Piperacillin/tazobactam versus carbapenems in patients with severe bacterial infections: A systematic review with meta‐analysis

Piperacillin/tazobactam or meropenem are often used to treat patients with severe bacterial infections. We aimed to compare the desirable and undesirable effects of empirical and/or definitive piperacillin/tazobactam versus carbapenems in patients with severe bacterial infections.


| BACKGROUND
Sepsis is one of the leading causes of death worldwide with 48.9 million cases and 11 million sepsis-related deaths each year. 1 Empirical antimicrobial therapy is recommended in the Surviving Sepsis Campaign guidelines for patients with sepsis or septic shock. 2 Broad-spectrum antimicrobials are recommended to ensure coverage of the likely pathogens. 2 Often, β-lactam/β-lactamase inhibitors (e.g., piperacillin/tazobactam) or carbapenems (e.g., meropenem) are used. 3 Antimicrobial resistance is a major threat to global health according to the World Health Organization, particularly the emergence of carbapenem-resistant Enterobacterales. 4 Piperacillin/tazobactam may be used as carbapenem-sparing agent to limit the spread of multidrugresistant bacteria and superinfections. 5 Piperacillin/tazobactam has been compared with meropenem in patients with bacteraemia with ceftriaxone-nonsusceptible Escherichia coli or Klebsiella species (spp.) in the MERINO non-inferiority trial, where the 30-day mortality was higher in patients receiving piperacillin/ tazobactam (12.3%) than meropenem (3.7%). 6 However, the mortality difference was less pronounced after excluding isolates non-susceptible to piperacillin/tazobactam. 7 Previous systematic reviews comparing empirical or definitive treatment with β-lactam/β-lactamase inhibitors and carbapenems have demonstrated similar effects on mortality suggesting that piperacillin/tazobactam and other β-lactam/β-lactamase inhibitors may be safe alternatives to carbapenems. [8][9][10][11][12][13][14][15] It is unclear if piperacillin/tazobactam is inferior to carbapenems for the empirical or definitive treatment of other severe bacterial infections. This systematic review compared the desirable and undesirable effects of empirical and/or definitive piperacillin/tazobactam versus carbapenems in patients with severe bacterial infections. We hypothesised that the use of piperacillin/tazobactam results in less favourable outcomes than carbapenems in this patient group.

| METHODS
The protocol was registered in the International Prospective Register of Systematic Reviews (PROSPERO, CRD42019123052) and published in journal before data extraction began. 16 Minor protocol deviations are presented with reasonings in the Supplement. This systematic review was prepared according to the recommendations from the Cochrane Collaboration 17 and reported according to the Preferred Reporting Items for Systematic Review and Meta-Analysis (PRISMA) statement (Supplement), 18

| Search strategy
The searches are presented in the Supplement. We systematically searched MEDLINE (PubMed interface, 1966 onwards), Embase (OVID interface, 1974 onwards), Cochrane Central Register of Controlled Trials (CENTRAL, Wiley interface, current issue), and Epistemonikos from inception to July 4, 2022 (Supplement). We also searched ClinicalTrials.gov, EU Clinical Trials Register, and ISRCTN Registry from inception to September 5, 2022. One trial 19 was identified through a reference list. 20

| Eligibility criteria
We included randomised clinical trials (RCTs) comparing piperacillin/ tazobactam with carbapenems in adult patients (as defined by the original trial) with suspected or confirmed severe bacterial infection (i.e., any bacterial infection requiring hospitalisation) regardless of publication status and language. We excluded quasi-randomised and cross-over trials.
We assessed both empirical and definitive treatment with piperacillin/tazobactam versus carbapenems in any timing, dosing, and duration. Empirical treatment was defined as any antibiotic treatment given before microbiological results were available due to clinical suspicion of infection, whereas definitive therapy was defined as any antibiotic treatment given according to microbiological results, including susceptibility. 21 We allowed for additional antimicrobial agents, if these were identical in both groups except for combinations of carbapenems and β-lactamase inhibitors (i.e., imipenem/cilastatin/relebactam, meropenem/vaborbactam), which we allowed in the carbapenem group only.
We assessed outcomes considered critical or important for decision-making according to Grading of Recommendations, Assessment, Development and Evaluation (GRADE) methodology. 22 The primary outcome was all-cause short-term mortality (≤ 90 days, including in-ICU and in-hospital mortality); the secondary outcomes were allcause long-term mortality (>90 days), adverse events (as defined in the original trials), quality of life (as defined in the original trials), days alive without or duration of mechanical ventilation, days alive without or duration of renal replacement therapy, days alive without or duration of vasopressors/inotropes, secondary infections (i.e., new infectious origin or clinical deterioration of the original infection with a different organism compared with the original infection), selection of fungi (e.g., Candida spp.) or bacteria with resistance against trial medication (e.g., Klebsiella spp., Eschericia spp.) or natural resistance (e.g., Enterococcus faecium, Clostridium difficile), and days alive out of hospital or hospital length of stay (LOS). We assessed outcomes at longest follow up if outcome data were available at multiple time points.

| Data extraction
Two authors (Marie Warrer Munch and Anders Granholm) independently extracted data using a standardised data extraction form on trial setting, population, interventions, and outcomes. Any disagreements were solved by discussion and involving a third author (Morten Hylander Møller) if necessary.
We aggregated outcome data if data for multiple components of one outcome were reported in several ways (i.e., different resistant bacteria), and the denominators were the same. We reported the outcome with the highest event count if an outcome was reported in more than one way with different denominators.

| Certainty of evidence
Two authors (Marie Warrer Munch and Anders Granholm) independently used the GRADE methodology to rate the certainty of evidence. 22,23 We downgraded the certainty of evidence if there were risk of bias, inconsistency, indirectness, imprecision, or publication bias, 22 and interpreted our results according to the GRADE recommendations. 24

| Statistical analysis
The meta-analyses were conducted in R version 4.2.0 (R Core Team, R Foundation for Statistical Computing, Vienna, Austria) with the meta (v. 5.5.0) and metafor (v. 3.4.0) packages.
For the primary outcome, a p-value of 0.05 was considered statistically significant. For the secondary outcomes, we adjusted the pvalue according to the number of available outcomes to control the family-wise error rate (i.e., 0.05 divided by the value halfway between one and the number of available outcomes). 25 As four secondary outcomes were available, a p-value of 0.02 was considered statistically significant for the secondary outcomes.
For each trial, we reported relative risks (RRs) with corresponding 95% CIs for dichotomous outcomes and mean differences with 95% CIs for continuous outcomes based on the raw event counts or means and standard deviations reported for each group. We included data from the intention-to-treat population where available.
For the meta-analyses, we reported RRs with CIs corresponding to the p-value thresholds defined above (i.e., 95% CI and 98% CIs for the primary and secondary outcomes, respectively). We conducted fixed effect models and random effects models and considered the most conservative analysis (i.e., the model with the highest p-value) as the primary. Of note, meta-analyses could only be conducted for the binary outcomes.
We used funnel plots to assess the risk of small trial bias 26 if 10 or more trials were included in the meta-analysis. To assess heterogeneity, we calculated both inconsistency (I 2 ) and diversity (D 2 ) statistics. 27

| Trial sequential analysis
We conducted trial sequential analyses (TSAs) to calculate the required information size for the meta-analyses to be able to conclusively confirm or reject an a priori 15% RR reduction. 16,28 We applied trial sequential monitoring boundaries using an alpha corresponding to the p-value thresholds defined above, beta 10% (power 90%), control event proportions as per the control arm of the included trials, and the estimated diversity (D 2 ) in the random effects models TSAs.

| Subgroup and sensitivity analyses
We planned to conduct subgroup analyses for the primary outcome of risk of bias, ICU setting, immunosuppression, primary sites of infection, treatment strategy, and type of carbapenem (Supplement). 16 The hypothesised subgroup effects are listed in the Supplement and the protocol. 16 As we were unable to conduct the subgroup analysis of risk of bias (due to lack of overall low risk of bias trials), we post hoc added a subgroup analysis of blinding (i.e., double-blind versus open-label trials). Moreover, as the typical dosing regimen for piperacillin/ tazobactam has changed from three times daily to four times daily, 29 we also post hoc added a subgroup analysis of lower (≤13.5 g per day corresponding to 4.5 g three times daily or less) versus higher daily doses (>13.5 g per day corresponding to 4.5 g four times daily) of piperacillin/tazobactam. We used the Chi-square test to assess the statistical heterogeneity across subgroups (test of interaction). A p-value of 0.10 was considered statistically significant.
To assess the change in treatment effect over time, we post hoc added a meta-regression of the primary outcome and the median time point (year) of the inclusion period for each trial.
We also conducted sensitivity analyses applying empirical continuity correction in the zero event trials. 30 To account for missing data, we conducted best/worst and worst/best sensitivity analyses ( Figure S2) for the primary outcome. 31

| Trial characteristics
Trial characteristics are presented in Table 1.

Included
Reports sought for retrieval (n = 1) F I G U R E 1 PRISMA 2020 flow diagram. a The search was performed three times with the latest search performed on July 4, 2022, in databases and on September 5, 2022, in trial registers. b One trial 19 was identified by handsearching the reference list of a previous systematic review. 20 6,48,50,51,57 and two both empirical and definitive treatment. 41,49 Two trials randomised episodes of febrile neutropenia instead of individual patients; 45,59 these were only included in the qualitative synthesis.

| Risk of bias
All trials were assessed as having overall high risk of bias ( Figure 2; Table S2, Figure S1).
No subgroup analyses demonstrated firm heterogeneity in the intervention effect of piperacillin/tazobactam versus carbapenems (Table 3; Figure S5). We were unable to conduct the subgroup analyses of risk of bias and ICU setting as there were no trials with an overall low risk of bias and no trials conducted in ICU patients only.
We found no conclusive evidence of a change in the treatment effect over time (RR change per year 1.01, 95% CI: 0.98 to 1.03, p = .47, R 2 = 0%, Figure S9).
Results from the best/worst and worst/best case sensitivity analyses were consistent with the main findings ( Figure S2).

| Secondary outcomes
No trials reported data for all-cause long-term mortality (>90 days), quality of life, and days alive without/duration of life support.  Table S2.

| Adverse events
F I G U R E 3 (A,B) All-cause short-term mortality (≤ 90 days). CI, confidence interval; pip/tazo, piperacillin/tazobactam. (A) Forest plot of all-cause short-term mortality (≤ 90 days) in 23 trials in 7125 patients. All trials were characterised as having an overall high risk of bias. The grey square represents the risk ratio, the size of the grey square represents the weight of the trial in the meta-analysis, and the horizontal bar represents the 95% confidence interval. (B) TSA of all-cause short-term mortality (≤ 90 days) in 23 trials in 7125 patients. All trials were characterised as having an overall high risk of bias. The analysis was computed using a two-sided alpha of 5%, beta of 10%, a priori risk reduction of 15%, and a control event proportion of 5.7%. The cumulative Z-score did not cross the conventional monitoring boundary or trial sequential boundary for benefit, harm, or futility, and the TSA is, thus, inconclusive. Pooled risk ratio 1.16 (random effects model) with TSA-adjusted confidence interval 0.82 to 1.65 (p = .09, D 2 = 0%). The required information size was not reached as only 25% of the required number of patients have been obtained.
T A B L E 2 GRADE evidence profile. No trials with a low risk of bias. c TSA was inconclusive as none of the trial sequential boundaries for benefit, harm or futility were crossed. The required information size was not reached as only 25% of the required number of patients have been obtained. The TSA-adjusted confidence interval for the absolute effect ranged from 10 fewer to 37 more. d TSA was conclusive as the trial sequential boundary for futility was crossed with 67% of the required number of patients obtained. The TSA-adjusted confidence interval for the absolute effect ranged from 41 fewer to 37 more. e TSA was inconclusive as none of the trial sequential boundaries for benefit, harm or futility were crossed. The required information size was not reached as only 5% of the required number of patients have been obtained. The TSA-adjusted confidence interval for the absolute effect ranged from 32 fewer to 160 more. f TSA could not be conducted as less than 5% of the required information size have been obtained. g We were unable to pool data in a meta-analysis and TSA as only one trial provided data for the outcome.
TSA demonstrated futility ( Figure S10). The certainty of evidence was moderate (downgraded for risk of bias, Table 2).
The sensitivity analysis with empirical continuity correction were in accordance with the primary meta-analysis ( Figure S12).
We were unable to conduct TSA as less than 5% of the required information size had been obtained to confirm or reject a 15% RR reduction ( Figure S13). The certainty of evidence was very low (downgraded for risk of bias and very serious imprecision, Table 2).
The sensitivity analysis with empirical continuity correction were in accordance with the primary meta-analysis ( Figure S14).

| Relation to current evidence
Our results for all-cause short-term mortality are in line with the MERINO trial, 6,7 but the results are uncertain, and benefits of piperacillin/tazobactam cannot be excluded.

| Strengths and limitations
The strengths of our review include adherence to the published protocol, 16 up-to-date searches, contact to authors in case of missing data, duplicate data extraction and risk of bias assessment, TSA, 28 and adherence to the PRISMA statement, 17 Cochrane Handbook, 18 and GRADE methodology. 22,24 However, the evidence exclusively comes from overall high risk of bias trials. Despite identifying 31 completed trials, the TSAs for most outcomes showed that the evidence was insufficient to draw firm conclusions as we could not confirm or reject an a priori 15% RR reduction or increase. Moreover, most of the secondary outcomes were not reported. As these are all ranked as critical or important for decisionmaking, the association between piperacillin/tazobactam versus carbapenems and patient-important outcomes has not been fully addressed yet.
The included trials were heterogenous in terms of population and interventions. All were done in patients with severe infections, but no trials were exclusively done in ICU patients, and the overall short-term mortality across trials was low (only 6.3%). The effects of piperacillin/ tazobactam as compared with carbapenems are likely somewhat different in critically ill patients, highlighting the need for high-quality trials particularly in ICU patients.
In many trials, 19,[32][33][34]37,42,44,47,49,50,53,55,58,60 piperacillin/ tazobactam was only administered two or three times daily which may not be sufficient, particularly not in critically ill patients. 67 However, a subgroup analysis did not demonstrate effect modification by the daily dose of piperacillin/tazobactam. strains non-susceptible to piperacillin/tazobactam. 6,7 This may explain some of the observed difference in mortality between the subgroups.
Finally, as the protocol was published in 2019, 16 and the data extraction and risk of bias assessment commenced in the Spring of 2020, we used the previous version of the Cochrane risk of bias tool. 18

| CONCLUSIONS
In this systematic review with meta-analysis and trial sequential analysis, we found that piperacillin/tazobactam as compared with carbapenems may be associated with less favourable outcomes in patients with severe bacterial infections, but both the quantity and certainty of the evidence were very low or low for most outcomes. The findings highlight the uncertainties about the effects of piperacillin/ tazobactam as compared with carbapenems on patient-important outcomes, including mortality and antimicrobial resistance. Our results stress the need for future high-quality trials addressing this important topic to ensure appropriate use of these broad-spectrum agents. Until more certain evidence emerges, it is reasonable to consider both piperacillin/tazobactam and carbapenems as effective treatment options for severe bacterial infections as we must consider the potential increase in antimicrobial resistance with increased consumption of broader-spectrum antibiotics.  had no influence on the study design; the collection, analysis, and interpretation of data; or the writing of the manuscript.