Risk-Adjustment Methodologies and Applications in the VA Amy K. Rosen, Ph.D. Director, Risk Assessment and Patient Safety Center for Health Quality, Outcomes and Economic Research A VA HSR&D Center of Excellence (Bedford, MA) & Professor, Health Policy and Management Boston University School of Public Health
60
Embed
Risk-Adjustment Methodologies and Applications in the VA Amy K. Rosen, Ph.D. Director, Risk Assessment and Patient Safety Center for Health Quality, Outcomes.
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Risk-Adjustment Methodologies and
Applications in the VA
Amy K. Rosen, Ph.D.
Director, Risk Assessment and Patient Safety
Center for Health Quality, Outcomes and Economic Research
A VA HSR&D Center of Excellence (Bedford, MA)
&
Professor, Health Policy and Management
Boston University School of Public Health
Purpose of Talk
Introduce concept of risk adjustment Describe two well-known diagnosis-based
risk-adjustment tools: Diagnostic Cost Groups (DCGs) and Adjusted Clinical Groups (ACGs)
Discuss applications in the VA Development of psychiatric risk-adjustment
measure for the VA
Why is Risk Adjustment Necessary?
Health status of population can vary significantly
Goal is to provide equitable compensation and make appropriate comparisons
Allocations based on efficiency and quality, not selection
Risk Adjustment
The process by which the health status of a population is taken into account when
evaluating patterns or outcomes of care or setting capitation rates
Predictive validity Subgroup fit Administrative feasibility Incentives for efficiency Resistance to gaming
Diagnosis-Based Risk Adjustment
Increasing use of risk adjustment based on diagnosis codes from administrative data
Persisting concerns with reliability and validity of diagnosis codes Outpatient data - not much known about reliability Variability in coding practices across providers and facilities Upcoding and diagnostic creep Tentative coding
Risk-adjustment measures minimize some of these (e.g., excluding ill-defined codes)
What are diagnosis-based risk-adjustment measures?
Diagnosis-based measures use demographics/diagnostic information from claims/encounters to:
Classify patients into clinically homogeneous groupsbased on expected need for resource utilization Create clinical profile Identify clinical needs Evaluate clinical management programs
Predict relative resource use Predict expenditures Same year as diagnosis (Concurrent Models) Subsequent year to diagnoses (Prospective Models)
Data Requirements
Defined population of patients Claims/encounter data available for all
members of the population (12 months) Unique patient identifiers (i.e., social security
numbers) Age and gender ICD-9-CM diagnosis codes from face-to-face
Linear additive formulas (OLS regressions) combine predictions based on HCCs and age/sex cells subject to: Hierarchical restrictions Exclusions of CCs in prospective models
that are not useful for predicting costs (minor injuries) vague and discretionary CCs based on concerns
about gaming in payment models
DCG Predictions:Relative Risk Score (RRS)
Illustrate annual resource use as determined from DCG cost weights
RRS calculated by adding cost weights of an individual’s HCCs and dividing by benchmark (i.e., Medicare) mean dollar amount
RRS normalized so that population mean = 1.00
Prospective Relative Risk Score Calculated
0.45 54 year old maleHCC
5.71 Diabetes with renal manifestation0.95 Type 1 diabetes1.84 Congestive heart failure0.90 Acute myocardial infarction0.89 Vascular disease with complication0 Vascular disease18.09 Dialysis status … …..0.46 Diabetes & congestive heart failure
43.30 Relative Risk Score
Health Score for Year 2
Which Providers are “More Efficient”?
Adjusted Clinical Groups (ACGs)
Clustering of morbidity is a better predictor of health care resource use than presence of specific diseases
Level of resources necessary for delivering health care services is correlated with the morbidity of that population
15,000 ICD-9-CM Diagnosis Codes
Step. 1: Adjusted Diagnosis Groups
Step 2: Collapsed ADGs
Step 3: CADGs combined into Major Adjusted Categories (MACs)
Step 4: Adjusted Clinical Groups
(32 ADGs)
(12 CADGs)
(26 MACs)
(106 ACGs)
AGE, GENDER
Generating ACG Output(Version 4.5)
Examples of ADGs and Their Common ICD-9-CM Codes
ADG Common Diagnosis (ICD-9-CM Code)
1 Time Limited: Minor Noninfectious Gastroenteritis (558.9)3 Time Limited: Major Phlebitis of Lower Extremities(451.2)9 Likely to Recur: Progressive mpaction of Intestine (560.3)
Malignant Hypertensive Renal Disease WithRenal Failure (403.01)Cerebral Thrombosis (434.0)Adult Onset Type II Diabetes w/ Ketoacidosis250.10)
10 Chronic Medical: Stable Essential Hypertension (401.9)Adult-Onset Type I Diabetes (250.00)
Clinical Vignette:40 year old woman: diabetes, hypertension (Release 4.5)
250.41, Diabetes with renal manifestations
401.9, Essential Hypertension
250.00, Adult Onset Diabetes, without complications
V70.0, Adult Routine Exam
ICD-9-CM ADG CADG MAC ACG
31: Preventative Administrative
10: Chronic Medical: Stable
9: Likely to recur: Progressive
5: Chronic Medical: Unstable
6: Chronic Medical: Stable
24: Multiple ADG Categories
4100: 2-3 other ADG combinations Age >34
Applying DCGs/ACGs in VA
Explore the feasibility of adapting diagnosis-based measures to the VA population
Examine how well each measure explains concurrent resource utilization and predicts future resource utilization in the VA
Evaluate their performance in clinically meaningful groups
Profile networks on their efficiency after adjustment for case-mix
ADG Categories in the VA and a Fee for Service Managed Care Population
ACC Categories in the VA and Medicare
Predictive Ratios for Patients with MH/SA Disorders
0.80 0.85 0.90 0.95 1.00 1.05 1.10
HCC 35 (LowerCost MentalDisorders)
HCC 33(Depression)
HCC 32(Psychosis)
DCG/HCC model
DCG/HCC model + dummy markers
Predictive Ratios For Subgroups of Veterans: Concurrent Models
Actual and Predicted Ambulatory Provider Encounters: Concurrent Models
ACG, DCG, and Unadjusted Efficiency Indices By Network
0.75
0.85
0.95
1.05
1.15
1.25
1.35
A C P B O J Q H L I N M D U E R S F G V T K
NETWORKS SORTED BY DCG PREDICTED RESOURCE USE
EF
FIC
IEN
CY
IN
DIC
ES
ACG DCG UNADJUSTED
Improved Special Population Data
*Note: A value greater than 1 means that the actual cost exceeds the predicted cost (or price).
What Weaknesses Remained?
Did not predict mental health costs well Did not explain long-term care costs Did not predict special population costs
Patient Safety Indicators (PSIs)
Developed by Agency for Healthcare Research and Quality (AHRQ)
Screen for potential safety events in the inpatient setting
Risk adjustment based on age, sex, age/sex interactions, DRGs, 27 comorbidities (AHRQ comorbidity software)
Examine observed and risk-adjusted PSI rates in VA 16 medical/surgical PSIs relevant to VA
AHRQ Comorbidities for “Decubitus Ulcer”
Congestive heart failure
Valvular disease
Pulmonary circulation disorders
Peripheral vascular disorders
Hypertension (combine uncomplicated and complicated)
Other neurological disorders
Chronic pulmonary disease
Diabetes, uncomplicated
Diabetes, complicated
Hypothyroidism
Renal failure
Peptic ulcer disease excluding bleeding
AIDS: Acquired immune deficiency syndrome I
Lymphoma
Metastatic cancer
Solid tumor without metastasis
Rheumatoid arthritis/collagen vascular diseases
Obesity
Weight loss
Blood loss anemia
Deficiency anemias
Alcohol abuse
Drug abuse
Depression
Additional VA comorbidities Paralysis
Liver disease
Psychoses
Characteristics of VA and NIS Samples: Discharges and Patients
PSI 3
FacilityVA Observed
AHRQ Expected
VA Expected
VA Obs / AHRQ Expt
VA Obs / VA Expt
A 19.45 22.92 17.60 0.85 1.11
B 18.16 24.10 17.99 0.75 1.01
C 19.42 24.21 20.50 0.80 0.95
“Decubitus Ulcer”
VA does well in non-VA comparison Within VA comparison changes direction
Conclusions
Despite different ways of evaluating model performance, model-based resource allocation for subgroups of veterans would not be adequate
Existing methods (ACGs/DCGs) generally underestimate health care costs of individuals with mental health/substance abuse (MH/SA) disorders
Non-VA based risk adjustment can be misleading in VA facility comparisons
Adequate Risk Adjustment: Important for Veterans with MH/ SA Disorders
The VA is the largest mental health service delivery system in the United States
Prevalence of mental disorders in VA: 30% Goal: develop and validate a psychiatric
diagnosis-based risk-adjustment measure (the “PsyCMS”) for veterans with MH/SA disorders
Guiding Principles
Incorporate all 526 adult MH/SA codes Develop clinically homogeneous categories
based on resource utilization Demonstrate face validity Include “manageable” # of categories Minimize “gaming” Predict concurrent/prospective utilization and
costs
Methods
Sample All veterans who received any health care in the
VA during Fiscal Year 1999 (October 1, 1998 through September 1, 1999) and had a MH or SA diagnosis (ICD-9-CM codes 290-312.9 or 316) (n=914,225)
Methods
Data Diagnostic and utilization data from VA inpatient
and outpatient administrative data Costs obtained from VA Health Economics and
Resource Center (HERC) FY99 data used for concurrent modeling; data
split into 60% development sample (n=548,535) and 40% validation sample (n=365,690)
FY00 data used for prospective modeling
Variables
Dependent Variables Total MH/SA costs: sum of costs associated with all
outpatient and inpatient MH/SA utilization Outpatient MH/SA encounters: sum of all visits associated
with any MH/SA diagnosis code, plus all visits in MH/SA specialty clinics
Inpatient MH/SA utilization: number of days a patient resided in any inpatient setting for MH or SA treatment
Independent Variables Age, gender, diagnostic information (all MH/SA primary
and secondary diagnoses)
Methods
Data Analysis (Four major steps):1. Classification and categorization of all MH/SA
codes into diagnostic classification system
2. Examined distribution of MH/SA disorders using PsyCMS
3. Assessed predictive validity of the PsyCMS using concurrent and prospective modeling
4. Compared performance of PsyCMS with ACGs and DCGs
58.03 other & unspecified anxiety states58.07 other & unspecified neurotic disorders59.03 non-dependent abuse of alcohol59.04 tobacco use disorder59.05 other nondependent drug abuse60.02 sexual deviations & disorders60.03 psychosomatic illness60.04 acute reaction to stress60.05 adjustment reaction, excluding prolonged depressive60.08 behavior disorder60.09 emotional disorders of childhood/adolescence60.10 other mental disorders60.13 attention deficit disorder, other hyperkinetic syndrome60.14 learning/development learning disorder
Drug/Alcohol Dependence/ Psychoses
Psychosis & Other Higher Cost mental
Disorders
Depression & Other Moderate
Cost Mental Disorders
Lower Cost Mental Disorders
Anxiety Disorders
Diagnostic Cost Group (DCG) Mental Health Groupings
ICD-9-CM Psychiatric Codes
ADG 23
Psycho-social: Time Limited, Minor
ADG 24
Psycho-social: Recurrent or Persistent,
Stable
ADG 25
Psycho-social: Recurrent or Persistent,
Unstable
CADG 10
Psycho-Social
MAC 10
Psychosocial
MAC 17
Acute: Minor and Psychosocial
MAC 24
All Other Combinations Not
Listed Above
•CADG 10 & CADG 1 & CADG 3
•CADG 10 & CADG 1 •CADG 10 & CADG
1 & CADG 2 & CADG 3
•CADG 10 & All Other Remaining CADG Combinations
ACG 1500
ACG 1400
ACG 2500
ACG 2700
Refer to MAC 24 Decision Tree
No
Yes
Yes
No
No
Yes
No
•CADG 10 Only
•CADG 10 & CADG 12 &
Anything Else
Yes
ACG 3700
MAC 21
Acute: Minor & Likely to Recur &
Psychosocial
ADG 25?
ADG 24?
ACG 1300
ADG 25?
ADG 24?
MAC 12
Pregnancy
Refer to MAC 12 Decision TreeACG 3500 ACG 2600
MAC 23
Acute: Minor & Acute: Major &
Likely to Recur & Psychosocial
Adjusted Clinical Group (ACG) Psycho-social Groupings
Results: Prevalence of Selected PsyCMS Categories
Table 1: Total MH/SA Costs for Selected Categories
Table 2: Model Goodness of Fit for Concurrent (FY99) Validation Samples
Table 3: Model Goodness of Fit for Prospective (FY00) Validation Samples
Conclusions
PsyCMS appears to be valid and reliable measure for MH/SA risk adjustment
PsyCMS performs better than other systems in predicting concurrent and prospective MH/SA costs/utilization
It can serve as risk-adjustment system for describing MH/SA populations, profiling MH/SA services, and budgeting future MH/SA resources
Rosen AK, Loveland S, Anderson J, Rothendler J, Hankin C, Moskowitz M, Berlowitz DR. Evaluating diagnosis-based case-mix measures: how well do they apply to the VA population? Medical Care 2001; 39(7): 692-704.
Rosen AK, Loveland S, Anderson J. Applying DCGs to Examine the Disease Burden of VA Facilities: Comparing the Six “Evaluating VA Costs” Study Sites to Other VA Sites and Medicare. Medical Care, June 2003: 41(6 suppl): II-91-II-102.
Rosen AK, Loveland S, Anderson J, Hankin C, Breckenridge J, Berlowitz DR. Diagnostic Cost Groups (DCGs) and concurrent utilization among patients with substance abuse disorders. Health Services Research, 2002: 37(4): 1079-1102.
Rakovski C, Rosen AK, Loveland S, Anderson JJ, Berlowitz DR, Ash A. Evaluation of diagnosis-based risk adjustment measures among specific subgroups: can existing measures be improved by simple modifications?" Health Services and Outcomes Research Methodology, 2002: 3(1): 57-74.
Rosen AK, Rakovski C, Loveland S, Anderson JJ, Berlowitz DR. Profiling resource use across providers: do different outcomes affect assessments of provider efficiency after case-mix adjustment? American Journal of Managed Care, 2002: 8(12): 1105-1115.
Rosen AK, Loveland S, Rakovski C, Christiansen C, Berlowitz DR. Do different case-mix measures affect assessments of provider efficiency? Lessons from the VA. The Journal of Ambulatory Care Management 2003: 26(3): 229-242.
Rosen AK, Reid R, Broemeling AM, Rakovski C. Applying a risk adjustment framework to primary care: can we improve on existing measures? Annals of Family Medicine, 2003: 1(1): 44-51.
References (cont’d)
Rosen AK, Trivedi P, Amuan M, & Montez M. The John Hopkins Adjusted Clinical Groups (ACGs) case-mix system: A risk-adjustment methodology currently available at the VA Austin Automation Center. VIReC Insights Vol. 4, No. 1. Hines, IL: VA Information Resource Center, 2003. Available at http://virec.research.med.va.gov.
Liu CF, Sales AE, Sharp ND, Fishman P, Sloan KL, Todd-Stenberg J, Nichol WP, Rosen AK, Loveland S. Case-mix adjusting performance measures in a VA population: pharmacy- and diagnosis- based approaches. Health Services Research, 2003: 38 (5): 1319-1338.
Warner G, Hoenig, H, Montez M, Wang F, Rosen AK. Evaluating diagnosis-based risk-adjustment methods in the spinal cord dysfunction population. Archives of Physical Medicine and Rehabilitation, 2004: 85(2): 218-226.
Rosen AK, Christiansen CL, Montez ME, Loveland S, Shokeen P, Sloan KL, and Ettner SL. Evaluating risk-adjustment methodologies for patients with mental health and substance abuse disorders in the Veterans Health Administration. International Journal of Healthcare Technology and Management, 2006: 7 (1/2): 43-81.
Sloan KL, Montez ME, Spiro A III, Christiansen CL, Loveland S, Shokeen P, Herz L, Eisen S, Breckenridge, JN, Rosen AK. Development and validation of a psychiatric case-mix system. Medical Care 2006: 44:568-580.
Montez ME, Christiansen CL, Ettner SL, Loveland S, Shokeen P, and Rosen AK. Performance of statistical models to predict mental health and substance abuse cost. BMC Medical Research Methodology, October 2006: 6:53.
Rosen AK, Zhaos S, Rivard P, Loveland S, Montez M, Elixhauser A, and Romano P. Tracking Rates of Patient Safety Indicators over Time: Lessons from the VA. Medical Care 2006: 44(9): 850-861.
www.dxcg.com
www.acg.jhsph.edu
Risk Adjustment for Measuring Health Care Outcomes, edited by Lisa Iezzoni, Health Administration Press, 3rd edition, 2003.