How the Pharma Industry is Using Predictive Analytics for Brand Marketing Teams

September 27, 2018 | 4 minute read
Text Size 100%:

It is hard to imagine a growing company that doesn’t utilize data analytics in its decision making. Most organizations use data to identify opportunities, monitor commercial activities, and pivot if necessary with the help of big data and analytics. According to a recent Forbes article, top industries adopting big data analytics are financial, telecom and technology, with healthcare following suit. The most common use case for big data technology still remains in data warehouse optimization, but we are starting to see more organizations turning to predictive analytics to obtain foresight and make more informed decisions.

AI Can Help Pharma Gain Competitive Edge

Pharmaceutical companies are especially interested in fast adoption of big data and analytics. For one, it has been increasingly more challenging to bring breakthrough treatments to the market at a consistent rate. As more treatment options lose patent protection and become generic, it can be harder for second line products to compete for the attention of healthcare professionals. In oncology, this dynamic started to emerge a few years ago when market leaders in the first-line targeted therapies went off patent. Oncology is now a complex market with multiple lines of treatment options per condition, intensified price competition coming from generic products, and a lot of information being presented to physicians. Traditional analytical approaches no longer provide the level of sophistication marketers require to keep up with price pressures, retain share of voice, and effectively market their products—all while keeping marketing spend on budget.

Similar to the pharmaceutical industry, data science and IT have also undergone significant changes thanks to major advancements in cloud infrastructure. As a result, machine learning has transitioned from R&D into production and is used for real-world clinical and commercial applications. Companies of all sizes can now run thousands of statistical algorithms in parallel and can do so repeatedly, reliably, and fairly inexpensively. There are many problems artificial intelligence can solve for oncology, but I will describe one that is especially dear to my heart and has started gaining traction in pharmaceutical industry – a problem of predicting a choice.

Predicting Physician’s Choice

Physicians make decisions about their patient’s health on a daily basis. In oncology, these decisions may extend patient’s longevity and improve quality of life or could lead to complications and side effects. These choices should be made with relevant and complete information and should follow a rigorous protocol. Physicians are very busy—not only do they treat patients and handle large amounts of paperwork, they must also stay on top of new learnings in medicine. Combine this with rapid changes in the drug market and information overload and you are looking at a doctor who only consumes information she deems relevant and tunes out everything else. As an oncology marketer how do you know what’s relevant to her? How do you know what, where, and when to reach her to make the biggest impact on her choice? This is where industrialized AI capabilities can make a difference.

One approach is to explore action sequencing as a potential solution to this business problem. These models started to emerge in the beginning of the 20th century and are associated with Russian mathematician Andrey Markov and his first stochastic model Markov Chain. It allows us to describe a sequence of events where the probability of each event depends on the state attained in the previous event. This means that we can predict a likelihood of a doctor to prescribe based on her most recent interaction with the brand. Markov chains have a few limitations and require significant amounts of data, which is why they have not been widely adopted by the pharma industry in the past. Over the last five years, we have been able to overcome challenges of handling big data using distributed cloud computing as well as gaining access to new AI algorithms that have significantly improved since 1900s. I am especially fascinated by the work that has been done in the field of reinforcement learning. It started with Google’s DeepMind and its AI engine AlphaGo that outplayed the world’s champion in Go—a game significantly more complex than chess.

Reinforcement Learning and Digital Marketing

Reinforcement learning trains a machine to learn the best next action through exploration and exploitation. The beauty of this method is that it is model-free. You don’t need to teach the machine complex relationships between an action and a consequence. You only need to provide frequent feedback. Let’s take an example where we teach a robot to walk. The robot doesn’t need to understand the landscape, space dimensions, or what objects prevent it from moving. All it needs is feedback in the form of a reward (ability to complete a step) or a punishment (inability to move further). After a while it starts identifying patterns of walking that lead to the destination in the shortest period of time while avoiding impediments.

The same concept could be applied to digital marketing. Traditional predictive analytical models that establish relationships between writing a prescription and marketing engagement could be replaced with reinforcement learning. Digital marketing and prescription activity data allow data scientists to build an algorithm that learns what patterns of digital engagement lead to a script for any given physician. The best part is that as we feed more data into an algorithm it starts getting smarter and learns specific marketing sequences that result in a prescription while avoiding unnecessary informational overload.

Reinforcement learning is a cutting-edge technology that still needs a lot of experimentation. Data science teams are actively working on understanding how it can be efficiently utilized to help brand marketers create truly personalized digital experience for physicians. But just like with big data and model-based machine learning, it is a matter of time when reinforcement learning will transition from the lab and gaming industry into robotics, finance, telecom and ultimately to healthcare.

Tatiana Sorokina

Tatiana Sorokina is an Advanced Analytics lead for Novartis US Oncology and supports brand teams with cutting-edge advanced analytics and machine learning products and services. Tatiana is an accomplished data science leader with deep expertise in AI, data and digital. She has a successful track record building global data science teams and analytical products using structured and unstructured data sources. Tatiana holds B.Sc. in Economics from Moscow State University and M.Sc. in Marketing Science from Columbia Business School. Prior to Novartis, Tatiana co-founded Parceed, an AI platform for medical education, and has led global Data Science team for Prognos, a healthcare AI company.

Show more

Previous Post

Challenges of Generalization in Machine Learning

Blaine Bateman | 13 min read

Next Post


Top 4 Reasons Data Scientists are Motivated to Change Jobs

Mary Ann Hunt | 4 min read