Healthcare analytics company Cotiviti has launched a COVID-19 tracker to predict infection outbreaks based on insurance claims data. It says its model has a high degree of accuracy and is useful for health ecosystem players and government authorities that need to decide where to allocate resources.
In this interview, I asked Cotiviti EVP, Jordan Bazinsky to explain.
- What are the needs for COVID-19 tracking in the US?
To make decisions on how to best protect their citizens and employees while also limiting harm to their economies, policymakers and business leaders need accurate and timely data about how far COVID-19 has spread in their communities. But unfortunately, we’ve seen that in too many areas, COVID-19 tests are not being administered to everyone who should receive one due to lack of resources. Therefore, in the absence of this testing, we sought to develop a model that would help to forecast which geographic areas were likely to see a substantial number of COVID-19 cases using other factors such as flu testing and flu diagnoses.
- How could this model impact/shift the U.S. approach to combating this health crisis (at private, federal, and state/local levels)?
We’re already seeing many states and local governments move to re-open their economies even as COVID-19 testing remains scarce. While this is understandable given the economic devastation this pandemic has caused for many, these decisions should be guided by accurate data to protect those who are most vulnerable. We hope this approach will encourage everyone to proceed with the utmost caution as they make decisions that could have far-reaching impacts.
- What is the unmet need you saw at Cotiviti? And what approach are you taking to address it?
This project originated the same day the WHO declared COVID-19 a pandemic. We assembled a team to explore how Cotiviti could help respond to the outbreak by using our Caspian Insights data and analytics platform. The team began examining leading indicators such as telemedicine, rapid flu testing, and chest x-rays, that can help predict potential areas of concern before COVID-19 testing takes place.
The primary deficiency we are aiming to solve is the widespread lack of COVID-19 testing resources, which has left states unable to confirm the true impact and reach of the virus in their communities. Instead, our approach relies on other leading indicators of COVID-19, such as flu testing and diagnosis. By comparing current flu testing data seen in the CPT codes processed through our systems against confirmed flu diagnoses seen in ICD-10 codes, we can spot significant discrepancies that could indicate a “hidden outbreak” is occurring.
- What are the use cases? Are people using it for purposes beyond what you originally envisioned?
Our focus is on helping all healthcare stakeholders to prepare for what’s ahead given the unpredictable nature of this virus. As healthcare organizations seek to gain more data and use that data to extract meaningful insights, we are offering this resource to supplement their existing resources.
We have fielded questions recently regarding how this data may support contact tracing. We have also had inquiries from retailers wanting to use this data to inform decisions about when to open stores in various parts of the country. While neither of these were uses we initially envisioned, they reflect the need from all stakeholders to have access to reliable, timely COVID-19 data.
Now that states are looking at loosening their social distancing mandates and re-opening previously shuttered businesses, Cotiviti has unveiled a second map that shows which states have seen a downward trend of influenza-like illness and COVID-like syndromic cases to aid in decision making. It will be critical to maintain active surveillance of any early spikes that may be predictive of COVID-19 resurgence.
- How does it compare with other initiatives, like the Johns Hopkins model?
While Johns Hopkins has assembled an excellent, informative COVID-19 dashboard that aggregates data to track cases around the world and show trends over time, it specifically focuses on confirmed cases, which can only be identified through COVID-19 testing. Similar dashboards and tracking tools released by other organizations are also limited to tracking confirmed cases. Our approach looks at where there are a significant population of unconfirmed but likely cases to help forecast the hidden impact of this outbreak.
- What are the data sources? How did Cotiviti ensure data quality and accuracy?
Our data source is Cotiviti’s Caspian Insights data and analytics platform—the engine behind our healthcare analytics solutions—which processes millions of claims per day and comprises longitudinal data for more than 130 million Americans. It combines financial and clinical information alongside a multitude of other healthcare data types, such as social determinants of health, medical records, pharmacy, dental, and lab information to give health plans and providers actionable information at their fingertips.
We have both automated data quality standards and rigorous processes to ensure data quality and accuracy at all levels. For example, healthcare data is known to be inconsistent across disparate systems—the same individual might be listed by different names in the different data feeds we receive. To overcome this challenge, we leverage a unique combination of probabilistic and deterministic models to establish linkages between data sources, while also ensuring the data is de-identified. Finally, we have a strong organizational commitment to quality at all levels of Cotiviti.
- How accurate are the predictions? How is that changing?
We made our first forecast on March 12, and 80 percent of our predictions were realized by March 22. We have continued to maintain this level of accuracy while refining our data and algorithms, and we continue to see indications of potential hidden outbreak in certain states. However, as COVID-19 testing becomes more available, we know that hidden outbreaks will diminish and transition to confirmed outbreaks. Therefore, our team is preparing to shift to a more sophisticated modeling approach that identifies “COVID-like illness” based on the unique care pattern of COVID-19 that can be seen in the claim, clinical, prescription, and lab data for a patient. This approach will allow continued monitoring of the virus until a vaccine is available.