For researchers, clinicians and healthcare practitioners, the news on COVID-19 has provided new lessons in public health. Researchers have pooled their collective knowledge to address the crisis, but the novelty of this virus has meant that predicting into the future often feels like a shot in the dark.
It has become clear that machine learning is one of the strongest tools for predicting how a virus will spread. These predictions have already been crucial in decision making on lockdowns and social distancing measures – lift restrictions too soon and we could face an intense second wave, too late and the public could become restless or distrustful.
When wide-scale behavioural changes are needed to slow the spread of COVID-19, it’s helpful for the public to see the story for themselves. After all, COVID-19 is a new story for all of us, one we’ve never read before – statistical modeling and machine learning could help us write a better ending.
In response, CHÉOS Scientist Dr. Ehsan Karim has leveraged his expertise to develop tools for visualizing country-specific COVID-19 data. Dr. Karim is an expert on big data, and believes that machine learning can improve a range of issues, from public health outcomes to patient-specific care plans. Dr. Karim is amongst the vanguard of researchers implementing machine learning in healthcare. Let’s dive into some of his research to better understand the role big data can play in health, as well as the role it might play in the current COVID-19 pandemic.
Big data vs. Traditional clinical trials
Clinical trials remain the gold standard in medical research, however they are limited in duration and scope; clinical trials often take place in a controlled environment that does not take into consideration changes over time or the impact of additional health variables. Moreover, in fast-moving pandemics like COVID-19, clinical trials aren’t able to keep up with the spread of the virus.
With the help of big data, machine learning has the potential to reach beyond these limitations, even helping to diagnose and treat complex or overlapping chronic illnesses while answering patients’ uncertainties in the long term.
Put simply, machine learning is computer models learning and improving based on access to large data sets. By inputting a large amount of available data, machines are now able to generate unique insights and make predictions about the future. Machine learning and artificial intelligence have given us such innovations as self-driving cars, email filters, facial recognition and a more fulsome understanding of the human genome.
Dr. Karim incorporates machine learning into his causal inference research, and believes that it has the potential to overcome many of the limitations of traditional clinical trials: “With the use of machine learning techniques, we can push boundaries and go beyond the standard statistical tools to make better and more meaningful treatment decisions,” he says.
Implementing machine learning in health care can be difficult without a proper understanding of the needs of each patient. Fortunately, large amounts of health care information are now collected by federal and provincial authorities, and more and more researchers take causality into consideration when analyzing this data. In today’s health care landscape, when it comes to assessing aggregate patient outcomes for chronic diseases, machine learning may be the breakthrough that we have been waiting for.
Multiple Sclerosis: Ready for intervention
Dr. Karim has spent much of his career applying these insights to Multiple Sclerosis, or MS. MS is a chronic condition where the immune system mistakenly attacks parts of the body that are vital to everyday function. MS damages the protective coverings of nerve cells, and leads to diminished function in the brain and spinal cord. Canada has the highest rate of MS in the world, and although there are several drugs prescribed to address symptoms of the disease, there is no known cure. A disease as slow and complex as MS is a prime target for a machine learning intervention.
Working with a group of experienced collaborators, Dr. Karim uses observational data going back to the early 1990’s to assess the longitudinal effects of MS therapies in BC. While most of these therapies originally underwent clinical trial testing in controlled settings, the data that Dr. Karim is working with come from thousands of real-world cases. This way Dr. Karim is able to assess these therapies from a different point of view to determine whether or not they worked in practice.
Designing and analyzing observational data is a difficult task. Dr. Karim first maps out the relationship of factors associated with the disease and treatment, such as age, sex and comorbidities, and then utilizes powerful computational techniques to obtain clinically relevant interpretations.
Dr. Karim has looked particularly at beta-interferon treatment, a treatment known to reduce the number of MS flare-ups. It remains unknown, however, what the long-term impacts of this treatment are, what the effects may be for patients above the age of 50, or whether the treatment actually reduces mortality rates and/or the overall progression of the disease. His model takes into consideration a number of unique factors to determine the best possible course of action for each patient. The goal of machine learning, combined with expert knowledge, is to determine what effect beta-interferon treatment will have on an individual MS patient in the long term. It will give doctors and caregivers a robust understanding of the long-term implications of this intervention.
Machine learning into the future
There remain some significant barriers to the full implementation of machine learning, especially gaps in publicly available health care data. These data sets are not gathered for the purpose of research and, therefore, must be repurposed to answer targeted research questions. But combined with expert knowledge, these emerging tools have the potential to revolutionize our understanding of treatment pathways, enabling us to obtain real-world evidence and make data-driven decisions.
COVID-19 is a particularly prescient case study in the application of big data in healthcare. For the first time, tools such as machine learning and data visualization have been implemented in pandemic response. Dr. Karim’s experience working with MS and other chronic conditions has set a strong path for analyzing data before any clinical trial evidence becomes available. Assessing and visualizing these data may be the best way to understand the spread of COVID-19 in the short term, while reducing the spread and morbidity of other novel diseases into the future. All the while Dr. Karim is proposing careful analysis plans with other CHÉOS collaborators to learn more about COVID-19 based on BC-specific observational databases, while honing novel machine learning tools.
