Comparing the Effectiveness of
a Consumer Directed Health
Plan with Traditional Cost
Sharing in Controlling Medical
Care Use and Cost
Roger Feldman/Steve Parente
University of Zurich, Switzerland
June 21, 2006
This research was supported by a grant from Pfizer, Inc. The
authors have no conflict of interest with the grantor.
School of Public Health
Outline
• What is a consumer directed health plan?
– General introduction and preliminary research
findings
• Theory of consumer behavior
– CDHP cost-sharing design creates a budget
constraint with 2 kinks
– Contrast with ‘standard’ health insurance that uses
coinsurance or deductible
– Determine expected effects on enrollee behavior
• Empirical model and results
School of Public Health
‘Classic’ CDHP Model – Definity
Health in Minneapolis Health Tools
and Resources
Health Tools and Resources
• Care management programs
• Internet tools to manage HRA, find
providers and services
$$
Definity
Health
Care
Advantage
School of Public Health
Annual
Deductible
HRA
Preventive Care 100%
Health Coverage
• Preventive care covered 100%
• Annual deductible in $1,000s
• Expenses above deductible covered
at 80-100%
Health
Coverage
Annual Deductible
Health Reimbursement Account (HRA)
• Employer creates tax-advantaged account
• Member directs HRA
• Account rolls over at year-end
• Account does not contain ‘real’ money
and does not belong to employee
Web- and
PhoneBased
Tools
The Health Savings Account (HSA)
An HSA account is owned by the
individual and used to pay for current
and future medical expenses
HSAs were enabled by the 2003
Medicare Modernization Act
Bush Administration has proposed
refundable tax credits for individuals
to purchase plans with HSAs
HSAs are offered by UnitedHealth, the
Blues, Aetna, Cigna, Humana, and
Kaiser
School of Public Health
$$
Annual
Deductible
HSA
Preventive Care 100%
Used with a high deductible health
plan
Health
Coverage
Annual Deductible
Both employee and employer can
make tax-free contributions to HSA
Who Chooses a CDHP?
• Strongest and most consistent conclusion:
CDHPs are preferred by highly-paid employees
(Parente, Feldman, and Christianson, 2004)
• A large employer that offered a PPO and POS
plan introduced Definity Health 2001:
– 38% of employees choosing Definity had income
above the firm’s 75% percentile
– 19% of POS and 29% of PPO enrollees were above
the 75th percentile of income
School of Public Health
Do CDHPs Experience
Favorable Selection?
• When the University of Minnesota offered a
CDHP in 2002, there was no evidence of
favorable selection (PFC, 2004)
• In the large employer previously mentioned,
CDHP enrollees had lower baseline illness
burden than the PPO but about equal to the
POS plan
• In our largest sample of 80,000 covered lives in
3 employers, there is evidence of mild
unfavorable selection against the CDHP
School of Public Health
Potential HSA Take-up
• Medicare Modernization Act of 2003 lets individuals and employers
contribute to tax-free health savings accounts up to the lesser of the
deductible (at least $1,050 for an individual or $2,100 for a family) or
a maximum set by law ($2,700 for a single person or $5,450 for a
family)
• 2006 State of the Union (SOTU) proposes to eliminate all taxes on
out-of-pocket spending through HSAs
• Low-income families would be offered refundable tax credits to
purchase health insurance policies with HSAs
• We project that tax credits would increase HSA enrollment from 3
million to 7 million (Feldman, et al., 2005) with much of the increase
coming from previously uninsured
School of Public Health
Price-elasticity* of uninsured take-up
with respect to HSA premium subsidy
Income Quartile
Single
Adults
Adults with
Dependents
0 to 25
0.080
0.039
25 to 50
0.138
0.107
50 to 75
0.498
0.250
75 to 100
0.754
0.378
All
0.205
0.107
* = % change in uninsured / % change in premium
School of Public Health
CDHP Effects on Cost and Use
• Evidence is very preliminary, but it
suggests that CDHPs with small gap
between HRA and deductible do not
control cost & use (PFC, 2004)
• CDHP disadvantage is most striking for
hospital cost versus POS cohort
• CDHP pharmacy cost trend is most
favorable
School of Public Health
Hospital Costs 2000-2002
$3,500
$3,000
$2,500
$2,000
CDHP Cohort
PPO Cohort
POS Cohort
$1,500
$1,000
$500
$0
2000
2001
2002
Note: costs are adjusted with 2-part model for age, gender,
case-mix, income, number of covered lives in contract, and
use of flexible spending account
School of Public Health
Why is Today’s Talk Different?
• Our prior work simply compared costs or
use for all enrollees in each cohort, e.g.
did the CDHP cohort spend more on
average than the PPO cohort?
• This research recognizes that the effects
may differ within the CDHP cohort
– Key insight: CDHP creates a budget
constraint with 2 kinks
– Kinked budget constraint has different
incentives for ‘low’ vs. ‘high’ users of services
School of Public Health
The Kinked CDHP Budget Constraint
Goods
Deductible Budget
e
a
b
c
f Co-Insurance Budget
d
CDHP Budget
Medical Care
HRA Deductible
School of Public Health
If I Only Lived One Year…
FSA = Flexible Spending
Account
Goods
With an FSA, You ‘Use It or
Lose It’ at the end of the
year
FSA Budget
Medical Care
FSA
School of Public Health
Key Difference Between FSA
and CDHP
• HRA lasts more than one year if you stay with
the employer and HSA is owned by enrollee
• Both versions of CDHP provide ‘insurance’
against cost of possible serious illness in the
future
• Risk-averse, far-sighted consumers in good
health may conserve medical care in a CDHP
School of Public Health
Model (1): Basics
2 periods and 2 states of the word: ‘good health’ and
‘serious illness:’
U t  W (Gt )  H ( M t ) GH
U t   SI ; M t  M
U t  W (Gt ) SI ; M t  M
Known probability p that healthy consumer will become
seriously ill next period
Fixed money income Y and employer contribution of C in
each period
C1 can only be spend on medical care; C2 can be spent on
medical care or goods
Insurance policy has deductible D and no coinsurance
School of Public Health
Model (2): CDHP Equilibrium
Healthy consumer’s objective function:
E( U ) W ( Y ) H ( C  S 1)
( 1  p ) [ W ( Y  C  S 1 M 2 )  H ( M 2 ) ] 
p W (Y  C  S1  D )
Solution involves finding optimal M2 given good
health, then recursively choosing how much to
save in the first period:
H1 / W1  E(W2) / W1
School of Public Health
CDHP Equilibrium
Goods
a
b
c
d
CDHP Budget
Medical Care
HRA Deductible
School of Public Health
Model (3): Deductible Plan
• Deductible plan (D-plan) has same F.O.C. but
different equilibrium value of M1 compared with
CDHP
• Suppose the D-plan enrollee had the same
value of M1 as the CDHP enrollee
• That would imply equal H1 but less saving
• Less saving implies a higher value of E(W2 )
• That contradicts the F.O.C.
• The D-plan enrollee cuts back M1 until H1 =
E(W2 )
School of Public Health
Model (4): Coinsurance plan
• Assume the coinsurance plan (C-plan) has c %
coinsurance up to a maximum limit on out-ofpocket expenditure
• At c = 1, the C-plan is identical to D-plan
• At c = 0, medical care in the C-plan is ‘free’ so
there is no reason to conserve
• The demand function is continuous, so there
must be 0 < c* < 1 at which M1 in the C-plan and
the CDHP are equal
School of Public Health
Predicted Spending by Budget
Region
Region 1 –
predicted spending
less than employer
contribution to
HRA
D-plan lowest
C-plan and CDHP
higher with
uncertain order
Region 2 –
predicted
spending above
HRA but below
deductible
D-plan = CDHP <
C-plan
School of Public Health
Region 3 –
predicted
spending above
deductible
D-plan = CDHP =
C-plan
Data
• Large employer added a CDHP to previouslyoffered PPO and POS Plans in 2001
• Quasi-experimental pre/post design
• We selected 3 cohorts of workers continuously
employed from 2000-2003:
– Always in PPO
– Always in POS
– PPO or POS in 2000, switched to CDHP in 2001 and
stayed in CDHP 2002 and 2003
School of Public Health
Plan Characteristics
PLAN
CDHP
CHARACTERISTIC
POS and PPO
Employer HRA
contribution
Not applicable
Deductible
Coinsurance/Co-pay
$1,000 single
$1,500 2-person
$2,000 family
$1,500 single
$2,250 2-person
$3,000 family
None
Rx coverage
Same as other
covered services
Preventive Care
Stop-loss limit
100% covered
$500 single
$750 2-person
$1,000 family
School of Public Health
None
$15 office visit co-pay
$100 inpatient co-pay
$10 generic
$20 formulary brand
$30 non-formulary brand
100% covered
$1,500 person (POS)
$3,000 family (POS)
$1,000 person (PPO)
$2,000 family (PPO)
Empirical Model – Step 1
• Predict employee’s 2000 spending region
on the basis of cohort, contract-level, and
employee demographic data
– Cohort stands in for unmeasured variables
that affect spending
– Control for health status using indicators for
34 ‘adjusted diagnostic groups’ (Starfield and
Weiner, 1991)
School of Public Health
2000 Cost Model
VARIABLE COEFFICIENT
INCOME
1.85E-07
AGE
0.01505
MALE
-0.16611
#LIVES
0.05793
FSA
0.10172
CDHP
-0.23575
PPO
-0.11845
R-SQUARE = .64
STANDARD
ERROR
1.59E-07
0.00262
0.04602
0.02137
0.053
0.06433
0.04877
tVALUE
1.16
5.75
-3.61
2.71
1.92
-3.66
-2.43
Pr > t
0.246
<.0001
0.0003
0.0068
0.0551
0.0003
0.0152
Dependent variable = ln(2000 total medical
expenditure in contract); regression controls
for 34 ADG categories
School of Public Health
Predicted 2000 Spending Regions
by Cohort
NUMBER PROBABILITY OF
COHORT of OBS.
REGION
CDHP
429
POS
1,249
PPO
1,025
1
2
3
1
2
3
1
2
3
School of Public Health
0.548
0.118
0.333
0.473
0.126
0.401
0.465
0.135
0.400
2001-2003 Cost Models
• We estimated 2-part models for total $,
physician $, Rx $, and proportion of Rx $
on brand-name drugs
• 1st part = probit analysis of any $
• 2nd part = ln($  $>0)
• Models include predicted region x Cohort
• Will present ‘key’ results
School of Public Health
Total Expenditure
PROBIT
VARIABLE
COEF.
POS x
REGION2
POS x
REGION3
CDHP x
REGION1
CDHP x
REGION2
CDHP x
REGION3
0.6373
SE
CHISQUARE
0.2808
5.1499
Pr > CHISQUARE
0.0232
CONDITIONAL ln(TOTAL
EXPENDITURE)
COEFF.
SE
t- Pr > t
VALUE
0.42986
0.07023
6.12 <.0001
1.1411
0.28
16.6112
<.0001
0.65593
0.04124
15.91 <.0001
-0.2248
0.1067
4.4411
0.0351
-0.11645
0.05238
-2.22
NA
NA
NA
NA
0.58771
0.12028
4.89 <.0001
NA
NA
NA
NA
0.76523
0.06473
11.82 <.0001
Regressions control for year, age, male, income, covered lives,
FSA use, concurrent ‘health shock’; omitted category = POS x
REGION1; NA = ‘not applicable’ because all observations = 1
School of Public Health
0.0262
Physician Expenditure
PROBIT
VARIABLE
POS x
REGION2
POS x
REGION3
CDHP x
REGION1
CDHP x
REGION2
CDHP x
REGION3
COEF.
SE
CHISQUARE
Pr > CHISQUARE
0.2155
0.2096
1.0575
0.3038
1.2256
0.2759
19.7412
-0.3139
0.1
9.8515
NA
NA
3.8598
83.4919
CONDITIONAL ln(PHYSICIAN
EXPENDITURE)
COEFF.
SE
t- Pr > t
VALUE
0.33135
0.062
5.34 <.0001
0.56323
0.03625
15.54 <.0001
0.0017
-0.02513
0.04642
-0.54
NA
NA
0.5407
0.1056
5.12 <.0001
0.0021
0.9631
0.67332
0.0569
11.83 <.0001
<.0001
Regressions control for year, age, male, income, covered lives,
FSA use, concurrent ‘health shock’; omitted category = POS x
REGION1; NA = ‘not applicable’ because all observations = 1
School of Public Health
0.5883
Rx Expenditure
PROBIT
VARIABLE
POS x
REGION2
POS x
REGION3
CDHP x
REGION1
CDHP x
REGION2
CDHP x
REGION3
Pr > CHISQUARE
CONDITIONAL ln(PHARMACY
EXPENDITURE)
COEFF.
SE
t- Pr > t
VALUE
COEF.
SE
CHISQUARE
0.6052
0.1467
17.0323
<.0001
0.4581
0.09006
5.09 <.0001
0.809
0.0978
68.4763
<.0001
0.74921
0.05297
14.14 <.0001
-0.2011
0.0714
7.9363
0.0048
-0.35918
0.07034
-5.11 <.0001
1.2198
0.4054
9.0515
0.0026
0.23713
0.1518
0.4822
0.1516
10.1168
0.0015
0.66084
0.08266
1.56
7.99 <.0001
Regressions control for year, age, male, income, covered
lives, FSA use, concurrent ‘health shock’; omitted category
= POS x REGION1
School of Public Health
0.1183
Brand Name Rx Proportion
VARIABLE
POS x
REGION2
POS x
REGION3
CDHP x
REGION1
CDHP x
REGION2
CDHP x
REGION3
Pr > t
COEFFICIENT
SE
t-VALUE
0.07377
0.01747
4.22
<.0001
0.02545
0.01028
2.48
0.0133
0.07243
0.01365
5.31
<.0001
0.15826
0.02945
5.37
<.0001
0.11147
0.01604
6.95
<.0001
Regressions control for year, age, male, income, covered
lives, FSA use, concurrent ‘health shock’; omitted category
= POS x REGION1
School of Public Health
Summary of Empirical Findings (1)
• CDHP enrollees predicted to be ‘low spenders’
consistently spent less in following years than a
comparison group with conventional cost
sharing
– This difference was found in all probit equations and
for cases with positive total expenditure and Rx
expenditure
• This finding is striking because CDHP enrollees
had no cost-sharing in this region
– HRA account provides insurance against future
expenses
School of Public Health
Summary (2)
• CDHP enrollees predicted to be in Region 2 or 3
spent more than the comparison POS group
– This finding is similar to our previous cohort study in
2001 and 2002 (PFC, 2004)
– CHDP enrollees in Region 3 have used their accounts
and face no cost-sharing at the margin  no incentive
to conserve on medical care
• The maximum out-of-pocket limit is too low
– Problem could be addressed by raising the limit and
introducing modest coinsurance above the limit
School of Public Health
Summary (3)
• Tiered pricing steers POS enrollees away
from brand name prescription drugs
• More results will be forthcoming (Parente,
Feldman, and Song, 2006)
School of Public Health
Final Thoughts
• Both CDHP and deductible plans
represent return to ‘demand side costsharing’ in US health care
• Is consumer directed health care just
‘deductible-lite’?
• Our models suggest that CDHP is
welfare-inferior to deductible plan because
it imposes an additional constraint on
enrollee
– But we ignored the tax subsidy for HRA
School of Public Health