Artificial intelligence (AI) and machine learning (ML) are more than trendy terms in the healthcare landscape; AI/ML technologies can provide consistency in the quality of day-to-day operations in order to focus resources more effectively towards strategy and analytics, building a positive lifecycle of continuous improvement for all stakeholders of a product or service.

The level at which one understands the functionality of AI/ML technology will greatly influence the impact that AI/ML has for business purposes. In the pharmaceutical and life sciences industry, imprecision as it relates to the use of technology can be costly, particularly when it comes to the delivery of value-based care.

With current patient access strategies optimized for persistence and repeatability, advanced digital technologies have the potential to vastly improve accuracy and precision of patient access services when they are employed appropriately. The accelerated pace of advancement in AI/ML, combined with the rapid increase in processing power and the expanding capabilities of connected devices, offers tremendous potential for manufacturers to tailor the patient’s journey. This can be a significant driver in transforming healthcare to achieve positive outcomes.

As we continue with this two-part series on intelligent interventions for specialty patient experiences, we’ll provide an overview of key AI/ML technologies, discuss patient access use cases, and gain a better understanding of key drivers for successful usage of AI/ML technologies through the help of IQVIA’s ValueTrak.

Depictions of AI/ML

Still in their infancies, AI/ML technologies are not interchangeable. ML refers to algorithms that improve automatically through experience and by the use of data. ML is also considered a part of AI, which is seen as any technique whereby computers mimic human intelligence. Examples of AI/ML include Google’s DeepMind, which can make connections and draw meaning without relying on predesigned behavioral algorithms. It becomes smarter and more aware over time.

Natural language processing, a subset of AI/ML, enables computers to understand the spoken word. Think of Alexa or Siri, devices that can predict and understand natural language questions and requests. Platforms such as Netflix utilize highly predictive technology based on viewing habits.

Deep learning is an evolution of ML where the software trains itself. With autonomous vehicles, for instance, multiple AI models work to inform how a car will act. Some deep learning models specialize in recognizing street signs, while others are trained in recognizing pedestrians.

Applications of AI are continuing to penetrate many aspects of daily life, from ride-sharing apps such as Uber and Lyft, which match drivers with destinations, to Gmail making recommendations on an email response and smart thermostats assessing living habits.

Thinking about each of these areas of penetration, manual repetitive tasks are eliminated, new opportunities arise, and results, such as customer satisfaction and outcomes, are achieved.

Implementing AI/ML

Starting slow with AI/ML is completely acceptable, and the AI/ML journey can be implemented progressively. The key is to be involved. Consider tasks that are manual, repetitive, or, worse, manual and repetitive. If outreach communications are based simply on frequency and persistence, there’s an opportunity for automation and personalization. If you’re spending time curating data to assess operational performance, there is an opportunity to standardize and focus time on drawing actionable insights.

This could be a multi-year journey. At the onset, seek to automate manual tasks. More processes can then be evaluated for efficiency, while resources are allocated to gain new insights and new opportunities can be pursued as costs decrease.

Implementation begins with forming questions that AI is expected to support. Questions should be clear, such as “Which prescribers should I call on to prevent prescription abandonment?” or “Which patient type is most likely to benefit from an additional outreach?” The data then needs to be assessed to determine if the right data is available to answer the question.

For example, in the first question, are prescriber, patient, prescription claims, communications, and adherence data offered? If not, the right data is needed. Once the question is posed and the data is acquired, algorithms must be created, fit for purpose, and adjusted as needed. There will be trial and error with algorithms, and insights must be generated with contextualization to be actionable. With prescription abandonment, it would be more meaningful to know when to intervene and which actions are recommended to prevent that abandonment. Those insights in that specific context would prove useful for AI.

In deployment, processes need standardization and scaling, as with any operation. Proof points should be established and validated, and rigorous quality assurances and controls should be implemented for continuous improvement. Data is constantly being collected and more data is always available to sharpen algorithms. Continuous testing, data gathering, and fine-tuning will improve technology.

AI/ML use cases and ValueTrak

Patient services offer many use cases for AI/ML. The following is a brief list to consider:

Healthcare provider alerts. Triggered alerts for the hub or field teams to coordinate with prescribers based on identified opportunities and recommended actions. Patient trends and patterns are utilized to predict outcomes and suggest interventions.

Personalized automated messaging. Interactions can be personalized, and messages can be triggered to help complete tasks or find information.

Operational alerts. Power operations to track and monitor operational health and performance with recommended actions to bring service in line with expectations.

Recovery suggestions. Lost revenue or missed opportunities can be recovered by prompting action.

With ValueTrak and its core capabilities in data aggregation and validation, and with its services in the proximity of specialty patient data, steps have been taken to translate data into insights:

• The Intelligent Intervention Suite has pre-built specialty dashboards that are customizable and provide an at-a-glance view of key patient, status, program, prescriber, and partner metrics with contextual trends. Suitable for case management teams, mid-management teams, and executives, the dashboards utilize aggregated and validated data to present information with ease. Users quickly understand the health of their programs and operations with the opportunity to drill into details.
• An alerts system can be set to identify when key metrics meet expected targets or fall from acceptable ranges. Alerts can also be used for strategic purposes, perhaps being used in conjunction with benchmarking. Alerts can be tactical, driving mid-management analysis of trends, strengths, and weaknesses. Along with core value-tracking capabilities, pre-built dashboards and reports provide the ability to react to performance of key metrics.
• Pattern and trend recognition of patient cohorts can be utilized to predict outcomes and recommend interventions. Teams can drill into the patient journey and be armed with the information necessary to engage with prescriber stakeholders.

Data ingestion alone can account for 60-80% of model development time. A reliable product and partner are required because ingesting data comprehensively is difficult but vital for operational efficiency and optimized results. Good governance is a must-have and always a vital best practice. A meaningful data partner can assist with creating a central repository, providing access, promoting data, allowing for data exploration, and encouraging the organization to build models and identify use cases.

In this way, AI/ML can be another lever to drive commercial value in addition to traditional methods and be a powerful enabler of commercial success. AI/ML has the potential to drive smarter decisions, unleash new opportunities, and help in discovering previously unseen insights efficiently.

We dug in deeper to this topic at the recent Fusion 2021 conference; if you missed this year’s sessions, click here to watch them on demand!