Study design, patient recruitment and selection

This observational cross-sectional study was conducted at the diabetes clinics of departments of Internal Medicine at Steve Biko Academic Hospital and Kalafong Hospital in Pretoria, South Africa between August 2012 and January 2013. A consecutive sampling method was used. A provisional list of patients eligible to participate in the study was made from the records. On clinic days, all eligible patients were approached and asked to participate in the study. Those who accepted to participate signed the informed consent after reading the patient information leaflet.

Patients who met the following criteria were eligible to participate: willingness to participate in the study by signing the informed consent, diagnosed T2DM, use of metformin for six months or more and ability to read and write in English. Patients were excluded from the study if one or more of the following was present: pernicious anaemia, alcoholism, gastrectomy, gastric bypass surgery, pancreatic insufficiency, malabsorption syndromes, chronic giardiasis, surgery involving small intestine or HIV infection. Intake of vitamin B12 or any multivitamin preparation during the past 6 weeks, estimated Glomerular Filtration Rate (e-GFR) < 50 mL/min/1.73 m² based on Cockcroft-Gault formula or diagnosed vitamin B12 deficiency were also exclusion criteria.

Variables and data collection

The following information was obtained from participants’ records: age, sex, ethnic group, date of diagnosis of T2DM, most recent HbA1c (glycated hemoglobin) level, duration of metformin use, dosage of metformin, smoking status, eGFR, Body-Mass Index (BMI), use of gastric acid suppressants or low-dose acetylsalicylic acid (both were linked to vitamin B12 deficiency [17, 18]), coffee consumption, presence and history of other medical conditions and alcohol consumption or any substance abuse. Participants were considered alcohol users when they consumed one drink or more per day. Coffee consumption was defined as daily consumption of one cup or more.

Serum vitamin B12 was measured in Ampath Laboratories by UniCel DxI 800 instrument (Beckman Coulter, United States). Vitamin B12 deficiency was defined as levels <150 pmol/L [3, 5, 6, 8].

Neuropathy Total Symptom Score-6 (NTSS-6) questionnaire was used to grade peripheral neuropathy. The investigator asked each participant about frequency and intensity of six peripheral neuropathic symptoms, and used the answers to complete the questionnaire. The sum scores for each patient were obtained (possible range between 0.00 and 21.96). Grading the frequency and intensity of symptoms was based on the definitions of Bastyr et al. [19].

To reduce occurrence of bias, scoring of peripheral neuropathy was performed by the same investigator prior to accessing any sort of data that gives information on metformin dose and/or duration of use. Serum samples were sent for analysis after completing data extraction and peripheral neuropathy scoring.

Statistical analysis

Sample size calculations showed that recruiting 120 participants would give a 7 % margin of error (95 % CI) around an estimated 20 % prevalence. STATA version 12 was used for statistical analysis. Two-sided P values ≤0.05 were considered statistically significant. Means (± standard deviations), actual numbers and percentages were used to describe data.

The prevalence of vitamin B12 deficiency was determined as the percentage of participants with vitamin levels <150 pmol/L. Relationship between vitamin B12 and peripheral neuropathy was examined by two methods. The first utilized chi-square test to investigate the relationship between the two binary variables created from the continuous variables of vitamin B12 (deficient for levels < 150 pmol/L, otherwise normal) and peripheral neuropathy (present for NTSS-6 scores > 6, otherwise absent). The second method used Spearman’s rank correlation coefficient (rho) to examine bivariate relationship between the two continuous variables of vitamin B12 levels and NTSS-6 scores.

Demographic and clinical characteristics of vitamin B12-deficient participants were compared to those with normal vitamin levels. Normality of continuous variables was examined by Shapiro-Wilk test. Unpaired t-test (for normally distributed data) and Mann–Whitney test (for non-normally distributed data) were used to compare the continuous variables. Chi-square test and Fisher’s exact test (when expected cell frequency was less than five) were used to compare categorical variables. Multiple logistic regression was performed where vitamin B12 status represented the binary dependent variable and all the variables with P values ≤0.25 in the univariate analysis were taken as independent variables. The variables metformin duration, metformin cumulative dose (product of total daily dose and duration of use) and T2DM duration were likely to be highly correlated. Three initial models, each contained one of these variables, were made to avoid the impact of the multicollinearity. Backwards stepwise manner was then used to reduce initial models to a final model, from which risk factors significantly associated with vitamin B12 deficiency were determined.

Prevalence of vitamin B12 deficiency

Relationship between vitamin B12 and peripheral neuropathy

The present study found no statistically significant difference in presence of neuropathy between those with normal and deficient vitamin levels (36.8 % vs. 32.3 %, P = 0.209). The levels of vitamin B12 and the NTSS-6 scores were not correlated (Spearman’s rho = 0.056, P = 0.54). Our results are in-line with those of a recently published study which reported that metformin use was not associated with the presence of diabetic peripheral neuropathy in T2DM patients [20]. Our findings are also in concordance with the results of the cross-sectional study of Chen et al. which revealed no significant differences between metformin users and non-users when neuropathy status was assessed by both objective (monofilament and neurothesiometry) and relatively subjective (questionnaires) measures [14]. Biemans et al. also reported no differences in occurrence of neuropathy in vitamin B12-deficient and –normal T2DM patients receiving metformin [15]. Our results are in contrast with the case–control study of Wile and Toth who reported more severe neuropathy among T2DM patients on metformin compared to non-metformin group [16]. Their results were obtained when the neuropathy severity was assessed by Toronto Clinical Scoring System (TCSS) and Neuropathy Impairment Score. The more objective electrophysiological tests showed no statistically significant differences among the two groups. Singh et al. also reported more severe neuropathy as assessed by TCSS in metformin-treated T2DM patients [7]. Our results also contrasts with the data of the cross-sectional study of de Groot-Kamphuis et al. who reported lower prevalence of neuropathy in metformin-treated T2DM patients compared to the non-metformin group [8].

Many animal studies have recently reported the glycemic control-independent neuroprotective impact of metformin. Mao-Ying et al. found that metformin reduced peripheral nerve endings loss and exerted protective effect against chemotherapy-induced peripheral neuropathy (CIPN) in mice [21]. Both peripheral neuropathy and CIPN share the glove-and-stocking distribution nature of sensory symptoms that includes parasthesia, dysthesia and pain [21]. Animal studies have also recently shown that metformin reversed induced neuropathic pain and protected against nerve injury [22], protected against neuronal apoptosis induced by ethanol [23], inhibited neuronal apoptosis in cortical cells [24], stimulated neurogenesis [25], and promoted neurogenesis following middle cerebral artery occlusion in mice [26]. Considering such impact of metformin, there can be two possible lines through which the medication affects the neuropathy status, excluding that related to glycemic control. One involves a positive impact through neuroprotective mechanisms, while the other induces neuropathy by enhancing vitamin B12 deficiency. Absence of association between vitamin B12 and peripheral neuropathy in our study may not thus totally preclude the potential of the medication to precipitate or worsen neuropathy through vitamin B12 deficiency. This theory may also explain the contradictory nature of results obtained by different studies.

Judicious interpretations of evidence around peripheral neuropathy as a clinical consequence of metformin-induced vitamin B12 deficiency may require considering certain medication-related features with possible significant methodological impacts. Being the cornerstone of the management of T2DM, it should be uncommon to encounter T2DM patients who are not on metformin. In observational studies that compare metformin users and non-users, T2DM patients who are not on metformin may hence be inherently different from those taking the medication. Being not on metformin is itself an abnormality with potential to make obtaining similar study groups a strenuous task. Having a control group may, therefore, negatively influence the validity in studies aiming at investigating the impact of metformin-induced vitamin B12 deficiency on neuropathy. The theory of the possible neuroprotective impact of metformin commented above may also lead to the same methodological conclusion. The possibility exists that neuroprotective effect of the medication group dilutes or predominates over its neuropathy-generating impact which is mediated by vitamin B12 deficiency. Comparing peripheral neuropathy in the control and metformin groups may thus produce distorted results that do not truly reflect the contribution of the metformin-induced vitamin B12 deficiency to the neuropathy status. From this perspective, designs that compare peripheral neuropathy among vitamin B12-deficient and –normal metformin-treated patients can theoretically result in more valid findings.

Risk factors for vitamin B12 deficiency

Low HbA1c was a significant risk factor for vitamin B12 deficiency in the final (Table 5) (OR = 0.97, 95 % CI: 0.95 to 0.99, P = 0.003) as well as the three initial models (Table 4). Kang et al. reported, but did not explain, similar results [10]. They reported stronger association between HbA1c and vitamin B12 deficiency (OR = 0.74, CI: 0.56 to 0.99). Their study included T2DM patients who were either on metformin plus insulin or metformin plus sulfonylurea. Such stricter inclusion criterion may explain the difference in magnitude of OR between the present study and Kang et al’s. We think the association between HbA1c and vitamin B12 status can be, at least partly, explained by compliance to metformin treatment. The well-documented gastrointestinal adverse metformin reactions, adding to the high prevalence of gastrointestinal complications in diabetes patients, may propose higher risk of non-compliance among metformin users. Higher doses of metformin may result in more prominent gastrointestinal adverse reactions and higher rates of non-compliance. Our theory is that patients with poor glycemic control (higher HbA1c) may have poor compliance to metformin and thus higher vitamin B12 levels. Raw data from the present study revealed that 21.5 % of participants were on the maximum daily dose of metformin, possibly supporting our proposed theory on the interaction between dose, non-compliance, vitamin B12 levels and glycemic control. Considering the daily dose of metformin and renal function as direct explanatory factors of the relationship between HbA1c and vitamin B12 is statistically inappropriate as the relationship existed despite adjusting for metformin dose and eGFR (Table 4).

Black South African descent was a significant protective factor for vitamin B12 deficiency in the final model (OR = 0.34, 95 % CI: 0.13 to 0.92, P = 0.033) (Table 5) as well as the three initial models (Table 4). This is the first study to report ethnic differences in vitamin B12 levels among metformin-exposed T2DM patients. Reinstatler et al. found no statistically significant differences in vitamin levels among black, white and Hispanic metformin-treated patients in the United States [4]. Higher vitamin B12 levels in black compared to white general populations were previously reported [27, 28]. Higher levels of the vitamin binding proteins transcobalamin II and haptocorrin in black individuals were described in South African settings, and explained their relatively elevated vitamin B12 levels [29].

Daily dose of metformin has shown borderline significance as a risk factor for vitamin B12 deficiency in the final model (OR = 1.96, 95 % CI: 0.99 to 3.88, P = 0.053) (Table 5). The association between the medication daily dose and the deficiency is sensible, and was previously reported by multiple regression analyses of many studies [8, 11, 30].

Limitations

Our study was conducted in tertiary academic specialist clinics, raising the possibility of over-representation of patients with complicated T2DM. Such patients may tend to have higher doses and durations of use of metformin. The study has only measured serum vitamin B12 levels to assess the vitamin status. Current recommendations suggest adding methylmalonic acid or homocysteine tests to better assess the intracellular status of the vitamin [31]. The absence of data on the compliance to metformin is also a limitation of this study. Compliance can have an impact on both the response to metformin and the levels of vitamin B12. To conclusively answer the secondary aim of investigating the relationship between peripheral neuropathy and vitamin B12 deficiency, a much larger sample size would have been needed to show a clinically important difference as small as 10 % to be statistically significant. NTSS-6 questionnaire, which is relatively subjective and symptom history-dependent, was the only used tool to assess neuropathy in the study. However, the questionnaire has been validated and was found suitable for evaluating peripheral neuropathy in clinical trials [19].