Baseline characteristics
A total of 301 patients with clinical diagnosis of CKD stages 3-5 were included in this study, including 167 males and 134 females. During the follow-up period until August 2022, there were 169 outcome events (56.1%), of which 1 (0.06%) developed unstable angina, 24 (73.4%) heart failure, 11 (6.5%) myocardial infarction, 21 (12.4%) coronary heart disease, 6 (3.6%) cerebral infarction, 3 (1.8%) cerebral hemorrhage, and 3 (1.8%) cardiogenic shock. Groups were stratified based on whether CVD events occurred within 5 years in patients with CKD stages 3-5, with group 1 being patients with CKD stages 3-5 without CVD events and group 2 being patients with CKD stages 3-5 with CVD events. The baseline characteristics of the 301 CKD patient subgroups are presented in Table 1. Compared with patients who did not experience a CVD event, those who experienced a CVD event had lower DBP and Hb levels, higher LDL, Scr, CRP, and IVS levels, and less use of antiplatelet agents, RAAS blockers, and lipid-lowering medications (P < 0.05). There were no significant differences in all other indexes between the two groups (P > 0.05).
Lasso regression of CVD events in patients with CKD stages 3-5 patients
This study defined whether CVD events occurred within 5 years in patients with CKD stages 3-5 as dependent variable. Age, gender, BMI, the history of diabetes mellitus, the history of hypertension, smoking, alcohol consumption, SBP, DBP and usage of β receptor blockers, antiplatelet agents, RAAS blockers, lipid-lowering agents, the levels of Hb, PLT, ALB, BUN, Scr, UA, total TC, TG, LDL-C, serum calcium, phosphorus, eGFR, 24h urinary protein and PTH, CRP and cardiac IVS were defined the independent variables. To reduce the problem of collinearity of the independent variables, the variables were screened by lasso regression analysis in R 4.1.3. It indicated that gender, presence or absence of DM, history of smoking, SBP and DBP, and use or not β blockers, the use of antiplatelet drugs, the use of RAAS blockers, the use of lipid-lowering drugs, the levels of Hb, Scr, serum calcium, eGFR, PTH and C-reactive protein, the thickness of the cardiac interventricular septum are risk factors for incident CVD within 5 years in patients with CKD stages 3-5 (Figure 1A and 1B).
Multivariate logistic regression of CVD events in patients with CKD stages 3-5 patients
The 16 variables selected as statistically significant by the lasso regression analysis were further included in the multivariate logistic regression analysis and it indicated the history of DM, SBP and DBP, and use or not β blockers, the use of RAAS blockers, the use of lipid-lowering drugs, the levels of Scr, eGFR, PTH and C-reactive protein, the thickness of the cardiac interventricular septum were risk factors for incident CVD events within 5 years in patients with CKD stages 3-5 (P < 0.05); However, gender, smoking history, whether antiplatelet agents were used, absolute hemoglobin value, and serum calcium were not statistically significant in multivariate logistic regression (P > 0.05) (Table 2).
Development and validation of the clinical prediction models
This study used 11 indicators that were statistically significant based on multivariate logistic regression analysis as independent variables and " CVD events within 5 years in patients with CKD stages 3-5 " as the dependent variable to construct a clinical risk prediction model. Based on the filtered variables, and combined with clinically relevant indicators, a total of 4 clinical risk prediction models were established, which were defined as the original model, inflammatory model, imaging model, and complete model, respectively, and sequentially referred to as Model 1, Model 2, Model 3, and Model 4. The original model incorporated the history of DM, SBP, DBP, and use or not β blockers, the use of RAAS blockers, the use of lipid-lowering drugs, the levels of Scr, eGFR, PTH, which were determined by reviewing the available literature and combining clinical conditions (13, 15, 20-22). The inflammation model was based on the original model with the addition of CRP and the imaging model with the addition of IVS, whereas the full model contained all indicators.
Regarding discrimination, plotting of the ROC curve was employed and the area under the curve (AUC) was calculated (Figure 1C). Compared with the original and inflammatory models, the imaging model and the full model performed better, with the highest C-statistic of 0.840 and 0.845, respectively. The AUC value of the imaging model was 0.840, with the 95% confidence interval (CI) of (0.796, 0.883). The AUC value of the full model was 0.845, with the 95% CI of (0.802, 0.888) (Tables 3). Regarding calibration, we plotted calibration curves (Figure 1D), calculated AIC and BIC values for the four models, and performed H-L tests on the four models. AIC values and BIC values were lowest for the imaging model and the full model compared to the original and inflammatory models. The AIC values were 321.821 and 311.531, and the BIC values were 353.600 and 356.017, respectively. The P-values of the H-L test of the four models were all greater than 0.05, and the goodness of fit test of the four models all passed, indicating better calibration (Table 3). In addition, the present study performed a DCA curve analysis to evaluate the clinical practical utility of the four models (Figure 1E). The results suggested that the clinical net benefit was significantly improved in all four models.
Ultimately, we performed a comprehensive evaluation combining the discrimination, calibration, and clinical utility of the model, and the complete model was selected as the optimal one. According to the sample size of this study, tenfold cross validation was selected for internal validation of the complete model, mainly by evaluating two aspects of model accuracy and consistency. The results suggested that the complete model had an accuracy of 0.728, and the accuracy was good; The kappa value of the complete model in consistency was 0.448, and the consistency was fair, which indicated that the complete model was established successfully.
A nomogram of the full model
Based on the above results, the full model exhibited better discrimination and calibration. Therefore, this study ranked the complete model as the optimal one. Based on the independent variables included in the full model, the history of DM, SBP and DBP, and use or not β blockers, the use of RAAS blockers, the use of lipid-lowering drugs, the levels of Scr, eGFR, PTH and C-reactive protein, the thickness of the cardiac interventricular septum were transformed according to functional relationships, after which the calibrated segments were used to construct nomograms in the plane according to the corresponding proportions in order for this risk prediction model to be better used in clinical practice. Using nomograms, we can visually and simply calculate the risk of CVD events over 5 years in patients with CKD stages 3-5 (Figure 2).