Predictive analytics has become a major force in insurance; particularly in health insurance. Health actuaries have traditionally used aggregate data for pricing and reserving. In the last 10–15 years, considerable developments have taken place in key areas of modeling: “case-finding” (identifying high-risk and cost patients for intervention), patient matching (for evaluating programs, drugs, devices and provider quality) and risk adjustment (financial transfers based on different populations’ relative risks). Health actuaries have focused on risk adjustment and application of public-domain models (HCC; CDPS) and commercial models. Other predictive applications have been performed by professionals such as statisticians and those with public health or clinical training.
Inexpensive computing power and increasingly powerful analytical tools become more widely available each year. One drawback for those interested in modeling applications has been accessing relevant datasets. For health actuaries, predictive analytics provides opportunities to ground traditional actuarial work in rigorous statistical methodology while bringing value to their organizations by combining techniques with traditional actuarial skills.
The Health Predictive Analytics Seminar, an interactive, hands-on event will provide attendees with the opportunity to apply practical knowledge of statistical techniques that are broadly relevant in actuarial work. Core techniques, like regression analysis, generalized linear models, decision trees, time series analysis and machine learning will be discussed. The curriculum will cover a refresher in basic statistics, hypothesis testing, tests of significance and interpretation of results. Attendees will learn an important component of health analytics work: the ability to understand and interpret claims data and construct appropriate algorithms. All techniques will be illustrated with health insurance applications and case studies.
While the seminar will cover statistical theory, it will focus on practical modeling. The medium of the seminar will be the open-source R statistical computing environment. Case studies will apply R code to different healthcare datasets (provided to attendees). For attendees not familiar with R, references will be provided for review prior to the course.
To further aid attendee understanding and for those who want to practice techniques after the seminar, instructors will be available for discussion during “office hours” via email, Skype or in person at University of Illinois at Chicago for one month after the seminar.
• To create an optimum learning environment, this seminar will be limited to 35 registrants.
• Attendees will be required to bring their own laptop computers to this seminar. Ideally, attendees will have (or be able to arrange) software installation privileges for the seminar’s purposes.
After this seminar, attendees will be able to:
• Apply and interpret standard statistical measures such as tests of significance;
• Understand the construction of typical algorithms for health data;
• Perform basic data manipulations that fit a variety of standard models in the R statistical computing environment;
• Graphically explore data to motivate various modeling choices and graphically criticize models and motivate model improvements;
• Interpret and critically examine standard model output;
• Test the performance of models on holdout data;
• Apply models in typical health insurance applications such as risk adjustment and time series; and
• Translate a business problem into the design of a data analysis strategy
This seminar is ideal for experienced actuaries who are unfamiliar, but interested in learning health data and health predictive analytics looking to gain an understanding by learning with a hands-on approach. It is designed for those with an intermediate knowledge of statistics and modeling (for example, those who attended the Society of Actuaries’ (SOA’s) Advanced Business Analytics Seminar, or who have similar knowledge and experience). The seminar combines a hands-on approach for those wishing to apply R models, with the opportunity for actuaries in leadership roles to identify appropriate models for solving business problems and to assess the suitability of models developed by analysts. Health, Long Term Care Insurance and Modeling Section members will also benefit from attending.
Level of Difficulty
The seminar is designed for attendees with a moderate level of experience in statistics and modeling (for example, current familiarity with ASA-level statistical concepts such as regression, analysis of variance and tests of significance). Knowledge of R is recommended. Attendees will have the opportunity to acquire basic R skills prior to the seminar.
Ian Duncan, FSA, CSPA, FCA, FCIA, FIA, MAAA
Adjunct Professor in the Department of Statistics & Applied Probability
University of California, Santa Barbara
Ian Duncan, FSA, CSPA, FCA, FCIA, FIA, MAAA, founded Solucia Inc. in 1998, a provider of health predictive modeling and outcomes evaluation for health care. After selling the company in 2010, he was head of clinical outcomes and analytics at Walgreens. Since 2011, Duncan has been an adjunct professor at the University of California, Santa Barbara, and president of Santa Barbara Actuaries Inc. He has published widely in Health Care Predictive Modeling and Outcomes.
Sreenivas (Si) Konda, Ph.D.
Assistant Professor in the Department of Epidemiology & Biostatistics
University of Illinois, Chicago (UIC)
Sreenivas (Si) Konda, Ph.D, was a visiting professor in the department of statistics & applied probability at University of California, Santa Barbara, from 2013–2015. He was previously a faculty member in the department of statistics and actuarial science at University of Waterloo from 2012–2013. Konda has taught several classes in R, SAS, regression models, multivariate analysis, health analytics and time series. He is currently collaborating with Duncan in a study of the application of Markov models to disease stage progression and survival time models. He also helps the cardiology division at UIC with statistics and bio-informatics consulting.
Nhan Huynh, MS
University of California, Santa Barbara
Nhan Huynh, MS, is a Ph.D. candidate in statistics at the University of California, Santa Barbara. She has worked with Duncan on many modeling assignments, and was a first prize winner for her presentation of a predictive modeling paper at the Actuarial Research Conference in 2016.