# 37. Metformin-associated B12 deficiency and further terminal loss

**Best judgment: about 19%; uncertainty range (sensitivity): 0.54–69%.** Long-term metformin causes biochemical B12 depletion. The unresolved question is how often this becomes a reversible cofactor deficit in the cells maintaining sensory terminals, and whether correcting it preserves enough physical terminals over five years. These are subjective probabilities of the complete causal proposition, not supplementation response rates. Reviewed 14 September 2026. [Probability method](probability_method.md) · [Collection index](README.md).

## Proposition, population and intervention

In adults with type 2 diabetes taking metformin for at least four years, metformin-associated B12 depletion creates an intracellular cofactor deficit before further terminal loss. Correcting that deficit while continuing metformin would preserve **at least 20% of five-year net peptidergic C-fiber terminal loss**, in a baseline group accounting for **at least 10% of new or progressive distal neuropathy in this exposed population**.

This is a **secondary deficiency and progression scope exception**. Participants may already have DPN and need not have recently diagnosed diabetes. The claim does not mean that B12 depletion explains ordinary incident DPN in early diabetes. A second B12-related cause is included in the neuropathy denominator; excluding it by definition would make this question unanswerable.

Follow fixed bilateral distal-leg and dorsal-foot skin territories. Identify peptidergic C fibers by stable neuronal identity, independently of current peptide staining. Net loss is physical disappearance minus newly established terminal entries, counting subsequent disappearance of replacements. Average territories within people, then people equally. Let L₀ and L_Q be five-year losses under contemporary care and Q, normalized to baseline terminal amount. For L₀ >0, preservation **B = (L₀ − L_Q)/L₀**.

A reference-care case is new bilateral length-dependent neuropathy with objective small-fiber loss or abnormal conduction, or established neuropathy with a further **≥10% baseline-normalized physical pan-small-fiber deficit** over five years. Isolated pain changes do not define progression. Let **C** be the fraction of these reference-care cases belonging to a qualifying baseline group. This is case coverage, not the fraction of metformin users with a low blood result.

Candidate groups are finite: all eligible users; serum B12 <150 pmol/L; B12 150–220 pmol/L; age ≥65; daily metformin dose ≥2,000 mg; and CUBN rs1801222 AA genotype. These are separate groups without additional intersections. Membership is fixed at entry before Q; no group is selected by later treatment response. A later claim requires at least one candidate satisfying the entire preceding prefix. Complete NO must exclude every remaining candidate.

**Q — selective cofactor repletion.** Throughout five years, raise deficient intracellular bioavailable methylcobalamin and adenosylcobalamin to the medians of their joint cell-type-specific distribution in B12-replete T2D peers matched for age, sex, BMI and kidney function. Leave concentrations already above their matched medians unchanged. Apply this rule throughout the body, including neurons, their axons and terminals, Schwann cells, satellite glia, skin and blood cells. The reference peers have serum B12 ≥220 pmol/L and normal direct cofactor-dependent enzyme function.

Q changes the availability of these two B12 cofactors; it does not directly normalize enzyme expression, ATP, methylation, myelin or nerve density. Keep metformin and other assigned care comparable. Natural downstream changes in glucose handling, anemia, repair, inflammation and compensatory pathways, including costs, count in the outcome. Oral 1 mg methylcobalamin is evidence about an implementable exposure, not proof that Q's intracellular target was reached.

For the local deficit, use either **MTR**, the methylcobalamin-dependent methionine synthase reaction, or **MMUT**, the adenosylcobalamin-dependent methylmalonyl-CoA mutase reaction. R2 requires deficient cofactor availability below the matched fifth percentile and a **≥20% increase in the corresponding cellular reaction flux with Q**, without directly changing substrate supply or enzyme amount at the comparison's outset. It must occur in at least 10% of people and on connected neuron/support-cell units serving at least 10% of their mapped terminals, before a further ≥5% physical net deficit lasting 30 days in year one. If established loss exists at entry, timing refers to additional loss.

R1 separately specifies the source: in a corresponding T2D metformin-initiation population with serum B12 ≥150 pmol/L, four years of metformin must increase the risk of B12 <150 pmol/L by **at least one percentage point** compared with no metformin and otherwise glucose-matched care. This source comparison and R2's actual local deficit are both required; a supplementation effect unrelated to depletion cannot substitute for them.

R2 also connects the source to those particular units: a counterfactual history without metformin from its initiation, with ideal glucose matching and the same B12 intake, must prevent **at least 20% of their Q-responsive reaction-flux deficit**. This permits additional dietary or absorption causes but excludes a wholly unrelated deficiency merely found in a metformin user. Baseline candidate membership is always defined under the metformin history and held fixed across this source comparison.

For both source comparisons, glucose matching means an ideal controller replaying the metformin-reference continuous blood-glucose trajectory, with other initially assigned drugs and B12 intake held equal. It is not an unspecified alternative medication. Non-glucose effects of metformin withdrawal may respond. This controller belongs to the source test; the five-year Q comparison continues metformin in both arms and permits glucose to respond naturally to cofactor repletion.

## Biological sequence and alternatives

Cobalamin must be absorbed, transported into cells and converted into active cofactors. Metformin can impair B12 availability through the intestinal route. Loss of either cofactor can disturb one-carbon metabolism or mitochondrial substrate handling. Axons could suffer directly, or through impaired supporting cells. Neither an MMA-toxin mechanism nor demyelination is mandatory.

Methylmalonic acid (MMA) can rise when the MMUT reaction lacks B12. It also varies with renal clearance and precursor metabolism. Total serum B12, circulating MMA and neuronal cofactor occupancy are different measurements. High MMA alone does not establish deficient neural B12, and a normal blood count does not exclude a neurological deficit.

An apparent supplementation benefit could instead be a pharmacological growth effect at high cobalamin exposure, an effect of another ingredient, improved large-fiber function, or altered symptoms without retained terminals. Conversely, a null motor-conduction trial in asymptomatic older adults does not resolve the effect in metformin-treated people with established neuropathy.

## Evidence and its limits

**Randomization supports the depletion step.** The HOME trial randomized **390 insulin-treated T2D patients** to metformin or placebo for 52 months. The target was 850 mg three times daily; actual mean exposure was about **2,050 mg/day**. Metformin reduced serum B12 by **19% relative to placebo, 95% interval 14–24% lower**. Deficiency below 150 pmol/L increased by **7.2 percentage points, interval 2.3–12.1**; the additional risk of levels 150–220 pmol/L was 11.2 points.

The biochemical analysis used last observations carried forward. Only **256 participants had final B12 measurements**, although 277 remained on study medication; percentages from the primary analysis must not be divided by the final laboratory n. The trial supports a causal blood-level effect despite attrition and baseline age imbalance. It did not measure terminal loss or intracellular neural cofactors. [de Jager et al., 2010](https://pmc.ncbi.nlm.nih.gov/articles/PMC2874129/).

DPP/DPPOS provides a second randomized exposure history in people initially at high diabetes risk. At five years, B12 deficiency occurred in **4.3% of metformin versus 2.3% of placebo-assigned participants**; at 13 years, the figures were **7.4% versus 5.4%**, with P=0.12 for the latter contrast. The metformin/placebo laboratory samples were **858/857 at five years and 764/756 at 13 years**. Later exposure included open-label and nonstudy treatment.

Among metformin participants with low B12 at year 13, **13/56 had abnormal monofilament results**. This was a small, post-treatment biomarker subgroup; the MNSI questionnaire and total examination score did not corroborate the difference. It is suggestive clinical context, not a randomized effect of deficiency or a C-terminal endpoint. [Aroda et al., 2016](https://pmc.ncbi.nlm.nih.gov/articles/PMC4880159/).

**A new absorption-related genotype strengthens source heterogeneity.** A 2026 study analyzed **487 B12-deficiency cases and 6,686 controls** among UK Biobank metformin users. For CUBN rs1801222, odds ratios were **1.56 for AG and 2.43 for AA**, versus GG. The outcome included clinical diagnoses or B12-injection prescriptions. Metformin-user deficiency proportions were approximately **6.0%, 8.0% and 12.8%** across GG, AG and AA.

Scottish and DPP analyses supplied replication, though small strata were less consistent. At DPP year five, the AA-versus-GG odds ratio was 3.98 in placebo and 3.58 in metformin; significance in only one arm does not itself prove an interaction. At year 13, AA metformin participants had substantially more deficiency. This makes a finite susceptibility group plausible without establishing its neural cofactor deficit or neuropathy coverage. No physical nerve endpoint was included. [Baldwin et al., 2026](https://pmc.ncbi.nlm.nih.gov/articles/PMC12957035/).

**The most relevant supplementation trial is favorable, but structurally indirect.** Didangelos randomized **90 people, 44 active and 46 placebo**, with established peripheral and autonomic diabetic neuropathy, metformin use ≥4 years and B12 <400 pmol/L. They received **1,000 µg/day oral methylcobalamin for 12 months** or placebo; no dropouts were reported. Mean diabetes duration was roughly 12–14 years and mean metformin duration 10–13 years. HbA1c remained near 6.8%; eGFR <50 was excluded.

B12 rose **232→777 pmol/L**, versus **231→243** on placebo. Sural action-potential amplitude changed **5.2→7.3 versus 5.1→4.6 µV**, a difference in changes of **+2.6 µV**. Conduction velocity's change advantage was approximately **4.01 m/s**, vibration threshold 6.7 V, and painDETECT score 2.9 points in the favorable direction. These are not measurements of peptidergic terminal preservation. PainDETECT is a neuropathic-symptom instrument, not a 0–10 pain-intensity scale.

The enrollment threshold also included many people within the laboratory's stated B12 reference range of **145–569 pmol/L**. MMA, homocysteine and holotranscobalamin were not measured. Thus a randomized functional benefit does not establish correction of a demonstrated cellular deficiency. [Didangelos et al., 2021](https://pmc.ncbi.nlm.nih.gov/articles/PMC7912007/).

There are material reporting inconsistencies. The active-arm Valsalva means and SDs allow a maximum paired-test P of about **0.182**, yet the table reports **0.983**; the placebo counterpart is similarly incompatible. Baseline folate means/SDs imply an ordinary Welch P near **0.00020**, versus reported 0.172. These discrepancies cannot be repaired without the underlying participant data.

They do not erase every favorable observation. The unadjusted 12-month amplitude difference is **2.7 µV, approximate 95% interval 0.87–4.53** using the reported independent-group SDs. A conservative Holm check across all 15 distinct Table 4 outcomes retains the smallest reported P values, while vibration and foot skin conductance become approximately 0.063 and 0.064. That calculation assumes the printed tests are otherwise valid; it cannot validate them. The registry was first submitted after trial completion, so its outcome list is not a prospectively fixed family. [Trial registry](https://clinicaltrials.gov/study/NCT04706377); [aggregate checks](../../../findings/damage_calibration_b12_quantitative.md).

**Engaged negative trials constrain a universal supplementation story.** Dangour's trial analyzed **201 eligible randomized adults aged ≥75**, explicitly excluding diabetes, anemia and very low B12. Participants had moderate biochemical deficiency and received 1 mg/day cyanocobalamin or placebo for a year. There were **191 participants with outcome data**, with 91 per arm in the main nerve table. B12 increased by 177%, active transport marker holotranscobalamin by 331%, and homocysteine fell 17%.

The primary posterior-tibial compound muscle action-potential effect was **−0.2 mV, interval −0.8 to +0.3**; sural sensory amplitude was **−0.6 µV, interval −1.5 to +0.2**. This argues against a large generic functional benefit from correcting moderate biochemical deficiency in asymptomatic older people. It does not bound Q's long-term peptidergic effect in the target T2D population. [Dangour et al., 2015](https://pmc.ncbi.nlm.nih.gov/articles/PMC4548176/).

Another randomized study of **140 people with MMA 0.40–2.00 µmol/L** reduced MMA and homocysteine with B12 over three months but found no improvement in neurological disability or symptoms. Its short duration and nonspecific endpoints leave the structural comparison open. [Hvas et al., 2001](https://doi.org/10.1093/clinchem/47.8.1396).

**The same HOME cohort supplies both metabolic concern and a net clinical null.** A post hoc analysis found that metformin increased MMA. Its mediation model assigned an adverse neuropathy-score pathway to MMA and a beneficial pathway to glycemic improvement, while the overall metformin effect was **0.032 score units, interval −0.121 to +0.182, P=0.34**. The mediator was not randomized. This is neither independent replication of HOME nor proof that MMA caused the adverse component. The net clinical null also cannot identify the benefit of continuing metformin while correcting only B12. [Out et al., 2018](https://doi.org/10.1016/j.jdiacomp.2017.11.001).

**Structural evidence exists, but does not isolate this human deficit.** A skin-biopsy study compared **10 B12-deficient people with pain and 10 without pain**, rather than deficient participants against concurrent B12-normal controls. Median epidermal density was **3.345 versus 6.20 fibers/mm**, both below the referenced age norms. This supports small-fiber involvement as a possibility, without determining metformin causation or replacement efficacy. [Güneş et al., 2018](https://doi.org/10.1007/s13760-017-0847-y).

In diabetic ZDF rats, four weeks of Metanx improved epidermal density with **8–10 animals per group**, but the treatment combined methylcobalamin, methylfolate and pyridoxal phosphate. In vincristine-treated rats, **0.5 mg/kg intraperitoneal methylcobalamin** preserved epidermal fibers, **n=5 per group**. The latter shows a structural action of methylcobalamin outside diabetes; neither study proves that correction of metformin-induced cofactor deficiency produced the effect. [Metanx study](https://pmc.ncbi.nlm.nih.gov/articles/PMC3402301/); [vincristine study](https://pmc.ncbi.nlm.nih.gov/articles/PMC4956006/).

**MMA genetics does not rule out B12-deficiency injury.** Molloy's study found that HIBCH rs291466 strongly raises circulating MMA, with approximately **46% higher MMA in one homozygous group**, independently of total B12, holotranscobalamin and homocysteine. Its mechanism involves valine metabolism. A neuropathy null for that allele therefore tests a different exposure from deficient B12-dependent enzyme function. It cannot establish that functional B12 deficiency is harmless, nor that every observational MMA association is confounding. The same distinction prevents treating all MMA elevation as a causal toxin exposure. [Molloy et al., 2016](https://pmc.ncbi.nlm.nih.gov/articles/PMC4863564/).

## Conditional probabilities

Each probability after R1 conditions on the whole preceding prefix. R1 is the source comparison; R2–R5 require a finite baseline group meeting all applicable requirements.

| # | Necessary claim | Conditional YES: low / best / high | Best cumulative |
|---|---|---:|---:|
| R1 | Four-year metformin exposure causes ≥1 percentage point excess risk of serum B12 <150 pmol/L in the specified initiation comparison. | 85 / **97** / 99% | 97% |
| R2 | The specified cofactor-responsive cellular deficit, with the required metformin-attributable component, occurs in the required people and connected units before further loss. | 30 / **65** / 90% | 63.05% |
| R3 | In such a group, L₀ >0 and selective repletion gives retained five-year peptidergic preservation B ≥5%. | 30 / **70** / 95% | 44.135% |
| R4 | The retained effect reaches B ≥20%. | 35 / **70** / 90% | 30.8945% |
| R5 | The group accounts for C ≥10% of reference-care new or progressive neuropathy. | 20 / **60** / 90% | **18.5367%** |

R1 has direct randomized biochemical support. R2 remains appreciably lower: circulating changes, CUBN association and peripheral function do not establish a neural cellular deficit of the specified magnitude. Given that deficit, basic cofactor biology, the favorable randomized functional trial and animal structural rescue justify R3 above one-half, while the engaged null trials and endpoint gaps prevent near certainty.

R4 asks about retained effect size after replacement and compensation, not simply whether B12 is needed. R5 depends on the distribution of preventable neuropathy among all exposed people. A 10% blood-deficiency prevalence neither proves nor disproves 10% case coverage.

The headline is **0.97 × 0.65 × 0.70 × 0.70 × 0.60 = 0.185367**. The source/local-deficit joint is 63.05%; the remaining structural-magnitude-coverage joint is 29.4% conditional on it. Splitting these requirements does not impose an independent penalty.

## Uncertainty and informative updates

The **lower scenario**, [85,30,30,35,20]%, gives **0.5355%**: biochemical depletion is real, but few cells have a relevant deficit, supplementation mostly changes function or a narrow subgroup, and little five-year terminal loss is prevented.

The **upper scenario**, [99,90,95,90,90]%, gives **68.56245%**: blood assays miss frequent neural cofactor deficiency, the functional trial reflects structural rescue, and susceptible groups account for enough progressive neuropathy. The negative trials then concern less relevant populations or time scales.

The range is approximately **68.0 percentage points wide**. It is sensitivity to these jointly defensible interpretations, not a confidence interval, probability distribution or measured variance. The preferred interpretation retains the direct positive while giving real weight to its reporting defects, absent engagement and missing structural endpoint. Five-point changes in individual judgments are not well resolved by the evidence.

A **large effect in a rare truly deficient group**, [99,90,90,85,15]%, gives **10.2242%**. A **common biochemical change with small structural benefit**, [99,80,45,35,80]%, gives **9.9792%**. These differ biologically despite similar complete probabilities.

Hypothetical demonstration of cell-level repletion, durable peptidergic preservation and sufficient case coverage could move the full row set to [99,90,90,85,80]%, giving **54.5292%**. Precise absence of structural preservation despite demonstrated local engagement could move it to [97,70,25,20,40]%, giving **1.358%**.

Changing R2 alone from 65% to 90% gives **25.6662%**; reducing it to 30% gives **8.5554%**. Certainty about R2 alone permits **28.518%** with other judgments fixed. This is an arithmetic sensitivity, not the expected value of investigating that link.

## Ideal experiments that would resolve each claim

Use unlimited selective control and complete measurement, with independent people as inference units. These comparisons define the truth conditions; they are not proposed treatment protocols. Preserve physical identities, observe replacement and downstream costs, and hold baseline group membership fixed. Exact population quantities decide YES or NO. Inadequate engagement, incomplete follow-up, marker-only outcomes or estimates crossing a boundary remain **unresolved**.

**R1 — causal source.** In the specified B12-normal T2D initiation population, compare four years of assigned metformin with no metformin and glucose-matched alternative care, with comparable B12 intake and direct measurement of absorption, transport and serum concentration. **YES:** the risk difference for B12 <150 pmol/L is ≥0.01. **NO:** it is <0.01. Absorption measurements locate the mechanism but are not an extra mandatory route. HOME approaches the duration and randomized contrast; its imperfect glycemic match and missing final laboratories prevent it from being the ideal comparison.

**R2 — actual cellular deficit and order.** In the exposed cohort's finite groups, measure both active cofactors, enzyme occupancy and MTR/MMUT reaction flux in connected neural and support compartments. Compare Q with reference conditions before the additional physical deficit, and resolve the counterfactual glucose-matched no-metformin history for those same units. **YES:** at least one group meets the ≥10% people/≥10% units criterion, deficient concentration threshold, ≥20% cofactor-responsive flux increase and ≥20% metformin-attributable share of that deficit before the specified loss. **NO:** every candidate fails at least one condition. Serum B12 or MMA alone cannot decide this. The CUBN study and HOME indicate exposure susceptibility; neither measured this local state or its source-attributable fraction.

**R3 — retained target-fiber effect.** In R2-qualified groups, compare five years of Q against reference care while metformin continues. Count cumulative disappearance, new entries and later loss using stable peptidergic identity. **YES:** a group has L₀ >0 and B ≥0.05. **NO:** no qualified group meets both. Supplementation trials approach assignment but provide function and symptoms; rodent histology lacks the specified human depletion context.

**R4 — material size.** Continue that same complete structural comparison. **YES:** an R1–R3 group has B ≥0.20. **NO:** every such group falls below 0.20. If reference care loses 30% of baseline terminals, Q must preserve at least six baseline percentage points. Biomarker normalization, pain improvement or increased density without baseline-normalized net accounting cannot substitute.

**R5 — exposed-population coverage.** Ascertain all new or progressive cases in a complete reference-care metformin cohort using the definition above, including B12-related second-cause neuropathy. **YES:** at least one R1–R4 group contains ≥10% of those cases. **NO:** none does, or no reference cases occur. A referral sample, percentage with low serum B12, or post-treatment subgroup does not supply the required denominator.

## Boundaries with related mechanisms

B12 deficiency can affect [energy handling](07_bioenergetic_insufficiency.md), [Schwann-cell support](17_schwann_metabolic_support.md) or [growth capacity](44_intrinsic_growth_restriction.md), but no single one is compulsory here. [Kidney failure](38_kidney_failure_uremic_exposure.md) can elevate MMA without the same cofactor deficit; kidney matching and direct flux measurement distinguish those possibilities.

Selective cofactor repletion is different from metformin withdrawal. The evidence supports a credible additional cause of neuropathy in some exposed people, while the specified physical preservation and coverage remain unresolved.