Association of sodium bicarbonate use during CPR with sustained return of spontaneous circulation in adult in-hospital cardiac arrest

Sodium bicarbonate (SB) is frequently administered during cardiopulmonary resuscitation (CPR) despite contemporary guidelines advising against routine use. Observational evaluation is challenging because SB is often given later in resuscitations, creating confounding by indication and resuscitation-time bias. We assessed the association between SB administration during CPR and sustained return of spontaneous circulation (ROSC). We performed a retrospective cohort study of adult (≥ 18 years) in-hospital cardiac arrests from January 1, 2020, to January 2, 2025, at a single academic hospital (~ 1,000 beds). Arrests were documented in real time on standardized Code Blue forms with stopwatch timestamps and supplemented with electronic medical record data. The primary outcome was sustained ROSC (≥ 20 min). Crude associations were estimated using risk difference (RD), risk ratio (RR), and odds ratio (OR). To account for time-varying confounding, we implemented a marginal structural model (MSM) with stabilized inverse probability of treatment weighting using discrete 2-minute intervals from CPR initiation to sustained ROSC or termination. SB was modeled as a time-varying initiation exposure. Among 1,301 events, 1,298 were included after excluding 3 with missing exposure/outcome. SB was administered in 565/1,298 (43.5%) cases; sustained ROSC occurred in 803/1,298 (61.9%) overall. In crude analysis, sustained ROSC occurred in 304/565 (53.8%) SB-exposed versus 499/733 (68.1%) unexposed cases (RD − 14.3%, 95% CI − 19.6 to − 9.0; OR 0.55, 95% CI 0.44–0.69; p < 0.0001). In the weighted MSM, SB initiation was not associated with higher sustained ROSC (aRR 0.98, 95% CI 0.91–1.06; marginal risk difference − 1.2%, 95% CI − 4.7 to + 2.3). Exploratory timing and 15-minute landmark analyses did not identify a clear administration window associated with higher sustained ROSC. In this cohort, SB use during CPR was associated with lower sustained ROSC in crude analyses, but the association attenuated toward the null after adjustment for measured time-varying confounding. Exploratory analyses did not identify a clear timing-dependent association between SB initiation and sustained ROSC. Because downstream outcomes and specific guideline-supported indications were incompletely captured, these findings should be interpreted as limited to sustained ROSC in an unselected in-hospital cardiac arrest cohort.