Heritability, the proportion of variation in a particular trait that is attributable to genetic factors, is key to predicting disease risk. A recent study reports the use of electronic health records (EHRs), which capture a range of clinical data, as a resource for studying the heritability of traits.

Credit: Katie Edwards/Alamy

Given that heritability underlies the predictive nature of family history for a particular disease or trait, it has been traditionally studied in family cohort studies, in particular twin studies, which are often of limited sample size. The authors recognized the opportunity provided by EHRs as a large-scale source of familial relationship data. While EHRs contain no record of genetic relatedness between patients, they do contain emergency contacts, which the team used to map the relatedness within three hospital’s patient populations.

The researchers obtained 3,550,598 inpatient EHRs used at the hospitals of Columbia University Medical Center, Weill Cornell Medical Center and Mount Sinai Health System, which provided 6,587,594 emergency contacts. Of these family members, 2,191,695 were patients themselves; 1,902,827 of these provided an additional 1,588,134 emergency contacts who were next of kin. Using a newly developed algorithm for extracting relationships, called relationship inference from the electronic health record (RIFTEHR), the researchers could identify 7.4 million familial relationships across the three medical centres, without compromising patient privacy.

Clinical and genetic data were used to validate the relationships, with positive predictive values between 66% and 99% among first-degree relatives. For example, PLINK was used to estimate relatedness for 302 patients with RIFTEHR-predicted relationships from Columbia, who had consented to re-use of their genetic data. Of 172 predicted relationships, 134 were parent–offspring relationships, nearly all of which showed the expected genetic relatedness of 50%.

Based on the more than 560,000 inferred pedigrees and using only medical data from the EHRs, Polubriaginof et al. then estimated the heritability of 500 traits adjusted for age and sex using sequential oligogenic linkage analysis routines (SOLAR). The resultant heritability estimates were similar across the three centres and matched those of previously published studies. The study also resolves an ongoing debate regarding the relative heritability of plasma levels of different cholesterol types, reporting that HDL cholesterol levels are more heritable than LDL cholesterol levels.

Taken together, this study showcases the utility of (accurate) EHRs for disease genetics, even without the genetics.