- Social media has always been about capitalizing on what’s trending, and for healthcare big data scientists, those top-ten tweets and hashtags may now help to save lives. Researchers at the University of Arizona are using predictive analytics algorithms to scan Twitter for mentions of asthma-related events that may drive patients to their local emergency departments in an effort to forecast utilization and help providers improve chronic disease management for these patients.
In a journal article set for publication in the IEEE Journal of Biomedical and Health Informatics, Dr. Sudha Ram, a professor of computer science, business intelligence, and big data analytics at the University of Arizona and her colleagues developed a predictive analytics model that can suggest patterns of asthma-related ED use with 75 percent accuracy.
"We realized that asthma is one of the biggest traffic generators in the emergency department," Ram said to UANews. "Often what happens is that there are not the right people in the ED to treat these patients, or not the right equipment, and that causes a lot of unforeseen problems."
Asthma affects eight percent of all US adults and nearly ten percent of children, according to recent CDC data, resulting in 14.2 million primary care visits and 1.8 million emergency department visits per year. Just a third of Medicaid-eligable pediatric patients regularly achieve medication-based control of their condition, and those who do not appropriately manage their chronic disease have a 21 percent higher risk of expericing an emergency department visit, a 2013 study found.
As healthcare providers increasingly dependent on value-based reimbursement try to reduce unanticipated crisis events that drive up costs and impact quality reporting measures, understanding why and when patients end up in the ED is an important part of reducing unnecessary spending and preventing avoidable hospitalizations.
To that end, Ram and her team looked beyond EHR and claims data to the world of Twitter, “a popular data source for disease surveillance using social media since it can provide nearly instant access to real-time social opinions,” the study says. Tweets can be easily categorized through the use of hashtags, and often contain geographical tagging and time stamp data that can be used to develop detailed models of patient behaviors. Through the use of authorized APIs, researchers can access some of this information for their work.
The researchers collected Twitter data from October through December 2013, identifying more than 1.3 million asthma-related messages, or more than 15,000 worldwide tweets each day mentioning key terms related to the disease.
“One of the challenges we needed to address was to extract signal from the noisy Twitter dataset i.e., to distinguish tweets that are relevant to asthma from tweets that mentioned asthma in an irrelevant context,” the study says. Using machine learning to extract a relevant dataset narrowed to English-language messages containing specific keywords, the team was eventually able to narrow down tweets by geographic areas of interest. They then correlated the Twitter data to known emergency department use information, environmental trigger data, and Google searches to produce a predictive analytics model that could identify increased social media activity with higher levels of ED visits.
“Although preliminary, the findings of this study are very promising for many reasons,” the study says. As healthcare big data analytics becomes more sophisticated in its use of non-traditional data sources in addition to clinical EHR data, historical claims data, or patient-generated health data, providers may have increased access to predictive models that help to allocate resources and better treat the individualized needs of patients.
“Interventions would be prioritized in time and place to reduce the risk for asthma ED visits,” Ram and her team predict. “For instance, public health resources could be used to reach out to patients from high-risk clusters or communities at any given time, and direct them towards less costly and more efficient care sites such as their primary care provider offices. Moreover, predicted risks could be spatially and temporally visualized, and made available to community stakeholders through various media sources.”
Through the use of social media and other real-time data that provides fodder for predictive analytics, population health management and chronic disease management programs could be more targeted towards patients at high risk of experiencing an adverse event. The researchers hope to validate their findings further before expanding their work into other chronic disease areas, such as COPD and diabetes, to help providers implement more comprehensive preventative and proactive treatment plans.