Data science and social good: Fighting opioid addiction


This is the second part of a two-part series that explores connections between advanced analytics and corporate responsibility.

Recently, two student researchers from the global DXC Internship Program harnessed an innovative analytics solution to unearth insights into the U.S. opioid crisis. The goal was to begin creating a framework for better understanding addiction risks.

Comments Brendan Stec, co-leader of the project along with Aiswarya Rangarajan, “One hundred and fifteen Americans die each day from overdosing on opioids, and many of those became addicted after they were initially prescribed an opioid painkiller following a serious surgery or injury. By using tools from data science and artificial intelligence, we hoped the data could tell us which types of individuals are either likely to develop an addiction or are already addicted (in real time), so the proper clinical decisions or interventions can be made going forward.”

According to Brendan and Aiswarya, knowing how to clean and prepare the data was actually the most important part of the entire project. The data our researchers started with was in a raw, unstructured format, and any machine learning algorithms wouldn’t work until the data was sufficiently prepared. They spent several weeks gathering all of the data they needed and figuring out what they could try to find out. Then, once the data was ready, it was time to search for interesting patterns.

Our intern team used an unsupervised machine learning algorithm to group patients into different clusters. Clustering allowed them to group patients without specifying what features they thought would be important beforehand. This strategy allowed the system the flexibility to discover completely new and unanticipated patterns of opioid addiction risk. Combining the data preparation with unsupervised machine learning and interpretation of the resulting patterns resulted in an artificial intelligence with potential to provide new insights into addiction risks.

What did our researchers discover? When treated with opioids, patients with second or third degree burns were almost twice as likely as non-burn victims to experience some kind of drug overdose. In addition, just being young, old, wealthy or poor did not make a person more or less likely to become addicted to opioids. But young patients with a certain type of medical history were at a high risk of addiction. A group of patients (about 150) between the ages of 15 and 24 with a history of concussions, anxiety and attention deficit disorder (ADD) were at a high risk of addiction if problems like broken wrists, clavicles and legs were treated with an opioid. Our interns thought it would be interesting for other researchers to further explore the connection between having a history of these conditions and the likelihood of future opioid addiction.

Ultimately, the project suggested that predicting the risk opioid addiction isn’t as simple as tracking trends in the amount of opioids taken over time. Just because someone has taken more opioids doesn’t make them a greater risk for overdose. Our researchers would see, for example, a patient prescribed opioids on four or five different occasions never show symptoms of addiction, while another patient developed symptoms after a single prescription. The findings indicated that opioid addiction occurs as the result of a complex interaction of factors, and that simple indicators of risk are difficult to find. A productive research strategy for the future would be to map these complex factors and design social and medical support systems to better serve at-risk patients.

For more on the ways in which data science can help illuminate today’s complex social and public health challenges, please read part one of this series, “Stronger communities, better healthcare.”

Speak Your Mind


This site uses Akismet to reduce spam. Learn how your comment data is processed.