Machine learning to predict notes for chart review in the oncology setting: a proof of concept strategy for improving clinician note-writing.

OBJECTIVE: Leverage electronic health record (EHR) audit logs to develop a machine learning (ML) model that predicts which notes a clinician wants to review when seeing oncology patients. MATERIALS AND METHODS: We trained logistic regression models using note metadata and a Term Frequency Inverse Document Frequency (TF-IDF) text representation. We evaluated performance with precision, recall, F1, AUC, and a clinical qualitative assessment. RESULTS: The metadata only model achieved an AUC 0.930 and the metadata and TF-IDF model an AUC 0.937. Qualitative assessment revealed a need for better text representation and to further customize predictions for the user. DISCUSSION: Our model effectively surfaces the top 10 notes a clinician wants to review when seeing an oncology patient. Further studies can characterize different types of clinician users and better tailor the task for different care settings. CONCLUSION: EHR audit logs can provide important relevance data for training ML models that assist with note-writing in the oncology setting.

Attention-based neural networks for clinical prediction modelling on electronic health records.

BACKGROUND: Deep learning models have had a lot of success in various fields. However, on structured data they have struggled. Here we apply four state-of-the-art supervised deep learning models using the attention mechanism and compare against logistic regression and XGBoost using discrimination, calibration and clinical utility. METHODS: We develop the models using a general practitioners database. We implement a recurrent neural network, a transformer with and without reverse distillation and a graph neural network. We measure discrimination using the area under the receiver operating characteristic curve (AUC) and the area under the precision recall curve (AUPRC). We assess smooth calibration using restricted cubic splines and clinical utility with decision curve analysis. RESULTS: Our results show that deep learning approaches can improve discrimination up to 2.5% points AUC and 7.4% points AUPRC. However, on average the baselines are competitive. Most models are similarly calibrated as the baselines except for the graph neural network. The transformer using reverse distillation shows the best performance in clinical utility on two out of three prediction problems over most of the prediction thresholds. CONCLUSION: In this study, we evaluated various approaches in supervised learning using neural networks and attention. Here we do a rigorous comparison, not only looking at discrimination but also calibration and clinical utility. There is value in using deep learning models on electronic health record data since it can improve discrimination and clinical utility while providing good calibration. However, good baseline methods are still competitive.

Development and characterization of a fecal-induced peritonitis model of murine sepsis: results from a multi-laboratory study and iterative modification of experimental conditions.

BACKGROUND: Preclinical sepsis models have been criticized for their inability to recapitulate human sepsis and suffer from methodological shortcomings that limit external validity and reproducibility. The National Preclinical Sepsis Platform (NPSP) is a consortium of basic science researchers, veterinarians, and stakeholders in Canada undertaking standardized multi-laboratory sepsis research to increase the efficacy and efficiency of bench-to-bedside translation. In this study, we aimed to develop and characterize a 72-h fecal-induced peritonitis (FIP) model of murine sepsis conducted in two independent laboratories. The experimental protocol was optimized by sequentially modifying dose of fecal slurry and timing of antibiotics in an iterative fashion, and then repeating the experimental series at site 1 and site 2. RESULTS: Escalating doses of fecal slurry (0.5-2.5 mg/g) resulted in increased disease severity, as assessed by the modified Murine Sepsis Score (MSS). However, the MSS was poorly associated with progression to death during the experiments, and mice were found dead without elevated MSS scores. Administration of early antibiotics within 4 h of inoculation rescued the animals from sepsis compared with late administration of antibiotics after 12 h, as evidenced by 100% survival and reduced bacterial load in peritoneum and blood in the early antibiotic group. Site 1 and site 2 had statistically significant differences in mortality (60% vs 88%; p < 0.05) for the same dose of fecal slurry (0.75 mg/g) and marked differences in body temperature between groups. CONCLUSIONS: We demonstrate a systematic approach to optimizing a 72-h FIP model of murine sepsis for use in multi-laboratory studies. Alterations to experimental conditions, such as dose of fecal slurry and timing of antibiotics, have clear impact on outcomes. Differences in mortality between sites despite rigorous standardization warrants further investigations to better understand inter-laboratory variation and methodological design in preclinical studies.

Machine learning identifies clusters of longitudinal autoantibody profiles predictive of systemic lupus erythematosus disease outcomes.

OBJECTIVES: A novel longitudinal clustering technique was applied to comprehensive autoantibody data from a large, well-characterised, multinational inception systemic lupus erythematosus (SLE) cohort to determine profiles predictive of clinical outcomes. METHODS: Demographic, clinical and serological data from 805 patients with SLE obtained within 15 months of diagnosis and at 3-year and 5-year follow-up were included. For each visit, sera were assessed for 29 antinuclear antibodies (ANA) immunofluorescence patterns and 20 autoantibodies. K-means clustering on principal component analysis-transformed longitudinal autoantibody profiles identified discrete phenotypic clusters. One-way analysis of variance compared cluster enrolment demographics and clinical outcomes at 10-year follow-up. Cox proportional hazards model estimated the HR for survival adjusting for age of disease onset. RESULTS: Cluster 1 (n=137, high frequency of anti-Smith, anti-U1RNP, AC-5 (large nuclear speckled pattern) and high ANA titres) had the highest cumulative disease activity and immunosuppressants/biologics use at year 10. Cluster 2 (n=376, low anti-double stranded DNA (dsDNA) and ANA titres) had the lowest disease activity, frequency of lupus nephritis and immunosuppressants/biologics use. Cluster 3 (n=80, highest frequency of all five antiphospholipid antibodies) had the highest frequency of seizures and hypocomplementaemia. Cluster 4 (n=212) also had high disease activity and was characterised by multiple autoantibody reactivity including to antihistone, anti-dsDNA, antiribosomal P, anti-Sjögren syndrome antigen A or Ro60, anti-Sjögren syndrome antigen B or La, anti-Ro52/Tripartite Motif Protein 21, antiproliferating cell nuclear antigen and anticentromere B). Clusters 1 (adjusted HR 2.60 (95% CI 1.12 to 6.05), p=0.03) and 3 (adjusted HR 2.87 (95% CI 1.22 to 6.74), p=0.02) had lower survival compared with cluster 2. CONCLUSION: Four discrete SLE patient longitudinal autoantibody clusters were predictive of long-term disease activity, organ involvement, treatment requirements and mortality risk.

Single cell characterization of myeloma and its precursor conditions reveals transcriptional signatures of early tumorigenesis.

Multiple myeloma is a plasma cell malignancy almost always preceded by precursor conditions, but low tumor burden of these early stages has hindered the study of their molecular programs through bulk sequencing technologies. Here, we generate and analyze single cell RNA-sequencing of plasma cells from 26 patients at varying disease stages and 9 healthy donors. In silico dissection and comparison of normal and transformed plasma cells from the same bone marrow biopsy enables discovery of patient-specific transcriptional changes. Using Non-Negative Matrix Factorization, we discover 15 gene expression signatures which represent transcriptional modules relevant to myeloma biology, and identify a signature that is uniformly lost in abnormal cells across disease stages. Finally, we demonstrate that tumors contain heterogeneous subpopulations expressing distinct transcriptional patterns. Our findings characterize transcriptomic alterations present at the earliest stages of myeloma, providing insight into the molecular underpinnings of disease initiation.

Comparison of growth features and cancer stem cell prevalence in intrahepatic and extrahepatic cholangiocarcinoma cell lines.

Aim of the study: Intra- and extrahepatic cholangiocarcinoma (I-CCA and E-CCA respectively) exhibit different growth features that contribute to different clinical outcomes. Cancer stem cells (CSCs) influence tumor growth and thereby may be responsible for these differences. The aim of this study was to document and compare the growth features of human I-CCA and E-CCA cell lines and determine whether any differences observed could be explained by differences in the prevalence and/or stem cell surface marker (SCSM) expression profiles of CSCs within the tumor cell lines. Material and methods: Six CCA cells lines, three I-CCA and three E-CCA, were studied. Tumor cell growth features including cell proliferation, colony/spheroid formation, migration and invasion were documented. CSC prevalence and SCSM expression profiles were examined by flow cytometry. Results: I-CCA cells had significantly increased proliferative activity, shorter doubling times and were more invasive than E-CCA cells, while colony/spheroid formation and migration were similar in the two cell populations. There were no significant differences in CSC prevalence rates or SCSM expression profiles. Conclusions: These findings suggest that I-CCA cells proliferate at a more rapid rate and are more invasive than E-CCA cells but the differences cannot be explained by differences in the prevalence or SCSM expression profiles of CSCs within the tumor cell population.

Assessment of a Crisis Standards of Care Scoring System for Resource Prioritization and Estimated Excess Mortality by Race, Ethnicity, and Socially Vulnerable Area During a Regional Surge in COVID-19.

Importance: Crisis standards of care (CSOC) scores designed to allocate scarce resources during the COVID-19 pandemic could exacerbate racial disparities in health care. Objective: To analyze the association of a CSOC scoring system with resource prioritization and estimated excess mortality by race, ethnicity, and residence in a socially vulnerable area. Design, Setting, and Participants: This retrospective cohort analysis included adult patients in the intensive care unit during a regional COVID-19 surge from April 13 to May 22, 2020, at 6 hospitals in a health care network in greater Boston, Massachusetts. Participants were scored by acute severity of illness using the Sequential Organ Failure Assessment score and chronic severity of illness using comorbidity and life expectancy scores, and only participants with complete scores were included. The score was ordinal, with cutoff points suggested by the Massachusetts guidelines. Exposures: Race, ethnicity, Social Vulnerability Index. Main Outcomes and Measures: The primary outcome was proportion of patients in the lowest priority score category stratified by self-reported race. Secondary outcomes were discrimination and calibration of the score overall and by race, ethnicity, and neighborhood Social Vulnerability Index. Projected excess deaths were modeled by race, using the priority scoring system and a random lottery. Results: Of 608 patients in the intensive care unit during the study period, 498 had complete data and were included in the analysis; this population had a median (IQR) age of 67 (56-75) years, 191 (38.4%) female participants, 79 (15.9%) Black participants, and 225 patients (45.7%) with COVID-19. The area under the receiver operating characteristic curve for the priority score was 0.79 and was similar across racial groups. Black patients were more likely than others to be in the lowest priority group (12 [15.2%] vs 34 [8.1%]; P = .046). In an exploratory simulation model using the score for ventilator allocation, with only those in the highest priority group receiving ventilators, there were 43.9% excess deaths among Black patients (18 of 41 patients) and 28.6% (58 of 203 patients among all others (P = .05); when the highest and intermediate priority groups received ventilators, there were 4.9% (2 of 41 patients) excess deaths among Black patients and 3.0% (6 of 203) among all others (P = .53). A random lottery resulted in more excess deaths than the score. Conclusions and Relevance: In this study, a CSOC priority score resulted in lower prioritization of Black patients to receive scarce resources. A model using a random lottery resulted in more estimated excess deaths overall without improving equity by race. CSOC policies must be evaluated for their potential association with racial disparities in health care.

The Potential For Bias In Machine Learning And Opportunities For Health Insurers To Address It.

As the use of machine learning algorithms in health care continues to expand, there are growing concerns about equity, fairness, and bias in the ways in which machine learning models are developed and used in clinical and business decisions. We present a guide to the data ecosystem used by health insurers to highlight where bias can arise along machine learning pipelines. We suggest mechanisms for identifying and dealing with bias and discuss challenges and opportunities to increase fairness through analytics in the health insurance industry.

Fibroblast Growth Factor 19 Induced Changes in Non-malignant Cholangiocytes.

BACKGROUND AND AIMS: Fibroblast growth factor (FGF)19 has been implicated in the pathogenesis of murine hepatocellular carcinoma. Whether it plays a role in the development or course of human cholangiocarcinoma remains to be determined. The aim of this study was to determine whether prolonged exposure to FGF19 results in the transformation of non-malignant human cholangiocytes into cells with malignant features. METHODS: Human SV-40 transfected non-malignant H69 cholangiocytes were cultured with FGF19 (0-50 ng/mL) for 6 weeks, followed by 6 weeks with medium alone. Cell proliferation, invasion, stem cell surface markers, oncofetoprotein expression, state of differentiation, epithelial-mesenchymal transition (EMT) and interleukin (IL)-6 expression were documented at various time intervals throughout the 12-week period. RESULTS: FGF19 exposure was associated with significant increases in cell proliferation, de-differentiation, EMT and IL-6 expression. However, each of these effects returned to baseline or control values during the 6-week FGF19 free follow-up period. The remaining cell properties remained unaltered. CONCLUSIONS: Six weeks of FGF19 exposure did not result in the acquisition of permanent malignant features in non-malignant, human cholangiocytes.

Trajectory Inspection: A Method for Iterative Clinician-Driven Design of Reinforcement Learning Studies.

Reinforcement learning (RL) has the potential to significantly improve clinical decision making. However, treatment policies learned via RL from observational data are sensitive to subtle choices in study design. We highlight a simple approach, trajectory inspection, to bring clinicians into an iterative design process for model-based RL studies. We identify where the model recommends unexpectedly aggressive treatments or expects surprisingly positive outcomes from its recommendations. Then, we examine clinical trajectories simulated with the learned model and policy alongside the actual hospital course. Applying this approach to recent work on RL for sepsis management, we uncover a model bias towards discharge, a preference for high vasopressor doses that may be linked to small sample sizes, and clinically implausible expectations of discharge without weaning off vasopressors. We hope that iterations of detecting and addressing the issues unearthed by our method will result in RL policies that inspire more confidence in deployment.

Pulse of the pandemic: Iterative topic filtering for clinical information extraction from social media.

The rapid evolution of the COVID-19 pandemic has underscored the need to quickly disseminate the latest clinical knowledge during a public-health emergency. One surprisingly effective platform for healthcare professionals (HCPs) to share knowledge and experiences from the front lines has been social media (for example, the "#medtwitter" community on Twitter). However, identifying clinically-relevant content in social media without manual labeling is a challenge because of the sheer volume of irrelevant data. We present an unsupervised, iterative approach to mine clinically relevant information from social media data, which begins by heuristically filtering for HCP-authored texts and incorporates topic modeling and concept extraction with MetaMap. This approach identifies granular topics and tweets with high clinical relevance from a set of about 52 million COVID-19-related tweets from January to mid-June 2020. We also show that because the technique does not require manual labeling, it can be used to identify emerging topics on a week-to-week basis. Our method can aid in future public-health emergencies by facilitating knowledge transfer among healthcare workers in a rapidly-changing information environment, and by providing an efficient and unsupervised way of highlighting potential areas for clinical research.

Ending Restraint of Incarcerated Individuals Giving Birth.

Advocates have long suggested making shackling incarcerated people during childbirth illegal. Yet exceptions would likely still allow prison personnel to implement restraint and leave clinicians no course for freeing a patient. This article argues that clinicians' assessments of laboring individuals' clinical needs must be prioritized, ethically and legally. This article also explains that, without strong policies in place, some clinicians will not feel empowered to demand that a patient be freed during labor. Beyond prohibiting restraint of laboring individuals, health care organizations must support clinicians seeking to execute their ethical duties to care well and justly for patients. Toward this end, this article proposes a model policy.

Predicting Response to Tocilizumab Monotherapy in Rheumatoid Arthritis: A Real-world Data Analysis Using Machine Learning.

OBJECTIVE: Tocilizumab (TCZ) has shown similar efficacy when used as monotherapy as in combination with other treatments for rheumatoid arthritis (RA) in randomized controlled trials (RCTs). We derived a remission prediction score for TCZ monotherapy (TCZm) using RCT data and performed an external validation of the prediction score using real-world data (RWD). METHODS: We identified patients in the Corrona RA registry who used TCZm (n = 452), and matched the design and patients from 4 RCTs used in previous work (n = 853). Patients were followed to determine remission status at 24 weeks. We compared the performance of remission prediction models in RWD, first based on variables determined in our prior work in RCTs, and then using an extended variable set, comparing logistic regression and random forest models. We included patients on other biologic disease-modifying antirheumatic drug monotherapies (bDMARDm) to improve prediction. RESULTS: The fraction of patients observed reaching remission on TCZm by their follow-up visit was 12% (n = 53) in RWD vs 15% (n = 127) in RCTs. Discrimination was good in RWD for the risk score developed in RCTs, with area under the receiver-operating characteristic curve (AUROC) of 0.69 (95% CI 0.62-0.75). Fitting the same logistic regression model to all bDMARDm patients in the RWD improved the AUROC on held-out TCZm patients to 0.72 (95% CI 0.63-0.81). Extending the variable set and adding regularization further increased it to 0.76 (95% CI 0.67-0.84). CONCLUSION: The remission prediction scores, derived in RCTs, discriminated patients in RWD about as well as in RCTs. Discrimination was further improved by retraining models on RWD.

Automated NLP Extraction of Clinical Rationale for Treatment Discontinuation in Breast Cancer.

PURPOSE: Key oncology end points are not routinely encoded into electronic medical records (EMRs). We assessed whether natural language processing (NLP) can abstract treatment discontinuation rationale from unstructured EMR notes to estimate toxicity incidence and progression-free survival (PFS). METHODS: We constructed a retrospective cohort of 6,115 patients with early-stage and 701 patients with metastatic breast cancer initiating care at Memorial Sloan Kettering Cancer Center from 2008 to 2019. Each cohort was divided into training (70%), validation (15%), and test (15%) subsets. Human abstractors identified the clinical rationale associated with treatment discontinuation events. Concatenated EMR notes were used to train high-dimensional logistic regression and convolutional neural network models. Kaplan-Meier analyses were used to compare toxicity incidence and PFS estimated by our NLP models to estimates generated by manual labeling and time-to-treatment discontinuation (TTD). RESULTS: Our best high-dimensional logistic regression models identified toxicity events in early-stage patients with an area under the curve of the receiver-operator characteristic of 0.857 ± 0.014 (standard deviation) and progression events in metastatic patients with an area under the curve of 0.752 ± 0.027 (standard deviation). NLP-extracted toxicity incidence and PFS curves were not significantly different from manually extracted curves (P = .95 and P = .67, respectively). By contrast, TTD overestimated toxicity in early-stage patients (P < .001) and underestimated PFS in metastatic patients (P < .001). Additionally, we tested an extrapolation approach in which 20% of the metastatic cohort were labeled manually, and NLP algorithms were used to abstract the remaining 80%. This extrapolated outcomes approach resolved PFS differences between receptor subtypes (P < .001 for hormone receptor+/human epidermal growth factor receptor 2- v human epidermal growth factor receptor 2+ v triple-negative) that could not be resolved with TTD. CONCLUSION: NLP models are capable of abstracting treatment discontinuation rationale with minimal manual labeling.

Pulse of the Pandemic: Iterative Topic Filtering for Clinical Information Extraction from Social Media.

The rapid evolution of the COVID-19 pandemic has underscored the need to quickly disseminate the latest clinical knowledge during a public-health emergency. One surprisingly effective platform for healthcare professionals (HCPs) to share knowledge and experiences from the front lines has been social media (for example, the "#medtwitter" community on Twitter). However, identifying clinically-relevant content in social media without manual labeling is a challenge because of the sheer volume of irrelevant data. We present an unsupervised, iterative approach to mine clinically relevant information from social media data, which begins by heuristically filtering for HCP-authored texts and incorporates topic modeling and concept extraction with MetaMap. This approach identifies granular topics and tweets with high clinical relevance from a set of about 52 million COVID-19-related tweets from January to mid-June 2020. We also show that because the technique does not require manual labeling, it can be used to identify emerging topics on a week-to-week basis. Our method can aid in future public-health emergencies by facilitating knowledge transfer among healthcare workers in a rapidly-changing information environment, and by providing an efficient and unsupervised way of highlighting potential areas for clinical research.

Alterations in chemokine receptor CCR5 activity influence tumor cell biology in human cholangiocarcinoma cell lines.

INTRODUCTION AND OBJECTIVES: Intrahepatic (I-CCA) and extrahepatic (E-CCA) cholangiocarcinoma (CCA) have different growth patterns and risks for tumor metastasis. Inhibition and/or activation of the chemokine receptor CCR subclasses have been reported to alter tumor cell biology in non-CCA cancers. In this study we documented CCR expression profiles in representative human I-CCA and E-CCA cell lines and the in vitro effects of CCR antagonists and agonists on tumor cell biology. MATERIALS AND METHODS: CCR expression profiles were documented by real-time reverse transcription polymerase chain reaction; cell proliferation by WST-1; spheroid formation by sphere dimensions in anchorage-free medium; cell migration by wound healing and invasion by Transwell invasion chambers. RESULTS: All 10 CCR motifs (CCR1-10) were expressed in the I-CCA, HuCCT1 cell line and six (CCR4, 5, 6, 8, 9 and 10) in the E-CCA, KMBC cell line. In HuCCT1 cells, CCR5 expression was most abundant whereas in KMBC cells, CCR6 followed by CCR5 were most abundant. The CCR5 antagonist Maraviroc significantly inhibited cell proliferation, migration and invasion in HuCCT1 cells, and spheroid formation and invasion in KMBC cells. The CCR5 agonist RANTES had no effect on HuCCT1 cells but increased cell proliferation, migration and invasion of KMBC cells. CONCLUSION: These results suggest that CCR expression profiles differ in I-CCA and E-CCA. They also indicate that CCR5 antagonists and agonists have cell-specific effects but in general, CCR5 inactivation inhibits CCA tumor cell aggressiveness. Additional research is required to determine whether CCR5 inactivation is of value in the treatment of CCA in humans.

Enhanced gemcitabine cytotoxicity with knockdown of multidrug resistance protein genes in human cholangiocarcinoma cell lines.

BACKGROUND AND AIM: Cholangiocarcinoma (CCA) is an often fatal primary cancer of the liver that tends to be resistant to chemotherapy. Multidrug resistance proteins (MRPs) contribute to the chemoresistance of these tumors. The objectives of the study were to document MRP expression profiles in two representative human intrahepatic and extrahepatic CCA cells lines (HuCCT1 and KMBC, respectively) and gemcitabine-induced cytotoxicity prior to and following MRP knockdown. METHODS: Multidrug resistance protein mRNA and protein expression were documented by real-time reverse transcription-polymerase chain reaction and western blots, respectively. MRP knockdown was achieved with lentivirus small hairpin RNA constructs. RESULTS: Prior to gemcitabine exposure, MRP1, MRP2, MRP4, MRP5, and MRP6 mRNA were expressed in HuCCT1 cells and MRP1, MRP3, MRP4, and MRP5 in KMBC cells. Following gemcitabine exposure, MRP5 and MRP6 expressions were significantly upregulated in HuCCT1 cells and MRP5 in KMBC cells. In HuCCT1 cells, although MRP5 knockdown had no effect, MRP6 knockdown significantly increased gemcitabine-induced cytotoxicity. In KMBC cells, MRP5 knockdown significantly increased gemcitabine cytotoxicity. CONCLUSIONS: Inhibition of MRP6 expression in intra-hepatic and MRP5 in extra-hepatic should be explored as potential treatments for CCA in humans.

Veno-venous extracorporeal membrane oxygenation allocation in the COVID-19 pandemic.

Rapid global spread of severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) and the resultant clinical illness, coronavirus disease 2019 (COVID-19), drove the World Health Organization to declare COVID-19 a pandemic. Veno-venous Extra-Corporeal Membrane Oxygenation (VV-ECMO) is an established therapy for management of patients demonstrating the most severe forms of hypoxemic respiratory failure from COVID-19. However, features of COVID-19 pathophysiology and necessary length of treatment present distinct challenges for utilization of VV-ECMO within the current healthcare emergency. In addition, growing allocation concerns due to capacity and cost present significant challenges. Ethical and legal aspects pertinent to triage of this resource-intensive, but potentially life-saving, therapy in the setting of the COVID-19 pandemic are reviewed here. Given considerations relevant to VV-ECMO use, additional emphasis has been placed on emerging hospital resource scarcity and disproportionate representation of healthcare workers among the ill. Considerations are also discussed surrounding withdrawal of VV-ECMO and the role for early communication as well as consultation from palliative care teams and local ethics committees. In discussing how to best manage these issues in the COVID-19 pandemic at present, we identify gaps in the literature and policy important to clinicians as this crisis continues.

A decision algorithm to promote outpatient antimicrobial stewardship for uncomplicated urinary tract infection.

Antibiotic resistance is a major cause of treatment failure and leads to increased use of broad-spectrum agents, which begets further resistance. This vicious cycle is epitomized by uncomplicated urinary tract infection (UTI), which affects one in two women during their life and is associated with increasing antibiotic resistance and high rates of prescription for broad-spectrum second-line agents. To address this, we developed machine learning models to predict antibiotic susceptibility using electronic health record data and built a decision algorithm for recommending the narrowest possible antibiotic to which a specimen is susceptible. When applied to a test cohort of 3629 patients presenting between 2014 and 2016, the algorithm achieved a 67% reduction in the use of second-line antibiotics relative to clinicians. At the same time, it reduced inappropriate antibiotic therapy, defined as the choice of a treatment to which a specimen is resistant, by 18% relative to clinicians. For specimens where clinicians chose a second-line drug but the algorithm chose a first-line drug, 92% (1066 of 1157) of decisions ended up being susceptible to the first-line drug. When clinicians chose an inappropriate first-line drug, the algorithm chose an appropriate first-line drug 47% (183 of 392) of the time. Our machine learning decision algorithm provides antibiotic stewardship for a common infectious syndrome by maximizing reductions in broad-spectrum antibiotic use while maintaining optimal treatment outcomes. Further work is necessary to improve generalizability by training models in more diverse populations.

Predicting human health from biofluid-based metabolomics using machine learning.

Biofluid-based metabolomics has the potential to provide highly accurate, minimally invasive diagnostics. Metabolomics studies using mass spectrometry typically reduce the high-dimensional data to only a small number of statistically significant features, that are often chemically identified-where each feature corresponds to a mass-to-charge ratio, retention time, and intensity. This practice may remove a substantial amount of predictive signal. To test the utility of the complete feature set, we train machine learning models for health state-prediction in 35 human metabolomics studies, representing 148 individual data sets. Models trained with all features outperform those using only significant features and frequently provide high predictive performance across nine health state categories, despite disparate experimental and disease contexts. Using only non-significant features it is still often possible to train models and achieve high predictive performance, suggesting useful predictive signal. This work highlights the potential for health state diagnostics using all metabolomics features with data-driven analysis.