Health Savings Accounts and Health
Savings Accounts and
Health Care Spending
p
g
Anthony T. Lo Sasso, University of Illinois at Chicago
Anthony
T. Lo Sasso, University of Illinois at Chicago
Mona Shah, UnitedHealth Group
Bianca K. Frogner, GWU
i
A lot of interest in CDHP, but still relatively little known
• CDHP
CDHP has attracted considerable attention from h
tt t d
id bl tt ti f
employers, government and individuals • Enrollment in HRAs and HSAs has been growing Enrollment in HRAs and HSAs has been growing
among those enrolled in employer sponsored plans as well as the non‐group
plans as well as the non
group market
market
– A recent survey found that 28% of large firms (>1000) offered a CDHP in 2009, up from 22% in 2008
– Overall 8% of workers were enrolled in such plans in 2008 up from 4% in 2006 • MMA
MMA introduced Medicare Part D which features i t d d M di
P t D hi h f t
a “donut hole” benefit design – though that hole is being plugged
is being plugged…
Our goal
Our goal
• The goal of our research is to examine effects h
l f
h
ff
of HSA enrollment on spending using a unique administrative data set with enrollees from over 700 small and large employers from a large national insurer
• We follow two cohorts of individuals before and after employers offered a CDHP: those that switch to a CDHP and those who stay in
that switch to a CDHP and those who stay in traditional plans
Data
• All
All data were extracted from UnitedHealth Care’s massive d t
t t df
U it dH lth C ’
i
data repository – Of course they offer a full complement of traditional plan y
p
p
designs, in addition to being the leading provider of CDHPs in the country
• We
We include employers that did not offer a CDHP in 2005 include employers that did not offer a CDHP in 2005
and then offered HSA plans on either a full replacement
basis or as an option alongside traditional plans in 2006 and 2007
• Employers included commercial groups with 2 to 5,000 employees offering both fully insured and ASO products
• 709 employers: 457 full replacement (all fully insured), 229 fully insured HSA option firms, and 23 (large) ASO option firms
Deductible Pre and Post‐HSA by Cohort, 2005‐2007
$1,800 $1 600
$1,600 $1,400 $1,200 $
, 00
$1,000 HSA
$800 T di i
Traditional
l
$600 $400 $200 $‐
2005
2006
2007
Out‐of‐Pocket Maximum Pre and Post‐
HSA by Cohort, 2005‐2007
$2,500 $2,000 $1,500 $
,
HSA
$1,000 Traditional
$500 $
$‐
2005
2006
2007
Methods are straightforward GLM difference‐in‐differences
• LLog( E(Y
( E(Yit ) ) = α
))
+ β
β1 Post
P it + β
β2 HSAit + β
β3 Post*HSA
P *HSAit + γ Smallit + δ Xit + εit, • Y is annual spending for person i
i
l
di f
i in period t
i
i dt
• X is enrollee characteristics (age, gender, region, and the type of plan in which the enrollee was enrolled in
the type of plan in which the enrollee was enrolled in 2005, 1‐digit industry code dummies, size of the employer, employer insurance type (full replacement,
employer, employer insurance type (full replacement, option‐fully insured, option‐ASO)
• Conveniently, with the log link function, the coefficient y,
g
,
estimate of interest (β3) can be interpreted directly as a multiplicative effect on total costs
More methodology details – concern over selection (levels vs. changes)
• By conditioning on pre‐HSA enrollment spending di i i
S
ll
di
we aim to minimize the potential selection problem that could exist if healthier employees bl
th t
ld i t if h lthi
l
opt to enroll in the HSA versus remaining in a traditional plan
traditional plan
– The basic presumption of the DD model is that while the levels of spending might differ between switchers the levels
of spending might differ between switchers
and stayers, the change in spending in the switcher ggroup would have been the same as the stayers
p
y had the switcher group not enrolled in an HSA (the counterfactual)
More methodology details – concern over selection (triple difference)
• W
We specify a triple‐difference (DDD) model to if t i l diff
(DDD)
d lt
test whether the effect of HSAs differs between full replacement enrollees and option enrollees:
full replacement enrollees and option enrollees: • Log( E(Yit ) ) = α + β1 Postit + β2 HSAit + β3
Post*HSA
Post
HSAit + β
+ β4 Post
Post*Full
Fullit + β
+ β5 HSA
HSA*Full
Fullit + +
β6 Post*HSA*Fullit + γ Fullit + δ Xit + εit
• This allows us to test whether we observe This allows us to test whether we observe
different HSA effects between full replacement enrollees (who presumably face a less voluntary switch to an HSA) and option enrollees, for whom h
) d
ll
f
h
selection is presumably most acute
More methodology details – concern over selection (intent to treat)
• All
All our enrollees were exposed to CDHP, but with a ll
d t CDHP b t ith
group that receives only CDHP and another group that receives a partial “dose” of CDHP we can estimate an p
intent‐to‐treat model
– The group receiving HSAs on a full replacement basis serves the function that a control group receiving no CDHP
serves the function that a control group receiving no CDHP would serve
– We estimate the treatment effect on the treated by estimating a model of the following model: – spend = a + b post + c Full + d post*Full
• This
This allows us to bypass individual‐level selection as a allows us to bypass individual level selection as a
potential source of bias because actual HSA enrollment (on an individual basis) never enters the estimation
Demographics of HSA switchers and traditional plan stayers, 2005
Sample Size
Female (%)
(SD)
Age
Child under 18 (%)
Firm Size
Firm Type
F ll insured,
Fully
i
d Full
F ll Replacement
R l
t
Fully insured, Option
ASO, Option
ALL
76,310
50.8
(50.0)
32.9
(17.7)
27.1
(44.5)
1 867
1,867
(1,699)
HSA Enrollees
23,587
50.4
(50.0)
33.2
(17.9)
26.9
(44.3)
942
(1,172)
Traditional Plan
Enrollees
52,723
51.0
(50.0)
32.8
(17.6)
27.3
(44.5)
2 281
2,281
(1,734)
17.33
17
32.1
50.6
56.00
56
34.2
9.8
31.2
68.8
Health status measures by firm type, 2005
Sample Size
All
76 310
76,310
Retrospective Risk
Score
1 09
1.09
Chronic Disease (%)
HSA
23,587
1.07
49.7
Full replacement
13 218
13,218
1 17
1.17
52 7
52.7
Insured, Option
8,068
0.96
46.3
ASO Option
ASO,
2 301
2,301
0 83
0.83
44 3
44.3
Traditional Plan
52,722
1.09
51.1
Full replacement
0
--
--
Insured, Option
16,433
1.07
52.3
ASO,, Option
p
36,290
,
1.10
50.6
50 7
50.7
DD GLM regression results
DD GLM regression results
VARIABLES
T t l Spending
Total
S di
Medical
M
di l
Spending
Pharmacy
Ph
Spending
Specification
p
1: average
g post
p period
p
effect
HSA*Post
-0.0459*
-0.0384
-0.0626***
(0.0277)
(0.0370)
(0.0199)
Specification 2: differential post periods year effects
HSA*Year 2006
HSA*Year 2007
-0.0570*
-0.0460
-0.0843***
(0.0302)
(0.0386)
(0.0204)
-0.0362
-0.0320
-0.0415*
(0.0313)
(0.0425)
(0.0234)
N = 228,930. Log-link used in GLM specification. Regressions also controls for 2005 plan type
dummies ((EPO,, HMO,, Indemnity,
y, POS,, PPO)) and 1-digit
g SIC dummies. Robust standard errors
(clustered at firm-level) in parentheses. *** p<0.01, ** p<0.05, * p<0.1
Other specifications
Other specifications
VARIABLES
T t l Spending
Total
S di
Medical
M
di l
Ph
Pharmacy
Spending
Spending
Specification
p
3: Triple-difference
p
two-part
p models testingg for differences in
HSA effect between full-replacement enrollees and add-on enrollees
HSA*Post*FullRep
-0.0485
-0.0430
-0.0444
(0.0417)
(0
0417)
Specification 4: intent-to-treat analysis
(0 0556)
(0.0556)
(0 0280)
(0.0280)
Post*Full Replace
-0.0611*
0 0611*
-0.0530
0 0530
-0.0721***
0 0721***
(0.0321)
(0.0432)
(0.0198)
-00.0731
0731
-00.0634
0634
-00.0862
0862
Implied TOT:
N = 228,930. Log-link used in GLM specification. Regressions also controls for 2005 plan type
dummies (EPO, HMO, Indemnity, POS, PPO) and 1-digit SIC dummies. Robust standard errors
((clustered at firm-level)) in p
parentheses. *** p
p<0.01,, ** p
p<0.05,, * p
p<0.1
TOT scaling factor equal to (1 – HSA) take-up in the option group (16.4%)
Our results point to three key findings: point 1
• HSA
HSAs are associated with a statistically significant and i t d ith t ti ti ll i ifi t d
economically meaningful relative decrease in spending when compared to individuals who remained in traditional plans
– Overall, enrollees in HSAs spent roughly 5‐7% less when p
p
compared to traditional health plan enrollees
– There is evidence that more of the relative reduction in spending occurred in the first year of enrollment
• Could
Could be consistent with prior year be consistent with prior year “benefit
benefit rush
rush”
• Learning effects
• Account balance accrual could counter‐act less generous benefit design
g
– Our estimates are probably conservative because in this naturalistic setting the comparison group benefits are getting g
stingier over time
Our results point to three key findings: point 2
• Pattern of our results suggests that HSAs had f
l
h
S h d
larger relative effects on pharmacy spending vs. outpatient and inpatient spending
t ti t d i
ti t
di
– For pharmacy spending, HSA enrollees spent 6‐9% less th t diti
than traditional plan enrollees
l l
ll
– Our findings are consistent with the notion that CDHP benefit designs affect decisions that are at the
benefit designs affect decisions that are at the discretion of the consumer, such as whether to fill or p
p
refill a prescription, but have less effect on care decisions that are (more) at the discretion of the provider
Our results point to three key findings: point 3
• Whil
While there is evidence consistent with favorable th
i
id
i t t ith f
bl
selection into HSAs based on the levels of spending we provide compelling evidence to
spending, we provide compelling evidence to suggest that there is not selection on changes (or trends) in spending
)
p
g
– The distinction is critical because we have longitudinal data spanning one year prior to the HSA being offered to two years post HSA introduction
to two years post‐HSA introduction
– We believe the combination of estimates provided e e ep ese
e s o ges e de ce p ese ed o
here represent the strongest evidence presented to date regarding the importance (or lack thereof) of selection when controlling for baseline spending