Machine learning for outcome prediction of acute ischemic stroke post intra-arterial therapy. PLoS One. 2014;9:e88225. [PMID: 24520356 PMCID: PMC3919736 DOI: 10.1371/journal.pone.0088225]

A Bayesian Network Model for Predicting Post-stroke Outcomes With Available Risk Factors

Bayesian network is an increasingly popular method in modeling uncertain and complex problems, because its interpretability is often more useful than plain prediction. To satisfy the core requirement in medical research to obtain interpretable prediction with high accuracy, we constructed an inference engine for post-stroke outcomes based on Bayesian network classifiers. The prediction system that was trained on data of 3,605 patients with acute stroke forecasts the functional independence at 3 months and the mortality 1 year after stroke. Feature selection methods were applied to eliminate less relevant and redundant features from 76 risk variables. The Bayesian network classifiers were trained with a hill-climbing searching for the qualified network structure and parameters measured by maximum description length. We evaluated and optimized the proposed system to increase the area under the receiver operating characteristic curve (AUC) while ensuring acceptable sensitivity for the class-imbalanced data. The performance evaluation demonstrated that the Bayesian network with selected features by wrapper-type feature selection can predict 3-month functional independence with an AUC of 0.889 using only 19 risk variables and 1-year mortality with an AUC of 0.893 using 24 variables. The Bayesian network with 50 features filtered by information gain can predict 3-month functional independence with an AUC of 0.875 and 1-year mortality with an AUC of 0.895. We also built an online prediction service, Yonsei Stroke Outcome Inference System, to substantialize the proposed solution for patients with stroke.

Machine Learning in Agriculture: A Review

Machine learning has emerged with big data technologies and high-performance computing to create new opportunities for data intensive science in the multi-disciplinary agri-technologies domain. In this paper, we present a comprehensive review of research dedicated to applications of machine learning in agricultural production systems. The works analyzed were categorized in (a) crop management, including applications on yield prediction, disease detection, weed detection crop quality, and species recognition; (b) livestock management, including applications on animal welfare and livestock production; (c) water management; and (d) soil management. The filtering and classification of the presented articles demonstrate how agriculture will benefit from machine learning technologies. By applying machine learning to sensor data, farm management systems are evolving into real time artificial intelligence enabled programs that provide rich recommendations and insights for farmer decision support and action.

Predicting Language Difficulties in Middle Childhood From Early Developmental Milestones: A Comparison of Traditional Regression and Machine Learning Techniques

PURPOSE

The current study aimed to compare traditional logistic regression models with machine learning algorithms to investigate the predictive ability of (a) communication performance at 3 years old on language outcomes at 10 years old and (b) broader developmental skills (motor, social, and adaptive) at 3 years old on language outcomes at 10 years old.

METHOD

Participants (N = 1,322) were drawn from the Western Australian Pregnancy Cohort (Raine) Study (Straker et al., 2017). A general developmental screener, the Infant Monitoring Questionnaire (Squires, Bricker, & Potter, 1990), was completed by caregivers at the 3-year follow-up. Language ability at 10 years old was assessed using the Clinical Evaluation of Language Fundamentals-Third Edition (Semel, Wiig, & Secord, 1995). Logistic regression models and interpretable machine learning algorithms were used to assess predictive abilities of early developmental milestones for later language outcomes.

RESULTS

Overall, the findings showed that prediction accuracies were comparable between logistic regression and machine learning models using communication-only performance as well as performance on communication and broader developmental domains to predict language performance at 10 years old. Decision trees are incorporated to visually present these findings but must be interpreted with caution because of the poor accuracy of the models overall.

CONCLUSIONS

The current study provides preliminary evidence that machine learning algorithms provide equivalent predictive accuracy to traditional methods. Furthermore, the inclusion of broader developmental skills did not improve predictive capability. Assessment of language at more than 1 time point is necessary to ensure children whose language delays emerge later are identified and supported.

SUPPLEMENTAL MATERIAL

https://doi.org/10.23641/asha.6879719.

Multimodal MRI-Based Triage for Acute Stroke Therapy: Challenges and Progress

Revascularization therapies have been established as the treatment mainstay for acute ischemic stroke. However, a substantial number of patients are either ineligible for revascularization therapy, or the treatment fails or is futile. At present, non-contrast computed tomography is the first-line neuroimaging modality for patients with acute stroke. The use of magnetic resonance imaging (MRI) to predict the response to early revascularization therapy and to identify patients for delayed treatment is desirable. MRI could provide information on stroke pathophysiologies, including the ischemic core, perfusion, collaterals, clot, and blood–brain barrier status. During the past 20 years, there have been significant advances in neuroimaging as well as in revascularization strategies for treating patients with acute ischemic stroke. In this review, we discuss the role of MRI and post-processing, including machine-learning techniques, and recent advances in MRI-based triage for revascularization therapies in acute ischemic stroke.

Simulation of patient flow in multiple healthcare units using process and data mining techniques for model identification

INTRODUCTION

An approach to building a hybrid simulation of patient flow is introduced with a combination of data-driven methods for automation of model identification. The approach is described with a conceptual framework and basic methods for combination of different techniques. The implementation of the proposed approach for simulation of the acute coronary syndrome (ACS) was developed and used in an experimental study.

METHODS

A combination of data, text, process mining techniques, and machine learning approaches for the analysis of electronic health records (EHRs) with discrete-event simulation (DES) and queueing theory for the simulation of patient flow was proposed. The performed analysis of EHRs for ACS patients enabled identification of several classes of clinical pathways (CPs) which were used to implement a more realistic simulation of the patient flow. The developed solution was implemented using Python libraries (SimPy, SciPy, and others).

RESULTS

The proposed approach enables more a realistic and detailed simulation of the patient flow within a group of related departments. An experimental study shows an improved simulation of patient length of stay for ACS patient flow obtained from EHRs in Almazov National Medical Research Centre in Saint Petersburg, Russia.

CONCLUSION

The proposed approach, methods, and solutions provide a conceptual, methodological, and programming framework for the implementation of a simulation of complex and diverse scenarios within a flow of patients for different purposes: decision making, training, management optimization, and others.

Towards phenotyping stroke: Leveraging data from a large-scale epidemiological study to detect stroke diagnosis

Objective

1) To develop a machine learning approach for detecting stroke cases and subtypes from hospitalization data, 2) to assess algorithm performance and predictors on real-world data collected by a large-scale epidemiology study in the US; and 3) to identify directions for future development of high-precision stroke phenotypic signatures.

Materials and methods

We utilized 8,131 hospitalization events (ICD-9 codes 430–438) collected from the Greater Cincinnati/Northern Kentucky Stroke Study in 2005 and 2010. Detailed information from patients’ medical records was abstracted for each event by trained research nurses. By analyzing the broad list of demographic and clinical variables, the machine learning algorithms predicted whether an event was a stroke case and, if so, the stroke subtype. The performance was validated on gold-standard labels adjudicated by stroke physicians, and results were compared with stroke classifications based on ICD-9 discharge codes, as well as labels determined by study nurses.

Results

The best performing machine learning algorithm achieved a performance of 88.57%/93.81%/92.80%/93.30%/89.84%/98.01% (accuracy/precision/recall/F-measure/area under ROC curve/area under precision-recall curve) on stroke case detection. For detecting stroke subtypes, the algorithm yielded an overall accuracy of 87.39% and greater than 85% precision on individual subtypes. The machine learning algorithms significantly outperformed the ICD-9 method on all measures (P value<0.001). Their performance was comparable to that of study nurses, with better tradeoff between precision and recall. The feature selection uncovered a subset of predictive variables that could facilitate future development of effective stroke phenotyping algorithms.

Discussion and conclusions

By analyzing a broad array of patient data, the machine learning technologies held promise for improving detection of stroke diagnosis, thus unlocking high statistical power for subsequent genetic and genomic studies.

Automatic recognition of gait patterns in human motor disorders using machine learning: A review

BACKGROUND

automatic recognition of human movement is an effective strategy to assess abnormal gait patterns. Machine learning approaches are mainly applied due to their ability to work with multidimensional nonlinear features.

PURPOSE

to compare several machine learning algorithms employed for gait pattern recognition in motor disorders using discriminant features extracted from gait dynamics. Additionally, this work highlights procedures that improve gait recognition performance.

METHODS

we conducted an electronic literature search on Web of Science, IEEE, and Scopus, using "human recognition", "gait patterns'', and "feature selection methods" as relevant keywords.

RESULTS

analysis of the literature showed that kernel principal component analysis and genetic algorithms are efficient at reducing dimensional features due to their ability to process nonlinear data and converge to global optimum. Comparative analysis of machine learning performance showed that support vector machines (SVMs) exhibited higher accuracy and proper generalization for new instances.

CONCLUSIONS

automatic recognition by combining dimensional data reduction, cross-validation and normalization techniques with SVMs may offer an objective and rapid tool for investigating the subject's clinical status. Future directions comprise the real-time application of these tools to drive powered assistive devices in free-living conditions.

Prediction of Hemorrhagic Transformation Severity in Acute Stroke From Source Perfusion MRI

OBJECTIVE

Hemorrhagic transformation (HT) is the most severe complication of reperfusion therapy in acute ischemic stroke (AIS) patients. Management of AIS patients could benefit from accurate prediction of upcoming HT. While prediction of HT occurrence has recently provided encouraging results, the prediction of the severity and territory of the HT could bring valuable insights that are beyond current methods.

METHODS

This study tackles these issues and aims to predict the spatial occurrence of HT in AIS from perfusion-weighted magnetic resonance imaging (PWI) combined with diffusion weighted imaging. In all, 165 patients were included in this study and analyzed retrospectively from a cohort of AIS patients treated with reperfusion therapy in a single stroke center.

RESULTS

Machine learning models are compared within our framework; support vector machines, linear regression, decision trees, neural networks, and kernel spectral regression were applied to the dataset. Kernel spectral regression performed best with an accuracy of $\text{83.7} \pm \text{2.6}\%$.

CONCLUSION

The key contribution of our framework formalize HT prediction as a machine learning problem. Specifically, the model learns to extract imaging markers of HT directly from source PWI images rather than from pre-established metrics.

SIGNIFICANCE

Predictions visualized in terms of spatial likelihood of HT in various territories of the brain were evaluated against follow-up gradient recalled echo and provide novel insights for neurointerventionalists prior to endovascular therapy.

Automating Construction of Machine Learning Models With Clinical Big Data: Proposal Rationale and Methods

Background

To improve health outcomes and cut health care costs, we often need to conduct prediction/classification using large clinical datasets (aka, clinical big data), for example, to identify high-risk patients for preventive interventions. Machine learning has been proposed as a key technology for doing this. Machine learning has won most data science competitions and could support many clinical activities, yet only 15% of hospitals use it for even limited purposes. Despite familiarity with data, health care researchers often lack machine learning expertise to directly use clinical big data, creating a hurdle in realizing value from their data. Health care researchers can work with data scientists with deep machine learning knowledge, but it takes time and effort for both parties to communicate effectively. Facing a shortage in the United States of data scientists and hiring competition from companies with deep pockets, health care systems have difficulty recruiting data scientists. Building and generalizing a machine learning model often requires hundreds to thousands of manual iterations by data scientists to select the following: (1) hyper-parameter values and complex algorithms that greatly affect model accuracy and (2) operators and periods for temporally aggregating clinical attributes (eg, whether a patient’s weight kept rising in the past year). This process becomes infeasible with limited budgets.

Objective

This study’s goal is to enable health care researchers to directly use clinical big data, make machine learning feasible with limited budgets and data scientist resources, and realize value from data.

Methods

This study will allow us to achieve the following: (1) finish developing the new software, Automated Machine Learning (Auto-ML), to automate model selection for machine learning with clinical big data and validate Auto-ML on seven benchmark modeling problems of clinical importance; (2) apply Auto-ML and novel methodology to two new modeling problems crucial for care management allocation and pilot one model with care managers; and (3) perform simulations to estimate the impact of adopting Auto-ML on US patient outcomes.

Results

We are currently writing Auto-ML’s design document. We intend to finish our study by around the year 2022.

Conclusions

Auto-ML will generalize to various clinical prediction/classification problems. With minimal help from data scientists, health care researchers can use Auto-ML to quickly build high-quality models. This will boost wider use of machine learning in health care and improve patient outcomes.

Artificial neural network models for early diagnosis of hepatocellular carcinoma using serum levels of α-fetoprotein, α-fetoprotein-L3, des-γ-carboxy prothrombin, and Golgi protein 73

More than 70% of hepatocellular carcinoma (HCC) cases develop as a consequence of liver cirrhosis (LC). Here we have evaluated the diagnostic potential of four serum biomarkers, and developed models for HCC diagnosis and differentiation from LC patients. Serum levels of α-fetoprotein (AFP), AFP-L3, des-γ-carboxy prothrombin (DCP), and Golgi protein 73 (GP73) were analyzed in 114 advanced HCC patients, 81 early stage HCC patients, and 152 LC patients. Multilayer perceptron (MLP) and radial basis function (RBF) neural networks were used to construct the diagnostic models. Using all stages, HCC diagnostic models had a higher sensitivity (>70%) than the individual serum biomarkers, whereas only early stage HCC diagnostic models had a higher specificity (>80%). The early stage HCC diagnostic models could not be used as HCC screening tools due to their low sensitivity (about 40%). These results suggest that a combination of the two models might be used as a screening tool to distinguish early stage HCC patients from LC patients, thus improving prevention and treatment of HCC.

Predicting two-year survival versus non-survival after first myocardial infarction using machine learning and Swedish national register data

BACKGROUND

Machine learning algorithms hold potential for improved prediction of all-cause mortality in cardiovascular patients, yet have not previously been developed with high-quality population data. This study compared four popular machine learning algorithms trained on unselected, nation-wide population data from Sweden to solve the binary classification problem of predicting survival versus non-survival 2 years after first myocardial infarction (MI).

METHODS

This prospective national registry study for prognostic accuracy validation of predictive models used data from 51,943 complete first MI cases as registered during 6 years (2006-2011) in the national quality register SWEDEHEART/RIKS-HIA (90% coverage of all MIs in Sweden) with follow-up in the Cause of Death register (> 99% coverage). Primary outcome was AUROC (C-statistic) performance of each model on the untouched test set (40% of cases) after model development on the training set (60% of cases) with the full (39) predictor set. Model AUROCs were bootstrapped and compared, correcting the P-values for multiple comparisons with the Bonferroni method. Secondary outcomes were derived when varying sample size (1-100% of total) and predictor sets (39, 10, and 5) for each model. Analyses were repeated on 79,869 completed cases after multivariable imputation of predictors.

RESULTS

A Support Vector Machine with a radial basis kernel developed on 39 predictors had the highest complete cases performance on the test set (AUROC = 0.845, PPV = 0.280, NPV = 0.966) outperforming Boosted C5.0 (0.845 vs. 0.841, P = 0.028) but not significantly higher than Logistic Regression or Random Forest. Models converged to the point of algorithm indifference with increased sample size and predictors. Using the top five predictors also produced good classifiers. Imputed analyses had slightly higher performance.

CONCLUSIONS

Improved mortality prediction at hospital discharge after first MI is important for identifying high-risk individuals eligible for intensified treatment and care. All models performed accurately and similarly and because of the superior national coverage, the best model can potentially be used to better differentiate new patients, allowing for improved targeting of limited resources. Future research should focus on further model development and investigate possibilities for implementation.

Artificial intelligence in healthcare: past, present and future

Artificial intelligence (AI) aims to mimic human cognitive functions. It is bringing a paradigm shift to healthcare, powered by increasing availability of healthcare data and rapid progress of analytics techniques. We survey the current status of AI applications in healthcare and discuss its future. AI can be applied to various types of healthcare data (structured and unstructured). Popular AI techniques include machine learning methods for structured data, such as the classical support vector machine and neural network, and the modern deep learning, as well as natural language processing for unstructured data. Major disease areas that use AI tools include cancer, neurology and cardiology. We then review in more details the AI applications in stroke, in the three major areas of early detection and diagnosis, treatment, as well as outcome prediction and prognosis evaluation. We conclude with discussion about pioneer AI systems, such as IBM Watson, and hurdles for real-life deployment of AI.

Use of Machine Learning Classifiers and Sensor Data to Detect Neurological Deficit in Stroke Patients

Background

The pronator drift test (PDT), a neurological examination, is widely used in clinics to measure motor weakness of stroke patients.

Objective

The aim of this study was to develop a PDT tool with machine learning classifiers to detect stroke symptoms based on quantification of proximal arm weakness using inertial sensors and signal processing.

Methods

We extracted features of drift and pronation from accelerometer signals of wearable devices on the inner wrists of 16 stroke patients and 10 healthy controls. Signal processing and feature selection approach were applied to discriminate PDT features used to classify stroke patients. A series of machine learning techniques, namely support vector machine (SVM), radial basis function network (RBFN), and random forest (RF), were implemented to discriminate stroke patients from controls with leave-one-out cross-validation.

Results

Signal processing by the PDT tool extracted a total of 12 PDT features from sensors. Feature selection abstracted the major attributes from the 12 PDT features to elucidate the dominant characteristics of proximal weakness of stroke patients using machine learning classification. Our proposed PDT classifiers had an area under the receiver operating characteristic curve (AUC) of .806 (SVM), .769 (RBFN), and .900 (RF) without feature selection, and feature selection improves the AUCs to .913 (SVM), .956 (RBFN), and .975 (RF), representing an average performance enhancement of 15.3%.

Conclusions

Sensors and machine learning methods can reliably detect stroke signs and quantify proximal arm weakness. Our proposed solution will facilitate pervasive monitoring of stroke patients.

Evaluating the lexico-grammatical differences in the writing of native and non-native speakers of English in peer-reviewed medical journals in the field of pediatric oncology: Creation of the genuine index scoring system

INTRODUCTION

The predominance of English in scientific research has created hurdles for "non-native speakers" of English. Here we present a novel application of native language identification (NLI) for the assessment of medical-scientific writing. For this purpose, we created a novel classification system whereby scoring would be based solely on text features found to be distinctive among native English speakers (NS) within a given context. We dubbed this the "Genuine Index" (GI).

METHODOLOGY

This methodology was validated using a small set of journals in the field of pediatric oncology. Our dataset consisted of 5,907 abstracts, representing work from 77 countries. A support vector machine (SVM) was used to generate our model and for scoring.

RESULTS

Accuracy, precision, and recall of the classification model were 93.3%, 93.7%, and 99.4%, respectively. Class specific F-scores were 96.5% for NS and 39.8% for our benchmark class, Japan. Overall kappa was calculated to be 37.2%. We found significant differences between countries with respect to the GI score. Significant correlation was found between GI scores and two validated objective measures of writing proficiency and readability. Two sets of key terms and phrases differentiating NS and non-native writing were identified.

CONCLUSIONS

Our GI model was able to detect, with a high degree of reliability, subtle differences between the terms and phrasing used by native and non-native speakers in peer reviewed journals, in the field of pediatric oncology. In addition, L1 language transfer was found to be very likely to survive revision, especially in non-Western countries such as Japan. These findings show that even when the language used is technically correct, there may still be some phrasing or usage that impact quality.

Outcomes and Complications After Endovascular Treatment of Brain Arteriovenous Malformations: A Prognostication Attempt Using Artificial Intelligence

PURPOSE

To identify factors influencing outcome in brain arteriovenous malformations (BAVM) treated with endovascular embolization. We also assessed the feasibility of using machine learning techniques to prognosticate and predict outcome and compared this to conventional statistical analyses.

METHODS

A retrospective study of patients undergoing endovascular treatment of BAVM during a 22-year period in a national neuroscience center was performed. Clinical presentation, imaging, procedural details, complications, and outcome were recorded. The data was analyzed with artificial intelligence techniques to identify predictors of outcome and assess accuracy in predicting clinical outcome at final follow-up.

RESULTS

One-hundred ninety-nine patients underwent treatment for BAVM with a mean follow-up duration of 63 months. The commonest clinical presentation was intracranial hemorrhage (56%). During the follow-up period, there were 51 further hemorrhagic events, comprising spontaneous hemorrhage (n = 27) and procedural related hemorrhage (n = 24). All spontaneous events occurred in previously embolized BAVMs remote from the procedure. Complications included ischemic stroke in 10%, symptomatic hemorrhage in 9.8%, and mortality rate of 4.7%. Standard regression analysis model had an accuracy of 43% in predicting final outcome (mortality), with the type of treatment complication identified as the most important predictor. The machine learning model showed superior accuracy of 97.5% in predicting outcome and identified the presence or absence of nidal fistulae as the most important factor.

CONCLUSIONS

BAVMs can be treated successfully by endovascular techniques or combined with surgery and radiosurgery with an acceptable risk profile. Machine learning techniques can predict final outcome with greater accuracy and may help individualize treatment based on key predicting factors.

Changing Management of Acute Ischaemic Stroke: the New Treatments and Emerging Role of Endovascular Therapy

OPINION STATEMENT

Urgent reperfusion of the ischaemic brain is the aim of stroke treatment, and the last two decades have seen a rapid advancement in the medical and endovascular treatment of acute ischaemic stroke. Intravenous tissue plasminogen activator (tPA) was first introduced as a safe and effective thrombolytic agent followed by the introduction of newer thrombolytic agents as well as anticoagulant and antiplatelet agents, proposed as potentially safer drugs with more favourable interaction profiles. In addition to chemo-thrombolysis, other techniques including transcranial sonothrombolysis and microbubble cavitation have been introduced which are showing promising results, but await large-scale clinical trials. These developments in medical therapies which are undoubtedly of great importance due to their potential widespread and immediate availability are paralleled with gradual but steady improvements in endovascular recanalisation techniques which were initiated by the introduction of the MERCI (Mechanical Embolus Removal in Cerebral Ischemia) and Penumbra systems. The introduction of the Solitaire device was a significant achievement in reliable and safe endovascular recanalisation and was followed by further innovative stent retrievers. Initial trials failed to show a solid benefit in endovascular intervention compared with IV-tPA alone. These counterintuitive results did not last long, however, when a series of very well-designed randomised controlled trials, pioneered by MR-CLEAN, EXTEND-IA and ESCAPE, emerged, confirming the well-believed daily anecdotal evidence. There have now been seven positive trials of endovascular treatment for acute ischaemic stroke. Now that level I evidence regarding the superiority of endovascular recanalisation is abundantly available, the clinical challenge is how to select patients suitable for intervention and to familiarise and educate stroke care providers with this recent development in stroke care. It is important for the interventional services to be provided only in comprehensive stroke centres and endovascular interventions attempted by experienced well-trained operators, at this stage as an adjunct to the established medical treatment of IV-tPA, if there are no contraindications.

Automatically explaining machine learning prediction results: a demonstration on type 2 diabetes risk prediction

Background

Predictive modeling is a key component of solutions to many healthcare problems. Among all predictive modeling approaches, machine learning methods often achieve the highest prediction accuracy, but suffer from a long-standing open problem precluding their widespread use in healthcare. Most machine learning models give no explanation for their prediction results, whereas interpretability is essential for a predictive model to be adopted in typical healthcare settings.

Methods

This paper presents the first complete method for automatically explaining results for any machine learning predictive model without degrading accuracy. We did a computer coding implementation of the method. Using the electronic medical record data set from the Practice Fusion diabetes classification competition containing patient records from all 50 states in the United States, we demonstrated the method on predicting type 2 diabetes diagnosis within the next year.

Results

For the champion machine learning model of the competition, our method explained prediction results for 87.4 % of patients who were correctly predicted by the model to have type 2 diabetes diagnosis within the next year.

Conclusions

Our demonstration showed the feasibility of automatically explaining results for any machine learning predictive model without degrading accuracy.

Predicting Appropriate Admission of Bronchiolitis Patients in the Emergency Department: Rationale and Methods

BACKGROUND

In young children, bronchiolitis is the most common illness resulting in hospitalization. For children less than age 2, bronchiolitis incurs an annual total inpatient cost of $1.73 billion. Each year in the United States, 287,000 emergency department (ED) visits occur because of bronchiolitis, with a hospital admission rate of 32%-40%. Due to a lack of evidence and objective criteria for managing bronchiolitis, ED disposition decisions (hospital admission or discharge to home) are often made subjectively, resulting in significant practice variation. Studies reviewing admission need suggest that up to 29% of admissions from the ED are unnecessary. About 6% of ED discharges for bronchiolitis result in ED returns with admission. These inappropriate dispositions waste limited health care resources, increase patient and parental distress, expose patients to iatrogenic risks, and worsen outcomes. Existing clinical guidelines for bronchiolitis offer limited improvement in patient outcomes. Methodological shortcomings include that the guidelines provide no specific thresholds for ED decisions to admit or to discharge, have an insufficient level of detail, and do not account for differences in patient and illness characteristics including co-morbidities. Predictive models are frequently used to complement clinical guidelines, reduce practice variation, and improve clinicians' decision making. Used in real time, predictive models can present objective criteria supported by historical data for an individualized disease management plan and guide admission decisions. However, existing predictive models for ED patients with bronchiolitis have limitations, including low accuracy and the assumption that the actual ED disposition decision was appropriate. To date, no operational definition of appropriate admission exists. No model has been built based on appropriate admissions, which include both actual admissions that were necessary and actual ED discharges that were unsafe.

OBJECTIVE

The goal of this study is to develop a predictive model to guide appropriate hospital admission for ED patients with bronchiolitis.

METHODS

This study will: (1) develop an operational definition of appropriate hospital admission for ED patients with bronchiolitis, (2) develop and test the accuracy of a new model to predict appropriate hospital admission for an ED patient with bronchiolitis, and (3) conduct simulations to estimate the impact of using the model on bronchiolitis outcomes.

RESULTS

We are currently extracting administrative and clinical data from the enterprise data warehouse of an integrated health care system. Our goal is to finish this study by the end of 2019.

CONCLUSIONS

This study will produce a new predictive model that can be operationalized to guide and improve disposition decisions for ED patients with bronchiolitis. Broad use of the model would reduce iatrogenic risk, patient and parental distress, health care use, and costs and improve outcomes for bronchiolitis patients.

Using Computational Approaches to Improve Risk-Stratified Patient Management: Rationale and Methods

Background

Chronic diseases affect 52% of Americans and consume 86% of health care costs. A small portion of patients consume most health care resources and costs. More intensive patient management strategies, such as case management, are usually more effective at improving health outcomes, but are also more expensive. To use limited resources efficiently, risk stratification is commonly used in managing patients with chronic diseases, such as asthma, chronic obstructive pulmonary disease, diabetes, and heart disease. Patients are stratified based on predicted risk with patients at higher risk given more intensive care. The current risk-stratified patient management approach has 3 limitations resulting in many patients not receiving the most appropriate care, unnecessarily increased costs, and suboptimal health outcomes. First, using predictive models for health outcomes and costs is currently the best method for forecasting individual patient’s risk. Yet, accuracy of predictive models remains poor causing many patients to be misstratified. If an existing model were used to identify candidate patients for case management, enrollment would miss more than half of those who would benefit most, but include others unlikely to benefit, wasting limited resources. Existing models have been developed under the assumption that patient characteristics primarily influence outcomes and costs, leaving physician characteristics out of the models. In reality, both characteristics have an impact. Second, existing models usually give neither an explanation why a particular patient is predicted to be at high risk nor suggestions on interventions tailored to the patient’s specific case. As a result, many high-risk patients miss some suitable interventions. Third, thresholds for risk strata are suboptimal and determined heuristically with no quality guarantee.

Objective

The purpose of this study is to improve risk-stratified patient management so that more patients will receive the most appropriate care.

Methods

This study will (1) combine patient, physician profile, and environmental variable features to improve prediction accuracy of individual patient health outcomes and costs; (2) develop the first algorithm to explain prediction results and suggest tailored interventions; (3) develop the first algorithm to compute optimal thresholds for risk strata; and (4) conduct simulations to estimate outcomes of risk-stratified patient management for various configurations. The proposed techniques will be demonstrated on a test case of asthma patients.

Results

We are currently in the process of extracting clinical and administrative data from an integrated health care system’s enterprise data warehouse. We plan to complete this study in approximately 5 years.

Conclusions

Methods developed in this study will help transform risk-stratified patient management for better clinical outcomes, higher patient satisfaction and quality of life, reduced health care use, and lower costs.

MLBCD: a machine learning tool for big clinical data

Background

Predictive modeling is fundamental for extracting value from large clinical data sets, or “big clinical data,” advancing clinical research, and improving healthcare. Machine learning is a powerful approach to predictive modeling. Two factors make machine learning challenging for healthcare researchers. First, before training a machine learning model, the values of one or more model parameters called hyper-parameters must typically be specified. Due to their inexperience with machine learning, it is hard for healthcare researchers to choose an appropriate algorithm and hyper-parameter values. Second, many clinical data are stored in a special format. These data must be iteratively transformed into the relational table format before conducting predictive modeling. This transformation is time-consuming and requires computing expertise.

Methods

This paper presents our vision for and design of MLBCD (Machine Learning for Big Clinical Data), a new software system aiming to address these challenges and facilitate building machine learning predictive models using big clinical data.

Results

The paper describes MLBCD’s design in detail.

Conclusions

By making machine learning accessible to healthcare researchers, MLBCD will open the use of big clinical data and increase the ability to foster biomedical discovery and improve care.

Population-based Stroke Atlas for outcome prediction: method and preliminary results for ischemic stroke from CT

BACKGROUND AND PURPOSE

Knowledge of outcome prediction is important in stroke management. We propose a lesion size and location-driven method for stroke outcome prediction using a Population-based Stroke Atlas (PSA) linking neurological parameters with neuroimaging in population. The PSA aggregates data from previously treated patients and applies them to currently treated patients. The PSA parameter distribution in the infarct region of a treated patient enables prediction. We introduce a method for PSA calculation, quantify its performance, and use it to illustrate ischemic stroke outcome prediction of modified Rankin Scale (mRS) and Barthel Index (BI).

METHODS

The preliminary PSA was constructed from 128 ischemic stroke cases calculated for 8 variants (various data aggregation schemes) and 3 case selection variables (infarct volume, NIHSS at admission, and NIHSS at day 7), each in 4 ranges. Outcome prediction for 9 parameters (mRS at 7th, and mRS and BI at 30th, 90th, 180th, 360th day) was studied using a leave-one-out approach, requiring 589,824 PSA maps to be analyzed.

RESULTS

Outcomes predicted for different PSA variants are statistically equivalent, so the simplest and most efficient variant aiming at parameter averaging is employed. This variant allows the PSA to be pre-calculated before prediction. The PSA constrained by infarct volume and NIHSS reduces the average prediction error (absolute difference between the predicted and actual values) by a fraction of 0.796; the use of 3 patient-specific variables further lowers it by 0.538. The PSA-based prediction error for mild and severe outcomes (mRS  =  [2]-[5]) is (0.5-0.7). Prediction takes about 8 seconds.

CONCLUSIONS

PSA-based prediction of individual and group mRS and BI scores over time is feasible, fast and simple, but its clinical usefulness requires further studies. The case selection operation improves PSA predictability. A multiplicity of PSAs can be computed independently for different datasets at various centers and easily merged, which enables building powerful PSAs over the community.