1 / 48

Adjusted Clinical Groups (ACGs): Concept and Method

Adjusted Clinical Groups (ACGs): Concept and Method. Barbara Starfield, MD Johns Hopkins University’s European ACG Conference Karlskrona, Sweden September 18, 2007.

andrew
Télécharger la présentation

Adjusted Clinical Groups (ACGs): Concept and Method

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Adjusted Clinical Groups (ACGs): Concept and Method Barbara Starfield, MD Johns Hopkins University’s European ACG Conference Karlskrona, Sweden September 18, 2007

  2. People and populations differ in their overall vulnerability and resistance to threats to health. Some have more than their share of illness, and some have less. Morbidity mix (sometimes called case-mix) describes this clustering of ill health in patients and populations. Starfield 03/06 CM 3372

  3. Clustering of morbidity is a result of a complex pattern of influences on health, extending far beyond biological vulnerability. Starfield 04/02 02-082 Starfield 04/02 CM 2108

  4. Co-morbidity is the concurrent existence of one or more unrelated conditions in an individual with any given condition. Multi-morbidity is the co-occurrence of biologically unrelated illnesses. For convenience and by common terminology, we use co-morbidity to represent both co- and multi-morbidity. Starfield 03/06 CM 3375

  5. Total morbidity is not the same as the sum of different diseases, because diseases cluster and are inter-related in various ways. A more accurate way of characterizing morbidity is to characterize the pattern of diseases in people and populations. Starfield 03/06 CM 3371

  6. Co-morbidity • is more common in socially deprived populations • is more common in children, as compared with its expected frequency based on frequency of diagnoses • has great impact on use of resources Starfield 01/07 CM 3569

  7. Morbidity Burden in Disadvantaged versus HMO Patients Source: HMO data from Forrest et al, BMJ 2002; 325:370-1. Starfield 12/04 CM 3090

  8. Ratio O/E Number of diseases Ratios of Observed and Expected (Co-)Occurrences of Diseases, Overall and for Different Ages Source: van den Akker et al, J Clin Epidemiol 1998; 51:367-75. Starfield 02/01 01-015 Starfield 02/01 CM 1817

  9. Co-morbidity, Inpatient Hospitalization, Avoidable Events, and Costs* Source: Wolff et al, Arch Intern Med 2002; 162:2269-76. Starfield 10/03 03-252 Starfield 10/03 CM 2632 *ages 65+, chronic conditions only

  10. Importance of Co-morbidity • Disease case management • Guideline relevance • Costs and complications • Orientation of health systems • Primary care vs specialty care • Appropriate use of specialty services • Quality of care: processes vs. outcomes Starfield 10/03 03-350 Starfield 10/03 CM 2730

  11. Conceptual Basis for ACGs • Individual diagnoses are less important in the care of patients and populations than are patterns and overall burdens of morbidity • Models of care need to be based on overall morbidity burdens rather than on specific diagnoses • Assessing the appropriateness of care needs to be based on patterns of morbidity rather than on specific diagnoses Starfield 01/07 AC 3544

  12. Clinical Observations Underpin the ACG System • Morbidity is NOT randomly distributed across individuals. 1) Morbidity “clusters”. 2) Diagnoses co-occur. • The “illness burden” of providers’ practices is NOT randomly distributed. 1) Some providers care for “sicker” patients. 2) Sick patients choose certain providers preferentially. Starfield 04/97 AC 1139 Starfield 1997 97-047

  13. Overview to Methodologic Basis for ACGs • A single ACG is assigned to a person based on his/her age, gender, and constellation of diagnosis codes. • ACGs are not a prior use measure per se; they are based only on diagnosis, not charges or procedures. Starfield 05/01 AC 1881 Starfield 2001 01-043

  14. Going from Diagnosis Codes to ACGs in Version 4.0 Diagnosis Codes (15-20,000) Adjusted Diagnosis Groups (32) Collapsed Adjusted Diagnosis Groups (12) Major Adjusted Categories (26) age, gender Adjusted Clinical Groups (83-100) Starfield 10/99 AC 1561 Starfield 1999 99-095

  15. Going from Diagnosis Codes to ADGs Criteria Used to Assign a Diagnosis Code to an ADG Expected resource intensity criteria • Expected need and cost of diagnostic or therapeutic procedures associated with the condition • Likelihood that return visits, specialty care, or hospitalization will be needed Clinical criteria • Persistence/recurrence over time • Clinical category: e.g., separate groups for injuries, signs/symptoms, pregnancy, malignancy Starfield 04/97 AC 1142 Starfield 1997 97-096

  16. Examples of the 32 ADGs Starfield 04/97 AC 1143 Starfield 1997 97-097

  17. Ambulatory Diagnostic Groups (ADGs) Time limited (4) Allergies Asthma Likely to recur (3) Malignancy Chronic medical (2) Chronic specialty (6) Dermatologic Injuries (2) Psychosocial/psychophysiologic (3) Signs/symptoms (3) Discretionary See and reassure Preventive/administrative Pregnancy Dental Total number of ADGs = 32 Starfield 12/97 AC 1260 Starfield 1997 97-018

  18. Going from ADGs to ACGs • During a single year, a patient’s diagnoses may fall into as many as 32 distinct ADGs. The potential permutations are vast. For practicality, a case-mix system must have a manageable number of mutually-exclusive categories • Clinically similar ADGs are combined into CADGs (collapsed ADGs). • Individual CADGs and the most common combinations are designated as MACs (Major Ambulatory Categories) with one additional MAC for “all other combinations” • ACGs are formed from the MACs, based upon relative contributions to resources use • Some ACGs are subgroups of a MAC based on • age and/or sex • total number of ADGs • total number of major ADGs Starfield 04/97 AC 1145 Starfield 1997 97-093

  19. To MAC 26tree MAC-26 Missing Age Age < 1 ACG 9900 Entire Population Age >= 1 Split into MACs,Based on CADGs MAC-1 MAC-3 MAC-5ACG 0800 MAC-7ACG 1000 MAC-9ACG 1200 MAC-11ACG 1600 MAC-13ACG 1800 MAC-15ACG 2300 MAC-17 MAC-19 MAC-21ACG 3500 MAC-23ACG 3700 MAC-25 MAC-2ACG 0400 MAC-4ACG 0700 MAC-6ACG 0900 MAC-8ACG 1100 MAC-10 MAC-12 MAC-14 MAC-16ACG 2400 MAC-18ACG 2800 MAC-20ACG 3400 MAC-22ACG 3600 MAC-24 To MAC 24tree To MAC 12tree Age ADG05 ? ADG25 ? Age ADG25? Age 1 or 2 input files? 1 Yes Yes No Yes No 1 2 1 1 ACG 0100 ACG 0600 ACG 1900 ACG 2900 ACG 5100 ACG 2500 ADG24? ADG24? Claims info? ACG 1300 2-5 No 2-5 2-5 ACG 0200 ACG 0500 ACG 2000 ACG 3000 Yes Yes Yes 6 + ADG05 ? 6 + ACG 1500 ACG 5110 ACG 2700 6 + ACG 3100 ACG 0300 No No No ACG 5200 ADG05 ? ACG 1400 Key ACG 2600 12 + Yes No MAC Major Ambulatory Category ADG Ambulatory Diagnostic Group CADG Collapsed ADGACG Ambulatory Care Group ACG 2200 ACG 2100 Yes No ACG 3300 ACG 3200 Decision Tree for ACGs Starfield 04/97 AC 1091 Starfield 1998 98-007 Source: JHU ACG Case Mix Adjustment System, V. 4.0, 1997.

  20. AC 1095 Starfield 04/97 AC 1095

  21. Distinguishing Characteristics of Adjusted Clinical Groups (ACGs) • Based on diagnoses, not procedures or prior utilization • Captures longitudinal dimension of patient health care • Requires routine insurance claims/encounter data only • Each ACG includes individuals with: 1) a similar pattern of morbidity 2) similar expected resource use Starfield 10/99 AC 1559 Starfield 1999 99-076

  22. There are about 100 types of co-morbidity groups (ACGs). For convenience, we have divided them into three groups (RUBs) according to the amount of resources people in them use in a year. Starfield 04/01 01-045 Starfield 04/01 AC 1852

  23. Applications of Morbidity-Mix Adjustment • Physician/group oriented • Characterizing and explaining variability in resource use • Understanding the use of and referrals to specialty care • Controlling for co-morbidity • Capitation payments • Refining payment for performance • Patient/population oriented • Identifying need for tailored management in population subgroups • Surveillance for changes in morbidity patterns • Targeting disparities reduction Starfield 03/06 CM 3373

  24. Rationale for ACGs: Co-morbidity (multi-morbidity) and disease variability are major challenges for health services, especially primary care. Starfield 01/07 CM 3543

  25. AC 1717 *Adults except * includes children **All with >50 people except 4940 Starfield 09/00 00-021 Starfield 09/00 AC 1717

  26. Expected Resource Use (Relative to Adult Population Average) by Level of Co-Morbidity, British Columbia, 1997-98 Thus, it is co-morbidity, rather than presence or impact of chronic conditions, that generates resource use. Source: Broemeling et al. Chronic Conditions and Co-morbidity among Residents of British Columbia. Vancouver, BC: University of British Columbia, 2005. Starfield 10/06 CM 3461

  27. 9 8 7 6 5 Mean Number of Visits 4 3 2 * * 0.81 0.79 0.83 0.52 1 0.40 0.33 0 Low-medium High Very high Morbidity Group Primary Care Physician Specialist Average Number of Visits Per Year to Primary Care and Specialists by Morbidity Burden, All Conditions, Managed Care Organizations, 1996 *p<.05 Starfield 07/03 03-131 Starfield 07/03 CM 2514 Based on data in Starfield et al, Ann Fam Med 2003; 1:8-14.

  28. 9 8 7 6 5 Mean Number of Visits 4 3 2 * * 1.12 0.92 0.86 0.91 0.85 0.75 1 0 Low-medium High Very high Morbidity Group Primary Care Physician Specialist Average Number of Visits Per Year to Primary Care and Specialists by Morbidity Burden,All Conditions, Medicare *p<.055 Starfield 07/03 03-133 Starfield 07/03 CM 2516 Source: Starfield et al, Ann Fam Med 2005; 3:215-22.

  29. * * * Average Number of Visits Per Year to Primary Care and Specialists by Morbidity Burden, Co-morbid Conditions, Managed Care Organizations, 1996 *p<.0001 Starfield 07/03 03-132 Starfield 07/03 CM 2515 Based on data in Starfield et al, Ann Fam Med 2003; 1:8-14.

  30. 8.95 9 * 8 6.57 7 6 5 4.32 * Mean Number of Visits 3.9 4 3 2.12 * 1.8 2 1 0 Low-medium High Very high Morbidity Group Primary Care Physician Specialist Average Number of Visits Per Year to Primary Care and Specialists by Morbidity Burden,Co-morbid Conditions, Medicare *p<.0001 Starfield 07/03 03-134 Starfield 07/03 CM 2517 Source: Starfield et al, Ann Fam Med 2005; 3:215-22.

  31. For those over 65, specialists play a major role except in people with low overall burdens of morbidity (at least in the US). Starfield 06/03 03-119 Starfield 06/03 CM 2503

  32. Morbidity Burden and Number of Primary Care Physicians Seen in a Year, Non-elderly Individuals All numbers are derived from the following population: People who were continuously enrolled in five health plans in 2001 and 2002 People with no hospitalizations in 2001 People with at least one identifiable generalist visit Starfield 09/07 CM 3850

  33. Morbidity Burden and Number of Specialists Seen in a Year, Non-elderly Individuals All numbers are derived from the following population: People who were continuously enrolled in five health plans in 2001 and 2002 People with no hospitalizations in 2001 People with at least one identifiable generalist visit Starfield 09/07 CM 3852

  34. Morbidity Burden and Number of Primary Care Physicians Seen in a Year, Elderly Individuals All numbers are derived from the following population: People who were continuously enrolled in five health plans in 2001 and 2002 People with no hospitalizations in 2001 People with at least one identifiable generalist visit Starfield 09/07 CM 3851

  35. Morbidity Burden and Number of Specialists Seen in a Year, Elderly Individuals All numbers are derived from the following population: People who were continuously enrolled in five health plans in 2001 and 2002 People with no hospitalizations in 2001 People with at least one identifiable generalist visit Starfield 09/07 CM 3853

  36. With high morbidity burden, the number of different physicians seen rises to a greater extent than is the case for number of visits, for both primary care and specialist care. Therefore, coordination of care is a major challenge for those with high morbidity burden. Starfield 09/07 CM 3855

  37. Controlling for Morbidity Burden • The more different specialists seen, the higher the total costs, medical costs, diagnostic tests and interventions, and types of medication. • The more DIFFERENT generalists seen, the higher the total costs, medical costs, diagnostic tests and interventions, and, to a lesser degree, number of types of medications. • The more generalists seen (LESS CONTINUITY), the more the number of DIFFERENT specialists seen. The effect is independent of the number of generalist visits. Starfield 09/07 CM 3854

  38. There is increasing concern that health systems and clinical practices themselves contribute to poor outcomes and excessive costs. Relevant considerations are: • Balance between primary care and specialty use • Quality concerns: errors of omission • Quality concerns: errors of commission (including overuse and adverse effects) • Disparities in health and in provided services Starfield 03/05 AC 3130

  39. What Could Population-Oriented Morbidity Assessment Accomplish? • Identifying sources of variation in health status/resource use that are NOT predicted by individual patient characteristics • Focus attention on the likelihood of systematic differences in predictability across population subgroups • Focus attention on health systems and provider characteristics that are associated with patterns of clinical care/resource use • Focus attention on the limitations of guidelines, especially in the presence of co- and multi-morbidity • Focus attention on adverse effects Starfield 03/05 AC 3135

  40. Expanding the Basic ACG System Starfield 01/07 AC 3556

  41. Risk Factors in the Johns Hopkins Diagnosis-Based Predictive Model (Dx-PM) Age Morbidity Burden (ACGs) Gender Risk Score Complicated Pregnancy Marker Pharmacy Use Marker (optional) Selected Medical Conditions Hospital Dominant Conditions Starfield 01/07 AC 3560

  42. Expanded Diagnostic Categories (EDCs) EDCs are groups of clinically similar diagnoses.  There are 264 groups that can be combined into 27 categories of clinical conditions, which can be further combined into 5 main types: Administrative; Medical; Surgical; Obstetric/Gynecological; Psychosocial. Starfield 01/07 AC 3557

  43. Percent Distribution by Degree of Co-morbidity for Selected Disease Groups, Non-elderly Population Starfield 09/03 03-213 Starfield 12/04 CM 3096 *About 20% have no co-morbidity.

  44. EDC Diagnosed Prevalence by Type of Population Served Facilities serving socially deprived populations have much higher diagnosed prevalence for 26 of the 27 EDCs. Facilities paid primarily by capitation have considerably (30%-400%) higher diagnosed prevalence of 22 of the 27 EDCs, as compared with facilities reimbursed by fee-for-service payments.   This accounts for the differences in co-morbidity, and therefore ACG distributions, in different populations and types of facilities. Source: The Johns Hopkins ACG System: Reference Manual, Version 8.0. Baltimore, MD: Johns Hopkins Bloomberg School of Public Health, 2006. Starfield 01/07 AC 3570

  45. Definition of Predictive Modeling (PM) Predictive modeling is a process that applies available data to identify persons who have high medical need and are “at risk” for above average future medical service utilization. Starfield 03/05 AC 3126 Source: Weiner J, 2004.

  46. Predictive modeling differs from forecasting based on prior use by employing characteristics of people’s medical diagnoses and thus their projected need for future services. Good PM (like the ACG system) recognizes that co-morbidity plays a major role in need for clinical interventions. Thus it is, essentially, a system for case-mix adjustment based on individual patients’ needs for better “case management”. Starfield 03/05 AC 3127

  47. Current Assumptions of Predictive Modeling/Case Management Use of information about patients is the key to achieving the aims of predictive modeling. That is, the best way to reduce costs/improve management is by identifying individuals, i.e., “cases” with special illness characteristics, NOT by changing systems and provider practices. Starfield 03/05 AC 3129

  48. Predictive Modeling and Case Management – New Approaches • Changes in health systems/provider performance are appropriate strategies to improve care. • Individual and group differences in “risk” can highlight systematic differences in health/ resource use by race, ethnicity, social class. • Identification of interventions that are harmful, e.g., adverse effects of medications THAT IS, PREDICTIVE MODELING IS A POTENTIALLY USEFUL STRATEGY TO ALTER CLINICAL MANAGEMENT/RESOURCE USE (CASE MANAGEMENT) ON A SYSTEMS LEVEL. Starfield 03/05 AC 3133

More Related