2020 Winter Symposium

Claims are a real godsend. They have since their inception at the beginning of this century transformed the very way we conduct analyses. More recently, they have been instrumental in pushing the envelope in Predictive Analytics and Machine Learning. The reason for this is that Claims provide the highest level of resolution possible: individual interactions between patient and the healthcare system.

Claims are far from perfect though. They have holes and bugs. By holes, we mean healthcare interactions that are missing in the data. By bugs, we mean the information conveyed in Claims that is plain wrong, which may come as a surprise to those new to Claims. In truth, we should not be complaining as Claims were not meant to be the analytical workhorse we have turned them into. They are after all simply invoices that Providers send to Payers to get reimbursed for services rendered to patients. In essence, we hijacked the Claims for analytical purposes.

Holes and bugs lead us to draw the wrong insights and make wrong decisions. It's even worse since we hold those insights and decisions to be the truth with a certainty that only comes from having diligently analyzed the data. Little surprise we are headed to catastrophe.

This talk is meant to guard against such an unfortunate outcome. It is organized in three parts. In the first part, we provide ample real-life examples of holes and bugs that wreak havoc in findings. These examples come from firsthand experience with working claims data over the past 15 years. We'll explain along the way that there are 2 types of holes, namely, longitudinal and pocket. Longitudinal holes are formed when some of the healthcare interactions of a patient are missing. These holes impact adherence, new to therapy starts, and line of therapy analyses. Pocket holes are formed when all the healthcare interactions of a patient go missing for a large number of patients. Pocket holes may pertain to an entire IDN, geography, setting (e.g., hospital), or Payer. They impact targeting and segmentation.

In the second part, we describe ways to identify holes and bugs. We also discuss when to use Open Claims and Closed Claims. In yet some other cases, we recommend staying away from Claims altogether especially when the data precludes us from getting a handle on what do not know we do not know. In the third and last part, we discuss a handful of techniques to plug holes and fix bugs. The first technique is filtering out and consists of developing and applying business rules to the raw data set to generate a more consistent data set for analyses. A second technique consists of using Machine Learning techniques such as Bayesian Reasoning, KNN (K Nearest Neighbor), and SVD (Singular Value Decomposition) to infer missing information from telltale markers that are present in the data. We'll also discuss other techniques such as CNN (Convolutional Neural Net) and HMM (Hidden Markov Model) to infer the presence of information that has never been present in the data. Case in point: Line of therapy and lab values.

Highlights of this session will include:

1. Predicting and finding underdiagnosed or misdiagnosed rare disease patients is one of key success factors in orphan drug marketing
2. Traditional method formulating this problem as a supervised learning task does not work well because of the extreme low prevalence rate and imbalance in the data
3. We developed a new semi-supervised machine learning model to improve the accuracy. The new method is able to leverage a vast majority of the unlabeled data and help to train a model with better precision.
4. In this study, we built a deep learning model with recurrent neural networks and generative adversarial networks, and in terms of prediction accuracy, the proposed method outperformed the traditional machine learning models such as random forest and extreme gradient boosting.

Data Analysts, Marketing Scientists, Data Scientists: Are they the same? Are they different? What is needed to be successful in different roles and at different stages of your career?

This panel session will discuss the current state of the union in our discipline and will offer perspectives and insights on the way in which roles have evolved. The panel will also provide snapshots of success stories and advice for how attendees can think about navigating their own careers. Small group discussions and insights sharing will also take place.

This session will provide additional insights and perspectives on themes discussed during the panel session on Day 1. We will also provide insight on best practices and lessons learned during recruiting and hiring. Finally, we will provide practical tools and some hands-on exercises to help empower you to shape your own development and begin to shape various career paths.

Being an expert in advanced analytics and data science might not be enough for you and your work to truly gain traction in your organization. In this session, we will provide tips and techniques to help you evolve from an Analyst to a Strategic Thought Partner, with real-life examples of how to increase your credibility and influencing skills.

We are all inherently visual communicators and stories are how people understand and relate to the world and they associate these stories with appropriate imagery. Yet presentations based on data are full of charts and tables and as we present we see people doing emails on their computer or looking at their phones. Our challenge is to find the story in the content and present it in such a way to keep our audience engaged and communicate insight so that action is possible. This means we may need to break some habits and gain some new ones. This session will share some tips and expose you to a few new habits you need to consider cultivating as presenters.