Dynamic changes in vasoactive-inotropic score and their prognostic value in severe sepsis and septic shock patients undergoing polymyxin B hemoperfusion: a real-world Asian ICU cohort study

Abstract

Severe sepsis and septic shock frequently require escalating vasoactive support, but the prognostic utility of vasoactive-inotropic score (VIS) trajectories during polymyxin B hemoperfusion (PMX-HP) in Asian ICUs remains unclear. We conducted a retrospective single-center cohort study in a tertiary medical ICU in Taiwan, enrolling consecutive adults with severe sepsis or septic shock who received PMX-HP between July 2013 and December 2019. VIS was calculated before PMX-HP (VIS1), immediately after PMX-HP (VIS2), and approximately 24 h post-treatment (VIS3); the primary outcome was 28-day all-cause mortality. In 64 patients, median VIS decreased from 28.8 (IQR 13.0-47.6) at VIS1 to 18.1 (IQR 7.8–45.0) at VIS2 and was 16.9 (IQR 1.3–45.1) at VIS3 among patients with available VIS3 data. VIS1, VIS2, and VIS3 were higher in non-survivors than survivors. A mixed-effects model using log(VIS + 1) confirmed significant effects of time, survival status, and their interaction. In an exploratory time-dependent Cox model, higher time-updated log(VIS + 1) was associated with 28-day mortality. The immediate VIS change showed limited discrimination for 28-day mortality (AUC 0.57, 95% CI 0.43–0.72), whereas VIS3 demonstrated better discrimination (AUC 0.78, 95% CI 0.66–0.91). Adding VIS3 to APACHE II improved discrimination compared with APACHE II alone (AUC 0.81 vs. 0.63; Delta AUC + 0.18; p = 0.020). Kaplan-Meier analysis using the ROC-derived VIS3 cut-off of 14.56 showed significantly worse 28-day survival when VIS3 remained above this threshold. Late post-treatment VIS monitoring may support bedside risk stratification in severe sepsis and septic shock undergoing PMX-HP; multicenter validation is warranted.

Clinical trial number Not applicable.

Introduction

Sepsis and septic shock remain major causes of morbidity and mortality in intensive care units (ICUs) worldwide, accounting for a substantial proportion of global deaths and healthcare utilization [1–3]. International guidelines emphasize time-sensitive resuscitation, early antimicrobial therapy, and prompt source control, and treatment delays have been associated with worse outcomes [4–6]. Contemporary estimates also indicate that the measured incidence of sepsis is sensitive to case-identification strategy [7]. Current consensus definitions conceptualize sepsis as life-threatening organ dysfunction arising from a dysregulated host response to infection, with septic shock representing a high-risk subset characterized by profound circulatory and cellular/metabolic abnormalities [8–10]. The clinical course is heterogeneous and dynamic, reflecting evolving interactions among inflammation, endothelial dysfunction, microcirculatory impairment, and organ failure trajectories [11, 12]. Accordingly, practical and reproducible bedside markers that capture the intensity and evolution of shock physiology are essential for risk stratification and for evaluating response to adjunctive therapies, alongside established illness-severity indices such as APACHE II [13].

Vasopressor and inotrope requirements are central operational signals in septic shock management, but single-agent dosing does not adequately represent multi-drug hemodynamic support. The vasoactive-inotropic score (VIS) integrates commonly used vasoactive and inotropic agents into a single quantitative metric, enabling standardized longitudinal assessment of shock severity and treatment intensity across time. In adult sepsis cohorts, higher VIS has been associated with worse outcomes, supporting its role as a pragmatic surrogate of cardiovascular failure burden [14]. However, in real-world settings where extracorporeal adjuncts are used, the prognostic meaning of VIS at clinically relevant peri-procedural time points - and the incremental value of VIS trajectories beyond single time-point measurements - remains incompletely defined.

Polymyxin B hemoperfusion (PMX-HP) was developed to adsorb circulating endotoxin and has been used as an adjunctive blood purification strategy in septic shock, particularly when Gram-negative endotoxemia is suspected [15, 16]. Endotoxin-mediated pathways plausibly contribute to vasoplegia and hemodynamic collapse, providing a mechanistic rationale for extracorporeal endotoxin removal [15]. Nevertheless, the clinical evidence base for PMX-HP remains controversial. Observational analyses and real-world reports have suggested possible hemodynamic improvements and potential outcome benefits in selected populations [16], whereas large randomized trials have failed to demonstrate consistent mortality reduction in broader septic shock cohorts, underscoring substantial heterogeneity in timing, patient selection, and biological target engagement [17]. Meta-analyses have likewise reported mixed conclusions, with effect estimates sensitive to study design, population definitions, and risk of bias, leaving unresolved questions regarding who - if anyone - derives clinically meaningful benefit from PMX-HP [18, 19]. In this context, physiologically grounded, dynamic markers such as VIS may be particularly valuable to characterize post-intervention trajectories and to identify high-risk phenotypes despite advanced supportive care.

Regional practice contexts further motivate focused evaluation. Sepsis epidemiology, case mix, and care pathways in Taiwan and other Asian health systems differ from many Western trial environments, and nationwide data indicate a high burden of severe infection and sepsis-related critical illness [20, 21]. PMX-HP is used in some Asian ICUs as part of routine care for refractory shock, but region-specific real-world data linking peri-procedural hemodynamic trajectories to outcomes remain limited [22, 23]. Clarifying how VIS behaves around PMX-HP and how VIS-based metrics relate to mortality may inform practical bedside risk stratification and guide escalation planning in settings where PMX-HP is deployed.

Therefore, we conducted a retrospective cohort study in a Taiwanese tertiary medical ICU with three prespecified aims: first, to characterize the temporal trajectory of VIS before PMX-HP, immediately after PMX-HP, and approximately 24 h after PMX-HP; second, to evaluate the association between serial VIS measurements and 28-day mortality using longitudinal and time-to-event analyses; and third, to determine whether post-treatment VIS provides incremental prognostic value beyond baseline APACHE II score.

Materials and methods

Study design and setting

We conducted a retrospective cohort study in the medical intensive care unit (ICU) of MacKay Memorial Hospital, Taipei, Taiwan, a tertiary referral center. The ICU is a 25-bed closed unit staffed by intensivists and provides advanced organ support, including mechanical ventilation and extracorporeal organ support. The study period spanned from 1 July 2013 to 31 December 2019. Polymyxin B hemoperfusion (PMX-HP) was available as part of routine clinical practice during the study period. No formal a priori sample-size calculation was performed; instead, the study cohort comprised all consecutive eligible patients treated during the prespecified study period.

Participants, case identification, and eligibility criteria

We screened consecutive adult ICU patients (age > = 20 years) who received at least one session of PMX-HP during their ICU admission between 1 July 2013 and 31 December 2019. Patients were eligible if they were treated for severe sepsis or septic shock requiring ICU-level care and underwent PMX-HP as an adjunctive therapy in routine practice. Given the long study timeframe, sepsis-related diagnoses were identified using contemporaneous clinical documentation and physiologic criteria consistent with Surviving Sepsis Campaign guidance during the study period. For earlier admissions (pre-2016), severe sepsis was considered infection accompanied by acute organ dysfunction. Septic shock was considered persistent hypotension requiring vasopressors despite adequate fluid resuscitation; for admissions after 2016, septic shock definitions were aligned with Sepsis-3 criteria when applicable (vasopressor requirement to maintain mean arterial pressure > = 65 mmHg and serum lactate > 2 mmol/L despite adequate fluid resuscitation).

Patients were excluded if they were younger than 20 years, pregnant, participating in another interventional clinical trial within one month, had undergone solid-organ transplantation within one year, had an anticipated survival of less than 30 days (e.g., advanced malignancy), had cardiopulmonary resuscitation within four weeks prior to ICU admission, had a do-not-resuscitate or palliative care order at admission, had known HIV infection, had active uncontrolled bleeding within 24 h, were brain dead at admission, had end-stage liver failure (Child-Pugh class C), had hemophilia, had a known allergy to polymyxin B, heparin, or extracorporeal circuit materials, or had received extracorporeal blood purification (e.g., continuous veno-venous hemofiltration/hemodialysis or plasma exchange) within 24 h prior to PMX-HP initiation.

PMX-HP procedure and treatment documentation

PMX-HP was performed using a polymyxin B-immobilized fiber hemoperfusion cartridge (Toraymyxin^®, PMX-20R; Toray Industries, Tokyo, Japan). Each treatment session typically lasted 2 h, with a blood flow rate of approximately 80–120 mL/min, in accordance with institutional practice. Anticoagulation was performed with systemic unfractionated heparin per ICU protocol unless contraindicated. The number of sessions and timing of PMX-HP initiation were determined by the attending intensivist based on hemodynamic status and overall clinical judgment. For each included patient, PMX-HP treatment characteristics were extracted from the electronic medical record, including the number of sessions, timing of initiation, and delivered session duration.

Data sources and variable collection

Clinical, demographic, laboratory, and microbiological variables were retrospectively extracted from the electronic medical record using a standardized data collection form. Variables included age, sex, body weight, major comorbidities, APACHE II score at ICU admission, infection source and causative pathogens (based on available culture results and clinical documentation), use of organ support (including continuous renal replacement therapy [CRRT] and extracorporeal membrane oxygenation [ECMO]), PMX-HP treatment characteristics, vasoactive/inotropic agent doses at predefined time points relative to PMX-HP, and paired lactate values when available. Mechanical ventilation status at baseline was recorded, and ventilator-duration fields were reviewed where available in the structured dataset. Mortality outcomes (28-day, ICU, and hospital mortality) were obtained from hospital records.

Vasoactive-inotropic score (VIS)

The vasoactive-inotropic score (VIS) was calculated at three predefined time points: before PMX-HP (VIS1), immediately after PMX-HP (VIS2), and approximately 24 h after PMX-HP (VIS3). VIS was computed using the following standard formula: VIS = dopamine (µg/kg/min) + dobutamine (µg/kg/min) + 100 x epinephrine (µg/kg/min) + 100 x norepinephrine (µg/kg/min) + 10,000 x vasopressin (U/min). Catecholamine and inotrope doses were normalized by body weight (µg/kg/min), and vasopressin was recorded as U/min. A VIS of 0 indicated no vasoactive or inotropic support at that time point.

To minimize measurement variability in this retrospective dataset, time windows were prespecified as follows: VIS1 was determined from the vasoactive/inotropic doses recorded closest to the PMX-HP start time within the preceding 1 h; VIS2 was determined from doses recorded closest to the PMX-HP end time within 1 h after completion; and VIS3 was determined from doses recorded closest to 24 h after PMX-HP completion within a window of 20–28 h. When multiple records were available within a time window, the value with the timestamp closest to the target time point was selected. If no documentation was available within the defined window for a given time point, VIS at that time point was treated as missing.

Outcomes

The primary outcome was 28-day all-cause mortality, defined as death from any cause within 28 days after initiation of the first PMX-HP session (day 0). Secondary outcomes included ICU mortality, hospital mortality, ICU and hospital length of stay, and VIS dynamics across the predefined time points, including the immediate change (DeltaVIS2-VIS1) and late change (DeltaVIS3-VIS1).

Statistical analysis

Continuous variables were summarized as mean +/- standard deviation (SD) or median (interquartile range, IQR), as appropriate, and categorical variables were summarized as counts and percentages. Comparisons between 28-day survivors and non-survivors were performed using Student’s t-test or the Mann-Whitney U test for continuous variables and the chi-square test or Fisher’s exact test for categorical variables. In addition to absolute VIS values and absolute VIS changes (DeltaVIS2-VIS1 and DeltaVIS3-VIS1), exploratory percentage-change metrics were calculated as [(VIS2-VIS1)/VIS1] x 100 and [(VIS3-VIS1)/VIS1] x 100 in patients with baseline VIS1 > 0, because percentage change is undefined when VIS1 = 0.

To address the repeated-measures structure of serial VIS, a linear mixed-effects model was constructed using log(VIS + 1) as the dependent variable because VIS was right-skewed and included zero values. Fixed effects included measurement time, 28-day survival status, and the time x survival-status interaction, with a random intercept for each patient. Likelihood-ratio tests were used to evaluate the overall effects of time, survival status, and their interaction.

An exploratory time-dependent Cox proportional hazards model was also constructed using VIS as a time-updated covariate across three predefined intervals: VIS1 from PMX-HP initiation to PMX-HP completion, VIS2 from PMX-HP completion to approximately 24 h post-treatment, and VIS3 thereafter until death or day-28 censoring. The time-dependent Cox model used log(VIS + 1), and an APACHE II-adjusted model was additionally examined.

Receiver operating characteristic (ROC) curve analysis and area under the curve (AUC) were used to assess binary-outcome discrimination. AUC 95% confidence intervals were estimated by DeLong’s method. The incremental prognostic value of VIS3 beyond APACHE II was evaluated by comparing the AUC of APACHE II alone with that of a logistic model including APACHE II and log(VIS3 + 1), using DeLong’s test for correlated ROC curves. Exploratory Kaplan-Meier analysis of 28-day survival was performed using the optimal ROC-derived VIS3 cut-off, with groups compared by log-rank test; a number-at-risk table was added.

To reduce survivorship bias in duration-based outcomes, ICU length of stay was summarized only among ICU survivors, and hospital length of stay was summarized only among hospital survivors. Between-group p-values for length-of-stay comparisons between 28-day survivors and non-survivors were therefore not calculated. Paired lactate change was explored in available cases, but exact synchronization with the VIS2/VIS3 sampling windows could not be guaranteed in this retrospective dataset. Cut-off values derived from the same cohort were regarded as exploratory. All statistical tests were two-sided, and p < 0.05 was considered statistically significant. Analyses were performed using IBM SPSS Statistics and R/Python statistical scripts, and figures were generated using standard statistical plotting functions.

Availability of materials, data, code, and protocols

This retrospective study used routinely collected clinical data. Due to patient confidentiality and institutional policy, the raw clinical dataset cannot be made publicly available. De-identified analytic datasets, statistical analysis code (R/Python scripts), and a summary of the institutional PMX-HP procedure protocol can be made available from the corresponding author upon reasonable request, subject to institutional requirements and applicable regulations. Any restrictions on access will be communicated at the time of request.

Ethical approval

This study was conducted in accordance with the Declaration of Helsinki and was approved by the Institutional Review Board of MacKay Memorial Hospital, Taipei, Taiwan (IRB No. 18MMHIS198e; approval date: 4 March 2019; approved study period: 4 March 2019 to 3 March 2020). The requirement for informed consent was waived due to the retrospective design and the use of de-identified clinical data.

Results

Patient characteristics

A total of 94 adult ICU patients with severe sepsis or septic shock who received polymyxin B hemoperfusion (PMX-HP) were screened. After applying the prespecified exclusion criteria, 64 patients were included in the final cohort and followed for 28-day outcomes (Fig. 1). Of these, 34 were 28-day survivors and 30 were non-survivors.

Fig. 1.

Study flowchart of screening, inclusion, and exclusion

Baseline characteristics, infection sources, and microbiology are summarized in Table 1. Compared with survivors, non-survivors more frequently required mechanical ventilation at baseline (80.0% vs. 52.9%, p = 0.04). Urinary tract infection was more common among survivors (29.4% vs. 3.3%, p = 0.02), whereas infection sources categorized as other were more common among non-survivors (26.7% vs. 5.9%, p = 0.04). APACHE II scores also tended to be higher among non-survivors, although the between-group difference did not reach statistical significance.

Table 1.

Baseline characteristics of the study cohort (N = 64)

Characteristic	All Patients (N = 64)	Survivors (N = 34)	Non-survivors (N = 30)	p-value
Age, years	66.1 +/- 12.3	64.3 +/- 13.8	68.2 +/- 10.3	0.20
Male sex, n (%)	43 (67.2%)	20 (58.8%)	23 (76.7%)	0.21
Body weight, kg	66.5 +/- 15.2	66.8 +/- 14.3	66.1 +/- 16.4	0.86
APACHE II score	26 (21–32)	25 (20–28)	28 (23–33)	0.10
Diabetes mellitus, n (%)	32 (50.0%)	16 (47.1%)	16 (53.3%)	0.80
Hypertension, n (%)	35 (54.7%)	21 (61.8%)	14 (46.7%)	0.34
CKD/ESRD, n (%)	26 (40.6%)	14 (41.2%)	12 (40.0%)	> 0.99
Source of infection
- Pneumonia, n (%)	19 (29.7%)	7 (20.6%)	12 (40.0%)	0.16
- Intra-abdominal infection, n (%)	19 (29.7%)	13 (38.2%)	6 (20.0%)	0.19
- Urinary tract infection, n (%)	11 (17.2%)	10 (29.4%)	1 (3.3%)	0.02
- Soft tissue infection, n (%)	5 (7.8%)	2 (5.9%)	3 (10.0%)	0.66
- Others, n (%)	10 (15.6%)	2 (5.9%)	8 (26.7%)	0.04
Major pathogens
- Escherichia coli, n (%)	20 (31.3%)	14 (41.2%)	6 (20.0%)	0.12
- Klebsiella pneumoniae, n (%)	14 (21.9%)	9 (26.5%)	5 (16.7%)	0.52
- Others, n (%)	30 (46.9%)	11 (32.4%)	19 (63.3%)	0.03
Mechanical ventilation use, n (%)	42 (65.6%)	18 (52.9%)	24 (80.0%)	0.04
CRRT use, n (%)	43 (67.2%)	19 (55.9%)	24 (80.0%)	0.061
ECMO use, n (%)	3 (4.7%)	1 (2.9%)	2 (6.7%)	0.596

Note: Continuous variables are presented as mean +/- standard deviation (SD) or median (interquartile range, IQR); categorical variables are presented as number (percentage). p-values were calculated using Student’s t-test, Mann-Whitney U test, chi-square test, or Fisher’s exact test, as appropriate. Abbreviations: UTI, urinary tract infection; CKD, chronic kidney disease; ESRD, end-stage renal disease; CRRT, continuous renal replacement therapy; ECMO, extracorporeal membrane oxygenation

VIS dynamics and group comparison

The vasoactive-inotropic score (VIS) was assessed at three predefined time points: before PMX-HP (VIS1), immediately after PMX-HP (VIS2), and approximately 24 h after PMX-HP (VIS3). Overall, the cohort-level median VIS decreased from 28.8 (13.0-47.6) at VIS1 to 18.1 (7.8–45.0) at VIS2 and was 16.9 (1.3–45.1) at VIS3 among patients with available VIS3 data (Table 2).

Table 2.

Distribution of VIS at three predefined time points by survival status

VIS time point	Available N	All Patients	Survivors	Non-survivors	p-value
VIS1 (pre-PMX-HP)	64	28.8 (13.0-47.6)	24.2 (6.1–35.2)	39.8 (20.8–70.4)	0.029
VIS2 (immediately post-PMX-HP)	64	18.1 (7.8–45.0)	10.7 (0.0-21.1)	36.3 (16.5–62.5)	0.001
VIS3 (~ 24 h post-PMX-HP)	60	16.9 (1.3–45.1)	5.2 (0.0-20.2)	41.4 (18.2–53.7)	< 0.001

Values are presented as median (interquartile range, IQR). Between-group comparisons were performed using the Mann-Whitney U test. Available N is shown because VIS3 was unavailable in four patients who died before or around the 24-hour measurement window. Abbreviations: VIS, vasoactive-inotropic score; PMX-HP, polymyxin B hemoperfusion

When stratified by 28-day survival, VIS was consistently higher among non-survivors than survivors across all three time points. Specifically, non-survivors had higher VIS1 (39.8 (20.8–70.4) vs. 24.2 (6.1–35.2); p = 0.029), higher VIS2 (36.3 (16.5–62.5) vs. 10.7 (0.0-21.1); p = 0.001), and higher VIS3 (41.4 (18.2–53.7) vs. 5.2 (0.0-20.2); p = < 0.001) compared with survivors (Table 2; Fig. 2).

Fig. 2.

Individual VIS trajectories by measurement time and 28-day survival status. Individual thin lines denote patient-level trajectories and bold lines denote group medians. The y-axis is log(VIS + 1) because VIS was right-skewed and included zero values. The p-value annotations were generated from the same analytic dataset as Table 2

Derived VIS metrics, percentage change, and exploratory prognostic analyses

Absolute VIS change and percentage VIS change were explored in addition to predefined VIS values. The immediate absolute change (DeltaVIS2-VIS1) did not differ significantly between survivors and non-survivors (-6.1 (-13.4-0.0) vs. -0.8 (-12.7-4.0); p = 0.315). In contrast, the late absolute change (DeltaVIS3-VIS1) was more favorable in survivors (-9.4 (-20.6–0.4) vs. -0.8 (-9.1-17.8); p = 0.019). Among patients with baseline VIS1 > 0, the immediate percentage change was not significantly different between groups (-33.3 (-73.3–10.0) vs. -6.7 (-50.0-0.0); p = 0.135), whereas the late percentage change was significantly more negative in survivors (-66.7 (-99.7–22.9) vs. -6.7 (-47.2-60.0); p = 0.002) (Table 3).

Table 3.

Derived VIS metrics, percentage change, and exploratory prognostic performance for 28-day mortality

Metric	Available N	Group comparison (survivors vs. non-survivors)	AUC (95% CI)	Additional notes
DeltaVIS2-VIS1 (immediate absolute change)	64	-6.1 (-13.4-0.0) vs. -0.8 (-12.7-4.0); p = 0.315	0.57 (0.43–0.72)	Early change alone poorly discriminatory
DeltaVIS3-VIS1 (late absolute change)	60	-9.4 (-20.6–0.4) vs. -0.8 (-9.1-17.8); p = 0.019	0.68 (0.54–0.82)	More favorable among survivors
DeltaVIS2-VIS1 (%)	58	-33.3 (-73.3–10.0) vs. -6.7 (-50.0-0.0); p = 0.135	0.61 (0.47–0.76)	Baseline VIS1 > 0 only
DeltaVIS3-VIS1 (%)	54	-66.7 (-99.7–22.9) vs. -6.7 (-47.2-60.0); p = 0.002	0.75 (0.61–0.88)	Stronger late prognostic signal
VIS3 absolute score	60	5.2 (0.0-20.2) vs. 41.4 (18.2–53.7); p = < 0.001	0.78 (0.66–0.91)	Optimal cut-off 14.56; sensitivity 0.88; specificity 0.71

Percentage-change analyses excluded patients with baseline VIS1 = 0 and used available-case analysis when VIS3 was missing. AUC 95% confidence intervals were estimated using DeLong’s method. ROC values and cut-offs were derived from the same cohort and should therefore be considered exploratory

Exploratory ROC analysis showed that the immediate absolute change in VIS had limited discrimination for 28-day mortality (AUC 0.57, 95% CI 0.43–0.72), supporting the interpretation that early post-PMX-HP VIS reduction alone is insufficient to reliably predict survival. By contrast, the absolute VIS3 score demonstrated better discrimination (AUC 0.78, 95% CI 0.66–0.91), with an optimal cut-off of 14.56 (rounded to 14.6) corresponding to sensitivity 0.88 and specificity 0.71. ROC curves for the immediate absolute change in VIS, late percentage VIS change, and the absolute VIS3 score are shown in Fig. 3.

Fig. 3.

ROC curves for immediate VIS change, late percentage VIS change, and absolute VIS3 predicting 28-day mortality. The legend reports AUCs with 95% confidence intervals estimated by DeLong’s method

Longitudinal mixed-effects and time-dependent survival analyses

In the linear mixed-effects model using log(VIS + 1), there were significant overall effects of measurement time (p = 0.016) and 28-day survival status (p = < 0.001), as well as a significant time x survival-status interaction (p = 0.001). This interaction indicates that VIS trajectories differed between survivors and non-survivors rather than merely reflecting a static between-group difference. In the exploratory time-dependent Cox model, higher time-updated log(VIS + 1) was associated with increased 28-day mortality (HR 1.46, 95% CI 1.23–1.74; p = < 0.001). This association remained similar after adjustment for APACHE II (HR 1.48, 95% CI 1.24–1.78; p = < 0.001) (Table 4).

Table 4.

Longitudinal and time-to-event analyses of VIS trajectories

Analysis	Model specification	Estimate	95% CI / df	p-value
Linear mixed-effects model	Overall time effect on log(VIS + 1)	Likelihood-ratio test	df = 2	0.016
Linear mixed-effects model	Overall survival-status effect on log(VIS + 1)	Likelihood-ratio test	df = 1	< 0.001
Linear mixed-effects model	Time x survival-status interaction	Likelihood-ratio test	df = 2	0.001
Time-dependent Cox model	Time-updated log(VIS + 1)	HR 1.46	1.23–1.74	< 0.001
Time-dependent Cox model	Time-updated log(VIS + 1), adjusted for APACHE II	HR 1.48	1.24–1.78	< 0.001

The mixed-effects model used log(VIS + 1) as the dependent variable and included a random intercept for patient. The time-dependent Cox model used start-stop intervals corresponding to VIS1, VIS2, and VIS3 measurement periods. HR, hazard ratio

Incremental prognostic value beyond APACHE II

To evaluate whether VIS3 added prognostic information beyond baseline illness severity, AUCs were compared in complete cases with both APACHE II and VIS3 available (n = 59). APACHE II alone yielded an AUC of 0.63 (95% CI 0.48–0.78). Adding log(VIS3 + 1) to APACHE II increased the AUC to 0.81 (95% CI 0.70–0.92), corresponding to a Delta AUC of + 0.18 (DeLong p = 0.020) (Table 5).

Table 5.

Incremental prognostic value of VIS3 beyond APACHE II for 28-day mortality

Model	Analysis set	AUC (95% CI)	Delta AUC vs. APACHE II	p-value
APACHE II alone	n = 59	0.63 (0.48–0.78)	Reference	-
VIS3 alone	n = 59	0.78 (0.66–0.90)	-	-
APACHE II + log(VIS3 + 1)	n = 59	0.81 (0.70–0.92)	+ 0.18	0.020

The combined model was a logistic model including APACHE II and log(VIS3 + 1). AUCs were compared using DeLong’s test for correlated ROC curves in the same complete-case analysis set

Kaplan-Meier analysis and secondary outcomes

Kaplan-Meier analysis using the ROC-derived VIS3 cut-off showed markedly lower 28-day survival when VIS3 remained > = 14.56 than when VIS3 fell below this threshold (log-rank p = < 0.001) (Fig. 4). The VIS3 < 14.56 group included 27 patients with 3 deaths, whereas the VIS3 > = 14.56 group included 33 patients with 23 deaths.

Fig. 4.

Kaplan-Meier 28-day survival according to the ROC-derived VIS3 cut-off of 14.56. The number-at-risk table is displayed below the survival curves. The cut-off is rounded to 14.6 in the text for clinical readability

Secondary and exploratory outcomes are summarized in Table 6. In response to the potential for survivorship bias, ICU length of stay was summarized only among ICU survivors, and hospital length of stay was summarized only among hospital survivors; no survivor/non-survivor p-values were calculated for duration-based outcomes. CRRT use was more common in non-survivors numerically, but the Fisher exact test did not reach conventional statistical significance. Paired lactate values were available in 39 patients. Exploratory paired lactate change was more favorable in survivors than non-survivors, and the percentage lactate change showed a similar pattern. Mechanical ventilation duration could not be robustly compared because the structured ventilator-duration fields were empty throughout the retrospective dataset.

Table 6.

Secondary and exploratory outcomes

Outcome	Analysis set / all patients	Survivors	Non-survivors	p-value
ICU length of stay, days	ICU survivors only (n = 31): 9.3 (7.2–20.2)	Not compared by 28-day survival	Not compared by 28-day survival	Not applicable
Hospital length of stay, days	Hospital survivors only (n = 30): 27.9 (18.2–51.4)	Not compared by 28-day survival	Not compared by 28-day survival	Not applicable
CRRT use, n (%)	43/64 (67.2%)	19/34 (55.9%)	24/30 (80.0%)	0.061
ECMO use, n (%)	3/64 (4.7%)	1/34 (2.9%)	2/30 (6.7%)	0.596
Paired lactate change (lac2-lac1)	-2.4 (-24.8-15.0)	-5.1 (-33.1–0.5)	3.5 (-10.4-72.4)	0.013
Paired lactate percentage change (%)	-7.0 (-56.1-32.6)	-21.1 (-59.1–3.5)	13.7 (-32.4-82.6)	0.029

Values are presented as median (interquartile range, IQR) or number (%), as appropriate. Length-of-stay analyses were restricted to ICU survivors or hospital survivors to avoid survivorship bias. Paired lactate analyses used available cases (n = 39). Structured ventilator-duration variables were completely missing and therefore are not shown

Discussion

Impact of PMX-HP on hemodynamic stability

In this retrospective single-center ICU cohort, median VIS fell from 28.8 before PMX-HP to 18.1 immediately after treatment, indicating a cohort-level reduction in vasoactive requirement during the early peri-procedural period. This observation is compatible with prior real-world reports suggesting hemodynamic improvement after PMX-HP in selected patients [16, 18, 19]. However, the finding should be interpreted as a descriptive signal rather than proof of causal treatment benefit. PMX-HP was delivered within a broader bundle of sepsis care. Concurrent vasopressor titration strategies, corticosteroid therapy, and perfusion-guided resuscitation may all have influenced short-term vasoactive needs [24–27]. The neutral overall result of the EUPHRATES trial remains an important counterbalance when interpreting early physiologic improvement [17].

Prognostic value of VIS dynamics

VIS maintained prognostic separation across all three time points, with non-survivors exhibiting higher scores before treatment, immediately after treatment, and at approximately 24 h. The mixed-effects model supported this observation statistically by demonstrating significant time, survival-status, and time x survival-status effects. The time-dependent Cox model further strengthened the analysis by treating serial VIS as a time-updated covariate rather than as isolated cross-sectional measurements. Together, these findings support VIS as a pragmatic bedside marker of shock intensity and persistent circulatory failure.

Importantly, the immediate change metric DeltaVIS2-VIS1 remained poorly discriminatory for 28-day survival, which clarifies that an early fall in VIS during the immediate peri-procedural period should not be interpreted as a reliable standalone predictor of survival. By contrast, the late absolute change and, more clearly, the late percentage change in VIS were more favorable among survivors, suggesting that sustained hemodynamic improvement over the ensuing 24 h is more informative than short-interval titration around a single procedure. This pattern is consistent with broader efforts to characterize sepsis recovery trajectories and clinically relevant phenotypes rather than relying on isolated early snapshots [28–30]. Exploratory ROC analysis of absolute VIS3 yielded an optimal cut-off of 14.56, and Kaplan-Meier curves showed marked survival separation across this threshold; however, this threshold was derived from the same cohort and requires external validation before clinical operationalization.

Incremental value beyond baseline severity

A key question is whether VIS3 simply mirrors baseline illness severity or contributes additional prognostic information. In the complete-case incremental analysis, adding log(VIS3 + 1) to APACHE II improved the AUC from 0.63 to 0.81. This suggests that late post-treatment vasoactive dependence may capture dynamic cardiovascular failure not fully represented by admission APACHE II. The result should nevertheless be considered exploratory because the cohort was small, the cut-off was internally derived, and VIS3 was measured after treatment initiation; therefore, VIS3 is best interpreted as a bedside reassessment marker rather than a baseline prognostic score [31].

Secondary outcomes and clinical paradox

Duration-based secondary outcomes require cautious interpretation in septic shock. In the earlier analysis, survivors appeared to have longer ICU and hospital stays than non-survivors, but this pattern is expected when early mortality truncates length of stay. To avoid this survivorship bias, the revised analysis summarizes ICU length of stay only among ICU survivors and hospital length of stay only among hospital survivors, without calculating misleading p-values across 28-day survival groups. Non-survivors more often required mechanical ventilation at baseline and numerically more often required CRRT during the admission, reinforcing the impression of greater multi-organ dysfunction. Exploratory paired lactate analyses moved in the same direction as VIS, with survivors showing greater lactate decline than non-survivors; however, lactate values were not timestamped precisely to the VIS2 and VIS3 windows and should therefore be interpreted cautiously.

Strengths and limitations

This study has several strengths, including consecutive case capture over 6.5 years, predefined VIS measurement windows, longitudinal modeling of serial VIS, time-dependent survival analysis, and evaluation of incremental prognostic value beyond APACHE II. The limitations, however, are substantial. First, the retrospective single-center design precludes causal inference and leaves important residual confounding. Baseline illness severity was high, APACHE II scores tended to be worse in non-survivors, and some outcomes may therefore have been driven by illness burden independent of PMX-HP. Second, VIS reflects the intensity of vasoactive therapy rather than comprehensive cardiovascular physiology; detailed hemodynamic variables such as cardiac output, vascular resistance, echocardiographic findings, and microcirculatory endpoints were not standardized within this real-world retrospective design, so unmeasured circulatory factors may have influenced the observed associations. Third, the number and timing of PMX-HP sessions were determined by the attending intensivist rather than a protocolized algorithm, introducing potential confounding by indication. Fourth, endotoxin activity assays were not available, limiting biological alignment between the intervention and its proposed mechanism. Other adjunctive extracorporeal or pharmacologic strategies in sepsis, including adsorptive CRRT filters and vitamin C-based regimens, have also yielded heterogeneous signals in the literature [32, 33]. Our findings should therefore be interpreted alongside the broader adjunctive-therapy literature in septic shock, where promising physiologic signals have not consistently translated into robust outcome benefit [34, 35]. Finally, although available-case analyses allowed exploratory evaluation of percentage VIS change, paired lactate change, time-dependent survival modeling, and time-to-event visualization, these analyses were constrained by baseline VIS1 = 0 in a subset of patients, incomplete lactate availability, absent structured ventilator-duration data, and imperfect synchronization of laboratory and hemodynamic time points.

Conclusions

In this retrospective single-center Taiwanese ICU cohort of adults with severe sepsis or septic shock treated with PMX-HP, higher serial VIS across peri-procedural time points was associated with non-survival. Although median VIS decreased immediately after PMX-HP, the early reduction alone did not reliably predict 28-day survival. Longitudinal mixed-effects modeling and time-dependent Cox analysis supported the prognostic relevance of dynamic VIS trajectories, and adding VIS3 to APACHE II improved discrimination for 28-day mortality. These findings support VIS as a pragmatic monitoring and risk-stratification tool rather than proof of PMX-HP efficacy. Prospective multicenter studies with protocolized treatment delivery, external validation of VIS3 thresholds, and complementary physiologic measurements are required before clinical operationalization.

Abbreviations

APACHE II

Acute Physiology and Chronic Health Evaluation II

CRRT

Continuous renal replacement therapy

ECMO

Extracorporeal membrane oxygenation

ICU

Intensive care unit

IQR

Interquartile range

PMX-HP

Polymyxin B hemoperfusion

ROC

Receiver operating characteristic

SD

Standard deviation

VIS

Vasoactive-inotropic score

Author contributions

W.-H.C. conceived and designed the study, collected the data, performed the statistical analyses, and drafted the manuscript. W.-J.W. and H.-F.S. contributed to data curation and interpretation. T.-Y.H. supervised the study and critically revised the manuscript for important intellectual content. All authors read and approved the final manuscript.

Funding

This study did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.

Data availability

The data presented in this study are available from the corresponding author upon reasonable request. Due to patient privacy and institutional regulations, the data are not publicly available.

Declarations

Ethics approval and consent to participate

This study was conducted in accordance with the Declaration of Helsinki and was approved by the Institutional Review Board of MacKay Memorial Hospital, Taipei, Taiwan (IRB No. 18MMHIS198e; approval date: 4 March 2019; approved study period: 4 March 2019 to 3 March 2020). The requirement for informed consent was waived due to the retrospective design and the use of de-identified clinical data.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.