Home healthcare serves a growing population of older adults with complex chronic conditions. Identifying patients at high risk for emergency department visits or hospitalizations is crucial for targeting preventive interventions, yet current risk prediction models relying on electronic health record (EHR) data have suboptimal performance. Patient-nurse verbal communication during home visits may provide a rich additional data source to enhance these models.
This presentation series explores the potential of integrating verbal communication data into home healthcare risk prediction. Dr. Song first introduces the topic, presenting a pilot study that refined natural language processing (NLP) algorithms to identify hospitalization and ED visit risk factors in patient-nurse conversations. The algorithms performed well on both human and auto-generated transcripts, with F1-scores over 0.79.
Next, Dr. Scroggins discusses using GPT-4 to generate synthetic patient-nurse conversations and automatically annotate them for common health problems. Augmenting real-world data with this synthetic data improved machine learning classifier F1-scores by an average of 7%, with the greatest gains for poorly-performing health problem categories.
Finally, Dr. Zolnoori presents a study integrating verbal communication features with structured EHR data and clinical notes in a risk prediction model. The best model achieved an AUC-ROC of 0.997 and F1-score of 0.941, with verbal communication features improving the F1-score by 26%. High-risk patients exhibited more interactions with risk-associated cues, greater sadness and anxiety, and longer silences.
Together, these studies highlight verbal communication as a critical missing data stream in current home healthcare risk models. Integrating NLP-extracted communication features, potentially augmented by synthetic data, could substantially improve model performance. Implementing routine communication recording into home healthcare workflows may enable more accurate risk prediction and targeted preventive care.
Presenters
Jihye Scroggins, PhD, RN
Jiyoun Song, PhD, RN
