Study design

UR of systematic reviews, meta-analyses (MA), and (when applicable) network meta-analyses (NMA) on IV vitamin C-based regimens in adults with sepsis or septic shock. This UR was developed following the PRIOR (Preferred Reporting Items for Overviews of Reviews) guidelines for umbrella reviews [21] (see S1 Table), and additionally the methodological framework proposed by Aromataris et al. [22] for this type of study. A completed PRISMA 2020 checklist is provided as Supporting Information to comply with reporting requirements for systematic reviews and evidence syntheses (see S1 Checklist PRISMA)

Search strategy

Structured searches were conducted in PubMed/MEDLINE, Scopus, Web of Science (including all collections), and EMBASE from the inception of each database until the project cutoff date. Search terms (combined with Boolean operators and MeSH/Emtree equivalents when applicable) included, as a guide: sepsis, septic shock, vitamin C, ascorbic acid, hydrocortisone, thiamine, HAT, randomized, systematic review, meta-analysis, network. No language restrictions were applied (See S2 Table).

Selection criteria

We included: (1) systematic reviews (SR) with or without meta-analyses (MA) and network meta-analyses (NMA) evaluating intravenous (IV) vitamin C as monotherapy, HAT (combination of hydrocortisone + vitamin C + thiamine), or vitamin C + thiamine in adults (≥18 years) with sepsis or septic shock; (2) studies reporting at least one outcome of interest: 28–30-day mortality (primary), ΔSOFA at 72–96 h (change in Sequential Organ Failure Assessment score), duration of vasopressors, ICU length of stay (ICU-LOS), hospital length of stay (hospital-LOS), acute kidney injury (AKI)/renal replacement therapy (RRT), and adverse events; (3) articles with clear design, explicit inclusion criteria, and reproducible literature search.

We excluded: (1) narrative reviews without reproducible criteria; (2) studies in pediatric or non-adult populations; (3) non-IV vitamin C interventions or combinations not aligned with the focus (e.g., nutraceutical cocktails where the specific effect of vitamin C cannot be separated).

Selection process and data extraction

Three investigators (VJVP, JJBC, LAMVS) independently performed screening and extraction in duplicate using Rayyan (review screening platform from QCRI). In the first phase (title/abstract), clearly irrelevant records were excluded, and the full-text review applied the inclusion/exclusion criteria. Discrepancies were resolved by consensus or with adjudication by a fourth reviewer (FEZM).

From each review, we extracted: study type (SR/MA/NMA), population, databases and search period, number of randomized controlled trials (RCTs) and non-RCTs, total sample size, intervention (dose, timing, and duration), comparators, reported outcomes with their estimators and precision; registration (e.g., PROSPERO), risk of bias (RoB) tool used for primary studies, publication bias assessment (funnel/Egger), and synthesis methods (random or fixed-effects model; NMA; trial sequential analysis—TSA).

Table construction

To ensure a homogeneous and auditable recording of the data, a data dictionary and a single template were developed, structured into three main blocks. The first block, “Identification and scope,” was used to record the fundamental information of each study, including a PDF number assigned for initial order and traceability, the reference in Vancouver format, the type of treatment studied (IV vitamin C monotherapy, HAT, or vitamin C + thiamine), as well as the population and clinical setting.

The second block compiled the “Methodological metadata” of each review. This section documented the databases searched, the search period, the designs of the included studies, the number of randomized controlled trials (RCTs) and non-RCTs, the total sample size, the comparators used, and the statistical synthesis model (random/fixed effects or network meta-analysis). Additionally, it was recorded whether a Trial Sequential Analysis (TSA) or a CINeMA (Confidence in Network Meta-Analysis) assessment had been performed, as well as the protocol registration information (e.g., in PROSPERO).

Finally, the third block focused on the “Quantitative results” for key outcomes: 28–30-day mortality, change in SOFA score (ΔSOFA) at 72–96 hours, and duration of vasopressor use. For each of these outcomes, the effect measure was recorded, such as risk ratio (RR), odds ratio (OR), mean difference (MD), or standardized mean difference (SMD), along with its 95% confidence interval (95% CI) and the heterogeneity statistic (I²), also specifying the model used in each review

When explicitly reported by the review, we added operational columns: initiation ≤24 h, duration 3–4 days, dose category 25–100 mg/kg/day vs > 100 mg/kg/day, and sepsis vs shock.

Extraction was performed in duplicate by three investigators (VJVP, JJBC and LAMVS) independently, after a pilot calibration phase (5 reviews), and any discrepancy was resolved by a fourth investigator (FEZM). The process was managed in Rayyan for screening and in a master sheet (auditable format) for data capture; any subsequent corrections left a trace (timestamp, author of change, and reason). When a review presented multiple time horizons (e.g., 28 d, in-hospital, ICU), 28–30 days was prioritized; when multiple models existed (fixed/random), the random-effects declared as primary were extracted. In network/component meta-analyses, we recorded the estimator for the relevant comparison (when available) and network parameters (e.g., consistency/coherence).

Risk of bias analysis

Methodological quality of each review was assessed with AMSTAR-2 (items 1–16), with special emphasis on critical items: prior protocol (2), list of excluded studies with justification (7), RoB assessment of primary studies (9), meta-analysis methods (11), consideration of RoB in interpretation (13), and publication bias (15) [23]. Three investigators applied AMSTAR-2 independently (VJVP, JJBC and LAMVS) and a fourth resolved disagreements (FEZM). We recorded item-by-item ratings (Yes/Partial/No/Not reported/N/A) and the overall synthesis (High/Moderate/Low/Critically low), which was compiled in S3 Table (with color codes for critical items).

As a complement, we documented how each review assessed RoB of RCTs (e.g., Cochrane RoB 2, Jadad), handling of publication bias (funnel, Egger), treatment of heterogeneity (I², sensitivity analyses), and when present, application of TSA or CINeMA assessment in NMA.

Certainty of evidence (GRADE)

To rate certainty with GRADE [24], we adopted an “anchor estimator” strategy when multiple reviews reported the same outcome with similar effect measures. Rather than performing a “meta-meta-analysis” or averaging results, we selected a single reference estimator for each critical outcome (28–30-day mortality) and therapeutic regimen (monotherapy, HAT, C + T). This anchor estimator consisted of a single RR/OR/MD/SMD with 95% CI (and I² when applicable) extracted from the most methodologically robust review, prioritizing high AMSTAR-2 quality, strict PICO alignment (28–30-day horizons over alternatives), random-effects model, and TSA availability.

The remaining reviews from the same silo were not statistically combined but served complementary functions to validate consistency of findings, explore subgroup analyses by dose (25–100 mg/kg/day vs > 100 mg/kg/day), timing of initiation (≤24h), duration (3–4 days) and population (sepsis vs septic shock), and perform sensitivity analyses to assess result robustness. GRADE application started from high certainty given these were RCTs synthesized in systematic reviews, with systematic downgrading across five domains: risk of bias (quality of individual RCTs and meta-analysis), inconsistency (I², directionality of effects, and confidence interval overlap), indirectness (inappropriate mixing of populations without stratification), imprecision (wide intervals crossing the no-effect threshold, optimal information size not reached, supported by TSA when available), and publication bias (funnel plot asymmetry or positive Egger test).

Importantly, the GRADE process was executed in duplicate by three investigators (VJVP, JJBC and LAMVS) with a fourth resolving disagreements (FEZM). The final presentation was delivered through SoF tables, separated by regimen (monotherapy, HAT, C + T), and included an anchor estimator (95% CI and I²), a final rating (High/Moderate/Low/Very low), and justification for each adjustment. Critical decisions (e.g., anchor estimator selection when multiple high-quality SRs existed) were recorded in a log with links to the corresponding figure/page in the PDF for peer review traceability.

A sensitivity analysis evaluating the stability of GRADE conclusions when the primary anchor is replaced by the second- and third-ranked reviews by AMSTAR-2 score is provided in Supplementary Material (S4 Table). The direction and certainty category of all conclusions remained unchanged across substitutions.

Overlap analysis

To avoid double counting and identify redundancy, we quantified RCT overlap between reviews for each “silo” (Vitamin C monotherapy; HAT; Vitamin C+thiamine) using the Corrected Covered Area (CCA) by Pieper et al. [25]. Primary articles were extracted by three investigators (VJVP, JJBC and LAMVS), with discrepancies resolved among themselves.

For each silo, unique RCTs and their appearances in each SR/MA were listed; we applied:

where No = total number of RCT appearances; Ns = unique RCTs; r = number of reviews; c = unique RCTs. Interpretation was: 0–5% slight, 6–10% moderate, 11–15% high, > 15% very high.

Statistical analysis

No new meta-analyses or meta-regressions were performed. Quantitative conclusions are based on estimators and TSA reported by included reviews. The objective was descriptive-critical, prioritizing quality (AMSTAR-2), certainty by outcome (GRADE), and consistency among reviews with very high CCA.

Study selection

The search in the four aforementioned databases retrieved 724 records with no additional sources identified through other means. After deduplication, 488 references were retained and screened by title and abstract; of these, 435 were excluded due to clear lack of relevance (e.g., non-adult populations, non-intravenous interventions, primary studies or narratives). Full-text assessment was performed on 53 articles, with 22 excluded for not meeting inclusion criteria (mainly: narrative review without reproducible methods, pediatric population, non-aligned intervention; for example, nutraceutical cocktails without separability of vitamin C effect, or absence of outcomes of interest). Finally, 31 reviews were selected [5–12,14,15,19,20, 26–44]. Fig 1 shows the PRISMA 2020 flow diagram of the study selection process. (Fig 1)

Download:

• PNG
  larger image
• TIFF
  original image

Fig 1. PRISMA 2020 flow diagram of the study selection process.

https://doi.org/10.1371/journal.pone.0351072.g001

Main characteristics of studies

Thirty-one reviews on intravenous vitamin C-based regimens in adults with sepsis or septic shock were included: 30 quantitative (28 systematic reviews with traditional meta-analyses and 2 network/component meta-analyses) and 1 qualitative (for methodological support) (See Table 1 and S5 Table). Within the quantitative corpus, we identified three analytical sets (“silos”): IV vitamin C monotherapy, HAT (hydrocortisone + vitamin C + thiamine), and vitamin C + thiamine without steroid. The body of trials feeding each silo is compact: 10 unique RCTs support HAT (VITAMINS, ACTS, VICTAS, ORANGES, Hwang, Chang, Mohamed, Lyu, Hussein, Jamshidi) [10–12,35], 10 unique RCTs support monotherapy (e.g., Zabet 2016; Fowler 2019; Wacker 2022; Rosengrave 2022; Lv 2021; El-Driny 2022) [5–7,37,40], and 7 unique RCTs support vitamin C + thiamine [13,41]. In monotherapy, 28–30-day mortality was the most frequent primary outcome; in HAT and vitamin C + thiamine, ΔSOFA at 72–96 h was often treated as primary and mortality as secondary [10–13]. Seven reviews employed trial sequential analysis (TSA) to anchor interpretation [6,10–13,27,29] and two used component analysis to disaggregate the effect of corticosteroid, vitamin C, and thiamine [20,34]. Comparators were predominantly placebo/standard care; in HAT, several RCTs contrasted against hydrocortisone alone (e.g., VITAMINS), relevant data for interpreting the absence of survival effect [10,34].

Download:

• PNG
  larger image
• TIFF
  original image

Table 1. Summary of included systematic reviews grouped by regimen and primary outcome focus.

https://doi.org/10.1371/journal.pone.0351072.t001

Objectives and doses used

Reviews converged on three clinical questions: whether IV vitamin C monotherapy reduces 28–30-day mortality and improves supportive outcomes; whether combinations (HAT and vitamin C + thiamine without steroid) add benefit over standard care; and which administration conditions (timing of initiation, duration, severity) modulate the effect [5–7,10–13,37]. In parallel, component meta-analyses examined which drug explains hemodynamic changes in combinations, consistently pointing to the glucocorticoid as the main determinant of vasopressor shortening and ICU length reduction, without clear translation into survival when adding vitamin C or thiamine [20,34].

Monotherapy was most frequently administered as 1.5 g every 6 hours (equivalent to 25–100 mg/kg/day) for 3–4 days, with initiation within 24 hours of sepsis recognition; under these conditions, signals of short-term mortality reduction and greater coherence between reviews are concentrated [5,7,36,37]. In contrast, very high doses (≥10 g/day or >100 mg/kg/day) and longer courses showed no reproducible advantage and, in several analyses, attenuated the signal [6,30]. HAT regimens followed relatively uniform protocols (hydrocortisone 50 mg/6 h + vitamin C 1.5 g/6 h + thiamine 100–200 mg every 6–12 h for 3–7 days) and vitamin C + thiamine replicated those doses omitting the steroid; in both cases, favorable effects were limited to ΔSOFA 72–96 h and vasopressor hours of small magnitude, without reproducible impact on survival [10–13].

Overlap index (CCA)

To quantify trial overlap between reviews in each silo, we calculated the CCA from the lists of RCTs included in each review. In HAT [10–12,28,35], we identified 10 unique RCTs appearing 45 times in total across 5 reviews; with Ns = 10, No = 45, r = 5 and c = 10, the CCA was 0.875 (87.5%), indicating very high overlap. In IV vitamin C monotherapy [5–7,37,40], we counted 10 unique RCTs appearing 40 times in 5 reviews; Ns = 10, No = 40, r = 5, c = 10 → CCA 0.750 (75.0%), also very high. In vitamin C + thiamine, 7 unique RCTs were recorded with 12 appearances in 2 reviews; Ns = 7, No = 12, r = 2, c = 7 → CCA 0.714 (71.4%), equally very high. These values confirm that reviews within each silo essentially synthesize the same body of trials; therefore, in the narrative synthesis we prioritized reviews of higher quality (high AMSTAR-2) and, when applicable, with TSA as main estimators, using the rest for sensitivity analysis and coherence, without additional quantitative aggregation. Inclusion matrices (RCT × review) used for calculation will be attached as supplementary material.

Methodological quality (AMSTAR-2)

The methodological quality of reviews supporting the main conclusion was predominantly high or moderate (See S3 Table). In monotherapy, several syntheses met critical items (prior protocol, risk of bias assessment, appropriate meta-analytical methods, consideration of bias in interpretation, and formal publication bias assessment) and achieved high AMSTAR-2 [6,7,37]. In HAT, meta-analyses with TSA [10–12] also showed high AMSTAR-2, as did the component/network MA that disaggregated component effects [34]. The remaining monotherapy and combination reviews generally obtained moderate AMSTAR-2 due to a non-critical weakness (e.g., incomplete list of excluded studies or partial reporting of primary study funding) [5,9,13,30,41]. A mini-MA in letter format lacked protocol and presented abbreviated domain assessment, resulting in low rating [43]. Strength and weakness patterns were consistent across silos: when reviews were pure RCTs and registered protocol with justified exclusion list, the rating was high, and when they integrated RCTs + observational studies without clear stratification or without discussing risk of bias impact, the rating descended to moderate [14,15,35].Explicitly, umbrella conclusions were anchored in reviews with high AMSTAR-2 and, when available, with TSA [6,10–12], while those of moderate quality were used for sensitivities (dose, initiation, duration, severity) and external coherence.

Certainty of evidence (GRADE)

Certainty for IV vitamin C monotherapy on 28–30-day mortality is low-to-moderate: several SR/MA of pure RCTs show coherent relative reductions (RR 0.60–0.75) when the intervention is administered early (≤24 hours), with intermediate dose (around 1.5 g/6 h; 25–100 mg/kg/day) and for 3–4 days [5–7,36]. Certainty is downgraded due to clinical/statistical heterogeneity, publication bias indicators in large meta-analyses (Egger tests), and still moderate sample sizes. Nonetheless, the directional consistency between RCT syntheses supports a greater clinical probability of benefit in the sepsis scenario compared to shock [9]. For surrogates with monotherapy (ΔSOFA, vasopressors), certainty is low, due to variability between studies and elevated I² in several analyses [6,27] (See Table 2).

Download:

• PNG
  larger image
• TIFF
  original image

Table 2. Summary of Findings (GRADE): Intravenous vitamin C monotherapy in adults with sepsis/septic shock.

https://doi.org/10.1371/journal.pone.0351072.t002

In HAT and vitamin C + thiamine, certainty for mortality is low (stable null effect or insufficient power according to TSA), while for ΔSOFA 72–96 h and vasopressor hours it is moderate: effects are consistent and statistically robust, but of small magnitude and without translation into length of stay or survival [10–13] (See Table 3). Component analyses confirm that the glucocorticoid explains vasopressor shortening and ICU day reduction in combined regimens, which does not modify certainty for survival when adding vitamin C or thiamine [20,34] (See Table 4).

Download:

• PNG
  larger image
• TIFF
  original image

Table 3. Summary of Findings (GRADE): HAT combination (hydrocortisone + vitamin C + thiamine) in adults with sepsis/septic shock.

https://doi.org/10.1371/journal.pone.0351072.t003

Download:

• PNG
  larger image
• TIFF
  original image

Table 4. Summary of Findings (GRADE): Vitamin C + thiamine (without corticosteroid) in adults with sepsis/septic shock.

https://doi.org/10.1371/journal.pone.0351072.t004

Main findings

IV vitamin C monotherapy showed a consistent signal of mortality benefit at 28–30 days when administered in sepsis (more than in shock), with intermediate doses and short courses; however, supportive outcomes were heterogeneous across reviews (reduction in vasopressor hours yes; ΔSOFA not reproducibly in pure RCTs). Conversely, combinations (HAT and vitamin C + thiamine) did not reduce mortality in reviews with the best methodological standards; their effect was limited to small and consistent improvements in ΔSOFA 72–96 h and vasopressor hours, without clear impact on length of stay or translation into survival.

Network/component meta-analysis confirmed that the corticosteroid is the main determinant of vasopressor shortening and slight ICU length reduction in combined regimens; adding vitamin C or thiamine did not modify survival in these schemes. Overall, the pattern is clear: monotherapy with probable short-term mortality benefit under specific conditions; combinations with modest hemodynamic benefit but no effect on mortality.

Interpretation of results

Interest in adjuvants like IV vitamin C stems from a dual goal: reducing deaths and, if not possible, improving organ function/hemodynamic stability with safe interventions [1,3]. Vitamin C, water-soluble with a good safety profile, has biological plausibility (cofactor of dopamine-β-hydroxylase for catecholamines, NO modulation, endothelial antioxidation, and immunomodulation), which could reduce vasopressor support and, under favorable conditions, impact survival [4].

The most coherent signal appears with monotherapy, early, intermediate dose, and 3–4 days in sepsis, which aligns with the pathophysiological window [5,7,36,37]. In contrast, very high doses or longer exposures provided no advantages and, in sensitivity analyses, diluted the possible effect, suggesting a narrow therapeutic window [6,30]. It should be noted that the hypothesis of a beneficial effect under conditions of early initiation, intermediate dose, and short course is largely derived from post hoc subgroup aggregation across independent reviews, and was not pre-specified in the primary trials. As such, it should be interpreted as a hypothesis-generating finding rather than a confirmed clinical signal. The consistency of this pattern across methodologically diverse reviews lends it some plausibility, but prospective validation in trials designed around these specific parameters remains necessary before any clinical inference can be drawn. For HAT or vitamin C + thiamine, small decreases in ΔSOFA and vasopressor hours are plausible (steroid effect; metabolic role of thiamine), but are not accompanied by replicable improvements in metrics that matter to patients/hospitals; furthermore, component-NMA attributes the hemodynamic benefit to the corticosteroid, not the vitamins [10–13,20,34,41].

Moreover, the “multiplication” of reviews is partially justifiable (new and large RCTs, need to update monotherapy vs combinations, sepsis vs shock, dose/timing/duration) [10–12,20,34]. However, our CCA showed very high overlap: many reviews re-synthesize the same set of RCTs within each silo. Therefore, their incremental value depends on whether they add high AMSTAR-2 with well-reported protocol/excluded list/publication bias, prespecified and plausible subgroups (25–100 mg/kg/d vs > 100 mg/kg/d; initiation ≤24 h; 3–4 days; sepsis vs shock), TSA clarifying sufficiency/futility, or component-NMA assigning the effect to corticosteroid rather than vitamins [10–12,20,34].

It should be noted that high CCA does not invalidate the findings of individual reviews nor the conclusions drawn from them. Its primary implication is one of evidence redundancy: the proliferation of reviews in this field has not expanded the primary evidence base, as most analyses draw from the same compact set of RCTs. The consequence for certainty grading is indirect but relevant: because GRADE ratings are determined by the characteristics of underlying primary trials — not by the number of reviews analyzing them — additional reviews over the same RCTs do not increase certainty. High overlap therefore reinforces the rationale for the anchor estimator strategy and explains why certainty ratings remain low-to-moderate despite the large number of available reviews.

Regarding evaluated outcomes, surrogates serve as early signals (more sensitive than mortality) and for process monitoring, but their value for adoption decisions requires: (1) clinically relevant size (in SOFA, ≥ 1–2 points) and (2) consistent relationship with hard outcomes or resource use (length of stay, readmissions, mortality). Here, observed differences (SOFA around 0.6–0.9 points; vasopressors −15 to −22 h) are small, sensitive to co-interventions, and do not translate into length of stay or improved survival; moreover, the corticosteroid explains most of the combination effect [10–13,34]. For managers, introducing an intervention based solely on surrogates is not justifiable; they may guide hypotheses or stops for futility/benefit in RCTs with TSA, but do not replace clinical outcomes.

Interpretation of certainty (GRADE)

Many SRs do not imply high certainty. With GRADE, certainty depends on underlying RCTs: clinical heterogeneity, imprecision from moderate sizes, and possible publication bias limit certainty of monotherapy on 28–30 d mortality to low-moderate [5–7]. In HAT and vitamin C + thiamine, mortality remains at low certainty (stable null effect or TSA with insufficiency/futility), while ΔSOFA and vasopressors reach moderate certainty due to consistency and robustness, but magnitudes are small and do not justify practice changes [10–13]. With a limited and overlapping core of RCTs, certainty is unlikely to increase without larger and better-designed trials.

Implications for clinical practice

Our findings do not support routine use of HAT or vitamin C + thiamine to reduce mortality; observed hemodynamic benefits are small and attributable to the corticosteroid [20,34]. We do not recommend IV vitamin C monotherapy as standard; at most, a conditional/weak recommendation only in early sepsis, intermediate dose, 3–4 days, under protocol with audit and safety, preferably in RCTs [5–7,45]. This position does not contradict current guidelines, which advise against routine vitamin C [45].

Limitations

The synthesis relies on published reviews; although we prioritized high AMSTAR-2 and documented very high overlap to avoid RCT double-counting, clinical heterogeneity, imprecision, and publication bias persist in some large meta-analyses. We did not perform de novo meta-analyses and thus depend on each review’s analytical decisions. An important methodological limitation is our approach of selecting a single anchor estimator for each outcome rather than performing formal synthesis of all available estimators. In contexts of very high CCA — as observed across all three silos in this review — averaging effect sizes from reviews that share the same primary RCTs would artificially inflate precision and violate the independence assumption underlying standard meta-analytic pooling. The anchor strategy was adopted to avoid this, prioritizing the most methodologically robust review per outcome and regimen. Nevertheless, this approach introduces subjectivity in anchor selection that could influence GRADE ratings. To address this, we present a sensitivity analysis in S4 Table, showing that substituting the primary anchor with the second- and third-ranked reviews by AMSTAR-2 score does not alter the direction or certainty category of any GRADE conclusion. The absence of prospective registration constitutes an additional limitation of this umbrella review, as it precludes independent verification that the protocol was not modified after data collection began. To partially mitigate this, we adhered to PRIOR reporting guidelines, documented all methodological decisions prior to analysis, and maintained an auditable extraction log with timestamps and author attribution, allowing critical evaluation and replication of our procedures. Finally, recent RCTs may not be included. However, given the consistent pattern and observed effect sizes, few isolated trials would substantially change the overall conclusions. We recommend periodic updates.

The high I² values observed for vasopressor duration (e.g., I² = 95% in Zhu 2022; I² = 45% in Ge 2021) and, to a lesser extent, for ΔSOFA, reflect sources of heterogeneity that go well beyond expected clinical variation. At least three operational factors contribute to this. First, the choice of first-line vasopressor differed across primary trials: norepinephrine was standard in most Western ICUs, whereas some Asian trials included dopamine or vasopressin as primary agents, directly affecting the baseline duration and dose-tapering trajectory. Second, the definition of “vasopressor duration” was not uniform: some trials recorded total hours of any vasopressor infusion, others reported time to cessation of the primary agent only, and at least one defined the endpoint as vasopressor-free days within a fixed window. Third, the timing and dose of vitamin C initiation varied considerably across trials included in each review, which interacts with vasopressor weaning in ways that are not linear. These distinctions are not merely semantic: pooling studies with different vasopressor definitions under a single MD or WMD produces a composite estimate of uncertain clinical meaning, which in part explains why vasopressor reduction was statistically significant in some SRs but not replicable in TSA-anchored analyses.

Conclusions and recommendations

The available evidence does not support routine use of intravenous vitamin C in adults with sepsis or septic shock, either as monotherapy or in combinations (HAT or vitamin C plus thiamine). In combinations, no mortality reduction is observed. The described hemodynamic effects—modest decreases in ΔSOFA and vasopressor hours—are small, of limited clinical relevance, and primarily attributable to the corticosteroid component. Regarding monotherapy, although some high-quality systematic reviews point to a possible signal of mortality benefit under specific conditions (early administration, intermediate dose, short courses, and in sepsis more than shock), overall certainty remains low-to-moderate and is constrained by insufficient sample sizes, clinical heterogeneity, publication bias indicators, and very high overlap among reviews, which precludes firm clinical recommendations.

Consequently, intravenous vitamin C—either alone or in combination—is not recommended as part of routine management of sepsis or septic shock. The signal observed with monotherapy warrants, instead, multicenter and rigorous clinical trials exploring defined scenarios (initiation ≤24 h, intermediate dose for 3–4 days, explicit distinction between sepsis and shock, and consideration of baseline vitamin C status), with clinically relevant outcomes (28–30-day mortality), standardized risk of bias assessment, and sequential analysis when appropriate. Furthermore, future syntheses are required that reduce overlap, integrate larger RCTs, and apply formal and transparent publication bias assessments, aiming to decrease current uncertainty and clarify the true role of this intervention.