Between January 2015 and December 2018, we retrospectively conducted a 4-year multicenter case-control study at the emergency departments (EDs) of three hospitals in southern Taiwan, namely a university-affiliated medical center (1,300 beds) and two teaching hospitals (one with 460 beds and one with 380 beds). The aimed patient was a naïve adult (i.e., without any antimicrobial and medical therapy prior to ED arrival) with community-onset bacteremia in the ED. According to the cut-off age suggested in the previously established study (10), the eligible patients were grouped as the following: middle aged (45–64 years), old (65–74 years), and very old (≥75 years). This study was approved by the Institutional Review Board of study hospitals, and the requirement of obtaining informed consent was waived. Clinical information was obtained and reported according to the Strengthening the Reporting of Observational Studies in Epidemiology (11).

From a database search, we identified adult patients who had undergone blood culture sampling at the ED. Among patients who had multiple episodes of bloodstream infections during the study period, we examined only the first episode. Of patients aged ≥45 years and experienced the bacterial growth of blood cultures, we excluded those with contaminated blood cultures, mycobacteremia, hospital-onset bacteremia, bacteremia diagnosed or treated before ED arrival, an uncertain mortality date, and incomplete chart records. Consequently, the study consisted of naïve patients aged ≥45 years experienced community-onset bacteremia was established.

Using a predetermined form, clinical data were jointly collected by a board-certified ED physician and an infectious disease (ID) physician who were well-trained by the IRB course and blinded to the aim and hypothesis of this study. Any discrepancy in data capturing was resolved through the discussion between two abstractors in periodic meetings. The captured ED information included patient demographics, vital signs, laboratory data, comorbidity types and severity, severity of bacteremia [Pitt bacteremia score (PBS)], timing and types of antimicrobial administration, and the length of stay. Further data regarding the length of hospitalization, durations and types of antimicrobial agents administered, imaging findings, timing and types of surgical and radiological interventions, causative microorganisms, bacteremia sources, and 30-day mortality were collected. The primary study endpoint was crude 30-day mortality after ED arrival (i.e., bacteremia onset).

Blood cultures were incubated in a Bactec 9240 instrument (Becton Dickinson Diagnostic Systems, Sparks, MD, United States). The causative microorganism was prospectively stored after identification by the Gram-Negative-Identification Card of the Vitek system (bioMérieux, Lyon, France) or the matrix-assisted laser desorption ionization time-of-flight mass spectrometry (MALDI-TOF MS) and susceptibility testing, which was performed using the disk diffusion method for aerobes and the agar dilution method for anaerobes, in accordance with the contemporary guideline of the Clinical and Laboratory Standards Institute (CLSI) (12). The susceptibility of Gram-negative aerobes to the following was tested: ampicillin/sulbactam, piperacillin/tazobactam, levofloxacin, moxifloxacin, cefazolin, cefuroxime, cefotaxime, ceftazidime, cefepime, ertapenem, and imipenem. For streptococci, staphylococci, and enterococci, the tested antibiotics were penicillin, cefoxitin, and ampicillin, respectively. Furthermore, the susceptibility of anaerobes to ampicillin/sulbactam, piperacillin/tazobactam, metronidazole, and moxifloxacin was tested. To determine the AAT timing for each patient, provided that empirical antibiotics administered were not included in the initial susceptibility panel, the susceptibility to the indicated antibiotic was tested.

Community-onset bacteremia refers to the bacteremic onset within the community (1, 3). Contaminated blood cultures were defined as those exhibiting the growth of potential contaminating pathogens, such as Gram-positive bacilli, Micrococcus species, Bacillus species, and Propionibacterium species (13). According to the previous suggestion (3), antimicrobial therapy was appropriate when the following criteria were totally fulfilled: (i) all bacteremia-causing pathogens in one episode were in vitro susceptible to the antimicrobials administered according to the contemporary CLSI criteria (12) and (ii) the route and dosage of antimicrobials was administered as recommended in the Sanford Guide 2021 (14). The period of AAT delays was defined as the gap between the onset of bacteremia (ED arrival) and the first dose of AAT administration (3).

The severity of bacteremia was determined according to the PBS, which was calculated during the first 24 h after ED arrival, and patients with a PBS of ≥4 were categorized as critically ill (3). Comorbidities were defined as described previously (15). Preexisting neurological diseases majorly included the Alzheimer’s disease; diseases resulting in dementia, multiple sclerosis, and demyelinating diseases; and stroke and cerebrovascular events (16). Hemato-oncological comorbidities consisted of hematological malignancies and solid tumors under active treatment. The severity of comorbidities was assessed based on the classification of McCabe and Jackson (rapidly fatal, ultimately fatal, and non-fatal) (17). Crude mortality was defined as death from all causes.

Consistent with general concepts (18), a patient was assigned to one of the bacteremia sources based on clinical diagnoses and/or isolation of pathogens. Based on the recommendation of the Surviving Sepsis Campaign (SSC) (13), complicated bacteremia was equated with whether that the bacteremia source is amenable to undergo source control, and the adequateness of source control for complicated bacteremia was determined by the board-certified ID physicians.

Statistical analyses were performed using the Statistical Package for Social Science for Windows (version 23.0; Chicago, IL, United States). Categorical and continuous clinical variables were compared using the Pearson chi-square test or the Fisher’s exact test, if the expected value was <5, and the independent t test, respectively. Spearman’s correlation was calculated to analyze the linear-by-linear association between the period of delayed ATT and mortality rates. A E-value was calculated to examine the potential effect of unmeasured confounders in our study (19). A two-sided P value of <0.05 was considered significant.

Two methods were adopted to investigate the prognostic effects of delayed AAT in the different age groups. First, the significant predictors of 30-day mortality were recognized by the univariable analysis. Consistent with previous methods (3, 20), the significant prognostic predictors and the continuous covariable of “the hour of delayed AAT” were jointly processed under the model of backward and stepwise logistic regression in each age group. Second, similar to the previous study (4), we performed the propensity-score (PS)-matched analysis to overcome baseline differences between three age groups. The PS was calculated by the independent determinants of 30-day crude mortality in overall patients, recognized by the model of backward and stepwise logistic regression. Patients in the different age groups were manually matched at a ratio of 1:1:1, by the closest total propensity-scores. Like the previous rule (21), the matching tolerance for the PS difference was P < 0.2. Focusing on each PS-matched group, the Kaplan–Meier curve and log rank test were utilized for the 30-day survival analysis, and the Pearson chi-square test were used to assess the prognostic effect of delayed AAT.

Of 5,786 patients with bacterial growth in blood cultures, 2,916 were eligible as the aimed population based on the inclusion and exclusion criteria; and 968 (33.2%), 683 (23.4%), and 1,265 (43.4%) were categorized in the middle-aged, old, and very old groups, respectively (Supplementary Figure 1). Of the total 2,916 patients, the majority (2,370, 81.3%) were admitted to general wards through the ED, 304 (10.4%) were admitted to intensive care units, 100 (3.4%) died in the ED, and only 142 (4.9%) were discharged through the ED and subsequently followed at outpatient clinics. The 609 (21.2%) patients were critically ill (i.e., PBS ≥ 4) upon ED arrival, and 15- and 30-day crude mortality rates were 12.0% (349 patients) and 15.4% (449), respectively.

Table 1 presents a comparison of clinical manifestations and outcomes between the middle-aged, old, and very old groups. Significant differences were noted in the proportion of male patients, nursing-home residents, comorbid severity (McCabe and Jackson classification), major comorbidities, severe bacteremia episodes (PBS ≥ 4 at onset), specific bacteremia sources (i.e., pneumonia, skin and soft-tissue infections, primary bacteremia, biliary tract infections, and liver abscess), and the presence of polymicrobial bacteremia among the three groups. Despite the similar periods of delayed AAT, intravenous (IV) antimicrobial therapy, and hospitalization as well as similar 15-day or 30-day mortality rates in the varied age groups, the proportions of approximately half of empirical antimicrobial classes differed in the varied age groups (Supplementary Table 1).

Of the total 3,281 causative microorganisms (middle-aged, old, and very old: 1,065, 762, and 1,454 isolates, respectively), the leading were Escherichia coli (1,222, 37.2%), Klebsiella pneumoniae (455, 13.9%), Staphylococcus aureus (375, 11.4%), Streptococcus species (355, 10.8%), Pseudomonas aeruginosa (108, 3.3%), Enterococcus species (106, 3.2%), Enterobacter cloacae (89, 2.7%), Proteus mirabilis (76, 2.3%), Salmonella enteritidis (48, 1.5%), and Aeromonas species (45, 1.4%). Notably, dissimilar specie distributions among the different age groups were observed in five common microorganisms, in terms of E. coli, K. pneumoniae, S. aureus, S. pneumoniae, and Enterococcus faecalis (Supplementary Table 2).

Overall, methicillin-resistant S. aureus, ampicillin-susceptible enterococci, and penicillin-susceptible streptococci accounted for 39.4% (148 isolates) of S. aureus, 85.8% (91) of enterococci, and 93.5% (332) of streptococci, respectively. Meanwhile, cefazolin, cefuroxime, ampicillin/sulbactam, moxifloxacin, cefotaxime, levofloxacin, ceftazidime, ertapenem, cefepime, piperacillin/tazobactam, and imipenem were active against 50.6, 71.2, 74.3, 81.3, 78.1, 82.2, 83.6, 88.9, 93.0, 94.6, and 99.3% of Gram-negative aerobes, respectively. Ampicillin/sulbactam, moxifloxacin, piperacillin/tazobactam, and metronidazole were active against 78.0, 87.5, 94.0, and 97.6% of the total anaerobes, respectively.

Of numerous predictors of 30-day mortality recognized by the univariable analysis (Supplementary Table 3), the following was independently associated with 30-day crude mortality using the multivariable regression model: nursing-home residence; polymicrobial bacteremia; critically ill (PBS ≥ 4) patients at onset; inadequate source control; bacteremic pneumonia; bacteremia due to urinary tract infections or liver abscess; fatal comorbidities (McCabe and Jackson classification); and comorbidities of hemato-oncology or liver cirrhosis. Notably, the hourly AAT delay was associated with an average increase of 0.3% in 30-day crude mortality rates [adjusted odds ratio (AOR), 1.003; P < 0.001].

A positive delayed AAT-related trend in 30-day mortality rates among the middle-aged (γ = 0.995, P < 0.001), old (γ = 0.969, P = 0.001), and very old (γ = 0.988, P < 0.001) groups was disclosed (Figure 1A). Among 968 patients in the middle-aged group (Table 2), after adjustment for the independent predictors of 30-day mortality determined in the multivariable analysis [i.e., the critical illness at onset, bacteremic pneumonia, and fatal comorbidities (McCabe and Jackson classification)], the prognostic impact of the hourly AAT delay was significant (AOR, 1.002; P = 0.03). Moreover (Table 2), the prognostic effect of 1-h AAT delay remained independently significant in the old age (AOR, 1.004; P < 0.001) and very old (AOR, 1.007; P < 0.001) groups, after respective adjustment for the independent predictors of 30-day mortality.

Based on the closest propensity scores using ten independent determinants of mortality (as shown in Supplementary Table 3), 582 patients in each age group were matched. After adequate PS-matching (Table 1), no significant differences of the patient proportion were exhibited between three age groups, in terms of patient characteristics, the proportion of nursing-home residents, fatal comorbidities (McCabe and Jackson classification), major comorbidities, major sources of bacteremia, polymicrobial bacteremia, inadequate source control, or severe bacteremia (PBS ≥ 4) at onset, the period of delayed AAT, IV antimicrobial therapy, or hospitalization, and 15-day or 30-day crude mortality rates. Of note, the dissimilar proportion between three age groups was only observed in comorbid cardiovascular diseases.

Focusing on the matched groups, the positive delayed-AAT-related trends in 30-day mortality rate among the middle-aged (γ = 0.921, P = 0.009), old (γ = 0.949, P = 0.004), and very old (γ = 0.982, P < 0.001) groups were disclosed (Figure 1B). Notably, the highest and lowest coefficient of correction in the very old and middle-aged groups were disclosed, respectively. Further analyses in survival curves, a significant difference between treatment with delayed AAT > 24 h (Figure 2A) or >48 h (Figure 2B) and treatment without delay was consistently demonstrated in three age groups. Of importance, the largest difference in survival curves in the very old group and the smallest difference in the middle-aged group occurred, in responses to inadequate empirical antibiotic therapy (i.e., delayed AAT > 24 or >48 h).

In further analyses on the matched groups (Table 3), the prognostic impacts of delayed AAT > 24 h remained significant in the middle-aged [odds ratio (OR), 1.73; P = 0.04], old (OR, 1.84; P = 0.03), and very old (OR, 1.87, P = 0.02) groups. Similarly, the prognostic impacts of delayed AAT > 48 h remained significance in the middle-aged (OR, 1.82; P = 0.04), old (OR, 1.95; P = 0.02), and very old (OR, 2.34, P = 0.003) groups. In sum, regardless of whether delayed AAT > 24 or 48 h, the vast prognostic impacts (i.e., the largest OR) of inadequate empirical antibiotic therapy occurred in the very old group.