Risk of acute kidney injury and survival in patients treated with Metformin: an observational cohort study

Metformin is a biguanide which has been used as first line therapy for the treatment of type 2 diabetes in the United Kingdom since the 1960s. In addition to its effectiveness as a treatment for type 2 diabetes, there is increasing evidence of its beneficial cardiovascular effects [1‐6]. There is, however, considerable debate regarding its use in patients with chronic kidney disease (CKD), defined as an estimated Glomerular Filtration Rate (eGFR) of less than 60 ml/min/1.73². Guidelines pertaining to metformin prescribing and renal impairment vary. The United States Food and Drug Administration (FDA) previously stated that metformin was contraindicated in males with a serum creatinine above 132.6 μmol/l and females above 123.8 μmol/l or abnormal creatinine clearance [7]. However, following a recent review of the literature, these recommendations were revised to indicate that metformin can now be used safely in patients with mild renal impairment (eGFR 45- 60 ml/min/1.73²) and in some with moderate renal impairment (eGFR <45 ml/min/1.73²) [8]. The National Institute for Health and Care Excellence (NICE) in the UK allows prescribing of metformin in patients with an eGFR of less than 60 ml/min/1.73² but recommends that the dose is reduced in patients with an eGFR of less than 45 ml/min/1.73² (or serum creatinine of greater than 132.6 μmol/l) and metformin discontinued if eGFR is less than 30 or serum creatinine greater than 150.3 μmol/l [9]. Current contraindications therefore exclude its use in a large proportion of patients with CKD [10]. Following the FDA’s announcement, the European Medicines Agency (EMA) have also now announced that metformin-containing medicines can be used in patients with moderately reduced renal function (eGFR 30 to 59 ml/min/1.73²) [11].

The advent of a universally agreed serum creatinine based definition for AKI allows identification of even mild (Stage 1) AKI without relying on hospital coding systems [12] allowing assessment of renal safety in patients with CKD.

The present study therefore aimed to examine whether current metformin use was associated with an increased rate of AKI as a proxy for lactic acidosis compared to non-use and to establish whether metformin exposure affected survival in patients with AKI.

Between 1st of January 2004 and 31st of December 2013, there were 25,148 patients with Type 2 diabetes in the Tayside region of Scotland forming the study cohort. 14,622 of these patients were treated with metformin at some point during the study period and 10,526 patients were never on metformin during the study period. Total person-time was 126,904 person years (pyrs) of which 60,738 pyrs were exposed to metformin. Median duration of follow-up was 8.1 years IQR (5.2-10.0). Demographic characteristics of the study population at the study midpoint (1st July 2008) are described in Table 1. The analysis conducted here is a time updated analysis of rates in person-time periods (rather than persons) associated with metformin exposure compared to person-time periods without exposure therefore we show the characteristics of person-time periods associated with current metformin exposure compared to person-time periods without current exposure in Table 2. Metformin exposure periods occurred at younger ages, higher HbA_1c, longer duration of diabetes and were associated with higher eGFR, p < 0.0002.

During follow-up, 4944 people had at least one episode of AKI during the study period. Of these AKI events, 4352 were Stage 1, 140 Stage 2 and 452 Stage 3 AKI events. Only person-time periods to the first event were included in analyses. The characteristics at entry into the study of those persons who had an event at any time during follow up versus those who did not are shown in Table 3. Patients who had any AKI event had lower eGFR at entry to the study than those who never developed an event (median 58 ml/min/1.73² IQR 46-70 vs 73 ml/min/1.73² IQR 61-88). AKI was strongly associated with pre-admission eGFR (Table 4).

Overall incidence of AKI was 39.0 cases/1000 pyrs. There were 32.4 cases of AKI/1000pyrs in person-time periods with current metformin use compared to 44.9 cases/1000pyrs during person-time periods in which metformin was not being used. Severity of diabetes was adjusted for by adjusting for diabetes duration and number of diabetes medicines. Current metformin use was associated with a lower AKI rate, HR 0.76 (95% CI 0.70, 0.82, p < 0.0001) after adjustment for sex, current age, current diabetes duration, and current number of diabetes drugs. When further adjusted for baseline renal function, current metformin use was no longer associated with any difference in AKI rate, HR 0.94 (95% CI 0.87, 1.02, p = 0.15). After including current or “ever” metformin use, person-time periods with current or “ever” exposure to metformin were associated with higher rates of AKI than person-time periods with no current or prior exposure; HR 1.13 (95%CI 1.05, 1.22, p = 0.001).

At study midpoint only 1.5% of people with an eGFR <30 were taking metformin compared to 72% of people with an eGFR >60. Clearly since current eGFR is a strong correlate of metformin exposure and predicts future AKI events, disentangling drug effects from allocation effects is difficult and negative confounding by allocation could explain the hazards ratio (HR) observed. In addition to adjusting for baseline eGFR in the Cox regression, we also examined whether, among strata of person-time periods defined by baseline eGFR, whether there was any evidence of an effect of metformin. We found no evidence of increased rates of AKI- even where baseline eGFR was <30, the rates of events in person-time periods exposed to metformin were not greater than those unexposed to metformin (Table 5). Furthermore we conducted a sensitivity analysis where we adjusted for lagged eGFR – i.e. for each time periods we included a time updated covariate which was eGFR 6 months prior to that time period as distinct from the baseline eGFR as that may have long preceded that person-time period. This analysis allows for a fuller adjustment of the effect of prevailing eGFR on the probability of metformin exposure. In this analysis the HR for current metformin exposure was 0.85 (95%CI 0.78, 0.92), p < 0.0001.

Note that since it could be argued that the probability of subsequent exposure to metformin (either for the first time or repeat prescription) may be altered when an AKI event has occurred, person- time was right censored at the first AKI event for this part of the analyses.

In this large cohort of over 25,000 patients with Type 2 diabetes, we found no evidence of any adverse effect of current metformin use on AKI incidence or on survival following AKI across different eGFR strata. We have shown risk of AKI does not differ during person-time periods where there was no current metformin exposure compared to those in which there was current metformin exposure.

Strengths of our study include the large cohort size with high metformin use. We identified our cohort using the SCI-Diabetes system. This system is used to record all diabetes clinical care in Scotland thereby capturing 99% of patients with diabetes [16]. Furthermore, prescribing data was obtained from pharmacy dispensed prescriptions (covering all pharmacies in the Tayside region) rather than prescriptions alone increasing reliability that the medicines are actually being taken. AKI was defined using the KDIGO creatinine based criteria from laboratory data that are only available at regional level [12].

Observational pharmaco-epidemiology studies have to contend with various potential biases or confounders. The most difficult bias for this study is negative confounding by allocation wherein those with a lower eGFR are more likely to develop AKI but less likely to be exposed to metformin. We attempted to address this by adjustment for eGFR in a number of ways and by conducting analyses by different strata of eGFR. None of the analyses showed any suggestion of an adverse effect. Since there are unlikely to be randomised trials of allocating metformin to those with low eGFR to assess safety, this observational study provides useful reassurance that we cannot see any adverse safety signals in the data that would warrant a change to the current guidelines. If anything the data suggest that the current guidelines may be too stringent and could be relaxed further. We used multivariable analysis to adjust for important clinical confounders in view of the difference between the metformin users and the non-metformin users. Furthermore, our data may not be applicable to other populations as the Tayside region of Scotland lacks ethnic diversity and is predominantly of European ancestry. In addition, we were unable to use direct data on co-morbidity due to problems of under reporting [17]. We therefore used number of prescribed medications prior to admission which has been shown to correlate well with co-morbidity [18]. A further limitation of our study was that we were unable to account for over the counter NSAID use in our population.

Reported incidence rates of metformin associated lactic acidosis are in fact very low further supporting its more widespread use. A Cochrane review showed no cases of fatal or non-fatal lactic acidosis in 70, 490 patient- years of metformin use [19]. Incidence rates of MALA were 3.3 per 100,000 person-years in a nested case control analysis of 50,048 patients from the United Kingdom General Practice Research Database compared with 4.8 per 100,000 person-years among sulphonylurea users [20]. Several studies have shown use of metformin in patients with renal impairment thereby suggesting non-adherence to guidelines [21‐25]. In a subgroup analysis of 6960 patients with eGFR 45-60 ml/min/1.73² from a Swedish cohort of 51, 675 patients, there was a significantly lower risk of acidosis/ serious infection in patients on metformin and a slightly higher but insignificant risk in the subgroup of 2044 patients with eGFR of 30-45 ml/min/1.73² [2]. A systematic review examining patients with CKD demonstrated the safety of metformin in patients with stable CKD (eGFR >30 ml/min/1.73²) including 150,000 patients. This concluded that rates of lactic acidosis were low and similar to that of sulphonylureas [26]. A more recent systematic review showed similar results suggesting that risk of lactic acidosis is low and indistinguishable from background risks and that available evidence supported expansion of metformin in patients with CKD with appropriate dose reduction and follow-up [27].

Despite almost 60 years of metformin use, there is a lack of pharmacokinetic studies of metformin in patients with CKD. Metformin inhibits the mitochondrial respiratory chain. This impairs energy metabolism aerobically leading to increased anaerobic metabolism thereby generating lactate [28]. Metformin is eliminated unchanged by the kidney therefore it can accumulate in patients with renal impairment increasing the risk of lactic acidosis. In a small study of 15 patients administered 850 mg of metformin, mild CKD (creatinine clearance, 60-90 mL/min) was associated with 23% to 33% reductions in medication clearance and moderate CKD(30-60 mL/min) with 74% to 78% reductions [29]. Whilst safety of metformin has been shown in patients with mild renal impairment, dose reduction is advisable. Further pharmacokinetic studies are required to establish safe dosing in patients with moderate to severe renal impairment.

It has previously been shown in animal models that metformin exerts a nephrotoxic effect [30]. However, there is a lack of data in humans to support this. A recent retrospective cross sectional study showed that in a small number of patients, with previously normal renal function, admitted with AKI, metformin use was associated with both lactic acidosis and worsening AKI in a dose related manner. They did not, however, demonstrate a difference in survival or need for RRT [31].

There is increasing evidence that AKI is associated with increased mortality even with stage 1 [32‐34] AKI. Our data found that survival rates were higher in patients with AKI previously treated with metformin compared to patients with AKI not previously on metformin. This is in line with several previous observational studies in which lower event rates in those exposed to metformin persists even in subgroups with impaired renal function [2, 4]. A Swedish study of 51,675 people with Type 2 diabetes found that those treated with metformin monotherapy had a lower risk of fatal or non-fatal cardiovascular disease events and all-cause mortality [2]. In a multicentre cohort of 19,691 diabetic patients with atherothrombosis, mortality rates were lower in patients treated with metformin compared to those not on metformin. This association persisted in a subgroup of patients with CKD (eGFR 30-60 ml/min/1.73²) [4]. A recent review of metformin use in patients with kidney disease recommended that metformin could be used in patients with an eGFR of 30-60 ml/min/1.73² with dose reduction [27]. Following this review the FDA have revised their recommendation stating that it can be used in mild renal impairment and in some with moderate renal impairment [8]. It is difficult to ascertain whether the survival benefit we have shown is due to selection bias in that patients who receive metformin may be a different cohort in that they are fitter. However, we have shown that in our data patients on metformin actually had a longer duration of diabetes and higher HbA_1c levels suggesting that the survival benefit may be due to the known beneficial effects of metformin on vascular endothelium rather than selection bias.

Whilst renal safety of metformin is important in its own right, we were also motivated to analyse AKI rates as a proxy for lactic acidosis. It is difficult to ascertain the true incidence of Metformin associated lactic acidosis (MALA). Observational studies may either over or underestimate incidence of MALA [35‐37]. Most cases of MALA are multifactorial with other factors contributing to lactic acidosis such as sepsis, cardiac, respiratory or hepatic failure. Studies can be subject to ascertainment bias with fitter patients being selected for treatment with metformin. In addition, patients on metformin are more likely to have lactate levels measured leading to the potential for overestimation of MALA. International Statistical Classification of Diseases and Related Health Problems 10th Revision (ICD −10) coding in isolation cannot be used in observational studies to establish cases of MALA due to a lack of specific code for lactic acidosis and many studies have not included patients with CKD. It is therefore difficult to establish the safety of metformin in patients with CKD. We therefore examined hospitalisations with AKI in patients treated with metformin as clinically significant lactic acidosis occurs in the presence of AKI. Whilst we cannot draw conclusions regarding the rates of MALA in patients with CKD, we have shown that metformin use does not adversely affect mortality which is reassuring. This is arguably more clinically significant than actual rates of MALA.

Our large cohort of metformin users provides a reassuring message of the safety of metformin in patients with or without a background of CKD and supports the recent revision of FDA and EMA guidance on metformin prescribing in patients with renal impairment. We have also demonstrated that metformin does not adversely affect survival in patients with AKI.