Characterizing shared and distinct symptom clusters in common chronic conditions through natural language processing of nursing notes

Abstract

Data-driven characterization of symptom clusters in chronic conditions is essential for shared cluster detection and physiological mechanism discovery. This study aims to computationally describe symptom documentation from electronic nursing notes and compare symptom clusters among patients diagnosed with four chronic conditions-chronic obstructive pulmonary disease (COPD), heart failure, type 2 diabetes mellitus, and cancer. Nursing notes (N = 504,395; 133,977 patients) were obtained for the 2016 calendar year from a single medical center. We used Nim-bleMiner, a natural language processing application, to identify the presence of 56 symptoms. We calculated symptom documentation prevalence by note and patient for the corpus. Then, we visually compared documentation for a subset of patients (N = 22,657) diagnosed with COPD (n = 3339), heart failure (n = 6587), diabetes (n = 12,139), and cancer (n = 7269) and conducted multiple correspondence analysis and hierarchical clustering to discover underlying groups of patients who have similar symptom profiles (i.e., symptom clusters) for each condition. As expected, pain was the most frequently documented symptom. All conditions had a group of patients characterized by no symptoms. Shared clusters included cardiovascular symptoms for heart failure and diabetes; pain and other symptoms for COPD, diabetes, and cancer; and a newly-identified cognitive and neurological symptom cluster for heart failure, diabetes, and cancer. Cancer (gastrointestinal symptoms and fatigue) and COPD (mental health symptoms) each contained a unique cluster.In summary, we report both shared and distinct, as well as established and novel, symptom clusters across chronic conditions. Findings support the use of electronic health record-derived notes and NLP methods to study symptoms and symptom clusters to advance symptom science.

Determining the Set of Items to Include in Breast Operative Reports, Using Clustering Algorithms on Retrospective Data Extracted from Clinical DataWarehouse

“…Each individual sentence was represented with TFIDF, Universal Language Encoder (USE), or Paraphrase model as proposed by the transformers library (PRP). These encoded sentences were then used as input for three clustering algorithms which have proven useful for text analysis in healthcare domain: k-means [4], DBSCAN [5], and Hierarchical Clustering [6]. This resulted in comparing 9 "couples" of token representation /clustering algorithm (3 token methods × 3 clustering algorithms).…”

“…We hypothesize that a structured model of medical reports could be learnt using machine learning algorithms applied to retrospective medical reports from clinical data warehouses (CDW) [3]. Machine learning techniques, and especially clustering algorithms, have shown their value in extracting medical information from medical reports, alleviating time and resource consumption for this task [4][5][6].…”

Natural Language Processing of Nursing Notes

“…18,38 The most common standard terminologies used were the SNOMED-CT (Systematized Nomenclature of Medicine, approximately 40%, n = 17) [12][13][14]16,22,25,27,29,31,32,36,41,42,45,[51][52][53] and the UMLS (Unified Medical Language System, approximately 35%, n = 15). 9,[12][13][14]19,22,25,[28][29][30]36,41,42,49,51 Nursing standard terminologies, such as the International Classification for Nursing Practice and the Omaha System, were used in only eight studies. 18,25,28,31,33,[35][36][37] Other standard terminologies were also used in some studies including the International Classification of Diseases, 18,25,28,31,33,35,<...…”

“…51,57 Seven studies were excluded from NLP system performance evaluation requirement because they implemented text mining methods that did not require NLP system evaluation. 17,24,34,38,41,44,49 Among the remaining eligible studies (n = 36), full performance evaluation metrics, based on the criteria described in Materials and Methods, were not reported for 42% (n = 15) 14,20,21,26,27,36,43,45,49,[51][52][53]56,57,59 of the studies. Full NLP evaluation metrics were reported for 61% (n = 22) from the 36 eligible studies.…”

Advancing Nursing Research Through Interactive Data Visualization With R Shiny

“…Research conducted by nurse scientists is highly innovative and tremendously impactful in improving health, recovery, quality of life, and wellness for individuals and communities. The evolution of nursing – as with other biomedical fields – has included a progressive focus on increasingly complex omic-based approaches, data science, and data analytics (Agarwal et al, 2022; Al-Zaiti et al, 2020; Barcelona et al, 2021; Harris et al, 2021; Heinsberg et al, 2021; Koleck, Topaz et al, 2021). In parallel, the way that researchers present and consume data is changing.…”