Building an Oncology Data Visualization Platform to Leverage Integrated Patient Data and Analytics

Building an Oncology Data Visualization Platform to Leverage Integrated Patient Data and Analytics
Gita Mishkin, MPH, Principal, Brand Analytics and Center of Excellence, Symphony Health; Paula Fullman, MSOD, Vice President, Patient Analytics Center of Excellence, Symphony Health; Don Faust, Lead Consultant, Patient Analytics Center of Excellence, Symphony Health
Abstract:
Understanding the patient journey and therapy utilization relies on familiarity with the granularity and complexity of an integrated dataset. It pulls together data from various resources, including, but not limited to, retail and mail order pharmacies, wholesalers, specialty pharmacies, hospitals, community-based offices, clinics and other healthcare delivery facilities, clinical registries, electronic medical records, and lab data. This allows pharmaceutical managers to develop strategy based on a holistic view of the brand (patient, payer, prescriber) experience without having to make leaps of faith across disparate data sources or to guess whether or not various data assets reveal different answers. To this end, the HealthCloud (an oncology data visualization application) allows for drag and drop insights and graphics for an otherwise analytically complex therapeutic area. Five milestones streamlined the process: (1) Identify key questions; (2) Develop business rules, including phases of implementation; (3) Design the core dataset needs; (4) Isolate data aggregations requirements to optimize performance; (5) Create dashboard for data visualization. The primary lesson was the importance of a development plan, including all the phases of release and roles and responsibilities.
 
Keywords: patient, integrated dataset, data visualization, oncology, dashboard, analytics
Background
Patient analytics
Patient analytics must be utilized as a foundation for driving healthcare monitoring and change, whether it be in market planning, marketing or operations. Four categorizations for patient analytics currently exist: (1) Prescriptive – identifies actions, (2) Predictive – examines likely scenarios, (3) Diagnostic – historical view of performance, and (4) Descriptive – what is happening now.1 Typically, the pharmaceutical industry utilizes diagnostic or descriptive analytics to monitor brand activity or marketing trends. Data is pulled from a variety of different forums, including Centers for Medicare and Medicaid Services, health plans or claims data aggregators. 

Integrated Data
Integrated data pulls together data from various resources, including, but not limited to, retail and mail order pharmacies, wholesalers, specialty pharmacies, hospitals, community-based offices, clinics and other healthcare delivery facilities, clinical registries, electronic medical records, and lab data. Integrated data utilizes a unique identifier to allow for linking of the information across all the different sources. By doing this, the addition of new data sources, such as biomarkers, becomes simpler. This type of data enables pharmaceutical managers to develop strategy based on a holistic view of the brand (patient, payer, prescriber) experience without having to make leaps of faith across disparate data sources or to guess whether or not various data assets reveal different answers. Understanding the healthcare dynamics within a particular disease state, such as a patient journey or therapy utilization, relies on familiarity with the granularity and complexity of an integrated dataset.

Data Visualization
Data visualization is a way to virtually present and manipulate data by using a business intelligence tool, such as Tableau® or Qlik®2. The visualizations vary in complexity from simple graphs and charts to heat maps and forecasting. The ability to drag and drop insights, filter by specific metrics or attributes and creating graphics specific to analytics of interest are all advantages to using a data visualization business intelligence tool. For the purposes of this specific initiative, Tableau® was utilized.

Oncology Therapeutic Area
Due to the complexity of the different tumor types and the medical regimens associated with each individual tumor type and the data complexity associated with various treatment delivery, the oncology therapeutic area was used to build the first HealthCloud. Oncology presented some unique opportunities to develop methodology for determining mechanisms for scaling across tumor types, such as starting with lower prevalence and characterizing complexities and nuances associated. For the purpose of planning the development of the dashboard, thorough research went into the definition of specific tumor types, including creating a comprehensive list of all the tumor types and estimating the potential magnitude of data, understanding what types of information was readily available via online resources, and engaging with a subject matter expert. In addition, the National Comprehensive Cancer Network® (NCCN) manages guidelines associated with many different tumor types, and their website was used as a primary source of information for each tumor type. These guidelines were accessed in order to identify the specific medications, develop treatment regimens, and understand physician and patient concerns. 

Based on the Oncology Therapeutic Area review, multiple myeloma was the first tumor type to be included in the first phases of the HealthCloud development. This is based on an estimate of approximately 115,000 multiple myeloma patients found in the 2016 data utilized for this analysis. In addition, the NCCN has a detailed patient guidelines book for review of treatment regimens.

Methods
The Oncology HealthCloud team identified five key milestones in order to streamline the process of development: (1) Identify key questions; (2) Develop business rules, including phases of implementation; (3) Design the core dataset needs; (4) Isolate data aggregations requirements to optimize performance; and, (5) Create dashboards for data visualization. The implementation of each stage was accomplished by either a design team (responsible for defining the overarching results and analytics to be displayed) or a development team (responsible for creating the datasets or the data visualizations). Each team member has a unique skill set to allow for cross functional expertise and input.

First, the design team identified some key business questions for developing the Oncology HealthCloud. These questions defined the purpose of the HealthCloud, as well as the desired metrics and outcomes to be displayed in the dashboard. These questions examined patient demographics, such as age, gender, race/ethnicity and geography, baseline diagnosis, baseline procedure, prescription metrics, line of therapy, treatment regimen and source of business.

The next step was for the design team to develop the business rules. These were broken out by the specific phases of implementation. During this step, two items were completed: (1) a spreadsheet clearly outlining all the metrics, attributes, and data filters requested for each of the analytics.  2) Visualization requirements for the analytics are best developed during this step as well. Identifying the types of graphics, whether maps or line graphs or tables, during this step will help streamline the following steps. The analytics document was reviewed in depth with the design and the development team.

The first phase contained several types of analytics, some basic and some more complex. Baseline counts, such as demographics of the patients and diagnosis, procedure, and prescription profiles (i.e. number of patients, number of claims, payment information) were created. Line of therapy, source of business, and plan control indices were also programmed into the first phase of the HealthCloud.

During step 3, the core dataset structure was designed by both the design and the development teams. Due to the magnitude of the integrated dataset used, it was important to identify specific fields to be included in a large data pull. This bolus of data was used as a large underlying dataset that could be housed on a data cloud. The large amount of data remains on the cloud with the ability to pull additional fields as needed for the HealthCloud. The development team created a technical document reflecting all the core dataset as requested by the design team.

Step 4 uses the data pulled from step 3. The magnitude and design of data pulled into a data visualization tool will directly impact its performance. Understanding the analytics designed for each phase, and how all of the different fields interact with each other, will help maximize the performance of the tool. One aggregated table gives the data visualization tool a smaller more limited view of the larger dataset. The data visualization tool does not need to search through a large dataset in order to find each requested field, but rather, a smaller, more specific dataset.

By step 5, the data visualization developer had the tools they need to create a workable dashboard. During this stage, ongoing meetings between the development and the design teams help streamline the process. The development team is responsible for creating the dashboard based on the analytics requirement document and the technical document. The design team was responsible for reviewing the dashboard as it is created and providing input for modifications, since they will present the dashboard to clients or internal stakeholders.
 
As each phase reached it final stages and the revisits to earlier steps slowed, the design team began to plan an internal training and roll out plan. To control the roll-out and change input, the design team chose a representative from each part of the commercial organization. This representative provided specific input on methodology and visualizations, which were later incorporated in the dashboards. Once modified, the design team conducted a larger training for the entire commercial organization. This included basic training on the data, Tableau® user training, and finally interpretation and management of the insights created in the HealthCloud. 

Discussion
Many lessons were learned during designing and developing this data visualization platform. The primary lesson learned was how to optimize project management across a number of stakeholders with many competing priorities and a process that was prone to both forward and backward movement. Executing the methodology steps was an iterative process and revisiting a previous step for modification was a regular occurrence. To overcome this obstacle, the design team created an itemized project timeline in Microsoft Project. This helped pinpoint delays, and allowed upper management to justify the need for additional resources to the project. In addition, the design team and the development team met once a week to review all the tasks required to create the dashboard. Future project phases and requirements could be started while waiting for resources to become available for some of the delayed steps.

A development plan clearly outlining all the phases of release and the roles and responsibilities of each team member optimized the design and development by capitalizing on individual knowledge and experience. The analytic plan clearly outlined the expectations for the output of the dashboard, and a technical document created a reference for database designers and the dashboard programmers.

Designed for internal use, development of this platform and tool often took a backseat to revenue generating work, and strong advocacy along with creative project management were required to keep this initiative moving forward. The project plan needs to provide for contingencies when priorities get shifted. The project leader frequently met with internal stakeholders’ for their agreement of action in cases where developer priorities were shifted. The ability to remain flexible, to shift gears to take advantage of resources when they were available and continuous monitoring of the work flow were key to success. 

The visualizations of the data also required a lot of review. A key learning early on was the difference in perspectives and knowledge about the data across the organization and the ability to align cross functionally to ensure that optimal utility within the visualizations required considerable time and effort along the way. There are several ways to represent data, and continuous review of the visualization helped provide ideas on how to alter the output to the most intuitive form. This also helped identify any issues with the data aggregations used to drive the visualizations. Often, restructuring the underlying tables was required to improve the display of information.

Conclusion
The oncology markets are complex and dynamic with new treatments and treatment paradigms evolving almost constantly. In order to effectively engage with and assist its clients who serve the oncology space, the need for an enhanced toolkit with the intention of providing proactive insights was identified and prioritized. The intention is to strengthen client knowledge and client relationships. As a market intelligence tool, it managed to make cumbersome, raw, integrated data more manageable with shorter analysis times and shorter time to insight. The analytical platform provides internal stakeholders and consultants with quicker access to patient, payer and prescriber related metrics for pharmaceutical and healthcare service related research. Client facing consultants are able to respond to client inquiries more quickly and to proactively offer insights to evolving market events. The faster insights allow for a better working relationship with clients or partners. In addition to helping provide faster information and insights to clients, this data, application and dashboards have also served as a training mechanism for employees. 

As mentioned in the discussion section, the HealthCloud development and design is an iterative process. As each new tumor type is added to the dashboard, and as new data or analytics are added to the platform, the HealthCloud will continue to change and expand. After completing the first phase of the dashboard (one tumor type and a set of specific analytics), this platform was rolled out to the larger internal organization, while still working on additional tumor types and analytics in later phases. To avoid misuse of the dashboard, or misrepresentation of the data contained within it, a large organization wide training was completed. While this maintained data security and integrity, it, equally important, allowed users to receive the proper education on the tumor types. The live user put the analytic platform through real use, and the real world feedback allows the development and design teams to modify the dashboard as necessary.

Each of the phases of analysis will grow increasingly more complex and include many more types of data and metrics. Lab data, biomarkers, and electronic medical records will all be added during the next phase of implementation. The analytics will be recalculated and recalibrated accordingly. Quality control reports will be created to help identify any changes in the underlying database. Persistency, compliance, length of therapy and medication possession ratio are all part of the phase 2 plans.

Although the steps involved in creating this dashboard were at times never-ending, internal stakeholders still maintained (and continue to maintain) the importance of this project. 

About the Authors
Gita Mishkin, MPH, is a data structure development and delivery professional in the pharmaceutical and life sciences industry.  She has over 10 years of experience in healthcare research, consulting, development and analytics.  She first joined Symphony Health in 2011 as a Practice Consultant with an emphasis on government relationships in 2011.  Since then, she has touched all sides of the SH business, including strategic partnerships, data provider contracting and onboarding, and brand analytics. Currently, Gita is both a Principal on the Brand Analytics team working with clients to develop datasets and analytics with the primary purpose of driving the client’s business forward successfully, and a Principal on the Patient Analytics Center of Excellence group tasked with project management and driving the development dashboards to showcase and monitor the Symphony Health data assets.  Key areas of expertise include integrated data, designing datasets and analytics within the confines of client rules and regulations, and brand management.  

Paula Fullman, MSOD, is a marketing analytics professional in the pharmaceutical and life science industries with more than 30 years’ experience in research and consulting. Paula has been with Symphony Health since 2003 and more than 13 years with the Brand Analytics Practice where she partnered with numerous pharmaceutical and biotech companies throughout the country to define business issues and develop cohesive, targeted solutions. Since 2017 Paula has led the newly created Patient Analytics Center of Excellence at Symphony where she continues to work with clients and to expand the use of patient data and analytics across the Symphony organization.  Key areas of expertise include patient and managed care analytics which assist in developing strategies and programs around brand management, marketing and reimbursement. 

Don Faust is a pharmaceutical and life sciences professional with over 20 years of experience, covering research and consulting, project management, and client services/operations.  Don joined Symphony Health in 2013 in the Brand Analytics Practice, most recently as Consulting Manager, working with various clients to help solve their business questions through problem identification, strategic solution recommendations and project design/implementation.  In 2017, Don joined the newly formed Patient Analytics Center of Excellence group as Lead Consultant, helping both clients and internal teams advance the use of the many data assets that Symphony Health has to offer.  Area of expertise includes the use of APLD data for brand management and marketing/market research uses.  
 
References

1 Four Types of Big Data Analytics and Examples of Their Use. 13 09 2017. http://www.ingrammicroadvisor.com/data-center/four-types-of-big-data-analytics-and-examples-of-their-use.

2 Data Visualization: What it is and why it matters. 13 09 2017. https://www.sas.com/en_us/insights/big-data/data-visualization.html.