QPPV.COM

Historical Signal Evaluation: SGLT2 Inhibitors and Diabetic Ketoacidosis

An evidence-based evaluation of the SGLT2 inhibitor–diabetic ketoacidosis signal, examining atypical presentation, pharmacological plausibility, spontaneous reports, clinical-trial evidence, observational studies, risk factors, PRAC assessment and regulatory risk minimisation.

Audio Lesson 18 min
Knowledge Assessment Test your understanding of this article. Take the assessment →

Historical Signal Evaluation: SGLT2 Inhibitors and Diabetic Ketoacidosis

Introduction

The association between sodium-glucose cotransporter-2 (SGLT2) inhibitors and diabetic ketoacidosis (DKA) is one of the more instructive modern examples of pharmacovigilance signal evaluation.

The signal is particularly valuable because it illustrates several problems that can make an adverse drug reaction difficult to recognise.

First, the event is uncommon.

Second, DKA was not initially expected to be a major adverse effect of a class of medicines developed to lower blood glucose.

Third, some affected patients had only moderately elevated blood glucose concentrations, producing what became widely recognised as an atypical or euglycaemic presentation.

Fourth, the early randomised clinical-trial evidence did not provide a strong statistical signal because the event was rare.

Finally, the biological mechanism is plausible but multifactorial. Some components of the mechanism are well established pharmacologically, while other proposed pathways remain incompletely understood.

The European Pharmacovigilance Risk Assessment Committee (PRAC) reviewed SGLT2 inhibitors in 2015–2016 and recommended measures to minimise the risk of DKA. The review ultimately led to changes in European product information and clinical recommendations. EMA stated that the benefits of SGLT2 inhibitors continued to outweigh their risks in the authorised type 2 diabetes population. 1

This case therefore demonstrates an important pharmacovigilance principle:

A signal does not need to begin with a statistically significant randomised-trial result to become clinically and regulatorily important.

The more useful question is:

How did the totality of evidence change the interpretation of the risk, and how did regulators translate that evidence into practical risk minimisation?

1. Define the Signal

The safety concern was diabetic ketoacidosis associated with SGLT2 inhibitor exposure.

The class acts by inhibiting SGLT2 in the renal proximal tubule, reducing glucose reabsorption and increasing urinary glucose excretion.

The initial pharmacological effect therefore appears straightforward:

More glucose is lost through the urine, reducing blood glucose.

The signal became more complicated when cases of DKA began to be reported in patients receiving SGLT2 inhibitors, including cases in which blood glucose was not markedly elevated.

This created two related but distinct pharmacovigilance questions:

  1. Do SGLT2 inhibitors increase the risk of DKA?
  2. Can SGLT2 inhibitors alter the clinical presentation of DKA so that the diagnosis is delayed?

The second question is especially important.

A medicine can create pharmacovigilance risk not only by increasing the incidence of an adverse event, but also by changing how the event presents.

2. Why the Signal Was Unexpected

DKA is classically associated with substantial insulin deficiency and marked hyperglycaemia.

The conventional clinical picture includes:

SGLT2 inhibitors reduce plasma glucose through urinary glucose loss.

That immediately creates a potential diagnostic problem.

A patient can develop substantial ketone production and metabolic acidosis while the expected degree of hyperglycaemia is attenuated.

The resulting presentation has commonly been described as:

euglycaemic diabetic ketoacidosis (euDKA).

The term should not be interpreted too rigidly.

Patients may have glucose concentrations that are normal or only modestly elevated rather than completely normal.

The clinically important point is that the glucose concentration may not be sufficiently high to trigger the usual immediate suspicion of DKA.

3. The Pharmacology: What Is Established?

The primary pharmacology of SGLT2 inhibition is well established.

SGLT2 is responsible for a substantial proportion of renal glucose reabsorption.

Inhibition of the transporter:

These effects explain the intended therapeutic action of the class.

They also provide a plausible metabolic context for ketoacidosis.

The key distinction is that the pharmacological effect is established, while the complete chain from SGLT2 inhibition to individual episodes of DKA is more complex.

4. Proposed Mechanisms of Ketoacidosis

Several mechanisms have been proposed to explain why SGLT2 inhibition can promote ketoacidosis.

The most commonly discussed pathway involves changes in the insulin-to-glucagon balance.

Reduced plasma glucose can reduce endogenous insulin secretion.

At the same time, SGLT2 inhibition has been associated with increased glucagon signalling.

A lower insulin-to-glucagon ratio favours:

The resulting ketonaemia can progress to ketoacidosis when compensatory mechanisms are overwhelmed.

Another contributing mechanism is volume depletion.

SGLT2 inhibitors cause urinary glucose and sodium loss, producing osmotic diuresis and natriuresis.

Volume depletion can increase counter-regulatory hormonal responses and may amplify the metabolic stress associated with DKA.

Renal handling of ketone bodies has also been proposed as a contributor.

However, not every mechanistic component has the same level of evidence.

A useful distinction is:

Mechanistic proposition Evidence status
SGLT2 inhibition increases urinary glucose excretion Established pharmacology
SGLT2 inhibition lowers plasma glucose Established pharmacology
SGLT2 inhibitors can promote osmotic diuresis Established pharmacology
Reduced insulin availability can favour lipolysis and ketogenesis Established metabolic physiology
SGLT2 inhibition alters glucagon signalling Supported by experimental and clinical studies
SGLT2 inhibition directly stimulates pancreatic alpha cells Proposed and supported by experimental evidence, but the clinical contribution remains incompletely defined
Altered renal ketone handling contributes materially to DKA Biologically plausible and supported by experimental evidence, but not sufficient by itself to explain every clinical case
A single mechanism explains all SGLT2-associated DKA Not established

This distinction matters.

The mechanism can be biologically persuasive without proving that the drug caused a particular patient's episode.

Mechanistic evidence should therefore be integrated with clinical and epidemiological evidence rather than substituted for it.

5. The Role of Catabolic Stress

One of the most important features of the signal is that SGLT2 inhibitor-associated DKA often occurs in the presence of additional metabolic stress.

Reported precipitants include:

These factors share a common physiological direction.

They can increase dependence on fat oxidation and ketone production or reduce available insulin.

This leads to a useful conceptual model:

SGLT2 inhibition may lower the metabolic threshold at which a susceptible patient develops clinically important ketogenesis during a catabolic stress state.

That statement is more defensible than claiming that SGLT2 inhibition independently causes DKA in every exposed patient.

The modern evidence suggests that susceptibility and precipitating factors matter.

6. The Initial Evidence: Spontaneous Reports

The early signal was driven importantly by post-marketing reports.

This is an important teaching point because spontaneous reports are sometimes dismissed because they lack a denominator.

That is the wrong interpretation.

A spontaneous report cannot normally estimate incidence or relative risk reliably.

It can, however, identify:

In this case, reports of DKA in patients receiving SGLT2 inhibitors became important because the presentation was sometimes atypical.

The cases were not merely reports of hyperglycaemic DKA in patients with uncontrolled diabetes.

Some cases involved relatively modest blood glucose concentrations.

That clinical feature increased the plausibility that the medicines were changing the phenotype of the event.

7. Why Randomised Trials Initially Provided Limited Reassurance

Randomised controlled trials are generally powerful tools for evaluating adverse effects.

However, their usefulness depends heavily on event frequency.

DKA is uncommon.

A 2017 meta-analysis of 72 randomised trials found that only nine trials reported at least one DKA event.

The analysis did not identify a statistically significant increased risk at the class level:

MH-OR 1.14 (95% CI 0.45–2.88).

The authors concluded that the trial evidence did not demonstrate an increased risk and considered the risk negligible when the medicines were properly prescribed.

This finding illustrates a critical pharmacovigilance lesson.

A negative randomised-trial analysis does not necessarily exclude a rare adverse reaction.

If the number of events is very small, the confidence interval can remain wide.

Furthermore, clinical trials may:

Therefore:

The absence of a statistically significant signal in clinical trials is evidence, but its ability to exclude a rare adverse reaction depends on event numbers and trial design.

8. Why Spontaneous Reports and Trials Can Appear to Disagree

At first glance, the evidence could appear contradictory.

Spontaneous reports suggested a serious adverse event.

Randomised trials did not clearly demonstrate an increased risk.

These findings are not necessarily inconsistent.

They answer different questions.

Spontaneous reporting

Useful for:

Weak for:

Randomised trials

Useful for:

Weak when:

A good signal evaluator should therefore integrate both sources rather than choosing one and dismissing the other.

9. The Regulatory Review

The European Commission initiated an Article 20 procedure for SGLT2 inhibitors in June 2015.

The review covered:

PRAC completed its review in February 2016.

The procedure was subsequently reviewed by the Committee for Medicinal Products for Human Use (CHMP), and the European Commission issued the legally binding final decision in April 2016. 2

The regulatory outcome was risk minimisation.

The medicines were not withdrawn.

The benefit-risk balance remained favourable for their authorised use in type 2 diabetes.

This is an important distinction.

A safety signal can result in:

The presence of a serious adverse reaction does not automatically imply that the medicine's overall benefit-risk balance is unacceptable.

10. What PRAC Identified

PRAC concluded that rare cases of DKA had occurred in patients receiving SGLT2 inhibitors.

Importantly, some cases were atypical because blood glucose was not as high as would ordinarily be expected in DKA.

EMA specifically warned healthcare professionals that DKA should be considered even when blood glucose was not particularly high.

This changed the practical diagnostic question.

Instead of:

Does this patient have very high glucose?

the clinician needed to consider:

Could this patient have ketoacidosis despite relatively modest glucose?

That is a significant pharmacovigilance outcome.

The signal therefore affected not only the estimated probability of an adverse event but also the clinical recognition pathway.

11. The Regulatory Risk Factors

PRAC identified several situations associated with increased risk.

These included:

The recommendations also addressed temporary interruption of SGLT2 inhibitor treatment during major surgery or serious acute illness.

This is an example of targeted risk minimisation.

The regulatory response did not simply state:

Do not use SGLT2 inhibitors.

Instead, it identified circumstances in which the risk could become more clinically important.

That is a more sophisticated pharmacovigilance response.

12. The Importance of Insulin Reduction

Insulin reduction deserves particular attention.

DKA fundamentally involves inadequate insulin action relative to metabolic demand.

An SGLT2 inhibitor can lower glucose sufficiently that a patient or clinician may perceive less need for insulin.

However, glucose control and suppression of ketogenesis are not identical physiological requirements.

Insulin has important anti-lipolytic and anti-ketogenic effects.

Therefore, reducing insulin too aggressively can increase ketogenesis even when glucose concentrations appear acceptable.

This is one reason the pharmacovigilance evaluation should not treat the drug as the only causal factor.

A useful causal model is:

SGLT2 inhibition + relative insulin deficiency + metabolic stress → increased susceptibility to ketogenesis

This is a conceptual model rather than a claim that every component is required in every case.

13. Euglycaemia as a Diagnostic Hazard

The most clinically important feature of the signal may be the potential disconnect between glucose concentration and severity of metabolic disturbance.

Consider two simplified patients.

Patient A

A patient with diabetes presents with:

DKA is immediately considered.

Patient B

A patient receiving an SGLT2 inhibitor presents with:

The glucose concentration may initially make DKA seem less likely.

Yet the second presentation can still represent serious ketoacidosis.

This is why the signal had clinical significance beyond a simple change in incidence.

It changed the diagnostic threshold.

14. Evidence From Observational Studies

As the signal developed, observational studies provided additional information.

One important question was whether the association persisted when SGLT2 inhibitors were compared with other glucose-lowering medicines.

A nationwide Korean cohort study included 56,325 new SGLT2 inhibitor users matched to 56,325 DPP-4 inhibitor users.

The study found no statistically significant increase in hospitalisation for DKA:

HR 0.956 (95% CI 0.581–1.572).

The authors did, however, observe variation in incidence over time and suggested that some clinical characteristics might identify patients with greater susceptibility.

This study is useful precisely because it did not simply confirm the suspected association.

A rigorous signal evaluation must include studies that weaken as well as strengthen the hypothesis.

A signal should survive attempts to falsify it.

15. Meta-Analysis of Randomised Evidence

Later randomised evidence provided a different picture from the earliest meta-analysis.

A 2020 systematic review and meta-analysis included:

DKA occurred in:

The pooled Peto odds ratio was:

2.13 (95% CI 1.38–3.27).

The absolute risk difference was approximately:

1.7 additional events per 1,000 patients over five years, with a 95% CI of approximately 0.6 to 3.4 additional events per 1,000.

The authors rated the evidence as high quality.

This is an important development.

The evidence base had moved from:

No clear statistical signal in early trials

toward:

A statistically detectable increase in DKA risk when a much larger accumulated trial population was considered.

This illustrates why signal evaluation should be dynamic.

An early negative analysis should not automatically become a permanent conclusion.

16. Why Absolute Risk Matters

The relative effect estimate can appear alarming.

An odds ratio of approximately 2 means that the relative odds are roughly doubled under the assumptions of the analysis.

But the event itself remains uncommon.

The 2020 meta-analysis estimated an absolute difference of approximately 1.7 additional events per 1,000 patients over five years.

That corresponds to a small absolute increase.

This matters because SGLT2 inhibitors provide important benefits in appropriate patients, including effects on glycaemia and, depending on the drug and indication, cardiovascular and renal outcomes.

A pharmacovigilance evaluation therefore needs to consider both:

The appropriate clinical question is not:

Is the relative risk increased?

It is:

How large is the absolute risk, in which patients is it concentrated, and how does it compare with the benefits of treatment?

17. Evidence Is Not Uniform Across Populations

The risk of DKA is not necessarily identical across all uses of SGLT2 inhibitors.

This is particularly important when considering type 1 diabetes.

SGLT2 inhibitors have been investigated in type 1 diabetes, but DKA risk is more prominent in this population because endogenous insulin reserve is limited.

A meta-analysis of placebo-controlled trials in type 1 diabetes found an increased risk of DKA and identified factors such as:

The analysis also found that treatment-related factors and baseline characteristics could explain substantial variation between studies.

This supports an important general principle:

Drug-associated risk may be modified strongly by the physiological reserve of the patient and the clinical context in which the drug is used.

18. The Mechanism Is Multifactorial

The clinical evidence is consistent with a multifactorial mechanism.

A simplified pathway is:

  1. SGLT2 inhibition causes glycosuria.
  2. Plasma glucose decreases.
  3. Insulin secretion may decrease.
  4. The insulin-to-glucagon balance shifts toward ketogenesis.
  5. Lipolysis increases.
  6. Free fatty acids reach the liver.
  7. Hepatic ketogenesis increases.
  8. Concurrent fasting, illness, dehydration or insulin reduction can amplify the process.
  9. Glucose remains lower than expected because of continued urinary glucose loss.
  10. Ketoacidosis may therefore develop without marked hyperglycaemia.

Much of this pathway is physiologically coherent.

But the complete human mechanism remains more complicated.

For example, the exact contribution of:

is not completely resolved.

This distinction should remain explicit in a pharmacovigilance assessment.

What is established

SGLT2 inhibition causes glycosuria and changes renal glucose handling.

What is strongly supported

SGLT2 inhibitors can alter the metabolic environment in a direction favouring ketogenesis, particularly when insulin availability is reduced.

What remains incompletely established

The precise contribution of each proposed pathway to individual episodes of SGLT2-associated ketoacidosis.

This is how mechanistic plausibility should be communicated.

19. Can the Signal Be Explained by Confounding?

Confounding is particularly important for DKA.

Patients prescribed SGLT2 inhibitors may have:

If these factors are not adequately controlled, observational studies can overestimate the association.

However, confounding cannot be assumed simply because an observational study is involved.

The evaluator should ask:

The appropriate conclusion is therefore neither:

Observational evidence proves causality.

nor:

Observational evidence is unreliable.

Instead:

Observational evidence can materially strengthen or weaken a signal when the design appropriately addresses major sources of bias.

20. The Problem of Exposure Windows

Timing is another important dimension.

If DKA occurs months after discontinuation of an SGLT2 inhibitor, a direct causal interpretation becomes less straightforward.

If it occurs during treatment or shortly after exposure in the setting of a recognised metabolic precipitant, temporal plausibility is stronger.

However, the persistence of pharmacodynamic effects after discontinuation means that the evaluator should not use an arbitrary exposure window without understanding the pharmacology.

The correct question is:

Is the selected exposure window biologically and clinically justified?

Exposure windows should therefore be based on:

21. Dechallenge and Rechallenge

Dechallenge and rechallenge can provide useful evidence in individual case assessment.

If a patient develops ketoacidosis while receiving an SGLT2 inhibitor and improves after discontinuation, that temporal sequence may support causality.

However, DKA treatment itself includes:

Therefore, improvement after stopping the medicine is not specific evidence of drug causality.

Rechallenge can theoretically provide stronger evidence, but intentional rechallenge after serious DKA would generally be clinically inappropriate.

Consequently, the absence of rechallenge evidence is not surprising and should not be treated as a major weakness in the signal.

This is another example of why causality frameworks must be applied intelligently rather than mechanically.

22. What PRAC Recommended

PRAC recommended several practical measures.

Healthcare professionals were advised to consider DKA in patients taking SGLT2 inhibitors who had symptoms consistent with the condition, even if blood glucose was not particularly high.

Risk factors were to be considered.

PRAC also recommended temporary interruption of treatment in patients:

If DKA was suspected or confirmed, treatment should be stopped immediately and should not be restarted unless another clear cause of ketoacidosis was identified and resolved.

The recommendations therefore targeted both:

  1. prevention;
  2. early recognition.

This is an important distinction.

Risk minimisation is not always about preventing the adverse reaction completely.

It can also reduce harm by shortening the time between onset and diagnosis.

23. Why the Benefit-Risk Balance Remained Positive

PRAC did not conclude that SGLT2 inhibitors should be withdrawn.

The benefits continued to outweigh the risks for their authorised use in type 2 diabetes.

That conclusion is understandable when the evidence is considered quantitatively.

The adverse event is serious but rare.

The relative risk appears increased.

The absolute excess risk is small in the overall treated population.

The risk can be reduced through:

At the same time, SGLT2 inhibitors provide clinically meaningful benefits.

The appropriate regulatory response was therefore targeted risk minimisation rather than withdrawal.

This is a useful example of benefit-risk reasoning in pharmacovigilance.

24. The Evolution of the Evidence

The history can be summarised as follows.

Stage Evidence Interpretation
Initial post-marketing period Serious DKA case reports, including atypical presentations Signal generation
Early clinical-trial evidence Few DKA events; no statistically significant increase in early meta-analysis Did not exclude a rare risk
Regulatory review 2015–2016 Case reports, clinical information and mechanistic considerations Sufficient concern for regulatory action
Subsequent observational studies Mixed results depending on comparator and population Added important context
2020 RCT meta-analysis 85 DKA events across 39 trials; increased risk detected Strengthened evidence for association
Later mechanistic literature Multifactorial metabolic explanation Increased biological plausibility
More recent meta-analyses Continued evidence of increased risk, with heterogeneity Refined magnitude and risk modifiers

This progression is exactly what a mature pharmacovigilance system should produce.

A signal starts as a hypothesis.

Evidence is accumulated.

The hypothesis becomes more precise.

Risk factors are identified.

The regulatory response is refined.

The signal remains subject to reassessment.

25. What the SmPC Evolution Teaches

The product-information changes provide another useful teaching example.

The regulatory wording evolved around several key concepts:

This is a good example of translating scientific uncertainty into practical clinical language.

The wording does not need to explain every mechanistic hypothesis.

It needs to communicate what clinicians and patients need to know to reduce harm.

That distinction is important.

A pharmacovigilance assessment may contain extensive mechanistic discussion, while the final SmPC may need only a concise and clinically actionable warning.

26. A Structured Evidence Assessment

A practical signal evaluation can be summarised using an evidence matrix.

Domain Evidence Assessment
Seriousness DKA can be life-threatening Strong concern
Temporal association Cases occurred during or after exposure Supportive
Unexpectedness DKA was not initially expected as a major class risk Important signal feature
Clinical phenotype Some cases had modest glucose elevation Strongly relevant
Biological plausibility Multiple metabolic pathways support ketogenesis Supportive
Spontaneous reports Multiple serious cases generated concern Signal-generating
Early RCT evidence Initially inconclusive because events were rare Did not exclude risk
Later RCT evidence Increased risk demonstrated in larger evidence base Supportive
Observational evidence Mixed across studies and comparators Supportive but heterogeneous
Confounding Diabetes severity and acute illness may contribute Important limitation
Risk factors Fasting, illness, dehydration and insulin reduction identified Supports targeted prevention
Rechallenge Generally unavailable and clinically inappropriate Limited relevance
Regulatory assessment PRAC recommended risk minimisation Supports regulatory significance
Benefit-risk Benefits remained favourable in authorised populations Supports targeted rather than prohibitive action

27. What Would Have Been a Poor Signal Evaluation?

Several analytical errors would have produced a weak assessment.

Error 1: Looking only at spontaneous reports

This would identify the signal but could not quantify the risk.

Error 2: Looking only at early RCTs

This could incorrectly conclude that no risk existed.

Error 3: Treating every DKA case as caused by the drug

This would ignore insulin reduction, fasting, infection and other precipitants.

Error 4: Treating euglycaemia as proof of causality

The atypical presentation is highly relevant, but it is not by itself proof that the drug caused the event.

Error 5: Ignoring the comparator

Different comparator populations can produce materially different estimates.

Error 6: Ignoring absolute risk

A relative risk increase can coexist with a small absolute increase.

Error 7: Treating a proposed mechanism as established fact

Mechanistic plausibility supports causality but does not prove individual causation.

Error 8: Assuming regulatory action means the causal question is completely settled

PRAC action means the evidence justified a regulatory response. It does not imply that every mechanistic or quantitative question has been resolved.

28. A Worked Case

Consider a patient with type 2 diabetes who is taking an SGLT2 inhibitor.

The patient develops an acute infection and eats very little for several days.

Insulin is reduced because glucose readings remain relatively low.

The patient subsequently develops:

A simplistic assessment might say:

Glucose is not high enough for DKA.

A pharmacovigilance-informed assessment asks a different set of questions:

Exposure

Is the patient receiving an SGLT2 inhibitor?

Temporal relationship

Did the metabolic event occur during treatment or within a biologically plausible period after exposure?

Alternative causes

Is there infection, fasting, dehydration, insulin deficiency or another precipitant?

Phenotype

Is this ketoacidosis despite relatively modest glucose?

Mechanism

Is the pharmacology capable of promoting a metabolic state favouring ketogenesis?

Regulatory relevance

Is this a recognised adverse reaction for the drug class?

Causality

Does the drug appear to be:

The most defensible conclusion may be:

SGLT2 inhibitor exposure was a plausible contributor to DKA in the setting of acute infection, reduced food intake, insulin reduction and dehydration.

That conclusion is more informative than either:

The drug caused DKA.

or:

The infection caused DKA.

Real pharmacovigilance cases are often multifactorial.

29. What This Case Teaches About Causality

The SGLT2-DKA signal demonstrates that causality is often probabilistic rather than binary.

The evidence can be considered as a convergence of:

No individual element is necessarily decisive.

The strength comes from convergence.

At the same time, alternative explanations must remain visible.

A patient with severe infection, prolonged fasting and insulin omission has strong independent reasons to develop DKA.

The presence of an SGLT2 inhibitor may increase susceptibility without being the sole cause.

This distinction is important for:

30. What a QPPV Should Take From This Case

30.1 A signal can emerge from an unexpected phenotype

The important observation was not simply "DKA occurred."

It was:

DKA occurred in a patient receiving an SGLT2 inhibitor, sometimes without the expected degree of hyperglycaemia.

That difference changed clinical recognition.

30.2 Rare events require evidence integration

No single evidence source was sufficient.

The signal required integration of:

30.3 A negative early trial result is not necessarily closure

When an event is rare, the absence of statistical significance may reflect limited information rather than evidence of no risk.

30.4 Risk factors can be more actionable than the overall relative risk

Identifying:

allowed regulators to design practical risk-minimisation measures.

30.5 Mechanism should support, not replace, evidence

The metabolic mechanism is compelling.

But mechanistic plausibility does not establish causality in an individual case.

30.6 Absolute risk belongs in the assessment

A serious event can have a meaningful regulatory significance even when its absolute incidence is low.

The evaluator should therefore report both relative and absolute measures where reliable estimates exist.

30.7 Risk minimisation can target diagnosis

The response to a signal may involve improving recognition rather than simply reducing exposure.

In this case, recognising DKA despite modest glucose concentrations was central.

30.8 Regulatory action can coexist with a positive benefit-risk balance

The correct response to a serious adverse event is not automatically withdrawal.

The appropriate question is whether risk can be adequately managed while preserving clinically meaningful benefit.

31. Overall Causal Assessment

The accumulated evidence supports a causal association between SGLT2 inhibitor exposure and an increased risk of diabetic ketoacidosis.

The evidence is stronger today than it was when the first reports emerged.

Several factors support causality:

Important qualifications remain.

DKA is multifactorial.

Acute illness, fasting, dehydration and insulin reduction can independently precipitate the event.

The contribution of individual mechanistic pathways is not completely established.

Risk is not uniform across all patients or indications.

The absolute risk remains low in the overall treated population.

The most defensible conclusion is therefore:

SGLT2 inhibitors increase the risk of diabetic ketoacidosis, including atypical presentations in which blood glucose may be only modestly elevated. The risk appears to be strongly influenced by patient physiology and precipitating factors such as insulin deficiency, reduced food intake, dehydration, acute illness and surgery. The pharmacological basis for increased ketogenesis is well supported, although the relative contribution of individual mechanistic pathways remains incompletely established. The regulatory response of targeted risk minimisation is consistent with the serious but uncommon nature of the event and the continued therapeutic benefits of the class.

32. Lessons for Signal Management

Lesson 1: Define the clinical phenotype

The signal was not simply DKA.

The atypical glycaemic presentation was itself an important part of the signal.

Lesson 2: Treat spontaneous reports as hypothesis-generating evidence

They cannot reliably quantify incidence but can reveal unexpected clinical patterns.

Lesson 3: Understand why trial evidence may initially be negative

Rare events can remain invisible in relatively small randomised datasets.

Lesson 4: Reassess when the evidence base grows

A conclusion reached from nine DKA-reporting trials should not be treated as permanently definitive when substantially more evidence becomes available.

Lesson 5: Examine precipitating factors

Multifactorial adverse reactions require identification of the conditions that convert susceptibility into an event.

Lesson 6: Separate established pharmacology from mechanistic hypotheses

Do not present every proposed pathway as established fact.

Lesson 7: Use appropriate comparators

Comparing with untreated patients and comparing with alternative glucose-lowering drugs answer different questions.

Lesson 8: Consider absolute risk

Relative risk alone is insufficient for benefit-risk interpretation.

Lesson 9: Consider the clinical consequences of altered presentation

A drug may increase harm by making an event harder to recognise, even if the event itself remains uncommon.

Lesson 10: Translate evidence into actionable risk minimisation

Risk factors, warning symptoms and temporary treatment interruption can be more useful than a generic statement that "DKA is possible."

Lesson 11: Keep the historical decision separate from current knowledge

The 2015–2016 regulatory review should be assessed using evidence available at that time.

Current signal assessment should incorporate subsequent evidence.

Lesson 12: Do not confuse regulatory significance with complete scientific certainty

A regulator may act when the evidence is sufficient to justify a warning or risk-minimisation measure even while quantitative and mechanistic uncertainty remains.

Conclusion

The SGLT2 inhibitor–DKA signal is a particularly useful pharmacovigilance case because it demonstrates how a serious adverse reaction can emerge through the interaction of pharmacology, patient susceptibility and clinical circumstances.

The initial signal was driven substantially by post-marketing reports of DKA, including cases in which hyperglycaemia was less pronounced than expected.

Early randomised evidence did not demonstrate a clear increase in risk, but the number of events was small.

Regulatory review nevertheless identified sufficient concern to justify risk minimisation.

Subsequent evidence strengthened the association and provided better estimates of absolute risk.

At the same time, later research reinforced the importance of confounding, patient characteristics and precipitating factors.

The mechanism is biologically coherent:

SGLT2 inhibition causes glycosuria and changes glucose and insulin physiology in a direction that can favour ketogenesis, particularly when metabolic stress or insulin deficiency is present.

But not every proposed mechanistic detail is equally established.

That distinction is important.

The strongest pharmacovigilance conclusion is therefore not simply:

"SGLT2 inhibitors cause euglycaemic DKA."

It is:

"SGLT2 inhibitors increase the risk of DKA, and they can produce an atypical presentation in which hyperglycaemia may be absent or modest. The risk is influenced by metabolic context and patient susceptibility, and the mechanism is multifactorial."

For signal evaluators, the broader lesson is even more important.

A good signal evaluation follows the evidence as it develops.

It asks:

That is the difference between reporting an adverse event and performing a pharmacovigilance signal evaluation.

References

  1. European Medicines Agency. SGLT2 inhibitors — referral. Article 20 procedure. Reference EMEA/H/A-20/1419.

  2. European Medicines Agency. SGLT2 inhibitors: PRAC makes recommendations to minimise risk of diabetic ketoacidosis. EMA/100751/2016. 12 February 2016.

  3. European Medicines Agency. EMA confirms recommendations to minimise ketoacidosis risk with SGLT2 inhibitors for diabetes. EMA/142655/2016. 26 February 2016.

  4. European Medicines Agency. SGLT2 inhibitors Article-20 procedure — Assessment report. EMA/PRAC/50218/2016. 10 March 2016.

  5. Monami M, Nreu B, Zannoni S, Lualdi C, Mannucci E. Effects of SGLT-2 inhibitors on diabetic ketoacidosis: A meta-analysis of randomised controlled trials. Diabetes Research and Clinical Practice. 2017;130:53–60. PMID: 28570924.

  6. Liu J, Li L, Li S, et al. Sodium-glucose co-transporter-2 inhibitors and the risk of ketoacidosis in patients with type 2 diabetes mellitus: A systematic review and meta-analysis of randomized controlled trials. Diabetes, Obesity and Metabolism. 2020;22(9):1619–1627. PMID: 32364674.

  7. Kim YG, Jeon JY, Han SJ, et al. Sodium-glucose co-transporter-2 inhibitors and the risk of ketoacidosis in patients with type 2 diabetes mellitus: A nationwide population-based cohort study. Diabetes, Obesity and Metabolism. 2018. PMID: 29569427.

  8. Fralick M, Schneeweiss S, Patorno E. Risk of diabetic ketoacidosis after initiation of an SGLT2 inhibitor. New England Journal of Medicine. 2017;376:2300–2302.

  9. Taylor SI, Blau JE, Rother KI. SGLT2 inhibitors may predispose to ketoacidosis. Journal of Clinical Endocrinology & Metabolism. 2015;100(8):2849–2852.

  10. Ogawa W, Sakaguchi K. Euglycemic diabetic ketoacidosis induced by SGLT2 inhibitors: Possible mechanism and contributing factors. Journal of Diabetes Investigation. 2016;7(2):135–138.

  11. Somagutta MR, Agadi K, Hange N, et al. Euglycemic Diabetic Ketoacidosis and Sodium-Glucose Cotransporter-2 Inhibitors: A Focused Review of Pathophysiology, Risk Factors, and Triggers. Cureus. 2021;13(3):e13665. PMID: 33824816.

  12. Goldenberg RM, Berard LD, Cheng AYY, et al. SGLT2 inhibitor-associated diabetic ketoacidosis: Clinical review and recommendations for prevention and diagnosis. Clinical Therapeutics. 2016;38(12):2654–2664.e1.

  13. Burke KR, Schumacher CA, Harpe SE. SGLT2 inhibitors: A systematic review of diabetic ketoacidosis and related risk factors in the primary literature. Pharmacotherapy. 2017;37(2):187–194.

  14. Dutta S, Kumar T, Singh S, et al. Euglycemic diabetic ketoacidosis associated with SGLT2 inhibitors: A systematic review and meta-analysis. Journal of Clinical Endocrinology and Metabolism. 2022.

  15. Risk of diabetic ketoacidosis associated with sodium glucose cotransporter-2 inhibitors: A network meta-analysis and meta-regression. Journal of Clinical Medicine. 2024;13:1748. PMID: 38541972.

  16. Assessing the risk of ketoacidosis due to sodium-glucose cotransporter-2 inhibitors in patients with type 1 diabetes: A meta-analysis and meta-regression. Diabetes Care. 2021. PMID: 33373368.

  17. Sodium-glucose co-transporter-2 inhibitors and the risk of ketoacidosis in patients with type 2 diabetes: systematic review and meta-analysis of randomised controlled trials. Diabetes, Obesity and Metabolism. 2020. PMID: 32364674.

  18. European Medicines Agency. Guideline on good pharmacovigilance practices (GVP), Module IX — Signal Management.

  19. European Medicines Agency. Questions and answers on signal management.

  20. European Medicines Agency. PRAC recommendations on safety signals and associated product-information recommendations.

Revision History

Related Articles

Historical Signal Evaluation: Fluoroquinolones and Aortic Aneurysm or Dissection

A detailed evaluation of the fluoroquinolone–aortic aneurysm or dissection signal, examining epidemiological evidence, biological plausibility, confounding, susceptible populations, regulatory assessment and changes to product information.

Historical Signal Evaluation: Metamizole and Agranulocytosis

A reconstruction of the 2024 EU metamizole–agranulocytosis review, from the regulatory trigger and available evidence through PRAC assessment, risk-minimisation conclusions and changes to product information.

Historical Signal Evaluation: Pioglitazone and Bladder Cancer

How the pioglitazone–bladder cancer signal evolved, why epidemiological studies produced conflicting results, how exposure duration and confounding affected interpretation, and how regulators translated uncertainty into risk-minimisation measures.