SLIDE 1
Bending the Cost Curve within a Pioneer ACO: The Role of Care Management
John Hsu 28 June 2016 AcademyHealth Annual Research Meeting Boston
1
SLIDE 2 Study Team and Disclosures
- John Hsu
- Maggie Price
- Christine Vogeli
- Richard Brand
- Michael Chernew
- Eric Weil
- Sreekanth Chaguturu
- Tim Ferris
Funding:
- NIA P01 AG032952
- Partners Healthcare
Speaker Disclosure:
- Hsu works at MGH, which is part of the Partners Healthcare System
2
SLIDE 3 Background
- Concerns about medical spending growth
- CMS: Alternative Payment Models (APMs)
– Movement away from FFS – Changes in the incentive structure
- Limited information on provider group strategies
- Underlying mixture of underuse and overuse
3
SLIDE 4 ACO Summary
- Several types/levels of ACOs
- Initial contracts between 2012-14
- Potential ACO provider group strategies/mechanisms:
– Improve matching of service level and need, e.g., clinic vs. ED care – Prevent need for clinically downstream services, e.g., hospitalizations – Reduce unnecessary use – Reduce unit prices
- Modest “savings” on average
- Comparison group challenges
4
SLIDE 5 Study Context
- Pioneer ACO
- Prior experience with Medicare High Risk Care Management Program
– Prior program in one large hospital within the system – Revised program for ACO, i.e., the integrated Care Management Program (iCMP=main ACO intervention)
- Local market with multiple Pioneer ACOs (5)
- State focused on spending growth reduction (Chapter 224)
5
SLIDE 6 Research Questions
- How did alignment to an ACO impact clinical event rates and
spending?
- Among ACO beneficiaries, how did entry into an integrated Care
Management Program (iCMP) impact clinical event rates and spending?
6
SLIDE 7 Methods
ACO Population
- Detailed data from one of the largest Pioneer ACOs (>82K beneficiaries
aligned)
– Examined beneficiaries newly aligned in January 2012 or January 2013 – Medicare claims available from 2009-14
- Followed beneficiaries until departures from Traditional Medicare program,
departure from the ACO catchment area, or death iCMP Identification and Entry
- High risk beneficiaries identified annually based on risk scores
– PCP review of annual lists for those with “modifiable” risks/spending – iCMP identification year is the year the beneficiary first appears on this list
- Beneficiaries were assessed by a care manager before starting the iCMP
program
- iCMP analyses focus on those beneficiaries that were on the lists and entered
the iCMP program (i.e. were assessed) in 2012-2014
7
SLIDE 8 Methods
– ED Visits: monthly counts – ED Visit Severity
- Used the NYU algorithm to classify the severity of ED visits
- Algorithm assigns a probability that a diagnosis falls in to each of four categories of
increasing severity; focused on the probabilities of the two lowest severity categories
- Visit defined as non-emergent of primary care treatable if >50%
– Hospitalizations: monthly counts – Medicare Costs
- Monthly total costs
- Standardized to 2012$
- Models:
– Negative binomial models for visit counts with individual-level fixed effects – Linear models for costs with individual-level fixed effects
8
SLIDE 9 Sensitivity Analyses
- “Dose” effect
- Historical secular trends
- Operational definitions, e.g., non-emergent ED visits
- Attrition
- Model fit
9
SLIDE 10
Baseline Characteristics by ACO Alignment Year
10
ACO Alignment Year 2012 2013 p-value Beneficiaries 42,417 19,649 Mean Age 72.6 71.2 <0.001 Female 61% 60% 0.114 Race: White 89% 89% <0.001 Black 5% 5% Other 6% 7% OREC= Aged 81% 80% 0.029 Mean CMS-HCC Score 1.1 1.2 <0.001 Dual 20% 21% <0.001
SLIDE 11
Baseline Characteristics by iCMP Identification Year
11
iCMP Identification Year 2012 2013 2014 p-value Beneficiaries 2,143 1,917 760 Mean Age 74.3 73.4 73.0 0.0143 Female 59% 62% 61% 0.076 Race: White 89 86% 94% <0.001 Black 6 8% 4% Other 5 6% 3% OREC= Aged 73% 71% 76% 0.027 Mean CMS-HCC Score 2.4 2.5 1.5 <0.001 Dual 24% 31% 25% <0.001
SLIDE 12 Comparable Pre-program Trends
12
0.00 0.02 0.04 0.06 0.08 0.10 0.12 0.14 0.16 0.18 0.20 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 2009 2010 2011 ED Visit Rate ACO Start = 2012 ACO Start = 2013 0.00 0.05 0.10 0.15 0.20 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 2009 2010 2011 ED Visit Rate iCMP Start: 2014: 1-6 iCMP Start: 2014: 7-12
SLIDE 13
Modest Changes in Clinical Event Rates with ACO Alignment
13 13
0.6 0.7 0.8 0.9 1.0 1.1 1.2 1.3 1.4 13+ mos 7-12 mos ACO Alignment (vs. pre): 1-6 mos ACO Alignment (post vs. pre) 13+ mos 7-12 mos ACO Alignment (vs. pre): 1-6 mos ACO Alignment (post vs. pre) 13+ mos 7-12 mos ACO Alignment (vs. pre): 1-6 mos ACO Alignment (post vs. pre) Hospitalizaitons Non-Emergent ED Visits ED Visits Relative Rate
SLIDE 14
Changes in Clinical Event Rates with iCMP Entry
14
0.6 0.7 0.8 0.9 1.0 1.1 1.2 1.3 1.4 13+ mos 7-12 mos iCMP Entry (vs. pre): 1-6 mos iCMP Entry (post vs. pre) 13+ mos 7-12 mos iCMP Entry (vs. pre): 1-6 mos iCMP Entry (post vs. pre) 13+ mos 7-12 mos iCMP Entry (vs. pre): 1-6 mos iCMP Entry (post vs. pre) Hospitalizaitons Non-Emergent ED Visits ED Visits Relative Rate
SLIDE 15 15
Changes in Monthly Costs Before and After ACO Alignment and iCMP Entry
15
- $700 -$600 -$500 -$400 -$300 -$200 -$100
$0 $100 13+ mos 7-12 mos iCMP Entry (vs. pre): 1-6 mos iCMP Entry (post vs. pre) ACO Alignment (post vs. pre)
Cost Difference
SLIDE 16 Conclusions
- Modest effects associated with entry into the ACO
- Additional, larger effects associated with entry into a care
management program
- Care management effects initially modest, but larger over time
16
SLIDE 17 Limitations
- Non-random assignment, thus potential selection bias from time-changing
unmeasured covariates
- Real-world clinical environments, thus while iCMP appears to be the main
program within the ACO, there are other smaller population health programs that could contribute to the observed effects
- Single health care system, albeit a large system with multiple hospitals and
thousands of physicians, thus unclear generalizability to other settings, e.g., non-ACOs, MSSPs
– Effect of specific mechanisms with much precision – Potential heterogeneity in effects across system or patient traits
- Medicare perspective, with no assessment of total spending including
program costs
17
SLIDE 18 Implications
- Promising findings for main “intervention” within the ACO
- Evidence consistent with shifts in care delivery/matching need with
site as one potentially effective strategy for reducing spending growth
- Population stability and time to observe “payoff”
18
SLIDE 19
Questions
John.Hsu@mgh.harvard.edu
19
