QT Prolongation, Ventricular Arrhythmia, and Cardiac Arrest Signals During Ceftriaxone Co-therapy with Individual Proton Pump Inhibitors: A Multidatabase Study

Abstract

The concomitant use of ceftriaxone and proton pump inhibitors (PPIs) is common in hospital practice. However, it is unclear whether individual PPIs differ in their effects on QT interval prolongation, ventricular arrhythmia, or cardiac arrest, collectively termed as QVC events. We conducted a two-stage, real-world study. First, we screened the United States Food and Drug Administration Adverse Event Reporting System (FAERS) and the Canada Vigilance Adverse Reaction (CVAR) database with standard disproportionality measures (reporting odds ratio and proportional reporting ratio) and six drug-drug interaction (DDI) algorithms to identify combination signals that exceeded component signals. Second, we validated signal-positive combinations in the Medical Information Mart for Intensive Care IV (MIMIC-IV) intensive care unit (ICU) electronic health record (EHR) cohort by assembling adult inpatients with overlapping ceftriaxone-PPI exposures. The primary outcome was 28-day QVC events. Multivariable Cox proportional hazards models were the main analysis and complemented by propensity score matching, inverse probability of treatment weighting, and Fine-Gray competing-risk models. To address external generalisability, an additional validation was performed using ECG-ViEW II, an Asian electrocardiogram-linked real-world database. The combination of ceftriaxone and lansoprazole was significantly associated with QVC events, revealing notable DDIs (e.g., in FAERS, Ω025 = 0.54). To validate these findings, a cohort of 5,594 patients receiving ceftriaxone combined with PPIs from the MIMIC-IV database was analyzed using Cox proportional hazards models. The analyses corroborated the initial findings (lansoprazole vs. other PPIs, multivariate HR = 1.30; 95% CI: 1.10–1.54), with the risk associated with the three PPI combinations ranked as lansoprazole > pantoprazole > omeprazole. ECG-ViEW II provided supportive Asian external validation, showing a higher QVC risk for ceftriaxone plus lansoprazole than for ceftriaxone plus other PPIs. Evidence from two national pharmacovigilance systems and an ICU EHR cohort indicated that PPI choice modified cardiac safety during ceftriaxone therapy. Lansoprazole co-use confers a higher risk of QVC, whereas omeprazole appears relatively safer. Therefore, prospective confirmation is warranted.

Graphical Abstract

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1007/s12012-026-10174-z.

Introduction

Ceftriaxone is a third-generation cephalosporin with long-acting properties against a broad spectrum of pathogens [1]. Proton pump inhibitors (PPIs) are widely used to treat peptic ulcers and gastroesophageal reflux diseases [2]. PPIs function by inhibiting H⁺/K⁺-ATPase and effectively preventing gastric acid secretion. However, increasing reports of adverse events (AEs) associated with PPIs, particularly kidney injury and hypomagnesemia, have raised concerns [3]. PPI-induced hypomagnesemia has been associated with QT prolongation, a significant factor in the development of torsades de pointes arrhythmias [4].

Ceftriaxone is frequently co-prescribed with PPIs, particularly in hospitalised and critically ill patients [5]. Specifically, the co-administration of ceftriaxone and lansoprazole has been associated with QT prolongation and increased risk of in-hospital mortality [6]. Satoru et al. [7] indicated that the concurrent use of ceftriaxone and lansoprazole was associated with a higher risk of ventricular arrhythmias and cardiac arrest than other drug combinations.

Prompted by the individual cardiac risks of ceftriaxone and PPIs, and a critical knowledge gap regarding their interaction, this study systematically evaluated the association between concurrent ceftriaxone use and specific PPIs (omeprazole, lansoprazole, pantoprazole, esomeprazole, and rabeprazole) on the increasing risk of QT prolongation, ventricular arrhythmias, and cardiac arrest (collectively termed QVC events). Our multi-stage approach involved first generating hypotheses through pharmacovigilance analysis of the U.S. Food and Drug Administration (FDA) Adverse Event Reporting System (FAERS) and Canadian Vigilance Adverse Reaction (CVAR) databases, and then validating the resulting signals within the Medical Information Mart for Intensive Care IV(MIMIC-IV) cohort, which provides granular data for rigorous confounder adjustment. To further address external generalisability beyond North American data sources, we additionally performed supportive external validation using ECG-ViEW II, an Asian electrocardiogram-linked real-world database.

Materials and Methods

This observational real-world pharmacovigilance cohort study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guidelines and the Reporting of a Disproportionality Analysis for Drug Safety Signal Detection Using Individual Case Safety Reports in PharmacoVigilance (READUS-PV) [8, 9].

Data Sources

Four publicly accessible databases were used in this study: MIMIC-IV, FAERS, CVAR, and ECG-ViEW II. Access to the MIMIC-IV database was granted after approval was obtained (Certification number: 13465720). All the databases used in this study were anonymized and de-identified. Therefore, this study was exempt from the institutional ethics committee review and did not require informed consent. The MIMIC-IV database [10] contains electronic health records from the Intensive Care Unit (ICU) and emergency patients at the Beth Israel Deaconess Medical Center (2008–2019), offering valuable clinical data. The FAERS and CVAR databases are publicly available spontaneous AEs reporting systems managed by the FDA and Health Canada, respectively [11, 12]. ECG-ViEW II is a freely accessible Asian electrocardiogram database containing structured ECG parameters linked with demographic information, diagnoses, prescriptions, comorbidities, and laboratory results, and was used as an additional external validation dataset [13].

Target Drugs

Ceftriaxone, cefuroxime, and PPIs that were not withdrawn from the FDA market as of July 1, 2024, were selected and coded using the World Health Organization (WHO) Anatomical Therapeutic Chemical (ATC) codes (eTable 1). In the FAERS and CVAR databases, we used cefuroxime with PPIs as a negative control, given its documented null interaction [14], and levofloxacin with amiodarone as a positive control for its known QT-prolonging effect [15]. For the clinical validation phase of MIMIC-IV, we aimed to further assess the signals observed in FAERS/CVAR analysis. Ceftriaxone, our primary drug of interest, was the second most frequently prescribed cephalosporin in the cohort (N = 52,782), providing a substantial sample for robust analysis. In contrast, although cefuroxime was included as a key comparator in our initial pharmacovigilance screening, its extremely low usage frequency in MIC-IV (N = 17) precluded any statistically meaningful validation.

Data Extraction and Processing

The inclusion criteria for the data in the MIMIC-IV and ECG-ViEW II databases were as follows: (1) ≥ 1 dose of ceftriaxone, (2) ≥ 1 dose of PPI during ceftriaxone therapy, and (3) overlap ≥ 24 h. Therapy initiation was set as the later of the first documented doses of ceftriaxone or PPIs, and termination occurred when either drug was discontinued for ≥ 24 h. Patients aged > 120 years or those with missing key baseline covariates were excluded. The PPIs were stratified into lansoprazole and other groups.

Data from January 2004 to September 2024 were extracted from FAERS and CVAR databases. Duplicate reports in the FAERS data were removed according to official rules. Reports were sorted by ‘CASEID,’ ‘FDA_DT,’ and ‘PRIMARYID.’ For identical ‘CASEID’ values, the report with the latest ‘FDA_DT’ was retained; if ‘FDA_DT’ values were also identical, the highest ‘PRIMARYID’ was kept. Drug names were standardized using MedEx-UIMA [16]. For drug names that MedEx-UIMA could not standardize, attempts were made to find the corresponding generic names using PubChem and DrugBank. The ultimate goal was to match each drug name in the FAERS database with its corresponding generic name as accurately as possible. In addition, we performed a thorough review of the reports with overlapping key fields. We also eliminated records with quality issues, including age > 120 years, cases where the same event code was linked to different sexes, and conflicting information regarding medications [17]. For descriptive summaries of FAERS/CVAR patient outcomes, a single adverse event report could contain multiple outcome labels. To avoid repeated counting of the same report in the baseline characteristics table, each report was assigned to the most severe outcome category according to a prespecified hierarchy based on the FAERS outcome classes [18]. Because cephalosporin–PPI interactions are rarely listed in commonly used drug–drug interaction (DDI) tools, reporters may not consistently code the target drugs as primary suspect, secondary suspect, or interacting drugs [19]. Therefore, all reports in the database were included, regardless of whether the doubts regarding the function of the target drugs in the AEs was Primary Suspect Drug (PS), Secondary Suspect Drug (SS), concomitant drug (C), or interacting drug (I). To prevent confounding from multiple drug interactions, we only analyzed cases limited to the concurrent use of one cephalosporin and one PPI. Cases involving three or more target drugs were excluded from the study. Reports from Canada were excluded from the FAERS database to avoid redundancy with CVAR.

Outcomes

The primary outcome was the incidence of QVC events. AE reports were extracted from the FAERS and CVAR databases using the Preferred Terms (PTs) mentioned in the Medical Dictionary for Regulatory Activities (MedDRA, version 27.1). QVC events were identified using Standardized MedDRA Queries (SMQs) and high-level term (HLT) levels. The selected outcomes were mainly composed of “Ventricular arrhythmias and cardiac arrest” (HLT code: 10047283), “Torsade de pointes/QT prolongation” (SMQ code: 20000001), and “Ventricular tachyarrhythmias” (SMQ code: 20000058), with specific details outlined in Supplement 2 eTable 2 [20]. For the MIMIC-IV and ECG-ViEW II databases, a QVC event was defined as a new-onset composite endpoint from post-medication electrocardiogram (ECG) data, requiring a patient to meet either of the following criteria: (1) a quantitative threshold (QTc interval > 500 ms or an increase of > 60 ms from baseline) or (2) a new documentation of a clinical diagnosis (“QT prolongation,” “ventricular arrhythmia,” or “cardiac arrest”), which was confirmed to be absent at baseline [21]. The primary outcome was the incidence of QVC events within 28 days of the initiation of the combination therapy. The secondary outcomes included individual components of the primary outcomes and mortality rates. Baseline measurements were defined as the most recent values recorded within three days before the initiation of medication.

Covariates

To control for confounding in the MIMIC-IV and ECG-ViEW II databases, we defined a comprehensive set of Propensity Score Covariates (PSCs): (1) demographic characteristics (age, sex, smoking status); (2) modified Charlson Comorbidity Index (CCI), excluding comorbidities regarded as separate risk factors; (3) risk factors for QT prolongation, such as history of myocardial infarction and other related heart diseases, chronic kidney disease, hypertension, and serum potassium abnormalities (< 3.5, > 5 mEq/L within 24 h of admission) (eTable 3) [22, 23]; and (4) application of QT prolongation high-risk drugs (including CredibleMeds Known-Risk (KR) or Possible-Risk (PR) drugs) (eTable 4) [23]; and (5) concomitant antithrombotic treatment variables, specifically anticoagulant use and antiplatelet pattern.

Statistical Analysis

To analyze the FAERS and CVAR databases, we first conducted a disproportionality analysis to identify potential pharmacovigilance signals by employing four established metrics: reporting odds ratio (ROR), proportional reporting ratio (PRR), Bayesian confidence propagation neural network (BCPNN), and multi-item gamma Poisson shrinker (MGPS) [24]. A positive finding was defined as a metric exceeding its specific threshold (eTables 5 and 6), representing a statistical association, rather than a confirmed causal effect. Recognizing the critical limitation that these standard measures cannot isolate the effects of drug-drug interactions (DDIs), we subsequently performed a dedicated secondary analysis. This analysis employed six published algorithms specifically designed for DDI detection: the Ω Shrinkage Measure Model, the Additive Model, the Multiplicative Model, the Combination Risk Ratio Model, the Chi-Square Statistics Model, and the Reporting Ratio Method [25, 26]. The objective of these models was to formally test whether the risk observed for the drug combination significantly exceeded the risk expected from the independent effect of each component. Detailed descriptions of these DDI algorithms are provided in eTables 7 and 8, respectively. The association between drug-event combinations (DEC) and DDI-positive signals was quantitatively analyzed, with the main parameter being the relative reporting ratio (RRR), with a higher RRR confirming a strong association [27]. The relevant parameters and algorithms are listed in eTable 9.

Regarding the MIMIC-IV and ECG-ViEW II databases, Kaplan-Meier curves and Cox proportional hazards models (with bidirectional stepwise variable selection) were used to estimate time-to-event risks, with results reported as hazard ratios (HRs) and 95% confidence intervals (CIs). Two analysis populations were used: (1) the original cohort (unmatched) and (2) a propensity score-matched (PSM) cohort, constructed by 1:2 nearest-neighbor matching with a caliper of 0.05 [28, 29]. Propensity scores were estimated via logistic regression with drug combinations as the dependent variable and PSC as the independent variable. Covariate balance was assessed using standardized mean differences (SMD), with SMD < 0.10 indicating adequate balance. Patients with missing key covariates were excluded from analysis.

To assess the risk of QVC events in patients receiving ceftriaxone in conjunction with a single PPI, those using multiple PPIs concurrently were excluded. Overlap Weighting (OW), PS-based causal inference, and generalized weighted estimation approaches have been applied for sample size adjustment [30]. Within the resulting weighted cohort, pairwise comparisons were conducted between treatment groups (e.g., lansoprazole vs. pantoprazole) using a previously outlined methodology. Statistical significance was defined as α < 0.05, with all tests 2-sided. Data were processed using R (version 4.4.3) and Excel (version 2021).

Sensitivity Analysis and Subgroup Analysis

Subgroup analyses of the FAERS and MIMIC-IV databases assessed the influence of demographic and clinical factors on the association between drug combinations and QVC events. In FAERS, ROR values were calculated for males vs. females and for patients younger than 18 years vs. those 18 years or older using 2 × 2 contingency tables, with Pearson’s chi-squared test determining significance (log₂ROR > 1, P < .05, for increased risk; log₂ROR < -1, P < .05, for decreased risk) [31]. For MIMIC-IV, a subgroup analysis was conducted in the PSM-adjusted population. Among PSC, race, and concomitant cardiovascular drug use, only those variables with at least 50 participants in the treatment group were included as subgroup variables. Additive models assessed the association between treatment assignment and subgroup variables, whereas multiplicative interaction models evaluated the interactions with continuous variables (age and CCI). PSC was included as a covariate in all models. Restricted cubic spline (RCS) curves generated using the interaction RCS package in R visualized the interaction effects on a continuous scale [32].

The robustness of our findings was confirmed by several sensitivity analyses. In FAERS, analysis is restricted to primary or secondary suspected drugs. In MIMIC-IV, we confirmed our results by limiting the cohort to patients on oral PPIs and employing alternative statistical methods, including 1:1 and 1:4 propensity score matching (PSM), stabilized inverse probability of treatment weighting (IPTW), and competing-risk models to account for all-cause mortality. Cumulative incidence function (CIF) curves and Fine-Gray models were applied to address the competing risk of all-cause mortality, with sub-distribution hazard ratios (SHRs) and 95% CIs reported. The effects of demographic and clinical factors on the link between drug combinations and QVC events were evaluated using the former, whereas the latter was used to verify the resulting robustness.

Results

Demographics

From Q1 2004 to Q3 2024, FAERS recorded 17,801,869 AE cases and 52,472,321 AE reports including 529,774 AEs linked to QVC events. Reports submitted by physicians comprised 37.80% of all the reports. The most frequently reported drug combination is ceftriaxone-pantoprazole. Within the CVAR group, 3.62% of the patients (n = 34) died. Additional summaries are provided in eTables 10 and 11.

In the MIMIC-IV database, the cohort included 5,594 patients: 437 in the lansoprazole group and 5,157 in the PPI group (Fig. 1; Table 1). After PSM, 437 and 872 patients were allocated to the two groups, respectively, and the matching and weighting diagnostics are shown in Fig. 1.

Fig. 1.

Flow chart of patient selection from the FAERS, CVAR, and MIMIC-IV databases. A MIMIC-IV flow chart; B FAERS flow chart; C CVAR flow chart. AE adverse event; AMD amiodarone; CRO ceftriaxone; CXM cefuroxime; DEMO patient demographic information file; ESO esomeprazole; LAN lansoprazole; LVX levofloxacin; OME omeprazole; PAN pantoprazole; PPIs proton pump inhibitors; REAC adverse reaction file (MedDRA Preferred Terms); RAB rabeprazole

Table 1.

MIMIC-IV baseline table before propensity score

Variable	Other PPIs N = 5157	Lansoprazole N = 437	P value	SMD
Age, Median (IQR)	69.61 (57.75, 81.63)	69.59 (58.54, 80.54)	0.848	0.012
Gender, n (%)			0.516	0.032
Female	2501 (48.50)	219 (50.11)
Male	2656 (51.50)	218 (49.89)
Smoker, n (%)			0.004	0.138
No	4004 (77.64)	313 (71.62)
Yes	1153 (22.36)	124 (28.38)
CCI, Median (IQR)	4.00 (3.00, 5.00)	4.00 (3.00, 6.00)	0.158	0.061
Hyperkalemia, n (%)			0.782	0.014
No	4930 (95.60)	419 (95.88)
Yes	227 (4.40)	18 (4.12)
Prior myocardial infarction, n (%)			0.883	0.007
No	4531 (87.86)	385 (88.10)
Yes	626 (12.14)	52 (11.90)
Heart failure, n (%)			0.403	0.041
No	3441 (66.72)	283 (64.76)
Yes	1716 (33.28)	154 (35.24)
Cardiomyopathy, n (%)			0.129	0.065
No	4981 (96.59)	416 (95.19)
Yes	176 (3.41)	21 (4.81)
Prior history of ventricular arrhythmia or cardiac arrest, n (%)			0.021	0.107
No	4792 (92.92)	393 (89.93)
Yes	365 (7.08)	44 (10.07)
Chronic kidney disease, n (%)			0.141	0.08
No	4488 (87.03)	391 (89.47)
Yes	669 (12.97)	46 (10.53)
Hypertension, n (%)			0.349	0.046
No	3127 (60.64)	255 (58.35)
Yes	2030 (39.36)	182 (41.65)
Known risk drugs, n (%)			0.112	0.076
No	1241 (24.06)	120 (27.46)
Yes	3916 (75.94)	317 (72.54)
Possible risk drugs, n (%)			0.668	0.021
No	2828 (54.84)	235 (53.78)
Yes	2329 (45.16)	202 (46.22)
Antiplatelet pattern, n (%)			1.000	0.001
No	3161 (61.30)	268 (61.33)
Yes	1996 (38.70)	169 (38.67)
Anticoagulant, n (%)			0.214	0.064
No	966 (18.73)	93 (21.28)
Yes	4191 (81.27)	344 (78.72)

Student’s t-test and Mann-Whitney U-test were used for normally and non-normally distributed continuous variables, respectively, whereas the chi-square test was applied for categorical data

CCI modified Charlson Comorbidity Index; PPI proton pump inhibitor; SMD standardized mean difference

Outcomes of FAERS and CVAR

The combination of ceftriaxone-lansoprazole yielded concordant QVC safety signals across all disproportionality and DDI algorithms in both databases, FAERS (ROR 4.97, PRR 4.78, EBGM 4.78, IC 2.26, and Ω025 0.54) and CVAR (ROR 5.58, PRR 5.41, EBGM 5.40, IC 2.43, and Ω025 0.18) (Tables 2 and 3). In addition, the levofloxacin-amiodarone pair was found across all six DDI algorithms for electrocardiogram QT prolongation (Ω025 0.29) (eTables 13 and 14). These results are consistent with those of previous studies and support the validity of the algorithms employed.

Table 2.

Signal strength of drug combinations associated with QVC events detected by at least one algorithm in FAERS and CVAR databases

Source	Drug	QVC cases	Non-QVC cases	ROR	95%CI for ROR	PRR	χ2	EBGM	95%CI for EBGM	IC	95%CI for IC
FAERS	OME	15,724	1,461,027	1.06	1.04 to 1.07	1.06	46.23	1.05	1.04 to 1.07	0.08	0.05 to 0.10
FAERS	LAN	6334	483,146	1.29	1.26 to 1.32	1.29	399.86	1.28	1.26 to 1.31	0.36	0.32 to 0.39
FAERS	CRO	2379	115,458	2.02	1.94 to 2.11	2.00	1203.69	2.00	1.93 to 2.07	1.00	0.94 to 1.06
FAERS	CXM	808	59,035	1.34	1.25 to 1.44	1.34	69.53	1.34	1.26 to 1.42	0.42	0.32 to 0.52
FAERS	CRO + PAN	287	13,942	2.02	1.80 to 2.27	2.00	144.52	2.00	1.81 to 2.20	1.00	0.82 to 1.17
FAERS	CRO + OME	164	8231	1.95	1.67 to 2.28	1.94	74.85	1.93	1.70 to 2.20	0.95	0.72 to 1.17
FAERS	CRO + LAN	155	3058	4.97	4.23 to 5.84	4.78	467.81	4.78	4.17 to 5.47	2.26	1.98 to 2.46
CVAR	CRO	133	7083	2.67	2.25 to 3.18	2.64	136.12	2.63	2.28 to 3.04	1.40	1.13 to 1.63
CVAR	CRO + LAN	67	1708	5.58	4.37 to 7.12	5.41	241.76	5.40	4.40 to 6.62	2.43	1.98 to 2.70
CVAR	CRO + RAB	4	110	5.16	1.90 to 14.00	5.02	12.95	5.02	2.18 to 11.56	2.33	0.15 to 2.79
CVAR	CXM + OME	2	36	7.89	1.90 to 32.76	7.52	11.39	7.52	2.29 to 24.77	2.91	-0.49 to 2.97

Bold values indicate the generation of positive signals by using the algorithm

CRO Ceftriaxone; CXM Cefuroxime; CVAR Canadian Vigilance Adverse Reaction; EBGM Empirical Bayesian Geometric Mean; ESO Esomeprazole; FAERS FDA Adverse Event Reporting System; IC Information Component; LAN Lansoprazole; OME Omeprazole; PAN Pantoprazole; PRR Proportional Reporting Ratio; QVC QT interval prolongation, ventricular arrhythmias, and cardiac arrest; RAB Rabeprazole; ROR Reporting Odds Ratio; χ² chi-square test value

Table 3.

Signal strength of drug-drug interactions for QVC events in FAERS and CVAR databases using multiple algorithms

Source	Drug	PTs	Ω 0.25	Additive model	Multiplicative model	CRR	χ	RRR
FAERS	CRO + LAN	QVC	0.54	0.03	7.49	2.38	9.36	0.02
FAERS	LVX + AMD	QVC	0.02	0.01	1.79	1.42	4.67	0.00
FAERS	CRO + PAN	QVC	-0.39	0.00	4.39	1.00	-0.05	-0.01
FAERS	CRO + OME	QVC	-0.60	0.00	3.68	0.97	-0.82	-0.01
CVAR	CRO + LAN	QVC	0.18	0.02	13.22	2.05	5.63	0.01
CVAR	CRO + RAB	QVC	-2.46	0.02	15.06	1.90	0.97	0.01
CVAR	CXM + OME	QVC	-3.37	0.02	1.71	3.56	1.26	0.01

Bold values indicate generation of positive signals in the algorithm

AMD amiodarone; CRO ceftriaxone; CXM cefuroxime; LAN lansoprazole; LVX levofloxacin; OME omeprazole; PAN pantoprazole; QVC QT interval prolongation, ventricular arrhythmias, and cardiac arrest; FAERS FDA adverse event reporting system; CVAR Canadian vigilance adverse reaction; PTs preferred terms; CRR combination risk ratio; Ω 0.25 Ω shrinkage model lower limit; χ chi-square statistic; RRR reporting ratio method; RAB rabeprazole

As shown in Fig. 2, we performed correlation analyses for DECs that produced a positive DDI signal in at least one algorithm within FAERS. A pronounced interaction was evident for ceftriaxone-lansoprazole: five adverse events exceeded the RRR_diff threshold of 0.75, with the strongest associations observed for long QT syndrome (9.41), electrocardiogram repolarization abnormality (7.96), and torsade de pointes (4.24). In contrast, ceftriaxone-omeprazole generated only two weak signals (long QT syndrome, 2.02; electrocardiogram QT prolonged, 0.78), whereas ceftriaxone-pantoprazole yielded only one signal (multiple organ dysfunction syndrome, 0.79). No positive signals were detected in the negative control group, whereas the positive control group produced six AEs with RRR_diff values above 0.75, most notably torsade de pointes (4.10).

Fig. 2.

Correlation analysis of Drug-Drug Interactions and target adverse events with positive signals. This figure illustrated the correlation between specific drug combinations and adverse events with significant drug-drug interaction signals. Red dots indicated statistically significant associations (RRR_diff > 0.75), whereas black dots represented drug-event combinations that did not reach statistical significance. The size of the dots reflected |RRR_diff|, indicating the relative strength of the association. AMD amiodarone; CRO ceftriaxone; CXM cefuroxime; ESO esomeprazole; LAN lansoprazole; LVX levofloxacin; OME omeprazole; PAN pantoprazole; QVC QT interval prolongation, ventricular arrhythmias, and cardiac arrest; RAB rabeprazole

Outcomes of MIMIC-IV

In the lansoprazole group, 33.87% of patients (148 cases) experienced QVC events, whereas in the other PPIs groups, 26.60% of patients (1,372 cases) experienced such events (eTable 15). This corresponded to an absolute risk increase of approximately 7.27 per 100 patients and a relative increase of approximately 27.33%. Multivariable Cox regression analysis showed that when compared with co-administration with other PPIs, the combined administration of ceftriaxone and lansoprazole was associated with a heightened risk of QVC event deterioration (HR = 1.30; 95% CI: 1.10–1.54; P=.002). After adjustment for PSM, the HR was 1.32 (95% CI: 1.08–1.62; P=.007) (Figs. 3 and 4).

Fig. 3.

Cumulative incidence of QVC events in the unmatched and PSM populations. (A) Kaplan–Meier curves in the unmatched population; (B) Kaplan–Meier curves in the PSM population; (C) CIF curves considering all-cause mortality as a competing risk in the unmatched population; (D) CIF curves considering all-cause mortality as a competing risk in the PSM population. The cumulative incidence of QVC events without considering competing risks was plotted using Kaplan–Meier curves, whereas the cumulative incidence of QVC events considering all-cause mortality as a competing risk was plotted using CIF curves. CIF cumulative incidence function; PPI proton-pump inhibitor; PSM propensity score matching; QVC QT interval prolongation, ventricular arrhythmias, and cardiac arrest

Fig. 4.

Cox regression and competing risk regression analysis results between PPIs groups and QVC events. The HR and SHR from the Multivariate, PSM, and IPTW models had been adjusted using PSC. HR hazard ratio; IPTW inverse probability of treatment weight; PPI proton-pump inhibitor; PSM propensity score matching; QVC QT interval prolongation, ventricular arrhythmias, and cardiac arrest; SHR subdistribution hazard ratio

In the single PPI subgroup, there were 2,665 individuals in the pantoprazole group, 2,112 in the omeprazole group, and 228 in the lansoprazole group (eTable 16). With OW, the risk of QVC events in the lansoprazole group exceeded that in the pantoprazole (HR = 1.29; 95% CI: 1.20–1.38; P<.001) and omeprazole groups (HR = 1.43; 95% CI: 1.33–1.53; P<.001). Furthermore, a decrease in the QVC event risk was noted in the omeprazole group, particularly in comparison with the pantoprazole group (HR = 0.90; 95% CI: 0.83–0.97; P=.008) (Figs. 5, eFigure2).

Fig. 5.

Cox regression and competing risk regression analyses of the association between PPI groups and QVC events among patients using a single PPI. (A) Univariate Cox regression analysis; (B) multivariate Cox regression analysis; (C) overlap-weighted Cox regression analysis; (D) univariate competing risk regression analysis; (E) multivariate competing risk regression analysis; (F) overlap-weighted competing risk regression analysis. The HRs and SHRs from the multivariate and overlap-weighting models were adjusted using PSC. HR hazard ratio; PPI proton-pump inhibitor; QVC QT interval prolongation, ventricular arrhythmias, and cardiac arrest; SHR subdistribution hazard ratio

Subgroup Analysis

In the FAERS database, sex- and age-stratified subgroup analyses identified three meaningful subgroup signals (Supplement 2, eFigure 3). First, within the ceftriaxone–lansoprazole combination group, QT prolongation-related events were reported more frequently in female patients than in male patients (38/1,383 vs. 18/1,540, P=.004). Second, multiple organ dysfunction syndrome was reported more frequently in female patients than in male patients (63/7,948 vs. 10/5,927, P<.001). Third, multiple organ dysfunction syndrome was reported more frequently in individuals aged ≥ 65 years than in those aged < 65 years (57/5,849 vs. 13/5,770, P<.001).

In the MIMIC-IV cohort, no significant interaction was found between drug combinations and any subgroup variable (interaction P>.05), as depicted in Supplement 2, eFigure4. A nonlinear interaction was found between age, CCI, and different drug combinations (Supplement 2, eFigure5).

Sensitivity Analysis

Regarding the FAERS database, the combination of ceftriaxone and omeprazole produced positive safety signals for electrocardiogram QT prolongation (Ω025 0.79) in all six interaction algorithms when the analysis was restricted to the primary suspect drug. The combination of ceftriaxone and lansoprazole showed positive signals in all six algorithms for ECG QT prolongation (Ω025 2.01), long QT syndrome (Ω025 0.82), and torsade de pointes (Ω025 1.33), and the corresponding correlations were significant (eTables 17 and 18; eFigure 6).

In the MIMIC-IV database, patients who received oral lansoprazole had a higher risk of QVC events than those treated with other proton pump inhibitors (PPIs) (P < .001) (eFigs. 7 and 8). The results from the 1:1 and 1:4 propensity score-matched (PSM) cohorts were consistent with those from the 1:2 PSM cohort (eTable 19, eFigures 9, and 10). Sensitivity analyses using IPTW and a competing risk model yielded similar estimates (Figs. 3 and 4, respectively).

In the ECG-ViEW II external validation cohort, 3,478 ceftriaxone–PPI episodes were included, comprising 89 episodes with ceftriaxone plus lansoprazole and 3,389 episodes with ceftriaxone plus other PPIs (eTable 20). The 28-day QVC event rate was higher for ceftriaxone plus lansoprazole than for ceftriaxone plus other PPIs (18/89, 20.2% vs. 345/3,389, 10.2%). In the main all-route analysis, ceftriaxone plus lansoprazole was associated with a higher QVC risk than ceftriaxone plus other PPIs in both the unadjusted Cox model (HR 1.94, 95% CI 1.21–3.12) and the adjusted Cox model (HR 2.64, 95% CI 1.58–4.42) (eFigure 13). Route-restricted and pairwise sensitivity analyses further supported a higher QVC risk for lansoprazole compared with pantoprazole or omeprazole, whereas pantoprazole and omeprazole did not differ significantly (eFigure 14).

Discussion

To the best of our knowledge, this is the first study to systematically evaluate QVC event signals associated with ceftriaxone co-therapy with individual PPIs by integrating two large-scale pharmacovigilance databases, a real-world ICU cohort, and an additional Asian ECG-linked external validation dataset. We first generated our hypothesis using the FAERS and CVAR databases, in which multiple algorithms consistently identified a risk signal for concomitant use of ceftriaxone and lansoprazole. This signal was the strongest among all ceftriaxone-PPI combinations ( FAERS ROR 4.97, Ω025 0.54). Guided by these pharmacovigilance signals, we validated this association in a cohort of 5,594 patients from the MIMIC-IV database. A comparison revealed that patients treated with ceftriaxone and lansoprazole had a significantly higher risk of QVC than those treated with ceftriaxone and other PPIs (aHR, 1.30; 95% CI: 1.10–1.54). The directionally consistent ECG-ViEW II findings extend the evidence beyond North American data sources and provide supportive validation in an Asian ECG-linked real-world dataset. These findings remained consistent across methodologically distinct stages, from broad risk signal detection to targeted clinical comparison, to minimize confounding bias. This not only provided more substantial evidence for the link between the concomitant use of ceftriaxone and lansoprazole and QVC events but also highlights the reliability of our multi-database research approach.

Our findings were corroborated by two large observational studies of the combination of ceftriaxone and lansoprazole. A propensity score-weighted Canadian cohort study [6] reported that the concomitant use of ceftriaxone and lansoprazole (vs. other PPIs) increased the risk of ventricular arrhythmia/cardiac arrest and in-hospital mortality (adjusted risk differences, 1.7% and 7.4%, respectively). Similarly, a Japanese claims database analysis using competing-risk models [7] found that ceftriaxone and lansoprazole increased the risk of ventricular arrhythmia/cardiorespiratory arrest (oral HR = 2.92, 95%CI: 1.99–4.29; intravenous HR = 4.57, 95%CI: 1.24–16.80). Collectively, despite the methodological differences, these studies converged on a consistent risk signal for this drug combination. Moreover, some studies have supported our findings [33]. Fan et al. further examined PPI-associated QT interval prolongation in 24,512 critically ill patients from MIMIC-III and reported a higher incidence of QT prolongation among PPI users than among histamine-2 receptor antagonist (H2RA) users or patients without acid-suppression therapy (8.5%, 3.3%, and 3.4%, respectively). After adjustment, PPIs remained associated with QT prolongation, and pantoprazole and lansoprazole showed higher risks than omeprazole [33].

We further stratified our analysis according to the type of PPI administered. Among the evaluated PPIs, omeprazole was associated with the lowest risk of QVC. This finding is consistent with a large analysis of the MIMIC-III database by Fan et al. [33], who reported that after multivariate adjustment and propensity score analysis, both pantoprazole (OR = 2.14; 95% CI: 1.52–3.03) and lansoprazole (OR = 1.80; 95% CI: 1.18–2.76) groups were associated with higher odds of QT interval prolongation than omeprazole.

The proposed mechanisms for PPI-induced cardiovascular events include disturbances in electrolyte homeostasis (sodium, potassium, and magnesium), endothelial dysfunction, and the blockade of human ether-a-go-go-related gene (hERG) potassium channels [5, 14, 34]. Lorberbaum et al. [14] showed that co-administration of ceftriaxone and lansoprazole significantly suppressed hERG currents, delayed ventricular repolarization, and prolonged QT interval. Lazzerini et al. [35] indicated that PPIs bind to different sites on hERG, producing variable degrees of channel blockade. Among the PPIs, pantoprazole showed the greatest inhibition (approximately 35–85%), followed by lansoprazole (20–50%), and omeprazole (10–30%). In addition to the PPI-mediated hERG effects, the physicochemical profile of ceftriaxone poses specific risks. Clinical cases [36] suggest that ceftriaxone may cause hypernatremia by disrupting sodium handling. Severe hypernatremia can precipitate fatal arrhythmias by altering the myocardial membrane potentials. In neonates, the precipitation of ceftriaxone with calcium has been linked to cardiopulmonary arrest due to microvascular embolism, a mechanism distinct from Kounis syndrome or hERG blockade [37]. Although this crystalline embolic risk is mechanistically distinct from PPI-mediated hERG inhibition, the co-administration of certain PPIs (e.g., lansoprazole) may produce synergistic toxicity.

Moreover, case reports [38] corroborate this mechanism: one patient developed QT prolongation after ceftriaxone plus lansoprazole, and the QT interval normalized after switching to pantoprazole, which is in agreement with the finding that the risk of QVC events was higher with ceftriaxone plus lansoprazole than with ceftriaxone plus pantoprazole. In contrast, an in vitro study of isolated rabbit hearts [39] did not observe QT prolongation with ceftriaxone plus lansoprazole, which may reflect the sex distribution or other unmeasured mechanisms.

Subgroup analysis indicated that older age and female sex were associated with a higher risk of QVC events, which is consistent with previous studies [7, 40]. Women with longer QT intervals and reduced repolarization reserves appear more susceptible to QT prolongation [41]. However, other reports [34] have identified the male sex as a risk factor, implying that the effects of sex on QT prolongation may depend on additional modifiers.

This study has several limitations that warrant consideration. First, the pharmacovigilance analyses relied on spontaneous reporting systems, which are inherently subject to reporting bias, underreporting, incomplete clinical information, and the absence of reliable exposure denominators. Second, as a retrospective observational study, causal relationships cannot be established, even after multivariable adjustment, propensity score matching, inverse probability of treatment weighting, and competing-risk analyses. Third, dose–response relationships could not be evaluated because of incomplete and inconsistent dosing information and limited exposure granularity across databases.

These findings suggest that the potential risk of QVC events associated with ceftriaxone in combination with PPIs, particularly lansoprazole, merits further clinical attention. Future work could (1) conduct prospective multicenter studies to validate the comparative effects of different PPIs with ceftriaxone on QVC events and examine pragmatic treatment parameters and (2) integrate in vivo and in vitro experiments to define the electrophysiological effects of ceftriaxone plus PPI regimens on cardiomyocytes and clarify the underlying molecular mechanisms.

Conclusion

This study provides clinically relevant evidence regarding antimicrobial practices. The co-prescription of ceftriaxone with lansoprazole was associated with a higher risk of QVC events than ceftriaxone with other proton pump inhibitors, whereas omeprazole showed the lowest observed risk among the evaluated agents. Although observational, these agent-level differences suggest a potential interaction that should inform antimicrobial stewardship when a proton-pump inhibitor is administered during ceftriaxone therapy. In such cases, clinicians should consider PPI selection in the treatment plan and implement targeted safety measures, including baseline and follow-up electrocardiograms and correction of potassium and magnesium abnormalities, particularly in patients with preexisting cardiac risk.

Supplementary Information

Below is the link to the electronic supplementary material.

Author Contributions

All authors contributed to the conception, design and original draft preparation of the study. Data curation: Mingnuo Zhao, Zirui Kong, Haobin Shen, Shuang Wang; Methodology: Dayu Chen, Yechao Chen, and Qiaoling Gu; Formal analysis and Investigation: Yechao Chen, Qiaoling Gu, and Mingnuo Zhao; Software: Aijin Zhao; Validation: Aijin Zhao, and Dayu Chen; Visualization: Mingnuo Zhao, and Aijin Zhao; Writing–review and editing: Dayu Chen, Yechao Chen, Qiaoling Gu, Peipei Liu, and Han Xie; Funding acquisition: Dayu Chen, Peipei Liu, and Haixia Zhang; Resources: Li Li, Yanan Zhang, and Aijin Zhao; Project Administration and Supervision: Peipei Liu, Haixia Zhang, and Dayu Chen. All the authors have read and agreed to the published version of the manuscript.

Funding

This work was supported by grants from the Jiangsu Pharmaceutical Association Research Foundation of Pharmacy [202495070] and Nanjing Municipal Health Bureau Medical Science and Technology Development Foundation [ZKX23020]. The funding sources had no relation to the study design, collection, analysis and interpretation of data, writing of the report and decision to submit the article for publication.

Data Availability

The datasets supporting the conclusions of this article are available in the FAERS repository (https://fis.fda.gov/extensions/FPD-QDE-FAERS/FPD-QDE-FAERS.html), the CVAR repository (https://www.canada.ca/en/health-canada/services/drugs-health-products/medeffect-canada/adverse-reaction-database.html), the MIMIC-IV repository (https://physionet.org/content/mimiciv/2.2/), and the ECG-ViEW II database (http://ecgview.org/default.asp). Access to the MIMIC-IV database requires user credentialing, and access was granted to the authors after approval was obtained (Certification number: 13465720). The analysis code has been made publicly available at https://github.com/family3253/mimic-ecg. Individual-level clinical data and restricted intermediate datasets are not redistributed by the authors because of database-specific access requirements.

Declarations

Conflict of interest

The authors declare no competing interests.

Ethical Approval

All databases employed rigorous anonymization and de-identification protocols for patient records to ensure compliance with privacy protection standards. Accordingly, this study was exempt from institutional ethics committee review and informed consent requirements under the applicable regulations. This observational real-world pharmacovigilance cohort study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guidelines and the Reporting of a Disproportionality Analysis for Drug Safety Signal Detection Using Individual Case Safety Reports in PharmacoVigilance (READUS-PV).