Abstract
Identifying significant associations between genetic loci and psychiatric disorders is dependent on very large sample sizes. Methods for diagnosing diseases on this scale, such as the use of self-assessment questionnaires and data from electronic health records, incorporate heritable variation unrelated to the disease of interest into the diagnosis. Consequently, genetic mapping will identify loci unrelated to the target disease while missing some that are related, and genetic correlations cannot be used to infer the genetic relationships between diseases and between cohorts. Furthermore, shared biases between different disorders appear as shared etiology. As sample sizes grow, such confounders propagate, and findings based on their presence are replicated and extended. Here, we draw attention to the problem, make suggestions for flagging affected cohorts, and discuss future data collection and machine learning approaches to mitigate the effects of heritable confounders in psychiatric disorders.