Data savvy precision health reinsurance analytics company leverages to access all sensitive health data at scale to vastly increase predictive ability value to self-insured employers to determine cost of healthcare for their people

Mission: To aid self-insured employers to cut costs and preserve vital class leading healthcare benefits for employees.

Problem: No safe access to vast data at an individual level to properly assess risk

Skona Re, a data-driven company specializing in managing healthcare insurance risk  for self-funded employers, is unable to timely obtain and analyze good data at scale. As health care costs continue to rise and spiral out of control, companies, especially self-insured employers, highly value any guardrails that they can place on these costs in order to deliver these important benefits to their people. Skona’s data security leader, Jonathan Harker, Head of Data Engineering, like many data professionals, finds the process time consuming and costly. Pushed by the company’s CEO to innovate faster to better address market opportunities, he was unable to improve workflow without throwing money and bodies to improve the data security gaps for received data. Even then, the amount of data was insufficient for their data 

First, there was a painstakingly slow legal process of trust. It took several months to get through the contract process that apportioned legal liabilities and indemnities if sensitive customer data that Skona received from its clients or business associates was breached. Then there were the periodic audits to make sure that the data received was kept in secure databases. This led to complaints from the data science and ML engineering teams who wanted to visualize and work with the data freely, to gain insights and build their models. “We want to play with this data in a safe sandbox”, said Ginger Bennigan, the irascible Software Development Manager at Skona, “These security measures are cramping our style! Our people want data of a portable nature to use on their laptops, so that they can work on this data at home.” “Don’t get me started on the password keys and access controls”, she added, “You’re limiting our access to some of the data. Don’t complain if our predictions are subpar!”

Second, there was the poor quality of the data that Skona received. The emphasis on security inevitably led to manipulated and historically old and stale data being shared between Skona and its clients; it did not produce the best models for predicting with most recent and comprehensive data sets. This was a constant source of frustration for Ginger and her team who were constantly blamed for the underperformance of their models and the poor predictions they produced. They on the other hand blamed the low quality data received from their clients. This back and forth finger pointing had created an impasse and some bad blood between Ginger’s team and the production engineering team on the client’s side. This adversely affected the spirit of collaboration when the ML model developed by Ginger’s team had to be deployed in the production environment on the client’s end and the teams had to work together.

Value of Democratizing Data

Jonathan Harker decided to try the tomtA atomic anonymizer. He downloaded tomtA’s on premise version for Windows and used it to invite Maria Antonelli, Chief Data Officer for the client to share data using tomtA’s atomic anonymizer filter, as well as invite the business associates of client to share the claims, provider, pharma and lab/biometric information of client’s employees. Based upon this information, Skona provided employer stop-loss health insurance. Both Jonathan and Maria had been suffering from the issues of sharing sensitive Agrinew employee health information—testudineous legal procedures and engineering impasse. The client’s employee health data included PII such as:

  • First and last name
  • Telephone number
  • Address
  • Social Security Number
  • Race
  • Age
  • Sex
  • Weight
  • Height
  • BMI
  • Blood pressure from last 4 doctor visits
  • Results from last 4 medical tests
  • All previous diseases
  • All previous medical procedures

This claims, provider, pharma and lab/biometric information constitutes highly regulated personal health information under HIPAA, and is only available to Skona if it enters into a business associates agreement.

Using tomtA, Skona eliminated the trust problem inherent in sharing sensitive data once and for all after the client was on-boarded by tomtA, which includes a review of tomtA’s technology and practices by client’s data security, compliance and legal teams. Now Maria could set up the tomtA atomic anonymizer in a private cloud to anonymize the latest sensitive employee health data and within a week’s time place the resulting anonymous data in Skona’s Anonymized Database. Not only was the tiresome two-month long legal process of establishing trust, apportioning legal liabilities, and follow-up audits eradicated, but also the reduction in time allowed for ML model building using fresher data. Thus in one fell swoop, two of the biggest impediments to a quick and good ML model creation had been dispatched. Maria Antonelli informed Jonathan Harker that the “ball was now in his court.”

Developing the Model

An excited Jonathan Harker could be seen speaking with Ginger Bennigan, Skona’s Software Manager that afternoon. “You mean we can take that entire dataset from the Anonymous Database and start playing with it in our sandbox without worrying a fig about security?” she asked. “Precisely!” said Jonathan, smiling.

Jonathan wanted to try out something else, something new. He knew that once Ginger’s team had finished building the ML model, they would need to deploy it in the client’s production environment. The deployment process was fraught with both technical and personnel problems, especially with the waning trust between the two engineering groups involved in the deployment process. Avoidance of any friction was dearly wished for by all.

In Production

tomtA’s “round-trip” capabilities would keep the deployment and the performance monitoring of the ML model all on the safe side – with Skona’s engineering department. It made perfect sense that the engineers who had created the ML model would be the ones best equipped for running it and monitoring its performance. If the performance cratered, then they would know best on how and when to retrain it. Ginger was elated with tomtA’s “round-trip” capability. She and her team were able to update the Skona-client anonymized database with predictions from their ML model within an hour of its availability. Using tomtA’s reverse filtering aka de-anonymization capability, the client production team under the supervision of Maria Antonelli could then suck up this public data from Skona and convert it to results for client employees. 

Happily Ever After

Over the next week Skona and Agrinew Engineering made sure that the entire “round-trip” functionality was automated and did not require any human intervention to function seamlessly. Furthermore, a process was set up whereby a periodic monitoring of Skona’s ML model and its updating if necessary could be done using the “Performance Monitoring” capability in tomtA. Business Development Director, Shin Watnabe is leading the effort on a new product offering from Skona based on this functionality, which he calls a “working prototype”.

Using tomtA, Skona increased critical accuracy and time to market in weeks instead of months or years by accessing prior inaccessible AI talent to address disruptive competition threats to its revenue. Important success metrics achieved:

  • Time to access needed sensitive data = A week instead of several months
  • Legal costs related to accessing needed data = 90% less (smart contracts)
  • Data analytics and ML model time to market/prediction = 3X Faster 
  • Freshness of data = one week or less instead of 2 months

The names and identities of the participants of this case study have been anonymized with the tomtA atomic anonymizer. Although the case study remains highly accurate, tomtA has safely preserved their identity.