This systematic review was conducted in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statement (Moher et al., 2009) and the guidelines from the Cochrane Handbook for Systematic Reviews of Interventions (Higgins and Green, 2008). PRISMA checklist is provided in Supplementary Table S1. The systematic review protocol was registered in the International Prospective Register of Systematic Reviews (PROSPERO), registration number: CRD42024554513 (Available from) (NIHR, 2025).

The primary research question addressed in this review was: “What is the effectiveness of sublingual/oral administration of Vitamin B12 alone or compared to other forms of administration (e.g., intramuscular injection)?”. The PICOs elements were defined as follows:

P (Participants): subjects of any age diagnosed with Vitamin B12 deficiency.

Three databases (PubMed, Embase, and Scopus) were searched up to 15 July 2024.

The search strategy was initially developed for PubMed using keywords and MeSH terms and was then adapted to the other databases. Search strings used for each database are presented in Supplementary Table S2. Reference lists of included studies and relevant reviews were screened manually to identify additional eligible studies. Rayyan software (Ouzzani et al., 2016) was employed for reference management, including the identification and removal of duplicate entries.

Reviewers underwent training at each step of the process, and a pilot screening was conducted to ensure consistency in applying the eligibility criteria. Three authors (MM, AN, and JV) independently screened titles and abstracts following de-duplication. Disagreements were resolved through discussion or by consultation with a fourth author (CS).

The same three authors proceeded to full-text screening, with any discrepancies resolved in the same manner.

The inclusion and exclusion criteria were defined in direct accordance with the PICOS framework. Specifically, eligible studies included participants (P) with a confirmed diagnosis of vitamin B12 deficiency, interventions (I) involving oral or sublingual vitamin B12 administration, comparators (C) using intramuscular supplementation, and outcomes (O) reporting changes in serum vitamin B12 or homocysteine levels. Only clinical study designs (S)—randomised controlled trials (RCTs), non-RCTs, cohort, or case-control studies—were included.

Inclusion criteria were: (a) RCTs; (b) N-RCTs; (c) case-control studies; (d) cohort studies; (e) cross-sectional studies with no time restriction, (f) only studies published in English were included due to resource limitations and to ensure accuracy in data extraction and risk-of-bias assessment.

Exclusion criteria were: (a) in-vitro RCTs; (b) lack of effective statistical analysis; (c) abstract and author debates or editorials; (d) review articles or non-clinical studies; (e) case reports and case series, ongoing and unpublished studies; (g) studies lacking a comparator group.

This review included both RCTs and observational studies (N-RCTs, cohort, and case-control) to assess the efficacy of various Vitamin B12 administration routes. While RCTs remain the gold standard for intervention assessment, observational studies provide complementary evidence, especially when using objective biomarkers such as serum Vitamin B12, MMA, and homocysteine. This approach is consistent with established methodologies adopted in previous systematic reviews assessing micronutrient interventions.

For each eligible study, two authors (MM and AN) independently extracted data, which was then reviewed by a third author (JV) using a piloted spreadsheet, in accordance with Cochrane Collaboration guidelines (Higgins and Green, 2008).

Where relevant data were missing, attempts were made to contact corresponding authors. Studies for which no response was received were excluded.

The following data were collected: publication year, country and continent, study setting (e.g., hospital, private practice, academic clinic), sample size, age range, mean age, study design, population characteristics, intervention details, outcomes, and effect measures (e.g., serum B12 levels). Subgroup analyses were planned based on age group (e.g., children, adults, older adults), baseline serum Vitamin B12, and the form of Vitamin B12 used.

In line with PRISMA guidelines, the assessment of methodological quality was conducted to evaluate the strength of evidence, recognising that methodological limitations may introduce bias.

For randomised clinical trials, the Jadad scale (Jadad et al., 1996), as used to assign a quality score ranging from 0 to 5, based on randomisation, blinding, and withdrawal reporting. A score of ≥3 was considered indicative of good quality.

To complement the Jadad score, the Cochrane Risk of Bias Tool (RoB 2) (Sterne et al., 2019) was also used for RCTs to provide a domain-based qualitative evaluation, assessing risks related to selection, performance, detection, attrition, and reporting.

For observational studies (cohort and case-control), the Newcastle–Ottawa Scale (NOS) (La Torre et al., 2014), was applied. This tool evaluates selection, comparability, and exposure or outcome assessment. A maximum of nine stars could be awarded, with a higher number indicating better quality.

Risk of bias was interpreted as follows: i) Low risk: All criteria met or no more than one rated as unclear; ii) Moderate risk: Two criteria rated as unclear, or one criterion not met; iii) High risk: Three or more criteria rated as unclear or at least two criteria not met.

The Grading of Recommendation, Assessment, Development, and Evaluation (GRADE) tool was used to assess the certainty of the evidence for the two primary biochemical outcomes—serum vitamin B12 concentration increase and homocysteine reduction—across all included studies, and separately for randomised controlled trials (RCTs) and observational studies.

Meta-analysis was performed using the R statistical program, R version 4.4.2 (The R Foundation for Statistical Computing, Wirtschaftsuniversitä̈t Wien, Vienna, Austria) (Sterne et al., 2016; Viechtbauer, 2010), employing a random-effect model via the metafor R package (Del, 2015). The mean change was used as effect estimate. Heterogeneity was assessed quantitatively using the I2-statistics and Cochran’s Q (Higgins and Thompson, 2002). Results were considered statistically significant at p < 0.05. Publication bias was evaluated using a funnel plot and Egger’s test for asymmetry (Egger et al., 1997).

The search strategy identified 2,183 potential articles: 1,299 from PubMed, 353 from Scopus, 527 from Embase, 4 from manual screening. After removal of duplicates, 1762 articles were screened. Subsequently, 1717 articles were excluded because they did not meet the inclusion criteria. Of the remaining 45 articles, 6 full texts could not be retrieved and 14 were excluded as they were out of scope. The remaining 25 articles were included in the qualitative synthesis, and 16 in the quantitative synthesis (Figure 1).

The studies included in the qualitative synthesis (Table 2) (Tanaka et al., 1981; Yamane et al., 1995; Kuzminski et al., 1998; Altay et al., 1999; Adachi et al., 2000; Bolaman et al., 2003; Sharabi et al., 2003; Roth and Orija, 2004; Yazaki et al., 2006; Castelli et al., 2011; Kim et al., 2011; Tillemans et al., 2014; Saraswathy et al., 2012; Parry-Strong et al., 2016; Metaxas et al., 2017; Schijns et al., 2018; Sezer et al., 2018; Bensky et al., 2019; Arıcan et al., 2020; Sanz-Cuesta et al., 2020; Tuğba-Kartal and Çağla-Mutlu, 2020; Orhan Kiliç et al., 2021; Ramos et al., 2021; Tandon et al., 2022; Korpeti et al., 2023) were published between 1981 and 2024 and comprised 13 randomised controlled trials, 2 case control study and 10 cohort studies (6 retrospective and 4 prospective). A total of 6,098 participants were included (mean: 244).

The mean age of the enrolled participants was 53.8 (SD = 9). Gender-specific data were available in some studies, reporting 268 males and 288 females.

The included studies were conducted in 10 Countries across 5 continents (6 studies from Turkey; 4 from United States, 3 from Japan; 2 each from India, Israel and Netherlands and 1 each from Greece, Spain, Switzerland, Brazil, New Zealand and Korea). Seventeen studies were conducted in hospital settings, and the remaining 8 in university hospital settings.

Among the 25 studies included in the qualitative synthesis (16 in the meta-analysis), 15 focused specifically on serum cobalamin changes (reported by baseline and post-treatment means and standard deviations in pg/mL) and 10 reported homocysteine outcomes (means and standard deviations in mmol/L). Regarding the form of vitamin B12 used, cyanocobalamin was the most common (n = 14), followed by methylcobalamin (n = 6), hydroxocobalamin (n = 4), and combinations including adenosylcobalamin in isolated cases.

Additionally, 15 of the included studies focused on B12 supplementation, with a total of 5,191 participants (mean: 346 per study). In contrast, 10 studies investigated the relationship between B12 supplementation status and homocysteine levels, encompassing a total of 191 participants (mean: 19.1 per study).

According to the Jadad scale for RCTs (n = 13) (Tanaka et al., 1981; Kuzminski et al., 1998; Bolaman et al., 2003; Sharabi et al., 2003; Yazaki et al., 2006; Castelli et al., 2011; Tillemans et al., 2014; Saraswathy et al., 2012; Parry-Strong et al., 2016; Metaxas et al., 2017; Schijns et al., 2018; Sanz-Cuesta et al., 2020; Tandon et al., 2022) the authors evaluated the quality of the clinical trial included in the qualitative synthesis, based on 5 items assessing the randomization process, blinding, and the dropout rate (i.e., the patients lost to follow-up). Quality scores ranged from 2 to 5 stars; additionally, a domain-specific assessment using the Cochrane Risk of Bias Tool (RoB 2) was also conducted and reported in Table 3 to provide a comprehensive appraisal. In the evaluation of RCT quality, 3 studies scored two points, indicating low-quality study and high risk of bias; 6 studies scored three points, and the remaining 4 scored five points, indicating medium and high-quality studies, respectively. Consistently, the risk of bias assessment, revealed a “moderate risk” for the first 6 studies scoring three points, and “high risk” for the 4 studies scoring five points.

According to the Newcastle–Ottawa Scale (NOS) for case-control studies (n = 2) (Yamane et al., 1995; Adachi et al., 2000) and cohort studies (n = 10) (Altay et al., 1999; Roth and Orija, 2004; Kim et al., 2011; Sezer et al., 2018; Bensky et al., 2019; Arıcan et al., 2020; Tuğba-Kartal and Çağla-Mutlu, 2020; Orhan Kiliç et al., 2021; Ramos et al., 2021; Korpeti et al., 2023), the authors assessed study quality based on object selection, comparability and exposure. Risk of bias (RoB) scores, assessed using the NOS, ranged from 5 to 9 stars. The case-control studies both scored six, indicating high-quality studies with a moderate risk of bias (Table 4). Among the cohort studies, 2 scored three and five points, indicating low-quality studies with high and moderate risk of bias, respectively. 3 studies scored six, indicating medium-quality studies with a moderate risk of bias. 2 studies scored seven, indicating medium-quality studies with low risk of bias. The remaining 3 studies, scored eight and nine, indicating high-quality studies with low risk of bias (Table 5).

The main sources of inconsistency in the NOS scale were sample representativeness and sample size. While all studies adequately ascertained exposure and ensured comparability between outcome groups by controlling for confounders, 42% (n = 5) did not provide adequate sample representativeness and size, and 17% (n = 2) did not report the non-response rate. In 67% of the studies (n = 8), the adequacy of follow-up of cohorts was not described.

The GRADE assessment showed moderate certainty of evidence for the increase in serum vitamin B12 levels and low certainty of evidence for homocysteine reduction, both across all studies and within RCT-only and observational-only subgroups. For observational studies reporting serum vitamin B12 levels, the certainty of evidence was upgraded from low to moderate due to the large and consistent magnitude of effect across studies, despite heterogeneity. High heterogeneity (I² >80%) and concerns regarding imprecision and risk of bias were the main factors contributing to downgrading (Supplementary Table S3).

Across 35 comparisons, the mean percentage increase in serum cobalamin following supplementation was approximately 293%, confirming the overall effectiveness of vitamin B12 interventions (Supplementary Table S4).

When stratified by administration route, the mean percentage changes were:

Intramuscular (IM): ∼307%

IM administration showed the highest average B12 increase, although oral and sublingual routes also yielded substantial improvements, supporting their use as effective and less invasive alternatives.

By dosage, the mean changes were:

500 mcg: ∼127%

While 1,500 mcg of vitamin B12 showed the highest average increase, the effect plateaued or declined at higher doses, suggesting a possible saturation of absorption pathways. These findings support a personalised approach, where absorption efficiency may be more relevant than absolute dosage.

Analysis of 10 comparisons showed that vitamin B12 supplementation was associated with a mean reduction of −35.8% in serum homocysteine concentrations (Supplementary Table S5).

This confirms the well-documented role of cobalamin in lowering homocysteine levels in deficient or at-risk individuals.

When stratified by administration route, the average percentage reductions were:

Intramuscular (IM): −48.3%

These results suggest that IM administration leads the most pronounced homocysteine reduction, however, sublingual and oral routes also demonstrate clinically relevant effects.

With respect to dosage, the average reductions were:

2000 mcg: −71.5% (single data point)

Although limited data are available for higher dosages, the findings support the efficacy of both oral and non-oral B12 formulations in modulating homocysteine levels. IM administration may provide the most robust response in clinical scenarios requiring rapid or substantial metabolic correction.

The included studies showed substantial heterogeneity (I² = 99.6%, p < 0.001), suggesting considerable variability in study outcomes that cannot be attributed to chance alone. Egger’s test for funnel plot asymmetry (p < 0.001) indicates potential publication bias, which may influence the observed effect sizes (Figure 2). The prediction interval (−237.1, 1,042.3) reflects a wide range of potential true effects in future studies, highlighting the high heterogeneity of the included trials. Sensitivity analysis identified one influential study (Castelli et al., 2011), based on both DFFITS and Cook’s distance. This study reported by far the largest mean difference and very high variability. Excluding this study narrowed the prediction interval to (−83.2, 768.4), but only slightly reduced the heterogeneity (I² = 99.2%). The increase in cobalamin levels remains significant (p < 0.001). There were no statistically significant differences in cobalamin levels between RCT and observational studies (p = 0.268).

The dependence of efficacy on administration route, underlying pathology, age and dosage was examined.

Some studies appear multiple times across the forest plots because they reported more than one independent comparison (for example, evaluating different administration routes or dosage subgroups within the same study population). In accordance with Cochrane and PRISMA recommendations, each comparison was treated as a separate data point in the meta-analysis to maintain data granularity and prevent loss of relevant information.

Figure 4 presents the forest plot illustrating the effect of vitamin B12 administration stratified by underlying pathology: gastrectomy, vitamin B12 deficiency of unspecified etiology, and other diseases (Kuzminski et al., 1998; Adachi et al., 2000; Bolaman et al., 2003; Sharabi et al., 2003; Castelli et al., 2011; Kim et al., 2011; Parry-Strong et al., 2016; Metaxas et al., 2017; Schijns et al., 2018; Sezer et al., 2018; Bensky et al., 2019; Tuğba-Kartal and Çağla-Mutlu, 2020; Orhan Kiliç et al., 2021; Ramos et al., 2021; Tandon et al., 2022). The pooled analysis demonstrates a significant increase in cobalamin levels across all subgroups. However, the difference between pathology groups was not statistically significant (p = 0.132), suggesting a comparable response to B12 supplementation across clinical conditions: i) Gastrectomy: this subgroup includes patients who have undergone partial or total gastrectomy, a condition known to impair intrinsic factor-mediated absorption of vitamin B12. The pooled effect estimate indicates a substantial increase in B12 levels following supplementation; ii) Vitamin B12 Deficiency (unspecified etiology): this category includes individuals diagnosed with B12 deficiency without a clearly defined cause. The effect size within this group is similar to that of gastrectomy, highlighting the broad applicability of B12 supplementation across different deficiency states; iii) Other Diseases: this smaller subgroup includes conditions in which B12 deficiency may be secondary to disease-related malabsorption or altered metabolism. The observed heterogeneity in effect size suggests variability in response within this group. These findings underscore the efficacy of B12 supplementation across a range of pathological conditions while emphasizing the need for further research to better define disease-specific responses and optimal dosing strategies.

This meta-analysis has several limitations that should be considered when interpreting the findings. The substantial heterogeneity observed across studies (I² > 80%) reflects differences in populations, dosages, supplementation duration, and analytical methods, which may have influenced the pooled estimates. Additionally, the presence of publication bias suggested by Egger’s test and funnel plot asymmetry indicates that smaller studies with non-significant results might be underrepresented. The methodological quality of the included studies, as assessed through the Jadad and Newcastle–Ottawa scales, varied from moderate to high, while the GRADE assessment indicated moderate certainty of evidence for serum vitamin B12 increase and low certainty for homocysteine reduction. These factors collectively suggest that the conclusions, while robust, should be interpreted with caution in clinical decision-making. Future research should focus on large-scale, well-controlled randomized trials with standardized dosing protocols, clearly defined deficiency thresholds, and long-term follow-up to strengthen the evidence base and guide clinical practice.