3.1 Characteristics of patients
The mean age, Assens insomnia scale, CFS score, AST/ALT ratio, prothrombin time, plasminogen, D-dimer, urea nitrogen and brain natriuretic peptide precursor in the frailty group were higher than those in the non-frailty group (P<0.05). Diastolic blood pressure,body weight,BMI, grip strength, upper arm circumference, calf circumference, glutamic aminotransferase, albumin, uric acid, high-density lipoprotein, hemoglobin, MNA nutritional score and MMSE score in the frailty group were lower than those in the non-frailty group (P<0.05). There were no significant differences in height, systolic blood pressure, heart rate, HbA1C, fasting blood glucose, AST, serum creatinine, total cholesterol, LDLC, triglyceride, retinol-binding protein(RBP), and platelet count between the two groups (P> 0.05). The baseline characteristics of the included patients are indicated in Table 1.
We randomly divided the patients into a training dataset (144 patients) and a validation dataset (62 patients) in the ratio of 7:3. The fundamental characteristics of the HFpEF patients in both sets were shown in Table 2. There were no statistically significant differences in any of the variables between two datasets (P> 0.05).
3.2 Echocardiographic indexes
The left atrial diameter (LAD), pulmonary artery pressure, E/e' and tricuspid regurgitant velocity in the frailty group were higher than those in the non-frailty group, the interval of early diastolic peak velocity in the ventricular septum and the lateral wall in the left ventricular lateral wall were lower than those in the non-frailty group, and the difference was statistically significant (P<0.05). There were no significant differences in aortic diameter(Aod), left ventricular end-diastolic diameter (LVDd), Left ventricular end-systolic diameter (LVDs), Interventricular septal thickness (IVS), left ventricular posterior wall thickness (LVPW), left ventricular short axis shortening rate (LVFS), left ventricular ejection fraction (LVEF) and E/A between the two groups (P> 0.05). The echocardiographic indexes of two groups are showed in Table 1.
3.3 Univariate analysis and Lasso regression
To determine the independent factors influencing the occurrence of frailty in HFpEF patients, an univariate analysis was conducted with the occurrence of frailty as the dependent variable and 52 clinical variables as independent variables (shown in Table 2). Variables with statistically significant differences (P<0.05) in the univariate analysis were subjected to Lasso regression to reduce the dimensionality of highly correlated independent variables and select variable combinations with high correlation to the dependent variable. Lasso regression was performed using 10-fold cross-validation, resulting in the selection of 9 predictive variables: age, weight, grip strength, MNA score, albumin, fibrinogen, E/e', lateral e', and tricuspid regurgitation velocity. The factors influencing the occurrence of frailty in HFpEF patients selected by Lasso regression are shown in Figure 1. Figure 1-A displays the λ values generated through 10-fold cross-validation, with the dashed line on the right corresponding to the λ value when the binomial deviation is one standard error away. The figure shows the 9 variables with non-zero coefficients obtained after compression. Figure 1-B reflects the variation of variable parameters with the change in compression parameter, where the x-axis represents the logarithm of λ and the y-axis represents the variable coefficients. The curve on the graph represents the process of variable coefficients changing with the logarithm of the harmonic parameter λ, and variable selection can be achieved when the regression coefficients of the variables gradually compress to 0. (Table 3, Figure 1)
3. Results
3.1 Clinical Data Comparison
In the frail cohort, the mean age, Athens Insomnia Scale scores, Chronic Fatigue Syndrome (CFS) scores, Aspartate Aminotransferase/Alanine Aminotransferase (AST/ALT) ratio, Prothrombin Time (PT), fibrinogen, D-dimer, Blood Urea Nitrogen (BUN), fibrinogen, D-dimer, and Pro-B-type Natriuretic Peptide (ProBNP) were significantly higher compared to the non-frail cohort. Conversely, the frail cohort exhibited lower diastolic blood pressure, weight, Body Mass Index (BMI), grip strength, upper arm circumference, calf circumference, ALT, albumin, Uric Acid (UA), High-Density Lipoprotein Cholesterol (HDL-C), Hemoglobin (Hb), Mini-Mental State Examination (MMSE) score, and CFS score compared to the non-frail group. These differences were statistically significant (P<0.05). No statistically significant differences (P>0.05) were observed between the two cohorts in terms of height, systolic blood pressure, heart rate, Hemoglobin A1C (HbA1C), fasting blood sugar, AST, creatinine, total cholesterol, Low-Density Lipoprotein Cholesterol (LDL-C), triglycerides, retinol-binding protein, and platelet count. See Table 1 for details.
3.2 Echocardiographic Index Comparison
In the frail group, left atrial diameter (LAD), pulmonary artery pressure, E/e', and tricuspid regurgitation velocity were higher than in the non-frail group. The early diastolic peak velocity at the interventricular septum (e') and the lateral wall (e') were lower in the frail group. These differences were statistically significant (P<0.05). There were no statistically significant differences (P>0.05) between the two groups in terms of aortic diameter (Aod), left ventricular end-diastolic diameter (LVDd), left ventricular end-systolic diameter (LVDs), interventricular septum (IVS), left ventricular posterior wall thickness (LVPW), fractional shortening (FS), left ventricular ejection fraction (LVEF), and E/A ratio. See Table 2.
3.2 Univariate and Lasso Regression
To determine the independent factors associated with frailty in HFpEF patients, univariate analysis was performed using the 52 clinical variables (Table 3). The variables with statistically significant differences (P<0.05) in the univariate analysis were further analyzed using Lasso regression to select the variables with high correlation with frailty. A 10-fold cross-validation was performed, and 9 predictive variables were selected: age, weight, grip strength, MNA score, albumin, fibrinogen, E/e', lateral wall e', and tricuspid regurgitation velocity. The Lasso regression identified the factors influencing frailty in HFpEF patients (Figure 1). Figure 1-A shows the λ values generated through 10-fold cross-validation, and the dashed line on the right corresponds to the λ value when the binomial deviation is one standard error away. The 9 variables with non-zero coefficients were obtained. Figure 1-B shows the change in variable parameters with the change in the regularization parameter λ. The x-axis represents the logarithm of λ, and the y-axis represents the variable coefficients. The curve on the graph represents the change in variable coefficients with the logarithm of the harmonic parameter λ. When the regression coefficients of the variables gradually compress to 0, variable selection can be achieved. See Table 3 and Figure 1.
3.3 Multivariable Regression Results
The 9 predictive factors selected by Lasso regression were used as independent variables, and whether HFpEF patients had frailty (no=0, yes=1) was used as the dependent variable in a multivariable logistic regression analysis. The results showed that age (OR=1.058, 95% CI: 1.025-1.091), grip strength (OR=1.058, 95% CI: 1.025-1.091), MNA score (OR=1.058, 95% CI: 1.025-1.091), albumin (OR=1.058, 95% CI: 1.025-1.091), and tricuspid regurgitation velocity (OR=1.058, 95% CI: 1.025-1.091) were independent risk factors for frailty in HFpEF patients. See Table 3.
3.4 Nomogram Model Construction
Based on the results of the multivariable logistic regression analysis, a nomogram was constructed to predict the risk of frailty in elderly HFpEF patients (Figure 1). By finding the corresponding values on the vertical line for each prognostic factor, the score for each factor can be obtained. The scores of each variable are added together to obtain a total score, which corresponds to the predicted risk of frailty in HFpEF patients below the column line chart. See Figure 2.
3.5 Molde Internal Validation
In the training set, The area under the curve (AUC) of the ROC curves was 0.950 (95% CI: 0.911-0.869) while in the validation set, it was 0.932 (95% CI: 0.882-0.867) (Figure 3). To ensure the predictive model’s reliability, we performed calibration using calibration plots and Hosmer–Lemeshow tests. The calibration curves exhibited an excellent fit between the predictive model and the validation set (figure 5). The Hosmer-Lemeshow goodness-of-fit test showed that the nomogram model had a good fit (χ2=4.761, P=0.783) . The calibration curve was plotted by performing internal validation using the "Bootstrap" method 1000 times, and the curve was close to the ideal curve, indicating good calibration of the model and good consistency between the predicted risk of frailty in HFpEF patients and the actual risk of frailty. To provide a comprehensive evaluation, the ROC curve of the prediction model was compared with the ROC curve of the model using a single predictive factor (Figure 6), and the AUC of the single predictive factor was always smaller than that of the prediction model, highlighting the robust performance of the model.
3.6 Clinical Value of the Prediction Model
The clinical net benefit of the model in predicting the risk of frailty was evaluated using the clinical decision curve analysis (DCA) (Figure 7-1). The black horizontal line represents the net benefit of including all heart failure patients without frailty, which is 0. The light gray line represents the net benefit of including all HFpEF patients with frailty, and the red line represents the net benefit of the model. The model showed good clinical net benefit, which is favorable for clinical decision-making. The clinical impact curve (CIC) was plotted (Figure 3-8), with the red curve representing the number of HFpEF patients predicted to have frailty at various threshold probabilities, and the blue curve representing the actual number of HFpEF patients with frailty at various threshold probabilities. The red curve covered the entire blue curve, indicating a high clinical effectiveness of the model.