2019 Poster Presentations

An All-Encompassing Machine-Learning Attribution Model to Plan, Predict and Measure Impact at the Channel and HCP Level

Mark Ortlepp, IQVIA
Justin Hawver, IQVIA

In today’s omni-channel world, analytics has come a long way in measuring and predicting impact at the channel level to understand where marketing dollars should optimally be spent. Questions about who to target, which channels, and how much to spend per channel will always be with us. To answer these questions and more, IQVIA has pushed the limits of machine learning to develop a multi-step ensemble model. After data pre-processing, Bayesian Machine Learning is used to update the decay parameters for each channel. Since we use a sampling algorithm to estimate the posterior distribution, this can be very time consuming. The attribution side of the model is established in the final step when estimating the impact of each channel at the HCP level. This is done by using the updated decay parameters with Bayesian Machine Learning. The HCP level model is a Bayesian regression model, built with constraints and prior knowledge of channel impact values applied from the channel level model. A separate model for each HCP is created, making the process time-intensive, but extremely valuable.

Compared to standard approaches, the attribution model has significantly more Rx impact prediction power for multichannel marketing campaigns. Using the HCP-level outputs of the model, micro-segmentation allows for customers to be grouped into viable segments to receive specific channels and/or creatives with customized messages that will resonate with the perspective audience. Non-linear channel response curves using the attribution model result in the ability for marginal attribution to be measured, allowing for investments to be optimally distributed to maximize ROI. IQVIA’s model can also be refreshed when a campaign is live to support campaign optimization decisions. This advanced modeling solution captures Rx impact behavior changes within a few months to show what channels, tactics and metrics need to be adjusted to improve campaign outcomes. Results from past and current campaigns are incorporated and are leveraged by marketers. These revised models and insights keep HCP preference and channel mix selections relevant and readily available for the next round of campaign planning, establishing a culture of continuous learning and optimization.

Big Data, Small Conditions: Challenges and Insights From Digital Measurement of Rare Condition Campaigns

Sean Bell McDermott, Analytics Supervisor, Crossix
Digital marketing via display, video, and mobile advertising have become increasingly important components of direct-to-consumer tactics. Traditional click & keyword measurement techniques can help optimize advertisers’ bid strategies for purchasing digital media, but it is not at all clear how well a consumer’s engagement with an ad is correlated with the underlying likelihood that a qualified patient will seek out a particular test or treatment. This problem is exacerbated in the case of rare conditions with complex patient pathways: it is difficult or impossible to derive any real success metrics, let alone ROI, from a campaign where patients need to go down a long, windy funnel from exposure to treatment.

This presentation will include a case study that shows some of the challenges and successes of measuring an unbranded campaign for a rare disease from end to end. This study shows that granular measurement of new patient starts is applicable to and can be feasible for campaigns well outside the large brand, common condition mainstream.

Delighting Customers with Automated Content Personalization

Michael Steward, Vice President – Analytics , Indegene
As more channels became available to pharma companies to interact with health care professionals (HCPs), the amount of content has increased proportionately. However, the abundance of information only generates more ‘noise’ for HCPs as they make every effort to remain up-to-date.

Pharma companies are looking at innovative approaches to streamline content that are most relevant and useful for HCPs by understanding customer journeys. Advanced capabilities in machine learning (ML) are increasingly being deployed at the enterprise level to generate actionable recommendations at an industrial scale.
  • Advance AI content creation and metadata generation - Advanced vision systems are trained to slice master assets into text and graphics components that are tagged by machine learning applications. The metatags become the foundation for the creation of new content by reusing existing components that are classified based on their media type, semantic category, and key message
  • ML graph algorithms to predict outcomes - Propensity models are developed and refined with useful touchpoint data in addition to the right content (knowledge graph), the right surround sound (relationship graph), and the right outcome (performance graph)

Determining Incidence for the First Time in a Growing Market with Limited Epidemiological Data

Varsha Damle, Director, Global Commercial Operations, Aimmune Therapeutics, Inc.
Peanut allergy (PA) is associated with high rates of severe reactions and impaired quality of life, yet epidemiologic data are limited. There has been no work done previously to understand the true rate of incidence in this therapeutic area. We combined a real-world, nationally representative cohort of patients’ medical and prescription claims, with findings from medical literature to determine incidence. We believe that this methodology will be of value to practitioners of commercial analytics in the pharmaceutical/biotech space with direct application to other emerging disease states with poorly understood incidence rates. Direct applications are market opportunity assessments for new product introductions, and accurate estimates of the volume of patients entering at the top of the forecast funnel to predict treatment rates and product revenues.

Dynamically Improving NLP & Theme Discovery Through Contextual Linguistics for Pharma

Jingfen Zhu, PhD – Chief Science Officer & VP, Analytics, Genpact
Kanishka Chatterjee – AVP, Commercial Analytics, Genpact

The paper deals with the science and the process of establishing a contextual linguistics program for pharma and healthcare companies to develop stronger insights into unstructured data - currently done through manual interventions or stock NLP engines. We have historically seen that NLP and text mining techniques have not yielded reliable results across industries, largely due to the lack of category based nuances within the LID (language identification layer). However, a credible solution to this challenge rests in developing and tuning a taxonomy that caters to the context of a particular therapy or category, making it possible for reliability and machine led insights to improve dynamically. Through this poster, we aim to present a glimpse of how we have been able to solve this challenge for a medical affairs function within a large Pharma and also potentially scale the solution to solve for the definitive unstructured data pool – social media.

How Can a Combination of Trial Subject-Level Data and Real World Data Improve Clinical Trial Design?

Michel Denarie, Sr Principal, IQVIA

How Payment Delays Can Affect Brand Marketing at Launch

Ashish Patel, Product Lead, CareSet Systems
Nilay Shastri, Field Operations Manager, Genentech

Is Data Fabric the Future for Data Lakes?

Ashish Sharma, Principal, Axtria
Rajesh Choudhary, Associate Director, Axtria

A Machine Learning Approach to Customer Acquisition

Brian Gibbs, PhD, Principal, Axtria Inc.

Multi-Channel, Multi-Perspective: Lessons Learned in Using Connected Patient Level Data to Optimize Mass Media DTC Campaign Performance

Adam Dubrow, Director, Advanced Analytics, Crossix
Mass media is alive and well, accounting for the largest share of direct to consumer (DTC) advertising spend. Recent innovations in analytics and the breadth and depth of data now available have made it possible to optimize mass media advertising efforts with greater accuracy and precision than ever before. This presentation will share lessons learned in helping pharma teams to measure and optimize the performance of their mass media DTC advertising efforts, with of focus on national TV, which is where the largest investments are made.

This presentation will share case studies demonstrating optimization insights uncovered for some actual mass media campaigns by harmonizing measurement and using connected data at the consumer level. These case studies leverage extensive connected patient level data sources to measure how marketing efforts work together across channels to drive patient outcomes and to enable data driven, forward looking optimization and scenario planning.

Oncology Local Market Influencer Analytics

Shiraz Hasan, Vice President, Precision Xtract
As healthcare providers and payers continue to broaden their influence on HCPs in oncology, it becomes necessary to integrate intelligence about these players into the HCP behavior framework. We combine knowledge of payer and provider anatomy – including organization of subsidiaries, channels of influence, and industry trends to create the Provider Master and Payer Spine, thus tying data observations to decision makers. Using this foundation we have built several models to understand brand utilization: single variate regressions illustrating correlation and spillover, multivariate regressions to explaining how payers and providers interact to push demand in a local market, and 3x3 influence charts tracking payer vs provider impact. We observed the impact of high-control health systems has spillover impact on the total health care delivery within the local markets in which they operate. This is critical to understand, as the drivers and levers of these systems, and the engagement model with these stakeholders (ie, partnerships, promotions, and contracts among stakeholders) have an effect not only within that health system but also on the overall local market.

Proactive Brand Performance Analytics

Dave Mahagnoul, SF Insights & Analytics, GSK
Bob Middien, Commercial Insights and Operations, Shire Pharmaceuticals
Mani Sethi, ZS Associates
Karthik Sourirajan, ZS Associates

Companies spend over $1.5 billion on R&D of a single drug and spend on launches in the US can be up to $250 million- it is imperative to establish Proactive Brand Performance Analytics that can help brand success by rapidly diagnosing root cause of business issues, identifying hidden opportunities, and enabling implementable actions. Often, it takes too long, with a lot of time spent in data collection and analysis. There is a critical need for an effective, agile, comprehensive and decision driven approach that identifies drivers of performance / opportunity areas, enables prospective (what-if) analysis, and facilitates collaborative decision making across functions.

In this presentation, we will showcase the journey of two companies in redefining their launch and growth monitoring capabilities to make agile, cross-functional decisions and the business impact they achieved. GSK started its journey with performance drivers analytics for in-line products to drive timely interventions. Encouraged by the impact generated, GSK applied the concepts for a launch brand. Shire started with the vision of building the best in class launch analytics to enable agile cross-functional decisions at launch. Encouraged by the impact generated for one brand, Shire has scaled the proactive launch approach as the “standard” capability for all follow-on launches as well as inline brands.