Continued uncertainty and risk in the healthcare industry is pressuring healthcare executives to keep a steady focus on reducing costs and transforming delivery models to improve the patient experience. Artificial intelligence (AI) manifested through machine learning algorithms is already transforming a variety of healthcare applications including medical imaging diagnosis, drug discovery and personalized medicine. The McKinsey Gold Institute estimates that applying big-data strategies to better inform decision making could generate up to $100 billion in value annually across the US health-care system, by optimizing innovation, improving the efficiency of research and clinical trials, and building new tools sets for physicians, consumers, insurers and regulators to meet the promise of more individualized approaches. 

How Machine Learning Analyzes Data 
The healthcare industry sits on a goldmine of data that is generated from many sources:  research and development (R&D); physicians and clinics; patients; caregivers; insurers and others. Healthcare organizations with large networks of web-enabled connected devices have the added challenge of continually capturing and storing immense amounts of data.  
To effectively manage and make sense of these enormous data sets, machine learning is becoming an increasingly important decision-making tool. Data is analyzed in two specific ways to support the process. The first method is to train the machine in an existing algorithm to browse live data faster and more efficiently than people. By combing through data stores, the machine can discover key indicators that map to a specific outcome and bring those items to the forefront. Applications such as predictive health trackers use this method to help monitor patients’ health status using real-time data collection. 
The second method analyzes historical data with specific outcomes and allows the machine learning algorithm to discover previously unknown indicators to predict outcomes. This method, also called predictive analytics, can help monitor patients and prevent emergencies before they occur. For example, health records for veterans returning from active duty could be rapidly scanned to help prevent a variety of possible health concerns, including physical issues, psychological issues and psychosocial issues concerning work and family. 

Practical Considerations for Machine Learning Projects 
Definitive Logic is leading research and development in both areas of machine learning using publicly available health records.  We have built customizable systems aimed at improving hospital performance, physician performance and other key aspects of the patient health process.  
When thinking about a machine learning project at your organization, we suggest 3 practical considerations to get started: 

  1. What Do You Want to Learn?

As simple as it sounds, the end is the best starting point. Know what problem you’re solving and ask the right questions to create the best model for your machine learning project.  

  1. Understand What’s Hard in a Machine Learning Project

A common misconception is that creating machine learning models/algorithms is the biggest challenge. In fact, data management and data cleansing require equal if not more skill and understanding. Poor data management and cleansing can lead to improper conclusions and worse, compromise HIPPA security and privacy.  

  1. Have the Right Resources in Place

Finally, have skilled data science resources that can interpret your data and apply it appropriately to solve problems. Less skilled resources risk “overfitting” data, i.e., producing an analysis that corresponds too closely or exactly to a set of data. For example, overfitted training data may not model real world outcomes because it corresponds too closely to the training model upon which it was built. 
Mike Lanouette is a Technical Architect focused on new analytic solutions at Definitive Logic.  He can be reached at 
Considering a machine learning project? Email us at