Effects of Part D on Medicare Part A/B
Concentrating on First Year
Authors: Melvin J. Ingber, Ph.D., Leslie Greenwald, Ph.D.,
Sara Freeman, M.S., Deborah Healy, Ph.D.
AcademyHealth Research Meeting
June 28, 2010
P j t ffunded
Project
d db
by th
the C
Centers
t
ffor M
Medicare
di
&M
Medicaid
di id S
Services
i
www.rti.org
RTI International is a trade name of Research Triangle Institute
Aspects of Study
• Descriptive Statistics - highlights to be presented
• Time series study of Part A/B spending and
utilization starting 2005
utilization,
• Impact analysis on A/B spending and utilization using
difference-in-difference
difference
in difference method, 2005
2005-2006
2006
• Caveats:
– 2006 had an extended enrollment period. Most Low-Income
Subsidy (LIS) enrollees were automatically enrolled at the
start. Enrollment for others ramped up through June.
– The study did not account for actual purchase or
consumption of drugs.
www.rti.org
June 28, 2010
2
Selected Statistics
• July 2006 sample 40,476,167 beneficiaries
–
86 % Fee-for-Service (FFS), of which
•
•
•
•
–
12 % Medicare Advantage (MA) HMO, PPO…, of which
•
•
•
•
–
www.rti.org
46% iin P
Partt D plan
l (PDP) – 50% LIS
LIS, 50% non-LIS
LIS
19% have Retiree Drug Subsidy plan (RDS)
16% have other known source (e.g., VA, TRICARE, FEHB) only
19% have no known coverage
94 % in a Part D drug plan (MA-PD) – 23% LIS, 77% non-LIS
<1% in RDS
2% have other known source (e.g., VA, TRICARE, FEHB) only
g
4% have no known coverage
2 % Private Fee-for-Service (PFFS).
June 28, 2010
3
Selected Statistics
Health Status of Enrollees – RxHCC risk scores*
scores for
2006, based on 2005 diagnoses, FFS mean=1
– FFS
•
•
•
•
•
1.033
1.169
1.031
0.925
0.914
non-LIS
LIS
RDS
Other known source (e.g., VA, TRICARE, FEHB) only
no known coverage
– MA (HMO, PPO …)
•
•
•
•
•
0.948
1.110
0.998
0 947
0.947
0.912
non-LIS
LIS
RDS
Oth known
Other
k
source (e.g.,
(
VA,
VA TRICARE,
TRICARE FEHB) only
l
no known coverage
*diagnosis-based
g
risk scores used to adjust
j
p
payment
y
for Part D p
plans
www.rti.org
June 28, 2010
4
Time Series FFS Medicare Spending/Utilization
• 5% Sample
• Tracked by month
– Total
T t l spending
di
– Inpatient spending and stays
• For
–
–
–
–
All fee-for-service beneficiaries
Enrollees/nonenrollees
Non-LIS subsample
CHF subsample
• Years tracked: 2005 – 2007
• Selected results follow
www.rti.org
June 28, 2010
5
Time Series - Inpatient use
• Inpatient use likely to be sensitive to introduction of
Part D
– Expenditures were adjusted for fiscal year updates
– Stays are unaffected by updates
– Pattern of use has high monthly variability with peaks in
February or March of each year
– No discernable trend is visible
www.rti.org
June 28, 2010
6
Time Series – Inpatient - FFS
Per capita monthly expenditures and stays/1000 benes
$300
35
$290
34
33
$280
32
$
$270
31
$260
30
$250
29
28
$240
$230
Inpatient Expenditures
Inpatient Stays
26
25
Ja
n0
M 5
ar
-0
M 5
ay
-0
5
Ju
l- 0
Se 5
p0
N 5
ov
-0
Ja 5
n0
M 6
ar
-0
M 6
ay
-0
6
Ju
l- 0
Se 6
p0
N 6
ov
-0
Ja 6
n0
M 7
ar
-0
M 7
ay
-0
7
Ju
l- 0
Se 7
p0
N 7
ov
-0
7
$220
27
www.rti.org
June 28, 2010
7
Time Series: Inpatient Stays, non
non-LIS,
LIS, FFS
www.rti.org
Ratio of Inpatient Stays: non-LIS Part D Enrollees/non-Enrollees
1 34
1.34
1.32
1.3
1.28
1.26
1.24
1.22
June 28, 2010
8
07
N
ov
-
7
Se
p0
7
Ju
l-0
M
ay
-0
7
7
M
ar
-0
Ja
n07
06
N
ov
-
6
Se
p0
6
Ju
l-0
M
ay
-0
6
M
ar
-0
6
1.2
Ja
n06
• Part D
enrollees
h
have
hi
higher
h
use than
nonenrollees
• LIS enrollees
have even
higher use
than non-LIS
displayed
here
Time Series: Regression of Stays on Time
andd Enrollment
E ll
t
Variable
Definition
D_ENROLL
Enrollee=1, nonenrollee=0
PARTD
Jun 06 – Dec 07 = 1
Post phase-in period
T
Month counter for trend. 1 - 24
PARTD_T
Interaction to shift trend, post
phase-in
PARTD_ENROLL_T
Interaction to shift trend for
enrollees post phase-in
www.rti.org
June 28, 2010
9
Time Series: Regression 2006
2006-2007
2007
Dep Var: Inpatient stays
Dep.
stays, Enrollees and Nonenrollees,
Nonenrollees 24 months
Variable
Coefficient
Intercept
27.79**
(1 14)
(1.14)
D_ENROLL
15.02**
(.761)
Enrollees have higher use on avg.
PARTD
−0.50
0 50
(.324)
Part D period
P
i d slightly
li h l llower, nonsignificant
T
−4.20**
(1 28)
(1.28)
General downward trend for all
PARTD_T
0.473
(.328)
Slightly positive increase in trend in full
Part D period, nonsignificant
PARTD_ENROLL_T
−0.060
(.054)
Tiny negative adjustment to trend for
enrollees, nonsignificant
** = 5% ssignificance,
g ca ce, * = 10%
0%
www.rti.org
Meaning
N=48
N
48, R-sq
R sq = 0.96
0 96
June 28, 2010
10
Time Series – Enrollees, LIS vs. non
non-LIS
LIS
•
•
There was no discernible change in inpatient use comparing all
enrollees to nonenrollees
p
utilization,, about 80% g
greater
LIS enrollees differ in inpatient
Stays per 1000 Beneficiaries: Ratio of LIS to non-LIS Enrollees
•
•
07
vNo
07
pSe
7
Ju
l-0
7
ay
-0
M
7
ar
-0
M
07
Ja
n-
06
vNo
06
pSe
6
Ju
l-0
6
ay
-0
M
6
ar
-0
M
Ja
n-
06
1.9
1.8
1.7
LIS enrollees had little to no change in drug coverage from 2005
because almost all were covered by Medicaid
Analysis
y then focused on non-LIS,, voluntarilyy enrolled,, nonMedicaid
population, who more likely had a change in coverage.
– Non-LIS enrollment went from 13% of FFS Part D in Jan. 2006 to 30% in
Dec. 2006 and was relatively stable through 2007
www.rti.org
June 28, 2010
11
Time Series Regression 2006
2006-2007
2007
D
Dep.
V
Var: IInpatient
ti t stays,
t
Non-LIS
N
LIS Enrollees
E
ll
and
d Nonenrollees,
N
ll
24 months
th
Variable
Coefficient
p
Intercept
25.94**
(1.01)
D_ENROLL
6.79**
(0.677)
Enrollees in Part D have higher use on
avg.
PARTD
−2.35**
(1.13)
Part D period slightly lower for all,
significant
T
−0.21
(0.287)
General downward trend,, nonsignificant
PARTD_T
0.17
(0.291)
Slightly increase in trend in full Part D
period, nonsignificant
PARTD_ENROLL_T
−0.02
(0.047)
Tiny negative adjustment to trend for
enrollees, nonsignificant
** = 5% significance
significance, * = 10%
www.rti.org
Meaning
N=48 R-sq = 0
N=48,
0.89
89
June 28, 2010
12
Difference-in-Difference
Difference
in Difference Method
• In these and other analyses, simple time series does
not reveal a Part D effect on Part A/B spending.
• A difference-in-difference, pre-post, approach was
also used, with many control variables, to determine
if there was an effect for Part D enrollees in 2006
2006.
• Only non-LIS enrollees were used, as they were most
likely to have had a coverage change
change.
• Panels of FFS beneficiaries and cross-sections of
beneficiaries were used in the analyses.
• Data from 2004, 2005 and 2006 were used.
www.rti.org
June 28, 2010
13
Difference-in-Difference
Difference
in Difference Method
• Dependent variables were total spending and
inpatient spending.
• Decedents were included or excluded in different
samples.
• Control
C t l variables
i bl iincluded:
l d d d
demographics,
hi
h
health
lth
status variables and statuses such as ESRD and
long-term institutionalized.
institutionalized
• Markers for Enrollment in Part D and other insurance
were included, as were indication of enrollment
duration.
www.rti.org
June 28, 2010
14
Difference-in-Difference Method – Non-LIS
P l no ddecedents,
Panel,
d t S
Selected
l t d Ch
Characteristics
t i ti
Sample has full year of eligibility in 2004 and live through
2006 to assure full health status measures
Characteristics
Sample total
Number (2005)
Number (2006)
935 603
935,603
935 603
935,603
% Change
Age < 65, 6.5%
60,356
56,225 -4.7%
Age ≥65, originally eligible by
disability, 5.1%
47,426
51,500 9%
Long-term care institutionalized, 0.4%
3,793
6,287 66%
Dialysis or transplant status,
status 0.3%
0 3%
2 946
2,946
3 655 24%
3,655
Post-graft status, 0.1%
1,171
1,257 7%
www.rti.org
June 28, 2010
15
Difference-in-Difference Method – Non-LIS
P l no ddecedents,
Panel,
d t S
Selected
l t d Ch
Characteristics
t i ti
Medicare spending
Physician/supplier
Mean ($)
(2005)
Mean ($)
(2006)
% Change
1,888
2,123 12%
98
109 11%
175
212 21%
2,201
2,608 18%
Skilled Nursing Facility
270
434 60%
Hospital Outpatient
762
881 16%
Home Health
253
347 37%
41
121 197%
Laboratory
Durable Medical Equipment
Inpatient
Hospice
Total
www.rti.org
5,689
June 28, 2010
6,835 20%
16
Difference-in-Difference Method – Non-LIS
P l no ddecedents,
Panel,
d t S
Selected
l t d Ch
Characteristics
t i ti
Drug coverage
Number
(2006)
%
Sample
Drug coverage
Number
(2006)
%
Sample
Creditable
coverage
165,717 17.7%
Month 7
55,770 6.0%
Retiree Drug
Subsidy
278,676 29.8%
Month 8
35,509 3.8%
Month 9
23,453 2.5%
M th 10
Month
20 214 2.2%
20,214
2 2%
Month 11
36,190 3.9%
Month 12
141,692 15.1%
Total Part D
coverage
322 091 34.4%
322,091
34 4%
Part D coverage
Months
Month 1
663 0.1%
Month 2
569 0.1%
Month 3
649 0.1%
Month 4
5,597 0.6%
Month 5
794 0.1%
o t 6
Month
991 0.1%
www.rti.org
June 28, 2010
17
Difference-in-Difference
Difference
in Difference Method - Model
Spending = (a1 × demog1 + a2 × demog2 + ….) × Year_2005
+ (a1 × demog1 + a2 × demog2 + …) × Year_2006 × md
+ (b1 × HCC1 + b2 × HCC2 + …)) × Year_2005
Y
2005
+ (b1 × HCC1 + b2 × HCC2 + …) × Year_2006 × mh
+ other clinical variables in similar form
+ (lti1 × long-term institutionalized) × Year_2005
+ (lti1 × long-term institutionalized) × Year_2006 × mlti
+ t1 × Year_2006
+ e1 × Part_D_enrollee
+ d1 × Part_D_enrollee
P t D
ll × Year_2006
Y
2006
+ f1 × Other drug coverage variables
www.rti.org
June 28, 2010
18
Difference-in-Difference Method – Total spending
P tDC
Part
Coefficients
ffi i t for
f P
Panel,l no ddecedents
d t
Months enrolled in Part D.
Variable marks person in
both 2005 and 2006
Estimate
Approximate
Standard
Error
t Value
Approximate
Pr > |t|
1 month
1,621
480.30
3.38
0.0007
2 months
1,084
518.40
2.09
0.0365
3 months
2,265
485.50
4.66
<.0001
4 months
th
479
167 40
167.40
2 86
2.86
0 0042
0.0042
5 months
1,477
439.00
3.37
0.0008
6 months
1,001
393.10
2.55
0.0109
7 months
631
58.56
10.77
<.0001
8 months
612
70.67
8.66
<.0001
9 months
,
1,101
84.83
12.98
<.0001
10 months
1,290
90.82
14.21
<.0001
11 months
1,313
70.15
18.71
<.0001
12 months
1 061
1,061
42 37
42.37
25 04
25.04
< 0001
<.0001
www.rti.org
June 28, 2010
19
Difference-in-Difference Method – Total spending
P tDC
Part
Coefficients
ffi i t for
f P
Panel,l no ddecedents
d t
Months enrolled in Part D.
Variable marks person in 2006
1 month
Estimate
Approximate
Standard
Error
t Value
Approximate Pr
> |t|
−11,230
230
679 20
679.20
−1
1.81
81
0 0702
0.0702
2 months
349
733.10
0.48
0.6342
3 months
−747
686.60
−1.09
0.2768
4 months
h
202
236 60
236.60
0 85
0.85
0 3935
0.3935
5 months
300
620.80
0.48
0.6285
6 months
87
555.90
0.16
0.8756
7 months
258
82.78
3.12
0.0018
8 months
132
99.88
1.32
0.1853
9 months
−5
119.90
−0.04
0.9682
10 months
−385
128.40
−3
0.0027
11 months
−528
99.16
−5.33
<.0001
12 months
−317
59.77
−5.31
<.0001
www.rti.org
June 28, 2010
20
Difference-in-Difference Method – Non-LIS
C
Cross-Section,
S ti no decedents,
d d t Ch
Characteristics
t i ti
Sample has full year of eligibility in 2004 for 2005
observations and full year of 2005 eligibility for 2006
observations to assure full health status measures
Characteristics
Sample total
Number (2005)
1 120 613
1,120,613
Number (2006)
% Change
1 086 892 −3%
1,086,892
3%
Age < 65, 8.2%
91,377
89,483 -4.7%
Age ≥65, originally eligible by
disability, 5.7%
63,517
61,537 -3%
Long-term care institutionalized, 1.4%
16239
15528 -4.4%
Dialysis or transplant status,
status 0.5%
0 5%
5049
5138 1.8%
1 8%
Post-graft status, 0.1%
1571
1657 5.5%
www.rti.org
June 28, 2010
21
Difference-in-Difference Method – Non-LIS
C
Cross-Section,
S ti no decedents,
d d t Ch
Characteristics
t i ti
Medicare spending
Physician/supplier
Mean ($) (2005)
Mean ($) (2006)
1,996
2,094 5%
Laboratory
101
109 8%
Durable Medical Equipment
210
222 5%
2,617
2,660 2%
Inpatient
% Change
Skilled Nursing Facility
419
466 11%
Hospital Outpatient
853
913 7%
Home Health
318
347 9%
Hospice
106
127 20%
Total
www.rti.org
6,620
June 28, 2010
6,938 5%
22
Difference-in-Difference Method – Non-LIS
C
Cross-Section,
S ti no decedents,
d d t Ch
Characteristics
t i ti
Drug coverage
Number
(2006)
%
Sample
Drug coverage
Number
(2006)
%
Sample
Creditable
coverage
179,940 16.6%
Month 7
64,202 5.9 %
Retiree Drug
Subsidy
301,846 27.8 %
Month 8
52,954 4.9 %
Month 9
28,461 2.6 %
M th 10
Month
25 504 2.3
25,504
23%
Month 11
45,651 4.2 %
Month 12
188,516 17.3 %
Total Part D
coverage
420 677 38.7
420,677
38 7 %
Part D coverage
Months
Month 1
1,535 0.1%
Month 2
1,096 0.1%
Month 3
1,423 0.1%
Month 4
7,474 0.7%
Month 5
2,009 0.1%
o t 6
Month
1,852 0.1%
www.rti.org
June 28, 2010
23
Difference-in-Difference Method – Total spending
P t D Coefficients
Part
C ffi i t for
f X-Sect,
X S t no decedents
d d t
Months enrolled in Part D.
Variable marks person in both
2005 and 2006
Estimate
Approximate
Standard
Error
Approximate Pr
> |t|
t Value
1 month
5 862
5,862
213 60
213.60
27 44
27.44
< 0001
<.0001
2 months
5,265
220.10
23.92
<.0001
3 months
5,088
208.60
24.39
<.0001
4 months
h
1 087
1,087
135 90
135.90
8
<.0001
0001
5 months
2,935
200.00
14.67
<.0001
6 months
2,695
197.80
13.62
<.0001
7 months
−476
57.82
−8.24
<.0001
8 months
−627
62.28
−10.07
<.0001
9 months
−71
78.30
−0.9
0.3666
10 months
143
81.79
1.75
0.08
11 months
140
65.61
2.13
0.0334
12 months
49
40.95
1.2
0.2318
www.rti.org
June 28, 2010
24
Difference-in-Difference Method – Total spending
P t D Coefficients
Part
C ffi i t for
f X-Sect,
X S t no decedents
d d t
Months enrolled in Part D. Variable
marks person in 2006 only.
1 month
Estimate
Approximate
Standard
Error
t Value
Approximate Pr
> |t|
−22,179
179
403 30
403.30
−5
5.44
< 0001
<.0001
2 months
−573
460.60
−1.24
0.2133
3 months
373
412.00
0.91
0.3649
4 months
h
317
207 70
207.70
1 53
1.53
0 1266
0.1266
5 months
1,050
360.00
2.92
0.0035
6 months
504
369.10
1.37
0.1721
7 months
1,475
82.83
17.81
<.0001
8 months
1,292
89.38
14.46
<.0001
9 months
1,283
114.60
11.19
<.0001
10 months
959
120.20
7.98
<.0001
11 months
772
94.68
8.15
<.0001
12 months
817
58.35
14.01
<.0001
www.rti.org
June 28, 2010
25
Summary
• All analyses show selection bias with Part D
enrollees sicker, even for non-LIS beneficiaries.
• The time series analyses do not show indications of
Part A/B utilization trends or levels changing with
implementation
p
of Part D.
• The difference-in-difference analysis shows some
spending decrease for longer enrollment durations
when a panel
panel, with or without decedents
decedents, is studied
studied.
• The difference-in-difference analysis shows some
spending increase when two cross-sections
cross sections are
studied.
www.rti.org
June 28, 2010
26