This systematic review and meta-analysis was performed in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses guidelines [12]. The study protocol was registered with the PROSPERO International Prospective Register of Systematic Reviews (https://www.crd.york.ac.uk/prospero/display_record.php?Recor-dID=376375). The following terms were used to search the available literature: [“antimicrobials” OR “antibiotics” OR “antibacterials”] AND [“de-escalation” OR “discontinuation” OR “narrowing” OR “step-down”]) AND [“hospital-acquired” OR “ventilator-associated” OR “healthcare-associated” OR “nosocomial”] AND [“pneumonia” OR “lower respiratory tract infection”]. Three electronic databases (PubMed, Embase, and the Cochrane Central Register of Controlled Trials) were searched for relevant articles published before January 2023. The reference section lists all research papers investigated, and appropriate reviews were examined manually to identify potentially relevant articles. As the present study is a systematic review of previous published articles, neither informed consent nor ethics approval was required.

This meta-analysis included all studies that met the following criteria: (1) the subjects were patients with culture-negative pneumonia; (2) an examination of the clinical outcomes of ADE was included; (3) in-hospital mortality outcomes were included; (4) relative risk ratios (RRs) and 95% confidence intervals (CIs) were reported or were able to be calculated using information in the article; and (5) the study was published in a peer-reviewed English-language journal. Review journals, case reports or series, commentaries, and extension or post-hoc articles were excluded.

The ADE strategy involves discontinuing broad spectrum antibiotics or narrowing the regimen based upon negative culture results following reassessment of a patient’s status after initiation of therapy [3]. Appropriate MDR-directed antimicrobial therapy was defined as a combination of at least one anti-pseudomonal beta-lactam antibiotic and at least one active anti-MRSA agent.

The primary outcome for this study was in-hospital mortality at the study index date. We also analyzed the duration of the initial antimicrobial therapy, total duration of antimicrobial therapy, and length of hospitalization.

Two reviewers independently screened the titles and abstracts of the studies retrieved from the search to identify relevant papers. We obtained full-text articles and extracted data according to the predefined inclusion criteria. The following variables were extracted: the first author, year of publication, study sites, number of patients, age, sex, initial intensive care unit (ICU) admission rate, type of pneumonia, ADE definition, ADE rate, patient mortality, duration of initial and total antimicrobial therapy, and primary outcomes.

As recommended by the Cochrane Collaboration, the methodological quality of included studies was evaluated using the in the Risk of Bias In Non-randomized Studies of Interventions (ROBINS-I) tool [13]. The ROBINS-I uses seven criteria to assess the potential risk of bias in each study including confounding, selection bias, classification of interventions, deviation from interventions, missing data, measurement of outcome, and selection of reported result. The risk of bias was graded as low, moderate, serious, critical, or no information [13]. The overall risk of bias was considered as serious or critical if the study achieved serious or critical ratings at least in one domain. Studies that obtained low in all domains were regarded as having a low risk of bias, and articles without any serious and critical ratings were regarded as having a moderate risk of bias [13].

The funnel plot was drawn to evaluate inherent publication bias and the Egger’s test was used to identify the evidence of bias. Any disagreements regarding study search, data extraction, and quality assessment were discussed and resolved by the authors.

We extracted the RRs and associated 95% CIs for the ADE strategy to evaluate clinical outcomes for dichotomous data compared with non-ADE treatments. For continuous data, the weighted mean difference (MD) with 95% CIs was extracted. Between-study statistical heterogeneity was evaluated using I² statistics on a scale of 0–100% (low heterogeneity for I² < 49%, moderate I² = 50–74% and high for I² > 75%, respectively) [14]. A random-effects model was used for moderate or high heterogeneity. A p value < 0.05 was considered statistically significant. Statistical analyses were performed with Stata statistical software (Version 14.2; Stata Corp LP, College Station, TX, USA) and Review Manager (Version 5.3; Nordic Cochrane Centre, The Cochrane Collaboration, Copenhagen, Denmark).

Figure 1 presents the results of the literature search. A total of 1,389 published articles were initially identified from three databases; 556 articles from PubMed, 137 articles from Embase, and 696 articles from the Cochrane Central Register of Controlled Trials. After removing duplicate articles, 1,048 potentially eligible articles were screened. A review of titles and abstracts resulted in the removal of 1,012 search records, and the remaining 36 articles were assessed by reading the full text. Thirty articles were excluded for the reasons presented in Figure 1, leaving six studies for our analysis [15–20].

Table 1 summarizes the features of the selected studies. Our systematic review and meta-analysis included 11,933 patients, of whom 1,152 received the ADE strategy and 10,781 did not receive ADE. The type of pneumonia was community onset pneumonia in 3 studies [16,18,20] and hospital-acquired pneumonia (HAP) in 3 studies [15,17,19], respectively. All studies were published between 2010 and 2021. A quality assessment of the included studies is reported in the Supplementary Table. Overall, the methodological quality evaluated by the ROBINS-I appeared to be acceptable.

Pooled estimates of strategy efficacy with respect to reduced in-hospital mortality rates by using ADE were weighed and combined with a generic inverse variance and random-effects model. Overall, the RR for in-hospital mortality indicated that ADE had a favorable effect compared with non-ADE (RR = 0.60, 95% CI = 0.38 to 0.93; Fig. 2), with a 40% absolute risk reduction. The heterogeneity was high (I² = 66%). The results of Egger’s test in the included studies indicated no significant publication bias (p = 0.335), although a visual inspection of the plot suggested asymmetry (Supplementary Fig.).

Because a substantial degree of heterogeneity existed among the included studies, we conducted subgroup and meta-regression analyses to explore heterogeneity according to strategy used in response to culture-negative pneumonia. Table 2 provides details of the subgroup and meta-regression analysis regarding the number of subjects (< 100 and ≥ 100), age (< 60 and ≥ 60 yr), ICU-acquired pneumonia, HAP or ventilator-associated pneumonia (VAP), and the type of ADE (discontinuation and narrowing spectrums). In the meta-regression analysis, probable sources of between-study heterogeneity for included studies were not found.

Four studies reported data on the duration of antibiotic therapy [15,16,18,19]. The ADE strategy resulted in shorter durations of total and initial antibiotic administration compared with non-ADE (MD = −3.88 days, 95% CI = −6.21 to −1.54, p < 0.01, I² = 98% and MD = −3.01 days, 95% CI = −6.93 to −0.91, p < 0.01, I² = 99%, respectively) (Fig. 3A, B). Data regarding the total lengths of hospital stay, which were available for three trials, were analyzed using the generic inverse variation method [15,16,20]. The duration in total length of hospital stays was significantly shorter with ADE compared with non-ADE (MD = −3.85 days, 95% CI = −7.52 to −0.18, p = 0.05, I² = 66%) (Fig. 3C). The mean ICU length of stay was similar between the ADE and non-ADE groups (MD = −0.93 days, 95% CI = −5.33 to 3.47, p = 0.09, I² = 64%) (Fig. 3D) [15,18].