Predictive value of genomic screening: cross-sectional study of cystic fibrosis in 50,788 electronic health records

Abstract

Doubts have been raised about the value of DNA-based screening for low-prevalence monogenic conditions following reports of testing this approach using available electronic health record (EHR) as the sole phenotyping source. We hypothesized that a better model for EHR-focused examination of DNA-based screening is Cystic Fibrosis (CF) since the diagnosis is proactively sought within the healthcare system. We reviewed CFTR variants in 50,778 exomes. In 24 cases with bi-allelic pathogenic CFTR variants, there were 21 true-positives. We considered three cases "potential" false-positives due to limitations in available EHR phenotype data. This genomic screening exhibited a positive predictive value of 87.5%, negative predictive value of 99.9%, sensitivity of 95.5%, and a specificity of 99.9%. Despite EHR-based phenotyping limitations in three cases, the presence or absence of pathogenic CFTR variants has strong predictive value for CF diagnosis when EHR data is used as the sole phenotyping source. Accurate ascertainment of the predictive value of DNA-based screening requires condition-specific phenotyping beyond available EHR data.

Keywords: Genetic techniques; Personalized medicine.

Fig. 1

EHR-based Genomic Screening for CF. In a database of 50,778 participants, there were 50,754 negatives, namely cases without bi-allelic pathogenic CFTR variants, and 24 positives, namely cases with such variants. Amongst the negatives, eight had EHR data suggesting a CF diagnosis, six of those had no CFTR pathogenic variants, and two had a single pathogenic variant. Open chart review concluded that one of the heterozygotes was a false negative and one was a true negative. The remainder of the negatives, those without bi-allelic CFTR variants or EHR data consistent with CF were also considered true negatives. Open chart review was pursued for all 24 positive cases. The diagnosis of CF was confirmed in 21 cases. In the remaining three cases the diagnosis could not be confirmed due to insufficient available evidence

Fig. 2

Data concordance in clinical and research CFTR variant results. Twenty of the 22 confirmed CF patients in the cohort also received their care in the Geisinger CF clinical program and had clinical CF genetic testing results available in their EHR. In the 20 cases with DNA data from both sources there was 100% pathogenic variant calling concordance