Study design and participants
This study used a retrospective observational design. Participants for the present study were adults with HF who visited the outpatient departments of two large tertiary medical centres operating HF outpatient clinics in two cities of metropolitan area (Seoul and Suwon city), South Korea, for regular medical follow-ups between July 2017 and August 2019. We selected 119 participants who had performed relevant serum blood tests, echocardiography, and stress tests and responded to the surveys about self-management behaviour and the QoL on the same day. We collected their data retrospectively by electronic medical record review.
Self-management behaviour was measured using the European Heart Failure Scale , a 12-item questionnaire related to self-care behaviour in HF. It includes the questions of consulting behaviours (i.e., “How often do you call your doctor/nurse in case of shortness of breath, ankle swelling, weight gain, or fatigue?”) and adherence with the regimen (i.e., “How often do you weigh yourself, try to drink less water, follow a low-sodium diet, regularly take medication, or exercise?”). Also, their QoL was assessed using a measuring tool provided by the World Health Organization (WHOQOL-BREF) , a 26-item questionnaire on the individual’s perceptions of their health and well-being. The participants’ stress levels were measured using the heart rate variability (HRV) measurement tool of uBioMacpa (Biosense Creative, Seoul, Korea), which displays stress level on a scale of 0 to 100.
All participants underwent a comprehensive transthoracic echocardiographic evaluation, a standard 2-dimensional and Doppler echocardiographic examination, according to the recommendations of the American Society of Echocardiography . Left ventricular systolic function was defined using the left ventricular ejection fraction (EF), calculated according to the modified Simpson’s method (i.e., subtracting left ventricular end-systolic dimension from left ventricular end-diastolic dimension). Left ventricular diastolic function was defined as the early mitral inflow velocity to early diastolic mitral septal annular velocity (E/E’), calculated using pulsed-wave Doppler and tissue Doppler echocardiography. The evaluation was conducted using GE Vivid 7 (GE Healthcare, Horten, Norway) or iE33 (Philips Medical Systems, Andover, MA, USA), performed by 6 sonographers and 2 echocardiologists in one medical centre. In the other medical centre, it was conducted using Vivid E95 (GE Healthcare, Horten, Norway) or EPIQ CVX (Philips Medical Systems, Andover, MA, USA), which was performed by 8 sonographers and 2 echocardiologists. In this study, we only collected EF for cardiac systolic function and E/E’ for cardiac diastolic function from the participants’ echocardiographic results.
Electronic medical record review was performed to collect the participants’ sociodemographic and clinical characteristics, anthropometric data, and serum blood test results, including hemoglobin A1C (HbA1C), high-density lipoprotein (HDL), low-density lipoprotein (LDL), total cholesterol, triglyceride, and high sensitivity C-reactive protein (hs-CRP).
Data were analysed using SPSS version 25.0 (IBM Corporation, Armonk, NY, USA).
Descriptive statistics (frequencies, percentages, means, and standard deviations) were used to explain the participants’ sociodemographic and clinical characteristics, levels of stress, self-management behaviour, and QoL. Independent samples t-tests and χ2 tests were conducted to identify the differences in the variables according to the levels of low and high QoL. The two QoL levels were created by using a median split for the QoL measure.
Multiple linear regression analysis was performed to examine the relationships among QoL, EF, E/E’, and self-management behaviour. The choice of these variables for the regression analysis was based on the significance in the univariate analysis to identify the major factors that predict the QoL.
Lastly, the predictive model for QoL of participants was developed using decision tree analysis. Decision tree analysis is a data-mining technique designed to partition the whole dataset into subgroups based on splitting criteria . We used the classification and regression tree (CART) method , which presents a hierarchical model structured as a tree for predicting the QoL of the participants. The tree model structure is made up of root nodes, splitting nodes (parent nodes), and terminal nodes (child nodes). Firstly, the dataset was partitioned into two subsets based on a predictor variable with the score of QoL. The process was repeated on each derived subset in an iterative (recursive partitioning) manner. This method looks for subgroups in the dataset in which the predictor variable is relatively homogeneous. At each node, the recursive partitioning identifies a predictor variable and a split by which may be subclassified .