Deep Learning Approaches for Predicting Glaucoma Progression Using Electronic Health Records and Natural Language Processing

• DOI: 10.1016/j.xops.2022.100127
• Publication Date: 2022-02-12T07:48:44Z
• Authors: Sophia Y. Wang; Benjamin Tseng; Tina Hernandez-Bo...
• Keywords: Health records; Electronic health record

Advances in artificial intelligence have produced a few predictive models glaucoma, including logistic regression model predicting glaucoma progression to surgery. However, uncertainty exists regarding how integrate the wealth of information free-text clinical notes. The purpose this study was predict requiring surgery using deep learning (DL) approaches on data from electronic health records (EHRs), features structured and natural language processing notes.Development DL an observational cohort.Adult patients with at single center treated 2008 through 2020.Ophthalmology notes were identified EHRs. Available included patient demographic information, diagnosis codes, prior surgeries, intraocular pressure, visual acuity, central corneal thickness. In addition, words patients' first 120 days mapped ophthalmology domain-specific neural word embeddings trained PubMed abstracts. Word used as inputs subsequent surgery.Evaluation metrics area under receiver operating characteristic curve (AUC) F1 score, harmonic mean positive value, sensitivity held-out test set.Seven hundred forty-eight 4512 underwent that incorporated both well input achieved AUC 73% 40%, compared only features, (AUC, 66%; F1, 34%) 70%; 42%). All outperformed predictions specialist's review (F1, 29.5%).We can successfully which will need EHRs unstructured text. Models incorporating those inputs. Future should make use within improve performance. Additional research is needed investigate optimal methods imaging into future well.