Predictive Analytics Model Identifies Illicit Online Pharmacies

A predictive analytics model could spot illicit online pharmacies, enhancing drug safety and patient health, a study published in JMIR revealed.

Online pharmacies have grown in popularity in recent years, mainly due to their convenience, lower prices, and access to drugs that may be otherwise unavailable. However, researchers noted that consumers have limited awareness of illicit online pharmacies, which are estimated to represent 67 to 75 percent of web-based drug merchants.

Illicit online pharmacies can pose a serious threat to patient health, with many of them providing substandard medications to customers without their knowledge, among other potential issues.

"There are several problems with illicit online pharmacies,” said Soundar Kumara, the Allen E. Pearce and Allen M. Pearce Professor of Industrial Engineering. “One is they might put bad content into a pill, and the other problem is they might reduce the content of a medicine, so, for example, instead of taking 200 milligrams of a medication, the customers are only taking 100 milligrams — and they probably never realize it." 

Researchers set out to develop a predictive analytics model that could identify and monitor illicit online pharmacies. The team conducted a traffic analysis based on web-collected data, which evaluates the means through which customers access online pharmacies and their level of engagement with these sites.

The group then developed a tool to predict the status of online pharmacies based on the referral websites to them. Researchers designed the predictive analytics model to approach the problem of distinguishing good online pharmacies from bad similar to the way that people make comparisons.

“The essential question in this study is, how do you know what is good or bad — you create a baseline of what is good and then you compare that baseline with anything else you encounter, which normally tells you whether something is not good," said Kumara.

"This is how we recognize things that might be out of the norm. The same thing applies here. You look at a good online pharmacy and find out what the features are of that site and then you collect the features of other online pharmacies and do a comparison."

Researchers examined several attributes of online pharmacies, but ultimately identified the relationships between the pharmacies and other sites as a critical attribute in determining whether the business was legitimate or not.

Additionally, if a pharmacy is mainly reached from referral websites that mostly link to or refer illicit pharmacies, then this pharmacy is more likely to be illicit.

"One novelty of the algorithm is that we focused mostly on websites that link to these particular pharmacies,” said Sowmyasri Muthupandi, a former research assistant in industrial engineering and currently a data engineer at Facebook. “And among all the attributes we found that it’s these referral websites that paint a clearer picture when it comes to classifying online pharmacies."

The most significant challenge in developing the predictive tool was the immense amount of variable information related to online pharmacies, researchers noted.

"It's very challenging to develop these tools for two reasons," said Hui Zhao, associate professor of supply chain and information systems and the Charles and Lilian Binder Faculty Fellow in the Smeal College of Business.

"First is just the huge scale of the problem. There are at least 32,000 to 35,000 online pharmacies. Second, the nature of online channels because these online pharmacies are so dynamic. They come and go quickly — around 20 a day."

The team pointed out that researchers could develop a warning system that alerts the consumer before a purchase that the site may be an illicit online pharmacy. Moreover, search engines, social media networks, online markets, and payment companies could use the algorithm to filter out illegitimate pharmacies online, or take the status of online pharmacies into consideration when ranking search results.

Policymakers, government agencies, and drug manufacturers could also use the model to identify and monitor illicit online pharmacies.

The many applications for this predictive model could hold significant implications for drug safety and patient health, the team stated.

“Given that this is a critical area of concern to patients’ health and the integrity of the drug supply chain, we hope this study will inspire additional efficient and effective prediction models or additional applications for the prediction models developed. On a larger scale, we hope to inspire more research in other aspects to fight illicit online pharmacies,” researchers concluded.

“Our literature review also reveals that literature on automatic prediction/identification of websites selling counterfeit products (not limited to drugs) is also very scarce, although selling counterfeit products on the web is a prevalent problem. Our framework and prediction models can be applied to other products, and we hope to inspire research in this general area as well.”