Association Mining with Clinical Data: Phenotype-Wide Association Studies (PheWAS)

This chapter covers association mining methods that are used to perform phenotype-wide association studies (PheWAS) using electronic health records (EHR) datasets. PheWAS and association mining are popular methods that are used throughout the research community, and in health analytics. We will describe how when utilizing a PheWAS one can explore a variety of hypotheses and their impact on the phenotype/condition/outcome of interest. This approach is considered a hypothesis generating method because the results of these analyses lend their way to future research studies that investigate the findings in greater detail. Because of the wide variety of hypotheses tested as part of a PheWAS, we will also discuss methods for multiple-hypothesis correction that adjust for this increase in hypotheses tested. Popular multiple-hypothesis testing correction methods covered in this chapter include Bonferroni, Holm’s method, Benjamini-Hochberg, and false discovery rate correction. We will also cover popular visualization methods, including Manhattan plots for visualizing the resulting p-values and forest plots for visualizing the effect estimates (e.g., odds ratios). We will also cover how to adjust for comorbidities within a PheWAS study. After reading this chapter, you should be able to confidently answer the following questions: