AI-Driven Predictive Models are Necessary for Effective Population Health

Healthcare is going through a major transformation where providers and payers are focusing on value-based care rather than fee-for-service compensation. This transformation has yielded innovation and adoption of technologies that have proven effective in other industries. Does AI-driven predictive analytics have a place in healthcare like it has in the financial and technological industries?

 

The Complexity of the Problems in Population Health Makes AI a Necessary Tool

At its heart, population health management (PHM) is a multidimensional optimization problem with each of the Triple Aim Population Health Objectives being a variable that directly impact a program’s profits. Providers and payers are navigating an evolving healthcare environment. The complexities that stakeholders in any value-based care organization face are multifaceted. PHM organizations must use limited resources to invest in programs that will be most effective.

  • COST REDUCTION - Prioritized patient lists for case management that take into account readmission risk and member propensity to engage with care managers
  • COST AVOIDANCE - Prioritized patients lists for proactive disease management
  • ER VISITS - Predict the members who will misuse the ER and the types of medical events that will cause them to have an ER visit
  • NETWORK LEAKAGE - Predict which patients are most likely to go out of network and which providers and procedures are most susceptible to network leakage
  • QUALITY MEASURES - Predict which members will be non-compliant for each quality measure and payer
  • ONCOLOGY MANAGEMENT - Predict which oncology members will be more vulnerable to ER visits and ER admissions after chemotherapy
  • MEDICAL ADHERENCE - Predict which members will be non-compliant for proactive interventions

“The application of machine learning to patient risk stratification advances the precision of stratification and presents a more holistic profile of risk” as compared to rule-based methods.”

- IDC PERSPECTIVE DECEMBER 2017 MACHINE LEARNING IMPROVES PRECISION OF RISK STRATIFICATION

 

Considerations for Using AI-driven Predictive Models in Value-Based Care

It is our belief that value-based care is an environment that is ideal for AI-driven projects. Healthcare organizations have a mass of unstructured data that include clinical and non-clinical factors. Aggregating this data and making sense of it is a problem best suited for machine learning. We have found that our customers embrace a different mind-set once they see the value of analytics. They base all their decisions on data. The problems that healthcare organizations encounter are multi-dimensional and often require non-intuitive decision making. AI is the solution this market needs.

VitreosHealth has 8 years of experience in this field, and our expertise spans a multitude of types of organizations. There are several problems with the way most vendors approach AI models for population health. We think that there is too much focus in the market on big data and not enough on dark data. Claims and EMRs are hiding a wealth of data that most of our competitors can’t even extract, let alone use to build more accurate models. We also see vendors that tout that their models incorporate 100,000 different variables. In our experience, it is more important to choose the right variables for your population. Many organizations want to build one perfect model that will help them in all their population health efforts, when in fact, they should be building an ensemble of models that each have an output that corresponds with your population health programs.

That being said, our approach to using AI effectively is three-fold:

  1. We utilize an AI-based ETL process to create a 360˚ Member Record
  2. Our data scientists and clinical analyst identify the right features or input variables using a multi-dimensional OLAP model. This ‘feature extraction’ is a very important step in building and training an AI model.
  3. As we get more data from your population, we aim to maximize predictive and prescriptive accuracy by implementing machine-based learning

 

Answering the WHO, WHY, and WHAT is Integral in Improving Triple-Aim Objectives

Our AI Models can improve patient experience, reduce cost of care, and result in better health outcomes. We do this by training your AI models on your localized data and helping you answer the following questions:

  1. WHO are the riskiest members in your population? We give you prioritized patients lists that rank your whole population in-terms of predictive risk. We want to show you who is at-risk for avoidable emergent events.
  2. WHY are these people at-risk? We can give you a 360-degree member view and highlight the features that resulted in the member being prioritized. That way, your care managers are better educated in the story of that member.
  3. WHAT can your care management team do about it? Our AI models give your care managers a prioritized list of recommendations for each member. These recommendations become part of each member’s care plan.

 

How do Our AI Models Work?

  1. We use AI to revolutionize the ETL process and incorporate disparate data sources to build a 360-degree member record.
  2. We build a multi-dimensional OLAP model to give data analyst the ability to drill down into the data.
  3. We use machine learning to maximize our model’s predictive and prescriptive accuracy, so we can tell you with a high-level of confidence which patients you should be targeting.

 

How Predictive Accuracies impact ROI of Population Health Programs?

Our predictive models incorporate a multitude of data sources including claims data, HRAs, EHR, and external data. Being able to ingest these sources of data allows us to increase the predictive power of our models. We have noticed that there has been a lot of interest in utilizing the social-economic determinants of care in the market in the past few months. We have been using socio-economic data for more than 8 years.

Every data source added onto claims data increases the predictive power of our models. This is one of the main reasons we have a predictive accuracy that is 2-3 times better than our competitors.

VitreosHealth’s AI-driven customized models are delivering accuracies that are 2-3 times better than those of CMS HCC, CDPSRx, etc. These accuracies translate to accomplishing the same results in outcomes with half the number of care managers. See Figure 4 to illustrate the ROI in care management resources to achieve a targeted savings through proactive member engagement.

VitreosHealth’s AI-driven customized models are delivering accuracies that are 2-3 times better than those of CMS HCC, CDPSRx, etc. These accuracies translate to accomplishing the same results in outcomes with half the number of care managers. See the below image to illustrate the ROI in care management resources to achieve a targeted savings through proactive member engagement.

Care management productivity is directly tied to predictive accuracy. In this graph, I am trying to illustrate how four different models with different predictive accuracies can give you different results. Suppose you only have the resources to target 20% of your population for proactive outreach. You would want to choose a model that helps you pick the top 20% of members that are most at risk for avoidable emergent visits. If you choose which members to focus outreach efforts on by using CMS HCC, you will most likely only prevent 15% of avoidable ER events, while if you used Vitreos Regression models, you could prevent around 38% of emergent events. Our new AI models will give you the ability to prevent 65% of ER events. A perfect model (powered by whatever deity you choose to believe in) will probably help you prevent 85% of emergent visits (in our experience, the top 20% of members result in 85% of avoidable cost). 

For a typical Medicare Advantage Plan with 10,000 members, preventing 40% of ER events results in $6 million in cost savings. You need a smaller care management team to achieve that $6 million if you are using higher accuracy AI predictive models (since you only need to touch around 10% of your population) as compared to HCC models (since you need to touch about 40% of your population). In the above Medicare Advantage example, you need a care management team with 17 members if you are using CMS HCC risk stratification compared to only 8 care managers if you are using Vitreos AI-based risk stratification predictive model.

We improve population health outcomes through innovative data warehousing, analytics, and improvement services.

Ready to Talk