We enrolled 75 maintenance HD patients in this single-center, prospective observational study conducted at Yoshikawa Internal Clinic between January and September 2025. For each patient, one HD session was analyzed with synchronized assessments at 5 standardized time points (predialysis; 1, 2, and 3 hours; postdialysis). Patients with complete data at all 5 time points were included. Exclusion criteria were: (1) history of atrial fibrillation, (2) moderate-to-severe valvular disease or regional wall motion abnormality, (3) left ventricular ejection fraction <50%, and (4) inadequate echocardiographic image quality.

HD was performed 3 times weekly for 4 hours per session, using high-flux membranes and standard dialysis equipment. Unless contraindicated, blood flow was set at 200 mL/min (the standard in Japan), with dialysate potassium 2.0 mmol/L and calcium 1.5 mmol/L. Dry weight (DW) was determined based on blood pressure changes, physical findings, chest radiography, and cardiothoracic ratio. For each session, DW, UF volume, treatment time, blood pressure, and heart rate were recorded. This standardized HD setting provides a controlled environment for studying rapid changes in preload, afterload, and cardiac filling.

Echocardiography was performed at the same 5 time points (pre, 1 hour, 2 hours, 3 hours, post). Echocardiograms were acquired using a diagnostic ultrasound system (Aplio a Verifia, model CUS-AA000; Canon Medical Systems Corporation, Otawara, Japan) by experienced sonographers. Analyses followed the recommendations of the American Society of Echocardiography (ASE) and the European Association of Cardiovascular Imaging (EACVI). ^,

At each time point, blood was drawn from the arteriovenous shunt. Concentrations of hANP and BNP were measured using a chemiluminescent immunoassay. All samples were analyzed with reagents from the same lot, under a standardized quality control program at the clinic.

We analyzed the data in long format (ID, time, marker, and value). To place all sessions on a common time scale, intradialytic time was normalized within each patient so that predialysis was 0 and postdialysis was 1. In our standard 4-hour sessions, 1 unit of normalized time corresponds to approximately 4 hours (for example, 0.25 corresponds to about 1 hour). All markers (echocardiographic indices and biomarkers) were standardized to z-score within each index; this linear rescaling does not change the timing of key events.

For each patient and marker, the predialysis value was taken as baseline. Using spline-based smoothed curves, we then defined 3 temporal indices: t20 (time to 20% of the total change from baseline), t50 (time to 50% of the total change), and t_peak (time of maximal deviation from baseline). Notably, t20 and t50 are defined as proportions of each marker’s within-session total change (amplitude) rather than absolute clinical decision thresholds. Lag was calculated as t_peak − t20. These indices were restricted to the prepeak portion of the curve so that “early” thresholds always refer to truly early changes rather than boundary artifacts.

Because IVC–CI can show erratic early fluctuations, we applied additional prespecified safeguards to this marker (brief smoothing around baseline and a fallback rule when no clear peak was present). These safeguards did not alter the basic definitions of t20, t50, or t_peak. Analyses were performed on complete cases, and no additional outlier removal was undertaken beyond these rules.

Peak (t_peak): For each smoothed curve, we identified the time point at which the absolute deviation from baseline was greatest, excluding trivial peaks at the very beginning or end of the session.

Amplitude (A): The total amplitude was defined as the maximum absolute deviation from baseline up to t_peak. t20 and t50: Within the interval from baseline to t_peak, we determined the first times at which the curve reached 20% (t20) and 50% (t50) of A.

Lag: Lag was defined as the interval between t20 and t_peak (Lag = t_peak − t20).

These definitions were chosen to capture how early each index responds and how long it takes to reach its peak.

For each index, we modeled the trajectory over normalized time using linear mixed-effects models with natural cubic splines (typically 4 degrees of freedom) for time. All models included patient-level random intercepts, and random slopes were added when they improved model fit. Models were estimated by restricted maximum likelihood, and predicted values were obtained on a fine grid over the 0–1 time axis.

For each patient and marker, temporal indices (t20, t50, t_peak, and Lag) were derived from patient-specific model-predicted trajectories and summarized as median [IQR] across the cohort ( Table 2 ). Sensitivity analyses, including UF-rate adjustment, are provided in the Supplementary material .

To examine robustness, we repeated key analyses using alternative definitions of early response. Specifically, we varied the early-change threshold to 15% and 25% of the total amplitude and evaluated simple alternative algorithms (e.g., focusing on the first half of the session or slightly perturbing spline-knot locations). Agreement with the primary estimator was quantified using Spearman’s rank correlation and the mean absolute error of marker-level medians. As an additional sensitivity analysis, we adjusted temporal indices for UF rate (mL/kg/h) using within-marker linear regression centered on the cohort mean UF rate ( Supplementary Table S4 and Figure S2 ).

To examine the clinical relevance of temporal indices, we analyzed intradialytic systolic blood pressure (SBP), which was recorded hourly at standardized time points during each HD session. SBP nadir was defined as the lowest SBP observed during the session, and maximal SBP decrease (ΔSBPmax) was calculated as baseline SBP minus SBP nadir. Normalized time difference was defined as SBP nadir time minus BNP t20 on the same 0–1 normalized time scale. Sessions were excluded from the time-difference analysis if BNP t20 was not identifiable (i.e., if BNP did not reach the predefined 20% threshold on the smoothed curve). Linear regression analysis was performed to assess the association between normalized time difference and ΔSBPmax.

All markers were processed using a uniform, prespecified pipeline. Simple safeguards were invoked only under predefined conditions, such as minor within-curve interpolation on the model-predicted fine grid over normalized time for temporal-index derivation (not imputation of missing observed time points) or fallback rules when curves were essentially monotonic and lacked a clear internal peak. When t20 or t50 could not be identified reliably, they were set equal to t_peak (yielding Lag = 0), and when multiple peaks had identical magnitude, the earliest was chosen.

To assess measurement reliability, intra- and interobserver reproducibility were evaluated in a randomly selected subset of participants (n = 20). Measurements were repeated offline by the same observer (intraobserver analysis) and by an independent second observer blinded to the initial results (interobserver analysis).

Intraclass correlation coefficients (ICCs) were calculated using a two-way random-effects model with absolute agreement. Reproducibility was assessed for E/A, E/e′, and LAVI.

For each index, t20, t50, and Lag were summarized as medians with interquartile ranges (IQR). To compare timing across indices, adjacent comparisons were performed using linear mixed-effects models on logit-transformed indices with Holm adjustment for multiple testing ( Supplementary Tables S1–S2 ).

All analyses were conducted in R version 4.5.0, using standard packages for mixed-effects modeling and data handling (including lme4, lmerTest, splines, emmeans, and the tidyverse suite). Linear mixed-effects models were estimated by restricted maximum likelihood with patient-level random intercepts and, when supported by model fit, random slopes. A two-sided p value <0.05 was considered statistically significant. Continuous variables are presented as mean ± SD or median [IQR], as appropriate. Reporting adhered to the STROBE guidelines for observational studies.

The study was approved by the institutional ethics committee of Yoshikawa Internal Clinic (approval date: December 2024) and conducted in accordance with the Declaration of Helsinki. Written informed consent was obtained from all participants.

Baseline characteristics are summarized in Table 1 (mean age 70.5 ± 12.7 years; 69% male; diabetes 37%; hypertension 76%). Mean serum albumin was 3.6 ± 0.3 g/dL. Baseline BNP (160.1 ± 151.0 pg/mL) and hANP (192.3 ± 132.2 pg/mL) showed wide interindividual variability. Echocardiographic indices also demonstrated broad distributions: E/A 0.87 ± 0.31, E/e′ 14.0 ± 4.7, LAVI 47.3 ± 11.0 mL/m², and LVMI 146.6 ± 27.3 g/m².

Table 1

Baseline characteristics

	All (N = 75)
Demographic characteristics
Age (years)	70.5 ± 12.7
Male sex, n (%)	52 (69%)
Body mass index (BMI, kg/m²)	21.5 ± 3.3
HD duration (years)	10.9 ± 10.6
Risk factors n, (%)
Diabetes	28 (37%)
Hypertension	57 (76%)
Past history n, (%)
Coronary artery disease	46 (61%)
Cerebrovascular disease	14 (19%)
Predialysis blood pressure (mm Hg)
Systolic (mm Hg)	146.6 ± 22.1
Diastolic (mm Hg)	75.5 ± 11.7
Laboratory values
Albumin (g/dL)	3.6 ± 0.3
Creatinine (mg/dL)	9.7 ± 2.3
Hematocrit (%)	33.6 ± 3.5
Biomarkers
BNP (pg/mL)	160.1 ± 151.0
hANP (pg/mL)	192.3 ± 132.2
Echocardiographic data
LAVI (mL/m²)	47.3 ± 11.0
LVMI (g/m²)	146.6 ± 27.3
E/A (ratio)	0.87 ± 0.31
E/e′ (ratio)	14.0 ± 4.7
LVEF (%)	61.0 ± 7.6
GLS (%)	−19.4 ± 2.6
TRPG (mm Hg)	25.4 ± 7.5
IVC–CI (%)	67.0 ± 8.0

Only gold members can continue reading. Log In or Register to continue

Share this:

• X
• Facebook

Related

Related posts:

1. Utilization and Outcomes of Dual Antiplatelet Therapy in Patients With Active Cancer Presenting With Acute Myocardial Infarction: A Global Registry Study
2. Chagas Cardiomyopathy in the United States: Testing and Diagnosis Across Real-World Datasets
3. Finerenone in Cardio-Oncology Patients With HFmrEF and HFpEF: Potential Clinical Benefits
4. Nanotechnology in Cardiology: A New Era of Targeted Therapies and Advanced Diagnostics—A Review

Stay updated, free articles. Join our Telegram channel

Join