Antibiotic de-escalation patterns and outcomes in critically ill patients with suspected pneumonia as informed by bronchoalveolar lavage results.

Introduction

Pneumonia is a leading cause of morbidity and mortality worldwide [ 1 ]. Antibiotic stewardship and de-escalation for pneumonia in critically ill patients is crucial given the rise of multi-drug resistant (MDR) organisms [ 2 ]. Despite the most recent guidelines from the Infectious Diseases Society of America (IDSA) advocating for antibiotic de-escalation in patients with no evidence of resistant organisms, substantial provider hesitancy in antibiotic de-escalation exists, even when the microbiology workup yields negative results [ 3 ]. Such provider hesitancy is partly due to the limited diagnostic yield of infectious workup for bacterial pneumonia, as well as concerns for patient safety and outcomes.

Current IDSA guidelines recommend noninvasive respiratory tract sampling, such as sputum sample and endotracheal aspirate, to identify causative organisms in patients with community-acquired pneumonia (CAP) who are at risk for resistant organisms or severe disease, as well as in patients with hospital-acquired pneumonia (HAP) and ventilator-associated pneumonia (VAP) [ 4 , 5 ]. However, high-quality, uncontaminated sputum samples can only be successfully obtained from a limited percentage of patients with CAP and HAP [ 4 , 6 , 7 ]. In a study of hospitalized patients with CAP, only 62% could provide a sputum sample, out of which only 57% met criteria for high-quality specimens [ 6 ]. In a study of patients with HAP, only 29.4% of patients had a sputum sample collected, out of which 64% were contaminated [ 7 ]. Sputum Gram stains have a rapid turnaround time, but the sensitivity is suboptimal. For example, a positive stain for Gram-positive diplococci is only 68.2% sensitive for identifying S. pneumoniae [ 6 ]. For patients with VAP, endotracheal aspirate can sometimes show discordant Gram stain and semiquantitative culture results with lower respiratory tract samples, indicating limited diagnostic accuracy. In a study focusing on the concordance between diagnostic workup performed on endotracheal aspirate and bronchoalveolar lavage (BAL) fluid, Gram stain results on endotracheal aspirates and BAL fluid only agreed about 60% of the time, and semiquantitative cultures performed on endotracheal aspirates can have a sensitivity as low as 38% [ 8 ]. Blood cultures are another commonly used method of noninvasive infectious sampling, but their diagnostic yield is low in pneumonia regardless of the severity and presence of risk factors for resistant organisms [ 4 , 9 ]. Invasive sampling of the lower respiratory tract with BAL often has better diagnostic yields, but such tests are performed at variable rates (proportions ranging from 0% to 25.8% of mechanically ventilated patients in one study) [ 10 ], even in critically ill patients [ 11 – 13 ]. BAL samples can be the substrate of multiple tests crucial to diagnosing and characterizing pneumonia, including cell count and differential, quantitative cultures, and multiplex PCR rapid diagnostic tests (RDT). While not included in current pneumonia guidelines, multiplex PCR RDTs performed on BAL fluid provide rapid results and effectively identify infectious organisms [ 14 – 16 ].

Ideally, antibiotic de-escalation in critically ill patients with pneumonia should be based on a rapid, sensitive, and specific microbiology diagnostic test. Combining multiplex PCR RDTs and quantitative cultures performed on BAL samples can potentially achieve such diagnostic goals and can be incorporated into antibiotic de-escalation strategies [ 17 ]. When given this sensitive and detailed data, clinicians in our center often de-escalate antibiotics [ 11 ]. Various ways of quantifying antibiotic administration have been proposed such as days of therapy [ 18 ]. Here we use the Narrow Antibiotic Therapy (NAT) score, which quantifies the breadth of antibiotic coverage as well as days administered [ 11 ]. In this study, we describe the antibiotic prescription pattern across different types of pneumonia (i.e., bacterial, viral, mixed bacterial/viral, microbiology-negative) in critically ill pneumonia patients when antibiotic de-escalation is informed by the combination of multiplex PCR RDT and quantitative cultures performed on BAL samples.

Methods

Study cohort

Patients were enrolled in the Successful Clinical Response In Pneumonia Therapy (SCRIPT) study (IRB STU00204868), a single-center prospective cohort study of patients with suspected pneumonia admitted to the medical intensive care unit (MICU) of Northwestern Memorial Hospital and requiring mechanical ventilation [ 11 , 19 , 20 ]. Patient families or legally authorized representatives consented to participate in the study. All demographic data, including sex, race, and ethnicity, were based on self-reported information entered into the electronic health record (EHR). The demographic data reflects the demographic breakdown of the patients admitted to the Northwestern Memorial Hospital MICU, and disaggregated data were reported when appropriate. If a patient was readmitted to the hospital, they were treated as a new study subject and given a new identifier. The SCRIPT research team reviewed patient charts on study enrollment to determine who fulfills set criteria of immunocompromised conditions (see Table 1 footnote).

Pneumonia episodes

BAL was performed for suspected pneumonia episodes at the discretion of the treating MICU team. At our center, BAL samples were routinely acquired to guide clinical decision-making. BAL is either performed by physicians through fiberoptic bronchoscopy or by respiratory therapists through a non-bronchoscopic approach (NBBAL). BAL samples were sent for routine clinical tests including cell count and differential, gram stain, quantitative culture, and many were sent for multiplex PCR RDT (BioFire ® FilmArray ® Pneumonia Panel) to identify possible causative organisms. At our institution, the PCR RDT has a runtime of approximately one hour, with results typically posted to the EHR within four hours. In contrast, cultures require around 24 h for Gram stain and preliminary findings, and 48 h for organism speciation. Susceptibility testing can take 72 h or longer. Bacterial cultures are generally held for five days to allow final results, while fungal cultures are incubated for four weeks, and AFB cultures for up to eight weeks. Blood cultures were routinely performed in the SCRIPT cohort when there was a suspicion for bacteremia. When there was further suspicion for fungemia or bacteremia with acid-fast bacteria, fungal cultures and AFB cultures were also sent. Blood culture results were used in conjunction with BAL results to determine antibiotic treatment plan. Our hospital’s guidelines provide general recommendations for antibiotic de-escalation when there is no evidence of resistance and in the context of clinical improvement, in compliance with the current recommendations from IDSA and ATS. Intensivists and pharmacists at our institution are both encouraged to follow these guidelines [ 21 ].

A panel of critical care physicians reviewed all available clinical data from the EHR and adjudicated suspected pneumonia episodes based on a pre-determined multi-reviewer protocol [ 22 ]. Based on the BAL results and chart review, they categorized each episode into bacterial pneumonia, viral pneumonia, mixed bacterial/viral pneumonia, microbiology-negative pneumonia, and non-pneumonia control. Microbiology-negative pneumonia was adjudicated based on pulmonary-specific features such as chest imaging, BAL neutrophilia, and ventilator parameters, and required clinical consensus from multiple reviewers. Presence of extra-pulmonary infection was also reviewed based on each patient’s EHR. Each bacterial pneumonia episode was further adjudicated on day 7 following BAL to determine whether the pneumonia had been clinically cured based on improvements in features such as fever curve, leukocytosis, and ventilator settings. Specific details of the adjudication protocol and criteria for clinical cure have been published [ 22 ]. Bacterial pneumonia episodes were categorized into resistant and susceptible based on the results of quantitative culture and multiplex PCR. An episode was defined as resistant if the causative organism tested positive for any antibiotic resistance gene ( mec A/C, MREJ, OXA-48-like, CTX-M, KPC, NDM, IMP, VIM), or if the quantitative culture result indicated a resistant pathogen (such as an ESBL-positive organism or MRSA). The code for processing PCR and quantitative culture results can be found at https://github.com/NUSCRIPT/mz_abx_deescalation_2024 .

Characterizing antibiotic stewardship pattern with the narrow antibiotic treatment (NAT) score

We quantified the spectrum of antibiotic therapy administered to each patient on each day using the NAT score. Briefly, a score of 0 corresponded to empiric guideline-recommended CAP coverage of ceftriaxone and azithromycin, −1 corresponded to monotherapy with a narrow-spectrum antibiotic, and −2 corresponded to no antibiotic treatment. Antibiotic therapies with broader spectrums, such as for resistant HAP or VAP, were assigned higher scores [ 11 , 23 ]. A description and examples of how NAT scores were assigned and analyzed relative to post-BAL days are included in Fig. 1 .

Antibiotic therapy spectrum for each pneumonia etiology was visualized by plotting the median and interquartile range (IQR) of NAT score against day, from two days before to seven days after the episode-defining BAL. If a patient died or was discharged less than seven days after the episode-defining BAL, then only the NAT scores during the inpatient days were included. Antibiotic de-escalation patterns were visualized by plotting the average NAT score over the first to the seventh day after the episode-defining BAL, or to the day of death or discharge, whichever came first.

Patient outcomes

For outcome analysis, we only included the patients with exactly one suspected pneumonia episode in order to avoid concurrent contribution from multiple different pneumonia etiologies to patient outcomes. The primary outcome was a composite of in-hospital mortality, discharge to hospice, or requiring lung transplantation during hospitalization, which we referred to as unfavorable outcomes. The secondary outcomes were duration of ICU stay, duration of intubation, and Clostridium difficile during admission. Clinical outcomes were compared across different pneumonia etiologies with the Mann–Whitney U test and Fisher’s exact test.

The patient cohort flow diagram is presented in Fig. 2 .

Statistical analysis

Average NAT scores were compared using Mann–Whitney U tests, with p < 0.05 as the cutoff for statistical significance. Patient mortality across different pneumonia etiologies was compared using Fisher’s exact test. Duration of ICU stay and duration of mechanical ventilation were compared using Mann–Whitney U tests. Statistical analyses and plotting were performed with R version 4.3.1. Data were analyzed with dplyr version 1.1.2 and tidyr version 1.3.0. All plots were generated with ggplot2 version 3.4.2 and ggpubr 0.6.0. Preprocessing was performed with Python version 3.9. Code is available at repository https://github.com/NUSCRIPT/mz_abx_deescalation_2024 .

Results

Clinical characteristics of study cohort

686 patients enrolled in SCRIPT between June 2018 and April 2023 had complete clinical adjudication information at the time of analysis ( Table 1 ). Among these patients, 409 (59.6%) were male, the median age was 62 years, and 206 (30.0%) were immunocompromised. These patients had 927 clinical episodes that were subsequently adjudicated to be bacterial pneumonia ( n = 288), viral pneumonia ( n = 176), mixed bacterial/viral pneumonia ( n = 198), microbiology-negative pneumonia ( n = 151), and non-pneumonia control ( n = 114). The adjudication was based on the combination of quantitative culture and PCR RDT results if available. These episodes were included in the visualization and analysis of antibiotic stewardship patterns. Of these episodes, 150 episodes were adjudicated as CAP, 257 as HAP, and 406 as VAP. Of the viral and mixed bacterial/viral pneumonia cases, 76% cases were SARS-CoV-2. Of the 486 bacterial and bacterial/viral episodes, 104 had resistant pathogens as identified using our logic ( Supplemental Table 1 ). The organisms detected are presented in Supplemental Table 2 . A detailed break-down of all pneumonia episodes by category and etiology is summarized in Supplemental Table 3 . The frequencies of other viruses causing pneumonia episodes can be found in Supplemental Table 4 .

Of all 686 patients, 534 had exactly one suspected pneumonia episode and were included in the clinical outcome analysis. Out of these patients, 164 had bacterial pneumonia, 93 had viral pneumonia, 76 had mixed bacterial/viral pneumonia, 107 had microbiology-negative pneumonia, and 94 were adjudicated not to be pneumonia (non-pneumonia control) ( Table 2 ).

The agreement and discrepancy between quantitative cultures and PCR RDT performed on BAL samples

Multiplex PCR result reporting was integrated into our EHR system on April 7, 2020. Of 927 episode-defining BAL samples, 657 were collected after that date, with 635 (96.7%) having both quantitative cultures and multiplex PCR performed, indicating a high compliance rate among physicians to send both quantitative cultures and multiplex PCR from BAL samples.

We also examined the agreement and discrepancies between PCR and quantitative culture results. Among these 635 BAL samples, 390 (61.4%) showed complete agreement between multiplex PCR and quantitative cultures in identifying bacterial organisms and resistance. 158 (24.9%) had organisms or resistance detected by culture but missed by PCR, but of these were 139 (21.9%) cases where the culture-identified bacteria were not part of the multiplex PCR panel. 61 (9.6%) had organisms or resistance detected by PCR but missed by culture. 26 (4.1%) had discrepancies, with PCR and culture each identifying different organisms.

Antibiotic stewardship pattern in treating pneumonia of different etiologies

The median NAT score on the day of BAL sample collection was 2 for bacterial pneumonia, microbiology-negative pneumonia, and non-pneumonia control episodes, which corresponded to empirical coverage for HAP or VAP ( Fig. 3 ). While some of the non-pneumonia control episodes were adjudicated to be completely noninfectious (e.g. heart failure exacerbation), some of these episodes were adjudicated to co-exist with extrapulmonary infections (e.g. pyelonephritis). When we further stratified non-pneumonia control episodes based on the presence or absence of extra-pulmonary infection ( n = 36 vs. n = 75), non-pneumonia control episodes with concurrent extra-pulmonary infection had a median NAT score of 2 on the day of BAL collection, while those without extra-pulmonary infection had a median score of 1 ( P = 0.014, Mann–Whitney U test). For viral pneumonia and mixed bacterial/viral pneumonia, the median NAT score on the day of BAL collection was 1, which corresponds to targeted HAP or VAP coverage. Notably, an antibiotic de-escalation pattern was evident in all pneumonia etiologies except resistant bacterial pneumonia on days 1–2 following BAL collection. This pattern was also reflected by the difference in average NAT scores of each pneumonia etiology ( Fig. 4 ). No significant difference in average NAT scores was observed between men and women ( Supplemental Fig. 1 ). For the viral pneumonia episodes, 44% had complete antibiotic cessation by post-BAL day 5.

To determine the association between pneumonia etiology, antibiotic stewardship patterns, and pneumonia outcome we further stratified bacterial, mixed bacterial/viral, and microbiology-negative pneumonia episodes based on the cure status on day 7 following BAL collection. Similarly, we stratified bacterial, viral, mixed bacterial/viral, microbiology-negative pneumonia episodes and non-pneumonia control episodes based on the presence or absence of concurrent extra-pulmonary infections. The cured pneumonia episodes of all etiologies demonstrated lower average NAT scores than uncured episodes ( Fig. 5 ). The episodes with concurrent extra-pulmonary infections demonstrate higher average scores, except for resistant bacterial pneumonia and mixed bacterial/viral pneumonia ( Fig. 6 ).

The extent to which the treatment was modified by PCR RDT results versus quantitative culture results

We analyzed the antibiotic de-escalation timeline in cases of susceptible bacterial pneumonia and microbiology-negative pneumonia, where de-escalation was most clearly observed. This analysis included 205 episodes of susceptible bacterial pneumonia and 136 episodes of microbiology-negative pneumonia, each lasting at least four days post-BAL collection. Among the 205 bacterial pneumonia episodes, 69 (33.7%) underwent antibiotic de-escalation on day 1, likely guided by multiplex PCR, while 115 (56.1%) were de-escalated by day 4, likely due to a combination of multiplex PCR and quantitative culture. Similarly, among the 136 microbiology-negative pneumonia episodes, 42 (30.9%) had de-escalation on day 1, and 78 (57.4%) by day 4. Therefore, we estimated that about 60% of susceptible bacterial pneumonia cases and 53.8% of microbiology-negative pneumonia cases that had antibiotic de-escalation by post-BAL day 4 received earlier de-escalation likely based on PCR results.

Blood cultures collected during suspected pneumonia episodes mostly did not influence antibiotic management

Among the 686 patients (15,287 ICU patient-days) included in the analysis, blood cultures were obtained on 2,337 patient-days, including 80 with fungal cultures and 6 with AFB cultures. Of these, 1,093 patient-days fell within the timeframe from two days before to seven days after an episode-defining BAL, encompassing 585 suspected pneumonia episodes in 467 patients. The majority (1,030 of 1,093) of these patient-days had negative blood culture results. In most pneumonia episodes, blood cultures did not lead to a change in management. Only 47 suspected pneumonia episodes had positive blood cultures, of which 9 were adjudicated as non-pneumonia control cases. Of these non-pneumonia cases, 4 had intra-abdominal infections, one had a genitourinary infection, one had a line associated infection. Among the 38 confirmed pneumonia episodes, 9 had consistent microbiology results between BAL and blood cultures. In 20 episodes, clinically significant (i.e., not Candida albicans ) bacterial or fungal organisms were identified in BAL but not in blood cultures, while 10 episodes had additional organisms detected in blood cultures.

Clinical outcomes

Unfavorable outcomes, duration of ICU stay, and duration of mechanical ventilation among patients with exactly one suspected pneumonia episode are reported in Table 3 . All categories of pneumonia had comparable unfavorable outcome rates to microbiology-negative pneumonia (Bacterial: OR = 1, P = 1.00; Resistant bacterial: OR = 1.19, P = 0.71; Susceptible bacterial: OR = 0.94, P = 0.89; Viral: OR = 1.09, P = 0.78; Bacterial/viral: OR = 1.18, P = 0.65). Patients with bacterial pneumonia and microbiology-negative pneumonia had overall comparable durations of ICU stay and intubation, but patients with viral pneumonia and mixed bacterial/viral pneumonia had significantly longer durations of ICU stay and intubation than those with microbiology-negative pneumonia. No major difference in clinical outcomes was observed between males and females ( Supplemental Table 5 ).

Of all 686 patients, only 14 (2.0%) tested positive for Clostridium difficile during the admission. Among the 14 patients, two had two bacterial/viral pneumonia episodes, one had one bacterial/viral pneumonia episode, two had one susceptible bacterial pneumonia episode, two had two microbiology-negative pneumonia episodes, three had one viral pneumonia episode, one had two resistant bacterial pneumonia episodes and one microbiology-negative pneumonia episode, and three had one non-pneumonia control episode.

Discussion

Our study describes an antibiotic prescription pattern where early antibiotic de-escalation in critically ill pneumonia patients is informed by PCR RDT and quantitative cultures performed on BAL samples and reports on the safety and feasibility of this practice. This study is one of the first to quantify the de-escalation of antibiotic therapy on consecutive ICU days with the NAT score. The NAT score distinguishes between broad-spectrum and narrow-spectrum antibiotics and is sensitive to the number of antibiotics given per day, providing more granular information compared with binary features such as antibiotic-free days. We describe an antibiotic management strategy from our institution where de-escalation occurred promptly after a BAL sample resulted in no resistant pathogens detected by multiplex PCR and quantitative cultures [ 11 ]. With early antibiotic de-escalation informed by BAL studies, we treat critically ill patients with microbiology-negative pneumonia similarly to or less aggressively than those with susceptible bacterial pneumonia and achieve comparable clinical outcomes. The overall incidence of Clostridium difficile in this patient cohort is low. The presence of clinical improvement and extra-pulmonary infections is taken into consideration when implementing antibiotic de-escalation. Our study found no significant association between sex and antibiotic de-escalation or clinical outcomes, suggesting that treatment decisions are primarily driven by factors unrelated to sex. These findings support generalizability across sexes, though unmeasured biological or social influences may still play a role in other settings.

Interestingly, the patients with viral-only pneumonia received the greatest antibiotic de-escalation (sometimes complete discontinuation) after BAL collection, which differs from the current published CAP guidelines that recommend continuing antibiotics in this scenario. The viral-only cases had the lowest average NAT score, and although the duration of mechanical ventilation and ICU stay were longer, still had comparable mortality to other categories of pneumonia. Our viral-only cases mostly comprised patients with SARS-CoV-2 pneumonia, which is known to have longer ventilation and ICU durations [ 20 , 24 ].

Due to their critical illness, clinical hesitancy to de-escalate antibiotics in ICU patients despite negative microbiology tests leads to inappropriately prolonged broad-spectrum antibiotic therapy [ 25 , 26 ]. The antibiotic prescription patterns described in this study suggest how BAL sampling encourages clinical practice consistent with the recommendations from the IDSA and enables early de-escalation of antibiotics following infectious workup without evidence of a resistant pathogen [ 23 ]. The recommendation on antibiotic stewardship in microbiology-negative pneumonia is a recent addition to the guidelines, as previous guidelines did not provide specific recommendations for this clinical scenario [ 27 ]. Despite this recent update to guidelines, many patients with microbiology-negative pneumonia do not receive antibiotic de-escalation in a timely fashion. In a retrospective cohort study, which included 164 hospitals across the United States, only one out of every seven patients with microbiology-negative pneumonia received antibiotic de-escalation by hospital day 4 [ 3 ].

Limitations to our study include its occurrence at a tertiary academic center with access to state-of-the-art rapid multiplex PCR, a robust BAL practice, including a significant NBBAL subset performed independently by respiratory therapists [ 28 ], and an established comfort level with basing antibiotic management on test results [ 13 ]. This was an observational study, and thus reflect heterogeneity of practice by different physicians in terms of specific antibiotic choices and processes of care. Since clinical improvement was taken into consideration when implementing antibiotic de-escalation, it was possible that less aggressive antibiotic strategies were preferentially selected for patients with improving clinical trajectory. It is important to recognize that the lack of a control group in this study prevents attributing the observed results directly to BAL culture and PCR testing. Additionally, the relatively low number of pneumonia episodes involving multidrug-resistant bacteria limits the ability to determine how applicable these findings are to settings where such infections are more prevalent. In addition, since we used the combination of multiplex PCR and quantitative cultures to de-escalate antibiotics, pinpointing the extent to which treatment was modified by PCR versus Gram stain and preliminary culture results in retrospective analysis is difficult due to the similar timeframe of results availability.

Conclusion

Here we quantify and describe day-by-day antibiotic de-escalation patterns from our institution in different etiologies of pneumonia episodes. Over a quarter of the time in viral pneumonia episodes, antibiotics were completely discontinued. Future prospective studies investigating the efficacy of a protocolized de-escalation strategy versus routine clinical care will be helpful to validate the feasibility and safety of more aggressive de-escalation.

Supplementary Material

Supplementary Information The online version contains supplementary material available at https://doi.org/10.1007/s10096-025-05144-2 .

Funding

SCRIPT is supported by NIH/NIAID U19 AI135964. CAG is supported by NIH/NHLBI K23HL169815, a Parker B. Francis Opportunity Award, and an ATS Unrestricted Grant. BDS is supported by the NIH (R01HL149883, R01HL153122, P01HL154998, P01 AG049665, and U19 AI135964). RGW is supported by NIH grants U19 AI135964, U01 TR003528, P01HL154998, R01HL149883, R01LM013337. CIP is supported by NIAID (U19 AI135964), Northwestern University Clinical and Translational Sciences Institute (5 KL2 TR001424-09). The funding sources did not have a role in the design, execution, or prior review of the study or in the data presented in this manuscript. Opinions expressed in this work do not necessarily reflect those of the funding sources.

Footnotes

Competing interests B.D.S. holds US patent 10,905,706, “Compositions and methods to accelerate resolution of acute lung inflammation,” and serves on the scientific advisory board of Zoe Biosciences, in which he holds stock options. Other authors declare no conflicts of interest.

Ethics approval This study was approved by the Northwestern University Institutional Review Board with study ID STU00204868.

Consent to participate Informed consent was acquired from all study participants or their surrogates.

Data availability

A significant portion of this data has been already made available through PhysioNet at https://physionet.org/content/script-carpediem-dataset/1.1.0/

A future update will include new patients and updated data since the publication of the original dataset. Code is available at https://github.com/NUSCRIPT/mz_abx_deescalation_2024 .

References

1. World Health Organization. The top 10 causes of death. https://www.who.int/news-room/fact-sheets/detail/the-top-10-causes-of-death . Accessed 30 Apr 2025

2. Wunderink RG, Srinivasan A, Barie PS et al. (2020) Antibiotic stewardship in the intensive care unit: an official American thoracic society workshop report in collaboration with the aacn, chest, cdc, and sccm. Ann Am Thorac Soc 17(5):531–540. 10.1513/AnnalsATS.202003-188ST

3. Deshpande A, Richter SS, Haessler S et al. (2021) De-escalation of empiric antibiotics following negative cultures in hospitalized patients with pneumonia: rates and outcomes. Clin Infect Dis 72(8):1314–1322. 10.1093/cid/ciaa212

4. Metlay JP, Waterer GW, Long AC et al. (2019) Diagnosis and treatment of adults with community-acquired pneumonia. Am J Respir Crit Care Med 200(7):E45–E67. 10.1164/rccm.201908-1581ST

5. Kalil AC, Metersky ML, Klompas M et al. (2016) Management of adults with hospital-acquired and ventilator-associated pneumonia: 2016 clinical practice guidelines by the Infectious Diseases Society of America and the American Thoracic Society. Clin Infect Dis 63(5):e61–e111. 10.1093/cid/ciw353

6. Miyashita N, Shimizu H, Ouchi K et al. (2008) Assessment of the usefulness of sputum Gram stain and culture for diagnosis of community-acquired pneumonia requiring hospitalization. Med Sci Monit 14(4):171–176

7. Naidus EL, Lasalvia MT, Marcantonio ER, Herzig SJ (2018) The diagnostic yield of noninvasive microbiologic sputum sampling in a cohort of patients with clinically diagnosed hospital-acquired pneumonia. J Hosp Med 13(1):34–37. 10.12788/jhm.2868

8. Scholte JBJ, Van Dessel HA, Linssen CFM et al. (2014) Endotracheal aspirate and bronchoalveolar lavage fluid analysis: inter-changeable diagnostic modalities in suspected ventilator-associated pneumonia? J Clin Microbiol 52(10):3597–3604. 10.1128/JCM.01494-14

9. Zhang D, Yang D, Makam AN (2019) Utility of blood cultures in pneumonia. Am J Med 132(10):1233–1238. 10.1016/j.amjmed.2019.03.025

10. Wayne MT, Seelye S, Molling D et al. (2022) Variation in U.S. hospital practices for bronchoscopy in the intensive care unit. Ann Am Thorac Soc 19(6):1061–1065. 10.1513/AnnalsATS.202110-1141RL

11. Pickens CO, Gao CA, Cuttica MJ et al. (2021) Bacterial superinfection pneumonia in patients mechanically ventilated for COVID-19 pneumonia. Am J Respir Crit Care Med 204(8):921–932. 10.1164/rccm.202106-1354OC

12. Gadsby NJ, Russell CD, Mchugh MP et al. (2016) Comprehensive molecular testing for respiratory pathogens in community-acquired pneumonia. Clin Infect Dis 62(7):817–823. 10.1093/cid/civ1214

13. Paonessa JR, Shah RD, Pickens CI et al. (2019) Rapid detection of methicillin-resistant Staphylococcus aureus in BAL: a pilot randomized controlled trial. Chest 155(5):999–1007. 10.1016/j.chest.2019.02.007

14. Strålin K, Korsgaard J, Olcén P (2006) Evaluation of a multiplex PCR for bacterial pathogens applied to bronchoalveolar lavage. Eur Respir J 28(3):568–575. 10.1183/09031936.06.00006106

15. Baudel JL, Tankovic J, Dahoumane R et al. (2014) Multiplex PCR performed of bronchoalveolar lavage fluid increases pathogen identification rate in critically ill patients with pneumonia: a pilot study. Ann Intensive Care 4(1):1–9. 10.1186/s13613-014-0035-7

16. Affolter K, Schumann DM, Tamm M et al. (2018) Multiplex PCR on the bronchoalveolar lavage fluid of immunocompromised patients. Chest 154(3):722–725. 10.1016/j.chest.2018.06.018

17. Darie AM, Khanna N, Jahn K et al. (2022) Fast multiplex bacterial PCR of bronchoalveolar lavage for antibiotic stewardship in hospitalised patients with pneumonia at risk of Gram-negative bacterial infection (Flagship II): a multicentre, randomised controlled trial. Lancet Respir Med 10(9):877–887. 10.1016/S2213-2600(22)00086-8

18. Benić MS, Milanič R, Monnier AA et al. (2018) Metrics for quantifying antibiotic use in the hospital setting: Results from a systematic review and international multidisciplinary consensus procedure. J Antimicrob Chemother 73:vi50–vi58. 10.1093/jac/dky118

19. Gao CA, Markov NS, Stoeger T et al. (2023) Machine learning links unresolving secondary pneumonia to mortality in patients with severe pneumonia, including COVID-19. J Clin Investig 133(12):e170682. 10.1172/JCI170682

20. Grant RA, Morales-Nebreda L, Markov NS et al. (2021) Circuits between infected macrophages and T cells in SARS-CoV-2 pneumonia. Nature 590(7847):635–641. 10.1038/s41586-020-03148-w

21. Northwestern Memorial Hospital Suggested Empirical Antimicrobial Therapy by Site of Infection - Respiratory. https://adsp.nm.org/uploads/1/4/3/0/143064172/empiric_prescribing_guidelines_respiratory.pdf . Accessed 7 Apr 2025

22. Pickens CI, Gao CA, Bodner J et al. (2023) An adjudication protocol for severe pneumonia. Open Forum Infect Dis 10(7):1–8. 10.1093/ofid/ofad336

23. Wang AA, Pickens CO, He H et al. (2020) The narrow-spectrum antibiotic treatment score: a novel quantitative tool for assessing broad- and narrow-spectrum antibiotic use in severe community-acquired pneumonia. Am J Respir Crit Care Med 201:A2929. 10.1164/ajrccm-conference.2020.201.1_meetingabstracts.a2929

24. Nolley EP, Sahetya SK, Hochberg CH et al. (2023) Outcomes among mechanically ventilated patients with severe pneumonia and acute hypoxemic respiratory failure from SARS-CoV-2 and other etiologies. JAMA Netw Open 6(1):e2250401. 10.1001/jamanetworkopen.2022.50401

25. Chiotos K, Tamma PD, Gerber JS (2019) Antibiotic stewardship in the intensive care unit: challenges and opportunities. Infect Control Hosp Epidemiol 40(6):693–698. 10.1017/ice.2019.74

26. Thomas Z, Bandali F, Sankaranarayanan J, Reardon T, Olsen KM (2015) A multicenter evaluation of prolonged empiric antibiotic therapy in adult ICUs in the United States. Crit Care Med 43(12):2527–2534. 10.1097/CCM.0000000000001294

27. Mandell LA, Wunderink RG, Anzueto A et al. (2007) Infectious diseases society of america/american thoracic society consensus guidelines on the management of community-acquired pneumonia in adults. Clin Infect Dis 44(Supplement_2):S27–S72. 10.1086/511159

28. Walter JM, Helmin KA, Abdala-Valencia H, Wunderink RG, Singer BD (2018) Multidimensional assessment of alveolar T cells in critically ill patients. JCI Insight 3(17):e123287. 10.1172/jci.insight.123287

Associated Data

Supplementary Materials

Data Availability Statement

A significant portion of this data has been already made available through PhysioNet at https://physionet.org/content/script-carpediem-dataset/1.1.0/

A future update will include new patients and updated data since the publication of the original dataset. Code is available at https://github.com/NUSCRIPT/mz_abx_deescalation_2024 .