A hybrid deep learning approach for phenotype prediction from clinical notes

Identifying patient cohorts from clinical notes in secondary use of electronic health records is a fundamental task in clinical information management. The patient cohort identification process requires identifying the patient phenotypes. However, the growing number of clinical notes makes it difficult to analyze the data manually. Therefore, automatically extracting clinical concepts is crucial to identify patient phenotypes correctly. This paper's proposed hybrid model is based on a neural bidirectional sequence model (BiLSTM or BiGRU) and a Convolutional Neural Network (CNN) for identifying patients' phenotypes in clinical notes. Furthermore, an extra CNN layer is run parallel to the hybrid proposed model to extract more features related to each phenotype. We used pre-trained embeddings such as FastText and Word2vec separately as the input layers to evaluate other embedding's performance in identifying patient phenotypes. We also measured the effect of adding additional data cleaning steps on discharge reports to identify patient phenotypes using deep learning models. Results demonstrated the proposed hybrid model extracts more features than existing methods of patient phenotype extraction and provides a better F1-score. We show that complementing the proposed hybrid model with an extra CNN in identifying different phenotypes improves the F1 scores. In addition, eliminating punctuation, numbers, and stop words in discharge reports before training hybrid models increased model performance.