Abstract:
Antimicrobial resistance (AMR) represents an urgent global health crisis exacerbated by the frequent empirical use of broad-spectrum antibiotics. AMR is exacerbated by inherent delays in obtaining culture results and antimicrobial susceptibility data after sample collection. In this study, we developed and validated Machine Learning (ML) models using routinely collected EHR data from inpatient and outpatient encounters to predict antibiotic resistance at the time of blood, urine or respiratory bacterial culture collection. The models demonstrated robust predictive accuracy, particularly in inpatient settings where clinical data was more consistently available. Notably, the model independently identified patterns that predict resistance, similar to how a clinician would attempt to predict resistance using prior culture and susceptibility data combined with their clinical training and knowledge of microbiological resistance patterns. Integrating these predictive tools into clinical workflows could significantly enhance empirical antibiotic selection, reduce unnecessary broad-spectrum antibiotic use, and meaningfully advance antimicrobial stewardship efforts.
Introduction:
The introduction of antibiotics marked one of the most transformative medical breakthroughs of the 20th century, profoundly reducing mortality from bacterial infections and extending the average human lifespan by approximately 20 years 1. Antibiotics have allowed clinicians to effectively manage previously life-threatening infections, revolutionizing healthcare and establishing these agents among the most frequently prescribed medications today, with over 236 million prescriptions annually in the United States alone 2,3. However, widespread antibiotic use has led to a rapid emergence and dissemination of antimicrobial resistance (AMR), threatening the gains achieved over the past century and posing an urgent global health crisis 4,5. In fact, AMR infections are projected to cause over 10 million deaths annually worldwide by 2050 6.
Unfortunately, the discovery and development of new antibiotics has significantly declined since the antibiotic “golden age.” Repeated rediscovery of existing antibiotics, combined with failed attempts at developing effective synthetic alternatives, have discouraged major pharmaceutical companies from investing in antibiotic discovery due to limited profitability compared with medications for chronic conditions 1. Consequently, there has been increasing attention in antibiotic stewardship to combat antimicrobial resistance through the responsible use and preservation of existing antibiotics,7. The gravity of the antimicrobial resistance crisis is recognized globally and has led to comprehensive initiatives such as the U.S. National Action Plan for Combating Antibiotic-Resistant Bacteria 8. The Centers for Disease Control and Prevention (CDC) and the World Health Organization (WHO) have strongly advocated for antimicrobial stewardship programs (ASPs), identifying these programs as the cornerstone strategy to slow resistance development, optimize patient outcomes, and safeguard existing antibiotic therapies 7, 9. Stewardship practices, including evidence-based prescribing guidelines, antimicrobial de-escalation, and surveillance of local resistance patterns, seek to balance effective treatment against the risk of resistance emergence 10.
Yet clinicians frequently face uncertainty in empiric antibiotic prescribing due to the inherent delays in receipt of culture and susceptibility results. This uncertainty can lead clinicians to prescribe broader-spectrum antibiotics (e.g., piperacillin-tazobactam, cefepime, or carbapenems) out of concern for under-treatment, exacerbating selective pressures that drive AMR. Currently available stewardship resources, such as institutional antibiograms and clinical guidelines, provide essential yet generalized recommendations based on aggregate data, typically lacking patient-specific factors known to influence susceptibility, including prior antibiotic exposures, co-morbidities, demographics, and socioeconomic conditions 11. The inability of existing tools to incorporate such personalized data frequently results in suboptimal empiric antibiotic selection, further driving unnecessary broad-spectrum antibiotic use and accelerating resistance.
Recent advances in artificial intelligence (AI) and machine learning (ML) offer promising avenues to enhance antimicrobial stewardship through more precise, individualized antibiotic prescribing 12. Recent studies have demonstrated the promise of data-driven predictive approaches, including the UTI SmartSet (UTIS), an EHR-integrated tool designed to reduce antibiotic mismatches for urinary tract infections, and machine learning algorithms that predict antimicrobial resistance directly from MALDI-TOF mass spectrometry profiles 13, 14. While promising, many of these approaches remain limited by narrow clinical contexts and lack of generalizability.
In this study, we expand upon previous efforts by developing and evaluating machine learning models using routinely collected EHR data to predict antimicrobial resistance patterns in both inpatient and outpatient settings for a range of commonly prescribed antibiotics 15. Our approach integrates comprehensive patient-specific variables, including clinical, demographic, and socioeconomic factors, and applies explainable ML methods, such as SHapley Additive exPlanations (SHAP), to interpret model-derived predictions. Through this work, we aim to provide clinicians with practical, scalable, predictive model, thereby enhancing targeted antibiotic selection, improving antimicrobial stewardship practices by incorporating previously overlooked factors associated with resistance.
Methods:
Study cohort:
For this analysis, we utilized EHR data from 283,715 patients aged 18 years and older who had a positive blood, urine, or respiratory culture collected during hospitalization (inpatient) or an outpatient visit, which are the most ordered cultures in both settings. To ensure data integrity, we excluded culture orders if the patient had a positive result within the preceding 14 days, reducing the likelihood of capturing repeat or persistent infections from the same episode.
The primary unit of analysis in this study was patient-encounter-order, representing each unique culture order linked to a patient encounter within the hospital. All data used in this study is publicly available 16. Patients included in this dataset were seen at Stanford’s quaternary hospital and affiliated community hospital between 1999 and 2024. Patient characteristics are provided in Table 1. The outcome of each model indicates whether the culture organism is sensitive to antibiotics. Our clinical team carefully reviewed and curated a list of commonly prescribed antibiotics. Table 2 presents these antibiotics along with their corresponding resistance prevalence.
Table 1.
Study cohort patients’ characteristics.
| Patient Characteristics | Overall | Positive culture | Negative culture |
|---|---|---|---|
| N patients | 283715 | 67798 | 257661 |
| Demographics | |||
| Age (median) | 55-64 years | 55-64 years | 55-64 years |
| Sex (%) | |||
| Male | 93763 (33%) | 18089(26.7%) | 88409(34.3%) |
| Female | 189864(66.9%) | 49677(73.2%) | 169191(65.7%) |
| Race (%) | |||
| White | 144952(51.1%) | 36437(53.8%) | 131795(51.2%) |
| Asian | 48588 (17.1%) | 10765 (15.9%) | 44281(17.2%) |
| African American | 12436 (4.4%) | 2617(3.9%) | 11539 (4.5%) |
| Other | 77739(27.4%) | 17979 (26.5%) | 70046 (27.2%) |
| Ethnicity (%) | |||
| Hispanic or Latino | 46113 (16.2%) | 11086 (16.3%) | 41999 (16.3%) |
| Non-Hispanic | 218897 (77.2%) | 53138(78.3%) | 199141(77.3%) |
| Unkown | 18705(6.6%) | 3574(5.3%) | 16521(6.4%) |
| Ordering Mode (%) | |||
| inpatients | 172234(60.71%) | 43780(65.57%) | 154028(59.78%) |
| outpatients | 140381(49.48%) | 28480(42.01%) | 126282(49.01%) |
Table 2.
Commonly Prescribed Antibiotics and Their Resistance Prevalence in the Dataset
| Antibiotic | Inpatient Resistance Prevalence (%) | Outpatient Resistance Prevalence (%) | Antibiotic | Inpatient Resistance Prevalence (%) | Outpatient Resistance Prevalence (%) |
|---|---|---|---|---|---|
| Cefepime | 7% | 8% | Ampicillin/Sul bactam | 37% | 38% |
| Ampicillin | 53% | 47% | Cefazolin | 24% | 13% |
| Gentamicin | 9% | 8% | Cefoxitin | 18% | 6% |
| Ceftazidime | 9% | 7% | Tobramycin | 10% | 7% |
| Ceftriaxone | 12% | 9% | Tetracycline | 45% | 40% |
| Ciprofloxacin | 22% | 19% | Levofloxacin | 23% | 18% |
| Nitrofurantoin | 13% | 11% | Amoxicillin/ Clavulanic Acid | 16% | 8% |
| Moxifloxacin | 45% | 29% | Penicillin | 45% | 47% |
| Doxycycline | 74% | 64% | Imipenem | 11% | 24% |
| Cefotaxime | 10% | 8% | Clindamycin | - | 41% |
| Cefpodoxime | 10% | 7% | Erythromycin | - | 51% |
Model development
For this analysis, we focused exclusively on the first culture order for an inpatient stay (i.e., subsequent cultures in the same hospitalization were excluded). A separate predictive model was developed for each antibiotic for each setting, and the corresponding data were split chronologically into training (pre-2020), validation (2020 and 2021), and test sets (2022 onwards) to maintain temporal integrity. To account for temporal shifts in clinical practice, we assessed covariate drift between pre- and post-pandemic periods by measuring Kullback-Leibler (KL) divergence and found most features remained stable; however, increased resistance to certain classes of antibiotics during the COVID-19 era likely reflects broader use of antibiotics and evolving prescribing patterns.
The predictive models incorporated a range of clinical and demographic features, including patient demographics (age, sex, etc.), laboratory results from the past two weeks, vital signs recorded within the last 48 hours, patient location at the time of culture order (ED, ICU, or Ward), Area Deprivation Index (ADI) score as a socioeconomic factor, and recent healthcare exposures in the past six months (e.g., visiting a nursing home, prior procedure, history of infected organisms, previous antimicrobial resistance, prior medication history, and antibiotic class).
Feature selection for model development was performed using Recursive Feature Elimination (RFE)17 with XGBoost18, a widely used Gradient Boosting Machine (GBM) implementation. XGBoost was selected for its efficiency in handling large datasets, inherent ability to manage missing values, and built-in regularization techniques that improve model robustness and mitigate overfitting, as demonstrated in prior studies. In our previous work19, we compared XGBoost to Logistic Regression on a subset of high-prevalence antibiotics. While Logistic Regression demonstrated consistent trends, it generally yielded slightly lower AUCs compared to XGBoost.
RFE is an iterative feature selection method that recursively removes the least important features, improving model interpretability and performance. We performed a grid search to identify the optimal number of features for each antibiotic, selecting the configuration that maximized Area Under the Curve (AUC) performance on the validation set. For hyperparameter optimization, we performed a randomized five-fold cross-validation search, using AUC as the scoring metric. To address class imbalance, we applied class weighting (scale_pos_weight) and incorporated Lasso and Ridge regularization to enhance model generalizability. The same procedure was applied to both inpatient and outpatient settings. Our analysis revealed that a patient’s history of microbial resistance within the last six months was a highly informative feature for prediction. However, since this information may not be available for new patients or those in the emergency department (ED) at the time of culture order, we also developed separate models that excludes prior microbial resistance as a feature.
To interpret model and features’ contribution, we utilized SHAP (SHapley Additive exPlanations) values. SHAP values estimate the contribution of each input feature to individual predictions. We computed SHAP values using the TreeExplainer from the SHAP Python library on the final models, and report the top contributing features for resistance predictions.
Results:
Table 3 presents the model’s performance in predicting antibiotic susceptibility for both inpatients and outpatients. We report the AUC (Area Under the Curve), positive predictive value (PPV), and negative predictive value (NPV). We also assess each model specificity at a threshold where sensitivity is at least 80%. Additionally, we evaluate the model’s performance when incorporating a patient’s microbial resistance history from the past six months into the feature set.
To interpret feature importance, we utilized SHAP values. For ease of interpretation, we adjusted the labels in the SHAP analysis so that the positive class corresponds to predicted resistance to Cefepime. Figure 1 illustrates the aggregated feature importance for Cefepime resistance prediction in the inpatient population using SHAP, comparing models with and without prior antimicrobial resistance history. Our SHAP results indicate that a history of Escherichia coli (E. coli) resistance to Cefepime is the strongest predictor of future resistance to Cefepime. Additionally, prior Pseudomonas aeruginosa resistance to Cefepime is also a significant predictor, aligning with clinical observations that P. aeruginosa often develops multidrug resistance, limiting treatment options. We also found that a history of receiving antibiotics particularly from the Macrolide-Lincosamide and Nitroimidazole classes is associated with an increased likelihood of future Cefepime resistance. While these antibiotic classes do not directly select for beta-lactam resistance, their use may indicate prior infections requiring broad-spectrum antibiotics, which can drive resistance through microbial selection pressure.
Figure 1.
SHAP values for predicting inpatient resistance to Cefepime in a model (a) without prior antimicrobial resistance history and (b) with historical Resistance data
Among socioeconomic factors, a higher Area Deprivation Index (ADI) score was associated with an increased probability of resistance to Cefepime. This finding suggests that social determinants of health, such as limited healthcare access, delayed care, and increased antibiotic exposure in certain communities, may contribute to antimicrobial resistance patterns.
From a clinical perspective, we observed that elevated heart rate (a potential marker of sepsis or systemic infection), high platelet count (possibly indicating inflammation or infection-related thrombocytosis), and increased blood urea nitrogen (BUN, which may reflect kidney dysfunction or severe infection) were all associated with higher resistance probabilities. Conversely, a lower lymphocyte count was linked to increased resistance risk, consistent with the idea that immune suppression or dysregulation may predispose patients to resistant infections.
Discussion:
Our study demonstrates that machine learning models can effectively predict antibiotic susceptibility patterns using routinely collected electronic health record (EHR) data. By stratifying our analysis between inpatient and outpatient cohorts, we observed some differences in predictive accuracy, likely driven by variations in data completeness and quality. Additionally, we found that prior resistance to antibiotics, particularly Cefepime, Ceftriaxone, Ciprofloxacin, and Levofloxacin, was highly predictive of future resistance, reinforcing the importance of historical antimicrobial susceptibility in forecasting resistance patterns.
Our SHAP-based feature analysis revealed that the model successfully identified key microbiological and clinical predictors of resistance. Prior resistance to Escherichia coli (E. coli) and Pseudomonas aeruginosa were among the most influential predictors of future Cefepime resistance, which is consistent with known resistance mechanisms in these organisms. The model also captured the predictive value of prior exposure to certain antibiotic classes, particularly Macrolide-Lincosamide and Nitroimidazole, suggesting that past antimicrobial treatment may serve as a proxy for broader resistance trends. Beyond microbiological predictors, our findings highlight the role of clinical and socioeconomic factors in shaping antibiotic resistance patterns. A higher Area Deprivation Index (ADI) score was associated with an increased likelihood of resistance to cefepime, suggesting that disparities in healthcare access and antibiotic prescribing patterns may contribute to resistance development. Additionally, we observed that physiological markers, including elevated heart rate, increased platelet count, and higher blood urea nitrogen (BUN), were associated with greater resistance risk. These factors may reflect underlying infection severity or systemic inflammatory responses, which could influence both microbial selection pressure and treatment efficacy. While the SHAP-based feature importance analysis did not reveal novel or unexpected predictors, it emphasized key variables (such as ADI score, prior organism-specific resistance, and physiologic markers) that may be underrecognized in real-time clinical decision-making. For example, although socioeconomic disparities are known to influence resistance patterns, the prominence of ADI in our model highlights the need to more routinely incorporate social determinants into antimicrobial risk assessments. Similarly, prior resistance to specific antibiotics and organisms remains a strong predictor, reinforcing the importance of reviewing a patient’s antimicrobial history when tailoring empiric therapy.
Our findings underscore the importance of integrating EHR data to improve predictive accuracy in antimicrobial resistance models. While prior microbial susceptibility data remained the most informative predictor, demographic, clinical, and socioeconomic factors also provided valuable insights. Future work should focus on refining models to improve generalizability across diverse patient populations, data modality and care settings.
Our study has several limitations. First, data completeness remains a challenge, particularly in outpatient settings where documentation of vital signs and laboratory results is often incomplete. Second, while prior antibiotic resistance history significantly improved predictive accuracy, it is not always available in real-world clinical settings. Our findings suggest that alternative clinical and demographic factors can still provide meaningful predictive value in the absence of prior resistance data, but further refinement is needed to optimize predictions under these constraints. Additionally, our reliance on structured EHR data limits the model’s ability to capture nuanced clinical information from provider notes. Natural language processing (NLP) techniques could enhance predictive accuracy by extracting relevant clinical details, such as infection severity, symptom progression, or unstructured documentation of prior resistance patterns. Future work should explore how NLP-driven feature extraction could complement structured data in improving predictive performance.
Finally, our study is subject to inherent biases in EHR-derived data, particularly those related to healthcare process signals. The presence or timing of an order may reflect clinician suspicion and decision-making, which can inadvertently serve as proxies for underlying infection risk. This can cause the model to learn patterns associated with provider behavior rather than patient physiology alone, potentially limiting interpretability and portability. Future work should consider methods to disentangle patient-level risk from healthcare process signals, such as causal inference frameworks or explicitly modeling provider behavior.
External validation across diverse healthcare settings is necessary to assess the model’s robustness and generalizability. Although our study utilized data from multiple hospitals, differences in prescribing patterns, laboratory workflows, and patient populations may influence model performance when applied in new clinical environments. As a next step, we are conducting external validation at two additional healthcare institutions to assess generalizability.
While clinical implementation was beyond the scope of this study, future work will focus on integrating the model into EHR systems to support real-time decision-making. Key challenges (e.g., data availability, clinician trust, institutional variability, and performance monitoring) will need to be addressed to enable safe and effective deployment.
Conclusion:
Our study highlights the potential of machine learning algorithms to harness electronic health record data for accurately predicting antimicrobial resistance patterns. By identifying clinically relevant resistance trends, akin to the insights of infectious disease specialists, our approach has the potential to enhance clinical decision-making by guiding more precise antibiotic selection. Moreover, our model extends beyond traditional susceptibility data by incorporating demographic and socioeconomic factors—often underutilized in clinical practice—fostering a more comprehensive and equitable approach to resistance prediction. This integration of predictive tools into clinical workflows could optimize empirical antibiotic prescribing, minimize unnecessary broad-spectrum antibiotic use, and support antimicrobial stewardship efforts, ultimately contributing to the fight against antibiotic resistance.
Figures & Tables
Table 3.
Performance metrics for antibiotic-specific susceptibility predictive models in inpatient and outpatient settings, including ROC-AUC with and without historical susceptibility.
| Inpatient | Outpatient | |||
|---|---|---|---|---|
| Antibiotic-Specific Model | ROC-AUC with historical susceptibility (Resistance) data | ROC-AUC without historical susceptibility (Resistance) data | ROC-AUC with historical susceptibility (Resistance) data | ROC-AUC without historical susceptibility (Resistance) data |
| Cefepime | 0.74 | 0.66 | 0.74 | 0.59 |
| Ampicillin | 0.62 | 0.61 | 0.62 | 0.58 |
| Gentamicin | 0.58 | 0.57 | 0.60 | 0.57 |
| Ceftazidime | 0.69 | 0.64 | 0.69 | 0.64 |
| Ceftriaxone | 0.70 | 0.66 | 0.71 | 0.61 |
| Ciprofloxacin | 0.68 | 0.64 | 0.68 | 0.63 |
| Clindamycin | - | - | 0.64 | 0.64 |
| Nitrofurantoin | 0.68 | 0.68 | 0.68 | 0.66 |
| Ampicillin/Sulbacta m | 0.58 | 0.55 | 0.55 | 0.50 |
| Cefazolin | 0.71 | 0.67 | 0.71 | 0.62 |
| Cefoxitin | 0.67 | 0.67 | 0.67 | 0.70 |
| Tobramycin | 0.63 | 0.57 | 0.65 | 0.63 |
| Tetracycline | 0.55 | 0.53 | 0.57 | 0.57 |
| Levofloxacin | 0.69 | 0.63 | 0.70 | 0.66 |
| Amoxicillin/Clavula nic Acid | 0.66 | 0.64 | 0.68 | 0.63 |
| Penicillin | 0.53 | 0.53 | 0.57 | 0.51 |
| Cefotaxime | 0.65 | 0.65 | 0.64 | 0.64 |
| Cefpodoxime | 0.66 | 0.66 | 0.62 | 0.62 |
| Doxycycline | 0.70 | 0.62 | 0.80 | 0.58 |
| Erythromycin | - | - | 0.65 | 0.64 |
| Moxifloxacin | 0.72 | 0.70 | 0.74 | 0.67 |
| Imipenem | 0.71 | 0.68 | 0.84 | 0.63 |
References:
- 1.Hutchings MI, Truman AW, Wilkinson B. Antibiotics: past, present and future. Curr Opin Microbiol. 2019;51:72–80. doi: 10.1016/j.mib.2019.10.008. doi:10.1016/j.mib.2019.10.008. [DOI] [PubMed] [Google Scholar]
- 2.Teillant A, Gandra S, Barter D, Morgan DJ, Laxminarayan R. Potential burden of antibiotic resistance on surgery and cancer chemotherapy antibiotic prophylaxis in the USA: a literature review and modelling study. Lancet Infect Dis. 2015;15(12):1429–1437. doi: 10.1016/S1473-3099(15)00270-4. doi:10.1016/S1473-3099(15)00270-4. [DOI] [PubMed] [Google Scholar]
- 3.Centers for Disease Control and Prevention (CDC) Antibiotic prescribing and use in doctor’s offices. CDC. website. Published [Year]. Accessed [Date]. Available from: https://www.cdc.gov/antibiotic-use/hcp/data-research/antibiotic-prescribing.html.
- 4.Hawn MT, Richman JS, Vick CC, Deierhoi RJ, Graham LA, Henderson WG. Timing of surgical antibiotic prophylaxis and the risk of surgical site infection. JAMA Surg. 2013;148(7):649–657. doi: 10.1001/jamasurg.2013.134. doi:10.1001/jamasurg.2013.134. [DOI] [PubMed] [Google Scholar]
- 5.Teillant A, Gandra S, Barter D, Morgan DJ, Laxminarayan R. Potential burden of antibiotic resistance on surgery and cancer chemotherapy antibiotic prophylaxis in the USA: a literature review and modelling study. Lancet Infect Dis. 2015;15(12):1429–1437. doi: 10.1016/S1473-3099(15)00270-4. doi:10.1016/S1473-3099(15)00270-4. [DOI] [PubMed] [Google Scholar]
- 6.Naghavi M, Vollset SE, Ikuta KS, Swetschinski LR, Gray AP, Wool EE, et al. Global burden of bacterial antimicrobial resistance 1990–2021: a systematic analysis with forecasts to 2050. Lancet. 2024;404(10459):1199–1226. doi: 10.1016/S0140-6736(24)01867-1. doi:10.1016/S0140-6736(24)01867-1. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Centers for Disease Control and Prevention (CDC) Core elements of outpatient antibiotic stewardship. CDC. website. Available from: https://www.cdc.gov/antibiotic-use/hcp/core-elements/index.html.
- 8.U.S. Department of Health and Human Services (HHS) National Action Plan for Combating Antibiotic-Resistant Bacteria, 2020-2025. Available from: https://aspe.hhs.gov/reports/national-action-plan-combating-antibiotic-resistant-bacteria-2020-2025.
- 9.World Health Organization (WHO) Promoting antimicrobial stewardship to tackle antimicrobial resistance. WHO Europe. website. Available from: https://www.who.int/europe/activities/promoting-antimicrobial-stewardship-to-tackle-antimicrobial-resistance.
- 10.Centers for Disease Control and Prevention (CDC) Core elements of hospital antibiotic stewardship programs. CDC. website. Available from: https://www.cdc.gov/antibiotic-use/hcp/core-elements/hospital.html.
- 11.Isaac T, Zheng J, Jha A. Use of UpToDate and outcomes in US hospitals. J Hosp Med. 2012;7(2):85–90. doi: 10.1002/jhm.944. doi:10.1002/jhm.944. [DOI] [PubMed] [Google Scholar]
- 12.Sakagianni A, et al. Using machine learning to predict antimicrobial resistance—A literature review. Antibiotics. 2023;12(3):452. doi: 10.3390/antibiotics12030452. doi:10.3390/antibiotics12030452. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Shapiro Ben David S, Romano R, Rahamim-Cohen D, et al. AI-driven decision support reduces antibiotic mismatches and inappropriate use in outpatient urinary tract infections. npj Digit Med. 2025;8:61. doi: 10.1038/s41746-024-01400-5. doi:10.1038/s41746-024-01400-5. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Weis C, Cuénod A, Rieck B, et al. Direct antimicrobial resistance prediction from clinical MALDI-TOF mass spectra using machine learning. Nat Med. 2022;28:164–174. doi: 10.1038/s41591-021-01619-9. doi:10.1038/s41591-021-01619-9. [DOI] [PubMed] [Google Scholar]
- 15.Corbin CK, Medford RJ, Osei K, Chen JH. Personalized Antibiograms: Machine Learning for Precision Selection of Empiric Antibiotics. AMIA Jt Summits Transl Sci Proc. 2020 May 30;2020:108–115. PMID: 32477629; PMCID: PMC7233062. [PMC free article] [PubMed] [Google Scholar]
- 16.Nateghi Haredasht F, Amrollahi F, Maddali M, Marshall N, Ma SP, Cooper LN, et al. Antibiotic Resistance Microbiology Dataset (ARMD): A de-identified resource for studying antimicrobial resistance using electronic health records. arXiv preprint arXiv:2503.07664. Published 2025. Available from: https://arxiv.org/abs/2503.07664.
- 17.Guyon I, Weston J, Barnhill S, Vapnik V. Gene selection for cancer classification using support vector machines. Mach Learn. 2002;46(1):389–422. doi: 10.1023/A:1012487302797. [Google Scholar]
- 18.Chen T, Guestrin C. XGBoost: A scalable tree boosting system. Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. 2016:p. 785–94. doi:10.1145/2939672.2939785. [Google Scholar]
- 19.Haredasht FN, Maddali MV, Ma SP, Chang A, Kim GYE, Banaei N, Deresinski S, Goldstein MK, Asch SM, Chen JH. Enhancing Antibiotic Stewardship: A Machine Learning Approach to Predicting Antibiotic Resistance in Inpatient Care. AMIA Annu Symp Proc. 2025 May 22;2024:857–864. PMID: 40417584; PMCID: PMC12099390. [PMC free article] [PubMed] [Google Scholar]