Abstract
Multimorbidity, cancer, and symptoms: Using electronic health record data to cluster patients in multimorbidity phenotypes
Journal of clinical oncology, Vol.37(31_suppl), pp.130-130
11/01/2019
DOI: 10.1200/JCO.2019.37.31_suppl.130
Abstract
130
Background: Cancer-related symptoms are associated with decreased quality of life, increased health care utilization, and shorter life expectancy. There is limited understanding of how multiple chronic conditions (MCC) contribute to variability in symptoms experienced in the context of cancer. Data mining the EHR will allow us to use real clinical data to identify multimorbidity phenotypes based on the clinical similarity of patients. Purpose of this study is to identify distinct subgroups of patients based on the MCC and cancer diagnoses and describe differences across these subgroups. Methods: EHR data was extracted from adult patients (n=2977) newly diagnosed with cancer in 2017 at one academic medical center. The SEER cancer site/histology list was used to group cancer diagnosis. MCC present for >6 months on the problem list or ICD-10 billing data were used. K-Means and K-Modes clustering procedures, with K equaling 7, were used to cluster patients based on MCC. Results: The sample consisted of 58% women, 93% white, with mean age of 62.4 (16.1) years. The most frequent cancers were GI (17%), gynecological (14%), and pulmonary (10%). The most frequent MCC were hypertension (33%), anemia (24%), and metabolic diseases (21%). Seven clusters correspond to following primary cancer sites: GI, pulmonary, urinary, gynecological, breast, endocrine, and skin. The MCC rates varied significantly across different primary sites with hypertension being present in call clusters, but anemia was present only in GI and urinary system cancers clusters. Conclusions: K-Means and K-Modes clustering procedures, with K equaling 7, produced similar clusters of cancer primary sites and MCCs, indicating our findings are stable and replicable. Our data extraction methods and clustering techniques worked well and can be expanded upon. Our next step is to repeat the data extraction and clustering analysis with the full data from the data warehouse (>30,000 records). The identified multimorbidity phenotypes will be used as inclusion criteria for prospective research with patients to explore the relationships among MCCs and symptoms in the context of cancer.
Details
- Title: Subtitle
- Multimorbidity, cancer, and symptoms: Using electronic health record data to cluster patients in multimorbidity phenotypes
- Creators
- Stephanie Gilbertson-White - University of IowaSanvesh Srivastava - University of IowaYunyi Li - University of IowaElyse Laures - University of IowaSeyedehtanaz Saeidzadeh - University of IowaChi Yeung - University of IowaSena Chae - University of Iowa
- Resource Type
- Abstract
- Publication Details
- Journal of clinical oncology, Vol.37(31_suppl), pp.130-130
- DOI
- 10.1200/JCO.2019.37.31_suppl.130
- ISSN
- 0732-183X
- eISSN
- 1527-7755
- Grant note
- DOI: 10.13039/100000002, name: U.S. National Institutes of Health
- Language
- English
- Date published
- 11/01/2019
- Academic Unit
- Internal Medicine; Nursing; Statistics and Actuarial Science
- Record Identifier
- 9984293097702771
Metrics
9 Record Views