Sodium/Glucose Cotransporter 2 Inhibitors and the Risk of Diabetic Ketoacidosis

An Example of Complementary Evidence for Rare Adverse Events

Wajd Alkabbani; Ryan Pelletier; John-Michael Gamble

Am J Epidemiol. 2021;190(8):1572-1581.

Abstract and Introduction

Abstract

Evidence from observational studies may be considered complementary to that of randomized controlled trials (RCTs), particularly when assessing rare outcomes of drug therapies. Sodium/glucose cotransporter 2 (SGLT-2) inhibitors are a novel class of antidiabetic agents that have been linked to an increased risk of diabetic ketoacidosis (DKA). We conducted a systematic review and separately meta-analyzed data from RCTs (n = 18; 2013–2019) and cohort studies (n = 7; 2017–2020) to assess the consistency of the magnitude of association between SGLT-2 inhibitors and DKA risk. We illustrate the strengths and weaknesses of the 2 designs. Results from RCTs and observational studies consistently showed almost a doubling in the risk of DKA among patients using an SGLT-2 inhibitor as compared with placebo or an active comparator. In a random-effects model, the pooled relative risk was 2.08 (95% confidence interval (CI): 1.28, 3.40) from placebo-controlled RCTs and 0.82 (95% CI: 0.25, 2.68) from active-comparator RCTs. The pooled adjusted hazard ratio from observational studies was 1.74 (95% CI: 1.28, 2.38). Notably, the 2 designs complement each other in several domains, including external and internal validity and power. This demonstrates a need for more comprehensive evidence when assessing rare adverse events for both sources.

Introduction

Randomized controlled trials (RCTs) are considered the gold standard of clinical evidence, due to the role of randomization in causal inference.^[1] Nonetheless, there has been much development over recent years of methods used in observational drug-effect studies, much of which has focused on causal inference.^[2] Despite infamous examples of conflicting findings,^[3,4] observational studies are argued to provide complementary evidence when RCTs are impossible (due to ethical and/or legal considerations), inadequate (low external validity, irrelevant comparisons), or inappropriate (a rare outcome, long-follow-up required).^[5,6]

Studying rare adverse effects of drug therapies is one scenario wherein evidence from observational studies complements that from RCTs. A contemporary example is that of the risk of diabetic ketoacidosis (DKA). In 2015, the Food and Drug Administration published drug safety communications warning that a new class of diabetes therapies called sodium/glucose cotransporter 2 (SGLT-2) inhibitors may lead to DKA,^[7] albeit with an atypical presentation whereby not all cases involved hyperglycemia but rather some involved euglycemia.^[8] Herein, using this timely and clinically relevant example, we assess the consistency of the magnitude of association of SGLT-2 inhibitor exposure with DKA between pooled data from randomized clinical trials and pooled data from observational studies. We also illustrate how strengths and weaknesses of the 2 evidentiary sources are ameliorative in multiple domains, specifically when RCTs have insufficient power and suboptimal adjudication of events.

1 2 3 4

Next Section

Abstract and Introduction
Methods
Results
Discussion
References

Table 1. Characteristics of Studies Included in a Systematic Review of Sodium/Glucose Cotransporter 2 Inhibitors and Risk of Diabetic Ketoacidosis, 2013–2020
First Author, Year (Reference No.)	Duration of Study	Total No. of Participants	Mean Age, years	% Female	Mean Duration of Diabetes Mellitus, years	SGLT-2 Inhibitor (Dose) or Exposure Group	Comparator (Dose)	Definition of DKA
Randomized Controlled Trials
Barnett, 2014 (12)	104 weeks	741	63.9	41.7	NR	Empagliflozin (10 mg or 25 mg)	Placebo	Assessed through the reporting of adverse events up to 7 days after the last dose of the study drug; coded using MedDRA (https://www.meddra.org/) version 15.0
Bode, 2015 (13)	52 weeks	714	63.6	44.5	11.7	Canagliflozin (100 mg or 300 mg)	Placebo	Data on events were collected through nonsystematic assessment; coded using MedDRA versions 14.0 and 16.0
Roden, 2015 (14)	76 weeks	899	55.0	38.7	NR	Empagliflozin (10 mg or 25 mg)	Placebo/sitagliptin (100 mg)	Assessed through the reporting of adverse events; coded using MedDRA version 16.0
Rosenstock, 2014 (15)	52 weeks	563	56.7	54.5	NR	Empagliflozin (10 mg or 25 mg)	Placebo	Assessed through the reporting of adverse events up to 7 days after the last dose of the study drug; coded using MedDRA version 16.0
Tikkanen, 2015 (16)	12 weeks	823	60.2	39.9	NR	Empagliflozin (10 mg or 25 mg)	Placebo	Assessed through the reporting of adverse events; coded using MedDRA version 15.0
Wilding, 2013 (17)	52 weeks	469	56.8	49.0	9.6	Canagliflozin (100 mg or 300 mg)	Placebo	Assessed through the reporting of adverse events; coded using MedDRA version 14.0 for week 26 and MedDRA version 15.0 for week 52
Zinman, 2015 (18)	206 weeks	7,020	63.1	28.5	NR	Empagliflozin (10 mg or 25 mg)	Placebo	Assessed on the basis of adverse events that occurred during treatment or within 7 days after the last dose of the study drug; coded using MedDRA version 18.0
Perkovic, 2019 (19)	245 weeks	4,401	36.0	33.9	15.8	Canagliflozin (100 mg)	Placebo	All ketone-related events, adjudicated by an independent adjudication committed on the basis of clinical presentation and predefined biochemical parameters; coded using MedDRA version 21.0
Wiviott, 2019 (20)	280 weeks	17,160	63.9	37.4	10.5	Dapagliflozin (10 mg)	Placebo	Adjudicated by the TIMI Clinical Events Committee, who were unaware of treatment assignments; coded using MedDRA version 21.0
Neal, 2017 (21)	52 weeks	10,142	63.0	36.0	13.5	Canagliflozin (100–300 mg)	Placebo	Events adjudicated by the Baim Institute for Clinical Research (Boston, Massachusetts); coded using MedDRA version 19.1
McMurray, 2019 (22)	127 weeks	4,744	66.0	24.0	NR	Dapagliflozin (5 mg or 10 mg)	Placebo	Independent, blinded DKA adjudication committee assessed all potential ketoacidosis events (definite or probable); coded using MedDRA version 22.0
Pollock, 2019 (23)	24 weeks	461	64.5	29.3	17.9	Dapagliflozin (10 mg)	Placebo	Assessed through the reporting of adverse events; coded using MedDRA version 21.0
Lavalle-González, 2013 (24)	52 weeks	1,284	55.4	52.9	6.9	Canagliflozin (100 mg or 300 mg)	Placebo/sitagliptin (100 mg)	Assessed through the reporting of adverse events; coded using MedDRA version 14.0 for week 26 and MedDRA version 15.0 for week 52
Frías, 2016 (25)	28 weeks	685	54.2	52.1	7.4	Dapagliflozin (10 mg)	Exenatide (2 mg)	Assessed through the reporting of adverse events; coded using MedDRA version 20.1
Hollander, 2018 (26)	52 weeks	1,325	58.2	51.5	7.5	Ertugliflozin (5 mg or 15 mg)	Glimepiride	Assessed through the reporting of adverse events; coded using MedDRA version 20.0
Gallo, 2019 (27)	104 weeks	621	56.6	53.6	7.9	Ertugliflozin (5 mg or 15 mg)	Placebo/glimepiride	Potential cases of ketoacidosis were reviewed by a blinded internal case review committee; coded using MedDRA version 20.0
Pratley, 2018 (28)	52 weeks	1,232	55.1	46.1	6.9	Ertugliflozin (5 mg or 15 mg)	Sitagliptin (100 mg)	Assessed through the reporting of adverse events; coded using MedDRA version 19.0
Haering, 2015 (29)	76 weeks	666	57.1	49.1	NR	Metformin + sulfonylureas + empagliflozin (10 mg or 25 mg)	Metformin + sulfonylureas	Assessed through the reporting of adverse events; coded using MedDRA version 16.0
Observational Studies
Fralick, 2017 (30)^a	180 days	76,090	54.6	47.3	NR	New users of SGLT-2 inhibitors between April 1, 2013, and December 31, 2014	New users of DPP-4 inhibitors between April 1, 2013, and December 31, 2014	Hospitalization for DKA (using the primary position ICD-9 code) within 180 days after drug initiation
Kim, 2018 (31)^a	3.5 years	112,650	53.2	44.8	NR	New use of SGLT-2 inhibitors, from 2014 to 2017, with no use in the prior 365 days	New use of DPP-4 inhibitors, from 2014 to 2017, with no use in the prior 365 days	Hospitalization for DKA
Hamblin, 2019 (32)^b	26 months	162	64.8	41.9	≤NR	Users of SGLT-2 inhibitors	Nonusers of SGLT-2 inhibitors	Emergency department visit with DKA or development of DKA during hospital admission using the Australian modification of the ICD-10 (codes E1010, E1012, E1015, E1016, E1111, E1112, E1115, E1116, E1311, E1312, E1411, and E1412); then verified by study physicians to confirm that DKA was present as defined by the ADA criteria (blood pH < 7.30, bicarbonate concentration ≤ 18 mmol/L, and elevated ketone levels). Euglycemic DKA was defined as blood glucose concentration < 250 mg/dL (<14 mmol/L).
McGurnaghan, 2019 (33)	12 years	238,876	65.8	43.5	9.0	Any initiation of dapagliflozin between November 2012 and 2016	No prescription for dapagliflozin from 2004 to 2016	DKA events were captured using linkage to national hospitalization records and death data; ICD-10 codes were used for cause of admission (code E10.0 or E10.1)
Ueda, 2018 (34)^a	3.5 years	34,426	61.0	39.0	NR	First prescription for an SGLT-2 inhibitor during the study period, without any previous use	First prescription for GLP-1 receptor agonists during the study period, without any previous use	Any hospital admission using ICD-10 codes (Sweden: codes E110A and E111A; Denmark: codes E111, E131, and E141)
Wang, 2017 (35)^a	18 months	60,196	53.8	NR	NR	First prescription for SGLT-2 inhibitors with no use in at least the prior 12 months	First prescription for sulfonylureas, DPP-4 inhibitors, GLP-1 receptor agonists, thiazolidinedione, insulin, and other miscellaneous antidiabetic agents with no use in at least the prior 12 months	DKA ascertained by hospital or emergency room diagnosis
Wang, 2019 (36)^a	Approximately 4–4.5 years	739,993	56.4	47.1	NR	New prescription dispensing record for SGLT-2 inhibitors, without prior use of the respective AHA ever recorded in the database	New prescription dispensing record for an insulinotropic AHA (DPP-4 inhibitors, GLP-1 receptor agonists, sulfonylureas, meglitinides) without prior use of the respective AHA ever being recorded in the database	DKA event that occurred after the index date of an AHA therapy, identified from a diagnosis code recorded in inpatient or emergency room claims. To be considered an incident event, DKA occurring after exposure to a new AHA had to occur ≥30 days after any preindex DKA event.
Douros, 2020 (37)^a	5 years	417,514	63.9	41.5	NR	New user of an SGLT-2 inhibitor, alone or in combination with other antidiabetic drugs, between 2013 and 2018	New users of DPP-4 inhibitors from 2013 to 2018, with no use in the prior 365 days	Hospitalization with a primary diagnosis of DKA or an emergency department visit using ICD-10 (Canadian adaptation) diagnostic codes E11.10, E11.12, E13.10, E13.12, E14.10, and E14.12 (or E11.1, E13.1, and E14.1 in the CPRD).

Abbreviations: ADA, American Diabetes Association; AHA, antihyperglycemic agent; CPRD, Clinical Practice Research Datalink; DKA, diabetic ketoacidosis; DPP-4, dipeptidyl peptidase 4; GLP-1, glucagon-like peptide 1; ICD-9, International Classification of Diseases, Ninth Revision; ICD-10, International Classification of Diseases, Tenth Revision; MedDRA, Medical Dictionary for Regulatory Activities; NR, not reported; SGLT-2, sodium/glucose cotransporter 2; TIMI, Thrombolysis in Myocardial Infarction.
^aMatched cohort.
^bReported for DKA cases only.

Table 2. Numbers of Events and Quality of Studies Included in a Meta-Analysis of Sodium/Glucose Cotransporter 2 Inhibitors and Risk of Diabetic Ketoacidosis, 2013–2020
First Author, Year (Reference No.)	SGLT-2 Inhibitor Users		Controls		Study Quality
First Author, Year (Reference No.)	No. of Events	Total No. of Participants	No. of Events	Total No. of Participants	Study Quality
Placebo-Controlled RCTs
Barnett, 2014 (12)	0	419	1	319	Good
Bode, 2015 (13)	1	477	0	237	Good
Roden, 2015 (14)	1	447	1	229	Good
Rosenstock, 2014 (15)	1	375	1	188	Good
Tikkanen, 2015 (16)	0	552	1	272	Good
Wilding, 2013 (17)	1	313	0	156	Good
Zinman, 2015 (18)	4	4,687	1	2,333	Good
Perkovic, 2019 (19)	11	2,200	1	2,197	Good
Wiviott, 2019 (20)	27	8,574	12	8,569	Good
Neal, 2017 (21)	12	5,790	3	4,344	Good
McMurray, 2019 (22)	3	2,368	1	2,368	Good
Pollock, 2019 (23)	1	145	0	148	Good
Active-Comparator RCTs
Roden, 2015 (14)	1	447	0	233	Good
Lavalle-González, 2013 (24)	0	735	1	549	Good
Frías, 2016 (25)	0	233	1	230	Good
Hollander, 2018 (26)	1	888	0	437	Fair
Gallo, 2019 (27)	1	205	0	209	Good
Pratley, 2018 (28)	1	985	0	247	Good
Haering, 2015 (29)	1	1,652	1	1,050	Good
Observational Studies
Fralick, 2017 (30)	55	38,045	26	38,045	Good
Kim, 2018 (31)	29	56,325	34	56,325	Good
McGurnaghan, 2019 (33)	13	8,516	664	230,360	Good
Ueda, 2018 (34)	19	17,213	11	17,213	Good
Wang, 2017 (35)	37	30,196	18	30,196	Fair
Wang, 2019 (36)	754	192,261	528	192,261	Good
Douros, 2020 (37)	372	202,186	103	202,186	Good