IDENTIFYING SERIOUSLY ILL DEMENTIA PATIENTS FOR EARLY PALLIATIVE CARE INTERVENTIONS. (16th November 2018)

• Record Type: Journal Article
• Authors: Wang, L; Sha, L; Lakin, J; Hong, P; Bates, D; Zhou, L
• Is Part Of: Innovation in aging. Volume 2(2018)Supplement 1
• Page Start: 950
• Page End: 950
• Publication Date: 2018-11-16
• Subjects: Aging -- Periodicals; Gerontology -- Periodicals; 612.67
• DOI: 10.1093/geroni/igy031.3525
• Languages: English
• ISSNs: 2399-5300

Abstract

Abstract: Early palliative care interventions drive high-value care, improving patient and family experiences while ensuring more efficient use of healthcare resources. However, palliative care is currently an underutilized resource in dementia care. One reason for this is challenging identification of dementia patients for early palliative care. The objective of this work is to develop a mortality prediction model as a proxy for identifying patients with advanced, progressive Alzheimer's Disease and Related Dementias (ADRD) for care interventions. A retrospective cohort of 8, 569 patients with ADRD were identified from our local (Partners Healthcare) longitudinal electronic health records (EHR) using a collection of ICD billing codes with at least 10 days of notes between 2011–2017. We obtained death information from Massachusetts death certificate files and the EHR and used it to label the dataset. We used latent Dirichlet allocation topic modeling and a natural language processing tool, called MTERMS, to generate topics from clinical notes. We subsequently took the topics and demographics information as features, developed a 12-month mortality prediction model using a recurrent neural network. The best model reached an area under the receiver operating characteristics (AUROC) curve of 0.915. There is no significant difference in AUROC values by changing the number of topics generated from the clinical notes. Among hundreds of topics, top ranked predictive factors associated with one-year mortality are among healthcare utilization (e.g., hospitalization, healthcare encounters, intensive care management program activity, emergency department visits), clinical problems (e.g., cancer, pain, anxiety, respiratory infections), surgical procedures, and medication delivery.