Machine Learning in Healthcare: What C-Suite Executives Must Know to Use it Effectively in Their Organizations

Machine learning (ML) is gaining in popularity throughout healthcare. ML’s far-reaching benefits, from automating routine clinical tasks to providing visibility into which appointments are likely to no-show, make it a must-have in an industry that’s hyper focused on improving patient and operational outcomes.
This executive report—co-written by Microsoft Worldwide Health and Health Catalyst—is a basic guide to training machine learning algorithms and applying machine learning models to clinical and operational use case. This report shares practical, proven techniques healthcare organizations can use to improve their performance on a range of issues.

The Top Three Recommendations for Successfully Deploying Predictive Analytics in Healthcare

As data availability and open source tools make predictive analytics increasingly accessible for health systems, more organizations are adopting this advanced capability. Organizations won’t, however, use predictive analytics to its full potential—making it routine, pervasive, and actionable—without a deployment strategy that scales the technology.
Three recommendations can help health systems successfully deploy predictive analytics and leverage data experience to improve data-driven interventions and outcomes:

Fully leverage your analytics environment.
Standardize tools and methods using production quality code.
Deploy with a strategy for intervention.

How to Use Text Analytics in Healthcare to Improve Outcomes—Why You Need More than NLP

Given the fact that up to 80 percent of clinical data is stored in unstructured text, healthcare organizations need to harness the power of text analytics. But, surprisingly, less than five percent of health systems use it due to resource limitations and the complexity of text analytics.
But given the industry’s necessity to use text analytics to create precise patient registries, enhance their understanding of high-risk patient populations, and improve outcomes, this executive report explains why systems must start using it—and explains how to get started.
Health systems can start using text analytics to improve outcomes by focusing on four key components:

Optimize text search (display, medical terminologies, and context).
Enhance context and extract values with an NLP pipeline.
Always validate the algorithm.
Focus on interoperability and integration using a Late-Binding approach.

This broad approach with position health systems for clinical and financial success.

Understanding Risk Stratification, Comorbidities, and the Future of Healthcare

Risk stratification is essential to effective population health management. To know which patients require what level of care, a platform for separating patients into high-risk, low-risk, and rising-risk is necessary. Several methods for stratifying a population by risk include: Hierarchical Condition Categories (HCCs), Adjusted Clinical Groups (ACG), Elder Risk Assessment (ERA), Chronic Comorbidity Count (CCC), Minnesota Tiering, and Charlson Comorbidity Measure. At Health Catalyst, we use an analytics application called the Risk Model Analyzer to stratify patients into risk categories. This becomes a powerful tool for filtering populations to find higher-risk patients.

Questions You Should Ask When Selecting a Healthcare Analytics Platform

As vice president of technology for a healthcare IT company, I’m often asked what should be considered when selecting a solution for healthcare analytics. Healthcare organizations have many choices when selecting a healthcare data warehouse and analytics platforms. I advise them to consider the following fundamental criteria: 1) time-to-value (measured in months, not years), 2) experience as a predictor of future success, and 3) extensibility to meet your needs.

The Top 4 Ways De-Identified Data Improves Research

Level 1 of the Healthcare Analytics Adoption Model is providing de-identified data marts and self-service tools. Researchers navigating the complex research process can use de-identified data in each step of the process to increase their chances of having more successful research projects.
Using de-identified data not only removes research roadblocks, but also enables researchers to navigate the four fundamental research steps with more ease. There are four specific ways de-identified data improves research:

Enables early discovery and exploration to test the validity of the hypothesis before committing time and resources.
Enables researchers to explore and refine their cohorts and determine whether the patient population can support the scope of the project.
Enables researchers to put together strong grant applications without having to tax the resources of enterprise data analysts—and without having to wait for analysts to answer relatively straightforward questions.
Enables researchers to come to the IRB with a strong, fully supported application. A data-driven research process ensures that both researchers and IRB reviewers don’t have to waste their time on projects that may not be viable.

Turn Research Into Care Delivery Improvements Using the Research Analytics Adoption Model

Research is a complex yet vital component of improving care delivery, and it can be hindered by a variety of organizational and technical roadblocks:

Insufficient tools and processes
Poor infrastructure
No single source of truth for data

Health systems can overcome these common research roadblocks and turn analytics-powered research into care delivery improvements by using the Research Analytics Adoption model as a strategic roadmap.
The model consists of 8 levels designed to align operations and research priorities:

De-identified tools and data marts
Delivery of customized data sets
EDW-facilitated study recruitment
Centralized, research-specific data collection
Automated research operations reporting
Biobank/genomic data integration
Multi-site data sharing
Translational Analytics

The 3 Challenges of Translational and Clinical Research Data Management and a Strategy to Succeed

Researchers are facing problems with clinical research data management. These challenges include: 1. Accessing healthcare data due to technology barriers, regulatory barriers, and organizational barriers; 2. Inefficient use of time and resources when working with the data because of poor study recruitment, data cobbling with Excel and Access databases, and materials waste when samples can’t be found.; and 3. Translating research discovery into clinical practice because systems aren’t in place to move new best practices into everyday clinical care.

Improving Population Health Outcomes: From Airplanes to Doctors’ Offices

A staple of inflight magazines, the “Best Doctors” ad showcases individual doctors for specialties in healthcare. Yet, there are no “Best Pilots” ads. That’s because healthcare functions as a craftsmanship practice, while aviation operates using a standard of production. The craftsmanship mentality in medicine leads to a wide variation in results for patients, even those facing the same diagnoses. To improve population health systematically, three systems are required: 1. The Best Practice System (including best practices identified and agreed upon), 2. The Adoption System (meaning how those practices are used across the enterprise), and 3. The Analytics System (in part, measuring how well those best practices are being implemented). Taken together, these systems will move healthcare toward an effective system of production and improve outcomes for patients.

The Three Systems Critical to Healthcare Analytics Success

Health systems looking to improve the lives of patients through analytics often face problems that prevent them from making the improvements they desire. But by using the three systems: Analytics, Best Practice, and Adoption, organizations can be successful. The analytics system ensures that data is aggregated, easy-to-access, and distributed efficiently by implementing a data warehouse. The best practice system provides the framework for best practices and baselines, which provide context and actionable insight to metrics provided by the analytics system. Finally, the adoption system consists of a permanent, multidisciplinary team to enact those actionable insights from the best practice system. All three systems together form the base for organizations to make the journey to data-driven improvement successful.

Healthcare Data Stewardship: The Key to Going from Information Poor to Information Rich

Poor healthcare data stewardship is part of the problem for health systems that feel like they are “data rich and information poor.” But this can be fixed two ways: implementing a data warehouse and improving data stewardship. Without appropriate healthcare data stewardship, even the best infrastructures become underutilized and poorly understood by knowledge workers who could be generating value with the data every day. Data stewards will become critical partners to the data warehouse team in creating a thriving user base. They are the data librarians who advise and guide users, and help them get the most value out of the enterprise data warehouse.

Best Solution to Aggregate Healthcare Data Including Clinical, Financial, Research, Population Health, and More

Health systems generate and collect enormous amounts of healthcare data. Health systems also need to analyze the data for many different needs, such as quality improvement, operations, research, and financial analytics. The best solution an organization can use to aggregate all of this data is an enterprise data warehouse with the following five qualities: new source data feeds that can be developed quickly, a flexible architecture, data definitions that match their context, a single source of truth to support all use cases, and customized data access.

Data Collection Tools in Healthcare: What You Need

Data collection tools in healthcare are supposed to make analyzing data from disparate sources easier. But in the real world, we often see stop-gap solutions we call spreadmarts—that conglomerate of one-off Access databases and Excel spreadsheet. There are three big problems with spreadmarts: Data quality (keeping accurate, consistent data); Collaboration (the spreadmarts become yet another silo of data); Security (it’s a challenge to ensure security on free-flowing spreadsheets). The solution is the Instant Data Entry Application (IDEA). With this tool there is no opportunity for manual data entry errors. The application is on a central server enabling collaboration. And security is much more controllable because it sits on the secure server behind a firewall.

Quality Improvement in Healthcare: Where Is the Best Place to Start?

One of the biggest challenges providers face in their quality improvement efforts is knowing where to get started. In my experience, one of the best ways to overcome that “where do we begin?” factor is by using data from an enterprise data warehouse to look for high-cost areas where there are large variations in how health care is delivered. Variation found through the KPA is an indicator of opportunity. The more avoidable variation that is reflected in a particular care process, the more opportunity there is to reduce that variation and standardize the process. Suppose after performing a KPA you discover three areas of opportunity. How do you determine which one to pursue, especially if it’s your first journey into process improvement? The most obvious answer would seem to be the one with the largest potential ROI. That may not always be the best course to pursue, however. You will also want to take into consideration the readiness/openness to change in each of those areas.

Keys to a Successful Health Catalyst Data Warehouse Platform and Analytics Implementation

We have found that many customers have similar questions about how the implementation process works when rolling out a Health Catalyst Late-Binding ™ data warehouse platform and analytics solutions.  So, we thought it would be useful to produce a document that we hope will answer the majority of these and other common questions. The keys for a successful Health Catalyst implementation are outlined step-by-step format.
Pre-step (most important): Identify key personnel resources needed on the health system side, 1) Implementation Planning, 2) Deploy Hardware, 3) Technical Kickoff Meeting with the Client and Health Catalyst Deployment Teams, 4) Access Source Data, 5) Install Platform, 6) Load Data, 7) Install Foundational Applications, 8) Install Discovery Applications, and 9) Install Advanced Applications
At the beginning of the project, Health Catalyst will begin a collaborative implementation planning process resulting in a timeline tailored to each project. Some projects can be accelerated, with the initial phase completed in 90 days.  Your health system will have questions specific to your organization and your circumstances. We are happy to answer those in person.

Selecting a Business Intelligence Healthcare Platform: 3 Criteria to Consider

Healthcare organizations have many choices when selecting a business intelligence healthcare platform. As the Vice President of Technology, I’m often asked what should be considered in that choice. I recommend looking at the Healthcare Analytics Adoption Model. It starts with a foundation of a data warehouse infrastructure and includes other criteria around implementation that can make or break success. The Analytics Adoption Model gives organizations a roadmap for understanding and leveraging the capabilities of healthcare analytics.

Late-Binding™ vs. EMR-based Models: A Comparison of Healthcare Data Warehouse Methodologies

 
Baffled by the options for healthcare data warehouses? Here, Eric compares two models: Late-Binding™ and EMR-based. Many organizations are taking a wait-and-see approach with analytics solutions provided by EMR vendors and other out-of-the-box solutions. In this post, Eric compares two models of a data warehouse: Late-Binding™ and EMR-based. He also outline important factors to consider when planning for long-term success in data warehousing and analytics.