Transcript Document

The effect of increases to
concession card income eligibility
thresholds on pharmaceutical
consumption
Work in Progress
Peter Siminski, PhD student, UNSW
Presented at CAER Summer Workshop in Health
Economics 1st Feb 2007
Outline
 A ‘natural experiment’ – change in eligibility
rules for Commonwealth Seniors Health Card
(CSHC)
 Methods
– Diff-in-diff to measure effect on consumption
using National Health Surveys
– Average price decreased by at least 66%
 Results: very low estimated price elasticity
Commonwealth Seniors Health Card (CSHC)
 CSHC -> concession price for PBS drugs for
people of age pension age
 Income eligibility threshold for CSHC almost
doubled in 1999
 What was impact on PBS consumption for the
affected?
CSHC income eligibility threshold, couples $2001 $’000s p.a.
Jan-06
Jan-05
Jan-04
Jan-03
Jan-02
Jan-01
Jan-00
Jan-99
Jan-98
Jan-97
Jan-96
Jan-95
90
80
70
60
50
40
30
20
10
0
Jan-06
Jan-05
Jan-04
Jan-03
Jan-02
Jan-01
Jan-00
Jan-99
Jan-98
Jan-97
Jan-96
Jan-95
A ‘Natural Experiment’
90
80
70
60
50
40
30
20
10
0
Treatment group
A ‘Natural Experiment’
90
80
70
60
50
40
30
20
10
0
Treatment group
Jan-06
Jan-05
Jan-04
Jan-03
Jan-02
Jan-01
Jan-00
Jan-99
Jan-98
Jan-97
Jan-96
Jan-95
Comparison group
A ‘Natural Experiment’
90
80
70
60
50
40
30
20
10
0
Comparison group 2
Treatment group
Jan-06
Jan-05
Jan-04
Jan-03
Jan-02
Jan-01
Jan-00
Jan-99
Jan-98
Jan-97
Jan-96
Jan-95
Comparison group
& 4th comp
group
Income Thresholds Used in the Analysis
Couples
2000
1800
1600
1400
1200
1000
800
600
400
200
0
Singles
1200
1000
Comp Gp 2
Treat Gp
800
600
Comp Gp
400
200
0
actual
1995
proxy
2001 2004-05
proxy
proxy
actual
1995
proxy
2001
proxy
2004-05
proxy
Sample Size
1995
2001
2004-05
Total
166
189
203
558
Comp Gp 1 (low income)
3897
2205
2995
9097
Comp Gp 2 (high income)
111
64
92
267
Comp Gp 3 (M 50-64; F 50-59)
5116
2588
3890
11594
Total
9290
5046
7180
21516
Treatment Group
A Complication – Treatment Group is Contaminated
100%
80%
60%
no card
card
40%
20%
0%
1995
2001
2004-05
Concession card coverage recorded in each year, but
CSHC not distinguishable from other cards
Contaminated Treatment Group
100%
80%
60%
40%
20%
0%
1995
2001
2004-05
no card '01 & '05
(contaminators)
no card in '01 & '05
(non-take up)
no card 1995; card
later
card in 1995 (safety
net)
card in 1995
(contaminators)
Accounting for the contamination
 Contaminators are present in each year and are
unaffected by the intervention
 If estimated treatment effect is a change of x%,
it can be shown that the effect on the affected is
x( p1 d1  p 2 d 2 )
p1 d1
Where p2 is the proportion of contaminators,
p1 =1-p2 and
d is mean PBS consumption
Dependent Variable
 Number of PBS drugs taken for selected
conditions in previous two weeks
 As reported by respondent
 Drug data collected inconsistently between
years
– NHS 2001 & 2004-05 recorded if taken for selected
conditions
– NHS 1995 recorded all drugs taken
– ABS classifies drugs into types commonly used for
specific conditions
 Generic drug names checked against PBS
Schedule
Dependent Variable (Cont.)
 Conditions for which drug data are available in
all three surveys
– Heart and circulatory conditions
– Asthma
– Diabetes and high sugar levels
 These account for 41% of all PBS drug
consumption in 2001
 NHS 1995 and 2004-05 also have data for
– Arthritis and osteoporosis
– Mental wellbeing
Mean PBS Consumption by Group and Year (excl 04-05)
1.4
1.2
1
0.8
0.6
0.4
0.2
0
'95
'01
Treatment group
'95
'01
Comparison group
(low income)
'95
'01
Comparison group
(high income)
'95
'01
Comparison group
(middle aged)
Distribution of Dependent Variable (all groups)
%
70
60
50
40
30
20
10
0
1995
2001
2004-05
0
1
2
3
4
5
6
7
8
number of PBS drugs taken
9 10
Distribution of Dependent Variable (Treatment Gp)
%
60
50
40
30
20
10
0
1995
2001
2004-05
0
1
2
3
4
5
6
number of PBS drugs taken
7
8
Modelling Approach
 Count data models
– Poisson regression model
  exp( 0  1Gr   2Yr   3 Int  X )
Where
  E ( y | Gr, Yr, Int, X )
– Negative Binomial regression model (NegBin)
~  exp( 0  1Gr   2Yr   3 Int  X   )
– Negative Binomial / Logit Hurdle model (not yet
implemented)
 Model each combination of year and
comparison group separately and then pooled
Goodness of Fit (NegBin)
%
70%
60%
50%
40%
Actual
30%
NegBin fitted
20%
10%
0%
0
1
2
3
4
5
6
7
8
9
10
number of PBS drugs taken
 Formally, NegBin model rejected by Pearson Chi-square
test (p < 0.001), suggesting misspecification
 Hurdle model might be more appropriate
Main results
 Negbin: 3 = -0.007 in pooled model
 -1% change in PBS consumption after adjusting
for contaminators (95% CI: -54%, 51%)
 Poisson: -2% change (95% CI: -53%, 49%)
Out-of-pocket Price of PBS Drugs ($1995)
5
10
15
20
25
30
35
40
45
50
55
60
65
70
75
80
85
90
95
100
105
$16
$14
$12
$10
$8
$6
$4
$2
$0
script number
Treatment group 1995
Treatment group 2001
What is the average marginal price change?
 Depends on distn of the number of scripts
purchased in calendar year prior to interview
 For treatment group average annual
consumption estimated to be 30.5 scripts
 Assume over-dispersed count variable
distribution (negative binomial)
 Consider a range of possible distributions
Possible Distributions (Treatment Group) – end of year
mean = 30.5
10%
8%
Thresholds
6%
4%
neg bin
poisson
82
73
64
55
46
37
28
19
10
1
2%
0%
Possible Distributions (Treatment Group) – mid-year
mean = 15.25
12%
10%
8%
6%
4%
2%
0%
Thresholds
neg bin
82
73
64
55
46
37
28
19
10
1
poisson
Possible Distns (Treatment Group) – throughout year
6%
5%
4%
thresholds
3%
neg bin
2%
poisson
1%
80
72
64
56
48
40
32
24
16
8
0
0%
Price Change (Cont.)
 A maximum of 12.8% of treatment group
experienced price increase
 Average marginal price fall was at least 66%
 Similarly, average marginal price increased by
at least
– 15% for comparison group 1
– 17% for comparison groups 2 & 3
 Modelled price change = -71%
or –119% using midpoint formula
Price Change (Cont.)
 Calculate price elasticity using midpoint formula
(C1  C0 )
( P1  P0 )
E

(C1  C0 ) 2 ( P1  P0 ) 2
 How important is the price change?
– Average person in treatment group purchases 30
scripts per year -> saves $360 p.a. = 1% income
– Person purchasing 10 times average saves $1083
p.a. = 3% income
– consider non-monetary costs + cost of seeing GP
Estimated Price Elasticity
1.6
1.4
1.2
1.0
0.8
0.6
0.4
0.2
0.0
-0.2
-0.4
-0.6
-0.8
-1.0
2001
2004-05
Comp 1
2001
2004-05
Comp 2
2001
2004-05
Comp 3
POOLED
Threats to Validity
 Quantity taken in fortnight might not be proportional
to drugs purchased. Price change could lead to:
– Changes in frequency or dosage of consumption
– Using out-of-date or substitute medications
– leads to underestimation of elasticity
 Endogeneity of Health Care Card Status
– People may reduce income to qualify for card, but unlikely
given low relative value of concession
– leads to overestimation of elasticity
 Possible overestimation of price change:
– PBS premiums not accounted for
– Non-monetary costs + costs of GP consultation
Conclusion
 Results are preliminary, next step is
implementation of hurdle model.
In the interim:
 Policy is efficient. Does not seem to induce
excess consumption
 Recipient value might be higher than govt cost
(on average) due to insurance value
 Middle income older people do not need
concessions to purchase sufficient quantity of
pharmaceuticals