Pharmaceutical Management Science Association

As the customer experience takes center stage, it has become increasingly important to transform customer engagement through the use of analytics and engaging stakeholders across multiple channels. Axtria CustomerIQ^TM is built to help you orchestrate customer-led, personalized omnichannel experiences. Axtria CustomerIQ^TM enables a unified planning, decisioning, and operations ecosystem that brings the following capabilities to life:

• Dynamic Customer Personas
• Journey Mapping
• Experimentation and Simulations
• Next Best Engagement
• Content Tagging
• Measurement and Optimization

Throughout this session, attendees will have the opportunity to see a demo and experience the platform firsthand by working on realistic use cases that you see in your day-to-day efforts.

Attendance is limited to pharmaceutical manufacturer representatives only.

The workshop highlights how WhizAI can equip commercial pharma companies with actionable insights to inform decision-making, support market access and pricing strategies, and improve the overall competitiveness of their products. Join us to get a hands-on experience of key payer analytics use cases and experience the platform in action:

• Identify untapped opportunities and barriers by utilizing payer, plan, provider and patient medical claims data
• Evaluate and prioritize payers and plans
• Accelerate time to market by embedding payer control into contracting strategy

Note: To participate in the hands-on portion, attendees should bring their laptops.

Attendance is limited to pharmaceutical manufacturer representatives only.

Troy has worked in global pharmaceutical R&D and Commercial businesses for the past 25 years. He has led teams for major drug programs and has expertise in clinical development, data science, RWE, digital health, precision medicine, launches, market access, start-up models and major business transformations across multiple therapeutic areas. He is a recipient of the prestigious Johnson Medal Award as part of the XARELTO® team effort to bring this innovative medicine to millions of patients around the world. He continues to build and lead high-performing teams dedicated to improving the lives of patients.

As Chief Commercial Data Science Officer, Janssen Pharmaceuticals, Troy is responsible for accelerating the global expansion of industry-leading commercial Data, Data Science, RWE, Precision Medicine and Digital Health capabilities for Janssen in order to drive the growth for our $50+ Billion worldwide commercial pharmaceutical business. He partners with our commercial, technology and R&D leaders across the globe to build future-facing teams and capabilities and to advance go-to-market strategies. He believes we are on the verge of realizing the true potential of advanced analytics, machine learning and AI solutions to have a direct impact on millions of patients and to provide a key differentiator to business growth in the coming years.

AI/ML methods have gained significant adoption in commercial operations across pharmaceutical companies, particularly for multi-channel, omni-channel, and Next Best Action (NBA) customer engagement strategies. However, the lack of robust customer-level data, particularly historical sales data for use at launch, poses a challenge for training these AI/ML methods with traditional ML-driven NBA models built on at least 6-12 months post-launch data. It is also widely recognized that a successful launch plays a key role in determining the future reach of a brand. This creates an opportunity for an improved NBA design for launch that adapts to the uniqueness of a brand and serves the unmet needs of customers from the day of launch. Fortunately, a substantial amount of primary market research (PMR) is typically conducted pre-launch and available for use. In this initiative, we sought to translate PMR-based insights, combined with customer preferences derived from product usage data, to build a precision targeting capability for products at launch, particularly in the rare disease space. This approach offers incremental benefits with improved timeliness, increased precision for marketing and better-informed launch readiness.

Pfizer has been challenged with rapidly navigating an unprecedented situation with the world’s leading Covid-19 vaccine (Comirnaty) and therapeutic (Paxlovid) through an ongoing pandemic. Never before has a vaccine been made available for widespread use in under a year after a novel disease was identified. For those infected, Paxlovid can greatly lower the risk of severe disease or death but must be made available to patients in a timely manner through novel distribution channels, all while facing high-profile barriers to uptake. To support the objective of expanding access to Covid-19 vaccination and treatment to as many people as possible in such a dynamic landscape, Pfizer has piloted and implemented a number of innovative analytics approaches that often leverage non-standard data sources to answer non-standard questions. These approaches have been supporting access throughout the treatment journey: starting with prevention by supporting vaccine booster uptake, through awareness to support more informed behavior by healthcare providers, to ensuring convenient and equitable consumer access to treatment. Actions driven by these analytical insights have helped us make a pivotal impact on not just Pfizer’s traditional customers but also various government stakeholders and public health priorities in support of our collective response to the Covid-19 pandemic.

Clinical Trials are extremely important as they are the gateway to FDA approval. Put simply, no clinical trials, no products to sell. Clinical trials are far from being a walk in the park. Indeed, up to 80% of studies fail to recruit the requisite number of patients in the agreed upon time. This talk is about our contribution to solve this problem. Let’s briefly discuss the business challenges, the business value, the solution, and then say a few words about the Predictive Model we developed.

Business Challenges. Site selection is paramount and, yet, it has never relied on a data-driven approach, which may explain the dismal level of failure. What’s more, a lot of time is unnecessarily wasted in the feasibility stage and, at the end, the selected sites just don’t measure up.

Business Value. They are as follows. Accelerate the site selection process to pick out Sites and Principal Investigators. Ensure racial and ethnic diversity, as this has taken center stage lately. Avoid the cost of choosing non-productive sites. Move towards owning the data and the expertise involved in site selection. Revisit our working relationships with CROs.

Solution. First, provide our teams with data that enables them to make informed site selection decisions. Second, deliver the data in dashboard format, so it is easy to use. Third, use AI and ML to assist in the discovery of effective sites.

The Model. It takes as input a description of the study to conduct and the profile and track record of the site under consideration, and produces as output the probability that the site will be successful in recruiting the requisite number of patients in the allotted time.

The Predictive Model operates as a module within a larger system which presents itself to the user as a GUI that displays a whole host of background information regarding the socio-economic, racial and ethnic, and SDOH information on the inhabitants in the neighborhood of the site.

The Model does two things. First, it predicts the odds the site will be able to recruit the requisite number of patients in a timely manner. This information is presented to the user who is the ultimate decision maker. By the way, the accuracy and F1-score of the Model are upwards of 90%.

Second, the Model provides assistance in monitoring the progress of sites and proposes course correction interventions as needed. This is best based explained with an example. Say 90 days have elapsed since the site was activated. Assume the site is to recruit 100 patients in 180 days but as of 90 days into the study, the site has recruited only 24 patients. What are the new odds of success? This is what the Model computes. It considers extending the deadline and lowering the target patients to recruit and suggests an optimal combination of the two. If that’s not promising (the odds of success is still low), the Model may suggest closing down the site and point to candidate sites to pick up the slack.

Coming soon!

The capabilities of Large Language Models and generative AI make a compelling case for wider AI adoption by enterprises in all fields including life sciences. ChatGPT has accentuated this case by bringing AI to the masses. AI offers several promising applications to enhance business strategy, knowledge management, and operational effectiveness. Expert domain understanding of the pharma industry is critical to apply AI successfully to its full potential in the pharma world. This presentation discusses ChatGPT, Large Language Models (LLMs), Generative Adversarial Network (GAN) and AI trends and the vision/implications for the future of Pharma Industry.

2022 has been catalogued as the year of AI in Pharma. This has been driven by the sharp increase in digital advertising spend because of the COVID-19 pandemic. Consequently, it appears appropriate to look at how Pharma embraced AI, particularly its application towards digital transformation and omni-channel marketing, as some of the most consequen-tial way of charting a new path towards a re-imagined customer experience, and marginal com-petitive advantage in an industry which has been trailing others in the field. AI driven recommender systems became an essential piece of the omni-channel marketing tech stack, and many actors quickly learned there is no silver bullet approach to developing one. Instead, key foundational compo-nents were identified as critical to the success of such endeavor: Business understanding and objec-tives framing, data infrastructure and datasets, scal-able customized AI solution, and business value measurement.

The post-pandemic era brought further evidence to the effectiveness of digital channels when syner-gized with field promotions in an omnichannel strat-egy. In this session we aim to propose an AI driven automated framework for omnichannel marketing to drive business impact by synchronizing various promotional efforts, both digital and field, targeted at influencing HCP behavior to meet short term and long-term objectives.

HCP classification is valuable only when the data is accurate, specific, and timely. In this presentation, we discuss the problems with the current taxonomy used to classify healthcare provider (HCP) specialties and propose an evidence-based approach to accurately classify HCPs based on transactional data and HCP-Health Care Organization affiliations. The proposed approach uses machine learning to handle the complexity and volume of the data and to learn patterns that resemble each specialty. The model was trained using physician-level summaries to assign the "likelihood of being a specialist" to all generalists (e.g, family medicine, NP/PAs, etc.), which helps segregate them into acting specialists and true generalists. This approach can provide a more complete, accurate, and granular view of HCPs and their true specialty.

It has long been the case that organizations measure their direct marketing and sales activities’ effect on physician behavior. External factors also play a role in physician decision-making. These factors can include competitors’ promotional spend, their sales force size and structure, and social conditions to name a few. With machine learning algorithms and higher computing capabilities, it is now possible to estimate the importance of these factors and their long term effects on the brand. This information can then be used to develop more effective strategies for inline brands, pipeline, and launches. This session will describe new approaches for determining the effects of external factors on physician and patient behavior as well as the means to simulate potential responses.

This session will be take audience thru a novel approach from macro level budget planning to marketing tactics and partners level promotional effectiveness measurement. Industry wide application on the judicious resources utilization, alignment on key performance indicators and pushing the envelope by focusing more on outcome based measurements vs activity based performance indicators.

The session be very useful for Marketing Leaders, Data Science/AI Leaders with finance leaders who are responsible for the budgetary allocation within their organizations to ensure timely & robust outcomes for the business.

In this analysis, we present a new omnichannel measurement framework that provides brands with channel level attribution for consumer media, HCP media and field force engagement, along with optimal frequency and sequencing across channels.

Industry sources mentions that up to 10% of the pharmacy claims submitted may fall under fraudulent claims. In order to minimize the impact to the bottom-line, it is crucial for pharmaceutical companies to invest in data & analytics driven initiatives to monitor and detect suspicious pharmacy activity before payment is made to these pharmacies. By using supervised/unsupervised AI/ML techniques, pharmacies indulging in fraudulent activities can be flagged for investigation and action.

The dollars saved can continue to support other patient support programs including co-pay and voucher to pass on the benefit to intended stakeholder i.e., “Patient”.

A measurement plan should be a mandatory component of every pharmaceutical marketing campaign we plan. But how do we do that? One option is setting up a simple A/B test and looking at its results before a large-scale rollout of our campaigns. In Tech, this may be a standard, no-brainer. But in many practical situations in pharma, we often find ourselves needing to measure effectiveness after the fact when the campaign has already run. At this stage, we can't conduct an A/B test. In such cases, we can employ non-experimental causal inference techniques, for example, PSM (propensity score matching) and Propensity score re-weighting (PSW), for conducting a retrospective A/B test. The intuition for propensity score matching is we look at each person who saw our campaign, find their long-lost counterfactual identical twin, and check for any difference in their outcome. We can discuss this further in the session!

John is an international technology executive with over 35 years of experience in the fields of data, advanced analytics, and artificial intelligence (AI). He is currently the Global Head of AI at EY.

John has been responsible for the global advanced analytics and AI function at a leading biopharmaceutical company where he led a team that developed and deployed over 25 analytical applications in 4 years. John was an Executive Partner at Gartner, where he was management consultant to market leading companies in the areas of digital transformation, data monetization and advanced analytics. Before Gartner, John was responsible for the advanced analytics business unit of the Dell Software Group.

Novo Nordisk is expanding into rare disease areas, such as Hemophilia A and B with inhibitors, Glanzmann Thrombasthenia, and Primary Hyperoxaluria Type 1, etc. by using real-world data and predictive analytics to identify potential or likely patients and link them to relevant clinician networks. Rare diseases can pose unique challenges, and Novo Nordisk's commitment to expanding its portfolio and investing in research and development can address unmet needs. Identifying patients with rare diseases is challenging, particularly in the absence of specific diagnosis codes. Novo Nordisk applies multiple filters and medical doctors' domain knowledge to create a cohort of confirmed patients. The company utilizes AI/ML algorithms and mapping hierarchy rules to prioritize physicians and generate a target list for multi-channel promotion. The ultimate goal is to provide better care for patients with rare diseases, who represent one of the largest underserved patient communities in the world.

The researchers from IQVIA and Horizon Therapeutics present an AIML approach to uncover referring providers who are not captured by the medical claims data for a set of infusion products. The advantages of performing this approach are obvious: 1) the model outperforms the rule-based approach; 2) significant features picked by the models can help researchers understand key drivers in referrals in the target market, 3) it also minimizes the reliance on human knowledge in a specific market, so it poses less pressure to the analyst. This innovation will help complete the referral networks that marketers and sales professionals can leverage in their commercial practices.

Asthma patients are at risk of acute exacerbations. Those exacerbations range in severity from mild and moderate flare ups which can be managed at home and outpatient clinics, to severe and life-threatening exacerbations that require encounters in acute care settings. Across all severities, healthcare providers seek to modify the risk of these events to decrease the risk of morbidity and mortality, as well as to in-crease the quality of life.

With an estimated prevalence of 26 million, the bur-den of asthma in the U.S. is considerable. Despite years of progress in mitigating the risk of acute ex-acerbations, it’s clear that unmet needs persist. Those needs are greatest for higher-risk sub-groups, and defining those high-risk strata through risk strati-fication methodologies is paramount to developing more targeted and effective risk reduction strategies.

To stratify the heterogeneous asthma population, we utilized a combination of predictive and clustering models. This approach helps overcome challenges of working with real world data and enables the dis-covery of distinct patient subgroups. Applying these techniques can offer an important source of intelli-gence for developing clinical trial selection criteria, supporting drug development and commercializa-tion, and identifying opportunities to improve patient outcomes. We hereby discuss the application of these methods in the asthma market.

Commercial investments are one the most expensive line items for organizations. Understanding its effectiveness and how it interacts with all channels is crucial for brands to succeed. With the shift toward digital messaging and engagement, the Unified Commercial Measurement and Optimization approach led Regeneron’s data analytics teams to unlock incremental returns on their investments. The unified approach eliminated a siloed measurement approach of top-down marketing mix and bottom-up HCP/DTC media attribution. This newly adopted analytics framework allowed Regeneron to start seeing the co–dependencies between digital media, professional promotions, and commercial drivers to optimize their campaigns in-market.

More specifically, the unified commercial measurement and optimization enabled Regeneron commercial analytics to:

• Balance traditional and digital marketing plans across partners, tech platforms, and content to drive market share – optimizing DTC, personal and non-personal promotions, and salesforce
• Measure omnichannel impacts of digital marketing initiatives on patient acquisition and adherence, accounting for both payer and competitive dynamics
• Test and implement next best actions across both patient and healthcare provider journeys to drive net conversion and adherence
• Programmatically activate digital media opportunities by prioritizing high-value audiences, maximizing qualified reach, and building better outcomes with marketing partners

Thyroid Eye Disease (TED) is a rare disease that often requires patients to visit multiple specialists for proper diagnosis and treatment. In the rare disease space sales forces are small and HCP time is limited, thus traditional targeting approaches may not be appropriate. A more efficient approach is identifying therapy-appropriate patients via medical claims and dynamically targeting their physician(s) at key points in their journey. There is a trade-off between the number of physicians reached, and the quality of the patient-physician pair.

This abstract proposes an optimization algorithm that matches therapy-appropriate patients to physicians by maximizing both the global number of pairs, and the strength of their relationship.

To that end, a subset of therapy-appropriate TED patients and their managing physicians were organized into a bipartite graph. The strength of their relationship was described by an assignment score, considering the length and number of visits, and the physician’s specialty or familiarity with the disease space.

The Hungarian algorithm was employed to maximize both the number of unique leads and their strength, and was shown to outperform manual selection. This methodology is adaptable to business-specific pairings of patient and physician types and leads performance feedback can inform future matches.

Case study showing how EMR-equivalent data provides high quality, longitudinal patient data to support product launch in an uncommon disease in Europe.

Case study showing how EMR-equivalent data provides high quality, longitudinal patient data to support product launch in an uncommon disease in Europe.

Uncommon disease background
Rare disease <2,000 patients in each major European country. Treatment is loaded towards a subset of the institutions life science companies have a list of target institutions.

Diagnosis via complex combination of genetics, diagnostic tests, plus S&S.

Progressive condition with acute relapses. Acute episodes are treated differently from progression – a key data target.

Until recently no labelled treatments.
Client need and data objectives
Detailed patient journey:

• Diagnostics and interaction with comorbidities
• Progression and treatment
• Relapse dynamics

Data objectives:

• Market structure
• Dynamic market evolution
• Validation of existing data sources
• Registries

The client need defined as follows:

• Longitudinal patient clinical records tracking from pre-diagnosis (EMR data)
• Data from target institutions
• Forward-looking tracking to monitor shifts in treatment behaviors and new treatments

Data capture and analysis
Proprietary data capture tools used to achieve:

• Diagnostic journey and adherence with guidelines
• High response from target tiers
• Representative data coverage
• Full EMR records on each patient, allowing for complex multi-dimensional analytics

Conclusions
The paper will show how this method can be applied

A large pharma company was preparing to launch a first-in-class oncology treatment targeting a very specific histology of a cancer type. As there are no ICD codes for the histology of the cancer, it was not easy to understand the patient’s journey or market, making it difficult to feed insights into a forecast. The pharma company’s intent was to identify the potential patient universe to assess the total market opportunity as well as find key physicians likely to engage with.

Business Objectives:

• Estimate the size of the potential patient universe by identifying patients who have this histology of cancer from diagnosis to treatment, as well as physician specialties treating this cohort of patients.
• Enhance the current understanding of patient characteristics, disease progression, and treatment dynamics to inform launch planning, including the sales force.
• Extrapolate the data findings to a national level to understand the base of patients for forecasting

AI, if left unchecked, can reinforce societal, racial and other biases that exist in the data, with the developer or in the model development process itself. AI models can de-anonymize data intentionally or unintentionally leading to proxy discrimination. This necessitates standard operating principles such as privacy, fairness, transparency and explainability. To this end we have developed a set of guiding principles and resources for the responsible development of AI-based solutions. Next steps include the codification of these guidelines into Standard Operating Procedures.

Coming soon!

The session looks to inform and review the implementation of an AI-Powered marketing strategy. The ecosystem is created by designing and using predictive models to effectively target healthcare professional (HCP) with marketing efforts. The process creates a methodology that is more accurate in identifying high ROI targets vs. targeting strategy based on volume. It is flexible with data availability to deploy a brand’s marketing budget and resources in a way that will maximize the ROI across all sales and marketing efforts. Furthermore, the algorithm adjusts the approach as brands transition from their nascent state to maturity.

The scores can be a mechanism to gain strategic insight on how to reach each target, deciding which doctors are better reached through emails or digital marketing can decrease wasteful rep details. Afterwards, using unsupervised learning techniques to create macro and micro-segments that align to their RX-Value by channel.

The final product is a target list created by artificial intelligence that identifies the highest potential engagers, the most effective marketing channel, and the necessary effort to convert. Most importantly it is easily replicable and has the potential to bring a much stronger ROI to marketing efforts.

An overview of the approach used to create a forecast which leveraged a unique, internally developed data point captured and reported from the field. The outputs & insights which were used to develop and deploy strategic initiatives for the field to drive adoption of the brand during its launch year. Working closely with a cross-functional team we were able to create and set goals for the field to drive the call to action, getting patients in the funnel while having a tangible way of tracking success.

Rare diseases affect over 350 million people worldwide and are often misdiagnosed due to their low prevalence. There is a growing interest in using AI/ML models to identify patients with rare diseases using their Electronic Health Records (EHR) data. However, obtaining sensitive and regulated data sources, such as patient images, ECG, and HCP notes, for inference violates the Data Protection Act and puts patients' privacy at risk. Federated Learning (FL) paradigms have recently gained attention as they allow different medical institutions or clients to train a model collaboratively without any data leakage. The current talk will focus on patient identification problem in the Federated Learning environment with a focus on communication and statistical heterogeneity of data. The research extends the RareBERT transformer-based model to an FL-based framework named FedRareBERT. The session will uncover key insights around FedRareBERT performance through an ablation study on learning mechanisms, data heterogeneity, and aggregation which impact model performance.

For some learning models, the objective is to train a model that can mimic the decisions a human might make given training data comprising a sequence of actions and corresponding observations. The actions the human takes and the observations are how the environment responds. For example, in omni-channel marketing the actions might comprise reaching out to a physician via email, a video call or an in-person sales call. Observations might comprise the treatment initiation or switch, or prescription filled. The decision task is to choose the next best promotional channel and in some cases marketing or sales tactic to engage each HCP on consumer based on a history of such actions and observations. Imitation learning means the objective is to choose actions that mimic human decision-making given a history of such actions and observations for multiple different physicians and a machine learning method. The problem with imitation learning is that it’s not possible to “jump out of the system” of human decision-makers. This presentation will explain how combining utility with probability might facilitate such a jump to making better decisions than the humans on the system was trained.

The depth and prevalence of racial disparities are coming to the forefront of U.S. healthcare. Measuring and ultimately addressing care disparities facing patients of color requires a deeper view of real-world patient data.

This case study leverages race and ethnicity data to examine the disparity of care in retinal disease patients across various demographics. After stratifying by age and diagnosis, stark relative treatment differences were observed.

Diabetes-related indications are disproportionately seen in patients of color, and certain retinal conditions are overwhelmingly seen in minority populations. Among patients 65+, minority populations were up to 33% less likely to be treated with any anti-VEGF and up to 64% less likely to be treated with Eylea as compared to White patients.

Join this session to learn how to leverage expanded demographic data and the full patient journey to shape solutions to improve health equity.

Healthcare providers and patients are increasingly accustomed to finding pharmaceutical information at their fingertips in their preferred channel of communication. Accordingly, pharmaceutical companies aspire to integrate communication across all the channels into a seamless experience for the customers to actively steer them towards a positive outcome. The impact of evolving digital promotional media (channels like email, phone, digital programmatic media, Electronic Health Record systems, social communities and growing) translates into a complicated and complex matrix of all possible ways an organization may integrate these channels into a well-orchestrated omnichannel strategy.

In this presentation, we will share learnings from experience working with four leading multinational pharmaceutical firms on global omnichannel marketing. This experience spanned all stages of the omnichannel journey covering strategy and program design, model building, deployment into production, and ongoing operationalization. We discuss the leading algorithmic approaches for turning data into actionable omnichannel insights. These approaches will include a mix of established as well as advanced methodologies (e.g. reinforcement learning) and innovative technologies (e.g. quantum computing). We will discuss key learnings for successful organization adoption and deployment, to turn insights into action.

This presentation will review the medical records curation approaches and their impact on data quality and clinical research insights.

During the discussion, the audience will be introduced to patient registry concept, the types of data collected, and research that can be supported. The patient registries, otherwise, also known as disease-based communities, require engagement with the participating healthcare institutions and enrolling qualified patients to aid understanding of the population characteristics, as well as therapy protocols and regimens, time to and on therapy, discontinuation trends and reasons, and healthcare outcomes. Since the data collection includes Electronic Medical Records with the ability to append patient reported outcomes questionnaires, as well as any other data sources relevant to the condition study, the resulting datasets provide in-depth insights on the patient population and their treatment pathways. KOLs and treating physicians are the Principal Investigators, managing and monitoring the patient progression and treatment care.

Furthermore, the data curation approaches and management, from manual to Natural Language Processing, will be reviewed and their advantages and dis-advantages summarized. For example, manual data curation provides the opportunity for in-depth understanding of selected data elements, such as scans and provide notes, but often requires appropriate amount of time for curation and proper training for the professionals curating the data points. On the other hand, leveraging validated Medical Language Processing and Machine Learning algorithms provides the ability to reliably convert the unstructured EMR information to structured data to create a unified data model while also allowing for insights’ generation and limiting the amount of time needed for data onboarding, processing, and standardization. The data quality as a result of selecting or applying either approach can differ and proper statistical methods for data validation needed to be selected as well.

In summary, this podium presentation will provide audience with a good understanding of the currently available approaches to medical data curation and their impact on the data availability, quality, and resulting research applications.

We propose two models for estimating dynamically-changing HCP engagement rates for different promotion types (channels and/or content). The two models reflect the subtle but critical information about the availability of data for different types of channels.