Fig 1.
Simplified workflow of the study.
For each measured pressure waveform, model personalization (calibration) was performed. An iterative optimization procedure was employed to tune the values of parameters describing the function of the heart (i.e. Emax–maximal value of the elastance function, and tm–time to the onset of constant elastance) as well as terminal compliances and resistances (Sc and SR, respectively) that would minimize the error between the measured and simulated pressure waveform in the radial artery. After model personalization, the model-simulated blood flow waveform in the ascending aorta was used to estimate stroke volume (SV). Finally, model-estimated and reference SV values have been compared.
Fig 2.
Exemplary model simulations of the pressure waveform in the radial artery (upper panels) and the corresponding blood flow waveform in the ascending aorta (lower panels) in four subjects: Two HD patients and two subjects from the control group.
Fig 3.
The quality of model fits: Mean absolute percentage error (MAPE) between the measured and model-simulated pressure waveforms in the radial artery for (a) control group and (b) HD patients.
For HD patients, the results are divided based on either the length of the interdialytic break before the HD session (a long, 3-day break vs short, 2-day break) or the time of the measurement (before/after the start of the HD session and before/after the end of the session).
Fig 4.
Comparison between the model-estimated (computed) stroke volume (SV) and bioimpedance-based (measured) SV values for (a) control group and (b) HD patients (data shown separately for the HD sessions after a long and short interdialytic break).
Solid and dashed lines represent linear regression, and 95% confidence intervals, respectively.
Fig 5.
Bland-Altman plots comparing the stroke volume estimated by the model versus estimated using bioimpedance cardiography (PhysioFlow) for (a) control group and (b) hemodialysis patients.
Dashed horizontal lines represent the 95% limits of agreement, straight horizontal line represents the mean difference. Before plotting, the data have been logarithmically transformed. SD denotes standard deviation.
Fig 6.
Box-plots for the estimated cardiovascular parameters in the hemodialysis (HD) and control groups.
The description of parameters and their units are provided in Table 1.
Table 1.
Summary of the estimated patient-specific values of the model parameters for hemodialysis (HD) patients and control group.
The data are shown as means and standard deviations (SD).
Table 2.
Summary of the estimated values of the cardiovascular parameters in hemodialysis (HD) patients depending on the length of the interdialytic break before the studied HD session (a long, 3-day break vs a short, 2-day break).
The data are presented as means (± standard deviation).
Fig 7.
Correlation matrix between the parameters estimated by the model (Y axis) and cardiovascular parameters or indices derived by SphygmoCor (X axis) for (a) control group of healthy subjects and (b) hemodialysis patients, * p < 0.05, ** p < 0.01, *** p < 0.001.
CSV–computed stroke volume, BMI–body mass index, PSP–peripheral systolic pressure, CSP–central systolic pressure, PDP–peripheral diastolic pressure, CDP–central diastolic pressure, CAI–central augmentation index, CAP–central augmentation pressure, PAI–peripheral augmentation index, CESP–central end-systolic pressure, PESP–peripheral end-systolic pressure, PF SV–stroke volume estimated by PhysioFlow.
Table 3.
Characteristics of the study subjects.
Data are reported as means ± standard deviation. The data reported for hemodialysis (HD) patients were assessed after the mid-week HD hemodialysis session.
Fig 8.
Graphical summary of the timeline of measurements in HD patients.
Measurements of the pulse wave in the radial artery were performed in 35 HD patients at 4 time points during two HD sessions (after a long and a short interdialytic break). In the control group the measurements were performed at one time only.