Start Your Search

K. Lane



Author of

  • +

    OS30 - Enabling High Value Analytics (ID 44)

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

      OS30.04 - Predicting Personalized Surgical Outcome Through Analytics & Machine Learning (ID 145)

      K. Lane, Techna Institute, University Health Network; Toronto/CA

      • Abstract
      • Slides

      Purpose/Objectives: Up to one quarter of all patients who undergo knee arthroplasty surgery are not satisfied with the operation results and can continue to experience pain and suboptimal joint function. Despite being a common procedure, no algorithms which bring together predictors in a unified model to be used at the point of care exist. Prediction algorithms provide an opportunity to identify the patients who are at risk of poor surgical response. Having these analytical models available at the point of care to predict personalized risk before the procedure enables better case selection, patient preparation and improved clinical outcomes. The arthritis data science team at the University Health Network (UHN) has developed analytical models to predict post surgical pain, function and satisfaction. Each patient assessed is provided with a personalized risk report prior to having the procedure. Machine learning allows for the identification of predictors and enables the data models developed to continually evolve.

      Methodology/Approach: The UHN arthritis data science team leveraged MicrosoftÂ’s Cortana Intelligence Suite to ingest, process and transform relevant data for the creation of predictive models and to visualize the data insights. The team leveraged 4 existing clinical databases with clinical and patient outcome data to derive predictive models for personalized risk, post surgical pain, function and satisfaction. Patient reports were created and provide a real-time personalized visualization of predictive outcomes, risk and contributing factors at the point of care. solution architecture.png

      Finding/Results: Over 9,500 records from four data sets were included in the data analysis to create the predictive models. Data included in the analysis was collected between January 2011 and August 2017. MicrosoftÂ’s Azure and Cortana Intelligence platforms were leveraged to identify the predictors of success or failure of a surgery. In addition, the probability of surgery success, likelihood of post surgical pain, function, patient satisfaction and 30-day hospital re-admission was also derived through the creation of multiple experiments and the development of models. All of the models developed were trained and scored to measure and compare accuracy. The most accurate models were visualized at the point of care by providing patients with a personalized probability of surgical success based on the derived model predictors.

      Conclusion/Implications/Recommendations: Machine learning and predictive analytics provides an opportunity to leverage large data from multiple siloed datasets and focus on outcomes in healthcare. Data scientists require keen business and data understanding to develop and train the most precise predictive models. Data Preparation and quality is essential for accurate processing and transformation.

      140 Character Summary: The arthritis data science team at UHN has developed analytical models to predict the probability of surgery success by leveraging machine learning technologies.

      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.