Finding your way into Advanced Analytics and AI, a CIO’s Guide

As hospitals settle into their new EMRs, ERPs, and other enterprise applications, the requests for data are exploding.  Finding the insights and value through innovative and valuable ways is becoming the priority but it is also challenging.  Now, more than ever, CIOs and CDOs need to be deliberate in creating a data strategy that encompasses the advanced analytics and artificial intelligence capabilities and competencies.  One approach is to start with the end in mind, thinking from the perspective of outcomes versus pure technology.

In the past year, we have worked with the leaders of several different hospital systems to build their artificial intelligence strategies.  As you can imagine, each hospital had a different approach, level of maturity and mindset in terms of getting a strong strategy in place.  Some examples of the directives:

• Our executives want to be in AI within 12 months. How do we do that?
• We have talented developers and data scientists. How do we bring our capabilities and competencies to the organization’s attention?
• We’ve been on our new EMR for 12 months, and our providers want to do some predictive analytics with the data. How do we go about valuing these ideas and prioritizing?
• We have dozens upon dozens of data marts all over the organization, and most aren’t being used. How do avoid building complex advanced analytics that don’t get used months later?
• We have gaps in our data. How do we ingest and incorporate some of the smaller data sets including external data sets?

While these organizations have different reasons for engaging on a data strategy, there were some common themes:

• Most of our data is well governed. It’s of high integrity, consistently refreshed, and well defined.
• Requests for data analytics are coming fast and furious, faster than can be kept up with.
• Most requests for data are one-off’s. They are seemingly unrelated to the other requests in our queue.
• Is our organization ready to embrace this technology?
• How do we make our intake-prioritization-development-release process more strategic?

In this post, we will address those items and provide a framework that will help you organize your assessments and journey into a data strategy.

A Healthy Look into the Mirror

Regardless where you are on your analytics journey, there can be a lot of value in assessing the major areas that make up a strong analytics program.  The roadmap depicted below focuses on the areas of Culture, Use Cases, People, Governance, Technology, Change Enablement, and Responsibility and Ethics.  These fundamental components are critical to building a highly functioning advanced analytics capability.  We will highlight a few of these.

Consider a Data First strategy by incorporating your enterprise analytics into your source systems build and deployment. Advanced Analytics projects are different from most other IT projects.  When upgrading your EMR, changing out access points, or moving to a new talent management system, success can be achieved with well defined requirements, project management, etc.  With Advanced Analytics, the ideas are a hypothesis, far from guaranteeing value or success in every project.  A culture that embraces “batting average” success rates can sustain the ebbs and flows of some ideas working, while others are flawed. It is important to celebrate the successes and learn as much as possible from the failures. This is a fail fast, fail forward opportunity.

Use Cases for Advanced Analytics may seem like an obvious important component, but there are two major dimensions to consider.  First, how are you engaging the organization to participate?  Ideas for advanced analytics should be coming from all over – administration, clinical operations, research, patient experience, etc.  The goal is inclusion, so you want to educate on what advanced analytics are, explain the reality versus the hype, how it can be useful, and how to engage with ideas. 

Next comes the structure and discipline in developing those use cases that can be a differentiator for your organization.  First, all use cases should be rooted in a value statement.  How will this project impact the organization?  How will it be measured?  When you are able to quantify your value statements, you are able to remove the bias and gut-feelings that allow for failing projects to continue longer than they should.  Use cases should also be developed with iterative goals.  Let’s say the use case is to predict a patient’s propensity to pay, helping you better budget bad debt.  The first development efforts may be to generate a report looking at historical bad debt by inpatient vs. ambulatory, another report looking at diagnosis and procedures, and yet another report looking at the patient profiles of employer, payor, etc. 

In analyzing those reports, you then may be able to identify potential variables to test with regression analysis, paving the way for the start of machine learning. Using this method may also uncover gaps and additional data requirements that would make your insights more relevant, accurate and valuable.  Those iterations are important for all of the reasons the Agile Methodology has gained so much traction, getting tangible data to the customers quickly to allow for regular evaluation of the project. Pace matters today. The need may come and go very quickly. Speed to value must be considered in your strategy.

A mature governance process is a must have.  While this may be the dullest topic and feel a bit like a unicorn, this function is responsible for allocating funding, evaluation of use cases, prioritization, project initiation, project evaluation, and continuing to evolve the overall strategy. The challenge is that you can’t wait for perfect governance at the peril of delivering value to the organization. Consider an approach that focuses a on well governed versus highly governed approach. Some organizations have built entire data governance departments and teams that could grind your speed to value to a halt. Balance the components of data governance with the pursuit of value and delivery of insights to the business.

When it comes to Responsibility and Ethics, transparency is paramount.  It is important that the insights being generated can be trusted.  Trust in artificial intelligence comes in a few ways.  First, ensure that the data items being used are ethically acceptable.  For example, using Race as a variable to gain insights on patient experience would violate ethical use.  However, using Race in research may be completely appropriate.  Data context is also important.  Understanding in what context the data was collected may have a large impact on how it should be used in machine learning.  Using Blood Pressure that was collected after administering a drug versus collecting Blood Pressure when a patient is at rest, has very different contexts.

Think Outcomes versus Technology

Analytics are different than process-based systems.  Process-based systems, like the EMR and ERP, are required use for those trying to do their job functions.  The use of analytics is generally above and beyond the day-to-day responsibilities; consequently, the analytics need to provide extra value so that the extra time spent at utilizing these insights is impactful.  The incorporation of third-party data into process-based systems may require effort that would be better spent on an enterprise approach incorporating information from many sources.  For example, patient scheduling optimization could benefit from telephone systems information, weather, or patient tracking information.  Ingesting that into the scheduling system may be somewhat counterproductive versus utilizing an enterprise analytics platform.

Be mindful of the bias to consider technologies first when starting an advanced analytics project.  This can lead down a path of placing the focus on the tools.  The organization needs to understand the technological capabilities, how natural language processing can unlock dark data or how image recognition can interpret scans, but the value to the users of advanced analytics is the outcomes.  Machine learning and deep learning help make a difference. They can speed up enrollment for clinical trials, individualize patient profiles to improve their experience, assist recruiting in matching candidates to appropriate positions … the opportunities are broad. With an outcomes mindset, the technology teams can focus on the mission and making a difference.

Michael Antonoff

Michael Antonoff has accumulated broad experiences during his 16 years in leadership roles.  He currently is vice president of advanced analytics and AI at The HCI Group: A Tech Mahindra Company.  Previously, he worked at The University of Texas, MD Anderson Cancer Center, in Houston, Texas, as executive director, Enterprise Business and Data Services.  In this role, he was responsible for the institution’s non-clinical systems and for building the new Big Data department for cognitive computing.  In 2017, he partnered with two physicians to start a new healthcare company, Patients We Share.  PWS is a patient navigation system designed for everyone involved in patient care to improve his or her efficiency and effectiveness in referrals, treating patients, and communicating with other providers about patient care. 

Chris Belmont

Chris Belmont, executive vice president of operations and strategy at The HCI Group: a Tech Mahindra Company, has more than 35 years of leadership in some of the world’s most respected patient care organizations. He is the former senior vice president and CIO of MD Anderson Cancer Center in Houston, ranked the No. 1 cancer center in the nation by U.S. News & World Report.  He also previously served as CIO and senior vice president of Ochsner Health System, Louisiana’s largest nonprofit, academic, multispecialty healthcare delivery system. In addition, he has held senior IT positions at top healthcare technology companies, including Siemens Medical Solutions and IBM Global Business Solutions as well as CEO of Intelligent Retinal Imaging Systems.