Demystifying predictive analytics
By: Nadea Leavitt | May 22, 2017
Analytics hold great promise for the pharma industry – but you have to understand how they work to reap the benefits.
As data becomes ever more complex and voluminous, the potential to gain deeper insight from data is growing. Realising this potential requires application of modern predictive analytics methods. As demand for innovative predictive analytics grows, so too does the need for greater clarity on how and when to use these methods.
Predictive analytics is one of those technologies that promises to revolutionize the way companies make business decisions. Industries around the world — including pharma — are eager to harness its potential. Despite being a relatively new trend, the global predictive analytics market was valued at roughly US$3.5 billion in 2016, and that value is expected to more than triple to US$11 billion by 2022.
But for all its promise, predictive analytics is still shrouded in mystery and hence its potential is still largely unrealized.
At this year’s International Society for Pharmacoeconomics and Outcomes Research (ISPOR) meeting in Boston (May 20-24), my colleagues and I will provide more clarity on this topic in a workshop on Monday afternoon entitled: Demystifying predictive analytics and an introduction to recent methodological innovations. The goal of the workshop is to empower healthcare researchers with a greater understanding of predictive analytics, and to explore case studies and innovative methodologies that are impacting clinical and commercial results in our industry today.
Predictive analytics isn’t a single piece of software or process. It is an outcome that draws on inspiration from the related fields of machine learning, data mining, statistics and artificial intelligence (AI) to analyze historic patterns in data to make predictions about the future or otherwise unknown events. It can be useful in almost any business setting, though its value over more traditional analytical methods comes to the fore where data is highly complex and large-scale, and where maximising predictive accuracy is of upmost importance.
Pharma is already leveraging predictive analytics at different points throughout the product life cycle, from accelerating drug discovery to optimizing salesforce outreach. The volume and diversity of healthcare studies and applications involving predictive analytics is rapidly growing, from risk stratification tools for major chronic conditions, to predictive algorithms to screen for patients with undiagnosed rare conditions.
In one example, the QuintilesIMS analytics team worked on a project in the UK to develop an algorithm which could be used to speed-up time to diagnosis for patients with tuberous sclerosis complex (TSC). TSC is a rare disorder that causes benign tumors to form throughout the body causing seizures, kidney failure, epilepsy and other, often fatal, conditions. These patients often go undiagnosed for a long time, leading to poor outcomes for patients and often avoidable costs for health systems arising from misdiagnoses and unnecessary treatments.
We used a database of over five million patients with linked information from primary care and hospital records. We started with building rules to find TSC patients based on a synthesis of interviews with clinical experts. This identified a high-risk group containing twenty-five non-TSC patients for every correctly identified TSC patient – not too bad given the rarity of the disease. We then developed a model based on conventional epidemiological statistical methods and managed to improve the TSC detection rate to one in thirteen. Finally, we developed a machine learning algorithm which produced a very impressive detection rate of one in five.
This established a valuable proof point for application of machine learning to pin-point hard-to-find patients, an approach we have since applied to great effect across many disease areas in both the US and Europe. Disease detection is just one example of where these highly innovative analytical methods can have a demonstrable impact resulting in better, faster and more informed decisions. As we continue to gather more biologics data, the impact on precision medicine will also be profound.
Predictive Analytics can be equally valuable in commercial applications, to determine which markets have the most potential, how to best engage with physicians and patients, and what data payers will need to support pricing expectations.
We are at a tipping point with predictive analytics in pharma, where there is a large enough volume of accessible data to make meaningful predictions, and powerful enough technology to mine and analyze that data in a matter of minutes rather than days.
However, these are not plug-and-play solutions, at least not yet. They still require a team of experts who understand the technology, and can craft targeted algorithms that ask the right questions and generate the most meaningful results. As these tools evolve, so too will our ability to answer more questions with greater accuracy, so we can speed-up drug discovery, optimise trial design and execution, improve disease management and optimise treatment allocation.
As demand for predictive analytics applications increases, and as data becomes ever more voluminous and complex, the need for greater clarity and understanding of rapidly evolving modern predictive analytics methods is paramount. It is important that all key stakeholders across the life science industry understand the benefits these tools can bring, so they can understand where and when they are most applicable, and assemble the right team of experts to deliver the most value driven results.