John Giannouris, Vice President, Specialty Data Strategy, and Consulting Services
In previous blogs, I touched on the wealth of data sources available to support specialty therapeutics, as well as the data disciplines essential to making the most of that information. In this post, I’d like to hone in on something a bit more technical: the potential role of machine learning in maximizing specialty therapies’ commercial potential.
First, a brief, non-technical explanation of “machine learning”: At the core, machine learning is the process of a computer poring over deep data sets to learn the “lessons” hidden within from the trends and patterns it recognizes. These are insights that take human eyes and minds significant time to identify across information sets of this size and depth.
Once the computer (machine) has learned from a data set, it can turn those learnings into models that are able to run against other data sets. In other words, a machine may be able to identify the constellation of factors that result in a patient being slow in initiating a specialty therapy. It may also be able to identify which patients are at risk of not initiating therapy at all – or predict which ones are mere days away from halting therapy.
Consider a patient starting a specialty therapy for Crohn’s disease. It could take three full weeks for that patient to be approved – from gathering all the required information and completing multiple forms, to addressing questions between the provider, the specialty pharmacy, and the payer. By using machine learning, within five days of the prescription being ordered, it may be possible to know that the patient has a mid-90% probability of not starting the therapy. More important, it may also be possible to identify a way to prevent that from happening.
At ValueCentric, we call these “intelligent interventions” – specific, data-informed recommendations for how you can increase the likelihood of new patient starts, decrease time to fill, and support ongoing adherence.
Machine learning models may help you to understand patients at-risk, recommend interventions to improve outcomes, and even suggest the right communication channel as well as the optimal time of day to intervene. For example, the model may predict that a certain patient is more likely to miss treatment, kicking-off a cascade of computational activities to suggest the next best action. It’s quite possible that the patient fits a risk cohort that requires same-day Hub outreach to minimize drop-off and a case-level performance metric is generated to be met by the Hub. Perhaps the patient would benefit from being reminded about the balance available on the manufacturer’s copay offer right around the time that their refill would be due. Or possibly, that the patient should be sent educational content on their mobile device to maintain engagement.
Once patients are on specialty therapy, machine learning models can illuminate adherence challenges and suggest interventions intelligently to maximize treatment continuity. With robust data, the continued analysis can also identify a patient that may be at risk of stopping therapy. Resources can be allocated and targeted towards those patients who are more likely to fail. Perhaps with the refill, the model has identified that the prescriber, rather than the pharmacy, is the optimal provider to intervene. Alternately, it could be possible that the model recognizes that for a very specific patient, the refill reminder should be ideally timed at 9AM on a Saturday, at three days before medicine is exhausted, to increase likelihood of refill fulfillment.
To get there, you need data cohorts and proven models for “learning” from them. It doesn’t happen overnight , but it can’t happen at all unless you’re collecting the data attributes needed to start building the cohorts and training the models. That all starts with understanding the data sources and mastering the data disciplines – and finding the right partner to guide you on the journey.