The Mutual Contribution of 3-NT, IL-18, Albumin, and Phosphate Foreshadows Death of Hemodialyzed Patients in a 2-Year Follow-Up

Abstract

Patients with chronic kidney disease (CKD), especially those who are hemodialyzed (HD), are at significantly high risk of contracting cardiovascular disease and having increased mortality. This study aimed to find potential death predictors, the measurement of which may reflect increased mortality in HD patients, and then combine the most promising ones in frames of a simple death risk assessment model. For this purpose, HD patients (n=71) with acute myocardial infarction in the last year (HD group) and healthy people (control group) as a comparative group (n=32) were included in the study. Various laboratory determinations and non-invasive cardiovascular tests were performed. Next, patients were followed for two years, and data on cardiovascular (CV) deaths were collected. On this basis, two HD groups were formed: patients who survived (HD-A, n=51) and patients who died (HD-D, n=20). To model HD mortality, 21 out of 90 potential variables collected or calculated from the raw data were selected. The best explanatory power (95.5%) was reached by a general linear model with four variables: interleukin 18, 3-nitrotyrosine, albumin, and phosphate. The interplay between immuno-inflammatory processes, nitrosative and oxidative stress, malnutrition, and calcium-phosphate disorders has been indicated to be essential in predicting CV-related mortality in studied HD patients. ClinicalTrials.gov Identifier: NCT05214872.

Keywords: IL-18; hemodialysis; inflammation; mortality; nitrosative stress; phosphate; predictive model.

Conflict of interest statement

The authors declare no conflict of interest.

Figures

Figure 1

Stages of variable selection for the creation of the final model. We have distinguished 10 categories of variables: blood morphology, lipids, metalloproteinases, iron, cardiovascular, metabolism, oxidative stress, calcium and phosphates, inflammation, and general metrics. Four categories ware completely rejected based on the Kruskal–Wallis test results. Spearman correlation and experimental model fitting helped to find the best combination of variables that differentiate the HD-A and HD-D groups (4 out of 21).

Figure 2

Boxplot depicting differences between HD-D (red), HD-A (blue), and Control (green) for the chosen variables: age (A), albumin (B), 3-Nitrotyrosine (C), alkaline phosphatase (D), Interleukin-18 (E), and phosphates (F). Additional swarmplot was overlaid onto the original plot in order to mark the obtained measurements for each of the patients (dots). Outlier values are represented by diamons.

Figure 3

Boxplot depicting differences between HD-D (red), HD-A (blue), and Control (green) for the chosen variables: intima media thickness (A), NT-pro-brain natriuretic peptide (B), central end systolic pressure (C), advanced oxidation protein products (D), peripheral end systolic pressure (E), and central pressure at T2 (F). Additional swarmplot was overlaid onto the original plot in order to mark the obtained measurements for each of the patients (dots). Outlier values are represented by diamons.

Figure 4

Boxplot depicting differences between HD-D (red), HD-A (blue), and Control (green) for the chosen variables: peripheral systolic pressure (A), peripheral P2 (B), central augmented pressure (C), peripheral mean pressure (D), central mean pressure of systole (E), and central mean pressure of diastole (F). Additional swarmplot was overlaid onto the original plot in order to mark the obtained measurements for each of the patients (dots). Outlier values are represented by diamons.

Figure 5

Boxplot depicting differences between HD-D (red), HD-A (blue), and Control (green) for the chosen variables: central pulse height (A), central systolic pressure (B), and peripheral pulse pressure (C). Additional swarmplot was overlaid onto the original plot in order to mark the obtained measurements for each of the patients (dots). Outlier values are represented by diamons.

Figure 6

Three-dimensional visualization of spacious relationships between the predictor variables. The HD-A and HD-D groups are marked with blue and red colors, respectively. For the sole purpose of creating a better perception of distances between positions, a color transparency gradient, which highlights depth of the images, has been added.

Figure 7

Visualization of the model results with respect to all the four chosen predictors. The HD-A and HD-D groups are marked with blue and red colors, respectively. The vertical dashed line represents a change in the predicted fate of patients (the negative value of the model prediction can be treated as predicting that a patient will survive in the next 2 years, but a positive value foreshadows the patient’s death).

Figure 8

Residual diagnostic plots for the models with and without the outlier patient. Outliers can be spotted as outfits in residual vs. fitted plots. The significance of outliers can be analyzed by residuals vs. leverage plots. Sole observations at the end of the blue line may have a significant impact on the model if they cross the dashed lines drawn for Cook’s distance. In the presented case, the outlier has not fallen outside the Cook’s boundaries, and as it is presented in the subplots on the right, its deletion does not change the model in any significant manner.