2020 Poster Presentations

Analog Intelligence: Growing Both Players in a Duopoly Market

Haejin Chung, Vice President (Former), Commercial Operations, Seqirus and Debbie Glick, CEO, Cello Health Advantage Inc.
History often repeats itself. Strategic commercial decisions in our industry are well served by studying success stories in similar environments before investing in costly primary research or piloting marketing or sales programs. Duopoly markets exist in many therapeutic categories. A leading vaccine manufacture was seeking an understanding of how both players in a market can grow in an environment similar to one in which they compete. To that end, a disciplined process was undertaken to identify, select, and study two analogs in duopoly markets. The duopoly players’ marketing, sales, and pricing strategies; investment choices; and tactical and distribution support were developed into case studies. Learnings from those case studies were extracted and applied to guide planning for the coming years.

Applying the Concept of Marginal Engagement and Annoyance Limit to Plan Digital Touchpoints

ZS

Building a Modern and AI-ready Data Foundation

Antares Pharma

Combining Multiple Supervised Classification Learning Problems to Drive Patient Adherence with Patient Hub Data

Veeva

Decentralized Federated Learning for Electronic Health Records

Yunlong Wang, Associate Director, Advanced Analytics, IQVIA
Access to large, high dimensional datasets to provide sufficient training data for deep learning models can be a challenge, especially for single entities who only have access to their own patients. Federated learning (FL) –a machine learning technique where a model is trained on data samples across a network of individual institution electronic health records (EHR) with no data sharing between institutions– can solve this problem. This novel method was tested in the setting of predicting progression to late-stage disease among early-stage Alzheimer’s patients. The resulting shared predictive model converged to the optimum solution with 90% fewer communication rounds than the traditional model. The ability of the FL algorithm to track the full gradient resulted in a ten-fold increase in accuracy over the classical approach of centralized processing with the same number of iterations. Advantages are three-fold: 1) data privacy is preserved; 2) computational burden is reduced; and 3) communication efficiency is increased.

Enhancing Model Interpretability and Accuracy for Disease Progression Prediction via Phenotype-Based Patient Similarity Learning

Yue Wang, Associate Data Science & Advanced Analytics Manager, Advanced Analytics, IQVIA
Recent years have witnessed a growth of longitudinal claim data, which has motivated the use of machine learning, especially deep learning in disease prediction and progression tasks. One challenge is to enhance the model interpretability to the general population in the healthcare industry. Models have been proposed to extract temporal patterns from longitudinal electronic health records (EHR) for clinical predictive models. However, the common relations among patients (e.g., receiving the same medical treatments) were rarely considered. In this paper, we propose to learn patient similarity features as phenotypes from the aggregated patient-medical service matrix using non-negative matrix factorization. On real-world medical claim data, we show that the learned phenotypes are coherent within each group, and also explanatory and indicative of targeted diseases. It helps to understand how the potential market patients look like, and consequently sending an advanced alert to a group of providers to increase the market share. We conducted experiments to predict the diagnoses for Chronic Lymphocytic Leukemia (CLL) patients. Results show that the phenotype-based similarity features can improve prediction over multiple baselines, including logistic regression, random forest, convolutional neural network, and more.

Empowering Clinical Decisions Using Machine Learning Prediction of Prognostic Biomarkers for Patient Disease Progression

Ariel Han, Vikram Singh, Rick Rosenthal, and Ewa Kleczyk, Symphony
This poster presents the use of a machine learning model to predict lab results, and its value in clinical support and long-term disease management. We examined the application of a biomarker called PSA (Prostate Specific Antigen), and its clinical significance in predicting time to metastasis for prostate cancer patients. We developed a mechanism to further strengthen the prediction by adding multiple dimensions from other healthcare data sources, and to replicate PSA’s predictive power in an environment where PSA is not available to further evaluate prostate cancer progression. This was achieved by analyzing patients’ history with the use of classification algorithms.

Identifying Undiagnosed and Untreated Patients for PAH: Bayesian Reasoning vs. Probabilistic Approach

Tanya Kulkarni, Business Analytics Manager, Actelion; Jack Lin, Director Business Analytics, Actelion; Rung Lin, PhD, Principal Consultant, Bayser; and Jean Patrick Tsang, PhD & MBA (INSEAD), President of Bayser
To identify undiagnosed and untreated patients, we can use a simple approach that consists of going after physicians that have a large number of patients. Assuming the prevalence of the disease of interest is uniformly distributed, the larger the number of patients, the higher the odds the physician will have such patients. Alternatively, we can deploy a more sophisticated approach such as Bayesian Reasoning to leverage specific pieces of information that are available on each individual patient. This includes diagnoses, drugs, lab tests, procedures, hospital admissions and the like. In this study, we compare the 2 approaches and shed light on the learnings we gleaned.

Managing Data As a Strategic Asset

Novartis

Mapping Referral Networks for Exploration of Commercial Network Interventions

Teis Kristensen, Senior Associate, Axtria Inc.
Sales and marketing experts within the pharmaceutical industry use commercial interventions based on the mapping of referral networks to identify influential physicians. The targeting of influential physicians is seen as a potential source of rapid growth for market entrants and a way to ensure a stable position in saturated markets. The underlaying assumption is that some physicians hold a disproportional influence on others and that mechanisms of social contagion helps spread the adaptation of products. This work describes how proxies of physician referral networks can be mapped from prescription claims data, reviewing the type of network configurations and metrics found to be associated with the diffusion of innovations, and providing a taxonomy of the types of targeting interventions that have been found to boost rapid adaptation of new products and continued growth in mature markets.

Optimizing Social Determinants for Improved Patient Outcomes

Kathy Johnson, Senior Director of Sales - Life Sciences, LexisNexis and Rich Morino, Director, Strategic Solutions, LexisNexis
This presentation will discuss why Social Determinants of Health matter, illustrate the impact of SDOH data on health outcomes, and explore the data sources often tapped for these insights. It will also address the SDOH-driven patient insights can be delivered across settings. The speaker will also use real-world examples to show the impact of using SDOH data within a diabetic population, highlighting the utility of this data to identify barriers to improved outcomes, to predict populations and find ways for partnerships with other healthcare stakeholders to improve health outcomes.

Oncology Launch Analytics and Predictive Triggers Using Patient Data

Exelixis and ZS