Big data in the ER
A grad student’s research project aims to use data and machine learning techniques to predict and prevent emergency room visits.
Biostatistician Colin Weaver takes the same kind of algorithms and machine learning techniques that land a certain blender on your Amazon page or a suggested movie in your Netflix account, and he applies them to something slightly more consequential: health care.
A master’s student in community health, Weaver’s work combines his expertise with statistics and a long-standing interest in improving health care. For his current research, he will use data and machine learning techniques to predict which patients are most likely to end up in the emergency department based on patterns in, for example, doctor visits, diagnoses, or prescription combinations.
“There are emergency visits that are unavoidable, but the visits we are interested in predicting are the ones that could perhaps be pre-empted or headed off, generally thought to be visits resulting from poorly managed chronic conditions,” Weaver says.
“If we can find which health care events increase someone’s risk of ending up in emergency, we will be able to identify patients early and hopefully provide them with some additional care to prevent their emergency visit.”
Weaver, who also earned his BSc in statistics from the University of Calgary, was one of nine UCalgary students to earn an Alberta SPOR Graduate Studentship, co-funded by Alberta Innovates and the Canadian Institutes of Health Research (CIHR) for his work. The awards were announced earlier this month.
This month it was also announced he has been awarded the CIHR Canada Graduate Scholarship – Master’s in support of his research project.
“Colin’s work demonstrates the kind of leadership and innovation in health research that brings about real change in the lives of Canadians,” says Lisa Young, dean and vice-provost (graduate studies). “We are so proud of his remarkable achievements in winning both the SPOR studentship and the CIHR Canada Graduate Scholarship; these funding programs can be the difference-maker in the success of our graduate students — they catalyze important research breakthroughs and fuel innovation.”
Digitized patient and hospital data a fresh resource
Weaver’s supervisor is Tyler Williamson, PhD, an assistant professor of biostatistics in the Cumming School of Medicine’s Department of Community Health Sciences. He is one of Canada’s top experts on primary care data. “What’s so exciting in Colin’s work is we’ve just never had data of this depth and quality before,” Williamson says.
“It’s only been in the last few years that primary care physicians gather data electronically, for one thing,” he says. “A pioneering organization in this endeavour is the Canadian Primary Care Sentinel Surveillance Network (CPCSSN). They are really the reason we have access to this rich data source. And after a few years grappling with who owns the data, privacy and ethics concerns, we will very soon have that data linked to Alberta hospital data. The two together are a very rich source of information.”
Potential for improved outcomes, cost savings
Weaver will obtain the data shortly to conduct his research. He will have access to electronic medical records from Alberta family doctors who are members of CPCSSN. The anonymized data include information on patient demographics, diagnoses, lab results, prescriptions and number of doctor visits. He’ll then link these data to emergency department and hospital visit data held by Alberta Health Services.
To turn those data into a useful predictive model, Weaver will apply the tools of his trade, specifically more traditional regression techniques and then machine learning methods, similar to those used by Amazon, Netflix, and others. Like them, he will be looking to predict preferences, or outcomes, based on what’s been seen before; the more data fed into the algorithms, the better they will be able to predict emergency visits.
“There is tremendous interest in a predictive model like this,” Williamson says. “It has the potential to both improve patient care, and offload the burden placed on the system by avoidable visits.”
Williamson adds this is all so new, there are no guarantees: “It might not be possible to predict hospitalization based on the data, but we sure are going to try,” he says. “Colin is an exceptionally well-prepared student — one of the best I’ve encountered. If someone is going to make this discovery, it’s going to be him.”