Introduction
Antibiotics are one of the most widely prescribed medication classes and have been largely successful in treating bacterial infections. Antibiotics are prescribed to over 33% of patients treated in the outpatient setting and nearly 50% of patients treated in the inpatient setting [Reference Chen, Li and Wang1]. However, disparities in antibiotic prescribing based on patient factors, such as race, ethnicity, age, or sex, have been documented in many settings and disease states [Reference Young, Strey and Lee2,Reference Kim, Kabbani and Dube3]. These disparities in prescribing may lead to inequitable care, higher healthcare cost, and increased antimicrobial resistance [Reference Kim, Kabbani and Dube3–Reference Menting, Redican, Murphy and Bucholc5]. The Centers for Disease Control and Prevention (CDC) estimate that antibiotic-resistant infections are responsible for more than 3 million infections and over 35,000 deaths in the United States, highlighting the serious threat posed by antimicrobial resistance [6]. The CDC and prior studies have classified Methicillin-resistant Staphylococcus aureus (MRSA) and Pseudomonas aeruginosa as leading causes of antibiotic-resistant infections, noting their significant contribution to hospital costs and mortality [6–Reference Bassetti, Magnasco, Vena, Portunato and Giacobbe11].
Community-acquired bacterial pneumonia (CABP) is a leading cause of inpatient mortality and high hospital costs [Reference Vaughn, Dickson, Horowitz and Flanders12–Reference Divino, Schranz, Early, Shah, Jiang and DeKoven14]. For patients admitted to the hospital with CABP, the cost of treatment per patient is approximately $17,000 on average, with an average length of stay of 6 days and inpatient mortality exceeding 6% [Reference Divino, Schranz, Early, Shah, Jiang and DeKoven14–Reference Weycker, Moynahan, Silvia and Sato16]. In 2019, the Infectious Disease Society of America (IDSA) and the American Thoracic Society (ATS) jointly published guidelines for the management of CABP, which outline recommendations for antibiotic therapy [17]. Guideline-concordant therapy is associated with improved outcomes for patients with CABP and is important for reducing antimicrobial resistance among causative pathogens [Reference Egger, Myers, Arnold, Pass, Ramirez and Brock18–Reference Capelastegui, España and Quintana24].
While the use of and disparities in guideline-concordant therapy have been studied in CABP, no studies have focused on guideline-concordant therapy among CABP due to drug-resistant pathogens. Further, patients with CABP due to antibiotic-resistant infections, such as MRSA or P. aeruginosa, experience worse outcomes or may develop higher severity pneumonia (i.e., necrotizing pneumonia) [Reference Siddiqui and Koirala25,Reference Sando, Suzuki and Ishida26]. Evaluating disparities in guideline-concordant antibiotic prescribing is important in identifying inequities in patient outcomes and care, which may inform the future development of targeted interventions that minimize these disparities. The objective of this study was to evaluate disparities in guideline-concordant antibiotic prescribing based on patient factors for patients admitted to the hospital with CABP due to MRSA or P. aeruginosa.
Methods
Study design and patient population
This was a retrospective cohort study of patients admitted to the hospital with a diagnosis of pneumonia due to MRSA or P. aeruginosa from 01/01/2011 to 10/01/2023, in version 8 of the All of Us database (cutoff date: 10/1/23). The screening dates for pneumonia due to MRSA or P. aeruginosa correspond to the publication year of the 2011 IDSA/ATS MRSA pneumonia treatment guidelines [Reference Liu, Bayer and Cosgrove27]. Patients were included in this study if they had a diagnosis of pneumonia (SNOMED 233604007) present in their record, were treated in the inpatient setting, and were prescribed antibiotics during the same encounter as the pneumonia diagnosis. Billing codes used to determine MRSA attribution were 124691000119101 (SNOMED), J15.212 (ICD-10), and 482.42 (ICD-9). Codes used to determine P. aeruginosa attribution were 41381004 (SNOMED), J15.1 (ICD-10), and 482.1 (ICD-9) [Reference Evans, Fortin-Leung and Kumar28,29]. Patients were excluded if they had documented causative pathogens other than MRSA or P. aeruginosa or if they had hospital-acquired pneumonia (HAP). For this study, HAP was defined as pneumonia with an initial diagnosis more than 48 hours after an inpatient admission. If a patient had multiple pneumonia visits on their record, only the earliest visit was counted.
Data source
We used the controlled tier of version 8 of the All of Us database (released February 2025), which includes individual-level data from electronic health records (EHRs) and surveys collected through 10/01/2023. The All of Us research program is an initiative spearheaded by the National Institutes of Health (NIH) to increase the speed of medical research discoveries [30]. To date there are 861,000+ participants consented and 746,000+ enrolled [31]. Of those enrolled, 40% are from racial or ethnic minority groups, and many are from groups underrepresented in biomedical research [31]. Additionally, All of Us contains data on antibiotic prescribing and subtypes of pneumonia, making this study possible.
Data collection
All tools used were provided in the cloud-based All of Us Researcher Workbench. Data were collected from the Controlled Access tier All of Us database using the cohort builder tool and dataset builder tool. Concept sets used for the dataset builder tool were “Pneumonia,” “Antibacterials for Systemic Use,” “Demographics,” and “The Basics Survey.” Data were then analyzed using Jupyter Notebook analysis environment and the R coding language inside the Researcher Workbench. Data were stored in the permanent storage area in the Researcher Workbench called the “Workspace Bucket,” which is a part of Google Cloud Storage.
Study variables and end points
Patients were stratified into groups based on whether they received guideline-concordant antibiotic therapy or guideline-discordant antibiotic therapy. Anti-MRSA guideline-concordant antibiotic therapy was defined as vancomycin or linezolid [17,Reference Liu, Bayer and Cosgrove27,Reference Kalil, Metersky and Klompas32]. Antipseudomonal guideline-concordant antibiotic therapy was defined as piperacillin-tazobactam, cefepime, ceftazidime, imipenem, meropenem, or aztreonam [17,Reference Kalil, Metersky and Klompas32]. Guideline-discordant antibiotic therapy included any other antibiotics prescribed. Only antibiotics received within 48 hours of the pneumonia diagnosis were considered. If a patient received any guideline-concordant antibiotic therapy within 48 hours of the pneumonia diagnosis, then they were considered to have received guideline-concordant antibiotic therapy for CABP for the purpose of this study.
Demographic characteristics were determined using the prepackaged “Demographics” and “The Basics” concept set, which contains information on sociodemographic (age, sex, race, ethnicity) and behavioral (marital status, tobacco, and alcohol use) factors. Race, ethnicity, and behavioral characteristics were self-reported during the All of Us enrollment process [33]. Patients who reported their ethnicity as Hispanic in the Self-Reported Category column or Ethnicity column were classified as Hispanic. Patients who were not classified as Hispanic and reported their race as Black or White in the Self-Reported Category column or Race column were classified as Non-Hispanic Black (NHB) or Non-Hispanic White (NHW), respectively. All other patients who did not meet the above classification criteria for race and ethnicity were classified as Other due to the small sample size of other groups. The All of Us database provides patient-provided information on both gender identity and sex assigned at birth. For this study, we determined sex from the sex-assigned-at-birth question.
The primary outcomes for this study were receipt of anti-MRSA guideline-concordant antibiotic therapy or the receipt antipseudomonal guideline-concordant antibiotic therapy.
Statistical analysis
Statistical comparisons were conducted in Jupyter Notebook inside the All of Us Researcher Workbench. Patient ages were compared using the Wilcoxon rank sum test. The chi-square test was used to compare dichotomous variables such as baseline characteristics. Counts of less than or equal to 20 are displayed as n ≤ 20 to comply with the All of Us data and statistic dissemination policy [34]. The actual count was used for statistical comparisons.
Multivariate logistic regression models were created with race and ethnicity (NHB vs. NHW, Hispanic vs. NHW, Hispanic vs. NHB) as the independent variable, receipt of either anti-MRSA guideline-concordant antibiotic therapy or antipseudomonal guideline-concordant antibiotic therapy as the dependent variable, and age (≥65 years vs. <65 years), sex (male vs. female), marital status (yes vs. no), alcohol use (yes vs. no), cigarette use (yes vs. no), and e-cigarette use (yes vs. no) as covariates. The second variable listed for each covariate serves as the reference variable for all regression models. Odds ratios and 95% confidence intervals for the receipt of anti-MRSA or antipseudomonal guideline-concordant antibiotic therapy were calculated using the outputs of the multivariate models. Patients who were classified as “Race and Ethnicity: Other” or did not provide an indication for any of the above factors were excluded from multivariable logistic regression analysis due to small sample size. The alpha level for significance was 0.05.
Regulatory and ethics
All data collection, analysis, and dissemination were conducted in accordance with the All of Us Data User Code of Conduct, Data and Statistics Dissemination Policy, Publication and Presentation Policy, Policy on Stigmatizing Research, and all other policies therein [34–36]. Inclusion in this study was not based on age, sex, race, or ethnicity. Institutional Review Board approval was obtained from the University of Texas Health Science Center at San Antonio Institutional Review Board (IRB) and Office of Clinical Research (OCR) before beginning this study (Protocol number: 20220682EX).
Results
A total of 336 patients with CABP due to MRSA (152) or P. aeruginosa (184) met study criteria (Figure 1).
Figure 1. Study inclusion flowchart.
Methicillin-Resistant Staphylococcus aureus
Guideline-concordant antibiotic therapy was prescribed to 70% of CABP-MRSA patients; all were prescribed vancomycin (98%) and/or linezolid (n ≤ 20). Other anti-MRSA agents prescribed include ceftroline (n ≤ 20) and tigecycline (n ≤ 20). No patients received tedizolid. We found no significant differences in baseline characteristics between the guideline-concordant and discordant groups (Table 1). There were no significant associations in the logistic regression analysis of patient factors and guideline-concordant anti-MRSA prescribing (Table 2).
Table 1. Baseline characteristics for CABP-MRSA patients prescribed guideline-concordant and discordant antibiotics
Bold italics indicates statistically significant findings.
Counts of less than or equal to 20 are displayed as n≤20 to comply with the All of Us data and statistic dissemination policy [34].
The actual n was used for statistical comparisons. CABP: Community-acquired bacterial pneumonia. MRSA: Methicillin-resistant Staphylococcus aureus.
Table 2. Logistic regression analysis of patient factors and guideline-concordant anti-MRSA prescribing
Bold italics indicates statistically significant findings.
Patients who were classified as Race and Ethnicity: Other or did not provide an indication for any of the above categories were excluded from logistic regression of Patient Factors and Guideline-Concordant Anti-MRSA Prescribing (total n = 23) due to small sample size.
The second variable listed is the reference group used in this model.
MRSA: Methicillin-resistant Staphylococcus aureus, NHB: Non-Hispanic Black, NHW: Non-Hispanic White.
Pseudomonas aeruginosa
Guideline-concordant antibiotic therapy was prescribed to 57% CABP-P. aeruginosa patients; all were prescribed cefepime (33%), piperacillin-tazobactam (21%), ceftazidime (n ≤ 20), meropenem (n ≤ 20), aztreonam (n ≤ 20), and/or imipenem (n ≤ 20). Guideline-concordant antibiotic therapy patients had fewer NHB patients (p = 0.04) (Table 3). In the multivariate logistic regression analysis of patient factors and guideline-concordant antipseudomonal prescribing, NHB patients were less likely to receive guideline-concordant antibiotic therapy when compared to NHW patients (OR 0.30, 95% CI 0.12 –0.75) (Table 4).
Table 3. Baseline characteristics for CABP-Pseudomonas aeruginosa patients prescribed guideline-concordant and discordant antibiotics
Bold italics indicates statistically significant findings.
Counts of less than or equal to 20 are displayed as n ≤ 20 to comply with the All of Us data and statistic dissemination policy [34].
The actual n was used for statistical comparisons. CABP: Community-acquired bacterial pneumonia.
Table 4. Logistic regression analysis of patient factors and guideline-concordant antipseudomonal prescribing
Bold italics indicates statistically significant findings.
Patients who were classified as Race and Ethnicity: Other or did not provide an indication for any of the above categories were excluded from logistic regression of Patient Factors and Guideline-Concordant Antipseudomonal Prescribing (total n = 23) due to small sample size.
The second variable listed is the reference group used in this model.
P. aeruginosa: Pseudomonas aeruginosa. NHB: Non-Hispanic Black. NHW: Non-Hispanic White.
Discussion
The present study was the first to evaluate antibiotic prescribing trends for inpatients with CABP due to MRSA or P. aeruginosa in the All of Us database. Of the patients admitted to the hospital with CABP due to MRSA, nearly three-quarters (70%) were prescribed guideline-concordant antibiotic therapy with no observed differences based on patient factors. For patients admitted to the hospital due to CABP due to P. aeruginosa, 57% received guideline-concordant antibiotic therapy and NHB patients were less likely to receive antipseudomonal guideline-concordant antibiotics than NHW patients.
Health disparity in pneumonia
Previously, our research team investigated differences in guideline-concordant antibiotic prescribing for patients admitted to the hospital with general CABP in the All of Us database. Of the 1,608 patients included, less than half (45%) received guideline-concordant antibiotic therapy for CABP [Reference Gilmore, Lee, Schmidt and Frei37]. Further, we found that NHB patients were 36% less likely than NHW patients, and Hispanic patients were 34% more likely than NHW patients to receive guideline-concordant antibiotic therapy for the treatment of CABP [Reference Gilmore, Lee, Schmidt and Frei37]. Therefore, our findings in these two studies are consistent in a mixed CABP population (all pathogens) and a CABP-P. aeruginosa population.
In addition to the above study, a study of 5,820 inpatients with pneumonia by Evans et al. identified disparities based on patient race and ethnicity [Reference Evans, Fortin-Leung and Kumar28]. They found that younger non-Hispanic Black patients were less likely to receive anti-pseudomonal antibiotics than any other race group considered [Reference Evans, Fortin-Leung and Kumar28]. They also found no difference in receipt of anti-MRSA therapy by race [Reference Evans, Fortin-Leung and Kumar28]. Both of these findings are consistent with our results. Other studies of general CABP report no difference by race or ethnicity in receipt of guideline-concordant antibiotic therapy, both in the general inpatient ward and ICU, which is different than our results [Reference Frei, Mortensen and Copeland38,Reference Mortensen, Cornell and Whittle39]. Direct comparison between these studies and ours are difficult due to differences in patient populations and topics. Nevertheless, both studies still report other variations in care and health outcomes in CABP treatment that are based on patient factors, which is consistent with our findings [Reference Frei, Mortensen and Copeland38,Reference Mortensen, Cornell and Whittle39].
Health disparity in other disease states
Disparity in antibiotic prescribing based upon patient factors has been well documented in settings other than pneumonia [Reference Young, Strey and Lee2,Reference Kim, Kabbani and Dube3]. In one scoping review of health equity and antibiotic prescribing in the United States from 2000–2022, Kim et al. found that Black patients were less likely to receive antibiotics when compared to White patients, which is consistent with our results [Reference Kim, Kabbani and Dube3]. In a study evaluating health disparity in the ambulatory care setting, Young et al. found that Black and Hispanic patients had the highest rates of overall antibiotic prescribing and inappropriate antibiotic prescribing, and that the impact of patient factors on antibiotic prescribing may be additive [Reference Young, Strey and Lee2]. The cause of these disparities remains unclear, but it is likely multifactorial. Additional studies on the drivers of health inequity are needed.
Strengths
CABP is a significant contributor to inpatient mortality and rising hospital costs worldwide. The selection of proper antibiotic therapy, which is essential for improving clinical outcomes, is a complex issue further complicated by the emergence of drug-resistant pathogens. Evidence regarding guideline-concordant antibiotic prescribing for CABP due to drug-resistant pathogens like MRSA or P. aeruginosa is extremely limited, even more so for health disparity in prescribing practices. The present study is the first to evaluate guideline-concordant prescribing for inpatients with CABP due to MRSA or P. aeruginosa. Furthermore, clinical practice guidelines are often used to develop hospital policies and workflows. Evaluation of current guidelines and their use is necessary to ensure timely administration of therapy, selection of effective therapy, lower healthcare costs, and improve patient outcomes. Identification of differences in prescribing practices is the first step in understanding them and generating future intervention-based studies investigating health disparity, novel treatment strategies, and optimization of current treatment delivery algorithms.
Limitations
Due to its observational design, this study has some potential limitations, including limited generalizability to the entirety of the US, lack of local microbiological data, lack of disease severity data, inability to display counts of ≤20, inability to control for variations in decision workflows between institutions and individual providers, inability to evaluate health outcomes, and potential selection bias due to nonrandomization of study arms. Additionally, this study was subject to a small sample size due to specific inclusion criteria and reliance on billing codes of documented causative pathogens. Thus, it may contain potential type 2 errors in both the MRSA and P. aeruginosa populations and may be underpowered. To limit study inclusion to only patients with CABP, rather than HAP, only pneumonia cases diagnosed within the first 48 hours of an inpatient admission were included; however, some HAP patients might have still been included.
To overcome the lack of local microbiological data, this study only considered pneumonia cases that had been attributed to MRSA or P. aeruginosa and coded as such using SNOMED, ICD-9, or ICD-10 codes. However, using pathogen-specific billing codes likely underestimated the true prevalence of these pathogens and may have introduced selection bias, as cases might be from providers who specialized in these pathogens. Further, pathogen-specific billing codes are rarely used in practice [Reference Higgins, Deshpande and Zilberberg40,Reference Skull, Andrews and Byrnes41]. One large cross-sectional study of patients hospitalized with pneumonia found that pathogen-specific ICD-9 codes appear to have limited sensitivity but high specificity [Reference Higgins, Deshpande and Zilberberg40]. Another study examining the validity using ICD-10 codes to identify hospitalized pneumonia found that ICD-10 codes are useful for identifying all-cause pneumonia but less helpful in identifying subtypes of pneumonia [Reference Skull, Andrews and Byrnes41].
To ensure adequate analysis of prescribed antibiotics, all inpatient antibiotics prescribed within 48 hours of the pneumonia diagnosis were included in this study; however, duration of therapy was not included in this study due to incompleteness of duration data. Additionally, the definition of guideline-concordant used in this study was based on the receipt of at least one dose guideline-concordant antibiotics in the first 48 hours. However, we did not reclassify guideline-concordant patients if they also received guideline-discordant antibiotic therapy at any time, which may contribute to underestimation of guideline-discordant prescribing. Despite these limitations, this work provides valuable insights into prescribing practices for CABP due to MRSA and P. aeruginosa and can inform future intervention-based studies.
Conclusions
In the All of Us database, 70% of CABP-MRSA and 57% of CABP-P. aeruginosa patients were prescribed guideline-concordant antibiotic therapy. Cases were defined using billing codes of documented causative pathogens. Race was independently predictive of guideline-concordant antibiotic therapy for patients with CABP-P. aeruginosa, but not CABP-MRSA. NHB patients with CABP due to P. aeruginosa were less likely to be prescribed guideline-concordant antibiotic therapy when compared to NHW inpatients. Evaluation of the implementation of current guidelines is necessary to ensure timely administration of effective therapy and improve patient outcomes. Variations of care and health disparity in guideline-concordant antibiotic therapy prescribing for CABP due to drug-resistant pathogens lead to suboptimal patient outcomes. Identification of these disparities is essential in minimizing them and promoting health equity for all patients.
Acknowledgements
This research would not be possible without the All of Us Research Program and its participants.
Author contributions
Corbyn M. Gilmore: Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Project administration, Validation, Visualization, Writing-original draft, Writing-review & editing; Adriana Vargus: Formal analysis, Visualization, Writing-review & editing; Grace C. Lee: Formal analysis, Visualization, Writing-review & editing; Susanne Schmidt: Formal analysis, Visualization, Writing-review & editing; Kelly R. Reveles: Formal analysis, Visualization, Writing-review & editing; Carlos A. Alvarez: Formal analysis, Visualization, Writing-review & editing; Christopher R. Frei: Conceptualization, Formal analysis, Methodology, Resources, Supervision, Validation, Visualization, Writing-original draft, Writing-review & editing.
Funding statement
No funding was received for this study. CMG, GCL, KRR, SS, and CRF were partially supported by the National Center for Advancing Translational Sciences (NCATS), National Institutes of Health (NIH), through Grant UM1 TR004538. The views expressed in this article are those of the authors and do not necessarily represent the views of the Department of Veterans Affairs, the National Institutes of Health, or the authors’ affiliated institutions.
Competing interests
CRF’s institution has received research grants from Amgen and AstraZeneca, in the past three years, to conduct research in the areas of cancer and infectious diseases. All other authors declare no conflicts of interest.
Open access
