To describe the prevalence of biochemical B12 deficiency in adults with type 2 diabetes taking metformin compared with those not taking metformin and those without diabetes, and explore whether this relationship is modified by vitamin B12 supplements.
Analysis of data on U.S. adults ≥50 years of age with (n = 1,621) or without type 2 diabetes (n = 6,867) from the National Health and Nutrition Examination Survey (NHANES), 1999–2006. Type 2 diabetes was defined as clinical diagnosis after age 30 without initiation of insulin therapy within 1 year. Those with diabetes were classified according to their current metformin use. Biochemical B12 deficiency was defined as serum B12 concentrations ≤148 pmol/L and borderline deficiency was defined as >148 to ≤221 pmol/L.
Biochemical B12 deficiency was present in 5.8% of those with diabetes using metformin compared with 2.4% of those not using metformin (P = 0.0026) and 3.3% of those without diabetes (P = 0.0002). Among those with diabetes, metformin use was associated with biochemical B12 deficiency (adjusted odds ratio 2.92; 95% CI 1.26–6.78). Consumption of any supplement containing B12 was not associated with a reduction in the prevalence of biochemical B12 deficiency among those with diabetes, whereas consumption of any supplement containing B12 was associated with a two-thirds reduction among those without diabetes.
Metformin therapy is associated with a higher prevalence of biochemical B12 deficiency. The amount of B12 recommended by the Institute of Medicine (IOM) (2.4 μg/day) and the amount available in general multivitamins (6 μg) may not be enough to correct this deficiency among those with diabetes.
It is well known that the risks of both type 2 diabetes and B12 deficiency increase with age (1,2). Recent national data estimate a 21.2% prevalence of diagnosed diabetes among adults ≥65 years of age and a 6 and 20% prevalence of biochemical B12 deficiency (serum B12 <148 pmol/L) and borderline deficiency (serum B12 ≥148–221 pmol/L) among adults ≥60 years of age (3,4).
The diabetes drug metformin has been reported to cause a decrease in serum B12 concentrations. In the first efficacy trial, DeFronzo and Goodman (5) demonstrated that although metformin offers superior control of glycosylated hemoglobin levels and fasting plasma glucose levels compared with glyburide, serum B12 concentrations were lowered by 22% compared with placebo, and 29% compared with glyburide therapy after 29 weeks of treatment. A recent, randomized control trial designed to examine the temporal relationship between metformin and serum B12 found a 19% reduction in serum B12 levels compared with placebo after 4 years (6). Several other randomized control trials and cross-sectional surveys reported reductions in B12 ranging from 9 to 52% (7–16). Although classical B12 deficiency presents with clinical symptoms such as anemia, peripheral neuropathy, depression, and cognitive impairment, these symptoms are usually absent in those with biochemical B12 deficiency (17).
Several researchers have made recommendations to screen those with type 2 diabetes on metformin for serum B12 levels (6,7,14–16,18–21). However, no formal recommendations have been provided by the medical community or the U.S. Prevention Services Task Force. High-dose B12 injection therapy has been successfully used to correct the metformin-induced decline in serum B12 (15,21,22). The use of B12 supplements among those with type 2 diabetes on metformin in a nationally representative sample and their potentially protective effect against biochemical B12 deficiency has not been reported. It is therefore the aim of the current study to use the nationally representative National Health and Nutrition Examination Survey (NHANES) population to determine the prevalence of biochemical B12 deficiency among those with type 2 diabetes ≥50 years of age taking metformin compared with those with type 2 diabetes not taking metformin and those without diabetes, and to explore how these relationships are modified by B12 supplement consumption.
RESEARCH DESIGN AND METHODS
Design overview
NHANES is a nationally representative sample of the noninstitutionalized U.S. population with targeted oversampling of U.S. adults ≥60 years of age, African Americans, and Hispanics. Details of these surveys have been described elsewhere (23). All participants gave written informed consent, and the survey protocol was approved by a human subjects review board.
Setting and participants
Our study included adults ≥50 years of age from NHANES 1999–2006. Participants with positive HIV antibody test results, high creatinine levels (>1.7 mg/dL for men and >1.5 mg/dL for women), and prescription B12 injections were excluded from the analysis. Participants who reported having prediabetes or borderline diabetes (n = 226) were removed because they could not be definitively grouped as having or not having type 2 diabetes. We also excluded pregnant women, those with type 1 diabetes, and those without diabetes taking metformin. Based on clinical aspects described by the American Diabetes Association and previous work in NHANES, those who were diagnosed before the age of 30 and began insulin therapy within 1 year of diagnosis were classified as having type 1 diabetes (24,25). Type 2 diabetes status in adults was dichotomized as yes/no. Participants who reported receiving a physician’s diagnosis after age 30 (excluding gestational diabetes) and did not initiate insulin therapy within 1 year of diagnosis were classified as having type 2 diabetes.
Outcomes and follow-up
The primary outcome was biochemical B12 deficiency determined by serum B12 concentrations. Serum B12 levels were quantified using the Quantaphase II folate/vitamin B12 radioassay kit from Bio-Rad Laboratories (Hercules, CA). We defined biochemical B12 deficiency as serum levels ≤148 pmol/L, borderline deficiency as serum B12 >148 to ≤221 pmol/L, and normal as >221 pmol/L (26).
The main exposure of interest was metformin use. Using data collected in the prescription medicine questionnaire, those with type 2 diabetes were classified as currently using metformin therapy (alone or in combination therapy) versus those not currently using metformin. Length of metformin therapy was used to assess the relationship between duration of metformin therapy and biochemical B12 deficiency. In the final analysis, two control groups were used to allow the comparison of those with type 2 diabetes taking metformin with those with type 2 diabetes not taking metformin and those without diabetes.
To determine whether the association between metformin and biochemical B12 deficiency is modified by supplemental B12 intake, data from the dietary supplement questionnaire were used. Information regarding the dose and frequency was used to calculate average daily supplemental B12 intake. We categorized supplemental B12 intake as 0 μg (no B12 containing supplement), >0–6 μg, >6–25 μg, and >25 μg. The lower intake group, >0–6 μg, includes 6 μg, the amount of vitamin B12 typically found in over-the-counter multivitamins, and 2.4 μg, the daily amount the IOM recommends for all adults ≥50 years of age to consume through supplements or fortified food (1). The next group, >6–25 μg, includes 25 μg, the amount available in many multivitamins marketed toward senior adults. The highest group contains the amount found in high-dose B-vitamin supplements.
Statistical analysis
Statistical analysis was performed using SAS version 9.2 (SAS Institute, Cary, NC) and SUDAAN version 10.1 (Research Triangle Park, Durham, NC). Sample weights and variances were applied throughout the analysis to provide a representative sample of the U.S. population ≥50 years of age.
SAS survey procedures and SUDAAN “proc descript” were used to estimate means and proportions. SUDAAN “proc crosstab” was used to estimate the weighted prevalence adjusted for age, race, and sex. Tests of significance were performed using t tests for continuous variables and χ2 test for categorical variables. Population-attributable risk for metformin use on biochemical B12 deficiency was calculated from the cross-sectional data using the Fleiss equation (27).
Polytomous logistic regression was performed in SUDAAN using “proc multilog” with a trilevel B12-status outcome (vitamin B12 deficiency, borderline deficiency, and normal) to assess the association of previously identified risk factors with biochemical B12 deficiency and borderline deficiency in our study population. Risk factors previously identified for biochemical B12 deficiency were assessed as exposure variables along with metformin therapy in a full model and included age, race/ethnicity, sex, BMI (calculated as weight in kilograms divided by height in meters squared), and the use of proton pump inhibitors, H2 blockers, antacids, B12 supplements, alcohol, and tobacco (12,28). The final polytomous logistic model adjusted for age, BMI, insulin, and B12 supplement use. Alcohol use and smoking could not be included in the model as >60% of responses were missing.
RESULTS
In the final analysis, there were 575 U.S. adults ≥50 years of age with type 2 diabetes using metformin, 1,046 with type 2 diabetes not using metformin, and 6,867 without diabetes. The demographic and biological characteristics of the groups are shown in Table 1. Among metformin users, mean age was 63.4 ± 0.5 years, 50.3% were male, 66.7% were non-Hispanic white, and 40.7% used a supplement containing B12. The median duration of metformin use was 5 years. Compared with those with type 2 diabetes not taking metformin, metformin users were younger (P < 0.0001), reported a lower prevalence of insulin use (P < 0.001), and had a shorter duration of diabetes (P = 0.0207). Compared with those without diabetes, metformin users had a higher proportion of nonwhite racial groups (P < 0.0001), a higher proportion of obesity (P < 0.0001), a lower prevalence of macrocytosis (P = 0.0017), a lower prevalence of supplemental folic acid use (P = 0.0069), a lower prevalence of supplemental vitamin B12 use (P = 0.0180), and a lower prevalence of calcium supplement use (P = 0.0002). There was a twofold difference in the prevalence of anemia among those with type 2 diabetes versus those without, and no difference between the groups with diabetes.
The geometric mean serum B12 concentration among those with type 2 diabetes taking metformin was 317.5 pmol/L. This was significantly lower than the geometric mean concentration in those with type 2 diabetes not taking metformin (386.7 pmol/L; P = 0.0116) and those without diabetes (350.8 pmol/L; P = 0.0011). As seen in Fig. 1, the weighted prevalence of biochemical B12 deficiency adjusted for age, race, and sex was 5.8% for those with type 2 diabetes taking metformin, 2.2% for those with type 2 diabetes not taking metformin (P = 0.0002), and 3.3% for those without diabetes (P = 0.0026). Among the three aforementioned groups, borderline deficiency was present in 16.2, 5.5, and 8.8%, respectively (P < 0.0001). Applying the Fleiss formula for calculating attributable risk from cross-sectional data (27), among all of the cases of biochemical B12 deficiency, 3.5% of the cases were attributable to metformin use; and among those with diabetes, 41% of the deficient cases were attributable to metformin use. When the prevalence of biochemical B12 deficiency among those with diabetes taking metformin was analyzed by duration of metformin therapy, there was no notable increase in the prevalence of biochemical B12 deficiency as the duration of metformin use increased. The prevalence of biochemical B12 deficiency was 4.1% among those taking metformin <1 year, 6.3% among those taking metformin ≥1–3 years, 4.1% among those taking metformin >3–10 years, and 8.1% among those taking metformin >10 years (P = 0.3219 for <1 year vs. >10 years). Similarly, there was no clear increase in the prevalence of borderline deficiency as the duration of metformin use increased (15.9% among those taking metformin >10 years vs. 11.4% among those taking metformin <1 year; P = 0.4365).
Weighted prevalence of biochemical B12 deficiency and borderline deficiency adjusted for age, race, and sex in U.S. adults ≥50 years of age: NHANES 1999–2006. Black bars are those with type 2 diabetes on metformin, gray bars are those with type 2 diabetes not on metformin, and the white bars are those without diabetes. *P = 0.0002 vs. type 2 diabetes on metformin. †P < 0.0001 vs. type 2 diabetes on metformin. ‡P = 0.0026 vs. type 2 diabetes on metformin.
Weighted prevalence of biochemical B12 deficiency and borderline deficiency adjusted for age, race, and sex in U.S. adults ≥50 years of age: NHANES 1999–2006. Black bars are those with type 2 diabetes on metformin, gray bars are those with type 2 diabetes not on metformin, and the white bars are those without diabetes. *P = 0.0002 vs. type 2 diabetes on metformin. †P < 0.0001 vs. type 2 diabetes on metformin. ‡P = 0.0026 vs. type 2 diabetes on metformin.
Table 2 presents a stratified analysis of the weighted prevalence of biochemical B12 deficiency and borderline deficiency by B12 supplement use. For those without diabetes, B12 supplement use was associated with an ∼66.7% lower prevalence of both biochemical B12 deficiency (4.8 vs. 1.6%; P < 0.0001) and borderline deficiency (16.6 vs. 5.5%; P < 0.0001). A decrease in the prevalence of biochemical B12 deficiency was seen at all levels of supplemental B12 intake compared with nonusers of supplements. Among those with type 2 diabetes taking metformin, supplement use was not associated with a decrease in the prevalence of either biochemical B12 deficiency (5.6 vs. 5.3%; P = 0.9137) or borderline deficiency (15.5 vs. 8.8%; P = 0.0826). Among the metformin users who also used supplements, those who consumed >0–6 μg of B12 had a prevalence of biochemical B12 deficiency of 14.1%. However, consumption of a supplement containing >6 μg of B12 was associated with a prevalence of biochemical B12 deficiency of 1.8% (P = 0.0273 for linear trend). Similar trends were seen in the association of supplemental B12 intake and the prevalence of borderline deficiency. For those with type 2 diabetes not taking metformin, supplement use was also not associated with a decrease in the prevalence of biochemical B12 deficiency (2.1 vs. 2.0%; P = 0.9568) but was associated with a 54% reduction in the prevalence of borderline deficiency (7.8 vs. 3.4%; P = 0.0057 for linear trend).
Comparison of average daily B12 supplement intake by weighted prevalence of biochemical B12 deficiency (serum B12 ≤148 pmol/L) and borderline deficiency (serum B12 >148 to ≤221 pmol/L) among U.S. adults ≥50 years of age: NHANES 1999–2006

Table 3 demonstrates the association of various risk factors with biochemical B12 deficiency. Metformin therapy was associated with biochemical B12 deficiency (odds ratio [OR] 2.89; 95% CI 1.33–6.28) and borderline deficiency (OR 2.32; 95% CI 1.31–4.12) in a crude model (results not shown). After adjusting for age, BMI, and insulin and supplement use, metformin maintained a significant association with biochemical B12 deficiency (OR 2.92; 95% CI 1.28–6.66) and borderline deficiency (OR 2.16; 95% CI 1.22–3.85). Similar to Table 2, B12 supplements were protective against borderline (OR 0.43; 95% CI 0.23–0.81), but not biochemical, B12 deficiency (OR 0.76; 95% CI 0.34–1.70) among those with type 2 diabetes. Among those without diabetes, B12 supplement use was ∼70% protective against biochemical B12 deficiency (OR 0.26; 95% CI 0.17–0.38) and borderline deficiency (OR 0.27; 95% CI 0.21–0.35).
CONCLUSIONS
The IOM has highlighted the detection and diagnosis of B12 deficiency as a high-priority topic for research (1). Our results suggest several findings that add to the complexity and importance of B12 research and its relation to diabetes, and offer new insight into the benefits of B12 supplements. Our data confirm the relationship between metformin and reduced serum B12 levels beyond the background prevalence of biochemical B12 deficiency. Our data demonstrate that an intake of >0–6 μg of B12, which includes the dose most commonly found in over-the-counter multivitamins, was associated with a two-thirds reduction of biochemical B12 deficiency and borderline deficiency among adults without diabetes. This relationship has been previously reported with NHANES and Framingham population data (4,29). In contrast, we did not find that >0–6 μg of B12 was associated with a decrease in the prevalence of biochemical B12 deficiency or borderline deficiency among adults with type 2 diabetes taking metformin. This observation suggests that metformin reduces serum B12 by a mechanism that is additive to or different from the mechanism in older adults. It is also possible that metformin may exacerbate the deficiency among older adults with low serum B12. Our sample size was too small to determine which amount >6 μg was associated with maximum protection, but we did find a dose-response trend.
We were surprised to find that those with type 2 diabetes not using metformin had the lowest prevalence of biochemical B12 deficiency. It is possible that these individuals may seek medical care more frequently than the general population and therefore are being treated for their biochemical B12 deficiency. Or perhaps, because this population had a longer duration of diabetes and a higher proportion of insulin users compared with metformin users, they have been switched from metformin to other diabetic treatments due to low serum B12 concentrations or uncontrolled glucose levels and these new treatments may increase serum B12 concentrations. Despite the observed effects of metformin on serum B12 levels, it remains unclear whether or not this reduction is a public health concern. With lifetime risks of diabetes estimated to be one in three and with metformin being a first-line intervention, it is important to increase our understanding of the effects of oral vitamin B12 on metformin-associated biochemical deficiency (20,21).
The strengths of this study include its nationally representative, population-based sample, its detailed information on supplement usage, and its relevant biochemical markers. This is the first study to use a nationally representative sample to examine the association between serum B12 concentration, diabetes status, and metformin use as well as examine how this relationship may be modified by vitamin B12 supplementation. The data available regarding supplement usage provided specific information regarding dose and frequency. This aspect of NHANES allowed us to observe the dose-response relationship in Table 2 and to compare it within our three study groups.
This study is also subject to limitations. First, NHANES is a cross-sectional survey and it cannot assess time as a factor, and therefore the results are associations and not causal relationships. A second limitation arises in our definition of biochemical B12 deficiency. There is no general consensus on how to define normal versus low serum B12 levels. Some researchers include the functional biomarker methylmalonic acid (MMA) in the definition, but this has yet to be agreed upon (30–34). Recently, an NHANES roundtable discussion suggested that definitions of biochemical B12 deficiency should incorporate one biomarker (serum B12 or holotranscobalamin) and one functional biomarker (MMA or total homocysteine) to address problems with sensitivity and specificity of the individual biomarkers. However, they also cited a need for more research on how the biomarkers are related in the general population to prevent misclassification (34). MMA was only measured for six of our survey years; one-third of participants in our final analysis were missing serum MMA levels. Moreover, it has recently been reported that MMA values are significantly greater among the elderly with diabetes as compared with the elderly without diabetes even when controlling for serum B12 concentrations and age, suggesting that having diabetes may independently increase the levels of MMA (35). This unique property of MMA in elderly adults with diabetes makes it unsuitable as part of a definition of biochemical B12 deficiency in our specific population groups. Our study may also be subject to misclassification bias. NHANES does not differentiate between diabetes types 1 and 2 in the surveys; our definition may not capture adults with type 2 diabetes exclusively. Additionally, we used responses to the question “Have you received a physician’s diagnosis of diabetes” to categorize participants as having or not having diabetes. Therefore, we failed to capture undiagnosed diabetes. Finally, we could only assess current metformin use. We cannot determine if nonmetformin users have ever used metformin or if they were not using it at the time of the survey.
Our data demonstrate several important conclusions. First, there is a clear association between metformin and biochemical B12 deficiency among adults with type 2 diabetes. This analysis shows that 6 μg of B12 offered in most multivitamins is associated with two-thirds reduction in biochemical B12 deficiency in the general population, and that this same dose is not associated with protection against biochemical B12 deficiency among those with type 2 diabetes taking metformin. Our results have public health and clinical implications by suggesting that neither 2.4 μg, the current IOM recommendation for daily B12 intake, nor 6 μg, the amount found in most multivitamins, is sufficient for those with type 2 diabetes taking metformin.
This analysis suggests a need for further research. One research design would be to identify those with biochemical B12 deficiency and randomize them to receive various doses of supplemental B12 chronically and then evaluate any improvement in serum B12 concentrations and/or clinical outcomes. Another design would use existing cohorts to determine clinical outcomes associated with biochemical B12 deficiency and how they are affected by B12 supplements at various doses. Given that a significant proportion of the population ≥50 years of age have biochemical B12 deficiency and that those with diabetes taking metformin have an even higher proportion of biochemical B12 deficiency, we suggest that support for further research is a reasonable priority.
Acknowledgments
This study was supported by private donations to the Rollins School of Public Health of Emory University. No other potential conflicts of interest relevant to this article were reported.
L.R. researched the data, contributed to discussion, and wrote the manuscript. Y.P.Q. researched the data, contributed to discussion and reviewed and edited the manuscript. R.S.W., J.V.G., and G.P.O. contributed to discussion and reviewed and edited the manuscript. G.P.O. had full access to all of the data in the study and takes responsibility for the integrity of the data and accuracy of the data analysis.
The authors thank Dr. Theodore M. Johnson, II (Emory University School of Medicine) for suggesting they look at metformin’s role in biochemical B12 deficiency. Dr. Johnson did not receive compensation related to his contribution. The authors also thank the NHANES participants for making this study possible through their participation.