International Insights on Mortality, Population and the Public Interest Tuesday, October 3, 2017 Westin River North Hotel, Chicago IL Using Predictive Analytics to Better Understand Morbidity Merideth Randles, FSA, MAAA Principal and Consulting Actuary, Milliman 1
Definition Predictive analytics is a form of advanced analytics that uses both new and historical data to forecast future activity, behavior and trends. It involves applying statistical analysis techniques, analytical queries and automated machine learning algorithms to data sets to create predictive models that place a numerical value, or score, on the likelihood of a particular event happening. Source: http://searchbusinessanalytics.techtarget.com/definition/predictive-analytics 2
Associations What is advanced analytics? A continuum rather than absolute Statistical techniques Machine Learning Big Data Expanding IT/Data Capabilities Individual Level Data Analysis and Projection 3
Emerging Examples Insurer Applications More sophisticated claims projections Morbidity scores/risk scores Medical Research Prediction models may be built with fewer cases but more information Supports better prediction of the population that will benefit from a drug or therapy Clinical Care Identification of at risk patients and patient specific treatments Patients Genome analysis can identify potential health risks 4
New York Times The first gene therapy treatment in the United States was approved recently by the Food and Drug Administration, heralding a new era in medicine that is coming faster than most realize and that perhaps few can afford. The treatment, Kymriah, made by Novartis, is spectacularly effective against a rare form of leukemia, bringing remissions when all conventional options have failed With gene therapy, scientists seek to treat or prevent disease by modifying cellular DNA. Many such treatments are in the wings: There are 34 in the final stages of testing necessary for F.D.A. approval, and another 470 in initial clinical trials, according to the Alliance for Regenerative Medicine, an advocacy group. The therapies are aimed at extremely rare diseases with few patients; most are meant to cure with a single injection or procedure. Source: https://www.nytimes.com/2017/09/11/health/cost-gene-therapydrugs.html?utm_campaign=khn%3a%20daily%20health%20policy%20report&utm_source=hs_email&utm_medium=email&utm_content=5 6269810&_hsenc=p2ANqtz-8AGQD9V4_TEK-_f-7lONi07OM21N_2h4EYe44uuG5V0ZiIFeSMIzYrBIEelkOg4Aqv8lZVh-cOe3EjbDXnbU2wrNMLixuy5gEWuenBqmEsTVN1w8&_hsmi=56269810 5
Morbidity Risk Scoring Early Example Now Broadly used within Health Care OVERVIEW Built from extensive individual level claims and diagnoses data for intended population Utilizes regression modeling Produces an estimate for each specific individual based on their utilization and/or diagnoses history May be applied in conjunction with other rating factors 6
Risk Adjustment Example Chronic Illness and Disability Payment System (CDPS) a diagnostic classification system that Medicaid programs can use to make health-based capitated payments for TANF and disabled Medicaid beneficiaries Researched by University of California San Diego Claims from seven states were analyzed which provided payment weights that states can use when adjusting HMO payments. Source: http://cdps.ucsd.edu/ 7
Example CASE STUDY EXAMPLE: Going beyond conventional rating factors for health care claims projections Unless specifically cited, the following data within the case study are presented for purposes of illustration and should not be relied upon for any other purpose. 8
Historical Costs 9
Demographics Department of Veterans Affairs, Volume II, Medical Programs and Information Technology Programs, Congressional Submission FY 2017 Funding and FY 2018 Advance Appropriations, p. 183 10
Case Study Model Generalized Linear Regression Model Allows for the separation of effects from highlight correlated variables Additional variables can be easily introduced Inclusion/exclusion of certain variables allow for the specific trend observation Model run time is fairly short (hours) Conducted at a service specific level Inpatient Surgical Admission, Hearing/Speech Exams, etc. 11
Sample Variables Age/Gender Region/Market of Residence Disability & Eligibility Status Occupational Exposure/Hazard Enrollment Duration Calendar Year Birth Cohort Counts of Chronic Illness and Disability Payment System (CDPS)-based conditions Specialized Diagnosis-Based Indicators 12
Projection Enhancements The following examples highlight projection enhancements by service line: 1. Disability factor adjustments Reflect ongoing migration and dampening of factors 2. Birth year cohort adjustments 3. Diagnosis based indicators 4. Calendar Year Trends 13
Age/Gender Factors 14
1 Disability Level Factors Relative disability level impact movement is visible when calendar year is a variable 15
Disability Level Population CONCLUSION: Implement disability factor with migration adjustment over time Department of Veterans Affairs, Volume II, Medical Programs and Information Technology Programs, Congressional Submission FY 2017 Funding and FY 2018 Advance Appropriations, p. 184 16
2 Birth Year Cohorts Certain birth years relate to utilization irrespective of other rating factors (age, gender) CONCLUSION: Add testing and implementation of birth year cohort adj. 17
3 Special Conditions CDPS condition count Special condition categories A blindness diagnosis, as defined by a list of ICD 9 and ICD 10 diagnosis codes Spinal Cord Injury Conditions A cancer diagnosis, based on the CDPS definition; A central nervous system diagnosis, based on the CPDS definition and modified to include dementia (based on ICD 9 and ICD 10 diagnoses codes developed internally); or A psychiatric diagnosis, based on the CDPS definition. 18
4 Calendar Year Trends Measuring service specific trends CONCLUSION: Implement service specific trends 19
Historical Costs 20
Projected Costs 21
Demographics Department of Veterans Affairs, Volume II, Medical Programs and Information Technology Programs, Congressional Submission FY 2017 Funding and FY 2018 Advance Appropriations, p. 183 22
Future Where is predictive analytics going? Modeling becomes more sophisticated as data capture increases More and more individualized applications Privacy, regulatory, ethical, social issues What may be used to assess risk? How should this risk be passed on to the consumer? Consumer Data integration and use 23