Excited to share new preprint "Phenotypic signatures in clinical data enable systematic identification of patients for genetic testing" led by @MorleyTheodore and with @NancyGenetics @lisa_bastarache @lhan111 @jonmorra8 medrxiv.org/content/10.110…
Genetic testing is the standard path to a diagnosis for people with a rare disease yet many who should receive a genetic test still do not or do so after years of diagnostic odyssey.
There are thousands of characterized genetic diseases with unique combinations of symptoms. However, many share a particular pattern of having multiple rare phenotypes often spanning organ systems.
Here, we propose that individuals more likely to have a genetic disease and therefore needing a genetic test can be identified from longitudinal clinical information in the electronic health record #EHR .
To test this hypothesis, we build a #MachineLearning model using only structured diagnostic information to predict whether a patient received a genetic test (CMA). We show that this model performs well in a held out dataset and in a large hospital population. 
We then assess the ability to identify patients with putative pathogenic CNVs in a subset of patients with genetic data. Patients with known CNV syndromes had very high probabilities.
Further, in comparison to the reports from the genetic tests, our approach identifies a higher proportion of patients that likely have genetic disease. While, also capturing many patients that have no clinical suspicion of genetic disease.
Finally, we show that while the model performs best on the diseases it was trained on, it still performance much better than expected on a diverse set of unrelated genetic diseases.
We believe this approach could improve the process for genetic testing making it more equitable and systematic which should lower costs and improve health care for those with a #RareDisease
Lots of work following up this effort including getting more genetic information for these patients to improve the model as well as to identify potential novel diseases. Happy to get thoughts on how to make this more impactful for patients.
