Abstract
Background
Intradialytic blood pressure (BP) abnormalities are common in patients undergoing chronic hemodialysis (HD), however, their relationship with body composition remains unclear. We investigated the association between intradialytic BP patterns and body composition using an unsupervised clustering approach.
Methods
Adult outpatients receiving chronic HD who underwent segmental multifrequency bioelectrical impedance analysis were included. Systolic blood pressure (SBP) trajectories during HD were mapped to prespecified 30-minute time points from 0 to 240 minutes, and changes relative to predialysis SBP were clustered using k-means clustering. Body composition parameters were compared across BP clusters. Linear mixed-effects models with a random intercept for patient identifier were used to account for repeated HD sessions.
Results
Among 928 HD sessions from 179 patients, three SBP patterns were identified: marked decreasing (16.8%), mild decreasing (46.2%), and rising (37.0%) patterns. Body mass index did not differ across clusters. However, fat-free mass index and skeletal muscle mass index increased from the marked decreasing to the mild decreasing and rising clusters, whereas percent body fat decreased. The extracellular water-to-total body water ratio was highest in the rising cluster, and visceral fat area showed a decreasing trend from the marked decreasing cluster to the rising cluster. These associations remained significant after adjustment. Similar findings were observed in sensitivity analyses.
Conclusion
Unsupervised clustering identified reproducible intradialytic BP phenotypes that closely associated with volume status and body composition in patients undergoing chronic HD.
Keywords: Blood pressure, Body composition, Clustering, Hemodialysis
INTRODUCTION
Cardiovascular disease is the leading cause of morbidity and mortality in patients with end-stage kidney disease undergoing chronic hemodialysis (HD) [1]. Blood pressure (BP) is a major determinant of cardiovascular risk; therefore, it should be controlled within acceptable target ranges [2,3]. However, in patients on chronic HD, the association between BP levels and cardiovascular outcomes is less straightforward than in general population, particularly with the distinctive patterns of BP measurements influenced by the HD treatment [4]. In routine chronic HD care, office BP measurements, including peridialytic and intradialytic BP, are frequently recorded and are essential for ensuring safe and effective HD treatment. Despite their clinical utility, the mechanisms underlying HD-related BP changes remain incompletely understood, and the management of abnormal intradialytic BP patterns, such as intradialytic hypotension and hypertension, remains unsatisfactory. Moreover, heterogenous and arbitrary definitions of abnormal intradialytic BP patterns may further hinder a clear understanding of their nature.
Recent advances in body composition assessment have increased its accessibility in clinical practice and enhanced our understanding of its clinical implications. In many HD centers, body composition analysis is regularly performed to guide dry weight adjustment and identify abnormal conditions such as malnutrition and sarcopenia [5]. Emerging evidence suggests a strong connection between body composition and intradialytic BP patterns. During HD treatment, BP decreases as retained extracellular water (ECW) is removed [4]. Excessive ECW removal beyond an individual’s tolerance can precipitate a BP drop and contribute to the development of intradialytic hypotension [6], whereas ECW overload has been associated with intradialytic hypertension [7]. These findings support the role of volume status as a key determinant of abnormal intradialytic BP patterns. In addition, prior studies have reported the associations between intradialytic hypotension and low muscle and fat mass [8,9]. Taken together, body composition analysis can help characterize intradialytic BP patterns and provide insights into the pathophysiologic mechanisms underlying intradialytic hypotension and hypertension.
In this study, we investigated the association between intradialytic BP patterns and body composition in patients undergoingchronic HD. We classified intradialytic BP patterns into clusters using an unsupervised approach to minimize investigator-driven bias. We then explored how these clusters were related to volume status and to muscle and fat mass.
METHODS
Patients
This retrospective study included adult patients with end-stage kidney disease who underwent outpatient chronic HD and body composition analysis between October 2016 and October 2023. Among 185 eligible patients, six were excluded because of missing baseline laboratory data or insufficient BP measurements (BP recordings available for < 180 minutes). Ultimately, 179 patients receiving chronic HD were included, contributing 928 paired BP and body composition measurements.
This study was conducted in accordance with the Declaration of Helsinki and approved by the Institutional Review Board of Chung-Ang University Hospital (number: 2310-022-19496). The requirement for written informed consent was waived because of the retrospective nature of the study and the use of anonymized data.
Data collection
Demographic and clinical data were obtained from electronic medical records, including age, sex, dialysis vintage, cause of end-stage kidney disease, and comorbidities. Comorbidity burden was assessed using the modified Charlson Comorbidity Index [10]. Additionally, dialysis-related parameters including interdialytic weight gain, dry weight, ultrafiltration volume, and calculated ultrafiltration rate were retrieved. The number of prescribed antihypertensive medications was also collected. Fasting blood samples were obtained before the midweek dialysis session to measure hemoglobin, albumin, blood urea nitrogen, creatinine, calcium, phosphorus, uric acid, sodium, potassium, total carbon dioxide, and total cholesterol. Postdialysis blood urea nitrogen was measured to assess dialysis adequacy.
Body composition analysis
Body composition was assessed every 6 months in outpatients undergoing chronic HD using a segmental multifrequency bioelectrical impedance analysis (BIA) device (InBody S10; BioSpace, Seoul, Korea). Measurements were performed after midweek dialysis sessions in the supine position. Eight tactile electrodes were placed on the tips of the thumbs and middle fingers of both hands and between the malleoli and calcanei of both feet. The device was regularly calibrated and operated by trained nursing staff according to the manufacturer's instructions.
Parameters obtained from BIA included fat-free mass, skeletal muscle mass, percent body fat (PBF), visceral fat area (VFA), and the ratio of ECW to total body water (ECW/TBW). Body mass index (BMI), fat-free mass index (FFMI), and skeletal muscle mass index (SMI) were calculated as the corresponding mass divided by height squared (kg/m2).
BP measurement
HD was performed using either Fresenius 5008S (Fresenius Medical Care, Homburg, Germany) or Artis (Gambro Dasco SpA, Medolla, Italy) machines. Patients were treated with bicarbonate-based dialysate at standard temperature two or three times weekly for 4 hours per session, with a blood flow rate of 200–350 mL/min.
BP measurements, including predialysis, postdialysis, and intradialytic BP, were obtained in the supine position using automated devices built into the HD machines, which underwent regular calibration. BP data were collected on the same day as the BIA measurements. All systolic and diastolic BP measurements were recorded from predialysis (0 minute) to at least 180 minutes. Because all sessions were scheduled for 4 hours, incomplete BP series were primarily attributable to missed recordings or early session termination.
Clustering of intradialytic BP patterns
Systolic blood pressure (SBP) values were mapped to prespecified 30-minute time points (0, 30, 60, 90, 120, 150, 180, 210, and 240 minutes). For each time point, measurements obtained within a ±15-minute window were assigned, and the mean value was used when multiple measurements were available within the window. The change in SBP at each time point was calculated relative to the predialysis SBP (ΔSBP_t = SBP_t − SBP_0). For sporadically missing SBP values between recorded measurements, missing values were imputed using within-session linear interpolation. Because the last SBP value could not be imputed, sessions yielded three types of complete trajectories through 180, 210, or 240 minutes, depending on the availability of the final measurements.
Unsupervised clustering was performed using k-means in a complete-case subset of sessions that included ΔSBP values at all time points from 0 to 240 minutes. Nonstandardized raw ΔSBP values were used for clustering to preserve the clinical meaning of absolute BP changes and to allow cluster distinctions reflecting both trajectory magnitude and shape. Candidate numbers of clusters (k = 2–8) were evaluated using the within-cluster sum of squares (elbow method), and cluster separability was assessed using the mean silhouette width based on Euclidean distance. K-means was fitted with 50 random starts (nstart = 50) and a maximum of 1,000 iterations per run (iter.max = 1,000).
Sessions with complete data up to 180 or 210 minutes but missing later measurements were assigned to the nearest centroid derived from the 0–240-minutes k-means model. Euclidean distances between the session vector and each centroid were computed at the prespecified time points, and the cluster with the minimum distance was assigned.
Statistical analysis
Continuous variables were expressed as means ± standard deviations, and categorical variables as numbers (%). Between-group differences were compared using one-way analysis of variance for continuous variables and the χ2 test for categorical variables. Because multiple HD sessions could belong to the same patient, linear mixed-effects models with a random intercept for each patient were fitted to account for repeated sessions within individuals. In the adjusted models, cluster membership, age, sex, predialysis SBP, and ultrafiltration rate were included as fixed effects. Adjusted marginal means were estimated, and pairwise comparisons among clusters were performed using Bonferroni correction.
We performed sensitivity analyses to evaluate the robustness and reproducibility of the identified BP clusters. First, a repeated random one-session-per-patient resampling was conducted to address potential bias from multiple sessions per patient. For each of 200 iterations, one HD session was randomly selected from each patient from the complete-case training dataset. Second, we repeated the clustering using an alternative number of clusters to assess whether the associations between BP patterns and body composition were preserved regardless of the specific choice of k. All statistical analyses were performed using R software (version 4.4.2; R Foundation for Statistical Computing, Vienna, Austria). A two-sided p-value < 0.05 was considered statistically significant.
RESULTS
Study population and dataset preparation
A total of 928 HD sessions from 179 patients were included in the final analysis dataset. The median number of sessions per patient was 3 (1, 8). The mean age was 68.2 ± 13.0 years, and 55.9% of the patients were male. The prevalence of diabetes and hypertension was 58.1% and 80.4%, respectively.
Of the 928 HD sessions, 826 (89.0%) had complete intradialytic BP measurements through 240 minutes and were used for k-means model training. The remaining sessions were assigned to clusters using centroid-based classification if complete measurements were available through 210 minutes (n = 75 [8.1%]) or 180 minutes (n = 27 [2.9%]).
Intradialytic BP clustering
Clustering was performed using the 826 sessions with complete dataset. The elbow profile supported k = 3, with a pronounced decrease in the within-cluster sum of squares from k = 2 to k = 3, followed by progressively smaller reductions thereafter (Supplementary Fig. 1). The mean silhouette width also favored k = 3 over k = 4 (0.293 versus 0.237). Based on these criteria, k = 3 was selected as the primary clustering scheme, whereas k = 4 was retained for sensitivity analysis of phenotype characterization.
After integration of the trained clustering model with centroid-based assignment, all 928 HD sessions were classified into three clusters according to their characteristic intradialytic SBP patterns: marked decreasing (n = 156, [16.8%]), mild decreasing (n = 429 [46.2%]), and rising (n = 343 [37.0%]) (Fig. 1). Characteristics across the intradialytic BP clusters are summarized in Table 1. The marked decreasing cluster had significantly higher ultrafiltration rate than the mild decreasing cluster, whereas patients in the rising cluster were prescribed a significantly higher number of antihypertensive drugs.
Fig. 1. Intradialytic SBP trajectories according to clusters. Mean changes in SBP relative to predialysis value are shown for the three clusters identified by k-means clustering (k = 3). Error bars indicate standard deviations. The marked decreasing, mild decreasing, and rising clusters included 156, 429, and 343 hemodialysis sessions, respectively.
SBP, systolic blood pressure.
Table 1. Cluster-specific session-level clinical, dialysis-related, and body composition characteristics according to intradialytic blood pressure pattern.
| Variable | Marked decreasing (n = 156) | Mild decreasing (n = 429) | Rising (n = 343) | Overall p-value | |
|---|---|---|---|---|---|
| Demographic & comorbidities | |||||
| Age (yr) | 68.1 ± 11.3 | 69.7 ± 11.5 | 69.4 ± 12.3 | 0.286 | |
| Males | 77 (49.7) | 231 (54.6) | 187 (55.5) | 0.466 | |
| Hemodialysis duration (mo) | 70.2 ± 62.8* | 52.3 ± 57.8 | 40.1 ± 56.2* | < 0.001 | |
| Charlson Comorbidity Index | 2.7 ± 1.9 | 2.6 ± 2.0 | 2.8 ± 2.1 | 0.324 | |
| Diabetes | 103 (66.5) | 236 (55.8) | 170 (50.4) | 0.004 | |
| Hypertension | 119 (76.8) | 346 (81.8) | 264 (78.3) | 0.308 | |
| Laboratory results | |||||
| Hemoglobin (g/dL) | 10.8 ± 1.9 | 10.6 ± 1.5 | 10.1 ± 1.4* | 0.002 | |
| Albumin (g/dL) | 3.7 ± 0.4 | 3.8 ± 0.4 | 3.7 ± 0.5 | 0.118 | |
| Sodium (mmol/L) | 135.7 ± 3.4 | 136.2 ± 3.0 | 136.6 ± 3.2 | 0.009 | |
| Potassium (mmol/L) | 4.7 ± 0.6 | 4.7 ± 0.7 | 4.6 ± 0.7 | 0.397 | |
| Kt/Vurea | 1.7 ± 0.2 | 1.7 ± 0.3 | 1.7 ± 0.3 | < 0.001 | |
| Hemodynamic & dialysis parameters | |||||
| Predialysis systolic BP (mmHg) | 169.3 ± 21.8* | 144.8 ± 22.3 | 133.1 ± 20.8* | < 0.001 | |
| Number of antihypertensive drugs | 1.2 ± 1.0 | 1.2 ± 1.1 | 1.5 ± 1.2* | < 0.001 | |
| Interdialytic weight gain (kg) | 2.3 ± 1.0 | 2.1 ± 1.2 | 1.8 ± 1.5* | < 0.001 | |
| Dry weight (kg) | 57.5 ± 10.1 | 56.9 ± 11.3 | 56.8 ± 11.5 | 0.653 | |
| Ultrafiltration volume (kg) | 2.0 ± 1.0 | 1.8 ± 0.8 | 1.6 ± 0.9* | < 0.001 | |
| Ultrafiltration rate (mL/kg/hr) | 9.0 ± 4.3* | 8.3 ± 3.6 | 7.1 ± 4.1* | < 0.001 | |
| Body composition parameters | |||||
| BMI (kg/m2) | 22.7 ± 3.7 | 22.1 ± 3.6 | 22.1 ± 3.6 | 0.169 | |
| FFMI (kg/m2) | 15.8 ± 2.2 | 15.9 ± 2.3 | 16.4 ± 2.3* | 0.001 | |
| SMI (kg/m2) | 8.4 ± 1.4 | 8.5 ± 1.5 | 8.7 ± 1.4* | 0.009 | |
| PBF (%) | 29.0 ± 13.0 | 27.2 ± 10.8 | 24.5 ± 10.6* | < 0.001 | |
| VFA (cm2) | 88.2 ± 57.2* | 77.1 ± 44.2 | 69.0 ± 40.8* | < 0.001 | |
| ECW/TBW (%) | 39.8 ± 1.9 | 39.6 ± 1.5 | 39.9 ± 1.4* | 0.048 | |
Data are expressed as mean ± standard deviation or number (%). The p-values indicate the overall significance of differences across clusters.
BMI, body mass index; BP, blood pressure; ECW/TBW, extracellular water-to-total body water ratio; FFMI, fat-free mass index; PBF, percent body fat; SMI, skeletal muscle mass index; VFA, visceral fat area.
*Indicates a significant difference compared with the mild decreasing group after Bonferroni correction (p < 0.05). Because clustering was performed at the hemodialysis-session level, the table summarizes session-level profiles.
Association between intradialytic BP clusters and body composition analysis
We compared body composition parameters across intradialytic BP clusters (Table 1). Significant differences across clusters were observed for FFMI, SMI, PBF, VFA, and ECW/TBW, whereas BMI did not differ significantly.
After accounting for repeated measurements using linear mixed-effects models adjusted for age, sex, predialysis SBP, and ultrafiltration rate, the overall pattern remained evident (Fig. 2). Compared with the mild decreasing cluster, the rising cluster showed higher ECW/TBW, FFMI, and SMI, but lower PBF. In contrast, the marked decreasing cluster showed lower FFMI and SMI than the mild decreasing cluster. BMI did not differ across the clusters.
Fig. 2. Adjusted body composition parameters according to intradialytic blood pressure pattern clusters. Adjusted marginal means with 95% confidence intervals are shown for each cluster. (A) BMI did not differ across clusters. (B, C) FFMI and SMI were higher in the rising cluster and lower in the marked decreasing cluster. (D, E) PBF and VFA showed overall differences across clusters, although pairwise differences were attenuated after Bonferroni correction. (F) ECW/TBW was highest in the rising cluster.
BMI, body mass index; ECW/TBW, extracellular water-to-total body water ratio; FFMI, fat-free mass index; SMI, skeletal muscle mass index; PBF, percent body fat; VFA, visceral fat area.
*p < 0.05 versus the mild decreasing cluster after Bonferroni correction.
In addition, we further adjusted for ECW/TBW to determine whether the association between intradialytic BP patterns and muscle and fat mass was independent of volume status. The overall pattern of associations was largely preserved (Table 2).
Table 2. Adjusted body composition parameters after additional adjustment for ECW/TBW.
| Variable | Marked decreasing | Mild decreasing | Rising | p-value |
|---|---|---|---|---|
| BMI (kg/m2) | 22.4 (21.8–23.0) | 22.4 (21.8–23.0) | 22.4 (21.9–23.0) | 0.893 |
| FFMI (kg/m2) | 15.9 (15.5–16.2)* | 16.1 (15.8–16.4) | 16.5 (16.2–16.8)* | < 0.001 |
| SMI (kg/m2) | 8.4 (8.2–8.6)* | 8.6 (8.4–8.8) | 8.8 (8.6–8.9)* | < 0.001 |
| PBF (%) | 28.1 (26.5–29.7) | 27.1 (25.7–28.5) | 25.4 (23.9–26.8)* | < 0.001 |
| VFA (cm2) | 82.0 (74.8–89.2) | 77.1 (70.8–83.4) | 72.8 (66.4–79.2) | 0.004 |
Data are expressed as adjusted means (95% confidence intervals). The p-values indicate the overall significance of cluster differences across clusters in linear mixed-effects models adjusted for age, sex, predialysis systolic blood pressure, ultrafiltration rate, and ECW/TBW, with a random intercept for each patient.
BMI, body mass index; ECW/TBW, extracellular water-to-total body water ratio; FFMI, fat-free mass index; PBF, percent body fat; SMI, skeletal muscle mass index; VFA, visceral fat area.
*Indicates a significant difference compared with the mild decreasing group after Bonferroni correction (p < 0.05).
Results of sensitivity analysis
We evaluated whether the findings were maintained in sensitivity analyses. The repeated random one-session-per-patient resampling analysis with k = 3 preserved the general trajectory structure of the three clusters, but the degree of stability differed across clusters (Supplementary Fig. 2). The marked decreasing and rising centroids showed high similarity to the original centroids, with mean correlations of 0.965 and 0.981, respectively. In contrast, the mild decreasing cluster showed lower correlation-based stability (mean correlation, 0.054), although its root mean square error was 6.08 mmHg, consistent with relatively small absolute deviation for a flat trajectory. Overall, the repeated resampling analysis supported preservation of the three-trajectory structure.
In a 4-cluster solution, the overall phenotypic structure observed in the 3-cluster analysis was preserved. Among the 928 HD sessions, 61 (6.6%), 269 (29.0%), 374 (40.3%), and 224 (24.1%) were classified as severe decreasing, moderate decreasing, near-stable, and rising patterns, respectively (Supplementary Fig. 3). The 4-cluster solution represented a finer stratification of the same underlying trajectory structure.
Body composition differences across the 4-cluster solution were consistent with those observed in the primary 3-cluster analysis (Supplementary Table 1). In the adjusted mixed-effects models, FFMI and SMI remained highest in the rising cluster, whereas PBF and VFA were highest in the severe decreasing cluster. In addition, ECW/TBW was lowest in the severe decreasing cluster. BMI remained similar across the clusters. These findings support the robustness of the primary results.
DISCUSSION
In this study, we used an unsupervised clustering approach to objectively classify dynamic intradialytic BP trajectories and evaluated their independent associations with body composition parameters. We identified three distinct, reproducible intradialytic BP phenotypes: marked decreasing, mild decreasing, and rising clusters. These clusters were strongly and independently associated with the patients’ volume status, muscle mass, and fat mass, even after rigorous adjustment for dynamic fluid removal stress.
Intradialytic BP monitoring is fundamental for ensuring patient safety during HD. Importantly, extreme intradialytic BP deviations—such as intradialytic hypotension and hypertension—are well-established risk factors for subsequent all-cause and cardiovascular mortality [11,12,13,14]. Given this established prognostic significance, understanding the underlying pathophysiological drivers of these high-risk BP trajectories is of paramount clinical importance. However, clinical decision-making based on arbitrary diagnostic cutoffs and binary thresholds may overlook the continuous physiological heterogeneity among patients [15,16,17]. In contrast, machine learning techniques like clustering learn directly from data, which helps elucidate the true relationship between clinical phenotypes and underlying pathophysiology. Using an unsupervised clustering approach, we visualized that intradialytic SBP alterations exist on a continuous spectrum and demonstrated that these data-driven phenotypes are intricately linked to the patients’ body composition characteristics.
Through the k-means clustering algorithm, we identified three distinct intradialytic BP phenotypes based on their trajectories over the HD session. Notably, these trajectory features were consistently maintained both when the repeated random one-session-per-patient resampling analysis with 200 iterations and when a 4-cluster solution (k = 4) was selected in the sensitivity analyses. We initially anticipated that the algorithm might differentiate the phenotypes based on the specific timing of the BP changes (e.g., early versus late drops); however, the clusters were primarily distinguished by the absolute magnitude and direction of the BP alterations. The features were confirmed both in the repeated random resampling analysis and the 4-cluster solution. A previous study by Wang et al. [18] determined intradialytic BP patterns using a dynamic time warping algorithm and applied normalization to the BP sequences, proposing five patterns based on density peak clustering, distinguishable by the timing and shape of the BP change. In contrast, we considered both the specific time interval after HD initiation and the absolute amount of change in SBP to be clinically crucial. This methodological choice allowed our clustering to capture the true magnitude of hemodynamic shifts, effectively revealing the trajectories representing highly reproducible biological signatures with the exceptional numerical stability.
Our clustering approach revealed that the dynamic intradialytic BP alterations are intricately linked to patients’ fluid status and body composition. It is noteworthy that a continuous, stepwise gradient in volume, muscle, and fat parameters was observed across the clusters, which was robustly confirmed in our 4-cluster sensitivity analysis. ECW/TBW indicates volume status and has been used to guide dry weight adjustment [5]. Based on our results, ECW/TBW tended to increase as the trajectory shifted from a marked decreasing to a rising pattern, while the lowest interdialytic weight gain and ultrafiltration rate and the highest number of antihypertensive drugs were observed in the rising cluster. The association between hypervolemia and intradialytic hypertension has been well demonstrated in previous literature [7,14]. Taken together, these findings suggest a state of volume overload in which the prescribed dry weight may be higher than the true euvolemic state. On the other hand, the excessive or rapid removal of ECW can precipitate a profound BP decline during HD [6]. Our results support the importance of individualized dry weight assessment and volume management for stable HD treatment, with attention to both marked decreasing and rising intradialytic BP patterns.
Indicators of muscle mass, such as FFMI and SMI, exhibited a stepwise increase toward the rising cluster, while indicators of fat mass, such as PBF and VFA, a stepwise decrease These findings should be interpreted with extreme caution. Because BIA inherently estimates fat-free mass based on TBW, ECW excess can lead to a substantial overestimation of muscle mass [19]. However, there is research supporting the association between sarcopenia and intradialytic hypotension [8], and some evidence even shows the protective effect of intradialytic exercise against intradialytic hypotension [20]. Furthermore, visceral adiposity acts as an active endocrine organ secreting pro-inflammatory cytokines [21,22], which may exacerbate endothelial and autonomic dysfunction, rendering patients highly vulnerable to BP drops. Our finding aligns with a previous study by Tian et al. [9], which reported that sarcopenic obesity, the combination of high fat mass and low lean mass, conferred the highest risk of intradialytic hypotension. Our results support previous findings, highlighting the critical role of sarcopenic obesity and adipose tissue in hemodynamic alterations over the HD course.
This study has several limitations that should be acknowledged. First, this was a retrospective, single-center study, which limits the generalizability of our findings to broader HD populations. Furthermore, this study design precludes causal inference regarding the relationship between intradialytic BP patterns and body composition parameters. Second, this study assessed body composition using BIA, which can limit the interpretation since it is easily affected by volume status, although we used a multifrequency BIA and measured following the recommended protocol [19]. Third, BP was measured using automated devices built into the HD machines. This approach may introduce measurement discrepancies compared to standardized measurement.
In conclusion, our unsupervised clustering successfully classified the dynamic trajectories of intradialytic BP into three distinct, reproducible phenotypes: marked decreasing, mild decreasing, and rising patterns. These data-driven BP trajectories are intricately linked to the patients’ underlying body composition parameters. This insight provides actionable therapeutic targets, such as more rigorous volume assessment in the rising pattern and nutritional and physical interventions for sarcopenic obesity in the marked decreasing pattern. Ultimately, integrating routine body composition and volume assessments into clinical practice can improve the interpretation of intradialytic BP patterns, paving the way for more individualized management strategies for patients undergoing chronic HD.
Footnotes
Funding: None.
Conflicts of interest: All authors have no conflicts of interest to declare.
Data sharing statement: Data supporting the findings of this study are available from the corresponding author upon request.
- Conceptualization: JS.
- Data curation: HK, SK, JHH, JS.
- Formal analysis: JS.
- Investigation: JS.
- Supervision: SK, JHH, JS.
- Writing - original draft: HK, JS.
SUPPLEMENTARY MATERIALS
Adjusted values of body composition parameters according to the 4-cluster solution
Elbow plot for selecting the number of clusters for intradialytic BP trajectories. Total WSS is plotted against the number of clusters (k = 2–8) for k-means clustering based on raw intradialytic changes in systolic BP from sessions with complete 0–240 minutes measurements. A marked reduction in the WSS was observed from 1,816,039.9 at k = 2 to 1,383,281.5 at k = 3, followed by progressively smaller decreases thereafter, supporting k = 3 as the primary clustering solution.
The three intradialytic SBP clusters in repeated one-session-per-patient resampling. One hemodialysis session was randomly selected from each patient, and k-means clustering (k = 3) was repeated over 200 iterations. Gray lines indicate the cluster centroids from each resampled dataset, while black lines represent centroids from the original full training dataset. The overall three-trajectory structure was preserved across resampling iterations, although the mild decreasing pattern showed greater dispersion than the marked decreasing and rising patterns.
Intradialytic systolic blood pressure trajectories according to the 4-cluster solution. The 4-cluster solution identified four distinct phenotypes: severe decreasing, moderate decreasing, near-stable, and rising patterns, supporting finer stratification of the primary 3-cluster trajectory structure.
References
- 1.Go AS, Chertow GM, Fan D, McCulloch CE, Hsu CY. Chronic kidney disease and the risks of death, cardiovascular events, and hospitalization. N Engl J Med. 2004;351:1296–1305. doi: 10.1056/NEJMoa041031. [DOI] [PubMed] [Google Scholar]
- 2.Aftab RA, Sellappans R, Ming CK, Shaik I. Taking a step further in identifying ideal blood pressure range among hemodialysis patients: a systematic review and a meta-analysis. Front Pharmacol. 2020;11:729. doi: 10.3389/fphar.2020.00729. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Kim JE, Choi YJ, Hwang SY, et al. Target blood pressure in Korean hemodialysis patients for optimal survival. Kidney Res Clin Pract. 2025;44:310–323. doi: 10.23876/j.krcp.22.241. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Hirawa N. Blood pressure management in hemodialysis patients. Hypertens Res. 2023;46:1807–1809. doi: 10.1038/s41440-023-01279-x. [DOI] [PubMed] [Google Scholar]
- 5.Shin J, Kim H, Jeon HE, et al. Optimizing hemodialysis management using bioelectrical impedance analysis: emphasizing the significance of the phase angle. Ren Fail. 2025;47:2567037. doi: 10.1080/0886022X.2025.2567037. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Flythe JE, Kunaparaju S, Dinesh K, Cape K, Feldman HI, Brunelli SM. Factors associated with intradialytic systolic blood pressure variability. Am J Kidney Dis. 2012;59:409–418. doi: 10.1053/j.ajkd.2011.11.026. [DOI] [PubMed] [Google Scholar]
- 7.Nongnuch A, Campbell N, Stern E, El-Kateb S, Fuentes L, Davenport A. Increased postdialysis systolic blood pressure is associated with extracellular overhydration in hemodialysis outpatients. Kidney Int. 2015;87:452–457. doi: 10.1038/ki.2014.276. [DOI] [PubMed] [Google Scholar]
- 8.Son HE, Ryu JY, Lee K, et al. The importance of muscle mass in predicting intradialytic hypotension in patients undergoing maintenance hemodialysis. Kidney Res Clin Pract. 2022;41:611–622. doi: 10.23876/j.krcp.21.153. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Tian M, Zha Y, Qie S, Lin X, Yuan J. Association of body composition and intradialytic hypotension in hemodialysis patients. Blood Purif. 2020;49:334–340. doi: 10.1159/000504245. [DOI] [PubMed] [Google Scholar]
- 10.Hemmelgarn BR, Manns BJ, Quan H, Ghali WA. Adapting the Charlson Comorbidity Index for use in patients with ESRD. Am J Kidney Dis. 2003;42:125–132. doi: 10.1016/s0272-6386(03)00415-3. [DOI] [PubMed] [Google Scholar]
- 11.Inrig JK, Patel UD, Toto RD, Szczech LA. Association of blood pressure increases during hemodialysis with 2-year mortality in incident hemodialysis patients: a secondary analysis of the Dialysis Morbidity and Mortality Wave 2 Study. Am J Kidney Dis. 2009;54:881–890. doi: 10.1053/j.ajkd.2009.05.012. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Stefánsson BV, Brunelli SM, Cabrera C, et al. Intradialytic hypotension and risk of cardiovascular disease. Clin J Am Soc Nephrol. 2014;9:2124–2132. doi: 10.2215/CJN.02680314. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Flythe JE, Xue H, Lynch KE, Curhan GC, Brunelli SM. Association of mortality risk with various definitions of intradialytic hypotension. J Am Soc Nephrol. 2015;26:724–734. doi: 10.1681/ASN.2014020222. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Assimon MM, Flythe JE. Intradialytic blood pressure abnormalities: the highs, the lows and all that lies between. Am J Nephrol. 2015;42:337–350. doi: 10.1159/000441982. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Graber ML, Franklin N, Gordon R. Diagnostic error in internal medicine. Arch Intern Med. 2005;165:1493–1499. doi: 10.1001/archinte.165.13.1493. [DOI] [PubMed] [Google Scholar]
- 16.Bekker HL. Making choices without deliberating. Science. 2006;312:1472. doi: 10.1126/science.312.5779.1472a. [DOI] [PubMed] [Google Scholar]
- 17.Dijksterhuis A, Bos MW, Nordgren LF, van Baaren RB. On making the right choice: the deliberation-without-attention effect. Science. 2006;311:1005–1007. doi: 10.1126/science.1121629. [DOI] [PubMed] [Google Scholar]
- 18.Wang F, Zhou JY, Tian Y, et al. Intradialytic blood pressure pattern recognition based on density peak clustering. J Biomed Inform. 2018;83:33–39. doi: 10.1016/j.jbi.2018.05.013. [DOI] [PubMed] [Google Scholar]
- 19.Ikizler TA, Burrowes JD, Byham-Gray LD, et al. KDOQI clinical practice guideline for nutrition in CKD: 2020 update. Am J Kidney Dis. 2020;76:S1–SS107. doi: 10.1053/j.ajkd.2020.05.006. [DOI] [PubMed] [Google Scholar]
- 20.Rhee SY, Song JK, Hong SC, et al. Intradialytic exercise improves physical function and reduces intradialytic hypotension and depression in hemodialysis patients. Korean J Intern Med. 2019;34:588–598. doi: 10.3904/kjim.2017.020. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Hotamisligil GS. Inflammation and metabolic disorders. Nature. 2006;444:860–867. doi: 10.1038/nature05485. [DOI] [PubMed] [Google Scholar]
- 22.Kawai T, Autieri MV, Scalia R. Adipose tissue inflammation and metabolic dysfunction in obesity. Am J Physiol Cell Physiol. 2021;320:C375–C391. doi: 10.1152/ajpcell.00379.2020. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Adjusted values of body composition parameters according to the 4-cluster solution
Elbow plot for selecting the number of clusters for intradialytic BP trajectories. Total WSS is plotted against the number of clusters (k = 2–8) for k-means clustering based on raw intradialytic changes in systolic BP from sessions with complete 0–240 minutes measurements. A marked reduction in the WSS was observed from 1,816,039.9 at k = 2 to 1,383,281.5 at k = 3, followed by progressively smaller decreases thereafter, supporting k = 3 as the primary clustering solution.
The three intradialytic SBP clusters in repeated one-session-per-patient resampling. One hemodialysis session was randomly selected from each patient, and k-means clustering (k = 3) was repeated over 200 iterations. Gray lines indicate the cluster centroids from each resampled dataset, while black lines represent centroids from the original full training dataset. The overall three-trajectory structure was preserved across resampling iterations, although the mild decreasing pattern showed greater dispersion than the marked decreasing and rising patterns.
Intradialytic systolic blood pressure trajectories according to the 4-cluster solution. The 4-cluster solution identified four distinct phenotypes: severe decreasing, moderate decreasing, near-stable, and rising patterns, supporting finer stratification of the primary 3-cluster trajectory structure.