Recently, there’s been a lot of buzzy talk about “building the data lake,” with some companies trying to build out and manage massive internal “raw” data stores. That’s great as far as it goes, but these efforts often suffer from a set of faulty underlying assumptions – namely, that data is scarce (it’s not) and once corralled in the data lake, technology will magically clean it up and render it useful (it won’t). Just because you have a garage full of car parts doesn’t mean you have a car.
The reality is that most companies are awash in unstructured, misaligned data they can’t use. Absent the talent and expertise required to realize value from it, these data lakes can quickly become costly data swamps, bogging down data efforts and depleting internal support while chowing down cloud and people resources. Sooner or later, absent tangible results, (i.e. real revenue generation and cost efficiencies) procurement and then senior management starts asking tough questions.
As my colleague Saurabh put it in the first installment of this series, “not only is healthcare data particularly ‘dirty’ and difficult to clean and structure, the highly complex nature of the life sciences makes matching data sets tricky.” Establishing business rules and writing code that gets those data sets to play nice in the sandbox together demands a team that pulls in a number of skill sets – clinical, therapeutic, commercial and data science.
Here are some key factors for realizing actionable data outcomes that we’ve identified in our work with clients:
- Prioritize a few data sets that can best address a business question – and accept some level of imperfection. When it comes to data sources, more is not always better – complexity equals cost. Why use five data sources where three would suffice?
- Ensure you have the right mix of expertise to tackle data mastering and integration, including deep knowledge of the therapy area, a provider’s-eye view of the clinical terrain and an understanding of the commercial considerations (e.g., coding and reimbursement factors when looking at claims data) that may come into play.
- Build out the technological infrastructure to support data management, along with the data science expertise to enable the data to “speak” and to maintain it. Housing and managing massive amounts of data securely can be challenging even for larger companies – partner where possible to lighten the load.
Getting it right can be hugely rewarding. For example, the brand team behind a blockbuster drug for a chronic condition was seeing prescriptions slipping in some markets, but they didn’t know why or how to stop the slide. Their formulary tiers, position and pricing were unchanged, as were their relationships with insurers.
Final claims data couldn’t tease out the cause. We needed to see into the “chatter” between the switch and the insurer, and that called for a claims data set covering every event in the adjudication process, from prescription to pickup. A comprehensive set of claims data, covering every step for the geography and therapeutic class concerned, was selected and overlaid with EHR data. Triggers were defined and a reporting mechanism built.
Analysis revealed that a large number of prior authorization requests were being denied by health plans in select areas. This allowed the brand’s commercial team to map out which physicians were seeing substantial rejections. Their sales team quickly confirmed the hypothesis – that patients and their physicians were getting tripped up filling out the prior authorization form, and physicians were simply giving up and moving on to the next treatment. Reps distributed cards with an 800-number to these physicians, who passed them along to patients, enabling them to get help navigating the form and access a needed treatment.
Do you have a pile of data sitting around, not generating value? Take a look at our white paper, How to Make Data Work for Life Sciences Companies, for further insights on getting your data capabilities up to speed, and get in touch – we’d love to compare notes and share what we’ve learned in assembling our own data operations.