Session 121 PD, Medicare Advantage Risk Score Basics Moderator: Christine Sue Bach, ASA, FCA, MAAA Presenters: Christine Sue Bach, ASA, FCA, MAAA Gregory Joseph Herrle, FSA, MAAA
2015 SOA Annual Meeting Session 121 Medicare Risk Scores Beginning & Intermediate Topics October 13, 2015 Greg J. Herrle, FSA, MAAA Consulting Actuary Milliman, Inc. Gregory.J.Herrle@Milliman.com Chris Bach, ASA, MAAA, FCA Senior Consulting Actuary Wakely Consulting Group, Inc. ChrisB@Wakely.com
Agenda What and why of risk adjustment Risk adjustment example Risk adjustment models and CMS Risk score data timing Risk score projections Risk adjusted payments Payment Timing Risk score coding
What is a (Medicare) risk score? Relative numerical representation of expected future illness burden Unique to each individual Based on data gathered during prior year Source: ING
Why is risk adjustment needed? Benchmark payment rates developed for average Medicare beneficiary Plans will actually enroll varying levels of risk / illness burden If a plan enrolls a sick individual, the plan would otherwise be underpaid relative to the expected costs of that person Risk adjustment allocates payments to plans consistent with the risk of each plan s enrolled beneficiaries
What is a risk adjustment model? Uses enrollment, diagnoses, and claims to develop a method for quantifying an individual s illness burden Diagnosis information used to identify each person s conditions Using the identified conditions, regression model develops appropriate weight for each condition, including age, gender, and other potential demographic factors Can be concurrent or prospective
CMS Risk Adjustment Model Statistical Model Age Gender Medicaid dual status Disability status ESRD / Institutional status Low Income Status (Part D) Health status (diagnoses) Additive Prospective Calibrated on FFS population (Part C Only)
CMS Risk Adjustment Model (cont.) Demographic Information Diagnoses (ICD 9/10 codes) Age Gender Dual status Etc. Hierarchical Condition Category (HCC) Relative Risk Factor (RF) 1 HCC 1 HCC 15 HCC 80 RF 2 RF 3 RF 4 Risk Score = RF 1 + RF 2 + RF 3 + RF 4
Medicare Risk Score Data Timing Risk score data is collected via Risk Adjustment Processing System (RAPS) data submissions from health plans to CMS. Initial submission for a calendar year is March of the following year, with final submission nine months later Initial submission for a non-calendar year (July 1 June 30) is September of the following year, with final submission nine months later
Medicare Risk Score Data Timing Graphic of Calendar Year Data Submissions yy = dates of service for which diagnosis data is collected initial submission final submission yy yy + 1 yy + 2 J F M A M J J A S O N D J F M A M J J A S O N D J F M A M J J A S O N D
Medicare Risk Score Data Timing Graphic of Non-Calendar Year Data Submissions yy = dates of service for which diagnosis data is collected initial submission final submission yy yy + 1 yy + 2 J A S O N D J F M A M J J A S O N D J F M A M J J A S O N D
Medicare Risk Score Data Sources for Plans Monthly Membership Reports (MMRs) Monthly accounting of plan revenue and risk scores Sent to health plans monthly Accompanies CMS payment to plans Risk scores are presented on a normalized basis Based on most recent risk score timing
Medicare Risk Score Data Sources for Plans Annual Beneficiary Files Annual detailed file of risk scores by beneficiary by month CMS produces the files each April Risk scores are presented on an unadjusted, non-normalized basis sometimes referred to as the raw risk score Based on previous year membership and completed, calendar year risk score data submission
Calculating Base Period Risk Score to Include in BPT Worksheet 1
Options for Calculating Projected Risk Score to Include in BPT Option 1 CMS Preferred Methodology Based on annual beneficiary file data from previous year Option 2 Alternate Methodology Based on YTD MMR file data (usually first quarter of current year)
Why choose one method over the other? Credibility considerations Data quality considerations Unusual enrollment situations for new plans
Risk Score Credibility - CMS Guidance Risk score credibility guidance memo can be found here: http://www.cms.gov/medicare/health-plans/medicareadvtgspecratestats/bid-pricing-tools-and-instructions- Items/BidGuidance.html?DLPage=1&DLEntries=10&DLSort=0&DLSortDir=descending
MA Risk Score Projection PREFERRED METHOD (bene file starting risk score) Development of 2016 Projected MA Risk Scores 2014 Description Model A Starting Risk Score 1.0691 B Convert to Raw - remove Normalization 1.0000 C Convert to Raw - remove MA Coding Pattern Adj 1.0000 D Plan Specific Coding Trend 1.0465 E Starting Data Adjustments i) Transition from lagged to non-lagged diagnosis data 1.0000 ii) Incomplete reporting of diagnosis data 1.0000 iii) Seasonality 1.0000 F Plan Specific Adjustments 1.0607 G Risk Model Adjustment i) Raw 2015 HPMS Posted Data 1.0000 ii) Missing Diagnosis Adjustment 1.0000 iii) Raw 2014 HPMS Posted Data 1.0000 H Raw Risk Scores, Projected to 2016 1.1868 I MA Coding Pattern Adjustment 0.9484 J Normalization Factor 0.9780 K Frailty Factor (additive) - L Final Risk Score 1.1509
ALTERNATE METHOD (YTD MMR starting risk score) Development of 2016 Projected MA Risk Scores 2014 Description Model A Starting Risk Score 1.0376 B Convert to Raw - remove Normalization 1.0260 C Convert to Raw - remove MA Coding Pattern Adj 0.9509 D Plan Specific Coding Trend 1.0230 E Starting Data Adjustments i) Transition from lagged to non-lagged diagnosis data 1.0180 ii) Incomplete reporting of diagnosis data 1.0250 iii) Seasonality 0.9737 F Plan Specific Adjustments 1.0607 G Risk Model Adjustment i) Raw 2015 HPMS Posted Data 1.0000 ii) Missing Diagnosis Adjustment 1.0000 iii) Raw 2014 HPMS Posted Data 1.0000 H Raw Risk Scores, Projected to 2016 1.2343 I MA Coding Pattern Adjustment 0.9484 J Normalization Factor 0.9780 K Frailty Factor (additive) - L Final Risk Score 1.1969
Risk Score Projection Starting Risk Score Risk score projections must start with a raw risk score no coding pattern adjustment or FFS normalization. MA coding pattern adjustment CMS adjustment to account for coding improvement over time Varies by year FFS Normalization factor CMS adjustment to normalize the total risk scores back to 1.0 Varies by year
Risk Score Projection Starting Risk Score Development of Raw Risk Score Preferred Method Using the starting risk score from the beneficiary file No need to remove normalization or coding trend since beneficiary file is already raw Raw risk score = A * B / C
Risk Score Projection Starting Risk Score Development of Raw Risk Score Alternate Method A Starting Risk Score 1.0376 B Convert to Raw - remove Normalization 1.0260 C Convert to Raw - remove MA Coding Pattern Adj 0.9509 Using the starting risk score from YTD MMR files Need to remove current year FFS normalization and coding adjustment since MMR files contain both Raw risk score = A * B / C
Risk Score Projection Plan Specific Coding Trend Applies to both preferred and alternative methods Plans must determine the appropriate trend for risk scores Risk score coding trend represents the plan s expected annual improvement in coding Appendix K of the 2015 MA BPT Instructions includes discussion of considerations for developing risk score trend
Risk Score Projection Timing Adjustments E Starting Data Adjustments i) Transition from lagged to non-lagged diagnosis data 1.0180 ii) Incomplete reporting of diagnosis data 1.0250 iii) Seasonality 0.9737 Applies only to alternate method, since beneficiary risk scores used in preferred method are already calendar year and complete. Timing adjustment = E = E.i * E.ii * E.iii
Risk Score Projection Timing Adjustment Definitions Lagged to non-lagged data converts data to a calendar year Incomplete data adjusts for final data submission Seasonality converts a partial year of data to a full year
Risk Score Projection Plan Specific Adjustments Applies to both preferred and alternative methods Plans must determine if any population adjustments must be made for differences between base period and projection period populations When using population adjustments to the risk score, you must consider the need for a corresponding claims adjustment
Risk Score Projection Risk Model Changes Applies to both preferred and alternative methods Accounts for changes in CMS HCC models Missing Diagnosis Adjustment needs to be calculated for any plans that filtered diagnosis data in RAPS submissions
Risk Score Projection Risk Model Changes G Risk Model Adjustment i) Raw 2015 HPMS Posted Data 1.0000 ii) Missing Diagnosis Adjustment 1.0000 iii) Raw 2014 HPMS Posted Data 1.0000 Risk model adjustment = G = G.i * G.ii / G.iii Note: Sometimes CMS will instruct plans to use a blend of two different HCC models to mitigate transition changes.
Risk Score Projection Projected Raw Risk Score A Starting Risk Score 1.0691 B Convert to Raw - remove Normalization 1.0000 C Convert to Raw - remove MA Coding Pattern Adj 1.0000 D Plan Specific Coding Trend 1.0465 E Starting Data Adjustments i) Transition from lagged to non-lagged diagnosis data 1.0000 ii) Incomplete reporting of diagnosis data 1.0000 iii) Seasonality 1.0000 F Plan Specific Adjustments 1.0607 G Risk Model Adjustment i) Raw 2015 HPMS Posted Data 1.0000 ii) Missing Diagnosis Adjustment 1.0000 iii) Raw 2014 HPMS Posted Data 1.0000 H Raw Risk Scores, Projected to 2016 1.1868 Projected Raw Risk Score = H = A * B / C * D * E * F * G
Risk Score Projection Coding, FFS Normalization and Frailty Factor Applies to both preferred and alternative methods Raw projected risk score must be adjusted for projection year coding pattern and FFS normalization Frailty factor is an additive adjustment that applies to Fully Integrated Dual (FIDE) SNPs only.
Risk Score Projection Final Worksheet 5 Projected Risk Score H Raw Risk Scores, Projected to 2016 1.1868 I MA Coding Pattern Adjustment 0.9484 J Normalization Factor 0.9780 K Frailty Factor (additive) - L Final Risk Score 1.1509 Final Projected Risk Score = L = ( H * I / J ) + K
Risk Adjusted Payments Plan estimates the required revenue and risk level for future enrolled population CMS converts this to a revenue payment at a 1.00 risk score Plan is actually paid based on the calculated risk score of the enrolled population
Risk Adjustment Example Plan A/B Bid Revenue Requirement 800 Plan A/B Risk Score 0.80 Normalized Plan A/B Bid 800 / 0.80 = 1,000 Actual enrolled risk 0.90 Actual Plan A/B Revenue (excl. Rebates) 1,000 * 0.90 = 900
Payment Timing #1: January 2015 - MMRs Used for January 2015 June 2015 payments Lagged and incomplete Based on July 2013 June 2014 diagnoses #2: July 2015 - MMRs Used for July 2015 December 2015 payments and mid-year restatement payment Non-lagged, but still incomplete Based on January 2014 December 2014 diagnoses
Payment Timing (cont.) #3: April 2016 Beneficiary Level File Represents final risk score for 2015 Used for final settlement payment in July or August 2016 Non-lagged and complete Still based on January 2014 December 2014 diagnoses, but now submitted through early 2016 Used for 2017 bid development
Payment Timing Example January 2015 Risk score = 1.5 January 2015 June 2015 payment = 1.5 * $1,000 July 2015 Risk score = 1.4 July 2015 December 2015 = 1.4 * $1,000 Mid-year restatement (for Jan to June) = (1.4 1.5) * $1,000 * 6 April 2016 Beneficiary level file Risk score = 1.6 Final settlement (paid July or August 2016) = (1.6 1.4) * $1,000 * 12
Risk Score Coding Critical to success of health plans All risk scores are normalized to nationwide average risk score which increases each year Proactive: Code it right the first time Provider approach Educate and motivate Documentation prompting Member approach Each member seen each year High risk patient focus Home bound / institutionalized Track missed opportunities in databases Data Approach Ensure permanent conditions are reported each year Reactive: Make sure it is right Confirm / substantiate High scores Significant changes Chart Review Billing few codes per claim Historical documentation issues Proper Submission All settings All provider types
Minimum MA Coding Intensity Adjustment 3.41% for 2010 through 2013 Extends MA Coding Intensity Application Doing nothing or only a little is a losing proposition Year Minimum Coding Intensity Adjustment % 2014 4.91% 2015 5.16% 2016 5.41% 2017 5.66% 2018 5.91% 2019 5.91% Source: CMS
Opportunities for Risk Score Improvement Dropped and missing diagnoses Prioritization of HCCs Home visits Hire vendor Provider risk sharing arrangements
Questions