The current coronavirus crisis has taught us that a rapid response to a fast-moving pandemic is critical. These lessons can be applied to other public health threats in our society as well, such as the opioid epidemic, in which 128 people die every day from an overdose, including prescription and illicit opioids. In the USA, 30,000 people died from fentanyl overdoses in 2018 alone. To address this major challenge, new research out from The Alan Turing Institute, Warwick Business School and Oxford Internet Institute presents a novel method to predict drug use based on high-frequency sales data from drug markets on the “dark web”— the internet unreachable via normal browsers.

This innovative new model monitors drug use more accurately, allowing policymakers and public health officials to respond more quickly to health crises as they develop. The researchers scrape the drug markets on the dark web to create a dataset of global drug sales over time. They then model demand using Wikipedia page views for each drug, which are available daily rather than annually, to predict changes in demand. The paper, Predicting Drug Demand with Wikipedia Views: Evidence from Darknet Markets by Sam Miller, Abeer ElBahrawy, Martin Dittus, Joss Wright, and Mark Graham is published by The Web Conference. 

The researchers’ model is much more accurate and timelier than the typical method for collecting statistics on drug use, which is currently done via annual survey. Rapid changes in demand, production and distribution are hallmarks of the US opioid epidemic, and so current statistics aren’t frequent enough to capture these sudden shifts. In fact, with the US Fentanyl epidemic, the federal government only declared a national emergency in January 2017, which the researchers note was arguably too late. Additionally, certain drugs are so novel that they do not yet appear in these surveys, e.g. there were at least 36 Novel Psychoactive Substances discovered between January and August 2019 alone. Finally, people aren’t always truthful in survey responses about their substance use.

Lead author Sam Miller, Doctoral student at The Alan Turing Institute and Warwick Business School, adds:

“We saw an opportunity to generate much faster drug use statistics by combining two sources of real-time data: drug markets on the “dark web” and the volume of Wikipedia page views for each drug. If policymakers can look at search data—Wikipedia or Google trends—they may be able to get more advanced notice of what is going to happen in drug sales so they can better allocate resources.

We hope this model can be useful in anticipating looming crises, for instance in predicting new drugs coming on to the market. While this might be an unorthodox approach, we hope it equips us to be better able to anticipate and respond more quickly to the next drug epidemic, with the end goal being to save lives.”

Media requests:

Beth Wood
Press and Communications Manager
T  +44 (0)20 3862 3390
M +44 (0)75 3803 8168
[email protected]

Notes to Editors:

  • The paper ‘Predicting Drug Demand with Wikipedia Views: Evidence from Darknet Markets is published Monday 20 April 2020 in Proceedings of The Web Conference 2020 (WWW ’20), April 20–24, 2020, Taipei, Taiwan.
  • It is co-authored by Sam Miller (The Alan Turing Institute and Data Science Lab, Warwick Business School, University of Warwick), Abeer ElBahrawy (The Alan Turing Institute and City University, London), Martin Dittus (The Alan Turing Institute and Oxford Internet Institute, University of Oxford), Joss Wright (The Alan Turing Institute and Oxford Internet Institute, University of Oxford, and Mark Graham (The Alan Turing Institute and Oxford Internet Institute, University of Oxford).
  • The paper is available at the following URL: