Objective 4 Flashcards
Risk factors that indicate whether a person may have high claims
- Inherent risk factors: age, sex, race
- Medical condition-related factors
- Family history for inheritable conditions
- Lifestyle risk factors: smoking, exercise, nutrition
- External risk factors: industry, location, education
Types of medical management interventions
- Care coordination (system): case management, discharge planning, in-hospital care coordination
- Condition management (patient): disease management, risk factor management
- Provider management (provider): provider profiling, pay-for-performance, ACOs
Areas where condition-based models are used in healthcare financial applications
- Program management: ID high-risk individuals, financial modeling and resource allocation, program evaluation (savings)
- Provider or health plan reimbursement: normalizing populations (for payment or evaluation), profiling providers
- Actuarial and underwriting functions: pricing, underwriting, projecting future claim costs
Types of predictive models that are not based on medical conditions
- Age/sex
- Prior cost
- Combo of age/sex and prior cost
Sources of data for developing risk factors
- Claims data
- Self-reported data
- External data
Risk factors identified by a health risk assessment
- Personal disease history
- Family disease history
- Health screenings and immunizations
- Drinking
- Smoking
- Sunscreen
- Nutrition
- Exercise
- Weight management
- Injury prevention
- Stress and wellbeing
- Women’s health
- General health assessment
- Functional health status
- Mental health status
Types of data used for predictive modeling
- Physician referral/chart (high reliability / low practicality)
- Enrollment (high / high)
- Claims (medium / high)
- Pharmacy (medium / high)
- Lab values (high / low)
- Self-reported (low / low)
Drawbacks of using data from medical charts
- Don’t cover out-of-network services or drugs prescribed by out-of-network providers
- Don’t record patient compliance
- Transcribing / transferring to uniform format is time consuming and requires highly-trained staff
- Lack of uniformity in physician coding of conditions and severity
- Typically unavailable to health plan or actuary
Advantages and disadvantages of using diagnosis codes for identifying member conditions
Advantages
1. Almost always present on medical claims
2. Uniform
3. Useful for identifying conditions
Disadvantages
1. Generally only primary and secondary are populated
2. Coding errors
3. Upcoding
4. Different physicians follow different coding practices
Drawbacks of using survey data
- Must be commissioned, budgeted, executed
- Isn’t updated as medical events occur - can become stale
- Response bias
- Untruthful answers
Questions to answer when building a clinical identification algorithm
Set of rules applied to claims to identify conditions
- Where are diagnoses?
- Source of diagnosis?
- If claims, what claims to consider (IP/OP/Lab)?
- If more than one diagnosis, how many to consider?
- Over what time span and how often must diagnosis appear?
- What procedures for identifying severity?
- What drugs for identifying conditions?
Challenges when constructing a condition-based model
- Large number of procedure and drug codes
- Severity level at which to recognize condition
- Impact of co-morbidities for conditions often found together
- Degree of certainty
- Extent of data
- Underlying benefit design
Definitions of sensitivity and specificity
- Sensitivity - percent of true positives
2. Specificity - percent of true negatives
External sources of clinical identification algorithms
- HEDIS from NCQA
- DMAA (for chronic diseases)
- Grouper models
- Literature
Reasons for using commercially-available grouper models
- Lots of work to build from scratch
- Must be maintained to add new codes - even more work
- Commercially-available models are accessible to many users. May be a requirement for providers and plans
Common features of Medicare prospective payment systems
- System of averages
- Increased complexity
- Relative weights
- Conversion factors
- Outlier payments
- Updates
- Focus on access and quality
Challenges with patient classification systems based on coding systems
- Need for new DRGs
- ICD coding
- Upcoding
- New coding systems
Factors for choosing the right predictive model
- Correlation structure
- Purpose of analysis
- Nature of available data
- Characteristics of outcome variable
- Distribution of outcome variable
- Functional relationship
- Complex decision model
Steps of the data warehousing process
- Identify patients to include
- Identify data elements to merge on
- Create flags
- Attach derived variables and flags
Characteristics for assessing the quality of a model
- Parsimony
- Identifiability
- Goodness of fit
- Theoretical consistency
- Predictive power
Statistics for determining whether a model is good
- R^2 and adjusted R^2 = 1 - (1 - R^2) * (N - 1) / (N - k - 1)
- Regression coefficients (signs / significance)
- F-Test
- Statistics for logistic models:
a) Hosmer-Lemeshow
b) Somers’ D
c) C-statistic - Multicollinearity
- Heteroscedasticity
- Autocorrelation
Re-sampling methods for validating a model
- Bootstrap (sampling with replacement)
- Jackknife (leave out 1 observation at a time)
- Cross-validation (subsets held out)
- Permutation test (rearrange labels)
Factors used in developing risk scores in the CMS-HCC risk model
HCC = Hierarchical condition category
- Demographics - higher for duals
- Disabled
- Separate models for LTC / ESRD
- New enrollees - separate age/gender only factors
- Prospective risk-adjustment methodology
- Calibration every 2 years
- Health status factors from ICD-9 codes grouped into HCCs
Central features of Massachusetts health care reform
- Exchange
- Employers must establish Section 125 accounts
- Large subsidies for families under 300% FPL
- More limited plan available for families over 300% FPL
- Individual mandate
- Funding through federal funds for safety net hospitals / uncompensated care
Steps for developing predictive models for care management programs
- Choose a disease or condition
- Rank conditions based on intervenability
- Identify population
- Plan the intervention
- Perform economic modeling of program
- Develop the predictive model
- Test actual outcomes against predictions, modify model and program as needed
Metrics that should be recognized in the Risk Management Economic Model
- Number and risk-intensity of members targeted
- Types of interventions used
- Number of nurses / staff, program costs
- Methodology for contacting and enrolling members
- Rules for integrating the program with the rest of the care management system
- Timing and number of contacts, enrollments, interventions
- Predicted behavior of target population without intervention, predicted effectiveness of intervention at modifying