Genomic risk profile scores (GRPSs) have been shown to predict case-control status of schizophrenia (SCZ), albeit with varying sensitivity and specificity. The extent to which this variability in prediction accuracy is related to differences in sampling strategies is unknown. Danish population-based registers and Neonatal Biobanks were used to identify two independent incident data sets (denoted target and replication) comprising together 1861 cases with SCZ and 1706 controls. A third data set was a German prevalent sample with diagnoses assigned to 1773 SCZ cases and 2161 controls based on clinical interviews. GRPSs were calculated based on the genome-wide association results from the largest SCZ meta-analysis yet conducted. As measures of genetic risk prediction, Nagelkerke pseudo-R(2) and variance explained on the liability scale were calculated. GRPS for SCZ showed positive correlations with the number of psychiatric admissions across all P-value thresholds in both the incident and prevalent samples. In permutation-based test, Nagelkerke pseudo-R(2) values derived from samples enriched for frequently admitted cases were found to be significantly higher than for the full data sets (Ptarget=0.017, Preplication=0.04). Oversampling of frequently admitted cases further resulted in a higher proportion of variance explained on the liability scale (improvementtarget= 50%; improvementreplication= 162%). GRPSs are significantly correlated with chronicity of SCZ. Oversampling of cases with a high number of admissions significantly increased the amount of variance in liability explained by GRPS. This suggests that at least part of the effect of common single-nucleotide polymorphisms is on the deteriorative course of illness.