Research in Practice · Systematic review and Bayesian meta-analysis · Published 2 November 2026
Intravenous sodium bicarbonate is familiar in the intensive care unit, but its place in metabolic acidosis has remained uncertain. It reliably changes acid–base measurements; whether that biochemical effect improves outcomes that matter to patients is a different question.
Kitisin and colleagues brought together the contemporary randomised evidence using a Bayesian meta-analysis. The central finding is clinically useful but narrower than a simple “bicarbonate works” conclusion: current evidence suggests a high probability that bicarbonate reduces the initiation of renal replacement therapy (RRT), while it does not establish a short-term survival benefit.
Three takeaways
- RRT initiation was lower: pooled RR 0.74 (95% credible interval [CrI] 0.59–0.98), with a 97.2% posterior probability of benefit.
- Short-term mortality remained uncertain: RR 0.94 (95% CrI 0.73–1.22), with the interval compatible with benefit or harm.
- The result supports selective rather than routine treatment: trial populations, severity thresholds and dosing strategies varied, and RRT initiation is partly influenced by clinician judgement.
What question did the study address?
The review asked whether intravenous sodium bicarbonate, compared with placebo or usual care, improves mortality and kidney-related outcomes in critically ill adults with metabolic acidosis.
The authors searched MEDLINE, Embase and CENTRAL, screened trial registries and updated the search on 2 April 2026. Four randomised controlled trials met eligibility criteria: BICAR-ICU, BICARICU-2, the SODa-BIC pilot and SODa-BIC. Together they enrolled 1,544 adults; 768 received bicarbonate and 776 received control treatment.
Two trials were open label and two were double-blind and placebo controlled. The primary outcome was all-cause mortality at 28 or 30 days. Secondary outcomes included RRT initiation, 90-day mortality, RRT-free days, vasopressor-free days, lengths of stay and adverse events.
The analysis used a Bayesian random-effects model. Unlike a conventional confidence interval alone, this approach allowed the investigators to estimate a posterior probability that treatment was beneficial for each outcome. That probability is informative, but it does not compensate for limitations in the underlying trials.
What were the main results?
For short-term mortality, the pooled risk ratio was 0.94 (95% CrI 0.73–1.22). The posterior probability of any mortality benefit was 72.1%, but the credible interval was wide and crossed no effect. In practical terms, the evidence did not establish that bicarbonate improves survival and remained compatible with both clinically relevant benefit and harm.
The kidney-related result was more persuasive. Bicarbonate was associated with a lower probability of RRT initiation: RR 0.74 (95% CrI 0.59–0.98), with a 97.2% posterior probability of benefit. Results were broadly consistent across alternative prior assumptions and in a frequentist sensitivity analysis.
Other pooled outcomes—including 90-day mortality, RRT-free days, vasopressor-free days and lengths of stay—were small or imprecise. A post hoc analysis suggested an 82.7% probability of less RRT dependence at the longest available follow-up, but its 95% CrI included no effect. The authors found no consistent safety signal; hypernatraemia, metabolic alkalosis and hypocalcaemia were numerically more frequent with bicarbonate, but the credible intervals included no effect.
GRADE certainty was moderate for short-term mortality, RRT initiation and several secondary outcomes, and high for ICU and hospital length of stay.
Why might RRT fall without a survival benefit?
Severe acidaemia can itself contribute to a decision to start dialysis. Correcting it may remove one trigger for RRT, improve haemodynamic tolerance or allow time for kidney recovery. That mechanism could reduce dialysis exposure without necessarily changing the many other determinants of death in critical illness.
However, RRT initiation is not a fully objective endpoint. Protocols specified indications, but the final decision still depended partly on clinicians and local practice. This matters because BICAR-ICU and BICARICU-2 were open label: knowledge of treatment allocation could have influenced when dialysis was started. The two blinded SODa-BIC trials reduce, but do not eliminate, this concern because changes in pH, bicarbonate and sodium may reveal allocation.
The analysis also cannot determine whether avoiding RRT represents durable preservation of kidney function. Longer-term kidney recovery was not reported consistently, and the review used aggregate trial-level rather than individual participant data.
Which patients were represented?
The four trials were clinically heterogeneous. BICAR-ICU and BICARICU-2 enrolled patients with pH ≤7.20; BICARICU-2 additionally required stage 2 or 3 acute kidney injury. The SODa-BIC studies enrolled vasopressor-treated patients with pH <7.30 and base excess ≤−4 mEq/L, generally representing less severe acidaemia and lower event rates.
Treatment also differed. The BICAR trials used intermittent 4.2% bicarbonate boluses, whereas the SODa-BIC studies used protocolised continuous infusions. All aimed to correct acidaemia, but the dose, concentration and clinical context were not identical.
Subgroup analyses suggested a higher probability of mortality benefit in patients with more advanced acute kidney injury, but estimates were imprecise and there was no clear evidence of effect modification. These analyses should not be used to claim a proven survival advantage in a subgroup.
What should clinicians do differently?
First, continue to treat the cause of metabolic acidosis. Bicarbonate is not a substitute for source control, restoration of perfusion, management of sepsis or shock, treatment of toxic exposure, or timely RRT when urgent indications are present.
Second, consider bicarbonate selectively when severe acidaemia—particularly with acute kidney injury—may itself be contributing to haemodynamic instability or an impending dialysis decision. Account for ventilation, sodium and fluid burden, ionised calcium and potassium, and monitor the clinical response rather than treating a laboratory value in isolation.
Third, do not generalise these data to every acidosis syndrome. The review excluded diabetic ketoacidosis, toxic alcohol poisoning, paediatric populations and other specific indications. Contemporary sepsis guidance continues to advise against routine bicarbonate solely to improve haemodynamics in hypoperfusion-induced lactic acidaemia, while allowing a more selective role in severe metabolic acidaemia with acute kidney injury.
Bottom line
Across four randomised trials involving 1,544 critically ill adults, sodium bicarbonate had a high probability of reducing RRT initiation but did not show a clear short-term mortality benefit. The most defensible interpretation is selective, physiology-informed use—especially in severe acidaemia with acute kidney injury—rather than routine administration to all ICU patients with metabolic acidosis.
Reference
Kitisin N, Walker HGM, Jung B, Jaber S, Mouncey P, Ke L, Raykateeraroj N, Hikasa Y, Caroli A, Serpa Neto A. Sodium bicarbonate therapy for metabolic acidosis in critically ill patients: a Bayesian systematic review and meta-analysis of randomised controlled trials. Intensive Care Medicine. Published online 2 November 2026. doi:10.1007/s00134-026-08602-3.