Lessons from the Field

Denver Health is an integrated safety net health care system in Denver, CO. Denver Health was awarded a significant grant from The Center for Medicare and Medicaid Innovation to implement a “population health” approach to the delivery of primary care. This approach focused on understanding the needs of the total population, organizing the population into stratified tiers, and designing services to meet the needs of the people within the tiers. The following is a short description of how Denver Health developed and refined the tiers with specific focus on tiers 3 and 4, which comprise the high-need, high-cost segments. For a full description of Denver Health’s population health and patient stratification methods, read, Augmenting Predictive Modeling Tools with Clinical Insights for Care Coordination Program Design and Implementation.


“We tested our methods using a proprietary 3M CRG model or clinical risk groups. This predictive modeling tool used diagnoses, procedures, pharmacy and demographic data to assign every Denver Health patient in the past 18 months to a CRG.

The clinical risk groups are one through nine. Nine being an absolutely catastrophic illness. Eight being a malignancy, and then kind of going up to a CRG of one, which is essentially the lowest risk. Each CRG is assigned a tier, split between adults and pediatrics (19 years or older is an adult unless on pediatric registry).

We did a number of different iterations figuring out who should make it into that top utilizing tier, tier four, based upon clinical risk groups, and then who should fit into either tier two or tier three, and then the criteria by which we would automatically graduate or move someone up. We learned that even a tier one patient, with seemingly low level medical problems, but all of a sudden, had three hospitalizations, we needed the ability to move that patient up to tier four. We also made a sort of an institutional decision that for an adult high-risk patient, we were going to use the three hospitalizations as our criterion. We originally set the criteria of  three hospitalizations, or two hospitalizations and a serious mental health disorder in a rolling 365-day lookback, in addition to a severe mental health diagnosis. This latter criterion did not test out well and honestly, those are patients that we would often try and move into the Mental Health Centers of Denver high-risk program, for more of a behavioral health intervention with medical adjunct versus what we do which is really medical with a behavioral health kind of adjunct. In addition to the tiering we include a rather exhaustive clinical screening by our nurse program manager. She might spend about 15 or 20 minutes doing chart review, sometimes more, looking at the types of care that patients have had, looking for signs of end-stage organ disease, and things like that. We just recently went on to EPIC, but before EPIC, we had a cobbled together kind of health record that involved looking in three to four different pieces of a chart to get a full picture of a patient before we would even consider trying to recruit them in. So a lot of time and effort has gone into that screening process, and I think that has served as an excellent predictor, if you will, of who might work within our model.”Jeremy Long, MD, MPH, Division of General Internal Medicine with the Denver Health and Hospitals, Department of Medicine

bc-childrensBelow is an example from British Columbia (BC) Children’s Hospital and Sunny Hill Health Centre using iterative learning to design a process to identify individuals who are a good fit for their enhanced care program.

British Columbia (BC) Children’s Hospital and Sunny Hill Health Centre are agencies within the Provincial Health Services Authority of British Columbia, CA, specializing in providing tertiary and quaternary care for the children of BC and  providing a pediatric subspecialty focus in medicine, surgery, mental health, rehabilitation and child development. These three agencies are the only agencies in BC to provide this level of care and expertise. Three service areas are involved in this initiative: Neurology, Developmental Pediatrics and Mental Health. The team from BC went through the following process to develop a method to identify the children in their chosen population segment:

Population segment with complex needs and high costs: Children accessing or requiring services from 3 specific programs/agencies within the province of BC:

  • BC Children’s Hospital Neurology,
  • BC Children’s Hospital Child and Youth Mental Health
  • Sunny Hill Health Centre for Children

BC Children’s Hospital and Sunny Hill Health Centre took a step-wise approach to understand how to identify individual patients for enhanced care:

  1. Ran a list of kids who needed all 3 services.
  2. Performed a group case review with doctors, which produced good insights on identifying individual kids but was too time consuming to scale-up.
  3. Developed and tested an algorithm, building on what was learned in steps 1 and 2.
  4. Refined further.
  5. Performed an in-depth analysis of 37 of the 120 children identified by the algorithm to learned how to further refine the algorithm.