Creating alerts for asthma patients using a machine learning model

Digital health platforms help to create personalized care experiences for patients with chronic diseases. Patient apps can be created as a customizable mobile application with an AI enabled user interface that keeps patients engaged. It provides an intelligent decision support engine that helps patients follow physician recommended guidelines. The platform provides cloud based data driven machine learning models, powerful data analytics, real time insights via dashboards to optimize remote patient monitoring.

Use cases and examples will be shown to the audience for chronic diseases such as asthma. First, dummy data for Asthma app mobile users will be shown. Next, use of external data from other sources will be explained and described. Finally, use of Machine learning approaches will be explained in Predicting risk of asthma attack.

Each example will highlite different datasets and variables used, analytic approaches considered and its pros and cons, and how machine learning can predict and help in reducing visits to the emergency room or hospital for severe asthma patients.

Data collection:

Data includes patients peakflow, zones for asthma (yellow, green or red), symptoms, medications, symptom severity and answers to a 6 question survey (including number of hospital visits, medication change reason etc) all entered by the patient. External data includes air quality data , geographical location and pollen data.


Machine learning methods work by uncovering hidden relationships between the target and features that classify or predict a particular outcome. In the context of telemonitoring via an app, supervised classification algorithms can be used to yield a classifier that distinguishes between a stable disease state and disease trajectory that it is indicative of incipient exacerbation on the basis of patient characteristics collected during a predefined time frame.

Thus, from a machine learning prospective, telemonitoring data collected on a daily basis may be considered as features and each corresponding day's disposition with regard to exacerbation status (yes or no) can be considered as an outcome for predictive modeling. Within this framework, an initial predictive model can be continuously improved with increased numbers of cases.

Examples of classification algorithms for building classification models include: adaptive Bayesian network, naive Bayesian classifier, and support vector machines. The naive Bayesian classifier looks at historical data and calculates conditional probabilities for the class values by observing the frequency of attribute values and of combinations of attribute values. The second algorithm used, adaptive Bayesian network, was based on Bayesian networks, which use a directed acyclic graph consisting of nodes, where each node represents an attribute. Corresponding to each node are instances with conditional probabilities. The conditional probability of an instance is calculated by the relative frequencies of the associated attributes in the training data. The third algorithm we used is a support vector machine algorithm which uses a subset of training data as support vectors. The support vectors are the closest instances to the maximum margin hyperplane, which provides the greatest separation between the classes. The support vectors are determined by constrained quadratic optimization.

A receiver operating characteristic (ROC) will be shown to characterize comparative performance of classifying algorithms for asthma exacerbation prediction resulting from different training data sets . Our study demonstrates significant potential of machine learning approaches using telemonitoring data for early prediction of acute exacerbations of chronic health conditions.


Outline/Structure of the Demonstration

Background of digital platform and disease management in asthma (10min)

Description of how mobile app data is collected and external and internal sources of data ( 10 min)

Explain ML approaches used for Predicting risk of asthma attack (20 min)

Future work , Conclusions and Questions - (5 min)

Learning Outcome

Participants will learn how machine learning applications can help save a visit to the hospital and hence reduce costs for asthma patients

Participants will understand various ML algorithms and how to deal with data anomalies such as missing/incomplete data for your model

Participants will learn a healthcare usecase for 1 disease and how it can be applied to multiple chronic diseases such as diabetes etc

Target Audience

Those intersted in healthcare use cases for ML, Those who would like to build alerts and predict future risk for diseases in their data

Prerequisites for Attendees

Have basic understanding of Machine Learning algorithms

Download Keva health advisor app from google or apple playstore (useful but not mandatory)

schedule Submitted 9 months ago

Public Feedback

comment Suggest improvements to the Speaker
  • Vishal Gokhale
    By Vishal Gokhale  ~  9 months ago
    reply Reply

    Thanks for submitting your proposal Jyotsna. 

    ODSC India is a practitioner conference, hence it would be useful if you can share the details of ML algorithms (choices, application/adaptation, challenges, solutions) that make keva possible.

    • jyotsna mehta
      By jyotsna mehta  ~  9 months ago
      reply Reply

      Thanks very much Vishal for your comment.  We are a digital startup company and are currently working on these algorithms and prefer not to share something that is not fully baked.  Our hope was to be ready to present by Aug hence we submitted the broader outline of the project given the deadline of 4/15. We are happy to share the ML algorithm details prior to the meeting.

      • Dr. Om Deshmukh
        By Dr. Om Deshmukh  ~  8 months ago
        reply Reply

        Hi Jyotsna,

        Adding on to Vishal's/your comments: 

        What will help is to talk about why you chose a particular set of algorithms to begin with, what led to rejecting some and then the final choice. A focus on aspects specific to your data and your use case which can then be generalized to broader use cases would be welcome too.

        Right now, the abstract reads to generic. In the learning outcomes, if you can focus only on the second and/or third, that will be quite valuable for the attendees.




        • jyotsna mehta
          By jyotsna mehta  ~  8 months ago
          reply Reply

          Thanks very much for your comments. I have picked on 1 use case and modified/expanded on the analytics section and added in the proposal.  Specific details and results will be shown in the actual presentation.  

          • Dr. Vikas Agrawal
            By Dr. Vikas Agrawal  ~  8 months ago
            reply Reply

            Dear Jyotsna: Thanks for your proposal. Will you be sharing the specifics of the end to end process and the machine learning pipeline and algorithms that allow to create such a great system? Could you please add that to the proposal as well? That way potential audience can look at the proposal and decide how the talk is likely to help them to go back and apply the data science techniques to their own work. 

            Usually at ODSC the slides and videos are made public.  

            Warm Regards


    • 480 Mins

      Modern statistics has become almost synonymous with machine learning, a collection of techniques that utilize today's incredible computing power. This two-part course focuses on the available methods for implementing machine learning algorithms in R, and will examine some of the underlying theory behind the curtain. We start with the foundation of it all, the linear model. We look how to assess model quality with traditional measures and cross-validation and visualize models with coefficient plots. Next we turn to penalized regression with the Elastic Net. After that we turn to Boosted Decision Trees utilizing xgboost. Along the way we learn modern techniques for preprocessing data.

    • Liked Nirav Shah

      Nirav Shah / Ananth Bala - Data Analysis, Dashboards and Visualization - How to create powerful visualizations like a Zen Master

      90 Mins

      In today’s data economy and disruptive business environment, data is the new oil and data analysis with data visualization is vital for professionals and companies to stay competitive. Data Analysis and developing useful and interactive visualizations which provide insights may seem complex for a non -data professional. That should not be the case, thanks to various BI & data visualization tools. Tableau is one of the most popular one and widely used in various industries by individual users to enterprise roll out.

      In this hands-on training session you will learn to turn your data into interactive dashboards, how to create stories with data and share these dashboards with your internal/external stakeholders. We will begin with practices for creating charts and storytelling utilizing best visual practices. Whether your goal is to explain an insight or let your audience explore data insights, using Tableau’s simple drag-and-drop user interface makes the task easy and enjoyable.

      You will learn to use functionalities like Table Calculations, Sets, Filters, Parameters and predictive analytics using forecast functions . You will also learn mapping and other visual functionalities. We will demo few charts such as Waterfall charts, Pareto charts, Gantt Charts, Control Charts and Box and Whisker’s plot.

      We will focus on data Visualization workflows and best practices using zen master techniques.