The value of pleth variability index in predicting hypotension during maintenance hemodialysis: a prospective observational multicenter study (PVi-HD study)

Abstract

Background

Intradialytic hypotension (IDH) is a frequent and serious complication of hemodialysis, and existing prediction tools are inadequate. The Pleth Variability Index (PVi), a non-invasive dynamic parameter that reflects cardiorespiratory interactions and fluid responsiveness, has shown promise in acute care settings, but its value in hemodialysis remains unexplored. We prospectively assessed PVi and its dynamic changes as predictors of IDH in a multicenter maintenance hemodialysis cohort.

Methods

In this prospective, multicenter observational study, we enrolled 200 patients on chronic hemodialysis. PVi was measured at baseline (PVi baseline) and 60 min into the session (PVi 1 h). We calculated the absolute change (ΔPVi) and percentage change (ΔPVi%). The primary outcome was the predictive performance of these parameters for IDH, assessed using receiver operating characteristic (ROC) curve analysis. Multivariate logistic regression was used to identify independent predictors.

Results

IDH occurred in 61 patients (30.5%). While baseline PVi was similar between groups, PVi 1 h, ΔPVi, and ΔPVi% were all significantly higher in patients who developed IDH (all P < 0.001). Dynamic PVi parameters demonstrated excellent predictive value, with ΔPVi showing the highest area under the ROC curve (AUC) at 0.84 (95% CI: 0.78–0.89). Followed closely by ΔPVi% (AUC 0.83), for which the optimal cutoff of 23.53% yielded a sensitivity of 78.69% and a specificity of 74.82% for predicting IDH. These significantly outperformed traditional predictors, including baseline mean arterial pressure (AUC 0.71) and ultrafiltration volume (AUC 0.68) (all P < 0.001 for comparison). In multivariate logistic regression, both ΔPVi% (OR 1.02, 95% CI 1.01–1.03) and ΔPVi (OR 1.22, 95% CI 1.16–1.28) were identified as robust independent predictors for IDH (both P < 0.001). This association was held in a sensitivity analysis using a more stringent definition of IDH based on symptomatic episodes. Using an optimal cutoff, Cox regression analysis revealed that patients with a ΔPVi% > 23.53% had a 7.15-fold higher risk of developing IDH (HR 7.15, 95% CI 3.86–13.22).

Conclusions

Dynamic changes in PVi during the first hour of hemodialysis are strong, independent predictors of subsequent IDH. Monitoring PVi changes offers a promising and practical non-invasive tool for early risk stratification.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12882-026-04883-4.

Introduction

Intradialytic hypotension (IDH) remains one of the most frequent complications during hemodialysis (HD), affecting approximately 12-30.7% of all sessions [1–3] and up to 20.3–51.6% in elderly and diabetic populations [4, 5]. This adverse event is associated with significant morbidity and mortality by causing cerebral hypoperfusion and end-organ ischemia [2]. Studies show 23.5% of patients experience brain ischemia, with symptoms such as dizziness and vomiting, and every 10mmHg decline in mean arterial pressure (MAP) increases the ischemic risk by 3% (P < 0.001) [6]. Hypotension also leads to myocardial stunning, wall motion abnormalities, and reduced ejection fraction, raising cardiovascular death risk [7, 8]. A 5-year study found a U-shaped link between blood pressure and mortality. Risks increased 1.57-fold below 90 mmHg and 1.25-fold above 140 mmHg [9].

Despite the clinical importance of IDH, current prediction methods are inadequate. Heart rate variability analysis shows modest accuracy, with an area under the receiver operating characteristic (ROC) curve (AUC) of 0.726 [10], while global longitudinal strain achieves high sensitivity (83%) and specificity (87%) but is complex and unsuitable for routine use [11]. Bioimpedance spectroscopy and relative blood volume monitoring show limited utility, with poor correlations to blood pressure changes (R = 0.216–0.232) [12–15]. Continuous blood pressure monitoring is accurate but resource-intensive [16], and machine learning models, despite promising results, have poor clinical applicability [17–21]. These limitations highlight the need for more effective, practical predictive tools for IDH detection.

The pleth variability index (PVi) represents a potential solution as a dynamic parameter derived from the perfusion index, which reflects respiratory variations in the photoplethysmographic waveform [22]. Originally developed for predicting fluid responsiveness in mechanically ventilated patients [23], it has shown promising results across various clinical settings. Studies in critical care environments have demonstrated that PVi values predict fluid responsiveness with an AUC ranging from 0.717 to 0.93 [24, 25]. From a physiological perspective, PVi may serve as an effective surrogate marker for intravascular volume status and cardiovascular reserve, both of which are critical factors in IDH development [26]. Evidence supporting this potential application comes from a small-sample study examining a similar parameter (PI) for predicting hypotension in Intensive Care Unit patients with acute kidney injury undergoing renal replacement therapy. This research demonstrated good predictive properties, with AUCs of 0.721 [27]. Beyond its theoretical promise, PVi offers significant practical advantages, as its measurement requires only a pulse oximeter with specific software capabilities to obtain reliable readings. This non-invasive and efficient approach makes PVi particularly suitable for routine application in dialysis centers. However, despite these compelling theoretical foundations and practical benefits, PVi’s utility specifically in the chronic hemodialysis population remains largely unexplored.

Therefore, our multicenter prospective observational study (PVi-HD) aims to evaluate the predictive value of PVi for IDH in chronic hemodialysis patients. By identifying a reliable PVi threshold that precedes hypotensive episodes, we seek to establish a practical clinical tool enabling preemptive interventions. This approach potentially could reduce IDH frequency and associated complications, ultimately improving patient outcomes and quality of life.

Materials and methods

Ethical considerations

This study was approved by the Institutional Review Boards of all participating centers (Approval Nos. 2025‑61, 2025‑KY‑016, 20250124‑08, WHGJYY‑LLSP‑2025‑01, YW‑2025‑002). Written informed consent was obtained from all participants before enrollment. The study was conducted in accordance with the principles of the Declaration of Helsinki and Good Clinical Practice guidelines. This study was registered in ClinicalTrials.gov (NCT07022847, June 7 2025).

Inclusion and exclusion criteria

Inclusion criteria

• Adults aged 18 to 80 years.

• Patients who had been receiving maintenance hemodialysis for more than 3 months with stable dialysis regimens.

• Patients who provided written informed consent and were willing to comply with study procedures.

Exclusion criteria

• Patients with significant cardiac arrhythmias (such as persistent atrial fibrillation) that could interfere with reliable PVi measurements.

• Patients with severe peripheral circulation disorders, significant limb edema, or onychomycosis that could affect accurate PVi readings.

• Patients with severe hemodynamic instability during dialysis or inability to cooperate with study procedures.

• Pregnant or breastfeeding women.

• Patients with known allergies to dialysis filters.

• Dialysis sessions terminated for reasons unrelated to hypotension (e.g., technical issues, patient request).

• Patients who withdrew informed consent during the study.

Study protocol

This prospective, observational, multicenter study was conducted between February 2025 and June 2025 at five Blood Purification Centers in Wuhu, China. Patient recruitment was performed during routine clinic visits before scheduled hemodialysis sessions.

Upon arrival at the dialysis center, patients’ demographic information, clinical history, medication use, and relevant laboratory parameters were collected. Baseline blood pressure was recorded with patients in a supine position after at least 5 min of rest. Standard hemodialysis parameters, including dialysis duration, dialysate composition (sodium and calcium concentrations), and ultrafiltration volume, were documented. A Philips IntelliVue monitor equipped (Philips Medizin Systeme, Boeblingen, Germany) with a Masimo SET module was used, with a PVi sensor attached to the index or middle finger of the non-access arm according to manufacturer recommendations. The PVi reflects respiratory-related variations in the photoplethysmographic perfusion index, representing dynamic changes in stroke volume and venous return driven by intrathoracic pressure fluctuations [22, 23].

After a 5-minute stabilization period, baseline PVi (PVi baseline) was recorded immediately before initiating hemodialysis. The second PVi measurement (PVi 1 h) was obtained 60 min after the start of hemodialysis, with patients maintaining a stable position and regular breathing pattern during measurement. Signal artifacts due to patient movement, variations in peripheral perfusion, or temporary probe disconnection were identified through continuous visual inspection of the waveform and device alerts during measurement. When artifacts were detected, readings were repeated immediately under stabilized conditions; if correction was not feasible, the affected data points were excluded after verification. All operators underwent unified training sessions across participating centers, which included standardized probe placement, measurement timing, and patient positioning protocols. Adherence to these procedures was monitored periodically to reduce inter-operator variability and ensure consistency of PVi measurements.

Throughout the hemodialysis session, blood pressure was monitored at 30 ~ 60 min intervals and additionally whenever patients reported symptoms suggestive of hypotension. All measurements were performed by trained research staff using standardized procedures to ensure consistency across sites. The time of onset, duration, and management of hypotensive episodes were documented in detail. We calculated the absolute difference (ΔPVi) and percentage change (ΔPVi%) between baseline and PVi 1 h values using the following formulas:

1

2

IDH was classified into three categories based on established criteria [28–30]: (1) Overall hypotension: defined as a systolic blood pressure (SBP) reduction ≥ 20 mmHg from the pre‑dialysis baseline during hemodialysis [28]; (2) Asymptomatic hypotension: SBP reduction ≥ 20 mmHg from baseline without accompanying clinical symptoms.; and (3) Symptomatic hypotension: SBP reduction ≥ 20 mmHg or MAP reduction ≥ 10 mmHg from baseline, with accompanying symptoms such as dizziness, nausea, vomiting, or muscle cramping [29, 30].

Outcome measures

The primary outcome was the predictive value of PVi parameters (PVi baseline, PVi 1 h, ΔPVi, and ΔPVi%) for intradialytic hypotension, as determined by ROC curve analysis. Secondary outcomes included: (1) The incidence of intradialytic hypotension according to different criterion in the study population; (2) Comparison of PVi parameters with traditional volume assessment indices in predicting hypotension; (3) Correlation between PVi parameters and reductions in SBP, diastolic blood pressure (DBP), and MAP during hemodialysis, as well as interdialytic weight gain and ultrafiltration volume.

Statistical analysis

Sample size calculation was based on preliminary data indicating an expected AUC of 0.70 for PVi in predicting intradialytic hypotension. With an estimated symptomatic hypotension incidence of 8% in our maintenance hemodialysis population, 200 patients were required to achieve 90% power to detect a difference of 0.20 between the null hypothesis (AUC = 0.5) and the alternative hypothesis (AUC = 0.7) at a significance level of 0.05 [31].

Descriptive statistics were presented as median (interquartile range) for continuous variables and as frequencies and percentages for categorical variables. Patients were divided into two groups based on the occurrence of intradialytic hypotension (IDH group vs. non-IDH group). Between-group comparisons were performed using Mann-Whitney U tests for continuous variables and chi-square tests or Fisher’s exact tests for categorical variables. ROC curves were constructed to evaluate PVi parameters (PVi baseline, PVi 1 h, ΔPVi, ΔPVi%) and traditional indicators for predicting IDH. DeLong’s method was used to compare AUCs between different parameters for direct assessment of their predictive performance. Correlation analyses using Pearson’s methods were performed to assess the relationship between PVi parameters and the blood pressure reduction during hemodialysis. Multivariate logistic regression analysis was conducted to identify independent predictors of IDH. Sensitivity analyses using alternative definitions of hypotension to assess the robustness of our findings. Patients were stratified into high and low-risk groups based on the optimal PVi parameter cutoff values (Youden index). The Youden index was defined as J = sensitivity + specificity − 1, and the cutoff corresponding to the maximum J was selected. Kaplan-Meier curves were constructed to analyze time to hypotension, with differences between groups assessed by log-rank tests. Cox proportional hazards regression was used to estimate hazard ratios for IDH occurrence. All statistical analyses were performed using SPSS version 26.0 (IBM Corporation, Armonk, NY, USA) and MedCalc version 19.0 (MedCalc Software Ltd, Ostend, Belgium). Data visualization and multiple imputation of data were conducted using Python (Python Software Foundation, https://www.python.org). A two-sided P-value < 0.05 was considered statistically significant.

Results

Patient characteristics and incidence of intradialytic hypotension

Following eligibility screening, 200 maintenance hemodialysis patients were enrolled in this prospective multicenter study. The patient recruitment process is summarized in Figure S1 (study flowchart), which details the screening, exclusion criteria, and final enrollment. The study population had a median age of 57.0 years (interquartile range [IQR]: 52.0–63.0), with male predominance (55.5%). Participants had undergone maintenance hemodialysis for a median duration of 5.0 years (IQR: 2.0–10.0). Missing laboratory parameters were observed during data collection (Figure S2). Variables with significant missingness (e.g., vitamin D, transferrin) were excluded from analysis. Multiple imputation was performed for remaining incomplete parameters. Comparative analysis revealed no significant distributional differences between pre- and post-imputation datasets (Table S1), with density plots demonstrating preserved data distributions (Figure S3). IDH incidence was evaluated based on the first criterion, 61 patients (30.5%) experienced IDH during the observation period. When applying the more stringent second criterion (symptomatic hypotension), 22 patients (11.0%) developed IDH (Fig. 1A). The temporal distribution of hypotensive episodes revealed that IDH most frequently occurred during the second hour of hemodialysis treatment (n = 28, 45.9%), followed by the first hour (n = 16, 26.2%) (Fig. 1B).

Fig. 1.

Incidence and timing of intradialytic hypotension across multiple centers. (A) Incidence rates of intradialytic hypotension across different categories and hospitals. (B) Distribution of onset times for intradialytic hypotension among 61 cases. Overall hypotension: SBP decrease 20 mmHg during hemodialysis; Asymptomatic hypotension: SBP decrease of 20mmHg during hemodialysis without symptoms. Symptomatic hypotension: SBP decrease ≥ 20 mmHg or MAP decrease ≥ 10 mmHg with symptoms such as dizziness or nausea. SBP: Systolic blood pressure; MAP: Mean arterial pressure

Differences in dialysis hemodynamic indicators and PVi by IDH status

Baseline characteristics of both groups are detailed in Table 1. While age, sex distribution, and comorbidity profiles were comparable, the IDH group showed significantly higher baseline blood pressure (SBP: 148.0 vs. 134.0 mmHg; DBP: 82.0 vs. 75.0 mmHg; MAP: 103.0 vs. 95.33 mmHg; all P < 0.001), greater interdialytic weight gain (2.8 vs. 2.2 kg, P < 0.001), and required higher ultrafiltration volumes and rates during dialysis (P < 0.001).

Table 1.

Baseline characteristics in patients without and with intradialytic hypotension

Variables	Non-IDH group (n = 139)	IDH group (n = 61)	P value
Demographic Characteristics
Sex (Male, %)	73(52.52%)	38(62.3%)	0.260
Age (year)	57.0 (53.0, 63.0)	55.0 (46.0, 63.0)	0.143
HD Duration (year)	5.0 (2.0, 10.0)	5.0 (3.0, 10.0)	0.116
End-Stage Renal Disease Duration (year)	5.0 (2.0, 10.25)	6.0 (3.0, 10.0)	0.114
Height (cm)	164.0 (158.0, 170.0)	165.0 (158.0, 170.0)	0.917
Pre-HD Weight (kg)	59.2 (53.55, 69.2)	62.4 (55.0, 70.0)	0.396
Weight Gain Between Sessions (kg)	2.2 (1.6, 2.75)	2.8 (2.1, 3.4)	< 0.001
Systolic Blood Pressure Baseline (mmHg)	134.0 (124.5, 145.0)	148.0 (138.0, 161.0)	< 0.001
Diastolic Blood Pressure Baseline (mmHg)	75.0 (68.0, 83.5)	82.00 (74.0, 89.0)	< 0.001
Mean Arterial Pressure Baseline (mmHg)	95.33 (89.5, 102.67)	103.0 (97.33, 110.33)	< 0.001
Types of Antihypertensive Medications			0.053
0	26 (18.71%)	20 (32.79%)
1–2	77 (55.40%)	24 (39.34%)
3–5	36 (25.91%)	17 (27.87%)
Co-morbidity (%)
Hypertension	123(88.49%)	52(85.25%)	0.684
Diabetes	35(25.18%)	17(27.87%)	0.823
Coronary Artery Disease	19(13.67%)	10(16.39%)	0.775
Others	2(1.44%)	2(3.28%)	0.587
Etiology of end-stage renal disease (%)			0.287
Hypertensive Nephropathy	35(25.18%)	9(14.75%)
Chronic Glomerulonephritis	63(45.32%)	29(47.54%)
Diabetic Nephropathy	34(24.46%)	17(27.87%)
Others	7(5.04%)	6(9.84%)
Recent UF Volume (kg)
Previous Session 1	2.4 (1.75, 3.0)	2.9 (1.9, 3.4)	0.038
Previous Session 2	2.3 (1.6, 3.0)	2.7 (1.90, 3.3)	0.066
Previous Session 3	2.4 (1.75, 3.0)	2.5 (2.0, 3.2)	0.199
Current Session Hemodialysis Parameters
UF Volume (kg)	2.2 (1.7, 2.8)	2.8 (2.2, 3.4)	< 0.001
UF Rate (%)	3.74 (2.87, 4.48)	4.45 (3.66, 5.3)	< 0.001
HD Duration (hour)	4.0 (4.0, 4.0)	4.0 (4.0, 4.0)	0.296
Blood Flow Rate (mL/min)	260.0 (250.0, 260.0)	260.0 (250.0, 260.0)	0.491
Dialysate Flow Rate (mL/min)	500.0 (500.0, 500.0)	500.0 (500.0, 500.0)	0.871
Dialysate Temperature (°C)	36.9 (36.5, 37.0)	36.5 (36.5, 37.0)	0.185
Dialysate Glucose Concentration (mmol/L)	0.0 (0.0, 0.0)	0.0 (0.0, 0.0)	0.961
Dialysate Calcium Concentration (mmol/L)	1.5 (1.5, 1.5)	1.5 (1.5, 1.5)	0.471
HD Mode (%)			< 0.001
HD	69(49.64%)	21(34.43%)
Hemodiafiltration	15(10.79%)	21(34.43%)
High-flux HD	55(39.57%)	19(31.15%)

Data are presented as median (interquartile range) or frequencies with percentages. IDH: Intradialytic hypotension; HD: Hemodialysis; UF: Ultrafiltration

Though baseline PVi values were similar between groups (IDH: 12.0 vs. non-IDH: 14.0, P = 0.085), PVi 1 h showed marked differences (IDH: 20.0 vs. non-IDH: 12.0, P < 0.001). Both ΔPVi and ΔPVi% were significantly greater in the IDH group (7.0 vs. -1.0 and 53.8% vs. -7.14%, respectively; both P < 0.001) (Fig. 2A-D). Subgroup analyses confirmed the consistency of ΔPVi and ΔPVi% differences between IDH and non-IDH groups across various clinical parameters, including age, sex, co-morbidity, HD duration, end-stage renal disease duration, antihypertensive medication use, pre-HD weight, and HD mode (Table 2 and S3). The IDH group showed significantly lower SBP nadirs (122.0 vs. 130.0 mmHg, P < 0.001) and MAP nadirs (88.67 vs. 92.33 mmHg, P = 0.014), but similar DBP nadirs (73.0 vs. 74.0 mmHg, P = 0.460) compared to the non-IDH group (Fig. 2E-G). More importantly, the IDH group had greater reductions in all blood pressure parameters during hemodialysis (SBP: 26.0 vs. 5.0 mmHg; DBP: 8.0 vs. 0.0 mmHg; MAP: 13.67 vs. 2.33 mmHg; all P < 0.001) (Fig. 2H-J). While the non-IDH group maintained stable hemodynamics, the IDH group showed progressive SBP decline with partial recovery toward session end (Fig. 2K-M).

Fig. 2.

Dialysis hemodynamic and PVi parameters comparison between IDH and Non-IDH Patients. This figure compares blood pressure and PVi parameters between patients with IDH and non-IDH groups. Panels (A) and (B) show baseline and 1-hour into HD PVi, respectively, while (C) and (D) depict the change and percentage change in PVi. Panels (E) to (J) illustrate the lowest SBP, DBP, MAP during HD, and the respective decreases in these pressures. Additionally, panels (K), (L), and (M) display the dynamic trends of SBP, DBP, and MAP throughout the HD session. SBP: Systolic blood pressure; DBP: Diastolic blood pressure; MAP: Mean arterial pressure; PVi: Pleth variability index; IDH: Intradialytic hypotension; HD: Hemodialysis; ΔPVi: Change in PVi; ΔPVi%: Percentage change in PVi

Table 2.

Subgroup analysis of ΔPVi% median values between non-IDH and IDH groups

Variables	Subgroup Level	Non-IDH group (n = 139)	IDH group (n = 61)	Sample Size (Non-IDH group / IDH group)	P value
Sex	Female	-11.44 (-31.04, 11.11)	50.00 (24.90, 106.67)	66 / 23	< 0.001
	Male	0.00 (-18.18, 35.29)	55.49 (25.89, 96.67)	73 / 38	< 0.001
Age	≤ 60 years	-6.25 (-27.27, 22.22)	48.33 (22.92, 85.71)	85 / 42	< 0.001
	> 60 years	-7.42 (-25.00, 27.91)	86.67 (38.75, 146.43)	54 / 19	< 0.001
Co-morbidity	No	0.00 (-11.76, 22.22)	15.00 (-3.12, 47.22)	13 / 8	0.293
	Yes	-7.14 (-26.48, 25.69)	70.59 (33.33, 100.00)	126 / 53	< 0.001
HD Duration	≤ 10 years	-6.07 (-25.00, 28.93)	51.92 (25.00, 95.00)	108 / 48	< 0.001
	> 10 years	-18.18 (-36.31, 12.70)	70.59 (27.59, 100.00)	31 / 13	< 0.001
End-stage renal disease Duration	≤ 10 years	-6.07 (-25.00, 30.83)	50.00 (25.00, 87.78)	104 / 47	< 0.001
	> 10 years	-16.67 (-31.37, 12.70)	85.29 (29.02, 100.00)	35 / 14	< 0.001
Antihypertensive Medications	No	0.00 (-15.44, 27.31)	50.00 (18.75, 100.00)	26 / 19	0.002
	Yes	-7.14 (-27.27, 22.22)	55.49 (27.83, 100.00)	113 / 42	< 0.001
Pre-HD Weight (kg)	≤ 50 kg	-7.14 (-28.99, 37.50)	88.89 (83.33, 100.00)	23 / 9	0.001
	> 50 kg	-6.70 (-25.23, 22.22)	50.00 (25.00, 100.00)	116 / 52	< 0.001
HD Mode	HD	-5.88 (-25.00, 20.00)	50.00 (22.22, 86.67)	69 / 21	< 0.001
	Hemodiafiltration	-27.78 (-41.07, 10.10)	76.92 (46.67, 137.50)	15 / 21	< 0.001
	High-flux HD	-7.14 (-21.11, 34.31)	40.00 (2.50, 94.44)	55 / 19	0.002

Data are presented as median (interquartile range). IDH: Intradialytic hypotension; HD: Hemodialysis; ΔPVi%: Change in Pleth variability index percentage

A subgroup comparison between symptomatic and asymptomatic IDH was conducted. Symptomatic IDH patients had significantly lower baseline and nadir blood pressures (SBP 136.5 vs. 154.0 mmHg, P < 0.001; MAP 99.5 vs. 106.7 mmHg, P = 0.012), but showed comparable PVi values throughout dialysis (baseline 14.0 vs. 12.0, P = 0.354; 1 h 20.0 vs. 20.0, P = 0.857). Antihypertensive medication use was more common in the asymptomatic subgroup (82.0% vs. 45.5%, P = 0.007) (Table S4).

Multivariate analysis of factors associated with intradialytic hypotension

Multivariate logistic regression with progressive adjustment models identified ΔPVi% as a significant predictor of IDH (Fig. 3). This association remained robust from the unadjusted model (OR 1.02, 95% CI 1.01–1.03, P < 0.001) through increasingly comprehensive adjustments. In the fully adjusted model accounting for demographics, clinical history, weight parameters, baseline MAP, and laboratory values, ΔPVi% maintained its independent predictive value (OR 1.02, 95% CI 1.01–1.03, P < 0.001). Other significant independent predictors in the final model included elevated baseline MAP (OR 1.08, 95% CI 1.03–1.13, P = 0.001), greater interdialytic weight gain (OR 1.84, 95% CI 1.06–3.20, P = 0.03), and higher serum potassium (OR 2.45, 95% CI 1.19–5.04, P = 0.015) (Fig. 3A-D). When analyzing absolute ΔPVi instead of percentage change, similar results were observed (OR 1.22, 95% CI 1.16–1.28, P < 0.001 in the fully adjusted model) (Fig. 3E). Sensitivity analyses using symptomatic hypotension as an alternative IDH definition confirmed these findings, with ΔPVi% (OR 1.01, 95% CI 1.00-1.02, P = 0.008) and ΔPVi (OR 1.09, 95% CI 1.00-1.19, P = 0.013) maintaining significant associations with IDH across all adjustment models (Figure S4).

Fig. 3.

Multivariate analysis of factors associated with IDH during hemodialysis. Panels (A-D) display progressive adjustment models: (A) Model 1 with ΔPVi% alone; (B) Model 2 adding demographic factors and clinical history; (C) Model 3 further adjusting for weight parameters and baseline MAP; and (D) Model 4 incorporating laboratory values. Panel (E) shows a parallel analysis using ΔPVi. For all models in this analysis, IDH was defined as a SBP reduction of 20mmHg occurring during hemodialysis sessions. MAP: Mean arterial pressure; PVi: Pleth variability index; ΔPVi: Change in PVi; ΔPVi%: Percentage change in PVi

Predictive performance of PVi parameters for IDH

ROC analyses demonstrated excellent predictive value of PVi-derived parameters for intradialytic hypotension (Fig. 4). ΔPVi demonstrated the highest discriminative ability (AUC = 0.84, 95% CI 0.78–0.89), followed by ΔPVi% (AUC = 0.83, 95% CI 0.77–0.88). At the optimal cutoff of 23.53%, ΔPVi% yielded a sensitivity of 78.69% and a specificity of 74.82% for predicting IDH. PVi measurements at 1 h showed lower discriminative performance (AUC = 0.79, 95% CI 0.73–0.85). In contrast, hemodynamic parameters exhibited moderate predictive value (baseline MAP/SBP AUC 0.71), while ultrafiltration-related indices (AUC 0.56–0.68) and serum potassium (AUC 0.65) showed limited predictive capacity. Serum albumin had the poorest performance (AUC < 0.60). Pairwise comparisons (Figure S5) confirmed significantly better performance of ΔPVi and ΔPVi% versus all conventional parameters (maximum AUC difference 0.40, P < 0.05).

Fig. 4.

Forest plot of AUCs for various predictors of intradialytic hypotension. PVi: Pleth variability index; ΔPVi: Change in PVi; ΔPVi%: Percentage change in PVi; UF: Ultrafiltration

Correlation of PVi with hemodialysis variables

Correlation analyses revealed distinct associations between PVi parameters and conventional hemodialysis measures (Figure S6). Baseline PVi showed negligible correlations with conventional hemodialysis measures (all P > 0.05) (Figure S6.A-E). In contrast, PVi 1 h demonstrated stronger correlations with weight gain (R = 0.16, P = 0.027), ultrafiltration volume (R = 0.17, P = 0.015), and hemodynamic changes (dSBP R = 0.36, P < 0.001) (Figure S6.F-J). Dynamic PVi parameters exhibited the strongest associations: ΔPVi and ΔPVi% both significantly correlated with weight gain (R = 0.16–0.20), ultrafiltration (R = 0.13–0.16), and blood pressure changes (dSBP R = 0.41–0.42, P < 0.001) (Figure S7.K-T). The strongest correlations were consistently observed with SBP changes.

Clinical outcomes and risk stratification by ΔPVi%

To further evaluate the clinical significance of PVi changes, we stratified patients using a ΔPVi% cutoff value of 23.53% (Youden index), dividing the cohort into two groups: ΔPVi% ≤ 23.53% (n = 117) and ΔPVi% > 23.53% (n = 83). The high ΔPVi% group had more males (66.27% vs. 47.86%, P = 0.015), higher baseline SBP (145 vs. 133 mmHg, P < 0.001) and MAP (102 vs. 95.33 mmHg, P < 0.001), and greater weight gain (2.5 vs. 2.3 kg, P = 0.008) requiring higher ultrafiltration (2.6 vs. 2.3 kg, P = 0.016) (Table S5). While serum sodium levels were slightly lower in the high ΔPVi% group (137.8 vs. 139.1 mmol/L, P = 0.018), serum chloride was also modestly reduced (100.90 vs. 99.60 mmol/L, P = 0.032), and beta-2 microglobulin levels were lower in the high ΔPVi% group (35.10 vs. 31.90 mg/L, P = 0.047); all other laboratory parameters showed no significant intergroup differences (all P > 0.05) (Table S6). PVi analysis showed lower baseline (10 vs. 15, P < 0.001) but higher 1-hour values (19 vs. 12, P < 0.001) in the high ΔPVi% group, resulting in greater changes (ΔPVi:7 vs. -1; ΔPVi%:71.43% vs. -11.76%, both P < 0.001) (Figure S7A-D). Hemodynamically, this group had larger blood pressure reductions (SBP: 20 vs. 7, DBP:4 vs. 0, MAP:9.33 vs. 2.67 mmHg, all P < 0.05) with an initial steep decline followed by partial stabilization (Figure S7E-J). Kaplan-Meier analysis demonstrated significantly higher hypotension incidence in patients with ΔPVi% > 23.53% versus ≤ 23.53% (57.8% vs. 11.1% by session end; log-rank χ²=52.00, P < 0.001). The ΔPVi% > 23.53% group showed earlier (15.7% vs. 2.6% at 1 h) and more frequent hypotension (41.0% vs. 8.5% at 2 h), with a 7.15-fold increased risk (HR = 7.15, 95%CI 3.86–13.22) (Fig. 5).

Fig. 5.

Kaplan-Meier curves for hypotension-free survival by ΔPVi% (Cutoff: 23.53%). This figure presents a Kaplan-Meier curve comparing hypotension-free survival probability between patients with ΔPVi% >23.53 (blue line, n = 83) and ΔPVi% ≤23.53 (orange line, n = 117) during a 4-hour hemodialysis session. ΔPVi%: Percentage change in pleth variability index; HR: Hazard ratio; CI: Confidence interval

Discussion

This prospective multicenter study represents the first systematic investigation into the utility of PVi and its dynamic changes as predictive markers for IDH in maintenance hemodialysis patients. Our findings compellingly demonstrate that PVi 1 h and its dynamic alterations from baseline (ΔPVi, ΔPVi%) possess robust predictive capacity for IDH, evidenced by AUC values ranging from 0.79 to 0.84. Furthermore, an optimal ΔPVi% threshold of 23.53% effectively stratified patients into distinct risk categories, with individuals in the high-risk group exhibiting a nearly seven-fold increased hazard for developing IDH.

The rationale for investigating PVi in hemodialysis stems from its well-established predictive capacity for fluid responsiveness in acute care settings. In critical care, PVi demonstrates excellent performance (AUC 0.93) for predicting hypotension in mechanically ventilated patients [25]. Surgical studies confirm its utility during anesthesia, with AUCs of 0.615–0.824 for anticipating induction hypotension [24, 32]. Two meta-analyses (n = 1,437-2,115 patients) consistently show PVi outperforms static hemodynamic parameters (summary AUC ≈ 0.75) [33, 34]. Given this robust evidence in acute settings, we hypothesized PVi could similarly predict intradialytic hypotension given hemodialysis’s characteristic fluid shifts. Furthermore, recent evidence in spontaneously breathing hemodialysis patients supports this physiological link. Dağar and Uzunosmanoğlu reported that mean PVi values increased significantly from 20.7 ± 5% to 27.7 ± 6% after the hemodialysis session (P < 0.001), and the change in PVi strongly correlated with ultrafiltration volume (r = 0.744, P < 0.001) [35]. This study directly demonstrates that PVi reflects volume shifts during HD, lending mechanistic plausibility to our finding that greater increases in PVi are associated with IDH events.

This potential advantage becomes particularly compelling when contrasted with current clinical practice. Traditional indicators such as interdialytic weight gain [1, 4], ultrafiltration volume [1, 5], baseline blood pressure [1, 4], and blood volume monitoring [14] have consistently demonstrated limited sensitivity and specificity. For example, classical predictors such as vascular pedicle width and cardiothoracic ratio exhibited poor predictive accuracy with sensitivities below 25% and modest specificities [36]. Similarly, blood volume monitoring parameters have shown poor correlation with intradialytic MAP and systolic blood pressure changes and suffer from low inter-rater reliability [14, 37, 38]. While integrative scores such as the SOCRATE (cardiovascular sequential organ failure assessment, index capillary refill time and lactate level) score have achieved an AUC of 0.79 for hypotension prediction, they lack the granularity of continuous, beat-to-beat monitoring [39]. Although machine learning models for IDH prediction have demonstrated exceptional performance (AUROC: 0.89–0.94) in recent studies [17–21], they predominantly depend on aggregated clinical parameters rather than real-time physiological indicators such as PVi. The limited success of these traditional and some advanced static approaches largely reflects their inability to capture the dynamic cardiovascular responses during dialysis.

In contrast to static measures, dynamic indices offer a more physiologically relevant assessment of hemodynamic status during IHD [40]. For instance, Mostafa et al. demonstrated that peripheral perfusion index and heart rate variability yielded moderate predictive performance for IDH (AUCs of 0.72 and 0.76, respectively) [27], however, these are often considered indirect surrogates of hemodynamic responsiveness. PVi, as a dynamic and non-invasive plethysmographic index, directly reflects respiratory induced variations in perfusion [22], providing continuous insight into the patient’s cardiovascular adaptive capacity during the volume shifts inherent to hemodialysis. This capacity likely explains the superior predictive performance observed in our study, where PVi changes outperformed conventional markers by substantial margins in AUC improvement (0.13–0.40). Such enhanced predictive validity aligns with the pathophysiology of IDH, which is often characterized by impaired autonomic and vascular compensatory mechanisms under the stress of fluid removal [26], particularly in critically ill populations [41].

Our study revealed an important paradox: patients who developed IDH had higher pre-dialysis SBP yet experienced greater declines in MAP. This aligns with prior reports identifying both high (diastolic BP ≥ 90 mmHg; OR 1.81) and low (< 60 mmHg; OR 2.72) pre-dialysis pressures as independent IDH risk factors [28, 42]. Likely mechanisms include larger interdialytic weight gain and higher ultrafiltration rates in hypertensive patients, as well as autonomic dysfunction and arterial stiffness that blunt compensatory vasoconstriction [5, 26, 43]. While pre-dialysis BP serves as a static risk marker, it lacks precision for imminent event prediction. In contrast, our ΔPVi% > 23.53 threshold identified a 7.15-fold higher IDH risk, unmasking latent hemodynamic instability and offering a dynamic, actionable predictor beyond conventional BP measurement. However, despite this apparent advantage of ΔPVi%, it is crucial to interpret its predictive performance in terms of both sensitivity and specificity. In our cohort, the cutoff of ΔPVi% > 23.53%, derived from the Youden index, provided a balanced profile, with a sensitivity of 78.69%, a specificity of 74.82%, and an AUC of 0.83. Given the potential harm of IDH, clinicians may reasonably prioritize sensitivity to avoid missing high-risk patients; however, very low specificity would lead to unnecessary interventions and additional workload. The current threshold therefore represents a pragmatic compromise for risk stratification in this population. In different clinical settings, the cutoff could be shifted to emphasize either sensitivity or specificity according to local risk tolerance and resource availability, and external validation is warranted to refine these thresholds.

Several hemodynamic mechanisms likely underpin PVi’s predictive value in this context. Increased respiratory variations in stroke volume and pulse pressure indicate proximity to the steep portion of the Frank-Starling curve, where cardiac output becomes highly sensitive to preload changes [44, 45]. During hemodialysis, ultrafiltration progressively diminishes cardiac filling, potentially transitioning the heart toward this preload-dependent state in susceptible individuals. The one-hour PVi measurement, therefore, can be conceptualized as a “stress test,” capturing the individual’s cardiovascular compensation to initial volume removal and thereby predicting their tolerance to continued ultrafiltration. This mechanistic understanding explains why ΔPVi% not only outperformed static parameters but also correlated strongly with subsequent reductions in systolic blood pressure. Nevertheless, subanalysis of symptomatic and asymptomatic episodes indicated that PVi changes, while predictive of IDH occurrence, did not distinguish symptom status, suggesting that symptom manifestation is primarily determined by absolute blood pressure thresholds and individual neurovascular sensitivity rather than dynamic perfusion variability.

Our study possesses several methodological strengths and offers significant innovations. Its prospective, multicenter design, coupled with rigorous data collection and comprehensive statistical analysis, provides high-quality evidence regarding PVi’s clinical utility in the hemodialysis population. The proposed ΔPVi% threshold of 23.53% represents a simple, readily applicable metric for bedside risk stratification. Furthermore, the significant correlation observed between PVi parameters and conventional hemodynamic measurements validates its physiological relevance while simultaneously demonstrating its enhanced predictive capability over these traditional markers.

Despite these strengths, several limitations merit note. First, the cohort, though diverse, was from a single geographic region. Second, PVi may be affected by atypical cardiac rhythms and peripheral vascular disease, with residual confounding possible [46]. Third, the timing of measurements may affect interpretability: approximately 26% of patients developed IDH before the 1-hour PVi assessment. For these individuals, the 1-hour PVi cannot be interpreted as prognostic and likely reflects a contemporaneous or post-event state, which may bias associations toward reverse causality in this subgroup. To mitigate this limitation in future studies, we will implement earlier PVi monitoring (e.g., within 5 min after dialysis initiation or post-blood draw) to improve temporal resolution and pre-event detection. Fourth, PVi was obtained at discrete time points rather than continuously, potentially missing transient predictive changes. Finally, results were generated using the Masimo M3002A device and require validation with other platforms.

Conclusions

In conclusion, this study establishes that dynamic PVi parameters, particularly changes from baseline, are promising, non-invasive, and independent predictors of intradialytic hypotension. By enabling early identification of at-risk patients, PVi-guided management strategies hold considerable potential to improve dialysis safety and enhance patient outcomes. Further interventional studies are warranted to robustly validate the clinical impact of PVi-guided protocols in the routine care of hemodialysis patients.

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

AUC

Area under the receiver operating characteristic curve

CI

Confidence interval

DBP

Diastolic blood pressure

ΔPVi

Absolute change in pleth variability index

ΔPVi%

Percentage change in pleth variability index

HD

Hemodialysis

HR

Hazard ratio

ICU

Intensive care unit

IDH

Intradialytic hypotension

IHD

Intermittent hemodialysis

IQR

Interquartile range

MAP

Mean arterial pressure

PVi

Pleth variability index

PVi 1h

Pleth variability index at 1 h after the start of hemodialysis

PVi baseline

Pleth variability index at baseline (before hemodialysis)

ROC

Receiver operating characteristic

SBP

Systolic blood pressure

Author contributions

Y.W. participated in research design, the writing of the paper and data analysis. J.L. participated in research design, the writing of the paper, the performance of the research and data analysis. W.L. participated in research design and data analysis. J.Z. and C.T. participated in research design. Q.X. participated in the writing of the paper and data analysis. F.L., T.Y., Y.L. and C.G. participated in the performance of the research. H.C. participated in the data analysis.

Funding

Part of this study is funded by a grant from the Anhui Province Clinical Medical Research Translation Special Project (202304295107020001, 202304295107020006), and the Key Scientific Research Foundation of the Education Department of Province Anhui (2025AHGXZK30670).

Data availability

The datasets used and analyzed during the current study are available from the corresponding author on reasonable request.

Declarations

Ethics approval and consent to participate

The Institutional Review Boards approved this study of all participating centers: Scientific Research and New Technology Ethics Committee of Wannan Medical College Yijishan Hospital, the lead center (Approval No. 2025-61); Medical Ethics Committee of Wuhu Second People’s Hospital (Approval No. 2025-KY-016); Medical Ethics Committee of Wuhu Jinghu District Hospital (Approval No. 20250124-08); Medical Ethics Committee of Wuhu Guangji Hospital (Approval No. WHGJYY-LLSP-2025-01); and Medical Ethics Committee of Wuhu Traditional Chinese Medicine Hospital (Approval No. YW-2025-002). Written informed consent was obtained from all participants before enrollment. This study was conducted in accordance with the Declaration of Helsinki and Good Clinical Practice guidelines.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.