Efficacy and Safety of Methylene Blue in Septic Shock: A Systematic Review and Meta-Analysis of Randomized Controlled Trials

Introduction

Septic shock is characterized by profound vasodilation and refractory hypotension that persists despite adequate fluid resuscitation and vasopressor therapy, and remains a leading cause of morbidity and mortality in intensive care units worldwide. The pathophysiology of vasoplegia in septic shock is driven substantially by excessive nitric oxide (NO) production, mediated by upregulation of inducible nitric oxide synthase (iNOS) and downstream activation of soluble guanylate cyclase (sGC), resulting in pathological vascular smooth muscle relaxation.

Methylene blue (MB) is a NO-pathway inhibitor that acts through direct inhibition of iNOS and sGC, restoring vascular tone in vasoplegic states and providing a mechanism of blood-pressure support distinct from conventional catecholamine vasopressors [1]. MB has an established evidence base in post-cardiopulmonary bypass vasoplegia and in calcium-channel-blocker or beta-blocker poisoning, and has more recently been studied as an adjunctive, catecholamine-sparing agent in septic shock. Proposed benefits include reduced vasopressor requirement, improved mean arterial pressure (MAP), and potentially shorter durations of organ support; proposed harms include methemoglobinemia, serotonin syndrome (in combination with serotonergic agents), and reduced mesenteric perfusion at high doses.

The randomized evidence base for MB in septic shock consists of a small number of modestly sized trials. Several prior systematic reviews and meta-analyses have synthesized this literature with varying inclusion criteria and search dates — including Fernando et al. (2024), which pooled six trials (302 patients) and reported low-certainty evidence of a mortality benefit; Ballarin et al. (2024), which pooled three trials (141 patients) and reported reductions in vasopressor discontinuation time, ICU stay, and mechanical ventilation duration; and Ng et al. (2025), which pooled five trials (257 patients) focused on the hemodynamic effect of MB [2–4]. At least one additional randomized trial (Shaker et al., 2025) has since been published, comparing low- and high-dose MB regimens against placebo [5]. We conducted this systematic review and meta-analysis to synthesize the currently available randomized evidence, incorporating this more recent trial, and to characterize the effect of MB on mortality, hemodynamic, and organ-support outcomes in adult septic shock.

Methods

Search Strategy

We searched PubMed, the Cochrane Central Register of Controlled Trials (CENTRAL), and ClinicalTrials.gov from inception to the search date. The search combined terms for methylene blue (“methylene blue” OR “methylthioninium chloride”) with terms for septic shock (“septic shock” OR sepsis OR “vasodilatory shock” OR “systemic inflammatory response syndrome”) and terms restricting to randomized trial design (randomized OR randomised OR RCT OR trial). The search returned 18 records from PubMed, 62 from Cochrane CENTRAL, and 22 from ClinicalTrials.gov, for a combined 102 records before deduplication (Figure 1).

Figure 1: PRISMA 2020 flow diagram of study identification, screening, and inclusion.

Eligibility Criteria

We included randomized controlled trials enrolling adults (≥18 years) with septic shock, comparing intravenous MB at any dose, route, or timing to placebo or usual care without MB. We excluded non-randomized studies, studies without a comparator arm, pediatric or neonatal populations, and studies of vasoplegia from a non-septic cause (e.g., post-cardiopulmonary bypass, drug-induced vasoplegia).

Study Selection

After removal of duplicates, titles and abstracts were screened against the eligibility criteria above. Five trials had extractable outcome data meeting full eligibility criteria and were included in this synthesis: Ibarra-Estrada et al. (2023), Shaker et al. (2025), Memis et al. (2002), Kirov et al. (2001), and Dong et al. (2025) [5-8,10]. Two further records identified during screening were excluded on full-text review: Preiser (1995), which lacked a comparator group, and a trial evaluating combined MB and vasopressin rather than a pairwise MB-versus-placebo/usual-care comparison. One additional record is a registered trial protocol without posted results and was excluded on that basis.

Data Extraction

For each included trial we extracted study design, country, population and eligibility criteria, MB dose/route/timing, comparator, sample size, age, sex distribution, and the sepsis definition used. For binary outcomes (short-term mortality), we extracted the number of events and total patients per arm. For continuous outcomes (MAP at 6 hours, duration of vasopressor exposure, ICU and hospital length of stay, duration of invasive mechanical ventilation, and methemoglobin concentration), we extracted the mean and standard deviation per arm. Short-term mortality was defined as the longest mortality outcome reported up to 60 days, or in-hospital mortality, consistent with prior systematic reviews of this literature.

Dong et al. (2025) [10] was included in the qualitative synthesis and study characteristics table below but did not contribute to any of the pooled quantitative outcomes in this analysis; that trial’s primary endpoint was a sublingual microcirculatory parameter not captured by our prespecified outcome list, and its published abstract reports that MB improved microcirculatory parameters without a corresponding significant difference in organ function or survival outcomes, though per-arm event data were not extracted for pooling here.

Statistical Analysis

For binary outcomes, we calculated the risk ratio (RR) and 95% confidence interval (CI) for each trial using the number of events and total patients per arm, applying a continuity correction of 0.5 to any trial with a zero-event arm. For continuous outcomes, we calculated the mean difference (MD) and 95% CI from the perarm mean, standard deviation, and sample size. We pooled effect estimates across trials using an inverse-variance, random-effects model with the DerSimonian–Laird estimator for between-study variance (τ²), given the expected clinical and methodological heterogeneity across trials (differing MB dosing regimens, sepsis definitions, and enrollment eras). Heterogeneity was assessed using the I² statistic. For every outcome pooled across three or more trials, we performed a leave-one-out sensitivity analysis, sequentially excluding each trial and re-pooling the remainder, to identify results that were disproportionately influenced by a single study. Outcomes reported by only a single trial are presented as that trial’s point estimate and 95% CI without pooling; no I² or between-study heterogeneity statistic applies to these. Given the small number of trials per outcome (≤ 4), funnel plots and tests for small-study effects (e.g., Egger’s test) were not performed, consistent with current guidance that such tests are underpowered and unreliable with fewer than 10 studies.

Risk of Bias Assessment

Risk of bias was assessed independently by two reviewers using the revised Cochrane risk-of-bias tool for randomized trials (RoB 2), which evaluates five domains: (D1) bias arising from the randomization process, (D2) bias due to deviations from intended interventions, (D3) bias due to missing outcome data, (D4) bias in measurement of the outcome, and (D5) bias in selection of the reported result. Each domain was judged as low risk, some concerns, or high risk, and an overall risk-of-bias rating was derived per study. Discrepancies between reviewers were resolved by discussion and consensus.

Because RoB 2 is outcome-specific — in particular, Domains 3–5 can differ depending on which outcome is being assessed within the same trial — the assessment was performed separately for each of the seven prespecified outcomes, restricted in each case to the trials that actually contributed quantitative data to that outcome. Dong et al. (2025) did not contribute to any pooled quantitative outcome (Section 2.4) and was therefore not included in this riskof-bias assessment.

Results

Study Selection

Following a systematic search of PubMed, Cochrane CENTRAL, and ClinicalTrials.gov and title/abstract screening, five randomized trials with extractable outcome data met inclusion criteria and were included in this synthesis (Section 2.3), enrolling a total of 303 patients.

Study Characteristics

Table 1 summarizes the characteristics of the five included trials. Trial size ranged from 20 to 91 patients (median 72). Three trials enrolled patients meeting Sepsis-3 criteria (Ibarra-Estrada 2023, Shaker 2025, Dong 2025); Kirov (2001) used SIRS-based criteria (≥2 SIRS criteria and MAP <70 mmHg despite fluids/vasopressors), predating the Sepsis-3 consensus definition; the sepsis definition used by Memis (2002) was not specified in the extraction. MB dosing regimens varied substantially, from a single bolus to multiday continuous infusions (Ibarra-Estrada: 100 mg over 6 hours daily for 3 days), reflecting the absence of a standardized dosing protocol across this literature.

Table 1: summarizes the characteristics of the five included trials.

Short-Term Mortality

Four trials, conducted in Mexico, Egypt, Turkey, and Norway (Ibarra-Estrada 2023, Shaker 2025, Memis 2002, Kirov 2001), reported short-term mortality data, together enrolling 201 patients: 100 in the MB arm and 101 in the control arm. Pooled using a random-effects model, MB was associated with a directionally favorable but not statistically significant reduction in short-term mortality compared with placebo/usual care (RR 0.69, 95% CI 0.44–1.10; p = 0.12; I² = 0%, τ² = 0) (Figure 2). The absence of statistical heterogeneity (I² = 0%) indicates the four point estimates are reasonably consistent with one another despite differing MB regimens; however, the wide confidence interval reflects the small total number of events and patients pooled, and the absence of statistical significance should not be interpreted as proof that MB has no effect on mortality — the available trials are simply too small to estimate this outcome precisely.

Figure 2: Forest plot of short-term mortality, methylene blue versus control.

A leave-one-out sensitivity analysis (Table 2) showed the pooled estimate remained directionally consistent and non-significant regardless of which single trial was excluded, with heterogeneity remaining low (I² ≤ 30%) throughout.

Table 2: Leave-one-out sensitivity analysis for short-term mortality.

Mean Arterial Pressure at 6 Hours

Two trials (Kirov 2001, Memis 2002; n = 50) reported MAP at 6 hours post-administration. MB was associated with a statistically significant increase in MAP compared with control (MD +15.5 mmHg, 95% CI 8.0–23.0; p < 0.0001; I² = 0%) (Figure 3). Neither Ibarra-Estrada (2023) nor Shaker (2025) reported this specific timepoint in the current extraction. Because only two trials contributed, no leave-one-out analysis was performed for this outcome.

Figure 3: Forest plot of mean arterial pressure at 6 hours, methylene blue versus control.

Duration of Vasopressor Exposure

Three trials (Ibarra-Estrada 2023, Shaker 2025, Kirov 2001; n = 171) reported duration of vasopressor exposure, converted to hours. The pooled random-effects estimate favored MB but did not reach statistical significance (MD −16.3 hours, 95% CI −43.4 to 10.7; p = 0.24), with substantial statistical heterogeneity (I² = 86%) (Figure 4).

Figure 4: Forest plot of duration of vasopressor exposure, methylene blue versus control.

A leave-one-out sensitivity analysis (Table 3) shows this pooled result is highly sensitive to which trial is included. Excluding Shaker (2025) — whose own point estimate for this outcome was essentially null (MD +2.0 hours, not favoring MB) — yields a statistically significant reduction of 31.1 hours with no residual heterogeneity (I² = 0%), driven by the concordant estimates from Ibarra-Estrada and Kirov. Excluding Ibarra-Estrada (2023) instead sharply attenuates the pooled effect toward the null. Excluding Kirov (2001) worsens heterogeneity further (I² = 92.6%). This pattern indicates that the overall null, heterogeneous result is largely attributable to one trial (Shaker 2025) diverging from the other two, rather than reflecting a consistent absence of effect across the literature.

Table 3: Leave-one-out sensitivity analysis for duration of vasopressor exposure.

Duration of Invasive Mechanical Ventilation

Three trials (Ibarra-Estrada 2023, Shaker 2025, Memis 2002; n = 181) reported duration of invasive mechanical ventilation. MB was associated with a statistically significant reduction (MD −0.82 days, 95% CI −1.38 to −0.26; p = 0.004; I² = 0%) (Figure 5).

Figure 5: Forest plot of duration of invasive mechanical ventilation, methylene blue versus control.

Unlike the vasopressor-duration outcome, this result was robust on leave-one-out sensitivity analysis (Table 4): the pooled estimate remained statistically significant regardless of which single trial was excluded, with no heterogeneity introduced in any scenario (I² = 0% throughout).

Table 4: Leave-one-out sensitivity analysis for duration of invasive mechanical ventilation.

ICU and Hospital Length of Stay

ICU length of stay was reported by a single trial (Shaker 2025; n = 60): MD −4.0 days (95% CI −11.4 to 3.4), not statistically significant. Hospital length of stay was likewise derived from a single usable trial (Shaker 2025; n = 60): MD −14.0 days (95% CI −25.5 to −2.5). Memis (2002) also reported hospital length of stay, but the extracted values for this outcome were incomplete in the underlying data and could not be included; this should be corrected before the manuscript is finalized. Because each of these outcomes currently rests on a single trial, no pooled estimate, forest plot, or leave-one-out analysis is presented, and these findings should be considered exploratory.

Methemoglobin Concentration

Methemoglobin concentration was reported by a single trial (Memis 2002; n = 30): MD +0.16% (95% CI 0.159–0.161%). We flag this result as unreliable: the underlying per-arm standard deviations are implausibly small relative to the reported means and are very likely a data-entry or unit error rather than a true reflection of the original trial’s precision. We recommend this value be re-verified against the original publication before being reported in any final version of this manuscript. Because methemoglobin could not be quantitatively pooled across trials, the absence of a reliable pooled safety signal here should not be interpreted as evidence that MB is free of this risk; larger trials should systematically report methemoglobin concentrations and other adverse events.

Summary of Pooled Estimates

Table 5: Summary of pooled and single-trial estimates across all outcomes.

Risk of Bias Assessment

RoB 2 was assessed separately for each of the seven prespecified outcomes, restricted to the trials contributing quantitative data to that outcome (Section 2.6). Judgments and the corresponding summary figures are presented below for each outcome, followed by a crossoutcome synthesis and a supplementary study-by-outcome overview. Short-Term Mortality

Table 6: RoB 2 domain judgments for short-term mortality.

Figure 6: RoB 2 summary for short-term mortality.

Three trials were judged as having some concerns overall and Kirov (2001) was judged at high risk, driven primarily by its open-label design (Domain 2). The pooled estimate numerically favored methylene blue, but the confidence interval crossed the null value. Mean Arterial Pressure at 6 Hours

Table 7: RoB 2 domain judgments for mean arterial pressure at 6 hours.

Figure 7: RoB 2 summary for mean arterial pressure at 6 hours.

Kirov (2001) carries the main methodological concern for this outcome, since its open-label design is more consequential for a physiologic, treatment-responsive outcome such as MAP than for a harder endpoint such as mortality.

Duration of Vasopressor Exposure

Table 8: RoB 2 domain judgments for duration of vasopressor exposure.

Figure 8: RoB 2 summary for duration of vasopressor exposure.

The high statistical heterogeneity observed for this outcome (Section 3.5) should be considered separately from RoB 2 — the two are distinct forms of uncertainty. Kirov (2001) remains the highest-risk contributing trial, while Ibarra-Estrada (2023) and Shaker (2025) were judged as having some concerns.

Duration of Invasive Mechanical Ventilation

Table 9: RoB 2 domain judgments for duration of invasive mechanical ventilation.

Figure 9: RoB 2 summary for duration of invasive mechanical ventilation.

All three contributing trials were judged as having some concerns overall. The pooled effect for this outcome is statistically significant with no heterogeneity (Section 3.6), but consistency of effect does not by itself lower a risk-of-bias judgment — the two are assessed independently.

ICU Length of Stay

Table 10: RoB 2 domain judgments for ICU length of stay.

Figure 10: RoB 2 summary for ICU length of stay.

Only Shaker (2025) contributed study-level data for this outcome among the five included trials; the assessment is therefore limited to this single trial and does not infer judgments for non-contributing studies.

Hospital Length of Stay

Table 11: RoB 2 domain judgments for hospital length of stay.

Figure 11: RoB 2 summary for hospital length of stay.

As with ICU length of stay, only Shaker (2025) contributed usable data for this outcome; Memis (2002) also reported hospital length of stay but was excluded from quantitative synthesis due to incomplete extracted data (Section 3.7), and is therefore not included in this risk-of-bias display either.

Methemoglobin Concentration

Table 12: RoB 2 domain judgments for methemoglobin concentration.

Figure 12: RoB 2 summary for methemoglobin concentration.

Only Memis (2002) contributed data for this outcome. Note the separate data-quality concern raised in Section 3.8 regarding this trial’s reported variance for methemoglobin, which is independent of — and not captured by — its RoB 2 judgment.

Cross-Outcome Summary

Across these outcome-specific assessments, Domain 5 (selection of the reported result) was the most recurrent concern, appearing as “some concerns” for nearly every study-outcome combination — consistent with the fact that none of these trials, several conducted before prospective trial registration was standard practice, had a fully verifiable pre-specified analysis plan. Kirov (2001) was the most methodologically problematic trial where it contributed, driven specifically by its open-label design (Domain 2), which is most consequential for the physiologic and treatment-responsive outcomes (MAP, vasopressor duration) it informs. Statistical significance and heterogeneity are not themselves RoB 2 domains and were not used to upgrade or downgrade any domain judgment.

Figure 13 summarizes the overall RoB 2 judgment for each study against each outcome it contributed to, in a single view.

Figure 13: RoB 2 overall judgment by study and outcome (supplementary traffic-light summary).