We systematically searched the PubMed, Embase, Scopus, and Web of Science databases to identify relevant articles that (1) compared the HPI with other intraoperative monitoring methods, (2) were published since January 2019, and (3) were either randomized controlled trials (RCTs) or retrospective studies. The following keywords were used: “intraoperative hypotension”, “perioperative hypotension”, “surgery”, “hypotension prediction index”, “HPI”, and “early warning system”. No language restrictions were imposed. The final search was completed in February 2025.

Only studies that met the following criteria were included: (1) clinical studies, (2) patients who underwent surgery with HPI-guided management, (3) standardized protocols for intraoperative hemodynamic management, (4) parallel control groups using other monitoring strategies, and (5) at least one of the following outcomes: postoperative complications or intraoperative hypotension.

Two authors (Y L and B L) primarily reviewed the articles to exclude irrelevant studies according to the inclusion criteria. Any disagreement at any stage was discussed with a third author (W X).

The American Society of Anesthesiologists physical status scores, ages, types of surgery, and study types were extracted from the original articles. The following data were collected for analysis: (1) major complications (cardiac, neurological, renal injury, and mortality), (2) time-weighted average (TWA) of hypotension, (3) total duration of hypotension, (4) area under the hypotensive threshold (AUT), (5) intraoperative blood loss, and (6) length of hospital stay.

The methodological quality and risk of bias of the included RCTs were assessed by two authors (Y L and KX Y) using the Cochrane risk of bias tool (RoB), which consisted of the following seven aspects: random sequence generation, allocation concealment, blinding of participants and personnel, blinding of outcome assessment, incomplete outcome data, selective outcome reporting, and other sources of bias. The revised Cochrane Risk of Bias tool (RoB 2) was used to further judge the risk of bias, which consisted of the following five distinct domains: randomisation process, deviations from the intended interventions, missing outcome data, measurement of the outcome and selection of the reported result.

Non-randomized trials were assessed using the Newcastle-Ottawa Scale (NOS). The highest possible score on the NOS is a 9, and studies with scores ≥ 5 were considered adequate for further analysis (https://www.ohri.ca//programs/clinical_epidemiology/oxford.asp).

Statistical analyses were performed using Review Manager software (RevMan 5.4; the Cochrane Collaboration) or R software (version 4.1.3, The R Foundation), as appropriate.

When the original data were expressed as continuous variables, the meta-analysis was performed using mean differences (MDs). When data were reported as medians and interquartile ranges (IQRs), authors or statisticians of the articles were contacted to obtain means and standard deviations [16–18]. Dichotomous variables were determined using risk ratios (RRs) calculated using the Mantel-Haenszel method.

Major complications and intraoperative hypotension were analyzed through a trial sequential analysis (TSA; version 0.9 Beta, Copenhagen Trial Unit) to quantify the statistical reliability of the included data in this meta-analysis. To minimize the heterogeneity among different designed researches and avoid a higher risk of bias, only RCTs were included in the TSA analysis. Specifically, a TSA was conducted to estimate the required information size and assess the risk of type I and II errors [19]. The TSA method is used in meta-analyses to control for the risk of random errors resulting from repetitive testing. This helps determine whether the available evidence is sufficient to draw reliable conclusions or if more studies are needed. The TSA plot showed that the conventional Z-value (Z=1.96) corresponded to the standard significance level (P < 0.05) of a two-sided test. If the cumulative Z-curve crosses this line, this suggests statistical significance in conventional meta-analyses. The TSA monitory boundaries (TSA Z-values) are more stringent thresholds adjusted for the risk of random errors. Crossing these boundaries provides firm evidence that accounts for repeated testing and data sparsity. The cumulative Z-curve plots the Z-values from the meta-analyses as studies are sequentially added in chronological order.

When the cumulative Z-curve of the results exceeds the TSA boundary, a sufficient level of evidence for the anticipated intervention effect has likely been reached at the current condition and same population. However, if the Z-curve does not exceed the TSA boundary, the amount of information is insufficient to reach a firm conclusion. Two-sided tests with a type I error of 5% and a power of 80% were used.

Study heterogeneity was determined by estimating the I² statistic, which describes the percentage of total variation across studies due to heterogeneity as opposed to chance. Outcome data with low heterogeneity (I² ≤ 50%) were analyzed using a fixed-effects model. For outcome data with high heterogeneity (I² > 50%), a random-effects model was used.

Publication bias was evaluated visually using a funnel plot.

We searched the PubMed, Scopus, Embase, and Web of Science databases. A total of 559 investigations were retrieved. After removing duplicates, 236 articles remained. An additional 17 reviews and meta-analyses, six case reports, nine protocols, and 137 articles without surgery or an HPI apparatus were excluded after screening the titles and abstracts. Sixty-seven full-text articles were thus retrieved.

After a detailed review, we excluded 27 reports with a lack of clinical outcomes and effective control groups, 17 articles that were not research articles, and two that were letters to the editor. Two additional studies were excluded because they did not focus on intraoperative observations. A review of the relevant literature revealed an additional article conducted by Lorente et al. [20]; however, the study population overlapped with that of Ripollés-Melchor et al. [14] and was thus excluded. A total of 19 distinct articles (12 RCTs and 7 retrospective studies) with a total of 2570 recruited patients were thus selected for final analysis in the current study (Fig. 1) [14,21–38].

The key characteristics of the included studies are summarized in Table 1. We compared HPI-guided management with other monitoring strategies. For the control groups, 17 studies defined standard or conventional care, while the remaining two conducted goal-directed fluid therapy (GDFT) and a predefined goal-directed therapy (GDT) strategy. Schneck et al. [28] and de Keijzer et al. [29] conducted three-arm studies, and data from the HPI group and routine anesthetic care were collected for analysis.

A critical evaluation of study quality was performed by two reviewers (Y L and W X), and any disagreement was discussed with another co-author (B L). The risk of bias is shown in Supplementary Figs. 1 and 2 and in Table 1.

As the HPI was used to guide management, the investigators and anesthesiologists could not be blinded during the study; however, the patients participating in the study were blinded. As the HPI was used to guide management, the investigators and anesthesiologists could not be blinded during the study; however, the patients participating in the study and analyzers were blinded. Since the primary outcomes were hypotension and complications, which were unlikely to be influenced by observation bias, we considered the risk of blinding in the included RCTs to be unclear risk of bias. The included RCTs showed minimal risk of bias due to their careful design and execution, as noted in previous systematic reviews. Overall, the risk of bias in these included RCTs was judged as low. RoB2 assessment also showed unclear risk for the blinding related domains and low risk for the most of remaining domains.

Among the non-randomized trials, one was rated 7, two were rated 8, and the remaining four were rated 9 on the NOS scale, indicating high quality and low risk of bias.

The primary outcome for this review was the incidence of major complications, including cardiac, neurological, renal injury, and mortality. Fifteen articles reported complications (Fig. 2A). As no complications were observed in the study conducted by Yoshikawa et al. [34], only the complications in the remaining 14 studies were analyzed. Andrzejewska et al. [31] reported a case of transient limb paresis in a patient who underwent posterior spinal fusion for adolescent idiopathic scoliosis. Given that this patient underwent this procedure at such a young age, transient limb paresis was more likely attributable to the surgical procedure itself than intraoperative hypotension. A total of 265 events in the HPI group and 329 events in the control group were reported. The pooled relative risk (RR) was 0.79 (95% CI [0.69–0.90], I² = 0, P < 0.001). This indicates that the HPI group experienced a 21% reduction in the RR of postoperative complications compared with the control group.

We then performed a TSA for major complications using the included RCTs (Fig. 2B). According to the results, the Z-curve crossed both the conventional and TSA Z-curves before the estimated pooled sample size (2088). It indicated that the pooled complications in the HPI group were different from the control group, and the result was reliable at the current design in the same population.

To elucidate the effects of HPI-guided management on surgical procedures, we conducted a subgroup analysis to mitigate bias due to the study design. The results for RCTs (RR = 0.83, 95% CI [0.72–0.96], I² = 0, P = 0.01) and non-RCTs (RR = 0.61, 95% CI [0.42–0.89], I² = 0, P = 0.01) were presented. A robust reduction in complications was found in the HPI group. This indicates that HPI-guided management effectively reduces complications during the perioperative period, regardless of the trial design.

Subgroup analyses were performed to investigate the specific clinical impact of HPI-guided management on each complication (cardiac, neurological, renal injury, and mortality) (Supplementary Fig. 4). Cardiac complications were significantly decreased in the HPI group (RR = 0.68, 95% CI [0.50–0.92], I² = 0, P = 0.01). Additionally, neurological (RR = 0.72, 95% CI [0.42–1.25], I² = 0, P = 0.24) and renal (RR = 0.85, 95% CI [0.70–1.01], I² = 0, P = 0.07) complications showed decreasing trends in the HPI group. No statistically significant difference in mortality was found between the groups under the current conditions (RR = 0.88, 95% CI [0.51–1.52], I² = 0, P = 0.64).