Start Your Search

D. Connors



Author of

  • +

    OS27 - Disrupting Technology into the Next Decade (ID 43)

    • Event: e-Health 2018 Virtual Meeting
    • Type: Oral Session
    • Track: Technical/Interoperability
    • Presentations: 1
    • +

      OS27.02 - Deep Learning Techniques to Improve Patient Care with Neural Networks (ID 572)

      D. Connors, Allscripts Analytics; Denver/US

      • Abstract
      • Slides

      Purpose/Objectives: Neural Networks are biologically inspired learning algorithms. Artificial networks have come a long way and are now considered some of the most powerful and robust learning algorithms deployed in numerous emerging software-based innovations. The goal behind neural networks (deep learning) is to construct a large hypothesis space of functions that contains a good approximation to the underlying function that represents the deterministic behavior of a process that generates the data. In the case of healthcare, the are definitive use cases of applying advanced predictive models using similar patient data to train a neural network. With improved care opportunities that are guided by predictive systems, comes the added potential of better care management including case management, disease mangagement, and high-risk case identification. An example of a neural network is shown below. deepnetwork.png The talk will demonstrate the results of applying deep learning training techniques to patient electronic health data.

      Methodology/Approach: TensorFlowTMis an open source software library for numerical computation using data flow graphs. Nodes in the graph represent mathematical operations, while the graph edges represent the multidimensional data arrays (tensors) communicated be- tween them. The flexible architecture allows you to deploy computation to one or more computer systems with a single API. TensorFlow was originally developed by researchers and engineers working on the Google Brain Team within Google’s Machine Intelligence research organization for the purposes of conducting machine learning and deep neural networks research, but the system is general enough to be applicable in a wide variety of other domains. In the context of this presentation, TensorFlow was applied to patient records that were formulated by the Johns Hopkins Adjusted Clinical Groups (ACG) model. The ACG system measures health status by grouping diagnoses into clinically cogent groups. The goal of the ACG system is to assign each individual a single, mutually exclusive ACG value, which is a relative measure of the individual's expected or actual consumption of health services. ACGs are closely related to many health trends.

      Finding/Results: nn_vs_acg.png Results show nearly 50% reduction of error in cost prediction.

      Conclusion/Implications/Recommendations: Results show promise of neural networks to reduce the mean cost prediction error of patients by over 50%.

      140 Character Summary: In healthcare, there are definitive use cases of applying advanced predictive models using similar patient data to train neural networks.

      Only Active Members that have purchased this event or have registered via an access code will be able to view this content. To view this presentation, please login or select "Add to Cart" and proceed to checkout.

  • +

    RF03 - Meaningful Data Use and Benefits (ID 17)

    • Event: e-Health 2018 Virtual Meeting
    • Type: Rapid Fire Session
    • Track: Clinical Delivery
    • Presentations: 1
    • +

      RF03.05 - Analytics: Driving Outcomes with AI and Machine Learning (ID 502)

      D. Connors, Allscripts Analytics; Denver/US

      • Abstract
      • Slides

      Purpose/Objectives: •Show outcomes of artificial intelligence (AI) initiative to leverage Microsoft Azure solutions for predictive models for patient cost and care •Describe how analytics play a key role in the need for accurate risk stratification, predicting patients with rising costs, and enabling optimized care pathways for improved outcomes •Explain how incorporating insights into actionable information is key to bringing value to existing data assets •Demonstrate how utilizing predictive models can drive business decisions and promote adherence to current, evidence-based guidelines in chronic disease management •Highlight success stories and lessons learned in the race to derive value from big data

      Methodology/Approach: •Followed process of capturing data, creating predictive models, applying these models and using them at point of care •Selected and gathered necessary data based on analytics needs: collected from proprietary data as well as new information ?Proprietary data includes 50 million unique patients, 325 clients, linked cost data on 20 million lives •Incorporated machine learning on large amounts of health data to predict health outcomes

      Finding/Results: •Case studies on tracking and identifying factors for chronic conditions/high resource utilization, diabetes risk, and opioid abuse risk in patient populations •Use of predictive analytics enables us to identify gaps in care, optimize medical decisions, and prioritize highest risk patients into precision-medicine pathways Analytics Results in Chronic Conditions Case *US Chronic Conditions Condition Patients with Condition % Allscripts Lives National Prevalence Stats Annual Direct Cost Estimate Hypertension 11,100,000 30% 29.1% $64.5 billion Hypercholesterolemia 9,300,000 25% 31.7% $30 billion meds alone $400 billion (stroke & MI) Lower Back Pain 4,400,000 12% 12% $40 billion Allergic Rhinitis 3,800,000 10% 8.4% $18 billion GERD 4,200,000 11% 20% $10 billion Diabetes 4,500,000 12% 10% $56 billion Anxiety 3,900,000 10% 18% $42 billion Depression 3,200,000 9% 6.7% $45 billion Predictive Analytics to Identify High Resource Utilizers, based on Chronic Conditions* predictive analytics - chronic conditions map.jpg

      Conclusion/Implications/Recommendations: To grow and develop successful predictive modeling, consider the following future opportunities and vision: •Continue to leverage EHR neutrality and interoperability, promote data governance •Create holistic picture of health, bridging gaps from EHR data, including integration of alternative data (socioeconomic determinants of health) •Develop advanced analytics, sourced from clinical data, geo/social/environmental data, cost data, patient/consumer/social data and pharma and life science data •Use large scale predictive modeling and validation to facilitate precision medicine approach to care, embed into standard EHR workflows •Deliver on outbreak surveillance, risk stratification with longitudinal records, image recognition •Achieve outcomes including reduced cost of care, clinical performance optimization, and recorded impact of current and future therapies

      140 Character Summary: Highlights value of big data, while demonstrating how predictive models can promote adherence to evidence-based guidelines in chronic disease management.

      Only Active Members that have purchased this event or have registered via an access code will be able to view this content. To view this presentation, please login or select "Add to Cart" and proceed to checkout.