Metformin-Cimetidine Drug Interaction and Risk of Lactic Acidosis in Renal Failure: A Pharmacovigilance-Pharmacokinetic Appraisal

Metformin, a basic oral hypoglycemic agent widely used in the treatment of type 2 diabetes mellitus, has been shown to be effective and safe both as a monotherapy and in combination and has shown health economy benefits as well as in preventing cardiovascular complications (1). The main concern regarding the safety of metformin is lactic acidosis (LA) (1). While metformin-associated LA is extremely rare, 30% to 50% of patients with this complication die as a consequence (2).

Metformin-associated LA has mostly been observed in patients with renal failure (1,3). When these patients take metformin, the peak plasma metformin concentration increases threefold to 10-fold, resulting in metformin accumulation in the body, thus increasing the risk of LA (4,5). Because of this complication, the U.S. Food and Drug Administration (FDA) initially banned metformin use in patients with elevated (or even slightly elevated) serum creatinine. Subsequent observational evidence suggests that the incidence of LA in metformin users, including those with impaired renal function, is similar to the incidence in patients with type 2 diabetes mellitus not taking metformin (6,7). In April 2016, the FDA finally revised the contraindication label for metformin to allow its use in patients with mild-to-moderate renal failure (estimated glomerular filtration rate [eGFR] >30 mL/min/1.73 m²) (8). However, concerns remain regarding whether to prescribe metformin to patients with mild-to-moderate renal failure in clinical practice. A change in metformin pharmacokinetics caused by drug interactions may potentially increase the risk of developing LA in patients with type 2 diabetes mellitus who are often taking medications for multiple complications or comorbidities (9).

Metformin is neither metabolized in the liver nor excreted in the feces after oral absorption and is excreted unchanged in the urine via tubular transporter transport. Moreover, the transporters involved in its metabolism are mainly organic cation transporter 2 (OCT 2), multidrug and extrusion protein 1 (MATE 1), and MATE 2-K (9,10). Histamine H₂ receptor (H₂R) inhibitors are a class of oral antacids, of which cimetidine, ranitidine, and famotidine have been shown to inhibit OCT2 and MATEs (11–13). For example, when cimetidine is combined with metformin, it increases the plasma exposure to metformin by 50% and decreases renal clearance by 30% (14). Metformin's in vivo elimination time is prolonged when its plasma concentration exceeds 5 mg/L, a level that is closely associated with the risk of LA (15,16). A recent pharmacokinetic/pharmacodynamics study reported a more conservative threshold of 3 mg/L (17). Metformin plasma concentrations do not exceed the 3 mg/L threshold in healthy people taking metformin with cimetidine at usual doses (18). To our knowledge, no studies in the published literature have assessed the risk of LA development following regular doses of metformin and H₂R inhibitors in patients with mild-to-moderate renal failure.

Pharmacovigilance analysis of the FDA Adverse Event Reporting System (FAERS) data and utilization of physiologically based pharmacokinetic (PBPK) models have become valuable tools for evaluating drug side effects and predicting drug interactions (18,19). The combination of pharmacovigilance and PBPK techniques can better assess the safety of drugs and provide more comprehensive, accurate, and personalized medication guidance in clinical practice. In this study, we first obtained case data of LA from the FAERS database to assess the risk of LA caused by combining metformin and H₂R inhibitors. Furthermore, we used the PBPK model to predict changes in the pharmacokinetics of metformin after it was combined with cimetidine at conventional doses in patients with mild-to-moderate renal failure and explore its association with the risk of LA.

The FAERS database contains over 15 million reports from over 100 countries, with the majority originating from the U.S., making it one of the world's largest pharmacovigilance databases. The reports in FAERS are primarily submitted by consumers, health care professionals, and manufacturers. The available data/variables within this data set include demographic and administrative information, drug information, adverse drug event information, patient outcome information, information on the source of the reports, and drug therapy start and end dates. The information structure of the FAERS database follows the International Safety Reporting Guidelines published by the International Conference on Harmonization, and adverse events and medication errors are coded using terminology from the Medical Dictionary for Regulatory Activities (MedDRA). FAERS raw data are freely accessible to investigators for specific analyses of drugs and adverse drug reactions. It was not necessary to obtain institutional review board approval or informed consent for using FAERS data, because it is a publicly accessible, anonymized database.

This study analyzed 2012Q4 to 2022Q4 data, and the Medical Concept Extraction System for Unstructured Information Management Architecture (MedEx-UIMA) system was used to uniformly map drug names to generic drug names (20). The study medications were metformin and H₂R inhibitors (including cimetidine, ranitidine, and famotidine), none of which is a combination product, nor did the combination limit the drug role. LA was mapped to the MedDRA preferred term “lactic acidosis” (version 25.1, code 10023676). According to FDA recommendations for using the latest version of cases (21), duplicate reports were excluded prior to analysis. We extracted and analyzed the basic information of each report and used the reporting odds ratio (ROR) method as well as the Bayesian confidence progressive neural network method (22) to assess the risk of LA reporting for metformin alone. The risk signal of adverse drug event reporting was generated when the number of reports was more than five, and the lower limits of the 95% CI for ROR (ROR₀₂₅) and information component (IC) (IC₀₂₅) were considered statistically significant at >1 and 0, respectively. We further used the concomitant signal score (CSS) and omega shrinkage measure (Ω) to assess the reported risk of LA after the combination of metformin with H₂R inhibitors (22); CSS is based on the proportional reporting ratio (PRR) for drug interaction signal detection. A signal was considered when PRR >1, CSS >1, and Ω₀₂₅ (the lower limit of the 95% CI for Ω) >0 when the two drugs combined had at least five reports. Requiring a minimum of five reports helps reduce the probability of false positive signals, because safety signals driven by too few isolated reports are less reliable and reproducible; additionally, several studies have shown that the performance of different algorithms is largely comparable above this level (23–26). Based on restricted study periods and patients with renal failure, we performed a sensitivity analysis of the signal. As indicated in a previous study (27), in this study, MedDRA terminology was used to screen for the indication and adverse drug event information, to identify the population with renal failure. These data were then mapped to the MedDRA higher-level terms “renal failure and impairment” (version 25.1, code 10038443) and “renal failure complications” (version 25.1, code 10010180).

This mathematical model, which was based on the anatomy, physiology, biochemistry, and physical chemistry, simulates drug disposition in vivo by linking various tissues and organs of the body via blood fluid rings. The model has been widely applied in recent years, including in the development of individualized dosing regimens and prediction of drug interactions among combination drugs for special populations (28,29). PK-Sim and MoBi software (version 8.0, https://open-systems-pharmacology.org/) was used to create a publicly available metformin systemic PBPK model (18). The model describes the drug-gene interaction between metformin and the SLC22A2 808G>T gene, drug-drug interaction between cimetidine and metformin, and the metformin-cimetidine-gene interaction study in the normal population. The model has good scalability and is used to assess the day-to-day and interindividual variability of metformin pharmacokinetics (30). Our study further constructed a population model of chronic kidney disease (CKD) to test the effect of cimetidine on plasma exposure to metformin in patients with various degrees of renal failure. Because the SLC22A2 808G>T single-nucleotide polymorphism results in reduced metformin exposure, peak concentrations are reduced by approximately 13–20% (18). Therefore, we chose the SLC22A2 808GG genotype to construct the model. Specific settings for model parameters are detailed in Supplementary Tables 1–5. In this study, metformin and cimetidine were coadministered, and the peak plasma concentration of metformin was measured after 3 days of coadministration; metformin concentrations >3 mg/L were closely associated with LA risk. Here, we chose the peak plasma concentrations rather than mean steady-state concentration, to increase the sensitivity of the study. In this study, R 4.2.2 (R Institute for Statistical Computing, Vienna, Austria) was used for data processing and mapping.

Pharmacovigilance data can be found at https://fis.fda.gov/extensions/FPD-QDE-FAERS/FPD-QDE-FAERS.html. PBPK modeling data are derived from published studies.

In April 2016, the FDA amended the label to allow metformin to be used in patients with mild-to-moderate renal failure (8). This date was used to divide the study periods into two for comparative analysis (Table 1). From 2012Q4 to 2016Q1, there were 31 reported cases of LA associated with metformin in combination with H₂R inhibitors, with an average of 2.21 cases reported quarterly. The median age of the patients was 76 years (Q1–Q3: 59–82 years); females accounted for the majority (80.65%), and the most common reported source and country were health care professionals (80.65%) and Korea (35.48%), respectively. Patients with renal failure accounted for 29.03%, a single daily dose of metformin of 2,000 mg was administered the most (35.48%), and the most common combination with H₂R inhibitors was cimetidine (61.29%). In terms of reported outcomes, most cases (83.87%) resulted in hospitalization or prolonged hospitalization. From 2016Q2 to 2022Q4, 109 cases of LA associated with metformin combined with H₂R inhibitors were reported, with an average of 4.04 cases reported quarterly. The median age of patients was 70 years (Q1–Q3: 64–78 years). The percentages of male and female were 39.45% and 49.54%, respectively, and the most common reported source and country were health professionals (95.41%) and the U.K. (37.61%). Patients with renal failure accounted for 66.97%; single daily dose of metformin was up to 2,000 mg (25.69%); the most common combination of H₂R inhibitors was with ranitidine (55.05%). In terms of reported outcomes, most (92.66%) resulted in hospitalization or prolonged hospitalization.

During the overall study period (2012Q4 to 2022Q4), there was a signal of LA associated with metformin monotherapy in the overall population (ROR₀₂₅ 136.24; IC₀₂₅ 6.27). In further sensitivity analyses, the signal persisted after restricting the study period to include only patients with renal failure. The risk signal assessment and sensitivity analysis for LA reporting associated with metformin alone are detailed in Supplementary Table 6. No LA signals were detected in the overall study period (2012Q4 to 2022Q4) for metformin combined with H₂R inhibitors in the overall population (Ω₀₂₅ −2.60; lower 95% CI for PRR [PRR₀₂₅] 5.95; CSS 0.08). In the sensitivity analysis, LA signals associated with metformin in combination with cimetidine were detected only in model 2 (restricted study period, 2012Q4 to 2016Q1) (Ω₀₂₅ 1.14; PRR₀₂₅ 122.32; CSS 2.20). Reported risk signal evaluations and sensitivity analyses for LA associated with metformin in combination with H₂R inhibitors are detailed in Table 2. To compensate for the limitation of incomplete information on renal failure cases in the spontaneous reporting database and further validate the LA signal associated with metformin combined with H₂R inhibitors as mentioned above, in this study, we constructed a metformin-cimetidine interaction PBPK model in people with different CKD stages to assess the effect of cimetidine on metformin plasma exposure in the mild-to-moderate renal failure state to determine the risk of LA.

According to the drug label, metformin was administered as follows: 1) 500-mg b.i.d. tab per os (po); 2) 850-mg q.d. tab po; 3) 1,000-mg q.d. tab po; and 4) 1,000-mg b.i.d. tab po. Cimetidine was administered as follows: 1) 300-mg q.i.d. tab po; 2) 400-mg b.i.d. tab po; and 3) 800-mg q.d. tab po (Fig. 1). The simulation results showed that, in CKD stage 1, when metformin was administered at 1,000 mg b.i.d. and cimetidine at 300 mg q.i.d. or 400 mg b.i.d., the peak plasma concentration of metformin exceeded 3 mg/L. In CKD stage 2, when metformin was administered at 1,000 mg b.i.d. and cimetidine at either 300 mg q.i.d., 400 mg b.i.d., or 800 mg q.d., the peak plasma concentration of metformin exceeded 3 mg/L. In CKD stage 3a, besides metformin (850 mg q.d.) and cimetidine (800 mg q.d.) combination, the peak plasma concentration of metformin in other combination regimens exceeded 3 mg/L. In CKD stage 3b, the peak plasma concentration of metformin in different cimetidine combination dose regimens exceeded the threshold of 3 mg/L. PBPK modeling of CKD stages 4 and 5 showed peak plasma metformin concentrations for all combined regimens surpassing levels seen in stage 3b (Supplementary Figs. 1 and 2). In addition, we predicted the effect of cimetidine on the fractional urinary excretion of metformin in different CKD stages and showed that the fractional urinary excretion of metformin decreased when metformin was combined with cimetidine, regardless of the CKD stage (Supplementary Figs. 3–8). These results suggest a potential risk of LA in patients with mild-to-moderate renal failure taking metformin concomitantly with cimetidine.

Based on the above predictions, we attempted optimizing the metformin-cimetidine combination regimens for patients with CKD stages 3a and 3b to reduce the risk of potential LA (Fig. 2). When patients with renal failure had CKD stage 3a, the following combination regimens could achieve peak plasma metformin concentrations of <3 mg/L: metformin (500 mg b.i.d.) and cimetidine (200 mg b.i.d.), metformin (500 mg b.i.d.) and cimetidine (300 mg b.i.d.), and metformin (500 mg b.i.d.) and cimetidine (400 mg q.d.). In CKD stage 3b, these combination regimens were metformin (250 mg b.i.d.) and cimetidine (200 mg b.i.d.), metformin (250 mg b.i.d.) and cimetidine (300 mg b.i.d.), and metformin (250 mg b.i.d.) and cimetidine (400 mg q.d.).

Although metformin interactions are largely considered safe, impaired renal function may increase susceptibility, resulting in dangerous metformin levels (9). Clinical trials on interactions in renal failure remain scarce because of persistent challenges. By combining pharmacovigilance mining and model-based pharmacokinetics, we found that conventional doses of metformin-cimetidine may cause a risk of LA in mild-moderate renal failure. Our work enhances the understanding of variable individual responses to metformin.

A study conducted in 2019 examined 559 metformin-related LA cases in 247 articles and found that almost all cases had other risk factors (including liver or kidney dysfunction, dehydration, and infection) and concluded that metformin may not be effective at therapeutic doses or might only assist in the development of LA (3). Similarly, our pharmacovigilance evidence suggests that metformin was closely related to LA risk and that renal failure and drug interactions might play an important role. We found a clear increase in the proportion of patients with renal failure, among those reporting LA associated with metformin combined with H₂R inhibitors, following the April 2016 FDA approval of metformin for patients with mild-to-moderate renal failure. In our study on the reported risk signal for LA associated with metformin and H₂R inhibitors, we were surprised to discover that a significant LA signal associated with metformin and cimetidine only appeared during the study period from 2012Q4 to 2016Q1 in the overall population; metformin combined with other H₂R inhibitors (famotidine and ranitidine), as well as in the renal failure subgroup, exhibited a lack of LA signals. However, potential biases inherent in spontaneous report studies must be taken into consideration. Metformin use in renal failure follows specific eGFR as recommended by FDA guidance (8). FAERS lacks laboratory eGFR values. Identifying renal failure cases via MedDRA mapping imprecisely defines dysfunction severity compared with staging by typical diagnostic codes or eGFR. The reliability of using MedDRA codes for case identification also depends on reporters’ discernment of cases. Despite most reporters in our study being health care professionals, the possibility of misidentification must be considered given the spontaneous nature of the data. Additionally, extensive attention to a real or perceived safety issue in the medical community or in public media may disproportionately increase reporting rates for that drug-event pair. These inherent limitations of the FAERS data could lead to overreporting or underreporting biases for certain drug-event pairs. Finally, FAERS data only contain reported adverse events without a corresponding comparator, and associations based on FAERS data cannot be considered causal. Therefore, while FAERS supports safety signal detection, confirmatory studies are needed to fully characterize risks.

PBPK modeling can optimize dosing regimens by simulating pharmacokinetics during renal failure. Our model, built on a validated metformin model (18), predicted how cimetidine affects metformin exposure in mild-moderate renal failure, to verify pharmacovigilance signals. Previous studies have only found that cimetidine increases metformin plasma levels via inhibited renal clearance (11,14,18), but they could not assess LA risk. Determining a reasonable metformin concentration threshold for LA, despite intricacies like individual variation, is key given the high mortality. Plasma metformin concentrations >5 mg/L have long been considered to be strongly associated with the risk of high LA development (15,16). In studies using population pharmacokinetic models to predict metformin plasma concentrations in patients with LA before admission, it was determined that, in most of such patients with impaired renal function (92%), the plasma metformin concentrations did not exceed 5 mg/L before admission, and the proportion decreased to 76% when the threshold was decreased to 2.5 mg/L (3). It is important to note that this value of 2.5 mg/L was based on just one study (31). In 2020, Kuan et al. (17) included more metformin-related LA clinical studies to establish the relationship between plasma metformin concentrations and severe hyperlactatemia (lactate concentration >5 mmol/L) by population pharmacokinetic methods, and updated 3 mg/L as the threshold concentration for assessing the risk of LA development. It should be noted that, for the metformin-associated LA cases included in this study, generally, the time between taking metformin and admission for LA was not reported. Furthermore, the appropriate time for reporting plasma metformin concentration to assess non-LA cases was also unclear (17). Therefore, we conservatively assumed that peak concentrations would be 3 mg/L in this study, to improve the sensitivity of the prediction.

The 2023 American Diabetes Association (32) and 2022 Kidney Disease Improving Global Outcomes (KDIGO) (33) guidelines recommend reduced metformin dosing with eGFR below 45 mL/min/1.73 m² and contraindicate use below 30 mL/min/1.73 m². Nevertheless, the PBPK prediction results in this study suggest that patients with CKD stages 1–3 may be at risk for LA when taking metformin at conventional doses in conjunction with cimetidine. Three previous studies of LA caused by the combination of metformin and cimetidine that are sufficiently reliable to support this prediction have been reported (34–36). A specific dose reduction regimen for metformin and cimetidine was also determined based on the PBPK model in patients with CKD stages 3a and 3b, to reduce the risk of potential LA. It is noteworthy that the application of the PBPK model to transporter-based drug interaction studies has some limitations as well. Primarily, the limitation is that the activity and expression levels of transporters may vary greatly in different individuals and disease states, and, for some transporters, their specific functions and interaction mechanisms are not yet completely understood (28). As a result of this study, it is difficult to fully parameterize the transporters (OCT2 and MATE) under the different CKD stages as well as to capture and accurately simulate the kinetic profiles of metformin and cimetidine based on these transporters. Thus, while the PBPK model can provide quantitative predictions, its prediction results are still influenced by model assumptions and parameter settings and require further experimental validation and support of further clinical studies.

Summarily, the combined analysis of pharmacovigilance-PBPK in this study demonstrated LA risk in patients with mild-to-moderate renal failure who were taking combined metformin and cimetidine at conventional doses. However, these results still need further clinical studies to confirm.

Acknowledgments. The authors thank FAERS.

Funding. This study was funded by the National Natural Science Foundation of China (grant no. 82100872), Natural Science Foundation of Fujian Province (grant no. 2022J01283), and High-level Personnel Research Start-up Funding of Fujian Medical University (grant no. XRCZX2020003).

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. W.X., J.L., J.Z., and Y.Z. were responsible for study design, data acquisition and interpretation, statistical analysis, and writing and editing of the manuscript. C.K. and W.L. aided in data acquisition and interpretation and statistical analysis. All authors approved the final manuscript. Y.Z. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.