A market-wide longitudinal patient database creates great potential for driving business value for a specialty drug manufacturer. It can support a deep understanding of patients, their holistic situations and their treatment histories. It can enable laser-focused strategy and program development. It can create the ability to predict who will be future patients of specific therapies, who will make those treatment decisions and who will pay for them. Recent blog posts discussed Why and How, but conversations on this topic don’t end there. A frequent question about longitudinal patient data is, “what do I do with the data to find actionable insights?”. In other words, I don’t want to drown in a data tsunami!
Before we can look at the types of analytics that illuminate insights from this data, we must first establish some important assumptions. For our specialty product that is primarily pharmacy-benefit, we’ll assume the following:
- Specialty pharmacies provide daily Rx status updates (in process, shipped, denied, cancelled, discontinued, etc) with attributes that describe the regimen, de-identified patient, physician, payer and pharmacy
- A third party data vendor provides pharmacy claims data for all products in the market basket, and for all products used by patients with the relevant diagnosis
- A third party data vendor provides medical claims for HCP-administered products in the market basket, and for all care provided to patients with the relevant diagnosis
- All specialty pharmacy status, pharmacy claims and medical claims data is integrated with a comprehensive customer master that includes customer affiliations, and with a payer master
Now we can get to the heart of the matter, which is — what to do with this data to find meaning. The following types of analysis start as basic and foundational, and then add layers of analytical sophistication.
N of 1: Inspect the details of individual de-identified patients from a random sample of patients in the database. This establishes a foundation of understanding and empathy for patients, their history with the disease of interest and relevant treatments, their holistic situations with comorbidities and other treatments, and how treatment algorithms are being applied.
Characterize the cohorts: Identify the types and measurable attributes of patient cohorts in the database. Which patients and attributes are only in the specialty pharmacy data? Which patients and attributes are in the specialty pharmacy status and pharmacy claims data? Which patients and attributes are in the pharmacy claims and medical claims data? Which patients and attributes span all three core data sets: the specialty pharmacy status, pharmacy claims and medical claims data?
Scale-up models: The specialty pharmacy status, pharmacy claims and medical claims datasets represent subsets of the overall market. How will analytical outputs from the different cohorts be scaled up to support market-level interpretations and models?
High-level statistics: What are patient count and utilization distributions across payer type, benefit type, diagnosing physician specialty, treating physician specialty, physician affiliation, patient age, patient gender, geographical region, dispensing class of trade, etc?
High-level metrics: What is the performance on core metrics like sales trend, market share, adherence, persistency, specialty pharmacy fill rate, specialty pharmacy time to fill, etc?
Drill-down into metrics: How do the core metric results change when viewed by payer type, benefit type, physician affiliation, patient age, patient gender, geographical region, dispensing class of trade, specialty pharmacy, etc? How do the results change when viewed through combinations of attributes, e.g. payer types within a geographical region? What are the significant attributes and combinations of attributes that identify good or bad pockets of metric performance?
Metric performance prediction: What are the attributes and combinations of attributes that support reliable prediction of the prescriptions and patients that will likely have good or bad metric performance?
Scale up N of 1: Apply different patient journey and treatment algorithm segmentation approaches to the data, to create unique quantitative lenses into the market. What do the statistics and metrics look like through these segmentation lenses?
Drill-down into patient journey and treatment algorithm segments: How do the statistics and metric performance change when the patient journey and treatment algorithm segments are viewed by attributes or combinations of attributes? Which are significant?
Future patient prediction: By looking at the historical journeys and treatment histories of current and past patients, can we create reliable prediction models that point to the people today that will be our patients in the future? What will be the signals for those emerging starts on therapy?
These analytics will find the meaningful insights from a longitudinal patient database, but we are still not at the endpoint: ACTION. Multiple layers are still necessary to convert meaning into action. An obvious layer is organizational, i.e. the roles and responsibilities that align to the transition of data from sourcing to data management to analytics. Another layer is the business process layer. Converting meaning into action requires business processes that support an iterative and ongoing questioning and drill-down process. An example business process could be a Retention Task Force that presents retention analysis results to a cross-functional group of brand, functional and medical subject matter experts, facilitated by business analyst and data science experts, in order to digest the analytical results, create new hypotheses for the next round of analysis, and identify potential actions to take in the market.