• Hypertrophic Cardiomyopathy (HCM)
  • Vaccines: 2023 Year in Review
  • Eyecare
  • Urothelial Carcinoma
  • Women's Health
  • Hemophilia
  • Heart Failure
  • Vaccines
  • Neonatal Care
  • Type II Inflammation
  • Substance Use Disorder
  • Gene Therapy
  • Lung Cancer
  • Spinal Muscular Atrophy
  • HIV
  • Post-Acute Care
  • Liver Disease
  • Biologics
  • Asthma
  • Atrial Fibrillation
  • RSV
  • COVID-19
  • Cardiovascular Diseases
  • Prescription Digital Therapeutics
  • Reproductive Health
  • The Improving Patient Access Podcast
  • Blood Cancer
  • Ulcerative Colitis
  • Respiratory Conditions
  • Multiple Sclerosis
  • Digital Health
  • Population Health
  • Sleep Disorders
  • Biosimilars
  • Plaque Psoriasis
  • Leukemia and Lymphoma
  • Oncology
  • Pediatrics
  • Urology
  • Obstetrics-Gynecology & Women's Health
  • Opioids
  • Solid Tumors
  • Autoimmune Diseases
  • Dermatology
  • Diabetes
  • Mental Health

Predictive modeling, stratification improve disease management risk profiling


As disease management programs have grown in size and scope, the importance of justifying their expense by demonstrating financial savings has become critical.

On a conceptual basis, few managed care executives argue the value of disease management programs. From a practical perspective, however, most have a difficult time correlating dollars and cents to that value. They increasingly ask their leadership teams whether or not a specific program delivers return on investment (ROI). All too often, results are unclear and cannot be quantified adequately.

As disease management programs have grown in size and scope, the importance of justifying their expense by demonstrating financial savings has become critical. It is no longer sufficient to defend a program based on an illustrated ROI. Insurers seek information about which members are being identified and how, what interventions can be applied to them most effectively, and which mechanism leads to genuine behavior change and savings.

These data requirements will only increase in the future, which means that insurers and program architects alike must become smarter about economic optimization of disease management efforts. This will require intensive risk profiling, predictive modeling and stratification on the part of all who are involved in program design and execution.

Consider the insurer that administers a typical high-cost, high-risk disease management program. Members are deemed to be high risk because their care is high cost and because they meet certain clinical triggers. Managing these members at the disease stage during which the insurer intervenes is largely palliative, because the major harm has been done long before. In addition, insurer identification methods typically result in a relatively large number of members being referred for management by costly clinical resources.

A more efficient program would use risk profiling, prediction and economic modeling to identify high-risk members earlier, prompting intervention with those whose behavior can be changed. Including a preventive/wellness component would also contribute towards earlier identification and encourage self-management. The issue, then, is how to incorporate risk profiling and predictive modeling in designing a successful program.

One crucial facet of any model is opportunity assessment. A population being stratified on any dimension can be rank-ordered. A graphic depiction of this approach will produce a downward sloping line, from the northwest quadrant to the southeast, illustrating a high frequency of the given event for a small percentage of the population and a progressively diminishing frequency as a larger percentage of lower-risk individuals is included.

Whatever the event that is being predicted, each member of the population will have an expected event frequency. If the event is hospitalizations, for example, members would be ranked in order of the likelihood that they would experience a hospitalization, according to the specific predictive model used. Managed healthcare executives may test any model by applying it to historic data, and evaluating the actual vs. expected identification of members identified as high or low risk.

Information about the predicted rank-ordering of the population may be combined with information about the efficiency and effectiveness of the intervention program. For example, an intervention that aims to close the gaps in care or enhance compliance in a diabetic population may be measured on many different variables – including effectiveness of outreach and enrollment, numbers of indentified gaps closed, and percentage of the population that becomes compliant. The result of the intervention will be a reduction in the event frequency along the curve. This may be equal at all points (resulting in a shift in the curve downward), or more effective at either the high-risk end or the low-risk end (resulting in a rotation of the curve). If the program is expected to produce limited initial results in terms of risk-reduction, the analysis should be applied to future years, when the effect of risk-reduction becomes apparent.

Related Content
© 2024 MJH Life Sciences

All rights reserved.