NeoGx: Machine-Recommended Rapid Genome Sequencing for Neonates

Area of Science:

• Medical Informatics
• Genetics
• Neonatology

Background:

• Genetic diseases are prevalent in Level IV Neonatal Intensive Care Units (NICUs).
• Neonatology providers may not always recognize the need for genetic evaluation in neonates.
• Health record phenotypes can be used to train machine learning models for predicting genetic testing needs.

Purpose of the Study:

• To develop and validate a machine learning (ML) algorithm to predict the necessity of genetic testing for neonates.
• To assess the impact of ML-driven genetic testing predictions on diagnostic odyssey length and resolution time.

Main Methods:

• Extracted Human Phenotype Ontology (HPO) terms from clinical text using Natural Language Processing (NLP) for a decade of NICU patients.
• Trained and selected a classifier considering various feature sets, architectures, and hyperparameters.
• Validated the ML classifier on a cohort of 2,241 Level IV NICU admissions (born 2020-2021).

Main Results:

• The ML classifier achieved an ROC AUC of 0.87 and PR AUC of 0.73 for predictions within the first week of NICU admission.
• Accelerating initial genetic testing using ML reduced the median time to the first genetic test from 10 days to 1 day.
• ML prediction, coupled with rapid genetic sequencing (rGS), increased the resolution of diagnostic odysseys within 14 days by a factor of 3.8 compared to baseline.

Conclusions:

• Machine learning predictions can significantly accelerate genetic evaluation for neonates.
• Implementing ML tools aids providers in identifying the need for genetic testing.
• Earlier and targeted genetic testing leads to improved patient outcomes in the NICU setting.