All Office Visits Per Quarter

chocolatehookΑσφάλεια

30 Νοε 2013 (πριν από 3 χρόνια και 4 μήνες)

65 εμφανίσεις


1

Supplemental Digital Content 1
:
Prope
nsity Score Matching Approach


E
m
ployer
-
level selection bias
was

an important consideration for this study. To reduce this
potential bias, we included baseline costs, a proxy for use, in the propensity score matching
procedure.
Consequently, in our matched sample,
high
-
utilizing
HDHP members

are matched
to similar
high
-
utilizing HMO members
.

To further investigate this potential threat to validity, we
examined baseline trends in utilization for the main outcomes, below, and found the study
groups followed simil
ar baseline trajectories in use,
indicating the central

underlying assumption
of the difference
-
in
-
differences framework was met.


Our propensity score model included the following variables: age, sex, family or individual plan
status, health status (Adjusted Clinical Groups score
),
employer size, neighborhoo
d
socioeconomic characteristics, baseline copay levels, total member health plan costs

at
baseline
, and secular changes
.
We performed an exact match on the characteristic of having
one

of

four chronic diseases to ensure adequate representation of these gro
ups.
We also

performed
an exact match on

the

variable
of
association plan status
because

these very small
employers purchase health plans in a different manner (through small business associations
rather than directly from Harvard Pilgrim) and we have slig
htly less information on enrollees
(such as whether they had access to Health Reimbursement Arrangements).


We used propensity score matching to generate a similar comparison group,
although

we found
that residual baseline differences between the cohorts remained (Table 1).
To investigate
whether controlling for residual differences generated more valid effect estimates compared to
unadjusted estimates, w
e
ran adjusted and unadjusted models

using
Poisson regression.
(
This
approach of using propensity score balancing followed by regression adjustment has been
promoted by Robins and collaborators as protecti
ng

against residual baseline differences due to
imperfections in the propensity score m
odel.
)


However, w
e found that results were nearly
identical and that interpretation did not differ. We therefore present t
he unadjusted results
in the
paper
(Table 2).



-0
.2

0
0
.2

0
.4

0
.6

0
.8

1
1
.2

1
.4

1
2
3
4
5
6
7
8
V
i
s
i
t
s

/

M
e
m
b
e
r

/

Q
u
a
r
t
e
r

Quar
t
er
A
ll O
f
f
i
c
e
V
i
s
i
t
s
P
e
r
Q
u
a
r
t
e
r

H
DH
P
(
n=
7
9
5
3
)

C
o
nt
r
o
ls
(
n=
7
9
5
3
)

Di
f
f
e
r
e
nc
e



E
mp
lo
ye
r
-ma
nd
a
t
e
d

s
w
i
t
c
h t
o
H
DH
P




2








-0
.
1

0
0.
1
0
.2

0
.3

1
2
3
4
5
6
7
8
V
i
s
i
t
s

/

M
e
m
b
e
r

/

Q
u
a
r
t
e
r

Quar
t
er
A
c
u
t
e
H
i
g
he
r
-P
r
i
o
r
i
t
y V
i
s
i
t
s
P
e
r
Q
u
a
r
t
e
r

H
DH
P
(
n=
7
9
5
3
)

C
o
nt
r
o
ls
(
n=
7
9
5
3
)

Di
f
f
e
r
e
nc
e



E
mp
lo
ye
r
-ma
nd
a
t
e
d

s
w
i
t
c
h t
o
H
DH
P



-0
.
1

0
0.
1
0
.2

0
.3

1
2
3
4
5
6
7
8
V
i
s
i
t
s

/

M
e
m
b
e
r

/

Q
u
a
r
t
e
r

Quar
t
er
A
c
u
t
e
L
o
w
e
r
-P
r
i
o
r
i
t
y V
i
s
i
t
s
P
e
r
Q
u
a
r
t
e
r

H
DH
P
(
n=
7
9
5
3
)

C
o
nt
r
o
ls
(
n=
7
9
5
3
)

Di
f
f
e
r
e
nc
e





E
mp
lo
ye
r
-ma
nd
a
t
e
d

s
w
i
t
c
h t
o
H
DH
P




3











-0
.
1

0
0.
1
0
.2

0
.3

1
2
3
4
5
6
7
8
V
i
s
i
t
s

/

M
e
m
b
e
r

/

Q
u
a
r
t
e
r

Quar
t
er
C
hr
o
ni
c
H
i
g
he
r
-P
r
i
o
r
i
t
y V
i
s
i
t
s
P
e
r
Q
u
a
r
t
e
r

H
DH
P
(
n=
7
9
5
3
)

C
o
nt
r
o
ls
(
n=
7
9
5
3
)

Di
f
f
e
r
e
nc
e



E
mp
lo
ye
r
-ma
nd
a
t
e
d

s
w
i
t
c
h t
o
H
DH
P



-0
.
1

0
0.
1
0
.2

0
.3

1
2
3
4
5
6
7
8
V
i
s
i
t
s

/

M
e
m
b
e
r

/

Q
u
a
r
t
e
r

Quar
t
er
C
hr
o
ni
c
L
o
w
e
r
-P
r
i
o
r
i
t
y V
i
s
i
t
s
P
e
r
Q
u
a
r
t
e
r

H
DH
P
(
n=
7
9
5
3
)

C
o
nt
r
o
ls
(
n=
7
9
5
3
)

Di
f
f
e
r
e
nc
e

E
mp
lo
ye
r
-ma
nd
a
t
e
d

s
w
i
t
c
h t
o
H
DH
P



-0
.2

0
0
.2

0
.4

0
.6

0
.8

1
1
.2

1
.4

1
2
3
4
5
6
7
8
T
e
s
t
s

/

M
e
m
b
e
r

/

Q
u
a
r
t
e
r

Quar
t
er
P
r
e
v
e
nt
i
v
e
L
a
b
T
e
s
t
s
P
e
r
Q
u
a
r
t
e
r

H
DH
P
(
n=
7
9
5
3
)

C
o
nt
r
o
ls
(
n=
7
9
5
3
)

Di
f
f
e
r
e
nc
e



E
mp
lo
ye
r
-ma
nd
a
t
e
d

s
w
i
t
c
h t
o
H
DH
P



-0
.2

0
0
.2

0
.4

0
.6

0
.8

1
1
.2

1
.4

1
2
3
4
5
6
7
8
T
e
s
t
s

/

M
e
m
b
e
r

/

Q
u
a
r
t
e
r

Quar
t
er
Ge
ne
r
a
l L
a
b
T
e
s
t
s
P
e
r
Q
u
a
r
t
e
r

H
DH
P
(
n=
7
9
5
3
)

C
o
nt
r
o
ls
(
n=
7
9
5
3
)

Di
f
f
e
r
e
nc
e



E
mp
lo
ye
r
-ma
nd
a
t
e
d

s
w
i
t
c
h t
o
H
DH
P




4










-0
.2

0
0
.2

0
.4

0
.6

0
.8

1
1
.2

1
.4

1
2
3
4
5
6
7
8
T
e
s
t
s

/

M
e
m
b
e
r

/

Q
u
a
r
t
e
r

Quar
t
er
P
r
e
v
e
nt
i
v
e
L
a
b
T
e
s
t
s
P
e
r
Q
u
a
r
t
e
r

H
DH
P
(
n=
7
9
5
3
)

C
o
nt
r
o
ls
(
n=
7
9
5
3
)

Di
f
f
e
r
e
nc
e



E
mp
lo
ye
r
-ma
nd
a
t
e
d

s
w
i
t
c
h t
o
H
DH
P



-0
.2

0
0
.2

0
.4

0
.6

0
.8

1
1
.2

1
.4

1
2
3
4
5
6
7
8
T
e
s
t
s

/

M
e
m
b
e
r

/

Q
u
a
r
t
e
r

Quar
t
er
Ge
ne
r
a
l L
a
b
T
e
s
t
s
P
e
r
Q
u
a
r
t
e
r

H
DH
P
(
n=
7
9
5
3
)

C
o
nt
r
o
ls
(
n=
7
9
5
3
)

Di
f
f
e
r
e
nc
e



E
mp
lo
ye
r
-ma
nd
a
t
e
d

s
w
i
t
c
h t
o
H
DH
P




5










References

Bang H, Robins JM.
Doubly robust estimation
in missing data and causal inference models.

Biometrics
. 2005 Dec;61(4):962
-
73.


-0
.2

0
0
.2

0
.4

0
.6

0
.8

1
1
.2

1
.4

1
2
3
4
5
6
7
8
T
e
s
t
s

/

M
e
m
b
e
r

/

Q
u
a
r
t
e
r

Quar
t
er
R
a
d
i
o
lo
g
y T
e
s
t
s
P
e
r
Q
u
a
r
t
e
r

H
DH
P
(
n=
7
9
5
3
)

C
o
nt
r
o
ls
(
n=
7
9
5
3
)

Di
f
f
e
r
e
nc
e



E
mp
lo
ye
r
-ma
nd
a
t
e
d

s
w
i
t
c
h t
o
H
DH
P