A5017505484- Machine Learning in Healthcare
- Colorectal Cancer Screening and Detection
- ECG Monitoring and Analysis
- Chronic Disease Management Strategies
- Palliative Care and End-of-Life Issues
- Topic Modeling
- Heart Rate Variability and Autonomic Control
- Explainable Artificial Intelligence (XAI)
- Artificial Intelligence in Healthcare
- Non-Invasive Vital Sign Monitoring
- Health Systems, Economic Evaluations, Quality of Life
- Artificial Intelligence in Healthcare and Education
- Heart Failure Treatment and Management
- Systemic Lupus Erythematosus Research
- Electronic Health Records Systems
- Frailty in Older Adults
- Machine Learning and Data Classification
- Emergency and Acute Care Studies
- Text Readability and Simplification
- Fault Detection and Control Systems
- Atherosclerosis and Cardiovascular Diseases
- Healthcare Policy and Management
- Sepsis Diagnosis and Treatment
- Plant Water Relations and Carbon Dynamics
- Probabilistic and Robust Engineering Design
Stanford University
2017-2022
Access to palliative care is a key quality metric which most healthcare organizations strive improve. The primary challenges increasing access are combination of physicians over-estimating patient prognoses, and shortage staff in general. This, with treatment inertia can result mismatch between wishes, their actual towards the end life. In this work, we address problem, Institutional Review Board approval, using machine learning Electronic Health Record (EHR) data patients. We train Deep...
Natural language correction has the potential to help learners improve their writing skills. While approaches with separate classifiers for different error types have high precision, they do not flexibly handle errors such as redundancy or non-idiomatic phrasing. On other hand, word and phrase-based machine translation methods are designed cope orthographic errors, recently been outpaced by neural models. Motivated these issues, we present a network-based approach correction. The core...
Improving the quality of end-of-life care for hospitalized patients is a priority healthcare organizations. Studies have shown that physicians tend to over-estimate prognoses, which in combination with treatment inertia results mismatch between wishes and actual at end life. We describe method address this problem using Deep Learning Electronic Health Record (EHR) data, currently being piloted, Institutional Review Board approval, an academic medical center. The EHR data admitted are...
To analyze the impact of factors in healthcare delivery on net benefit triggering an Advanced Care Planning (ACP) workflow based predictions 12-month mortality.
We develop an algorithm that accurately detects Atrial Fibrillation (AF) episodes from photoplethysmograms (PPG) recorded in ambulatory free-living conditions. collect and annotate a dataset containing more than 4000 hours of PPG wrist-worn device. Using 50-layer convolutional neural network, we achieve test AUC 95% presence motion artifacts inherent to signals. Such continuous accurate detection AF has the potential transform consumer wearable devices into clinically useful medical monitoring tools.
Abstract Accurate and efficient identification of complex chronic conditions in the electronic health record (EHR) is an important but challenging task that has historically relied on tedious clinician review oversimplification disease. Here we adapt methods allow for automated “noisy labeling” positive negative controls to create a “silver standard” machine learning automate systemic lupus erythematosus (SLE). Our final model, which includes both structured data as well text processing...
SummaryTo facilitate the development of machine-learning (ML) models in care delivery, which remain poorly understood and executed, Stanford Medicine targeted an effort to address this implementation gap at health system by addressing three key challenges: developing a framework for designing integration artificial intelligence (AI) into complex work systems; identifying building teams people, technologies, processes successfully develop implement AI-enabled executing manner that is...
We present Natural Gradient Boosting (NGBoost), an algorithm for generic probabilistic prediction via gradient boosting. Typical regression models return a point estimate, conditional on covariates, but output full probability distribution over the outcome space, covariates. This allows predictive uncertainty estimation -- crucial in applications like healthcare and weather forecasting. NGBoost generalizes boosting to by treating parameters of as targets multiparameter algorithm....
Calibrated uncertainty estimates in machine learning are crucial to many fields such as autonomous vehicles, medicine, and weather climate forecasting. While there is extensive literature on calibration for classification, the classification findings do not always translate regression. As a result, modern models predicting regression settings typically produce uncalibrated overconfident estimates. To address these gaps, we present method that does assume particular distribution over error:...
Machine learning (ML) models for allocating readmission-mitigating interventions are typically selected according to their discriminative ability, which may not necessarily translate into utility in allocation of resources. Our objective was determine whether ML have different usefulness based on overall and ability. We conducted a retrospective analysis using claims data acquired from the Optum Clinformatics Data Mart, including 513,495 commercially-insured inpatients (mean [SD] age 69 [19]...
Probabilistic survival predictions from models trained with Maximum Likelihood Estimation (MLE) can have high, and sometimes unacceptably high variance. The field of meteorology, where the paradigm maximizing sharpness subject to calibration is popular, has addressed this problem by using scoring rules beyond MLE, such as Continuous Ranked Probability Score (CRPS). In paper we present \emph{Survival-CRPS}, a generalization CRPS prediction setting, right-censored interval-censored variants....
We develop an algorithm that accurately detects Atrial Fibrillation (AF) episodes from photoplethysmograms (PPG) recorded in ambulatory free-living conditions. collect and annotate a dataset containing more than 4000 hours of PPG wrist-worn device. Using 50-layer convolutional neural network, we achieve test AUC 95% show robustness to motion artifacts inherent signals. Continuous accurate detection AF has the potential transform consumer wearable devices into clinically useful medical...
Identifying patients who will be discharged within 24 hours can improve hospital resource management and quality of care. We studied this problem using eight years Electronic Health Records (EHR) data from Stanford Hospital. fit models to predict hour discharge across the entire inpatient population. The best performing achieved an area under receiver-operator characteristic curve (AUROC) 0.85 AUPRC 0.53 on a held out test set. This model was also well calibrated. Finally, we analyzed...
Machine learning has recently demonstrated impressive progress in predictive accuracy across a wide array of tasks. Most ML approaches focus on generalization performance unseen data that are similar to the training (In-Distribution, or IND). However, real world applications and deployments rarely enjoy comfort encountering examples always IND. In such situations, most models commonly display erratic behavior Out-of-Distribution (OOD) examples, as assigning high confidence wrong predictions,...
Abstract Objective To analyze the impact of factors in healthcare delivery on net benefit triggering an Advanced Care Planning (ACP) workflow based predictions 12-month mortality. Materials and Methods We built a predictive model mortality using electronic health record data evaluated ACP models’ predictions. Factors included non-clinical reasons that make inappropriate, limited capacity for ACP, inability to follow up due patient discharge, availability outpatient missed cases. also...
Improving the quality of end-of-life care for hospitalized patients is a priority healthcare organizations. Studies have shown that physicians tend to over-estimate prognoses, which in combination with treatment inertia results mismatch between wishes and actual at end life. We describe method address this problem using Deep Learning Electronic Health Record (EHR) data, currently being piloted, Institutional Review Board approval, an academic medical center. The EHR data admitted are...
Structured Abstract Importance Machine learning (ML) models for allocating readmission-mitigating interventions are typically selected according to their discriminative ability, which may not necessarily translate into utility in allocation of resources. Objective To determine whether ML ranked differently based on overall and ability. Design A retrospective analysis using claims data acquired from the Optum Clinformatics Data Mart. Setting Health plan all 50 states commercially-insured...