• Different PSPs’ contributions toward patient behavior were measured with increased accuracy.
• Exhaustive datasets were built based on patient info such as demographics and adherence.
• A standardized capability for model building was created to produce readily consumable outputs.

New opportunities unlocked with Math + Engineering + Design

• The program measurement was replicated across brands for multiple drugs.
• The framework doubled as a one-stop shop for all patient-related explorations.
• Process complexity was eliminated by evaluating and standardizing PSP impacts.
• User-friendly interfaces were built to enable intuitive user navigation.

Each year, healthcare and pharmaceutical organizations earmark exorbitant amounts of money to tackle one of the most widespread concerns in the healthcare industry today: patient non-compliance. This issue is particularly relevant in the context of long-term medication used to treat chronic diseases, as an estimated 50% of patients do not take their medications as prescribed, reducing the effectiveness of treatment regimens. This non-compliance costs businesses around $100 billion annually and could even result in fatalities [1].

There are numerous reasons for patient non-compliance, but they can be broadly classified into the following 3 categories:

• Patient-related: e.g., poor health literacy, lack of involvement in the treatment decision-making process etc.
• Physician-related: e.g., prescription of complicated regimens, communication gaps, treatment by multiple physicians etc.
• Healthcare-system–related: e.g., restrictions on office visit times, lack of access to care and health IT etc.

While the problem could not be attributed entirely to existing adherence intervention methods’ low efficiency, there was a need to change the way patient support was approached. In this regard, PSPs are critical since they assist healthcare and pharmaceutical organizations in ensuring systematic patient compliance. However, because the barriers to medical compliance are so numerous and varied, adherence techniques and solutions must take into account a number of factors such as side effects, patients’ financial situations, and PSP accessibility, not to mention the sheer volume of patient data that must be processed to extract any actionable insights.

Our client, one of the world’s largest biopharmaceutical companies, wanted to work with TheMathCompany to address this very issue. They wanted an exhaustive patient analytical dataset creation framework to understand which patient support programs (PSPs) were leading to the highest levels of compliance from patients. In order to systematically meet this requirement, TheMathCompany created a minimum viable product (MVP) as part of the complete project timeline’s initial phase. The MVP focused on patient compliance for one specific drug sold by the firm – a long-term prescription injection used worldwide. Used to treat chronic inflammatory conditions such as arthritis, the drug is administered through under-the-skin injections. However, post-administration adverse reactions – such as site reactions, nausea, and headaches – as well as potential side effects – risks of nervous system issues, allergic reactions, and immune reactions – had led to high rates of patient non-compliance.

The client company also required a framework with the capability to analyze various types of data, such as past medicine intake and patient drop-off data, to gain insights into patient behavioral patterns, various barriers patients face during medical therapy, as well as the PSP methods that could help improve patient outcomes. As such, we created the framework in such a way that it could be extrapolated for use to various other business use cases across brands with minimal coding required.

The framework was created so that customers could use user-friendly widgets to configure both the patient analytical dataset (ADS) and the modeling technique of the solution. This model-building capability could then be scaled across brands and business use cases. The base tool contained prebuilt modules, enabling users to create the ADS and build their own models with minimal intervention. Additionally, because of the client's global reach, the framework also had to be scalable across numerous medications and therapeutic use cases to account for all conceivable independent and multiple dependent variables. The client also wanted a no-code framework, along with a user-friendly consumption interface.

Figure 1: Project Roadmap

The project was divided into three phases, with each phase being built depending on feedback from the implementation of the preceding phase.

With the client’s problem statement in mind, our teams set out to build an ML-based analytical framework that accounts for various types of data and improves patient outcomes using a three-phased approach. The project timeline was defined as such:

1. The aim of Phase 1 was to build a minimum viable product (MVP) for the drug in question. The MVP had limited functionalities: a framework to formulate a master ADS through the analysis of a select choice of dependent (patient adherence) and independent variables using limited choice of techniques and input data. This can then be used in a modelling framework.
   
   
   
   
2. After demonstrating the MVP and receiving feedback on the same from the client, our experts scoped out the capability of the ADS to be extended to include more types of data such as geo data, HCP data, and web data using a feasibility study. The possibility of implementing a data quality management (DQM) layer with respect to impact measurement was also considered. As part of the second phase, a new exhaustive dataset was created, encompassing new dependent and independent variables along with modularized models involving parameterized inputs and a built-in tuning functionality.
3. In the final phase, the standardized framework was used to generate analytical capabilities for other drugs using different filters and parameters. End-to-end ML modelling capabilities were integrated and implemented to supplement data quality management and broader PSP measurement.