Insights from CREDENCE trial indicate an acute drop in estimated glomerular filtration rate during treatment with canagliflozin with implications for clinical practice

Please cite this article as: Oshima M, Jardine MJ, Agarwal R, Bakris G, Cannon CP, Charytan DM, de Zeeuw D, Edwards R, Greene T, Levin A, Lim SK, Mahaffey KW, Neal B, Pollock C, Rosenthal N, Wheeler DC, Zhang H, Zinman B, Perkovic V, Heerspink HJL, Insights from CREDENCE trial indicate an acute drop in estimated glomerular filtration rate during treatment with canagliflozin with implications for clinical practice., Kidney International (2021), doi: https://doi.org/10.1016/j.kint.2020.10.042.


Insights from CREDENCE trial indicate an acute drop in estimated glomerular filtration rate during treatment with canagliflozin with implications for clinical practice.
Long-term eGFR trajectories and overall and renal safety profiles were similar among subgroups of an acute drop in eGFR.

Introduction
Sodium glucose co-transporter 2 inhibitors (SGLT2) have been shown to reduce the risk of kidney and cardiovascular outcomes in people with type 2 diabetes. SGLT2 inhibitors increase distal tubular sodium and chloride delivery to the macula densa, which augments tubuloglomerular feedback causing reversal of afferent arteriolar vasodilatation. As a result, SGLT2 inhibitors cause an acute reduction in intraglomerular pressure and glomerular filtration rate (GFR). 1 Large cardiovascular outcome trials have demonstrated that the initial drop in eGFR following initiation of SGLT2 inhibition is followed by a stabilisation of longterm kidney function decline. [2][3][4][5] However, in clinical practice, an acute drop in eGFR may raise safety concerns that prevent clinicians from continuing SGLT2 inhibitors. A better characterisation of the acute drop in eGFR and its association with long-term eGFR trajectories and safety outcomes is required to support appropriate use of SGLT2 inhibitors in clinical practice.
We therefore performed a post-hoc analysis of the Canagliflozin and Renal Events in Diabetes with Established Nephropathy Clinical Evaluation (CREDENCE) trial to determine the predictors of an acute eGFR decline following initiation of the SGLT2 inhibitor canagliflozin and its associations with long-term eGFR trajectories and safety outcomes.

Effect of canagliflozin versus placebo on acute eGFR decline
Of the 4401 participants in the CREDENCE trial, 4289 individuals (97.5%) (2144 in the canagliflozin group and 2145 in the placebo group had eGFR measurements available at baseline and week 3 and were included in this analysis. All participants were using an ACEinhibitor or an angiotensin receptor blocker (ARB) as required by the protocol. At week 3, the mean acute change in eGFR was -7.0% (standard error [SE] 0.4) and 0.3% (0.4) in the canagliflozin and placebo group, respectively (mean difference -7.3%, 95% CI -8.5 to -6.1; p <0.001). More participants in the canagliflozin (956 [45%]) than in the placebo (450 [21%]) group experienced an acute eGFR drop of >10% (p <0.001) ( Table 1). An acute drop in J o u r n a l P r e -p r o o f eGFR of greater than 30% was rare and occurred in only 89 (4.2%) and 39 (1.8%) participants in the canagliflozin and placebo group, respectively (p <0.001).

Predictors of acute drop in eGFR
In the canagliflozin group, participants with an acute drop in eGFR were more likely to be older, had a longer duration of diabetes, a higher body mass index, systolic blood pressure, and eGFR, and were more likely to use diuretics at baseline, while they were less likely to have heart failure ( Table 1). In the placebo group, it was higher systolic blood pressure, eGFR, and UACR that predicted an acute drop in eGFR.
To identify participants particularly prone to experience a drop in eGFR >10% with canagliflozin rather than placebo we performed a logistic regression analyses to examine the consistency of effects of canagliflozin versus placebo on acute eGFR decline in subgroups defined by baseline characteristics. Overall, canagliflozin increased the likelihood of an acute eGFR drop >10% compared to placebo (odds ratios [OR] 3.03, 95% confidence interval [CI] 2.65, 3.47; p <0.001) ( Figure 1). In subgroup analyses across various baseline participant characteristics, participants with a history of heart failure were less likely to experience an eGFR decline >10% following canagliflozin initiation relative to placebo (p interaction =0.04). The likelihood of an acute drop in eGFR of >10% with canagliflozin compared to placebo was similar across other subgroups defined by the eGFR category at screening ( Figure 1). Other participant characteristics at baseline were also not associated with a higher likelihood of a drop in eGFR with canagliflozin compared to placebo. Finally, the effect of canagliflozin versus placebo on percentage change in eGFR during the first 3 weeks was also consistent across various subgroups defined by participant characteristics at baseline (Supplementary table 1).

Treatment discontinuation
Among these 4289 participants, none discontinued their assigned treatment during the first 3 weeks of the trial. After week 3, 519 (24.2%) in the canagliflozin and 629 (29.3%) in the J o u r n a l P r e -p r o o f placebo group discontinued their assigned treatment during follow-up. There was no statistically significant difference in number of participants who discontinued therapy across categories of an acute eGFR change in each treatment group (Table 2). Discontinuation rates were also similar within the canagliflozin group when finer categories of eGFR decline thresholds were used (Supplementary table 2).

Long-term eGFR trajectories
Compared to the placebo group, the annual decline in eGFR in the canagliflozin group was smaller in each category of an acute change in eGFR. After multivariable adjustment, the annual change in eGFR from week 13 to the last available measurements among participants receiving canagliflozin with an acute eGFR drop >10% was similar as compared with those receiving canagliflozin that experienced an acute modest drop between 0 to 10% or increase in eGFR (p = 0.11; Figure 2A). In the few participants with an acute drop in eGFR >30% after canagliflozin initiation long-term eGFR trajectory was similar as compared to the main analyses (Supplementary figure 1). Within the placebo group there was no statistically significant difference in an annual change in eGFR across categories of an acute change in eGFR (p = 0.09; Figure 2B). Further adjustment for acute changes in eGFR did not alter these findings. In a subgroup analysis, the long-term eGFR trajectories within the canagliflozin group were similar as compared to the main analysis in each subgroup of eGFR at screening (Supplementary table 3). The only exception was the subgroup with eGFR between 30 and 45 mL/min/1.73 m 2 . In this subgroup participants with a drop in eGFR had a significantly slower rate of eGFR decline during long-term follow-up compared to participants with an increase in eGFR (Supplementary table 3).

Safety outcomes during acute and chronic phases
During the first 3 weeks, the percentage of participants experiencing an adverse event was similar between the canagliflozin and placebo groups except for renal related adverse events (Supplementary table 4). Renal related adverse events were more frequently J o u r n a l P r e -p r o o f reported in those who experienced an acute eGFR drop of >10% with canagliflozin than placebo.
During the chronic phase after week 3, the event rates for each safety outcome were lower in the canagliflozin compared to placebo group regardless of the acute change in eGFR category (Figure 3). Within the canagliflozin and placebo groups separately, the event rates and multivariable adjusted hazard ratios for overall and serious adverse events, overall and serious renal related adverse events, acute kidney injury, and hyperkalemia did not differ across categories of an acute change in eGFR ( Figure 3). When finer categories of eGFR decline thresholds were used, results were generally similar with the exception that in the subgroup of participants with an acute drop of >30% using canagliflozin the overall adverse events (multivariable adjusted hazard ratio 1. were higher compared to participants with an acute increase in eGFR using canagliflozin (Supplementary figure 2). Similar findings were observed across screening eGFR subgroups (Supplementary table 5). In an additional analysis using the predicted likelihood of an acute eGFR drop >10%, the effects of canagliflozin relative to placebo on adverse events were consistent regardless of a likelihood of an eGFR drop >10% above or below the median (p for interaction >0.51) (Supplementary table 6).

Discussion
In this post-hoc analysis of the CREDENCE trial we demonstrated that an initial drop in eGFR in response to canagliflozin is common but a larger drop of 30% in eGFR was a rare event. We also demonstrated that the long-term eGFR trajectories as well as overall and renal safety profiles during canagliflozin treatment were similar regardless of the initial eGFR drop except when it unusually exceeded 30%, when adverse events and renal related adverse events were reported more frequently. These data suggest that an acute decrease in eGFR up to 30% can be tolerated after treatment initiation with canagliflozin.

J o u r n a l P r e -p r o o f
Understanding which patients with diabetes and CKD are more likely to experience an acute drop in eGFR upon SGLT2 inhibitors initiation is important to guide best use of SGLT2 inhibitors in clinical practice. Observational analyses done in just those assigned to canagliflozin identified older age, higher systolic blood pressure, higher eGFR, absence of heart failure, and use of diuretic treatment as predictors of an acute fall in eGFR. Acute drops in eGFR also occurred in some placebo group participants, which is likely explained by the large biological variability of eGFR and the phenomenon of regression of the mean. 6 The participant characteristics associated with an acute drop in eGFR in the placebo group were higher systolic blood pressure, eGFR, and UACR. In the randomised comparison between canagliflozin and placebo it was only a history of heart failure that predicted acute drop in eGFR with those with a history of heart failure less likely to experience an acute eGFR drop of at least 10%. We note that the interaction p-value for this comparison was of borderline statistical significance and not adjusted for multiple testing and the results should thus be cautiously interpreted, but it is possible that improvements in cardiac output following SGLT2 inhibition increase glomerular filtration and decreases the likelihood of a drop in GFR.
The likelihood of experiencing a greater than 10% decline in eGFR following canagliflozin was consistent regardless of the starting eGFR level. It should be noted that in the current analyses the initial decline in eGFR was expressed as percentage decline and the absolute acute eGFR decline was smaller at lower compared to higher eGFR as previously reported. 7 Additionally, among participants in the lowest eGFR category at screening, the long-term rate of eGFR decline was also less in participants with an acute eGFR decline compared to those with an increase. Taken together, we believe that practitioners may be reassured by these data that in conjunction with the renal and cardiac benefits seen across the eGFR spectrum of 30-90 ml/min/1.73m 2 support initiation of canagliflozin in patients with eGFR between 30 and 45 mL/min/1.73 m 2 .
The results of our study demonstrate that acute declines in eGFR up to 30% after initiation of canagliflozin treatment would not be a reason to discontinue treatment -eGFR J o u r n a l P r e -p r o o f trajectories during chronic treatment with canagliflozin were consistent within the canagliflozin group and significantly better than placebo regardless of the initial eGFR change. Moreover, safety outcomes within canagliflozin-treated subjects were consistent regardless of the initial eGFR change. Acute drops in eGFR greater than 30% were rare but were associated with a higher likelihood of adverse events during prolonged treatment that should prompt careful monitoring. This 30% eGFR threshold is close to the 30% serum creatinine increase, equivalent to a 27% eGFR decline, deemed tolerable upon initiation of renin-angiotensin-aldosterone system (RAAS) inhibition in light of the long-term kidney protective effects of these agents. 8,9 We note that although a greater than 30% drop in eGFR with canagliflozin was associated with a higher risk of adverse events we cannot answer the question whether reducing the dose or discontinuing canagliflozin would attenuate this increased risk given the nature of the data available to us.
During the first 3 weeks of the trial renal related adverse events were more frequently reported among individuals with an acute drop in eGFR. Renal related adverse events comprise a range of different investigator reported events, which include decreases in eGFR and increases in serum creatinine. The higher frequency during the early stage of the trial may be a reflection of investigator response to the acute drop in eGFR.
Rather than being interpreted as an adverse event, prior studies in patients with CKD with or without diabetes treated with ACE inhibitors or ARBs demonstrated that an acute drop in eGFR is associated with long-term preservation of kidney function, supporting the concept that the acute decline in eGFR is of hemodynamic origin and reflects a reduction in intra-glomerular pressure that is associated with long-term kidney preservation. [10][11][12][13] However, this pattern could not be replicated in other studies with ACE-inhibitors or ARBs nor in our trial. 14,12,15,16 It should be noted that the subgroup of participants with an acute drop in eGFR comprised participants with a canagliflozin induced acute (reversible) drop in eGFR as well as participants with pre-existing progressive kidney function loss in whom eGFR will likely continue to decline during the chronic phase of the trial thereby diluting the association. This complicates assessment and interpretation of the relationship between an acute drop in J o u r n a l P r e -p r o o f eGFR and long-term kidney function. A better characterization of the eGFR decline prior to initiation of SGLT2 treatment would help in future studies to distinguish the acute hemodynamic drop in eGFR from progressive kidney function loss.
The strengths of our study include the large number and diverse groups of participants, the sequential measurements of eGFR during the CREDENCE trial, and the ability to adjust for multiple important risk factors. However, our study has several limitations.
First, as our study cohort was derived from a clinical trial of patients with type 2 diabetes and chronic kidney disease, the results may limit generalisability to broader populations. Second, the analyses according to the acute change in eGFR were no longer randomized. Although we adjusted the analyses for differences in participant characteristics across drop in eGFR categories, residual confounding cannot be excluded. In addition, eGFR was not measured after treatment discontinuation. We therefore could not establish whether the acute drop in eGFR is reversible after treatment discontinuation. However, in patients with type 2 diabetes with established cardiovascular disease or cardiovascular risk factors it has been shown that eGFR is completely reversible 30 days after canagliflozin discontinuation 2 . Finally, serum creatinine was measured with the Jaffe method which is more susceptible to interferences than the enzymatic methods. 17 In addition, within-individual variability in eGFR is high. 6 Both factors may have led to misclassifications and limit the precision of the results.
In conclusion, although acute drops in eGFR >10% occurred in nearly half of all participants following initiation of canagliflozin, the benefit of canagliflozin compared with placebo was observed regardless of the acute eGFR decline. Safety outcomes were also similar among subgroups of initial drop in eGFR. Considering the beneficial effects of canagliflozin on renal and cardiovascular outcomes in patients with type 2 diabetes and CKD, our data suggest that an acute drop in eGFR of no more than 30% following canagliflozin initiation may be no reason for safety concern, but close monitoring of kidney function remains essential.

Study design and participants
We performed a post-hoc observational analysis from the CREDENCE trial (ClinicalTrials.gov no. NCT02065791) which was a multicenter, double-blind, placebocontrolled, randomized trial evaluating the effects of canagliflozin on renal outcomes in subjects with type 2 diabetes and chronic kidney disease. The design of the trial and primary outcomes have been published previously. 3,18 In brief, a total of 4401 individuals underwent randomization at 690 sites in 34 countries between March 2014 and May 2017. Participants were eligible if they were ≥30 years of age and had type 2 diabetes, with a glycated haemoglobin (HbA1c) level of 6.5 to 12.0%. They were also required to have chronic kidney disease, defined as eGFR of 30 to <90 ml/min/1.73 m 2 and urinary albumin-to-creatinine ratio (UACR) of >300 to 5000 mg/g (>33.9 to 565.6 mg/mmol). All the participants were to receive a stable dose of RAAS inhibitors for at least 4 weeks before randomization.
Participants were randomized to receive canagliflozin 100 mg daily or matching placebo using randomly permuted blocks with stratification by screening eGFR categories

Category of an acute change in eGFR
Serum creatinine was measured in a central laboratory by use of the Jaffe method with rate blanking at randomization, week 3, week 13, week 26, and every 26 weeks thereafter, as published previously. 3 In an additional analysis, we used finer categories of acute change in eGFR: >30% decrease, >20 to 30% decrease, >10 to 20% decrease, >0 to 10% decrease, or ≥0% increase. Since these subgroups were defined by a post-randomization variable (eGFR), these subgroups do not allow for randomized comparison.

Outcomes and follow-up
The outcome for this study was the on-treatment eGFR slope which was defined as the annual rate of change in eGFR based on all on-treatment eGFR measurements from week 13 to the last available measurements. Investigator reported safety outcomes included any adverse events, any serious adverse events, (serious) renal-related adverse events, acute kidney injury, and hyperkalemia. The definition of hyperkalemia includes the preferred terms of "hyperkalemia" and "blood potassium increased" according the Medical Dictionary for Regulatory Activities. Number of events were counted separately for the acute phase (from baseline to week 3) and for the chronic phase (from week 3 through 30 days after the last dose for the safety analyses) of the trial. Participants were followed from the end of the acute phase until the first of the study outcomes, death, or the end of follow-up

Statistical analyses
The on-treatment population, defined as all participants who took at least 1 dose of study medication, was used for our analysis. We summarised baseline characteristics according to categories of an acute change in eGFR in the canagliflozin and placebo group. Continuous variables were reported as means with SDs for variables with approximately symmetrical distributions. Results for variables with skewed distributions were presented as median and interquartile range (IQR) and were transformed into natural logarithms before analysis.
Linear trends across categories of an acute change in eGFR in each treatment group were tested by linear regression analysis and logistic regression analysis, as appropriate.
For assessment of predictors of acute eGFR declines following canagliflozin initiation, logistic regression models were used to estimate the univariate ORs and their corresponding 95% CIs of canagliflozin versus placebo for the risk of an acute eGFR drop >10% in the overall population and across the following baseline participant subgroups: age, sex, race or ethnic group, current smoking, history of hypertension, history of heart failure, history of cardiovascular disease, body mass index, systolic blood pressure, HbA1c, screening eGFR, UACR, and use of insulin and diuretic. A test for interaction across subgroups was performed by adding an interaction term between categorical values of each respective subgroup and treatment assignment in the logistic regression models. We also estimated the effects of canagliflozin versus placebo on an acute change in eGFR during the first 3 weeks across baseline participant subgroup using mixed effects models which were adjusted for baseline eGFR and treatment.
The on-treatment eGFR slope was analyzed by mixed effects model with random slopes and random intercepts. The model was adjusted for categories of acute change in eGFR, time, an interaction term for acute change in eGFR category by time as well as the following baseline covariates and an interaction term between these covariates and time: age, sex, race or ethnic group, current smoking, history of hypertension, history of heart failure, duration of diabetes, history of cardiovascular disease, body mass index, systolic blood pressure, HbA1c, eGFR, log-transformed UACR, high density lipoprotein cholesterol, J o u r n a l P r e -p r o o f low density lipoprotein cholesterol, log-transformed triglycerides, and use of diuretic and RAAS inhibitor. The linearity assumption was verified by visual inspection of the long-term eGFR trajectory.
Cox proportional hazard regression was performed to estimate the hazard ratio for safety outcomes according to categories of an acute change in eGFR. Safety outcomes that occurred within 3 weeks after randomization were excluded and follow-up time was calculated from the week 3 serum creatinine measurement until 30 days after the last administration of study medication. The models used an acute increase in eGFR as reference and were adjusted for baseline covariates as described above. Homogeneity of treatment effects were tested by adding an interaction term between treatment assignment and category of an acute change in eGFR as a continuous variable in the Cox regression models. The proportional hazards assumption was confirmed by including an interaction term between the acute change in eGFR category and time as a timevarying covariate in the Cox regression models. In an additional analysis, we predicted the likelihood of an acute eGFR drop >10% for each participant in the canagliflozin group using a logistic regression model based on the following baseline covariates: age, sex, history of heart failure, duration of diabetes, body mass index, systolic blood pressure, eGFR, log-transformed UACR, and use of diuretic. These covariates were selected since they were significantly different across categories in drop in eGFR. Using the logistic regression model we calculated the probability of a drop in eGFR >10% for each participant in the canagliflozin and placebo group. We finally stratified the population by median probability of drop in eGFR >10% and assessed the effects of canagliflozin versus placebo on adverse events among participants above and below the median likelihood of eGFR drop more than 10%.
To assess the impact of baseline eGFR, we estimated the long-term eGFR slope and safety outcomes according to screening eGFR categories (30-<45, 45-<60, and 60-<90 mL/min/1.73 m 2 ). All analyses were conducted using Stata, version 15 (Stata Corporation, College Station, TX, USA). A two-sided p value <0.05 was considered statistically significant.         Renal related adverse events include acute kidney injury, anuria, azotemia, blood creatinine increased, blood urea increased, glomerular filtration ratio decreased, nephropathy toxic, oliguria, renal failure, and renal impairment Hyperkalaemia was spontaneously reported by the investigator. The summary counts provided for the adverse event of hyperkalaemia include the MedDRA preferred terms of "hyperkalaemia" and "blood potassium increased."