Column Big data column大數據專欄

2020/07/09 Forecasting COVID-19 with Predictive Analytics, Big Data Tools

 - In the midst of a situation as uncertain as the COVID-19 pandemic, the healthcare industry has sought to use big data and predictive analytics tools to better understand the virus and its spread.

For more coronavirus updates, visit our resource page, updated twice daily by Xtelligent Healthcare Media.

Hospitals and health systems have leveraged predictive models to gain insight into COVID-19 risk, disease outcomes, and the virus’ potential impact on resources.

However, even with the most comprehensive data and predictive algorithms, the novelty of the pandemic means that the industry is still largely left in the dark.

“This is new to all of us. COVID-19 is a disease we've never seen – it's only been on the planet that we know of for maybe six to eight months,” said Joseph Colorafi, MD, System VP of Clinical Data Science for CommonSpirit Health.

 

“While there are some therapeutic regimens that show promise, there's no vaccine yet. Even a city like New York that was quite consumed by COVID-19, estimates show that only 20 to 25 percent of that population got infected, which means there's still 75 percent of that population that is vulnerable to the disease. And everywhere else in the United States is probably lower than that.”

Herd immunity can be achieved in two ways, Colorafi noted: Either the virus infects a higher percentage of individuals, which could result in millions of deaths; or leaders work to slow down the progression of the disease.

“We need to slow down the contact rate so that we spread out the disease over a longer period of time. And we're working under that umbrella and doing what we feel is safest for our patients and for our clinicians to flatten that curve,” Colorafi said.

As communities across the country continue to relax social distancing orders, it will be crucial for health systems to plan for changes in healthcare demand and surges of COVID-19 patients. With this in mind, the team at CommonSpirit Health built predictive models using de-identified cell phone data, public health information, and data from the system’s own care sites.

“We started looking at the first peaks of the virus, and there were enough cases for us to do some modeling and prediction. Now, just three months later, we're looking at the same virus as people are getting back together and the contact rate is going up. And we're trying to manage this resurgence in our markets,” Colorafi said.

Using these models, the organization can gain insight into the surges and dips of COVID-19 infection rates, allowing for better planning while hospitals work toward resuming their full spectrum of necessary services. The models factor in fixed data, such as population and availability of healthcare providers, as well as variable data like social distancing relaxation and normalized county new cases.

 

CommonSpirit is able to produce a predictive outlook for about 75 percent of its markets across the country, including communities in Texas, California, Arizona, and the Pacific Northwest. In many cases, the data reveals that a surge of COVID-19 cases can be expected about two to three weeks after social distancing requirements are relaxed.

The project has also allowed the CommonSpirit team to make associations between various data elements, Colorafi noted.

“What's interesting about this work is that we have the opportunity to be innovative with information and make connections between data points that we may not have otherwise considered,” he said.

“For example, we’re using de-identified cell phone data as a surrogate for mobility. We know if people are moving around more, they're social distancing less, and we can statistically correlate that mobility to the number of admits in a particular county. That's been very helpful in managing some of the hospital demand and the larger number of admits that are coming in these days.”

The team designed the tools to offer a look at what the healthcare landscape will look like in the near and not-so-near future, Colorafi explained.

“One model is for long-term forecasting. We're trying to forecast three to six months out, if not longer than that. The model looks at your at-risk population, how many get infected in that population, and the percentage of those individuals that would get admitted. And then once they're admitted, how long does it take those patients to move through the hospital if they survive?” he said.

“There are a certain number of days involved with all of those calculations and that movement of patients. So that's the model that we lapped on for a long-term forecast. We have shorter term forecasts that are more useful now as far as what's going to happen in the next month. That's what we're focusing on right now, is the shorter term.”

Although predictive tools like these are still far from infallible, the insight they offer into disease patterns and patient surges is vital in a situation like the COVID-19 pandemic.

“As far as epidemics are concerned, the world has cut its teeth on the ability for models to be helpful. They're never perfect and they're not going to predict everything, but the important thing is to understand what information we can use to help our patients and staff get through this,” Colorafi concluded.

Data Source: https://healthitanalytics.com/news/forecasting-covid-19-with-predictive-analytics-big-data-tools
Image Source :Getty Images