I recently read the Harvard Business Review article To Survive, Health Care Data Providers Need to Stop Selling Data (read that here). Harsha Madannavar, Todd Clark, and Joseph Johnson, the authors of this article, claim:
“Most data-driven healthcare IT (HCIT) providers aren’t going to survive. Their business models are at serious risk of failure in the next three to five years. To beat those odds, they need to evolve dramatically, and fast, to a point where they are not selling data at all… A better option is to evolve from providing data to providing insight”.
At VitreosHealth, we learned this lesson the hard way. We were one of the first vendors that brought predictive analytics to population health. Our goal is, and has always been, to predict, intervene, and avoid poor outcomes for members under risk-based or capitated programs. When we first started:
- We built customized predictive models for specific use cases like predicting avoidable emergency room visits, avoidable hospital admissions, and total costs for population health (just like we do now)
- We used local social determinants of health data along with medical claims, HRAs, and lab and clinical data from EMRs (just like we do now)
- Our predictive models had accuracies that were 2-3 times greater than the traditional rules-based risk stratification scores and actuarial-based scores that were embedded in EMRs and care management workflows (our accuracies are even higher now because our machine-based learning models are trained on very large data sets)
- We delivered these predictive models by leveraging an enterprise software delivery model (we do NOT do this anymore)
In 2013, we completed our pilot project for GlobalHealth, an Oklahoma based Medicare HMO that has a leadership team that values innovation and has aggressive plans for growth. The pilot results exceeded their expectations and we successfully delivered the predictive models to the customer (in enterprise software format), trained their users, developed online training materials, and implemented the solution leveraging cloud infrastructure. Two months later, we got a call from their Chief Operating Officer, David Thompson asking us to take back the enterprise software application. We were in a state of shock. When we asked why, he listed off the following reasons:
- He had no one to run this software application. He said they hired talented IT business analysts with clinical domain knowledge, trained them and in two months they got amazing employment offers from the higher-paying financial institutions across the street. We started hearing similar stories across the country from our other healthcare customers.
- The care managers (users) were overwhelmed with the data and the different risk scores we presented. CMs are typically nurses, MAs and PAs, and are not exposed to business intelligence (BI) and analytics applications. They were not able to use all the functions in the application, assimilate all the data, and use it to make the right decisions for their patients. They asked us questions like, “What does a high clinical risk score mean?” and, “What went into the predictive modelling?” Predictive models were a black box, and CMs spent a lot of time just trying to understand why the member had high risk scores even before reaching out to them. This negatively impacted their productivity.
- The end users wanted more. They did not just want us to provide a list of who were the “Rising Risk” members, but also wanted to know why these members were at risk and what the best course of action was regarding appropriate intervention. They wanted Prediction, Prescription, and Precision when it came to their analytics.
We also learned that the complexity in monolithic enterprise analytical applications was reducing their effectiveness; it raised more questions than answers. They told us they were un-interested in enterprise software because of the long implementation timelines, high costs, and immense investment in user training.
Bottom line, we were providing DATA which was almost useless. They wanted INSIGHTS that were actionable and digestible. This single event sparked a massive change within VitreosHealth.
The Transformation Process
This crisis became a call for transformation to change our business model. Considering the fragmented nature of our healthcare system, it was clear that healthcare customers cannot afford AI-based predictive analytics delivered in the traditional enterprise software model. Our internal mantra was, “enterprise healthcare analytics software is dead”. In the era of smartphones and lightweight Apps, could we deliver this intelligence outside of the enterprise application? We went back to the drawing board to come up with a new solution delivery model for predictive analytics.
We asked our customers about their ideal model of delivery. They described their challenges and their needs. One of the customers told us they wanted the Uber of analytics for population health. This reminded me of professor Theodore Levitt’s quote:
“People don’t want to buy a quarter-inch drill, they want a quarter-inch hole”.
We decided that to survive, we needed to deliver healthcare predictive Insights-as-a-Service (IaaS) applications for different users within the enterprise – care managers, physicians, quality managers, executives. This means we needed to learn how to deliver healthcare intelligence as a service.
Productizing Predictive Insights-as-a-Service
Our mission was to productize our predictive insights. We started by understanding what the key components of a consumer product are and applying this framework to specific use cases for value-based care. For the sake of illustration, let’s try to productize the insights from a Medicare predictive model that identifies rising risk members for avoidable Emergency Room (ER) visits.
Products have Specific Consumables. In our case, the consumable was monthly lists of prioritized, predicted members who were at high risk of avoidable emergency room visits and admissions in the next 6-12 months. This consumable was delivered via the Care Manager Insights application, a 3-click, read-only application that tells care managers who the risky members are, why they are risky, and what actions to take through a personalized care plan.
Products have Measurable Value. To track the performance of the Care Managers, and also keep improving our predictive models, we had to track our results. Internally, we follow a strict process of reviewing results every month and give constructive feedback to the modelling team on how we can improve the models. For that, we designed the Performance Insights Application as a 3-click, read-only application. We even developed a Predictive Accuracy Performance Scorecard which gives our customers the ability to measure how many of those predicted members were contacted, what gaps-in-care were closed, and measure the outcomes monthly.
Products have Tangible Benefits. Our goal was to deliver prioritized lists that are 2-3 times more accurate than formula-based risk scores and half the cost of building customized models internally. We need to be able to stand by our results. So, we started offering outcomes-based pricing and sharing in the risk with our clients.
Products deliver a Consistent Experience. A McDonalds Big Mac tastes the same no matter where you bought it from. So, our Care Manager Insights and Performance Insights application had to be consistent. We provide standard, compelling, and actionable storyboards which have a real impact on outcomes.
Product Packaging. We worked hard to make sure our applications look and feel the same every single time while also staying relevant month after month. All the Insights reports, dashboards, and scorecards have a standardized theme. This took us months of iterations to get this right. It really was the most difficult part. Our unique Insights reports and dashboards (2×2 Quadrant chart, mover analysis, waterfall charts) became iconic to VitreosHealth.
Products have Emotional Appeal. We wanted to make sure the customers had a positive experience every time they saw Vitreos predictive insights. This means that we need to provide Insights that were non-intuitive and high impact. We started delivering the Needle-in-the-Haystack Member List, which highlights members who have significant changes going on in their risk and utilization and need quick communication with a care manager. We started getting comments like, “Wow, we have no idea that this was happening in our population”. We leveraged learning across our customers to show them opportunities in their own populations even before they asked us. If we built an Opioid Abuse or New Enrollee predictive model for one Medicaid health plan customer, we would run it for all our Medicaid customers even before they asked at no charge. We believe, if we take care of our customers and impact their outcomes, they will take care of us. It shows in our growth in revenues per customer over time.
Products have Value-Based Pricing. It must be affordable. If we must win big against the giants like IBM Watson Services and Optum Health, we needed to have a better cost structure, be nimble and customer-centric, and deliver the solution at half the cost and in half the time. That was the goal from day 1. We realized that most of the health care providers and payers will not be able to pay the huge price that vendors like IBM Watson and Optum Health need to charge to cover their huge overhead and infrastructure costs.
Product have Strategic Value. We started having annual meetings with the executive leadership teams of each of our customers to understand their goals, understand areas of improvement, build predictive models that had the biggest impact, and provide insights to support the population health programs for the coming 12 months. Our governing mantra became ROI. We deliver standard monthly risk and outcomes variance analysis reports against their plan to track our performance towards the year-end goals.
The journey was painful. Customer like GlobalHealth have been our partners throughout the process. Their results are a validation of our business model. GlobalHealth doubled their revenues over the last 4 years. Scott Vaughn, CEO of GlobalHealth, says:
“We use VitreosHealth Insights to fuel our MLR reduction program, and so far, we’ve reduced per member per month (PMPM) cost for the target population by 16 percent and decreased emergent admissions by 18 percent and readmissions by 22%. Overall reduced MLR from 93% to 86%. Increase Medicare Advantage membership by 24%”
Accepting Insights-as-a-Service as a better delivery model is difficult for individuals who have grown up believing in the duality of enterprise software versus consulting service. The world is changing; it started with software-as-a-service, storage-as-a-service, infrastructure-as-a-service, and now it is moving to “Intelligence-as-a-Service”, otherwise called Insights-as-a-Service. The game is on for true disruption in healthcare, and I really think the market dynamics are well-aligned for this sort of change. The IaaS model provides predictive intelligence competencies to providers and payers who could not have access to such capabilities through the traditional enterprise software application approach. Predictive Insights-as-a-Service (IaaS) is becoming the great equalizer by helping mid-market payers and providers compete with large health plans and integrated delivery networks in the growing value-based care market.
See how our IaaS delivery model works: