BACKGROUND: There has been little examination of the stability and validity of polygenic risk scores (PRS) i.e., whether individuals identified as high-risk for a disorder with one PRS are identified as high-risk with another PRS, and whether high-risk individuals have the disorder. METHODS: The UK Biobank recruited 502,534 individuals aged 37-73 in the UK between 2006-2010. PRS were calculated for 408,853 white British individuals. PRS-CS, which uses SNP effect sizes under continuous shrinkage, was used to calculate three different PRS for major depressive disorder (MDD), alcohol use disorder (AUD), and type 2 diabetes (T2D), and two different PRS for schizophrenia (SCZ). PRS stability was measured using correlations between different PRS for the same disorder, and percentage of individuals consistently identified as high-risk (top 5% PRS). Sensitivity and specificity were used to measure PRS validity. RESULTS: Correlations between PRS ranged from low to high (SCZ: r=0.78
MDD: r=0.16-0.78
AUD: r=0.13-0.90
T2D r=0.29-0.77). Percentage of individuals consistently identified as high-risk (top 5% PRS) for schizophrenia with different SCZ PRS was 47.7%, i.e., less than half of individuals identified as high-risk with one PRS were identified as high-risk with another PRS. Percentages of individuals consistently identified as high-risk were 9.5-47.0% for MDD, 8.3-63.5% for AUD, and 14.1-45.2% for T2D. Sensitivity of PRS was moderate for MDD (66.1-74.4%) and AUD (72.3-74.2%), moderate/good for T2D (77.3-96.3%), and good for SCZ (90.2-93.3%). Specificity was low for all PRS (50.7-56.4%). CONCLUSIONS: Limited stability and specificity of PRS highlight their lack of clinical utility in psychiatry.