Predictive Analytics

Today in New York City, healthcare providers see not only their patients’ pasts, but also their likely futures. They even see the distribution of risk — for chronic diseases, for example — across the 16 million citizens served by Healthix, the region’s largest health information exchange (HIE).

This bridge between past and future — critically important for patient-centered population health management — is the work of HBI Solutions, Inc. and its Spotlight Data Solution predictive analytics engine. Running on the InterSystems HealthShare® Health Informatics Platform, Spotlight identifies elements, or “features,” in health records that could spell future trouble.

Eric Widen, co-founder and CEO of HBI Solutions, took a few minutes to discuss Spotlight and the likely future of care management.

What are typical use cases for Spotlight?
With the shift toward value-based care, providers are on the hook for keeping patient healthy and preventing unnecessary health service utilization. Providers need to know when to intervene and for which patients. Spotlight customers include health systems, physician practices, federally qualified health centers, accountable care organizations, health plans, public and private HIEs and technology vendors.

Spotlight can be wired two ways. For accountable care organizations or insurance panels, it will identify patients who would benefit from care management programs. For a hospital patient, it will predict the potential for readmission and complications like sepsis. And for ambulatory patients, it will predict events like heart attack, stroke or the onset of chronic disease.

Spotlight is underpinned by predictive models running in real time via HealthShare. It includes predictive risk models, dashboards, reports, and scorecards supporting population health, risk management, readmission management and quality improvement.

How do you develop the predictive models?
Our machine-learning processes take all clinical information available on patients to “learn” correlations between certain features in the health record — such as smoking — and risk. We look at age, comorbidities, socioeconomic factors, demographics, medications, and number and type of abnormal lab results. We can augment electronic health record (EHR) data with billing and claims data and use natural language processing to include clinical notes data. There are typically 50 to 200 relevant features for each risk model.

Historically, analytics largely centered on retrospective analytics, usually done to understand past performance. It was always a look backward. Our algorithms allow users to look forward. We’re generating new clinical information that doesn’t exist in the EHRs alone.

There are four major differentiators for Spotlight. First, it works in real time. Some of our competitors work with claims data that’s at best 30 days old — more likely 90 to 120 days old. Second, we provide a service to tune and calibrate the algorithms to the client’s local data set, because risk features vary from one location to another. Third, our methods are published in peer-reviewed journals, and the majority of our employees are computer scientists, statisticians, data scientists and machine-learning experts. Finally, our analytics are live on over 20 million patients.

Why did you select InterSystems HealthShare as your data management platform?
HealthShare Information Exchange and Health Insight bring together all the health information we need for our predictive models within a single environment. HealthShare integrated very well with Spotlight, and we had been talking to InterSystems for a while. We started collaborating for a common client, Healthix, which was already using HealthShare to create communitywide, multi-source health records that did the real-time, up-front integration work we needed to populate our algorithms.

Since then, we’ve expanded our partnership. Now our solution runs natively inside the InterSystems platform, and we can take advantage of HealthShare capabilities, like Information Exchange’s clinical event notifications. These let us send alerts to providers, payers, and care managers when a patient’s risk profile changes.

Also, InterSystems is very partner-oriented and is geared toward ensuring that we are successful. It even flew engineers from Cambridge, Mass., to our Palo Alto offices for face-to-face collaborations.

So, what’s next for HBI?
We’re continually expanding the number and types of risk models, including the development of over 100 new chronic-disease models and an early-warning system for hospital patients who need monitoring. We’ll continue to deliver new models for these areas and explore new predictive needs. With the right data, we can create models for anything predictable, including nonclinical events like missed appointments and bad debt.

We’re also integrating our risk output into care management to better automate workflow. For example, we can automatically place patients on lists or registries based on likelihood of disease or untoward events. Another example of ongoing work is developing algorithms to predict which patients diagnosed with HIV and AIDS won’t follow their treatment plans.

Of course, we’ll also be partnering with InterSystems to help mutual customers predict and manage the future.