Utilizing an Artificial Neural Network to Predict Self‐Management in Patients With Chronic Obstructive Pulmonary Disease: An Exploratory Analysis. Issue 1 (21st December 2020)

• Record Type: Journal Article
• Authors: Bugajski, Andrew; Lengerich, Alexander; Koerner, Rebecca; Szalacha, Laura
• Is Part Of: Journal of nursing scholarship. Volume 53:Issue 1(2021)
• Page Start: 16
• Page End: 24
• Publication Date: 2020-12-21
• Subjects: Artificial neural network -- chronic obstructive pulmonary disease -- self‐management -- symptoms; Nursing -- Periodicals; Nursing -- United States -- Periodicals; 610.73
• DOI: 10.1111/jnu.12618
• Languages: English
• ISSNs: 1527-6546

Abstract

Abstract: Purpose: The main objective of this study was to utilize an artificial neural network in an exploratory fashion to predict self‐management behaviors based on reported symptoms in a sample of stable patients with chronic obstructive pulmonary disease (COPD). Design and Methods: Patient symptom data were collected over 21 consecutive days. Symptoms included distress due to cough, chest tightness, distress due to mucus, dyspnea with activity, dyspnea at rest, and fatigue. Self‐management abilities were measured and recorded periodically throughout the study period and were the dependent variable for these analyses. Self‐management ability scores were broken into three equal tertiles to signify low, medium, and high self‐management abilities. Data were entered into a simple artificial neural network using a three‐layer model. Accuracy of the neural network model was calculated in a series of three models that respectively used 7, 14, and 21 days of symptom data as input (independent variables). Symptom data were used to determine if the model could accurately classify participants into their respective self‐management ability tertiles (low, medium, or high scores). Through analysis of synaptic weights, or the strength or amplitude of a connection between variables and parts of the neural network, the most important variables in classifying self‐management abilities could be illuminated and served as another outcome in this study. Findings: The artificial neural network was able to predict self‐management ability with 93.8% accuracy if 21 days of symptom data were included. The neural network performed best when predicting the low and high self‐management abilities but struggled in predicting those with medium scores. By analyzing the synaptic weights, the most important variables determining self‐management abilities were gender, followed by chest tightness, age, cough, breathlessness during activity, fatigue, breathlessness at rest, and phlegm. Conclusions: The results of this study suggest that self‐management abilities could potentially be predicted through understanding and reporting of patient's symptoms and use of an artificial neural network. Future research is clearly needed to expand on these findings. Clinical Relevance: Symptom presentation in chronically ill patients directly impacts self‐management behaviors. Patients with COPD experience a number of symptoms that have the potential to impact their ability to manage their chronic disease, and artificial neural networks may help clinicians identify patients at risk for poor self‐management abilities.