still have substantial clinical relevance by identifying critical biological pathways that inform novel treatment or prevention strategies. Development of models for both association and prediction assessments should be directly tied to clinical translation to move adoption forward to advance precision medicine.