Protecting Patient Privacy in Advertising with Machine Learning

Machine learning is highly effective at finding patterns in massive data sets and using learnings to optimize and improve the outcomes of systems and processes. 

Applied to healthcare, and specifically to patient data management, where the use of electronic medical records, genetic records, administrative data, and other types of data has skyrocketed in the last decade, machine learning has the power to transform the way we look at everything from diagnosis to treatment.

In fact, healthcare data accounts for 30% of the world’s total data volume. In that same vein, machine learning provides healthcare marketers with a new way to think about using data to drive personalized communications with audiences and delivering the right message to the right person at the right time — in a way that’s optimized to support better patient outcomes.

Top AI in Healthcare Update: KenSci Attains Highest Score Among Healthcare AI Vendors in New KLAS Report

However, in an era where privacy is top of mind, consumers are demanding the right to privacy and protection of their data. It’s not hard to imagine how, in the wrong hands, unprotected healthcare data could be abused by fraudsters and cybercriminals. Consumer privacy, especially when dealing with sensitive health data, must be protected at all costs. 

It’s incumbent upon organizations like mine that leverage health data to develop technology that allows AI or ML to extract critical insights and learnings from very sensitive datasets without the risk of protected data being used to identify individual patients. The key is to use privacy-preserving technologies, like differential privacy, which is an approach that mathematically guarantees the privacy of individual patients while allowing valuable, population-level insights to be understood and used in applications. 

When these systems are purposely built so that no one, including clients or data scientists or engineers, have access to identifiable patient data, patient privacy is maximized. In practice, this means that health data sets need to be manipulated just enough to obscure data in a way that fully minimizes the risk of a person analyzing datasets to trace such data back to an individual patient. This can be done by meeting HIPAA’s stringent de-identification standards, which ensures that there is a “very small” risk of a patient being identified through an evaluated data set. To meet this standard, these data are physically placed in tightly guarded, auditable data environments that have been certified by third-party statisticians. In these environments, analysts and data scientists are free to use advanced analytical tools and build models that extract patterns about specific conditions from real-world clinical data in a way that unlocks immense value and drives innovation for our industry. 

Top AiThority News: Algorand-Based Xfinite (XET) To Launch on MEXC Global

To use an example, let’s say that we need to reach consumers that may benefit from messaging for an upcoming ad campaign seeking to drive awareness of a newly launched therapy. Leveraging the combination of socioeconomic and health data sets, the system might generate a lookalike target audience of “men over 45-years-old, who live in the Northeast, and watch late-night television.” These models are then rigorously verified to be HIPAA-compliant before they can be used to serve ads to relevant consumers most likely interested in learning about the new therapy. 

Top Customer Intelligence Blog: Salesforce Brings Out New Sales Cloud Innovations Featuring AI-based Revenue Intelligence

Locking sensitive data behind a privacy firewall, or data clean room ensures organizations can glean critical insights for their business while preserving patient privacy. 

So how can patients or end-users know that patient privacy remains protected when considering machine learning? 

Statements emphasizing privacy are essential, but actions are even better. First, look to see that organizations are transparent about how they use clinical data. Next, check to see if their solutions have been verified for HIPAA compliance by outside organizations. Companies should be willing to provide evidence of this in the form of a letter or certificate. Ad tech companies leveraging machine learning that care about privacy will go to great lengths to test and demonstrate their compliance. 

Done correctly, healthcare and pharma advertisers that make use of privacy-safe machine learning can realize audience quality appreciation of more than 25% compared to traditional methods. At a time when just 35% of patients say they consider pharmaceutical advertising relevant to them, better audience targeting that combines the power of machine learning with differential privacy can dramatically increase effectiveness and lower costs for advertisers. Moreover, better personalization and optimization can fundamentally change the way consumers view ads as a form of direct engagement that makes them better informed about their health conditions and, ultimately, leads to better outcomes. 

[To share your insights with us, please write to sghosh@martechseries.com]