Effects of dapagliflozin on urinary output, fluid balance, and biochemistry in critically ill patients: a post-hoc secondary analysis of the DEFENDER trial

Abstract

Background

See More

Sodium-glucose cotransport-2 inhibitors (SGLT2i) have established benefits in diabetes mellitus, heart failure, and chronic kidney disease, but their physiological effects during critical illness remain unclear. We explored whether dapagliflozin affected urinary output, fluid balance, and other physiological parameters in critically ill patients with acute organ dysfunction.

Methods

This secondary analysis of the DEFENDER trial included 401 critically ill patients with acute organ dysfunction randomized to receive dapagliflozin 10 mg daily or standard care. We analyzed urinary output, fluid balance, electrolytes, acid–base status, glycemia, and vasopressor requirements over the first five days using Bayesian models.

Results

Dapagliflozin progressively increased urinary output (day 5: + 157 mL/day, 95% CrI -90 to 386, probability 90%) and decreased fluid balance (day 5: -290 mL/day, 95% CrI -564 to -27, probability 98%). Furosemide use was lower in the dapagliflozin group (overall -3%, 95% CrI -7% to 1%, probability 90%). Dapagliflozin had minimal effects on creatinine and electrolytes but was associated with progressive small decreases in pH (day 5: -0.02, probability 96%). Maximum glucose levels were consistently lower with dapagliflozin (-9 mg/dL overall, probability 83%). Norepinephrine requirements showed a time-dependent increase in the dapagliflozin group, with the expected dose difference reaching 0.034 mcg/kg/min by day 5 (probability 94%), and heterogeneity analysis revealed larger effects in patients with sepsis or on mechanical ventilation.

Conclusion

This exploratory analysis suggests dapagliflozin may enhance diuresis and reduce loop diuretic requirements in critically ill patients, potentially at the cost of increased vasopressor needs. Glucose levels were likely slightly lower with dapagliflozin. Given the study's limitations and heterogeneous treatment effects, these findings should be considered hypothesis-generating pending confirmation in prospective trials.

Introduction

Sodium-glucose cotransport-2 inhibitors (SGLT2i) are increasingly used in medical patients with several clinical conditions [1,2,3,4,5]. SGLT2i have been shown to improve relevant patient-centered outcomes in patients with type 2 diabetes mellitus, heart failure, and chronic kidney disease (CKD). SGLT2i may also have acute effects such as reducing occurrence of acute kidney injury (AKI) in some specific populations [4]. Trials of SGLT2i in acutely ill patients, however, have been limited in size and have yielded inconclusive results so far [5, 6]. On the other hand, the recent DEFENDER trial found a 90% probability that dapagliflozin reduced the need for renal replacement therapy [6]. Such observations suggest the need to better understand how dapagliflozin might have affected kidney relevant physiological variables.

SGLT2i have important physiologic effects relevant to critical illness. Specifically, besides the expected effects on glycemic control due glycosuria, SGLT2i may increase urinary output and, due to free water losses, also increase sodium levels [7,8,9]. Such diuresis and hypernatremia may affect fluid balance, pH and vasopressor requirements. Knowledge of such effects may help clinicians interpret clinical and laboratory findings and respond as clinically indicated.

In this secondary analysis of the Dapagliflozin for Critically Ill Patients With Acute Organ Dysfunction (DEFENDER) trial [6], we explored whether dapagliflozin, a SGLT2i, would affect urinary output, fluid balance, furosemide use, norepinephrine requirements, sodium and potassium levels, and glucose levels in the first five days after the initiation of therapy. We aimed to test the primary hypothesis that, compared with controls, dapagliflozin would increase urinary output and, thereby, lead to a more negative fluid balance.

Methods

Design and population

Secondary post-hoc analysis of a randomized clinical trial. Patients included in the DEFENDER trial from sites that had agreed to collect additional data for the first five days after enrollment. We included patients from participating sites who had any physiological data collected during the first five days, regardless of data completeness.

Overview of the DEFENDER trial

DEFENDER was a multicenter, open-label, randomized clinical trial that included 507 participants with at least 1 acute organ dysfunction (hypotension, acute kidney injury, or respiratory failure). Patients were randomized 1:1; for 10 mg of dapagliflozin up to 14 days or standard of care. The primary endpoint was a combined hierarchical endpoint of hospital mortality, initiation of renal replacement therapy (RRT), and intensive care unit– ICU—length of stay. The trial had neutral results for the primary endpoint (win ratio of 1.01; 95% confidence intervals from 0.90; 1.13; P = 0.89). The trial, however, also found a 90% probability that dapagliflozin reduced the use of renal replacement therapy (10.9% versus 15.1%).

Endpoints

The following daily parameters from day one up to day five were analysed: daily urinary output (not cumulative), daily fluid balance (not cumulative), furosemide use (daily, yes or no), norepinephrine dose (defined as the maximum daily dose of norepinephrine used by patients in mcg/kg/min), sodium and potassium (both defined as the maximum daily values), maximum and minimal glucose levels (in mg/dL), insulin use (daily, yes or no), and insulin dose (insulin units during a given day). Daily values of creatinine, in mg/dL, and pH and base excess (BE) are also reported. 28-day mortality, use of RRT, and days alive and free of the ICU at 28 days are also reported since the sample used is not the whole sample reported in the original trial.

Statistical analysis

Continuous outcomes were assessed through Bayesian linear mixed models with study day, study group, and their interaction as fixed effects, and participants as random intercepts. Urinary output and fluid balance models were adjusted for daily creatinine, daily furosemide dose, baseline admission type (sepsis, cardiovascular, other), mechanical ventilation, and vasopressor use. Furosemide use (binary) was analyzed using Bayesian logistic mixed models with similar adjustments. Creatinine was adjusted for daily furosemide dose, daily fluid balance, baseline admission type, mechanical ventilation, and vasopressor use. To assess the net clinical effect on fluid status (as opposed to the mechanistic diuretic effect), we also performed an alternative analysis of urinary output and fluid balance adjusting only for baseline characteristics (admission type, mechanical ventilation, and vasopressor use), without adjustment for time-varying furosemide dose or creatinine. This approach captures the overall impact of dapagliflozin on fluid balance, incorporating both its direct diuretic effects and indirect effects through reduced furosemide requirements. Creatinine was adjusted for fluid balance to distinguish true renal effects from hemoconcentration. For these four outcomes, heterogeneity of treatment effects was assessed using models with three-way interactions (study day × intervention × subgroup characteristic). Heterogeneity was considered meaningful if the probability of subgroup differences exceeding 50 mL (for fluid outcomes), 0.3 mg/dL (for creatinine), or 5% (for furosemide use) was > 80%. Other biochemical parameters (sodium, potassium, pH, base excess) were analyzed with a simpler adjustment considering baseline features (admission type, vasopressors, mechanical ventilation) besides study day, intervention, and their interaction. This is because we are more interested in safety here and not in the effects of, say, dapagliflozin on sodium that are independent of its effect on creatinine and fluid balance. Similarly, glucose models were not adjusted for covariates to assess for overall safety effects. For maximum glucose levels, an additional model was constructed adjusting for insulin dose, transformed using the inverse hyperbolic sine function—a log-like transformation that accommodates zero values. Insulin dose was analysed with logistic regression Bayesian model adjusted for day, intervention, and their interaction.

For norepinephrine, dose was set at zero for patients discharged alive to the ward (since norepinephrine use is not permitted outside the ICU); patients that died before day 5 were annotated and a two-step approach was used. First, we assessed whether mortality was different between arms using a mixed Bayesian logistic regression adjusted for day, intervention, and random intercept for the patient using a regularizing prior (log odds ratio– log(OR)– for the intervention being normally distributed (mean zero and standard deviation of 0.355) [10]; if the probability of a mortality increase of at least 1% was above 0.90 at any given day we would assume that an intervention effect on early mortality could not be ruled out and would attempt modelling norepinephrine use using a multistate model. If probability of a 1% increase in mortality with the intervention was low, we would assume no clear effect of the intervention on mortality from days 1 to 5 and would model norepinephrine dose daily using a hurdle model. We found that probability of increase in 1% mortality with intervention was always below 0.30 (highest value was 0.26 for day 5), and hence a hurdle gamma model was used. The hurdle gamma model was adjusted for baseline vasopressor use, admission type, mechanical ventilation status, and daily fluid balance to isolate the direct treatment effect and maintain consistency with other analyses. This model handles the zero-inflated nature of norepinephrine administration by separately modeling: 1) whether a patient received any norepinephrine (the hurdle component), and 2) the dose received when norepinephrine was administered (the gamma component). We assessed heterogeneity of treatment effects (HTE) by fitting hurdle gamma models with three-way interactions (treatment × time × subgroup) for key baseline characteristics: admission type (sepsis/infection, cardiovascular, other), mechanical ventilation status, and baseline vasopressor use. Heterogeneity was evaluated through examination of three-way interaction coefficients, with meaningful heterogeneity defined as > 0.01 mcg/kg/min for the gamma component (dose effect) and > 0.05 for the hurdle component (probability of use).

Overall mortality and RRT use were analysed using standard Bayesian logistic regression assuming a neutral prior for the intervention odds ratio (mean 0, standard deviation of 0.355). Days alive and free of the ICU was modeled as a Bayesian liner regression with a normal prior for the estimate with mean zero and standard deviation of 5.

Regularizing priors for the intervention were used to produce more conservative results; they were all centered as absence of difference. Missing data were handled through the likelihood-based approach of Bayesian mixed models, which uses all available data under the missing at random assumption. Results are presented as the model-based marginal probabilities that the interaction between day and intervention was present (probability of direction), the model-based marginal expected differences between intervention and control groups for all five-day period (with mean and 95% credible intervals– CrI) and the model-based marginal expected differences for each day. For all physiological outcomes, we report model-based absolute differences between groups: mean differences with 95% credible intervals (CrI) for continuous outcomes and probability differences with 95% CrI for binary outcomes. This approach provides clinically interpretable effect estimates throughout. Probability of direction for increase or decrease of effect was also reported. The reported posterior probabilities reflect the strength of evidence for the direction of effect rather than traditional statistical significance. A probability of 95% indicates that, given the data and model, there is a 95% probability the effect is in the stated direction. These probabilities directly quantify uncertainty and should be interpreted as continuous measures of evidence strength rather than dichotomous significance tests. We intentionally avoid rigid probability thresholds, allowing readers to judge the strength of evidence based on the full posterior distributions.

Results

Overall

A total of 401 patients (206 control, 195 dapagliflozin; 79% of the DEFENDER trial population with data from 10 sites) had data available for analysis. The mean age was 64.1(SD, 15.1) years, and 182 participants (45.4%) were female. The primary reasons for ICU admission were infection (n = 151; 37.7%) and cardiovascular conditions (n = 134; 33.4%). Compared to those without data available, participants included in this analysis were more likely to have severe presentations of critical illness, with higher rates of mechanical ventilation (51.9% vs. 25.5%), norepinephrine use (54.6% vs. 32.1%), and inotropic support (16.5% vs. 10.4%). Baseline characteristics comparing participants with and without available data are shown in eTable 1. Comparison of patients according to enrollment group in this analysis set is shown on Table 1.

Table 1 Comparison between patients randomized to dapagliflozin or control group in the analysis set

Clinical outcomes

87 patients died in hospital in each group (42% and 45% for control and dapagliflozin arm, respectively), resulting in an odds ratio for mortality of 1.07 (95% CrI 0.76 to 1.52; absolute difference of −0.06 with 95% CrI −0.06 to 0.10; 0.35 probability of benefit). 30 patients in control group (15%) and 22 patients in dapagliflozin group (11%) required RRT, with a resulting odds ratio of 0.84 (95% CrI 0.54 to 1.29; absolute difference of −0.01 with 95% CrI −0.07 to 0.03; 0.78 probability of benefit). The median number of days alive and outside the ICU was 10 (interquartile range 0 to 23) in the control group versus 16 (interquartile range between 0 and 23.5) in the dapagliflozin group resulting in a mean difference of 0.4 days (95% CrI −1.8 to 2.5 days-free; 0.63 probability of benefit).

Urinary output

The probability of interaction between dapagliflozin and urinary output over time was 0.86. Overall, dapagliflozin increased urinary output by 94 mL/day (95% CrI −52 to 229) with 91% probability. This effect remained consistent, with probabilities around 0.90 from day 2 to day 5 (Table 2, Fig. 1). By day 5, urinary output was 157 mL/day higher in the dapagliflozin group (95% CrI −90 to 386). Raw values are shown in eFigure 1 and subgroup analysis is shown in eFigure 2, with numerical values in eTable 2 and 3. Heterogeneity analyses revealed no major signs of heterogeneity for treatment effects on urinary output (heterogeneity probability 16–34%). In analyses adjusting only for baseline characteristics (without furosemide or creatinine adjustment), the net effect was 82 mL/day overall (95% CrI −52 to 223, probability 88%), demonstrating that dapagliflozin improves urinary output even when accounting for reduced diuretic use.

Table 2 Main results

Fig. 1

Differences in urinary output,nfluid balance, and furosemide use Dapagliflozin − Control Posterior distributions of treatment effects (dapagliflozin minus control) from Bayesian mixed models adjusted for baseline characteristics. (A) Urinary output (mL/day): Progressive increase in effect from day 1 to day 5, with posterior distributions shifting rightward over time. (B) Fluid balance (mL/day): Increasingly negative fluid balance (favorable effect) with dapagliflozin, shown by leftward shift of distributions. (C) Furosemide use (probability difference): Consistent reduction in furosemide requirements with dapagliflozin across all days. Shaded areas represent posterior density distributions for each day, with darker shades indicating earlier days. Vertical dashed lines indicate no effect. All models adjusted for baseline vasopressor use, admission type, mechanical ventilation, and creatinine levels

Fluid balance

The probability of interaction was 0.98, indicating a progressive effect. Overall, dapagliflozin reduced fluid balance by 163 mL/day (95% CrI −317 to −13) with 98% probability. The probability that dapagliflozin reduced fluid accumulation exceeded 0.95 for days 2–5 (Table 2, Fig. 1). By day 5, fluid balance was 290 mL/day more negative in the dapagliflozin group (95% CrI −564 to −27). Raw values are shown in eFigure 3 and subgroup analysis is shown in eFigure 4, with numerical values in eTable 2 and 3. Heterogeneity analyses revealed no major signs of heterogeneity (heterogeneity probability 17–33%). In analyses adjusting only for baseline characteristics, the net effect was −143 mL/day overall (95% CrI −295 to 8, probability 97%), confirming that dapagliflozin achieves meaningful reductions in fluid accumulation despite reduced furosemide use.

Furosemide use

The probability of interaction was 0.72. Overall, dapagliflozin reduced furosemide use by 3% (95% CrI −7% to 1%) with 90% probability. This effect was consistent across days, with 5–9% fewer patients requiring furosemide from days 1–5 (Table 2, Fig. 1). Raw furosemide use according to day and allocation arm is shown in eFigure 5, with subgroup analysis shown in eFigure 6 and numerical values on eTable 2 and eTable 3. Heterogeneity was found for furosemide use across all subgroup comparisons (heterogeneity probability > 80% for all comparisons), but the magnitude of these differences was small (up to 4.4% for a).

Creatinine

There was a probability of interaction of 0.89 for creatinine, but with minimal values (Table 2, Fig. 2). The effect showed a slight trend from higher values in the dapagliflozin group on day 1 (0.09 mg/dL) to lower values by day 5 (−0.05), but with probabilities of direction that did not strongly support either direction (ranging from 0.39 to 0.73). Raw values are shown in eFigure 7. No signs of heterogeneity were seen on subgroup analysis (eFigure 8).

Fig. 2

Posterior Distributions of Treatment Effects on Laboratoy Parameters by Day. Baseline−adjusted analyses; vertical dashed line indicates no effect.Posterior distributions of baseline-adjusted treatment effects by day from Bayesian mixed models. (A) Creatinine: Minimal effects across all days with distributions centered near zero. (B) Sodium: Slight trend toward higher values with dapagliflozin, most pronounced by day 5. (C) Potassium: No clinically meaningful differences between groups. (D) pH: Progressive decrease with dapagliflozin treatment, with distributions shifting leftward from day 3 onward. (E) Base excess: Time-dependent pattern shifting from higher values early to lower values by day 5, consistent with mild metabolic acidosis. All models adjusted for baseline admission type, mechanical ventilation, and vasopressor use. Vertical dashed lines indicate no effect

Sodium and potassium

The probability of interaction for dapagliflozin use and differences in daily sodium and potassium levels were 0.82 and 0.22, respectively. For sodium, changes minimally increased over the 5-day period with a maximum probability of difference at day 5 (0.86) with an effect size of 0.78 mEq/L (95% CrI −0.63 to 2.19). Probabilities of direction for daily potassium effect ranged from 0.38 to 0.73 across the five days and minimal effect sizes (Table 2, Fig. 2). Raw values for sodium and potassium are shown in eFigure 9.

pH and base excess

A strong probability of interaction (0.97) was found for pH over time, with increasing probability of dapagliflozin decreasing pH from day 3 onward (Table 2). However, the magnitude of this effect was minimal, with day 5 showing a difference of only −0.02 (95% CrI −0.04 to 0.00). Base excess showed a time-dependent pattern with dapagliflozin treatment (probability of interaction 0.94), shifting from a trend toward higher values on day 1 (+ 0.42 mEq/L, probability 0.74) to progressively lower values by day 5 (−0.52 mEq/L, probability of reduction 0.77). Results are shown in Table 2 and Fig. 2. Violin plot for raw values in shown in eFigure 10.

Glucose levels

Maximum glucose levels showed a consistent trend toward lower values in the dapagliflozin group across all days, but differences were small, with an overall effect of −9 mg/dL (95% CrI −24 to 6) and probability of 0.83 for this direction. The effect was most pronounced on day 1 (−10 mg/dL, 95% CrI −27 to 6, probability 0.89) and gradually decreased by day 5 (−7 mg/dL, 95% CrI −25 to 12, probability 0.76). Results remained consistent after adjustment for daily insulin dose (eTable 4). Minimum glucose levels showed a similar but weaker trend, with an overall effect of −2 mg/dL (95% CrI −9 to 5) and probability of 0.71 for lower values in the dapagliflozin group. The effect was minimal on day 1 (−1 mg/dL) and increased slightly by day 5 (−3 mg/dL). Results are shown in Fig. 3, and raw boxplot values shown in eFigure 11.

Fig. 3

Posterior Distributions of Difference in Glucose Values Vertical dashed line indicates no effect. Effects of dapagliflozin on glucose parameters. (A) Posterior distribution of differences in maximum glucose levels (mg/dL) between dapagliflozin and control groups by day. Distributions are consistently shifted to the left of zero, indicating lower maximum glucose values with dapagliflozin across all days. (B) Posterior distribution of differences in minimum glucose levels (mg/dL)

Insulin use and dose

Insulin use was lower in the intervention group, with an overall probability of reduced insulin requirement from days 2 to 5 of 0.77, and a probability of lower insulin dose in the dapagliflozin group of 0.85 among those requiring insulin use. No consistent daily trend in insulin use or dose was observed over the first five days. The mean between-group difference in insulin dose was small (–0.2 units; 95% CrI, –1.7 to 0.2). Results for insulin use and dose are shown in eTable 4 and eFigure 12.

Norepinephrine use and dose

After adjustment for baseline vasopressor use, admission type, mechanical ventilation, and fluid balance, dapagliflozin was associated with modestly higher norepinephrine requirements compared to control (overall difference: 0.018 mcg/kg/min, 95% CrI: −0.007 to 0.043, probability 91.8%). This effect demonstrated a clear time-dependent pattern, with minimal difference on day 1 (0.007 mcg/kg/min, probability 65.4%) increasing progressively to day 5 (0.034 mcg/kg/min, probability 94.4%). The hurdle gamma model revealed two distinct components: among patients receiving norepinephrine, those in the dapagliflozin group required progressively higher doses over time (interaction coefficient: 0.052, 95% CrI: −0.02 to 0.12, probability 91.6%), while the probability of requiring norepinephrine decreased more slowly compared to control (hurdle interaction: −0.12, 95% CrI: −0.32 to 0.09, probability 87.5%). Among patients not requiring vasopressors at baseline, the probability of initiating norepinephrine was progressively higher in the dapagliflozin group, reaching a 7.1% absolute difference by day 5 (95% CrI: −6.0% to 27.3%, probability 87%). Results are shown in Fig. 4 and Table 2. Use of norepinephrine according to groups (yes/no) at a given day is shown in eFigure 13. We found evidence of heterogeneity in treatment effects across subgroups (eTable 5, eTable 6, eFigure 14). At day 5, dapagliflozin was associated with increased norepinephrine requirements primarily in patients with sepsis/infection (0.054 mcg/kg/min, 95% CrI −0.002 to 0.244; probability of harm 97%), mechanically ventilated patients (0.047 mcg/kg/min, 95% CrI −0.001 to 0.221; probability of harm 96%), and those on baseline vasopressors (0.042 mcg/kg/min, 95% CrI −0.006 to 0.201; probability of harm 93%), while effects were more uncertain in other subgroups. These model-based estimates, which adjust for baseline vasopressor use, admission type, mechanical ventilation, and fluid balance, reveal the isolated treatment effect that may not be apparent in raw data comparisons due to baseline imbalances and confounding factors.

Fig. 4

Hurdle Model Analysis of Norepinephrine Requirements Dapagliflozin vs Control (Adjusted for baseline characteristics).Hurdle model analysis of norepinephrine requirements. Panel A shows the overall expected dose difference, combining both the probability of requiring norepinephrine and the dose when required. This effect increased progressively from day 1 (0.007 mcg/kg/min, 65% probability of increase) to day 5 (0.034 mcg/kg/min, 94% probability). The hurdle model decomposition reveals two contributing mechanisms: Panel B shows that dapagliflozin increased the probability of requiring norepinephrine (up to 7% absolute difference by day 5), while Panel C demonstrates that among patients receiving norepinephrine, those in the dapagliflozin group required progressively higher doses

Discussion

In this secondary analysis of the DEFENDER trial, we found that dapagliflozin produced several measurable physiological effects in critically ill patients with acute organ dysfunction. Dapagliflozin treatment was associated with progressively increased urinary output and more negative fluid balance over five days, while reducing furosemide requirements. Notably, these beneficial effects on fluid status persisted even in baseline-adjusted analyses that did not account for furosemide or creatinine changes, confirming a net clinical benefit beyond the mechanistic diuretic effect. These diuretic effects were accompanied by minimal changes in electrolytes and creatinine, modest reductions in glucose levels with insulin sparing, and progressive mild metabolic acidosis. Importantly, we observed time-dependent increases in norepinephrine requirements, with heterogeneity analyses revealing that patients with sepsis (97% probability of harm), mechanical ventilation (96%), or baseline vasopressor use (93%) were most susceptible to these hemodynamic effects.

The current findings align with and expand our understanding of SGLT2 inhibitor mechanisms in critical illness. The progressive increase in diuresis and negative fluid balance corresponds with the drug's known osmotic diuretic properties. In acute heart failure trials, SGLT2 inhibitors have shown modest increases in diuresis and natriuresis with reduced loop diuretic requirements [11,12,13]. Mechanistic studies suggest these effects are counterbalanced by vasopressin release and rapid tubular adaptation to increased distal sodium delivery [14,15,16]. Our findings extend this evidence to heterogeneous critically ill populations, demonstrating a detectable diuretic response even in this complex setting.

The enhanced diuresis and reduced loop diuretic requirements may help explain the 90% probability of reduced RRT observed in DEFENDER. By preventing fluid overload-related kidney injury and potentially influencing clinician perception of kidney recovery, these physiological effects could delay or avoid RRT initiation. The finding that RRT reduction occurred despite increased vasopressor requirements suggests the benefits of improved fluid management may outweigh hemodynamic risks in preventing severe kidney dysfunction, though this balance varies substantially across patient subgroups.

The metabolic effects merit particular attention. The mild pH decrease and shifting base excess patterns are compatible with subclinical ketogenesis [17] or increased chloride levels. While glucose reduction was modest, the insulin-sparing effect raises intriguing parallels with the TGC-Fast trial, where tight glucose control reduced kidney dysfunction [18]. However, unlike insulin-mediated glucose lowering, SGLT2 inhibition promotes metabolic flexibility and ketogenesis without hyperinsulinemia. These metabolic adaptations, rather than glucose reduction alone, may contribute to nephroprotection through improved cellular metabolism and reduced renal sodium retention.

Our heterogeneity findings have important implications for clinical practice. The observation that ICU-free days remained unchanged despite increased vasopressor use suggests a complex trade-off. Dapagliflozin may cause transient hemodynamic changes requiring vasopressor support while simultaneously facilitating recovery through improved fluid management, or different patient subgroups may experience divergent effects. Based on our findings, clinicians might consider dapagliflozin in hemodynamically stable patients with evidence of fluid overload, particularly those admitted with cardiovascular conditions who showed minimal vasopressor increases. Conversely, early initiation should be avoided in patients with septic shock or vasopressor requirements. Future trials should consider stratified randomization by baseline hemodynamic status or adaptive designs examining delayed initiation after stabilization.

This study has several strengths. It provides robust data on physiological effects of dapagliflozin in critically ill patients from a multicenter randomized trial, minimizing selection and ascertainment bias. Our Bayesian approach with regularizing priors provides conservative estimates while allowing probabilistic interpretation of effects. The high posterior probabilities observed (90–98%) despite skeptical priors strengthen confidence in our findings.

We acknowledge important limitations. As a post-hoc analysis of 79% of DEFENDER participants, these findings are hypothesis-generating. The included patients had higher illness severity, potentially amplifying both beneficial effects (enhanced diuresis) and adverse effects (vasopressor requirements). We analyzed only the first 5 days despite 14-day treatment duration, potentially missing later effects. Fluid balance calculations are inherently imprecise due to unmeasured insensible losses and incomplete intake recording. Missing data patterns were informative, as early death or discharge created incomplete follow-up. We lacked ketone measurements to clarify metabolic acidosis mechanisms and volume status markers to contextualize diuretic responses. Diuretics other than furosemide were not captured, nor were administration routes or institutional glucose protocols documented. The absence of standardized biochemical measurement sources introduces variability, particularly for pH and glucose values. Finally, the open-label design may have influenced clinical decisions, though the consistency of physiological effects suggests robust pharmacological mechanisms.

Conclusion

In critically ill patients, dapagliflozin therapy is likely to enhance diuresis, reduce fluid accumulation and furosemide requirements, and provide modest glucose lowering with reduced insulin needs. These benefits were accompanied by small decreases in pH and increased norepinephrine requirements, particularly in patients with sepsis or on mechanical ventilation. The clinical significance of these combined fluid and metabolic effects provides a potential mechanistic explanation for the reduced renal replacement therapy observed in DEFENDER and warrants further investigation.