Patient-Specific Predictive Antibiogram in Decision Support for Empiric Antibiotic Treatment

Background: The rising trend of antibiotic resistance imposes a heavy burden on healthcare both clinically and economically (US$55 billion), with 23,000 estimated annual deaths in the United States as well as increased length of stay and morbidity. Machine-learning–based methods have, of late, been used for leveraging patient’s clinical history and demographic information to predict antimicrobial resistance. We developed a machine-learning model ensemble that maximizes the accuracy of such a drug-sensitivity versus resistivity classification system compared to the existing best-practice methods. Methods: We first performed a comprehensive analysis of the association between infecting bacterial species and patient factors, including patient demographics, comorbidities, and certain healthcare-specific features. We leveraged the predictable nature of these complex associations to infer patient-specific antibiotic sensitivities. Various base-learners, including k-NN (k-nearest neighbors) and gradient boosting machine (GBM), were used to train an ensemble model for confident prediction of antimicrobial susceptibilities. Base learner selection and model performance evaluation was performed carefully using a variety of standard metrics, namely accuracy, precision, recall, F1 score, and Cohen κ. Results: For validating the performance on MIMIC-III database harboring deidentified clinical data of 53,423 distinct patient admissions between 2001 and 2012, in the intensive care units (ICUs) of the Beth Israel Deaconess Medical Center in Boston, Massachusetts. From ~11,000 positive cultures, we used 4 major specimen types namely urine, sputum, blood, and pus swab for evaluation of the model performance. Figure 1 shows the receiver operating characteristic (ROC) curves obtained for bloodstream infection cases upon model building and prediction on 70:30 split of the data. We received area under the curve (AUC) values of 0.88, 0.92, 0.92, and 0.94 for urine, sputum, blood, and pus swab samples, respectively. Figure 2 shows the comparative performance of our proposed method as well as some off-the-shelf classification algorithms. Conclusions: Highly accurate, patient-specific predictive antibiogram (PSPA) data can aid clinicians significantly in antibiotic recommendation in ICU, thereby accelerating patient recovery and curbing antimicrobial resistance.Funding: This study was supported by Circle of Life Healthcare Pvt. Ltd.Disclosures: None

Download Full-text

A Pragmatic Machine Learning Model to Predict Carbapenem Resistance

Antimicrobial Agents and Chemotherapy ◽

10.1128/aac.00063-21 ◽

2021 ◽

Author(s):

Ryan J. McGuire ◽

Sean C. Yu ◽

Philip R. O. Payne ◽

Albert M. Lai ◽

M. Cristina Vazquez-Guillamet ◽

...

Keyword(s):

Machine Learning ◽

Predictive Value ◽

Medical Center ◽

Academic Medical Center ◽

Tertiary Care ◽

Carbapenem Resistance ◽

The United States ◽

Gradient Boosting ◽

Negative Case ◽

Extreme Gradient Boosting

Infection caused by carbapenem resistant (CR) organisms is a rising problem in the United States. While the risk factors for antibiotic resistance are well known, there remains a large need for the early identification of antibiotic resistant infections. Using machine learning (ML), we sought to develop a prediction model for carbapenem resistance. All patients >18 years of age admitted to a tertiary-care academic medical center between Jan 1, 2012 and Oct 10, 2017 with ≥1 bacterial culture were eligible for inclusion. All demographic, medication, vital sign, procedure, laboratory, and culture/sensitivity data was extracted from the electronic health record. Organisms were considered CR if a single isolate was reported as intermediate or resistant. CR and non-CR patients were temporally matched to maintain positive/negative case ratio. Extreme gradient boosting was used for model development. In total, 68,472 patients met inclusion criteria with 1,088 CR patients identified. Sixty-seven features were used for predictive modeling. The most important features were number of prior antibiotic days, recent central venous catheter placement, and inpatient surgery. After model training, the area under the receiver operating characteristic curve was 0.846. The sensitivity of the model was 30%, with a positive predictive value (PPV) of 30% and a negative predictive value of 99%. Using readily available clinical data, we were able to create a ML model capable of predicting CR infections at the time of culture collection with a high PPV.

Download Full-text

An Interpretable Early Dynamic Sequential Predictor for Sepsis-Induced Coagulopathy Progression in the Real-World Using Machine Learning

Frontiers in Medicine ◽

10.3389/fmed.2021.775047 ◽

2021 ◽

Vol 8 ◽

Author(s):

Ruixia Cui ◽

Wenbo Hua ◽

Kai Qu ◽

Heran Yang ◽

Yingmu Tong ◽

...

Keyword(s):

Machine Learning ◽

Real World ◽

Time Series Data ◽

Time Window ◽

Medical Center ◽

Characteristic Curve ◽

Series Data ◽

Gradient Boosting ◽

Early Management ◽

Extreme Gradient Boosting

Sepsis-associated coagulation dysfunction greatly increases the mortality of sepsis. Irregular clinical time-series data remains a major challenge for AI medical applications. To early detect and manage sepsis-induced coagulopathy (SIC) and sepsis-associated disseminated intravascular coagulation (DIC), we developed an interpretable real-time sequential warning model toward real-world irregular data. Eight machine learning models including novel algorithms were devised to detect SIC and sepsis-associated DIC 8n (1 ≤ n ≤ 6) hours prior to its onset. Models were developed on Xi'an Jiaotong University Medical College (XJTUMC) and verified on Beth Israel Deaconess Medical Center (BIDMC). A total of 12,154 SIC and 7,878 International Society on Thrombosis and Haemostasis (ISTH) overt-DIC labels were annotated according to the SIC and ISTH overt-DIC scoring systems in train set. The area under the receiver operating characteristic curve (AUROC) were used as model evaluation metrics. The eXtreme Gradient Boosting (XGBoost) model can predict SIC and sepsis-associated DIC events up to 48 h earlier with an AUROC of 0.929 and 0.910, respectively, and even reached 0.973 and 0.955 at 8 h earlier, achieving the highest performance to date. The novel ODE-RNN model achieved continuous prediction at arbitrary time points, and with an AUROC of 0.962 and 0.936 for SIC and DIC predicted 8 h earlier, respectively. In conclusion, our model can predict the sepsis-associated SIC and DIC onset up to 48 h in advance, which helps maximize the time window for early management by physicians.

Download Full-text

Prediction of Acquired Antimicrobial Resistance for Multiple Bacterial Species Using Neural Networks

mSystems ◽

10.1128/msystems.00774-19 ◽

2020 ◽

Vol 5 (1) ◽

Cited By ~ 6

Author(s):

D. Aytan-Aktug ◽

P. T. L. C. Clausen ◽

V. Bortolaia ◽

F. M. Aarestrup ◽

O. Lund

Keyword(s):

Machine Learning ◽

Escherichia Coli ◽

Mycobacterium Tuberculosis ◽

Antimicrobial Resistance ◽

Bacterial Species ◽

Individual Species ◽

Learning Models ◽

Content Type ◽

Multiple Species ◽

Machine Learning Models

ABSTRACT Machine learning has proven to be a powerful method to predict antimicrobial resistance (AMR) without using prior knowledge for selected bacterial species-antimicrobial combinations. To date, only species-specific machine learning models have been developed, and to the best of our knowledge, the inclusion of information from multiple species has not been attempted. The aim of this study was to determine the feasibility of including information from multiple bacterial species to predict AMR for an individual species, since this may make it easier to train and update resistance predictions for multiple species and may lead to improved predictions. Whole-genome sequence data and susceptibility profiles from 3,528 Mycobacterium tuberculosis, 1,694 Escherichia coli, 658 Salmonella enterica, and 1,236 Staphylococcus aureus isolates were included. We developed machine learning models trained by the features of the PointFinder and ResFinder programs detected to predict binary (susceptible/resistant) AMR profiles. We tested four feature representation methods to determine the most efficient way for introducing features into the models. When training the model only on the Mycobacterium tuberculosis isolates, high prediction performances were obtained for the six AMR profiles included. By adding information on ciprofloxacin from the additional 3,588 isolates, there was no reduction in performance for the other antimicrobials but an increased performance for ciprofloxacin AMR profile prediction for Mycobacterium tuberculosis and Escherichia coli. In conclusion, the species-independent models can predict multi-AMR profiles for multiple species without losing any robustness. IMPORTANCE Machine learning is a proven method to predict AMR; however, the performance of any machine learning model depends on the quality of the input data. Therefore, we evaluated different methods of representing information about mutations as well as mobilizable genes, so that the information can serve as input for a robust model. We combined data from multiple bacterial species in order to develop species-independent machine learning models that can predict resistance profiles for multiple antimicrobials and species with high performance.

Download Full-text

Using satellite imagery to understand and promote sustainable development

Science ◽

10.1126/science.abe8628 ◽

2021 ◽

Vol 371 (6535) ◽

pp. eabe8628

Author(s):

Marshall Burke ◽

Anne Driscoll ◽

David B. Lobell ◽

Stefano Ermon

Keyword(s):

Machine Learning ◽

Sustainable Development ◽

Satellite Imagery ◽

Model Building ◽

Model Performance ◽

Training Data ◽

Learning Approaches ◽

Research Directions ◽

Development Outcomes ◽

Research And Policy

Accurate and comprehensive measurements of a range of sustainable development outcomes are fundamental inputs into both research and policy. We synthesize the growing literature that uses satellite imagery to understand these outcomes, with a focus on approaches that combine imagery with machine learning. We quantify the paucity of ground data on key human-related outcomes and the growing abundance and improving resolution (spatial, temporal, and spectral) of satellite imagery. We then review recent machine learning approaches to model-building in the context of scarce and noisy training data, highlighting how this noise often leads to incorrect assessment of model performance. We quantify recent model performance across multiple sustainable development domains, discuss research and policy applications, explore constraints to future progress, and highlight research directions for the field.

Download Full-text

A Microbiome-Based Solution to Address Alarming Levels of Drug-Resistant Bacteria in the Newborn Infant Gut

Infection Control and Hospital Epidemiology ◽

10.1017/ice.2020.1106 ◽

2020 ◽

Vol 41 (S1) ◽

pp. s439-s439

Author(s):

Giorgio Casaburi ◽

Rebbeca Duar ◽

Bethany Henrick ◽

Steven Frese

Keyword(s):

Antibiotic Resistance ◽

Antimicrobial Resistance ◽

Bacterial Species ◽

The United States ◽

Resistant Bacteria ◽

Delivery Mode ◽

Term Infants ◽

Shotgun Metagenomics ◽

Single Strain ◽

Breastfed Infants

Background: Recent studies have focused on the early infant gut microbiome, indicating that antibiotic resistance genes (ARGs) can be acquired in early life and may have long-term sequelae. Limiting the spread of antimicrobial resistance without triggering the development of additional resistance mechanisms would be of immense clinical value. Here, we present 2 analyses that highlight the abundance of ARGs in preterm and term infants and a proof of concept for modulating the microbiome to promote early stabilization and reduction in ARGs in term infants. Methods: Large-scale metagenomic analysis was performed on 2,141 microbiome samples (90% from pre-term infants) from 10 countries; most were from the United States (87%) and were obtained from the Comprehensive Antibiotic Resistance Database (CARD). We assessed the abundance and specific types of ARGs present. In the second study, healthy, breastfed infants were fed B. infantis EVC001 for 3 weeks starting at postnatal day 7. Stool samples were collected at day 21 and were processed utilizing shotgun metagenomics. Selected antimicrobial-resistant bacterial species were isolated, sequenced, and tested for minimal inhibitory concentrations to clinically relevant antibiotics. Results: In the first study, globally, 417 distinct ARGs were identified. The most abundant gene among all samples was annotated as msrE, a plasmid gene known to confer resistance to macrolide-lincosamide-streptogramin B (MLSB) antibiotics. The remaining most-abundant ARGs were efflux-pump genes associated with multidrug resistance. No significant association in antimicrobial resistance was found when considering delivery mode or antibiotic treatment in the first month of life. In the second study, the EVC001-fed group showed a significant decrease (90%) in ARGs compared to controls (P < .0001). ARGs that differed significantly between groups were predicted to confer resistance to β-lactams, fluoroquinolones, or multiple drug classes. Minimal inhibitory concentration assays confirmed resistance phenotypes among isolates Notably, we found resistance to extended-spectrum β-lactamases among healthy, vaginally delivered breastfed infants who had never been exposed to antibiotics. Conclusions: In this study, we show that the term and preterm infant microbiome contains alarming levels of ARGs associated with clinically relevant antibiotics harbored by bacteria commonly responsible for nosocomial infections. Colonization of the breastfed infant gut by a single strain of B. longum subsp infantis had profound impacts on the fecal metagenome, including reduction in ARGs and reduction of potential pathogens. These findings highlight the importance of developing novel approaches to limit the spread of ARGs among clinically relevant bacteria and the relevance of an additional approach in the effort to solve AR globally.Funding: Evolve BioSystems provided Funding: for this study.Disclosures: Giorgio Casaburi reports salary from Evolve BioSystems.

Download Full-text

Abstract 15076: County-level Subgroups With Unique Cardiovascular Outcomes: An Unsupervised Clustering Analysis

Circulation ◽

10.1161/circ.142.suppl_3.15076 ◽

2020 ◽

Vol 142 (Suppl_3) ◽

Author(s):

Matthew W Segar ◽

Shreya Rao ◽

Amy E Hughes ◽

Ambarish Pandey

Keyword(s):

Machine Learning ◽

Risk Factors ◽

Linear Regression ◽

Healthcare Access ◽

Model Performance ◽

The United States ◽

County Level ◽

Cvd Risk ◽

Unsupervised Machine Learning ◽

Cvd Mortality

Introduction: Regional heterogeneity in cardiovascular disease (CVD) risk has been demonstrated in the United States (US), however associations between county-level social and demographic risk factors and CVD mortality have not previously been described. To address this, we evaluated whether unsupervised machine learning-based clustering could identify distinct US county subgroups with unique CVD outcomes. Methods: The study included 2,676 counties from the 2020 County Health Rankings program. Unsupervised hierarchical clustering of 46 candidate variables encompassing demographic, health behaviors, socioeconomic factors, and healthcare access domains was used to derive phenogroups. The association of phenogroups and age-adjusted CVD mortality was assessed by linear regression. Results: Clustering identified 4 phenogroups based on within-cluster inertia. Cluster 1 (N=924; 24.5%) counties were largely white, suburban households with high income and access to healthcare. Cluster 2 counties (N=451; 16.9%) were large with predominantly Hispanic residents and below average prevalence of CVD risk factors. Cluster 3 (N=951; 35.5%) counties included rural, white residents with the lowest levels of healthcare access. Cluster 4 (350; 13.1%) counties were comprised of predominantly black residents with substantial cardiovascular comorbidities and physical and socioeconomic burdens. Age-adjusted cardiovascular mortality increased in a stepwise fashion from 247 in cluster 1 to 349 per 100,000 residents in cluster 4 (FigA) with the phenogroups demonstrating regionality across the US (FigB). Addition of phenogroup in linear regression improved model performance with a R 2 of 0.70. Conclusions: Unsupervised machine learning clustering based on demographic and behavior data can identify unique county phenogroups with differential risk of CVD mortality and may aid in identifying communities at highest risk for CVD-related adverse events.

Download Full-text

1007. Etiology of Sepsis; A Systematic Review of Emergency Department Sepsis

Open Forum Infectious Diseases ◽

10.1093/ofid/ofy210.844 ◽

2018 ◽

Vol 5 (suppl_1) ◽

pp. S300-S300

Author(s):

Ashley Husebye ◽

Caitlin Baxter ◽

Elizabeth Wesenberg ◽

Glen Hansen

Keyword(s):

Emergency Department ◽

Early Recognition ◽

Medical Center ◽

Bacterial Species ◽

The United States ◽

Species Group ◽

Blood Cultures ◽

Contamination Rate ◽

Klebsiella Pneumonia ◽

Systematic Analysis

Abstract Background Sepsis is a systemic response to an infection involving one or multiple organ failures frequently caused by bacteremia. Over a million cases of sepsis are reported in the United States annually with an estimated 25% mortality. Early recognition, diagnosis, and treatment of sepsis in the Emergency Department (ED) improves patient outcomes. Increased awareness of sepsis has fostered novel opportunities to improve diagnostics. EDs are increasingly targeted as areas of primary care for suspected septic patients. Understanding the etiology of ED sepsis supports empiric approaches and opportunities for targeted diagnostics. However, a systematic analysis of etiology of ED sepsis, spanning multiple years, is lacking. Methods A retrospective analysis conducted over 60 months at Hennepin County Medical Center, an inner-city level one trauma center with over 100,000 ED visits annually were examined. Positive blood cultures drawn in the ED were included in data analysis. Charts were reviewed for patient demographics and whether the culture was treated; infections that were not treated were considered contaminants, and relevant susceptibility patterns. Results A total of 8,013 blood cultures were drawn in the ED over an initial 12-month period. Of these, 8.4% (n = 674) were culture positive resulting in 731 microorganisms. Of these, 314 were treated as infections with the remaining considered contaminants. Overall contamination rate was 2.9%. Of clinically relevant positive blood cultures, 19.4% were Escherichia coli, 18.5% were Staphylococcus aureus, 27.1% were strep species (group A strep 5.4%, group B strep 4.8%, strep pneumonia 5.1%), 7.0% were Enterococcus faecalis, and 6.4% Klebsiella pneumonia. Among these species, they accounted for 78.4% of pertinent positive cultures. Gram-negative bacteremias accounted for 41% of infections compared with 59% for Gram-positive organisms. Conclusion A comprehensive understanding of the etiology of ED sepsis facilitates appropriate empiric antimicrobial prescribing for patients who present with sepsis in the ED. Data collected to date identifies five key bacterial species associated with over 78% of confirmed ED sepsis. Disclosures All authors: No reported disclosures.

Download Full-text

A Machine Learning Approach to Identify Predictors of Potentially Inappropriate Non-Steroidal Anti-Inflammatory Drugs (NSAIDs) Use in Older Adults with Osteoarthritis

International Journal of Environmental Research and Public Health ◽

10.3390/ijerph18010155 ◽

2020 ◽

Vol 18 (1) ◽

pp. 155

Author(s):

Jayeshkumar Patel ◽

Amit Ladani ◽

Nethra Sambamoorthi ◽

Traci LeMasters ◽

Nilanjana Dwibedi ◽

...

Keyword(s):

Machine Learning ◽

Older Adults ◽

The United States ◽

Gradient Boosting ◽

Study Cohort ◽

Real World Data ◽

Extreme Gradient Boosting ◽

Inflammatory Drugs ◽

Anti Inflammatory ◽

Using Data

Evidence from some studies suggest that osteoarthritis (OA) patients are often prescribed non-steroidal anti-inflammatory drugs (NSAIDs) that are not in accordance with their cardiovascular (CV) or gastrointestinal (GI) risk profiles. However, no such study has been carried out in the United States. Therefore, we sought to examine the prevalence and predictors of potentially inappropriate NSAIDs use in older adults (age > 65) with OA using machine learning with real-world data from Optum De-identified Clinformatics® Data Mart. We identified a retrospective cohort of eligible individuals using data from 2015 (baseline) and 2016 (follow-up). Potentially inappropriate NSAIDs use was identified using the type (COX-2 selective vs. non-selective) and length of NSAIDs use and an individual’s CV and GI risk. Predictors of potentially inappropriate NSAIDs use were identified using eXtreme Gradient Boosting. Our study cohort comprised of 44,990 individuals (mean age 75.9 years). We found that 12.8% individuals had potentially inappropriate NSAIDs use, but the rate was disproportionately higher (44.5%) in individuals at low CV/high GI risk. Longer duration of NSAIDs use during baseline (AOR 1.02; 95% CI:1.02–1.02 for both non-selective and selective NSAIDs) was associated with a higher risk of potentially inappropriate NSAIDs use. Additionally, individuals with low CV/high GI (AOR 1.34; 95% CI:1.20–1.50) and high CV/low GI risk (AOR 1.61; 95% CI:1.34–1.93) were also more likely to have potentially inappropriate NSAIDs use. Heightened surveillance of older adults with OA requiring NSAIDs is warranted.

Download Full-text

Genome-wide predictions of genetic redundancy in Arabidopsis thaliana

10.1101/2020.08.13.250225 ◽

2020 ◽

Author(s):

Siobhan A. Cusack ◽

Peipei Wang ◽

Bethany M. Moore ◽

Fanrui Meng ◽

Jeffrey K. Conner ◽

...

Keyword(s):

Machine Learning ◽

Arabidopsis Thaliana ◽

Model Building ◽

Expression Patterns ◽

Model Performance ◽

Machine Learning Algorithms ◽

Stress Conditions ◽

Loss Of Function ◽

Genetic Redundancy ◽

Gene Pairs

ABSTRACTGenetic redundancy refers to a situation where an individual with a loss-of-function mutation in one gene (single mutant) does not show an apparent phenotype until one or more paralogs are also knocked out (double/higher-order mutant). Previous studies have identified some characteristics common among redundant gene pairs, but a predictive model of genetic redundancy incorporating a wide variety of features has not yet been established. In addition, the relative importance of these characteristics for genetic redundancy remains unclear. Here, we establish machine learning models for predicting whether a gene pair is likely redundant or not in the model plant Arabidopsis thaliana. Benchmark gene pairs were classified based on six feature categories: functional annotations, evolutionary conservation including duplication patterns and mechanisms, epigenetic marks, protein properties including post-translational modifications, gene expression, and gene network properties. The definition of redundancy, data transformations, feature subsets, and machine learning algorithms used affected model performance significantly. Among the most important features in predicting gene pairs as redundant were having a paralog(s) from recent duplication events, annotation as a transcription factor, downregulation during stress conditions, and having similar expression patterns under stress conditions. Predictions were then tested using phenotype data withheld from model building and validated using well-characterized, redundant and nonredundant gene pairs. This genetic redundancy model sheds light on characteristics that may contribute to long-term maintenance of paralogs that are seemingly functionally redundant, and will ultimately allow for more targeted generation of functionally informative double mutants, advancing functional genomic studies.

Download Full-text

Machine learning prediction of dropping out of outpatients with alcohol use disorders

PLoS ONE ◽

10.1371/journal.pone.0255626 ◽

2021 ◽

Vol 16 (8) ◽

pp. e0255626

Author(s):

So Jin Park ◽

Sun Jung Lee ◽

HyungMin Kim ◽

Jae Kwon Kim ◽

Ji-Won Chun ◽

...

Keyword(s):

Machine Learning ◽

Alcohol Use ◽

Outpatient Treatment ◽

Medical Center ◽

Specific Treatment ◽

Mental Illnesses ◽

Patient Specific ◽

Dropping Out ◽

Support Vector ◽

Final Model

Background Alcohol use disorder (AUD) is a chronic disease with a higher recurrence rate than that of other mental illnesses. Moreover, it requires continuous outpatient treatment for the patient to maintain abstinence. However, with a low probability of these patients to continue outpatient treatment, predicting and managing patients who might discontinue treatment becomes necessary. Accordingly, we developed a machine learning (ML) algorithm to predict which the risk of patients dropping out of outpatient treatment schemes. Methods A total of 839 patients were selected out of 2,206 patients admitted for AUD in three hospitals under the Catholic Central Medical Center in Korea. We implemented six ML models—logistic regression, support vector machine, k-nearest neighbor, random forest, neural network, and AdaBoost—and compared the prediction performances thereof. Results Among the six models, AdaBoost was selected as the final model for recommended use owing to its area under the receiver operating characteristic curve (AUROC) of 0.72. The four variables affecting the prediction based on feature importance were the length of hospitalization, age, residential area, and diabetes. Conclusion An ML algorithm was developed herein to predict the risk of patients with AUD in Korea discontinuing outpatient treatment. By testing and validating various machine learning models, we determined the best performing model, AdaBoost, as the final model for recommended use. Using this model, clinicians can manage patients with high risks of discontinuing treatment and establish patient-specific treatment strategies. Therefore, our model can potentially enable patients with AUD to successfully complete their treatments by identifying them before they can drop out.

Download Full-text