Health risk prediction models incorporating personality data: Motivation, challenges, and illustration.

Abstract

The age of "big data" in health has ushered in an era of prediction models promising to forecast individual health events. Although many models focus on enhancing the predictive power of medical risk factors with genomic data, a recent proposal is to augment traditional health predictors with psychosocial data, such as personality measures. In this article we provide a general overview of the medical risk prediction models and then discuss the rationale for integrating personality data. We suggest three principles that should guide work in this area if personality data is ultimately to be useful within risk prediction as it is actually practiced in the health care system. These include (a) prediction of specific, priority health outcomes; (b) sufficient incremental validity beyond established biomedical risk factors; and (c) technically responsible model-building that does not overfit the data. We then illustrate the application of these principles in the development of a personality-augmented prediction model for the occurrence of mild cognitive impairment, designed for a primary care setting. We evaluate the results, drawing conclusions for the direction an iterative, programmatic approach would need to take to eventually achieve clinical utility. Although there is great potential for personality measurement to play a key role in the coming era of risk prediction models, the final section reviews the many challenges that must be faced in real-world implementation. (PsycINFO Database Record (c) 2019 APA, all rights reserved).