About Us
Click this icon for a list of key pages >
More than 25 years of R&D
No quick "Eureka!" discovery, the technology, methodology and testing behind Prevista has been 25 years in the making. Early work included image / target recognition and financial markets. As AI began to develop with sophisticated neural network software and Bart Kosko's fuzzy logic, we began to apply these techniques to real-world applications -- simple at first, then more complex.
Steady improvements in hardware performance and capacity were critical. Genetic programming, for example, is very computationally intensive. Until CPU speed moved into the Ghz range with multiple cores, GP was not practical for many problems. High RAM capacity and solid state storage were also key, to efficiently run large volumes of data through complex AI processes.
We began to apply this AI technology to predict healthcare risk, using standard claims data -- and found that our accuracy was significantly better than the leading products, e.g. ACGs™, CDPS™, Impact Pro™ and DxCG™. Intrigued, we ran tests using the same parameters in the 2007 Society of Actuaries study of risk adjustment products. R2 accuracy scores in this study ranged between 20.5 and 32.1 (20.5% and 32.1%) for the standard prospective prediction of costs for the future 12 months, based of the previous 12 months of claims data. All of the leading predictive / risk adjustment products delivered very similar results, clustered between 20 and 30 percent R2. Accuracy of 30% out of 100% is poor indeed.
Note: See table 1.1 on page 1 of the 2007 SOA study. For a PDF copy of the study click on this link::
We found that PreVista was able to make the same type of 12 month cost prediction with R2 accuracy of 68.5% to 99.7%, depending on the cost bucket, e.g. "$50,000 to $100,000" and "$500,000+". PreVista was also very accurate with higher cost members -- of particular importance for health plans and insurors, and a key weakness with standard risk / predictive products.
In addition, PreVista uncovered unusual results and questions. For example, we found that PreVista could identify groups of "high risk and cost" patients who would improve rather than decline -- contrary to every 'risk adjustment' score. This opens the door to new questions; why will these patients do much better than others with very similar data? In case management, for example, these patients will not need the same level of care and costly resources can be directed to others who will need more management and care.
For our team, this is probably the most interesting aspect of PreVista. What new questions and discoveries will PreVista uncover? What will we find with new EMR data, for example? What secrets are hiding in genomic data?
So, after years of study and testing, we are now we are ready to deliver PreVista to healthcare organizations, to improve the efficiency and effectiveness of the care process.
How can we help you?
East Bay Labs Inc.
East Bay is a 'public benefit corporation', which means that we operate under the rules for a for-profit, but include the public good in our mission and business decisions. Profit is not our primary motivation.
To contact East Bay Labs, please click on:
Mark Hays
CEO
The first IT company I co-founded sold personal computers with the CP/M operating system, back when Bill Gates sold Microsoft Basic and competed with Phillipe Kahn's Turbo Pascal. The chip in my digital cam will store 100,000 times more data than my first personal PC.
So... I have seen many changes in digital technology, the industries that sell it and the customers who use it. There has been one constant: innovation.
For typical CV info, please see my page
on LinkedIn:
Hut 8
This is the name for our central R&D operation. If you can guess the location, we will donate 5% of your initial invoice to a food bank or disaster relief organization in your area. (1% more if you can identify the original location of Hut 8 and why we chose this name.)
Prediction Predictive modeling forecast health risk redmission EMR HIS HIE claims reports acquisition custom accuracy 3M™ McKesson™
Ingenix™ Optum™ Milliman™ MedAI™ ISO™ DxCG™ DRG™ CMS ACG™ Verisk™ pharmacy fraud abuse actuary payer provider STARS healthcare