NEW YORK — Adding a polygenic risk score to clinical models of schizophrenia prognosis does not improve their predictive performance, a new study has found.
Schizophrenia is a highly polygenic disease and polygenic risk scores — which are thought to be a key aspect of personalizing medicine — can capture its genetic liability. However, whether polygenic risk scores can improve predictions of diseases or their outcomes in clinical practice is still under investigation.
Researchers led by the Icahn School of Medicine at Mount Sinai's Alexander Charney examined whether adding a polygenic risk score to clinical schizophrenia assessments could improve their ability to predict a range of outcomes. Charney noted in an email that schizophrenia polygenic risk scores are among the most powerful and explain more of the variance in disease than risk scores for other diseases can.
But as they reported Monday in Nature Medicine, the researchers' analysis of more than 8,500 individuals suggested that the addition did not improve the performance of clinical prediction models.
This outcome, Charney said, should not be surprising to people who have been doing research using polygenic risk scores, but could be to others who are familiar with such risk scores but haven't studied them personally. That, he added, is because polygenic risk scores tend to be described as a tool that is expected to have clinical utility. "There is a disconnect of sorts between the perspectives of people who work closely with PRS and the way PRS has been presented to the greater scientific community," he said.
For their analysis, Charney and his colleagues examined electronic medical record data from the BioMe biobank — a cohort of more than 30,000 individuals from New York's Mount Sinai Health System — to identify 762 individuals who had been billed for schizophrenia- or related psychosis-linked treatment and amassed clinical outcomes data. Of these, the researchers focused on six areas indicative of poor outcomes: inpatient treatment, prescription of multiple antipsychotics or of clozapine, aggressive or self-injurious behavior, and homelessness.
At the same time, the researchers generated a schizophrenia polygenic risk score using summary statistics from a genome-wide association study of the disease. This score could, with statistical significance, explain some of the variance in inpatient treatment and aggressive behavior.
But when the researchers examined how well models could predict these outcomes based on clinical features alone, clinical features plus the polygenic risk score, and polygenic risk score alone, the addition of the polygenic risk score did not improve predictive ability.
Additionally, in 7,779 cases from the multiethnic Genomic Psychiatry Cohort, the addition of a polygenic risk score also did not improve outcome predictions. For this cohort, the outcomes analyzed differed and were based on a validated checklist of symptoms, including suicidality, impairment, whether symptoms responded to anti-psychotic medications, and disease course.
For both cohorts, the researchers noted that the findings held across genetic ancestry groups.
These results suggested that a schizophrenia polygenic risk score does not improve predictions of patient outcomes better than information that is collected clinically. This, the researchers noted, is in line with findings from other studies and underscores the need for polygenic risk scores to not only be predictive but also provide additional information that cannot easily be gleaned elsewhere.
The researchers noted that their analysis relies on a polygenic risk score built from a genome-wide association study of schizophrenia, rather than an analysis of schizophrenia outcomes and that predictions using scores based on such studies might have performed better.
Charney added that he and his colleagues are now examining the relationship between the polygenic risk score and social determinants of health. Or put another way, he said, they are considering "the age-old question of nature versus nurture."