Event Sequence Modeling: Applying the Math from Digital Attribution to Map the Patient Journey

Healthcare stakeholders can improve patients’ journeys by applying the event sequence modeling techniques that are commonly used for conversion attribution in the digital marketing space.

As healthcare has become progressively more patient- and outcomes-centric, discussions of “the patient journey” have become ubiquitous in every corner of the industry. From payers to health systems to individual healthcare providers (HCPs), nearly every industry stakeholder has an interest in developing a robust understanding of how patients traverse the broader healthcare landscape.

But while “the patient journey” is a convenient rhetorical device, its monolithism obfuscates a truth that is incredibly important to acknowledge: “the” patient journey is a myth. The inherent complexity of the healthcare industry — and of health, itself — opens up countless paths from Point A to Point Z. Various permutations of factors like genetics, disease etiologies, comorbidities, referral patterns, time between appointments, and more result in thousands, if not millions, of distinct patient journeys.

That said, the heterogeneity of the patient journey need not preclude healthcare stakeholders from striving to improve each and every individual’s pursuit of health and wellness. The proliferation of data — and, critically, tools and techniques powerful enough to make use of it — in the wake of the digital revolution has rendered many previously prohibitively complex problems eminently solvable.

In the case at hand, by mirroring the digital marketing industry’s use of event sequence modeling algorithms, stakeholders in the healthcare industry can derive actionable insights from analyses of thousands upon thousands of unique patient journeys. These insights can then be leveraged to clear the way forward for both current and future patients, regardless of the twists and turns their individual paths may follow.

Striving for a Better Approach to Attribution

Conversion attribution is among the foremost challenges of digital marketing. Many consumer journeys are shaped by multiple digital touchpoints, and determining how much credit each touchpoint should receive for a conversion is a complicated — and often even controversial — endeavor.

If a consumer converts only after viewing a display ad, clicking through a social ad, and receiving an email ad, which touchpoint should receive credit for the conversion? The display ad, without which the consumer may not have gone on to engage with the product or brand being marketed? Or the email ad, which ultimately closed the deal? First-click attribution models dictate the former; last-click attribution models dictate the latter.

Of course, the truth of the matter is that nearly every touchpoint leading up to a conversion should receive some credit. By implementing a scoring methodology grounded in mature data analytics, marketers can weight the contributions of each touchpoint along successful consumer journeys. These analytics allow marketers to not only identify patterns in sequences of touchpoints, but aggregate similar sequences into journey “types,” articulate the differences between these types, and use these insights to guide their future decision-making.

The sequence of three touchpoints described above makes for a very different consumer journey than a sequence comprised of five consecutive display ads, and comparing and contrasting these sequences can teach a marketer a great deal about each consumer — and other consumers like them. Event sequence modeling techniques help marketers determine when multiple sequences of touchpoints are similar enough to be considered variants of the same type of consumer journey, and provide a roadmap on which marketers can plot a new sequence of touchpoints that comprises an incomplete (that is, conversionless) journey.

For instance, imagine that a marketer uses event sequence modeling to figure out that consumers whose journeys begin with Touchpoint A, progress through either Touchpoint B1, Touchpoint B2, or Touchpoint B3, and continue with Touchpoint C can almost certainly be driven to a conversion with either Touchpoint D1 or Touchpoint D2. The next time the marketer encounters a consumer who has been served Touchpoint A, Touchpoint B3, and Touchpoint C, they will only have to decide between two courses of action (serving Touchpoint D1 or serving Touchpoint D2), dramatically reducing the likelihood of an errant strategic decision.


In short, event sequence modeling enables marketers to build next best action recommendations that take in a given set of circumstances and spit out a set of probabilities that describe the next steps that have the highest chance of driving a conversion (or other desired outcome).

Applying Event Sequence Modeling to the Patient Journey

Just as marketers have access to boundless reserves of data on consumers’ behavior vis-à-vis digital touchpoints, healthcare stakeholders — payers and health systems in particular — have access to extensive data on patients’ and HCPs’ behavior vis-à-vis the healthcare system at large. And while the challenges of using this data in meaningful ways take different shapes in each context, the underlying analytical ask is the same: how does one make sense of complex sequences of events and use this understanding to predict optimal next steps?

As highlighted above, the phenomenon signified by “the patient journey” is, in fact, an amalgam of thousands, if not millions, of distinct patient journeys. If a consumer digital journey is defined by an individual’s progress from Touchpoint A through conversion, a patient journey can be defined by an individual’s undergoing Test A1 through Test A3, being referred to HCP B, receiving Diagnosis C, being prescribed Treatment D1 and Treatment D2, continuing to fill Treatment D1 while lapsing from Treatment D2, and so forth.

By analyzing event sequences sourced from thousands upon thousands of patient journeys, sequential modeling algorithms can identify trends that (1) no single HCP will ever have enough context to identify themselves and/or (2) are so subtle that even payers and health systems with panoramic industry visibility will never be able to identify via traditional analyses. This is especially true when it comes to the identification and treatment of rare diseases.

Outside designated Centers of Excellence, individual HCPs may only encounter one or two cases of a rare disease throughout their entire career. As a result, patients with these diseases often end up bouncing between referrals for years — sometimes even for decades — before receiving a definitive diagnosis. While this is not so much a failure of American medical schools as it is an inevitable byproduct of the probabilistic nature of the practice of medicine, it is nevertheless a serious issue.

Event sequence modeling algorithms enable healthcare stakeholders to engage with these patients’ HCPs earlier in the patients’ journeys. Overlaying patients’ genetic and demographic data with analyses of the sequences of events that constitute their journeys can help payers and health systems understand what tends to precede the diagnosis of a rare disease. Not unlike leveraging insights into consumer journey types lets marketers predict which final touchpoint is needed to drive a conversion, leveraging insights into patient journey types lets healthcare stakeholders predict what step is needed to precipitate a desired outcome.

For instance, an event sequence modeling algorithm might determine that patients who have undergone Test E1 through Test E4 in a certain order, been referred to a series of specialists according to Referral Pattern F, and been prescribed either Treatment G1 or Treatment G2 are likely to have Rare Disease H. Without such an algorithmic analysis, a patient may have to wait years for this diagnosis; with it, a payer or health system can inform the patient’s HCP that the sequence of events constituting the patient’s journey through Referral Pattern F adheres closely to the event sequence that typically precedes the diagnosis of Rare Disease H. The HCP can then evaluate the patient specifically for Rare Disease H — something they might not have ever thought to do if they had not seen the disease manifested in a patient before — potentially enabling the patient to receive proper treatment years before they would have otherwise.


Facilitating a Proactive Approach to Healthcare

It is difficult to overstate the value of standardizing this degree of proactivity across the healthcare industry. A world in which a health system is able to survey its claims data, map it against strategically abstracted “types” of patient journeys, and identify (and advise) HCPs whose patients’ claims histories suggest they may be progressing along specific types of journeys is a world in which healthcare resources are utilized more efficiently and, more importantly, patients receive the treatment they need as early as possible.

Bringing such a world into existence will not necessarily be easy, but, as conversion attribution modeling has illustrated in the digital marketing space, the tools and techniques for doing so are readily available. Event sequence modeling can give key industry stakeholders insight into critical patterns of behavior across the broader healthcare landscape, facilitating not only efficient diagnostics, but improved disease management, comprehensive medication adherence programs, and more precise, more empathetic healthcare marketing.

As Danish philosopher Søren Kierkegaard once mused, “Life can only be understood backwards…but it must be lived forwards.” Sequential clustering gives stakeholders across the healthcare industry the ability to mitigate this disconnect, applying the lessons of patients’ past behaviors to their current efforts to help patients live longer, better lives moving forward.

Vlad Ryvkin

Vlad Ryvkin

Associate Director, Data Science

I have been working in or studying healthcare for approximately 9 years. I have a B.S. in biotechnology, and an M.S. in biomedical sciences. After completing my studies, I was an analyst for EmblemHealth, New York’s largest non-profit health insurer. There, I worked on many of their health-related quality measures, including osteoporosis, rheumatoid arthritis, and medication adherence.

I joined the Saatchi team in April of 2016, where I have been working in the fields of women’s health, surgical products, and rare disease. I have worked with all kinds of medical and marketing data and am proficient in SQL, SAS, and R programming. My expertise lies in data science and business intelligence.

Gray_Mackenzie_headshot_smallMackenzie Gray

Senior Analytics Associate, Data Science

I have a B.A. in Biochemistry and M.S. in Data Science from New College of Florida. I have always been interested in both healthcare and machine learning, and have found the field of Data Science to offer a fascinating and powerful approach towards merging these interests.

I joined the SSW analytics team in the summer of 2018 as the Data Science & Engineering intern, and came on full-time in January 2019.  My work has focused on developing new analytics product offerings and supporting current ones with Python, R, and SQL, with an emphasis on network analytics and machine learning.

In my free time I enjoy watching basketball, movies, and playing jazz guitar.

What do you think?

This site uses Akismet to reduce spam. Learn how your comment data is processed.