Hyperkalemia and maintenance of renin–angiotensin system inhibitor therapy after initiating SGLT-2 or DPP-4 inhibitors

ABSTRACT

Background

Post hoc analyses of clinical trials suggest that sodium-glucose cotransporter-2 inhibitors (SGLT-2i) lower the risk of hyperkalemia and facilitate the use of renin–angiotensin system inhibitors (RASi) in people with type 2 diabetes. Whether this is also observed in routine care is unclear. We investigated whether SGLT-2i lowered the risk of hyperkalemia and RASi discontinuation as compared to dipeptidyl peptidase 4 inhibitors (DPP-4i).

Methods

Using the target trial emulation framework, we studied adults with type 2 diabetes (T2D) who started SGLT-2i or DPP-4i in Stockholm, Sweden (2014–2021). The outcomes were incident hyperkalemia (potassium >5.0 mmol/l), mild hyperkalemia (potassium >5–≤5.5 mmol/l), and moderate to severe hyperkalemia (potassium >5.5 mmol/l). Among RASi users, we studied time to RASi discontinuation through evaluation of pharmacy fills. Cox regression with inverse probability of treatment weighting was used to estimate per-protocol hazard ratios (HRs).

Results

In total, 29 849 individuals (15 326 SGLT-2i and 14 523 DPP-4i initiators) were included (mean age 66 years, 37% women). About one-third of participants in each arm discontinued treatment within 1 year. Compared with DPP-4i, SGLT-2i use was associated with a lower rate of hyperkalemia (HR 0.77; 95% CI: 0.64–0.93), including both mild (0.76; 0.62–0.93) and moderate/severe (0.53; 0.40–0.69) hyperkalemia events. Of 19 116 participants who used RASi at baseline, 7% discontinued therapy. Initiation of SGLT-2i vs. DPP-4i was not associated with the rate of RASi discontinuation (0.97; 0.83–1.14). Results were consistent in intention-to-treat analysis and across strata of sex, cardiovascular disease, use of MRA, and use of RASi.

Conclusions

In patients with diabetes managed in routine clinical care, the use of SGLT-2i was associated with lower rates of hyperkalemia compared with DPP-4i. Possibly because of a relatively high rate of treatment discontinuations, this was not accompanied by higher persistence on RASi therapy.

Graphical Abstract

Open in new tabDownload slide

BACKGROUND

People with type 2 diabetes (T2D) are at increased risk of experiencing hyperkalemia, particularly in the presence of comorbid conditions such as hypertension, heart failure, and chronic kidney disease (CKD) [1–3]. Hyperkalemia is the most common adverse event of renin–angiotensin system inhibitors (RASi), a first-line therapy for the management of cardiorenal diseases [4, 5]. Hyperkalemia can be potentially life-threatening, resulting in arrhythmias or cardiac arrest [1–3]. Because of hyperkalemia or fear for hyperkalemia, RASi treatments are often not initiated or discontinued prematurely, exposing patients to a higher risk of adverse clinical outcomes such as cardiovascular disease, heart failure, and death [6, 7]. Thus, strategies that lower hyperkalemia risks and allow prolonged RASi use are needed.

Recent trial findings show that sodium-glucose cotransporter-2 inhibitors (SGLT-2i), in addition to offering cardiorenal protection, may also lower the risk of hyperkalemia and improve persistence to RASi therapy [8–11]. However, it is currently unknown whether these effects are observed beyond the controlled settings of trials. Compared to trial participants, patients managed routine care tend to be older, with a higher prevalence of comorbidity, with more varying medication regimes, less adherence, and subjected to heterogeneous clinical practices. Two observational studies using routine care data have observed that new users of SGLT-2i experienced a lower rate of hyperkalemia when compared to other diabetes medications [12, 13], but they used insensitive diagnostic codes to define hyperkalemia and ascertained drug use by prescription, which can induce information bias due to primary non-adherence [14]. No routine care studies, to our knowledge, have investigated whether SGLT-2i therapy associates with persistence to RASi therapy.

Thus, the aim of this study was to compare the efficacy of SGLT-2i and dipeptidyl peptidase 4 inhibitors (DPP-4i) in reducing the risk of hyperkalemia in people with T2D, using health records and laboratory data from the region of Stockholm, Sweden. Furthermore, we investigated whether the utilization of SGLT-2i or DPP-4i facilitated the adoption of RASi.

MATERIALS AND METHODS

Data sources

We used data from the Stockholm CREAtinine Measurements (SCREAM) project, a healthcare utilization cohort including all adult residents in Stockholm between 2006 and 2021 [15]. The region of Stockholm had a population of 2.3 million citizens in 2019 and provides universal healthcare with a single unified health system. Administrative databases with complete information on demographic data, healthcare use, diagnoses and therapeutic/surgical procedures, and vital status were enriched with performed laboratory tests, dispensed prescriptions at Swedish pharmacies and validated kidney replacement therapy endpoints. Registries were linked and de-identified by the Swedish National Board of Welfare and are considered to have no or minimal loss to follow up. Because the study utilized de-identified data, it was deemed not to require informed consent and was approved by the regional ethical review boards and the Swedish National Board of Welfare.

Target trial specification and emulation

Following the target trial emulation framework [16, 17], we first specified the protocol of a target trial that would evaluate the comparative effectiveness of SGLT-2i versus DPP-4i on hyperkalemia risk in patients with T2D. The key components of the target trial protocol and emulation process are presented in detail next and in Supplementary Table S1

Study design

We used an active-comparator new-user design to mitigate the risk of confounding by indication and time-related biases [18, 19]. Based on similarity of indications of SGLT-2i during the study period, DPP-4i—another class of glucose-lowering medication with no known effect on hyperkalemia—was selected as the active comparator.

We included all adult (>18 years old) community-dwelling citizens with T2D from Stockholm who were new users of SGLT-2i or DPP-4i. New users were defined as patients who filled their first dispensation for SGLT-2i or DPP-4i between 1 January 2014, and 31 December 2021, and no previous recorded dispensation of either drug in the previous year. The date of the first SGLT-2i or DPP-4i dispensation was defined as the index date. Patients with a history of kidney failure (initiation of dialysis or kidney transplantation or eGFR of <15 ml/min per 1.73 m²), pancreatitis, cirrhosis, or acute hepatitis were excluded (Supplementary Table S2). Furthermore, we excluded patients who had a recorded serum/plasma potassium >5.5 mmol/l in the 6 months prior to treatment initiation, or dispensed a potassium binder in the 6 months preceding the index date to decrease the possibility that early hyperkalemia events during follow up would be related to a previous hyperkalemia event, or that the drugs under comparison were started differentially based on history of hyperkalemia. An overview of the longitudinal study design is presented in Supplementary Figure S1

Treatment strategies

We compared two treatment strategies: initiation of a SGLT-2i (i.e. dapagliflozin, empagliflozin, canagliflozin) and continuation of treatment during follow up versus initiation of a DPP-4i (i.e. sitagliptin, vildagliptin, linagliptin, or saxagliptin) and continuation of treatment during follow up. Discontinuation of SGLT-2i or DPP-4i treatment was defined as no further dispensation recorded within the 90 days after the estimated supply of the most recent dispensation. We focused on the effect of continuous treatment to offer more clinically relevant treatment effect estimates, since many patients discontinue their initial treatment in clinical practice.

Outcomes

The primary outcome was hyperkalemia, defined by the presence of an elevated serum/plasma potassium exceeding the commonly used clinical threshold of 5.0 mmol/l [20]. For completeness, we also evaluated mild hyperkalemia (5 < potassium ≤ 5.5 mmol/l) and moderate to severe hyperkalemia (potassium >5.5 mmol/l), separately. We examined both the first event and recurrent events (i.e. multiple hyperkalemia events). For the latter, abnormal potassium elevations within 7 days were grouped and considered the same event. Secondary outcome was discontinuation of RASi, defined as the absence of a RASi dispensation during the 90 days after the estimated pill supply of the most recent fill had ended. Detailed definitions of all outcomes are summarized in Supplementary Table S3

Patients were followed from index date until hyperkalemia, death, emigration from the region, or end of follow up (31 December 2021), whichever occurred first. Since the per-protocol effect was our main causal effect of interest, patients were additionally censored when no dispensations were received within the 90 days after the estimated end of pill supply from the most recent dispensation.

Confounders

Baseline confounders were ascertained at the index date and included demographics (e.g. age, sex), laboratory measurements [e.g. potassium level, eGFR, glycated hemoglobin [HbA1c], and urinary albumin-to-creatinine ratio], comorbidities (e.g. heart failure, psychiatric disorder and hypertension), other diabetes medication use, other medications use [e.g. angiotensin-converting enzyme inhibitor/angiotensin receptor blocker (RASi), mineralocorticoid receptor antagonists (MRA)], healthcare resource utilization, and calendar year. Definitions and look-back windows for each confounder are provided in Supplementary Table S4

Statistical analyses

Continuous variables are presented as mean ± standard deviation or median with interquartile range (IQR), and categorical variables as numbers and percentages. We used inverse probability of treatment weighting (IPTW) to adjust for baseline confounding [21, 22]. Using a multivariable logistic regression model, we estimated the probability of initiating SGLT-2i vs DPP-4i (propensity score, PS) as a function of the baseline covariates listed before. Patients in the SGLT-2i group were weighted by 1/PS and in the DPP-4i group by 1/(1–PS). Weights were stabilized by using the marginal probability of the received treatment as the numerator of the weights. We calculated standardized mean differences (SMDs) to assess the balance of covariates between the treatment groups before and after weighting, using a SMD of >0.1 as the threshold for meaningful imbalance.

Primary analyses

A weighted Cox regression model was used to estimate hazard ratios (HRs) for first hyperkalemia event, while for recurrent hyperkalemia events, incidence rate ratios (IRRs) were estimated by a weighted Poisson regression model. Weighted cumulative incidence curves were plotted using the Aalen–Johansen estimator, which does not overstate absolute risks in the presence of the competing risk of death [23]. The 95% confidence intervals for the absolute risks were calculated using a nonparametric bootstrap with 1000 samples.

In our cohort, potassium, HbA1c, albuminuria, and eGFR were missing in 10.9%, 4.7%, 20.9%, and 5.1%, respectively. We therefore used multiple imputation by chained equations to impute missing data. The imputation model used to derive 50 imputed data sets included the treatment variable, all covariates, and the Nelson–Aalen estimate of the cumulative hazard [24]. The within method was chosen to combine multiple imputation with IPTW [25]. This method uses the propensity score to derive effect estimates for each imputed dataset independently, which are then pooled to get an overall estimate.

Secondary analysis: persistence on RASi therapy

To analyze the persistence on RASi therapy, we restricted our study population to individuals who were on RASi therapy at the time of SGLT-2i or DPP-4i initiation. Ongoing RASi use was ascertained by the overlap between the date of SGLT-2i or DPP-4i treatment start and the estimated pill supply of the last recorded RASi dispense. In these analyses, we added a categorical variable of RASi adherence to our propensity score model; the category levels of this variable were “new users of RASi” (if treatment was initiated at the same time or within a year prior to the initiation of SGLT-2i or DPP-4i), “prevalent users with good adherence” (if there was a history of RASi dispensations for at least one year before index date and the proportion of days covered in that year was ≥75%), and “prevalent users with poor adherence” (if proportion of days covered <75%).

Subgroup and sensitivity analyses

Subgroup analyses were performed to test for potential effect modification by age (≥70 years vs. <70 years), sex, cardiovascular disease, use of RASi, and eGFR categories (<60 vs. ≥60 ml/min/1.73 m²). To evaluate whether the associations varied significantly by these factors, we performed likelihood ratio tests comparing multivariate models with and without multiplicative interaction terms. Finally, IPTW and multiple imputation using the within method were used for each subgroup to adjust for baseline confounding and imputing missing values, respectively.

To test the robustness of our findings, the following sensitivity analyses were performed. First, to investigate the influence of informative censoring we applied an intention-to-treat follow-up approach, where follow up was continued for a maximum of 1 and 2 years, regardless of treatment discontinuation or switch. Second, to detect the influence of potential unmeasured confounding, we investigated the effects of SGLT-2i vs. DPP-4i on two positive control outcomes: hospitalization for heart failure and major adverse cardiovascular events (MACE), which have a well-established nonnull causal association with SGLT-2i in clinical trials [26–28]. In addition, we evaluated as a negative control outcome diverticular disease, which we hypothesize it should not be associated with SGLT-2i use. Third, to investigate potential differential outcome ascertainment due to differences in the frequency of potassium testing between the SGLT-2i and DPP-4i arms, we calculated the number of potassium tests during follow up in the unweighted population.

RESULTS

Baseline characteristics

After applying inclusion and exclusion criteria, we included 29 849 adults with T2D of which 14 523 initiated DPP-4i and 15 326 initiated SGLT-2i therapy (Supplementary Figure S2). Main baseline characteristics are reported in Table 1, and all baseline characteristics used in our models are presented in Supplementary Table S5.

New users of SGLT-2i were younger than new users of DPP-4i (64 vs. 68 years), had higher HbA1c (63 vs 61 mmol/mol), and higher eGFR (90 vs. 83 ml/min/1.73 m²). SGLT-2i users had a higher prevalence of acute coronary syndrome (17% vs. 11%) and other ischemic heart disease (27% vs. 20%). SGLT-2i users also received more MRA (7% vs. 5%) and beta-blockers (44% vs. 40%). Compared to new users of DPP-4i, users of SGLT-2i had a higher frequency of outpatient visits to the diabetologist (19% vs. 17%) in the year prior to treatment initiation.

After weighting, all covariates were well balanced between treatment strategies, with SMDs < 0.1 (Supplementary Table S5, Supplementary Figures S3 and S4).

SGLT-2i vs. DPP-4i and rates of hyperkalemia

Discontinuation of treatments was not uncommon. As many as 32.4% of users of SGLT-2i and 34.2% of users of DPP-4i discontinued their treatments within a year (Fig. 1). During a median follow up of 0.9 years (IQR 0.4–1.9) on treatment, 995 individuals in the DPP-4i group and 477 individuals in the SGLT-2i group experienced at least one hyperkalemia event, corresponding to an incidence rate of first hyperkalemia event of 44.8 and 27.5 events per 1000 person-years, respectively (Table 2). SGLT-2i initiation, compared with DPP-4i initiation, was associated with a lower rate of hyperkalemia (adjusted HR 0.77; 95% CI 0.64- 0.93) (Table 2), including both mild (0.76; 0.62–0.93) and moderate/severe (0.53; 0.40–0.69) hyperkalemia. Weighted cumulative incidence curves (Fig. 2a–c) showed lower absolute risks of hyperkalemia in the SGLT-2i group throughout follow up. The 1-year absolute risks of hyperkalemia were 3.4% (95% CI: 2.8, 4.2%) in the SGLT-2i group and 4.5% (4.1, 4.9%) in the DPP-4i group, resulting in a weighted risk difference of −1.1% (95% CI: −1.9%, −0.2%).

In the primary cohort, 486 patients experienced multiple hyperkalemia events. Over a median follow up of 0.9 years (IQR 0.4–1.9), we detected 2554 repeated hyperkalemia events (at least 7 days apart). The adjusted IRR of hyperkalemia recurrence in the SGLT-2i vs DPP-4i group was 0.68 (95% CI, 0.62, 0.74) (Table 2). Similar lower IRR were observed for mild hyperkalemia (IRR 0.74; 95% CI, 0.68–0.82) and moderate/severe hyperkalemia (IRR 0.43; 95% CI, 0.35–0.53).

SGLT-2i vs. DPP-4i and persistence on RASi

A total of 19 116 (64%) participants used RASi at the time of initiation of SGLT-2i or DPP-4i and thus constituted to our secondary cohort. Their baseline characteristics are summarized in Supplementary Table S6 All baseline characteristics were well balanced after weighting (Supplementary Figure S5).

Over a median follow up of 0.9 (IQR: 0.4, 2.0) years, 7% of participants stopped RASi therapy. The incidence rates of RASi discontinuation were 55.1 (51.3–59.2) events per 1000 person-years in the SGLT-2i group and 47.9 (44.0–52.1) in the DPP-4i group (Table 3). There was no difference in the rates of RASi discontinuation between group (adjusted HR 0.97; 95% CI 0.83 to 1.14), nor in their 1-year absolute risk difference −0.1% (95% CI: −1.0, 0.9%) (Table 3).

Subgroup and sensitivity analyses

Subgroup analyses showed consistency of our main results across strata of sex, cardiovascular disease, use of MRA, and use of RASi (all P values for interaction >.05, Fig. 3). By contrast, there was a suggestion for heterogeneity across strata of age, diuretics, and eGFR. Specifically, there was no association between SGLT-2i and the risk of any hyperkalemia and mild hyperkalemia in populations using diuretics, aged ≥70 years, and eGFR <60 ml/min/1.73 m². However, there was consistency across all strata on lower risk of moderate/severe hyperkalemia.

Results of intention-to-treat analyses with 12- and 24-months follow up aligned with our main analyses, but with a smaller magnitude of effect (Supplementary Table S7 and S8). Consistent with trial evidence [26–28], the rates of heart failure (adjusted HR 0.78; 95% CI 0.66 to 0.92) and MACE (adjusted HR 0.74; 95% CI 0.62 to 0.89) were lower for SGLT-2i than DPP-4i (Supplementary Table S9). Conversely, there was no difference between treatment groups for the occurrence of diverticular disease (adjusted HR 0.99; 95% CI 0.80 to 1.24; Supplementary Table S9). No major differences were observed in the frequency of potassium testing across treatment groups (Supplementary Table S10).

DISCUSSION

In this large cohort study of >29 000 patients with T2D managed in routine care, we observed that use of SGLT-2i was associated with a lower rate of hyperkalemia compared with DPP-4i use. However, we found no difference in the rates of RASi discontinuation. These findings were consistent across subgroups and various sensitivity analyses.

Our study expands post hoc observations from clinical trials suggesting lower hyperkalemia rates in individuals randomized to SGLT-2i vs. placebo [29,30]. Our study also expands to European health systems two previous observational studies using routine care data: Fu et al. [12], evaluated US claims data of patients with T2D and CKD stages 3–4, and observed that the initiation of SGLT-2i vs DPP-4i was associated with a lower rate of hyperkalemia diagnoses; Wu et al. reported that initiation of SGLT-2i by Chinese patients with T2D was associated with a lower rate of moderate/severe hyperkalemia (potassium >5.5 mmol/l) compared to DPP-4i in an intention-to-treat analysis.

To this preceding evidence we add some novel findings. Hyperkalemia is a common complication in T2D, but many of these events are not coded with clinical diagnoses [31, 32]. Therefore, we regard as a study strength the capture of all potassium measurements performed in our region. This allowed for the novel exploration of mild hyperkalemia events (potassium >5–5.5 mmol/l), which may be less likely to receive a diagnosis code but more likely to prompt clinical decisions such as discontinuation of therapies [33]. The evaluation of recurrent hyperkalemia events is another novel finding that reinforces the plausibility of the observed associations. We also observed larger hyperkalemia reduction rates when our follow up was restricted to periods of drug use (i.e. per-protocol effect), compared to intention-to-treat approach, which supports the possibility of these effects being a direct consequence of SGLT-2i therapy. Another novelty of our analysis is that whereas we observed consistently lower rates of moderate/severe hyperkalemia (potassium >5.5 mmol/l) in relation to SGLT-2i use across a range of subgroups, the associations were less consistent for mild hyperkalemia (potassium >5.0–5.5 mmol/l). Specifically, we failed to observe any difference between treatment strategies in subgroups of old age (>70 years) and eGFR<60 ml/min/1.73 m². Previous trials and observational studies [12, 13, 29, 30] evaluating this effect have measured the rate of moderate/severe hyperkalemias, or hyperkalemia severe enough to deserve a diagnosis, and in this setting our results agree. However, we are not aware of other studies that have explored the effect of SGLT-2i on mild hyperkalemias and cannot confirm/refute our observations. We speculate that the lack of association with mild hyperkalemia events in our study suggests that SGLT-2i is less effective in managing the small variations in potassium levels potentially attributable to the multiple risk factors (i.e. diet, renal potassium excretion, acidosis, comorbidities, polypharmacy, etc.) that operate in older adults or in those with established CKD [20].

Our findings align with mechanistic evidence on the pleiotropic effects of SGLT-2i on potassium homeostasis. First, SGLT2i increase distal sodium and water delivery to the distal nephron, enhancing electronegativity in the tubular lumen, and thus leading to increase urinary potassium secretion [34]. Second, SGLT2i increase aldosterone secretion, which further directly and indirectly increases urinary potassium excretion, lowering the risk of hyperkalemia [35–37]. Third, rather than having a direct effect on potassium homeostasis, the reduction in the incidence of hyperkalemia observed with SGLT2 inhibitors might be also due to its intrinsic nephroprotective effect, i.e. maintenance of kidney function itself, or due to yet uncharacterized mechanisms.

An important novel finding in our analysis is that despite a lower hyperkalemia occurrence among SGLT-2i users, there was no apparent effect on RASi persistence in our study. This observation contrasts with a recent secondary analysis from the pooled CREDENCE and DAPA-CKD trials [11]. In that study, the authors observed that the relative risk for RASi discontinuation was 15% lower in patients randomized to receiving SGLT-2i (HR, 0.85; 95% CI, 0.74 to 0.99) compared to placebo. A potential explanation is that in our study, follow up was too short for patients to benefit enough from SGLT-2i: while patients on CREDENCE and DAPA-CKD had median 2.2 (IQR 1.6–2.6) years on therapy, patients in our study were median 0.9 (IQR 0.4–1.9) years. This short persistence to SGLT-2i is not an isolated finding, as it echoes multiple reports globally showing that at least 40% of patients discontinue SGLT-2i treatment within 1 year [38–42]. Drug efficacy and safety may importantly differ in routine care versus trial settings, where patients are carefully selected and drug adherence and accountability are closely monitored. The short duration of SGLT-2i treatments in current clinical practice forms a substantial barrier to the delivery of evidence-based care. Additionally, our design may introduce bias by censoring patients when stopping SGLT-2i, without evaluating if the medication cessation was permanent or temporary (for instance, a temporary cessation during the resolution of an adverse effect). However, the fact that we observe similar associations following an intention-to-treat design (i.e. assuming the medications to be consumed throughout the follow up) would argue against this.

Additional strengths of our study include the use of a target trial emulation—which mitigates time-related biases such as immortal and survivor bias—and the setting of a universal tax-funded health system, which minimizes selection bias from disparate access to health care. We ascertained treatment duration with precision by capturing the totality of pharmacy dispensations in our country, which is a more accurate surrogate of drug intake than a doctor’s prescription (i.e. intention). Limitations of our study include the lack of information of confounders such as dietary potassium intake, BMI, bicarbonate levels or the use of potassium-containing supplements. However, our modeling of positive and negative control outcomes and our comparison of rates of potassium testing between treatment strategies suggests our findings to be potentially robust to uncontrolled confounding and detection bias. A final limitation is the lack of information on race. Thus, our findings may be limited in terms of generalizability to other world regions with larger ethnic diversity.

To conclude, in this cohort study of patients with T2D managed in routine care, SGLT-2i initiation was associated with a lower risk of hyperkalemia compared to DPP-4i. However, this was not accompanied by higher persistence on RASi therapy.

ACKNOWLEDGEMENTS

The results of this study were presented at the 61st European Renal Association (ERA) Congress in Stockholm (Sweden).

FUNDING

Research reported in this publication was supported by the Swedish Research Council, the Swedish Heart and Lung Foundation, the the National Institutes of Health (R01 DK115534), the Young Scientists Fund, National Natural Science Foundation of China [Grant No. 82304245] and the Netherlands Organisation for Scientific Research.

DATA AVAILABILITY STATEMENT

The data underlying this article will be shared on reasonable request to the corresponding author.

CONFLICT OF INTEREST STATEMENT

The authors do not report any direct disclosure in relation to this study. Unrelated to the study, J.J.C. reports funding to Karolinska Institutet by AstraZeneca, Astellas, Amgen, Vifor Pharma, and NovoNordisk; personal honoraria for lectures by Fresenius Kabi, Baxter Healthcare, and Abbott, and being a member of advisory boards for Astellas, AstraZeneca, and GSK. ME reports funding from AstraZeneca and Astellas pharma, advisory boards from Astellas, and payment for lectures by AstraZeneca, Astellas, Boehringer-Ingelheim, and Vifor Pharma.

REFERENCES

Author notes