Historical Signal Evaluation: SGLT2 Inhibitors and Diabetic Ketoacidosis

The SGLT2 inhibitor–diabetic ketoacidosis signal is important for more than the existence of a rare adverse reaction. It shows how a medicine can alter the presentation of an event in a way that makes recognition more difficult. In some reported cases, ketoacidosis occurred without the degree of hyperglycaemia conventionally expected in diabetic ketoacidosis (DKA). A clinician who used high blood glucose as the principal diagnostic trigger could therefore be falsely reassured.

The signal also demonstrates why causality in pharmacovigilance often involves an interaction rather than a single isolated cause. Relative insulin deficiency, acute illness, fasting, dehydration, surgery, alcohol exposure or other catabolic stress may combine with SGLT2 inhibition to increase susceptibility. The evaluator must therefore distinguish a medicine's contribution to risk from the claim that the medicine independently explains every episode.

Clinical and Pharmacological Context

Sodium-glucose cotransporter-2 (SGLT2) is located in the proximal renal tubule and is responsible for much of the filtered glucose that is normally reabsorbed. SGLT2 inhibitors reduce renal glucose reabsorption, causing glycosuria and lowering plasma glucose. Their established pharmacological effects also include osmotic diuresis and changes in sodium handling.

DKA is a metabolic emergency characterised by ketone production and metabolic acidosis arising when effective insulin action is insufficient for metabolic demand. The conventional phenotype often includes marked hyperglycaemia, dehydration, gastrointestinal symptoms and compensatory rapid breathing. SGLT2 inhibition creates a distinctive diagnostic problem because urinary glucose loss may blunt the rise in plasma glucose while ketogenesis and acidosis continue.

“Euglycaemic DKA” is commonly used for this phenotype, but the term should not be interpreted as requiring literally normal glucose. The practical issue is that glucose may be normal or only moderately elevated relative to what would ordinarily be expected in DKA. The pharmacovigilance signal was therefore two-dimensional:

1. risk: SGLT2 inhibitors may increase susceptibility to DKA;
2. recognition: SGLT2 inhibitors may make DKA less obvious if clinicians rely heavily on hyperglycaemia.

That second dimension explains why risk communication was clinically important even before every estimate of relative risk had converged.

Mechanistic Plausibility and Susceptibility

SGLT2 inhibition lowers plasma glucose independently of insulin. Lower glucose availability may be accompanied by reduced endogenous insulin secretion, while changes in glucagon signalling and substrate utilisation can favour lipolysis and ketone production. Osmotic diuresis and volume depletion can add a counter-regulatory stress response. These mechanisms provide a coherent physiological context for ketoacidosis.

The complete causal pathway is nevertheless not reducible to one mechanism. Experimental work has proposed direct effects on glucagon secretion and altered renal ketone handling, but the contribution of each pathway to an individual clinical event is not identical or completely resolved. Mechanistic plausibility therefore strengthens the signal; it does not replace clinical evidence.

Precipitating factors are part of the causal model

A striking feature of reported SGLT2-associated DKA has been the frequent presence of additional metabolic stressors. Regulatory and clinical literature has repeatedly identified circumstances such as:

• reduced food intake or prolonged fasting;
• severe dehydration;
• acute illness or infection;
• major surgery;
• sudden or excessive reduction in insulin;
• increased insulin requirement during physiological stress;
• excessive alcohol use;
• low endogenous insulin reserve.

These factors are not merely “confounders” to be removed from the story. They may be effect modifiers or components of the causal pathway. A useful conceptual model is:

SGLT2 inhibition + insufficient insulin action + catabolic stress → increased ketogenesis → possible ketoacidosis.

The model does not mean that every component is required in every case. It explains why the same exposure can be tolerated in one clinical state and become hazardous in another.

How the Signal Became Visible

Post-marketing reports were central to recognition. Their value was not that they established incidence—they could not—but that they described an unusual clinical phenotype. Serious DKA in people treated with a glucose-lowering medicine was already noteworthy; cases with unexpectedly modest glucose concentrations were more informative because they suggested that the treatment might be altering the clinical pattern itself.

For individual case assessment, the most useful information includes:

• diabetes type and evidence of insulin reserve;
• SGLT2 inhibitor, dose and exposure timing;
• insulin or other glucose-lowering therapy and recent changes;
• glucose, pH or bicarbonate, anion gap and ketone measurements;
• fasting, vomiting, dehydration, infection, surgery or alcohol exposure;
• treatment and time to recovery;
• whether another precipitating cause was identified.

A case should not be dismissed merely because a recognised precipitant is present. The signal question is whether SGLT2 inhibition contributed to susceptibility or changed the phenotype. Equally, the presence of the drug does not justify ignoring a major alternative cause such as severe infection or profound insulin omission.

The EU Article 20 Review

The European Commission initiated an EU review in June 2015 under Article 20 of Regulation (EC) No 726/2004. PRAC assessed the safety concern and made recommendations in February 2016. The Committee for Medicinal Products for Human Use (CHMP) adopted EMA's final opinion, after which the European Commission issued legally binding decisions in April 2016 for the products concerned.

EMA concluded that rare cases of DKA, including life-threatening cases, had occurred with SGLT2 inhibitors and that some were atypical because blood glucose was not as high as expected. Product information was updated to communicate DKA as a rare adverse reaction for the products reviewed and to describe recognition and risk-reduction measures. EMA also stated that the benefits of the medicines continued to outweigh their risks in the authorised type 2 diabetes population.

This sequence should be described precisely. PRAC's recommendation is a pharmacovigilance committee recommendation; the CHMP opinion is EMA's scientific opinion for the centrally authorised medicines concerned; the European Commission decision provides the legally binding EU outcome of the Article 20 procedure. Treating all three stages simply as “PRAC made a law” would erase an important regulatory distinction.

Translating the Signal Into Risk Minimisation

The regulatory response focused on recognition, susceptibility and interruption of exposure rather than withdrawal of the class. EMA advised healthcare professionals to consider DKA when symptoms were compatible even if glucose was not markedly elevated. If DKA was suspected or confirmed, treatment was to be stopped; re-starting was not recommended unless another cause had been identified and resolved. The recommendations also highlighted recognised risk factors and temporary interruption during major surgery or serious acute illness.

This is targeted risk minimisation. The intervention was designed around the mechanism by which harm could occur and the reason diagnosis could be delayed. It did not imply that every exposed person required the same intensity of intervention or that every DKA episode was attributable solely to the medicine.

The history also shows why regulatory language must be read in its time and product context. In the 2016 referral outcome, EMA described DKA as rare—up to 1 in 1,000—for the reviewed SGLT2 inhibitor products. That regulatory frequency statement should not be repurposed as a universal contemporary incidence estimate across all indications, populations and members of the expanded class. Later use in heart failure, chronic kidney disease and different diabetes populations changes the denominator and clinical context.

Why Early Trial Evidence Could Appear Reassuring

Rare adverse events expose a fundamental limitation of clinical trials: randomisation controls confounding, but it does not guarantee adequate statistical power. If an event occurs only a few times across a development programme, a trial or meta-analysis may produce a wide confidence interval compatible with both no increase and a clinically meaningful increase.

This helps explain why early pooled randomised evidence did not always show a statistically significant DKA signal. Trial populations may also differ from routine care: insulin adjustments are more controlled, acute illness may be handled under protocol, high-risk patients can be excluded, and adverse events are observed within a defined follow-up period.

The correct conclusion from sparse negative evidence is therefore not “the event cannot be drug-related.” It is “the available trial data have limited ability to exclude a rare effect of this magnitude.” That conclusion leaves room for spontaneous reports and mechanistic evidence to remain important while better comparative data accumulate.

Later randomised evidence

As exposure and trial experience expanded, meta-analyses had more DKA events to analyse. A 2025 updated meta-analysis of 22 randomised trials involving 80,235 participants with type 2 diabetes reported a pooled relative risk of 2.32 (95% CI 1.64–3.27) for DKA with SGLT2 inhibitors compared with controls. Subgroup estimates varied by the type of trial and baseline characteristics.

The value of such evidence is not that it retroactively proves every spontaneous case. It shows how a rare-event signal can become more measurable once enough randomised exposure accumulates. The magnitude should still be interpreted with absolute risk, population and indication in mind. A relative increase in a rare event can remain small in absolute terms, while still being clinically important because ketoacidosis can be life-threatening and its atypical presentation can delay treatment.

Reconciling Different Study Designs

Signal evaluation often encounters studies that appear to disagree. Some observational analyses have reported little or no increase compared with particular active comparators; other randomised or pooled analyses have found an increased relative risk. These findings should not be resolved by counting how many studies are “positive.”

Instead, the evaluator should ask what each design estimates:

A negative active-comparator cohort may indicate that confounding, comparator choice or the treated population matters. A positive trial meta-analysis may indicate a genuine relative increase while still being unable to identify which clinical circumstances produce most events. Both can improve the model.

Exposure Context: Type 1 Diabetes and Other Indications

The 2016 European review concerned products authorised at that time for type 2 diabetes and specifically noted that SGLT2 inhibitors were not authorised for type 1 diabetes in that setting, although off-label cases had occurred. This historical statement should not be converted into a timeless claim about every SGLT inhibitor or every later jurisdictional authorisation.

The class has subsequently acquired major roles outside glucose lowering, particularly in heart failure and chronic kidney disease. That expansion changes the exposed population. A contemporary safety system therefore cannot simply inherit a 2016 diabetes-only risk model without asking whether susceptibility, concomitant therapy and diagnostic awareness differ in newer indications.

The enduring pharmacovigilance concept is broader: the risk model must follow the population in which the medicine is actually used. When indication, background disease or concomitant therapy changes, the safety question should be re-examined even if the labelled adverse reaction is already known.

Case-to-Signal Reasoning in Practice

Consider an illustrative case: a person with type 2 diabetes taking an SGLT2 inhibitor undergoes surgery, has reduced oral intake, continues the medicine, and develops nausea, tachypnoea, high-anion-gap metabolic acidosis and ketonaemia with only modest hyperglycaemia.

The signal evaluator should not ask only “did the drug cause DKA?” A better analysis separates the causal components:

• Was there objective ketoacidosis?
• Was SGLT2 inhibition pharmacologically active at the relevant time?
• Did surgery and fasting create a catabolic stress state?
• Was insulin reduced or physiologically insufficient?
• Did the modest glucose concentration plausibly delay recognition?
• Was there another dominant cause?
• Does the case add new information about timing, susceptibility, outcome or preventability beyond the already characterised risk?

For an established risk, the purpose of case review shifts. The organisation is no longer merely deciding whether the association exists. It is looking for changes in frequency, severity, risk factors, indication, preventability or the effectiveness of existing risk minimisation.

Failure Modes and Governance Lessons

Several errors recur when this signal is reviewed.

The first is glucose anchoring: excluding DKA because glucose is not markedly elevated. The signal itself demonstrated why that rule can fail in the presence of SGLT2 inhibition. The second is single-cause reasoning: treating surgery, fasting, infection or insulin reduction as proof that the medicine did not contribute. Those factors may interact with exposure rather than compete with it as mutually exclusive explanations.

The third is relative-risk-only communication. A two-fold relative increase can sound very large without the baseline frequency; conversely, describing DKA as rare can obscure the seriousness of delayed recognition. Good benefit-risk communication therefore preserves both dimensions: absolute frequency and clinical consequence.

A fourth failure is historical label drift. Risk statements from the 2016 type 2 diabetes review should not be copied into new populations as though indication, background disease and current authorised product information had not changed. Pharmacovigilance governance should verify the current label and the current exposed population.

Illustrative inspection questions

An inspector examining management of this type of established serious risk could reasonably ask:

• Can the organisation show how atypical DKA cases are medically identified and coded?
• Are cases reviewed for insulin status, fasting, dehydration, surgery, acute illness and other relevant circumstances?
• Can the company distinguish the original signal evaluation from later monitoring of an established risk?
• Were new indications and populations considered when the medicine's use expanded?
• Are product-information and risk-minimisation commitments traceable to the applicable regulatory procedure?
• If new evidence changes the estimated magnitude or susceptible population, is there a documented route for signal re-evaluation and escalation?

These are illustrative questions, not invented inspection findings or additional legal requirements. They test whether the pharmacovigilance system can demonstrate that known risk is actively understood rather than merely copied from an old assessment.

Key Takeaways

The SGLT2 inhibitor–DKA signal teaches five durable lessons.

First, a medicine can alter the phenotype of an adverse event as well as its probability. The possibility of ketoacidosis despite only modest glucose elevation was central to the regulatory response.

Second, precipitating factors do not necessarily disprove a drug contribution. For multifactorial metabolic events, susceptibility and exposure can interact.

Third, rare-event safety evidence evolves. Early trials may be underpowered; later pooled randomised data can materially change precision.

Fourth, risk minimisation should address the mechanism of preventable harm. In this case, recognition of atypical DKA, identification of risk factors and interruption of exposure in relevant acute circumstances were more useful than a generic warning detached from clinical context.

Finally, once a signal becomes an established risk, surveillance does not end. The question changes from “does this association exist?” to “has its frequency, severity, susceptible population, indication context or preventability changed?”

References

1. European Medicines Agency. SGLT2 inhibitors — Article 20 referral. Review initiated June 2015; PRAC recommendation 11 February 2016; CHMP opinion 25 February 2016; European Commission final decisions April 2016.
2. European Medicines Agency. EMA confirms recommendations to minimise ketoacidosis risk with SGLT2 inhibitors for diabetes. EMA/265224/2016. 28 April 2016.
3. European Medicines Agency. SGLT2 inhibitors: PRAC makes recommendations to minimise risk of diabetic ketoacidosis. EMA/100751/2016. 12 February 2016.
4. European Medicines Agency. SGLT2 inhibitors Article-20 procedure — Assessment report. EMA/PRAC/50218/2016. 2016.
5. Wang Y, et al. Sodium-Glucose Cotransporter-2 Inhibitors and Diabetic-Ketoacidosis in T2DM Patients: An Updated Meta-Analysis and a Mendelian Randomization Analysis. Clinical Pharmacology & Therapeutics. 2025. doi:10.1002/cpt.3615. PMID: 40070044.

Regulatory Note

The 2015–2016 EU review was an Article 20 procedure under Regulation (EC) No 726/2004. PRAC performed the pharmacovigilance assessment and made recommendations; CHMP adopted EMA's opinion; the European Commission issued the legally binding final decisions for the centrally authorised products concerned.

The clinical measures described in the historical referral outcome should be understood in the context of the authorised product information and indication at that time. Current clinical decisions must follow the current product information and applicable guidance for the specific medicine and jurisdiction. The analytical workflow and illustrative inspection questions in this article are recommended explanatory practice, not additional legal requirements.