Benchmarking genetic birth prevalence estimates against newborn screening data

Accurate prevalence estimates for rare congenital conditions are critical for understanding disease epidemiology and enabling drug development. They inform public health investment, identify communities with high disease burden or underdiagnosis, and highlight unmet clinical need. Large biobanks have enabled genetics-based models to estimate disease prevalence. Autosomal-recessive (AR) diseases are particularly suited to this approach, as birth prevalence can be inferred from pathogenic allele frequency in unaffected populations; however, this approach has not been validated against real-world clinical datasets at scale. Newborn screening (NBS), which tests for neonatal diseases using quantitative diagnostic methods, provides a uniquely robust comparator for birth prevalence with low ascertainment bias, large sample size, and low diagnostic variability. With the objective of benchmarking a commonly used approach for determining disease prevalence, we applied a genetic model to estimate birth prevalence for 28 AR diseases from NBS panels and compared them with reported birth prevalence from 23 million newborns in the United States. We found that concordance between the genetic estimate and NBS was impacted by the source of allele frequency estimates, ancestry-matching methodology, and pathogenic variant inclusion criteria. Incorporating these refinements, we demonstrate that a genetics-first approach can provide first-order estimates of AR disease birth prevalence for 25 of 28 NBS diseases (89%). However, we note a general underestimate of the genetic prevalence, suggesting that identifying additional sources of pathogenicity would improve concordance with NBS. Further, we assessed the impact of epidemiological and genetic variables, highlighting diseases where genetic prevalence estimates may not be accurate.