0906 Free-text Documentation of Sleep Disturbance in Acute Myeloid Leukemia: A Natural Language Processing Study

Abstract Introduction Sleep disturbance in acute myeloid leukemia (AML) patients is usually linked to cancer-related fatigue, affecting quality of life. Although addressing sleep disturbance might prevent fatigue, it is often overlooked. Limited knowledge about healthcare clinician’s observations of sleep disturbance in AML patients hampers evidence-based interventions. To address this gap, this study identified sleep disturbance documentation in patients with AML from free-text narrative notes in electronic health record (EHR) using a natural language processing (NLP). Methods The sample consisted of 692 patients with AML, who received 2,064 encounters at Midwestern academic hospital. We followed a stepwise process using NLP. 1) Identify a preliminary list of synonyms based on (a) 57 common symptoms from a previous study; (b) relevant literature; and (c) “synonyms” category of the Unified Medical Language System. We developed and validated algorithms using the publicly available NLP system, NimbleMiner, then an NLP algorithm was applied to extract sleep disturbance documentation using 316,038 EHR narrative notes of AML patients. The overall accuracy to identify documentation of sleep disturbance was high, with F score = 0.97. Results In general, this study involves older AML patients (average age, 59 years) and male (57%). When applying the NLP algorithm, 543/692 (78.5%) patients and 1,255/2,064 (60.8%) encounters in this sample were identified as having notes with language describing sleep disturbance. The most common language describing sleep disturbance was sleeping medication (40.5%), such as melatonin, olanzapine, zolpidem. Additionally, around 30% of languages mentioned sleep disorders including insomnia and obstructive sleep apnea (OSA). Conclusion We have demonstrated the feasibility of identifying sleep disturbance from EHR narrative notes with NLP in AML patients. While using sleeping pills are found as a common treatment option for sleep disturbance, it is important to consider the established first-line treatment for insomnia (cognitive behavioral therapy) and OSA (continuous positive airway pressure). Utilizing NLP-generated symptom indicators can assist in identifying high-risk patients for targeted sleep interventions. The impact of sleep disturbance on health outcome in AML patients should be further evaluated. Support (if any) This study was funded by Center for Advancing Multimorbidity Science: NIH/NINR 1 P20 NR018081-01 (MPI: Gardner & Rakel)