Early 8.4% sodium bicarbonate and in-hospital mortality in cardiogenic shock with severe acidemia and hyperlactatemia: A two-hour landmark overlap-weighted retrospective cohort study

Abstract

Objective

To evaluate the association between early concentrated 8.4% sodium bicarbonate and subsequent in-hospital mortality in patients with cardiogenic shock, severe acidemia, and hyperlactatemia.

Methods

We conducted a retrospective two-hour landmark cohort study using the single-center MIMIC-IV v3.1 database. True time zero was the first time pH ≤7.20 and lactate ≥4 mmol/L were both known within a two-hour pairing window. Patients with pre-time-zero bicarbonate exposure, renal replacement therapy, or diabetic ketoacidosis were excluded. Patients alive and still hospitalized at two hours were classified according to 8.4% sodium bicarbonate administration during the 0- to 2-hour window. Propensity-score overlap weighting was used to balance baseline covariates. The primary outcome was subsequent in-hospital mortality.

Results

Among 489 landmark patients, 110 received early 8.4% bicarbonate and 379 did not. After overlap weighting, the maximum absolute standardized mean difference was 0.091. Weighted mortality was 80.4% with early bicarbonate versus 70.2% without early treatment, corresponding to a risk difference of 10.2 percentage points (95% CI, −0.1 to 20.4; P=.052) and a risk ratio of 1.145 (95% CI, 1.001–1.310; P=.049). Sensitivity analyses showed broadly consistent point estimates, although confidence intervals and target overlap populations varied.

Conclusions

Early 8.4% sodium bicarbonate did not demonstrate a survival benefit among patients who survived to the two-hour landmark. Estimates were directionally compatible with higher mortality, but residual confounding and statistical uncertainty preclude a causal conclusion of harm.

1. Background

Cardiogenic shock remains a highly lethal syndrome despite advances in revascularization, vasoactive therapy, and temporary mechanical circulatory support.^1,2 Severe acidemia with hyperlactatemia identifies marked circulatory and metabolic failure and is consistently associated with poor prognosis.^3–5 In this setting, clinicians must address the cause of shock while managing the immediate cardiovascular consequences of low pH.

Sodium bicarbonate is attractive because acidemia may impair myocardial contractility, reduce vascular responsiveness to catecholamines, promote arrhythmias, and aggravate hyperkalemia.^6,7 However, bicarbonate does not correct pump failure or tissue hypoperfusion. It generates carbon dioxide, may reduce ionized calcium, and delivers a substantial sodium and osmotic load; these effects may be particularly relevant in patients with limited ventilatory reserve, pulmonary congestion, or right ventricular dysfunction.^6–8

Evidence remains uncertain. An international practice study showed wide variation in bicarbonate use. ⁹ The BICAR-ICU trial did not improve its primary composite outcome overall, although an acute kidney injury subgroup suggested less renal replacement therapy. ¹⁰ The subsequent BICARICU-2 trial did not reduce 90-day mortality in patients with severe metabolic acidemia and moderate-to-severe acute kidney injury. ¹¹ Recent observational studies and a target-trial emulation have reported heterogeneous associations, reflecting differences in populations, indications, and analytic methods.^12–17

Cardiogenic shock presents an additional methodological challenge because bicarbonate is preferentially given to patients with lower pH, higher lactate, hyperkalemia, escalating vasoactive support, or impending arrest. Exposure defined after physiologic deterioration may also introduce immortal-time or treatment-availability bias if follow-up starts before treatment classification is complete. ¹⁸ We therefore used a two-hour landmark design to estimate whether early 8.4% sodium bicarbonate was associated with subsequent in-hospital mortality among patients alive and hospitalized two hours after severe acidemia with hyperlactatemia was recognized.

2. Methods

2.1. Data source and study design

We conducted a retrospective two-hour landmark cohort study using deidentified MIMIC-IV version 3.1 data from Beth Israel Deaconess Medical Center (2008–2022), accessed through PhysioNet by a credentialed user under the data use agreement.^19–22 The complete available period was used, and all eligible admissions were included consecutively without random or selective sampling. Reporting followed STROBE. ²³ Because this secondary analysis involved only deidentified data without patient contact, intervention, or re-identification, additional institutional ethics approval and individual informed consent were not required. The study followed the Declaration of Helsinki(1975, as revised in 2024).

2.2. Cohort definition and true time zero

Adults aged ≥18 years with diagnosis-coded cardiogenic shock (ICD-9-CM 785.51 or ICD-10-CM R57.0) were identified, restricting analysis to the first ICU stay. Severe acidemia with hyperlactatemia required pH ≤7.20 and lactate ≥4 mmol/L. True time zero was the first time both criteria were known; when measured separately, the later timestamp was used, with measurements required within two hours. Available baseline surrogates, including mean arterial pressure, norepinephrine rate, pressor count, mechanical ventilation, mechanical circulatory support, inotrope use, and pH/lactate trends, were incorporated into the treatment model. Cardiac output/index, pulmonary capillary wedge pressure, cardiac power output, detailed ventricular function, coronary anatomy, revascularization success, and adjudicated shock etiology were not consistently available. Previous MIMIC-IV studies support the feasibility of cardiogenic-shock phenotyping in this database.^24,25

2.3. Exclusion criteria

Patients were excluded if they had received sodium bicarbonate before true time zero, if renal replacement therapy had already begun before true time zero, or if diabetic ketoacidosis was present. These exclusions were intended to remove patients in whom bicarbonate use reflected ongoing pre-baseline therapy, dialysis-related management, or a distinct metabolic indication rather than early treatment of shock-associated acidemia with hyperlactatemia. To avoid repeated-patient dependence, the analytic cohort retained only the first eligible hospitalization per patient.

2.4. Landmark population and exposure

Exposure was assessed from true time zero through two hours, and the landmark population included patients alive and still hospitalized at two hours. Early treatment was defined as any 8.4% sodium bicarbonate administration during this interval. Because medication records did not encode clinical indication, exposure was classified regardless of documented treatment intent. The comparator was no early 8.4% bicarbonate; later administration did not alter baseline classification. We restricted exposure to reliably identifiable concentrated 8.4% bicarbonate and excluded low-concentration or nonspecific bicarbonate products because their indications, administered volumes, osmotic loads, and documentation may differ.

2.5. Outcome

The primary outcome was subsequent in-hospital mortality after the two-hour landmark. We used in-hospital death as a binary endpoint rather than time-to-death because a subset of deaths lacked exact death timestamps. Treating those patients as censored or excluding them would have created outcome misclassification. Alive discharge was not treated as censoring for the primary endpoint.

2.6. Covariates and missing data

Baseline covariates included age, sex, pH, lactate, creatinine, potassium, bicarbonate, PCO₂, mean arterial pressure, norepinephrine rate, active vasopressor count, mechanical ventilation, mechanical circulatory support, inotrope use, pre-baseline pH/lactate trends, and ICU time before time zero. Creatinine, potassium, bicarbonate, and PCO₂ were the most recent values within 24 hours; mean arterial pressure was the median during the preceding four hours; and pH/lactate trends were calculated as the last minus earliest value during that interval. Norepinephrine was represented by the active infusion rate at time zero, whereas other vasopressors were summarized by active-agent count. Missingness was approximately 20%–22% for creatinine, potassium, bicarbonate, and mean arterial pressure and 1% for PCO₂. Missing continuous variables were median imputed with corresponding missingness indicators; when only one pH or lactate value was available, the trend was coded as zero.

2.7. Weighting and estimand

Propensity-score overlap weighting was used to estimate the average treatment effect in the overlap population.^26,27 Propensity scores were estimated by bias-reduced multivariable logistic regression ²⁸ including baseline covariates, missingness indicators, and selected nonlinear terms for pH, lactate, creatinine, potassium, norepinephrine rate, and ICU time. Overlap weights were 1−e(X) for early-treated patients and e(X) for controls, with no trimming. Balance and practical positivity were assessed using absolute standardized mean differences, effective sample sizes, and weight distributions; an SMD <0.10 was preferred.

2.8. Statistical analysis

Weighted mortality risks were estimated using normalized Hájek means, R̂a = ΣI(Ai=a)wiYi/ΣI(Ai=a)wi. The risk difference was R̂early − R̂no-early, and the risk ratio was R̂early/R̂no-early. Robust Hájek-type standard errors and the delta method were used for the risk difference and log risk ratio, respectively, with Wald confidence intervals and P values, treating estimated propensity scores and overlap weights as fixed.

A block bootstrap refitting the propensity-score model within resamples was explored, but repeated resamples failed convergence or practical-positivity checks; bootstrap percentile intervals were therefore not used for primary inference. A joint sandwich variance estimator accounting for first-stage propensity-score estimation was not implemented for the bias-reduced nonlinear treatment model. Thus, propensity-score estimation uncertainty was not fully propagated, and effect estimates were interpreted as associations within the overlap population rather than causal effects.

2.9. Sensitivity and alternative-specification analyses

Sensitivity analyses excluded baseline mechanical circulatory support, restricted the cohort to measured potassium <6.0 mmol/L, excluded cardiac-arrest codes, used complete cases, evaluated a hemodynamic-support-enriched phenotype, and used a one-hour landmark. Additional analyses omitted pre-baseline pH/lactate trends, required a 30-minute pairing window, and used five multiply imputed datasets. Propensity scores were refit for each analysis, with reduced models used when separation, nonidentifiable terms, or inadequate overlap occurred; model choice was based on treatment-model diagnostics rather than outcomes. Analyses were considered diagnostically acceptable with maximum weighted SMD ≤0.15, effective sample size ≥30 per group, and 99th-percentile weight <0.98, with SMD <0.10 preferred. Each analysis targeted its own overlap population and was treated as an exploratory robustness check without adjustment for multiple comparisons.

3. Results

3.1. Cohort assembly

Among 3,354 diagnosis-coded cardiogenic shock admissions, 3,106 first ICU stays were identified, of which 681 met the severe acidemia and hyperlactatemia criteria. After excluding pre-time-zero bicarbonate use, renal replacement therapy, or diabetic ketoacidosis, 522 hospitalizations remained, corresponding to 521 first eligible hospitalizations per patient. Twenty-five patients died within two hours; 3 (12.0%) had received early 8.4% bicarbonate. Seven additional records had internally inconsistent discharge or death timestamps and indeterminate landmark status; 2 had early bicarbonate exposure. The final two-hour landmark cohort included 489 patients (Figure 1; Additional file: Table S5).

Figure 1.

Cohort assembly and two-hour landmark design. (a) Selection of the final two-hour landmark cohort from diagnosis-coded cardiogenic shock admissions. (b) Definition of true time zero, the 0- to 2-hour exposure-assessment window, the two-hour landmark, and subsequent follow-up for in-hospital mortality. Exposure was classified as early versus no early administration of 8.4% sodium bicarbonate. Only patients alive and still hospitalized at the landmark were included in the primary analysis.

3.2. Exposure

Among the 489 landmark patients, 110 received early 8.4% sodium bicarbonate and 379 did not. Among treated patients, the median time to first administration was 38 minutes (IQR, 17–73), and the median cumulative dose during the 0- to 2-hour window was 100 mEq (IQR, 50–150); 71 of 109 patients with quantifiable dose received ≥100 mEq. During the exposure window, non-bicarbonate volume-expansion records were present in 42.5% of no-early and 93.6% of early-treatment patients, blood-product records in 6.6% and 10.0%, and vasoactive records in 43.3% and 89.1%, respectively. These descriptive event counts do not represent fluid balance or causal cointervention effects. Target bicarbonate was subsequently administered during hours 2–24 in 21.9% of no-early and 64.5% of early-treatment patients, indicating that the estimand concerns early initiation rather than never-use (Additional file: Table S5 and S6; Figure. S2 and S3).

3.3. Baseline characteristics and overlap

Before weighting, early bicarbonate recipients had more severe acid-base abnormalities, greater potassium derangement, higher norepinephrine rates, more active vasopressors, and more frequent mechanical circulatory support. After overlap weighting, the maximum absolute standardized mean difference across modeled terms was 0.091. Effective sample sizes were 197.4 in the no-early group and 93.5 in the early-treatment group (Table 1; Additional file: Figure. S1).

Table 1.

Baseline characteristics of the analytic landmark cohort.

Variable	No early 8.4% bicarbonate (n=379)	Early 8.4% bicarbonate (n=110)	Unweighted SMD	Overlap-weighted SMD
Age, years	68.00 [58.50, 77.00]	69.00 [60.00, 77.75]	0.044	0.005
Female sex, n (%)	163 (43.0)	54 (49.1)	0.122	0.015
pH at time zero	7.16 [7.11, 7.18]	7.10 [7.05, 7.16]	0.521	0.027
Lactate at time zero, mmol/L	6.80 [5.10, 9.40]	8.40 [6.80, 11.23]	0.419	0.018
Bicarbonate, most recent within 24 h before time zero, mmol/L	16.00 [14.00, 19.00]	15.00 [11.25, 16.00]	0.435	0.040
PCO2, most recent within 24 h before time zero, mm Hg	48.00 [39.50, 56.50]	43.00 [34.25, 52.00]	0.289	0.021
Creatinine, most recent within 24 h before time zero, mg/dL	1.90 [1.40, 2.20]	2.00 [1.73, 2.90]	0.233	0.002
Potassium, most recent within 24 h before time zero, mmol/L	4.70 [4.20, 5.00]	4.80 [4.50, 5.78]	0.471	0.032
Mean arterial pressure, median during 4 h before time zero, mm Hg	72.50 [67.75, 78.50]	72.50 [65.00, 77.00]	0.151	0.007
Norepinephrine infusion rate at time zero, μg/kg/min	0.00 [0.00, 0.10]	0.20 [0.00, 0.40]	0.632	0.040
Number of vasopressor agents at time zero	0.00 [0.00, 1.00]	1.50 [0.00, 2.00]	0.697	0.038
Mechanical ventilation at time zero, n (%)	44 (11.6)	18 (16.4)	0.137	0.011
Mechanical circulatory support at time zero, n (%)	28 (7.4)	17 (15.5)	0.256	0.007
Inotrope use at time zero, n (%)	20 (5.3)	7 (6.4)	0.046	0.004
pH change in 4 h before time zero	0.00 [-0.12, 0.00]	0.00 [-0.06, 0.00]	0.160	0.025
Lactate change in 4 h before time zero, mmol/L	0.00 [0.00, 1.95]	0.00 [0.00, 0.88]	0.166	0.019
ICU time before time zero, h	4.05 [1.09, 12.84]	1.76 [0.65, 11.49]	0.135	0.069

Continuous variables are median [IQR]; binary variables are n (%). The maximum SMD across all modeled terms, including nonlinear terms and missingness indicators, was 0.091. SMD, standardized mean difference.

3.4. Primary outcome

In the unweighted landmark cohort, in-hospital mortality was 80.9% (89/110) among early bicarbonate recipients and 64.1% (243/379) among patients without early treatment. After overlap weighting, mortality was 80.4% versus 70.2%, corresponding to a risk difference of 10.2 percentage points (95% CI, −0.1 to 20.4; P=.052) and a risk ratio of 1.145 (95% CI, 1.001–1.310; P=.049). The risk-difference confidence interval included the null, whereas the risk-ratio interval narrowly excluded it (Figure 2).

Figure 2.

Primary overlap-weighted estimates of subsequent in-hospital mortality. (a) Overlap-weighted mortality risks among patients with and without early 8.4% sodium bicarbonate administration. (b) Corresponding absolute and relative effect estimates, expressed as the risk difference and risk ratio with 95% confidence intervals. Confidence intervals were calculated using robust Hájek-type standard errors and the delta method, treating estimated propensity scores and overlap weights as fixed.

3.5. Weight and positivity diagnostics

Primary overlap weights showed no marked extremes. In the no-early group, median, 95th-percentile, 99th-percentile, and maximum weights were 0.108, 0.536, 0.695, and 0.876, respectively; corresponding values in the early-treatment group were 0.581, 0.889, 0.934, and 0.946 (Additional file: Table S7).

3.6. Sensitivity and alternative-specification analyses

Point estimates remained above 1.0 across sensitivity analyses, although confidence intervals and target overlap populations varied. RRs were 1.16 (95% CI, 1.01–1.34) after excluding baseline mechanical circulatory support, 1.13 (0.97–1.32) with potassium <6.0 mmol/L, 1.20 (1.03–1.40) after excluding cardiac-arrest codes, 1.09 (0.93–1.27) in complete cases, 1.14 (0.98–1.34) in the hemodynamic-support-enriched phenotype, and 1.11 (0.97–1.27) with a one-hour landmark. Similar estimates were obtained after omitting pre-baseline pH/lactate trends (RR 1.14, 95% CI 1.00–1.30), using a 30-minute pairing window (1.14, 0.99–1.30), and multiple imputation (1.14, 1.00–1.30). Overall, estimates were consistently nonprotective, although several confidence intervals included the null. These analyses were interpreted as robustness checks rather than independent confirmatory tests because they targeted different overlap populations and were not adjusted for multiple comparisons (Figure 3; Additional file: Tables S3 and S4).

Figure 3.

Primary and sensitivity analyses of subsequent in-hospital mortality. The forest plot presents overlap-weighted risk ratios comparing early versus no early 8.4% sodium bicarbonate administration in the primary two-hour landmark analysis and selected diagnostically acceptable sensitivity analyses summarized in Supplementary Table S3. Squares indicate point estimates, horizontal lines indicate 95% confidence intervals, and the dashed vertical line denotes the null value of 1.0. Each analysis used an analysis-specific propensity-score model and therefore targeted its own overlap population. CI, confidence interval; RR, risk ratio.

4. Discussion

4.1. Principal findings

In this two-hour landmark cohort of patients with cardiogenic shock and severe acidemia with hyperlactatemia, early 8.4% sodium bicarbonate was not associated with improved in-hospital survival. After overlap weighting balanced the included measured baseline covariates, treated patients had an estimated 10.2-percentage-point higher absolute risk and a 14.5% higher relative risk of death. The risk-difference confidence interval narrowly included the null, whereas the risk-ratio confidence interval narrowly excluded the null under the fixed-weight inference; therefore, the study does not establish causal harm. The defensible conclusion is that concentrated bicarbonate showed no survival benefit in this conditional landmark population and that clinically important harm cannot be excluded.

4.2. Clinical interpretation

Bicarbonate is often given when severe acidemia creates an urgent need to normalize pH. In cardiogenic shock, however, low pH and high lactate usually reflect the severity of pump failure and tissue hypoperfusion. Bicarbonate may improve extracellular pH without restoring coronary perfusion, reversing ventricular dysfunction, unloading a congested ventricle, or improving microcirculatory oxygen delivery. If the shock process continues, a more favorable blood-gas value may not translate into cellular recovery. Carbon dioxide generation may worsen intracellular acidosis when ventilation is limited; alkalinization can reduce ionized calcium and potentially impair contractility; and the sodium and osmotic load may worsen pulmonary congestion or filling pressures.^6–8 These mechanisms are not directly demonstrated here, but they explain why biochemical correction may not improve a patient-centered outcome.

4.3. Context of existing evidence

The findings align with persistent uncertainty in the bicarbonate literature. BICAR-ICU did not improve its primary outcome overall, although an acute kidney injury subgroup suggested less renal replacement therapy. ¹⁰ BICARICU-2 subsequently found no reduction in 90-day mortality among patients with severe metabolic acidemia and moderate-to-severe acute kidney injury, despite a lower use of kidney replacement therapy.^11,17 Contemporary observational studies have reported benefit, neutrality, or potential harm depending on phenotype, timing, treatment intensity, and analytic approach.^9,12–16 Such heterogeneity is clinically plausible because metabolic acidemia is a syndrome rather than a single disease. Most prior studies enrolled mixed ICU or septic populations and used 4.2% infusions or broadly defined bicarbonate exposure. They do not directly resolve whether early concentrated 8.4% bicarbonate is beneficial in pump-failure shock, where sodium load, carbon dioxide generation, congestion, and limited myocardial reserve may be particularly important.

4.4. Methodological considerations

The landmark design prevents follow-up from beginning before exposure classification is complete and thereby reduces a major time-related bias. ¹⁸ Patients otherwise must survive long enough to be classified as treated, whereas rapidly dying patients may have no opportunity to receive bicarbonate. Its tradeoff is conditional interpretation: results apply only to patients alive and hospitalized at two hours. If bicarbonate influenced death during the first two hours, conditioning on landmark survival could introduce selection bias. Overlap weighting also changed the target population. Early-treated patients were substantially sicker before adjustment; weighting emphasized patients with sufficient measured similarity to support a comparison in the region of clinical equipoise.^26,27 The resulting average treatment effect in the overlap population should not be generalized to patients who were almost certain to receive or not receive treatment. Moreover, balance in measured covariates does not remove confounding by clinician assessment of impending arrest, ventricular function, coronary anatomy, revascularization status, invasive hemodynamics, or rapidly changing vasoactive requirements. The database did not consistently provide cardiac index, cardiac output, pulmonary capillary wedge pressure, cardiac power output, detailed echocardiographic ventricular function, coronary anatomy, or revascularization success, so these clinically important hemodynamic dimensions could not be adjusted directly.

4.5. Robustness of the findings

Point estimates remained above the null after excluding baseline mechanical circulatory support, restricting to measured potassium below 6.0 mmol/L, excluding admissions with cardiac-arrest codes, using complete cases, enriching for hemodynamic support, and shortening the landmark window. This consistency reduces concern that the primary estimate resulted entirely from one subgroup definition or exposure window. However, the analyses used different propensity-score models and targeted different overlap populations. Several required reduced treatment models to maintain adequate overlap and stable weights, and several confidence intervals included the null. They should therefore be interpreted as robustness checks rather than independent causal confirmations or evidence of a dose-response relationship.

4.6. Clinical implications

This study should not be interpreted as a blanket prohibition against bicarbonate. Potential indications may include severe hyperkalemia, selected toxicologic conditions, substantial bicarbonate loss, or clinical circumstances in which ventilation or renal replacement therapy alters the risk-benefit balance. The present analysis addresses a narrower question: whether early administration of concentrated 8.4% sodium bicarbonate after recognition of severe acidemia with hyperlactatemia in cardiogenic shock was associated with subsequent in-hospital survival. In this conditional landmark population, the data did not demonstrate a survival benefit. Contemporary cardiogenic-shock guidance emphasizes rapid etiologic and hemodynamic phenotyping, serial assessment of perfusion, appropriate vasoactive support, urgent revascularization when indicated, and timely escalation or de-escalation of mechanical circulatory support.^29–38 Bicarbonate should therefore be considered a targeted adjunct for selected indications rather than a substitute for correction of the underlying shock state. Complex toxicologic presentations may further illustrate the importance of integrating acid-base therapy with the precipitating cause and evolving hemodynamics, as reported in a recent case of reversible atrial fibrillation and cardiogenic shock following acute aluminum phosphide poisoning. ³⁹

4.7. Limitations

Several limitations should be acknowledged. Residual and time-dependent confounding remains possible because bicarbonate use may have been triggered by clinical deterioration during the exposure window, including fluid administration, blood products, or vasoactive escalation. The landmark design excluded patients who died within two hours and therefore conditions inference on landmark survival; among 25 exact early deaths, 3 (12.0%) had received early bicarbonate. Seven additional records had inconsistent discharge, death, or observation timestamps and could not be reliably classified. Missing laboratory and hemodynamic data were handled with median imputation and missingness indicators, with directionally similar results in multiple-imputation sensitivity analysis, although unmeasured physiology cannot be recovered.

Cardiogenic shock was identified from ICD diagnosis codes rather than contemporaneous adjudication, and important hemodynamic and clinical features,including cardiac output, filling pressures, ventricular function, coronary anatomy, revascularization success, and impending arrest,were not consistently available. Medication records did not encode treatment indication, and later bicarbonate use was common, so the estimand reflects early initiation rather than never-use. Variance estimates treated propensity scores and overlap weights as fixed and therefore did not fully propagate first-stage estimation uncertainty; multiple comparisons were not adjusted, and the single-center setting limits generalizability. A formal dose–response analysis was not performed because cumulative dose was influenced by evolving severity, treatment opportunity, and survival, creating substantial time-dependent confounding and treatment-duration bias.

4.8. Future directions

Future studies should focus on physiologically defined subgroups, including severe acute kidney injury, refractory hyperkalemia, adequate ventilation for carbon dioxide clearance, and right-ventricular failure with congestion. A prospective trial should prespecify treatment triggers, concentration, dose, and infusion strategy; capture serial pH, lactate, ionized calcium, carbon dioxide, and hemodynamics; and assess both physiologic response and patient-centered outcomes. External validation in contemporary cardiac ICUs would also help determine whether the present association is reproducible.

5. Conclusions

Among patients with cardiogenic shock and severe acidemia with hyperlactatemia who survived to a two-hour landmark, early 8.4% sodium bicarbonate did not demonstrate a survival benefit. Estimates were directionally compatible with higher mortality, but residual confounding and statistical uncertainty preclude a causal conclusion of harm. These findings do not support routine early use of concentrated bicarbonate solely for the purpose of improving survival.

Appendix.

List of abbreviations

ATO

Average treatment effect in the overlap population

CI

Confidence interval

DKA

Diabetic ketoacidosis

ESS

Effective sample size

ICU

Intensive care unit

MCS

Mechanical circulatory support

MIMIC-IV

Medical Information Mart for Intensive Care IV

PCO2

Partial pressure of carbon dioxide

RD

Risk difference

RRT

Renal replacement therapy

RR

Risk ratio

SMD

Standardized mean difference.

ORCID iD

Cao Ma https://orcid.org/0009-0007-2979-8457

Ethical considerations

The collection of patient information and creation of the MIMIC-IV research resource were reviewed by the Institutional Review Board of Beth Israel Deaconess Medical Center, which approved the data-sharing initiative and granted a waiver of informed consent. This secondary analysis used only deidentified data accessed by a credentialed user under the PhysioNet data use agreement; no directly identifying patient details were extracted, reported, or redistributed. Because the present study involved only secondary analysis of deidentified data without patient contact, intervention, or re-identification, additional institutional ethics approval and individual informed consent were not required. The study was conducted in accordance with the ethical principles of the Declaration of Helsinki(1975, as revised in 2024).

Data Availability Statement

MIMIC-IV version 3.1 is available through PhysioNet to credentialed users who complete the required training and sign the data use agreement. The source data cannot be redistributed by the authors.*