The Case of Drug Expenditures in Taiwan Hospitals

advertisement
Shin-Yi Chou, Lehigh University
James A. Dearden, Lehigh University
Mary E. Deily, Lehigh University
Hsien-Ming Lien, National Cheng-Chi University
1

Two important but also complicated issues in NHI
 Drug expenditure
 Global budgeting system
2
3

Global budget
 a fixed maximum expenditure or a target, set by government, for a
defined set of health care services.

Intended purpose:
 Governments have control over health care expenditures and can
plan for revenue needs.

Widely implemented in European countries; several Asian
countries; and Veterans’ Administration in the U.S.
4

NHI was introduced in March 1995.
 Reimburses providers for most medical services and on a fee-for-
service basis

Adoption of GBS was required in the NHI Act, and was
designed to phase in by sectors.
 Dental services (1998)
 Traditional Chinese medical services (2000)
 Western-based medical clinics (July 2001)
 Hospital services (July 2002)
 Dialysis Treatment (2006)
5
Total Budget for Hospital Services
(Expenditure Cap)
Outpatient Care
(46.065%)
Inpatient Care
(45.502%)
Outpatient Dialysis
(5.722%)
Special Services
(2.711%)
Taipei Region
Northern Region
Central Region
Southern Region
Eastern Region
Kao-Ping Region
6

Expenditure target ~ “Soft cap”
 If actual total spending exceeds target, the spending gap is “shared”
between the health care providers and the insurer.
 The insurer shares some financial risk with health providers and
cannot control health spending perfectly

Expenditure cap ~ “Hard cap”
 If the actual spending is higher than target, the health care
providers are responsible for spending gap.
 The insurer bears no risks and controls health spending perfectly.

Taiwan adopts a Hard Cap GBS, with Sheltered Payments
7

Simple version:
 Ex post Service Point Value = GB/Total Service Points
▪ If service points are greater than expected, fixed budget does not
expand: instead, the point value decreases below 1; providers
reimbursed at a “discounted” rate

Version with sheltered payments
 Ex post Service Point Value to reimburse unsheltered services =
(GB – Payments for sheltered services valued at ex ante prices)
/(Total service points for unsheltered services)
▪ Payments for sheltered services are not at risk of being discounted
▪ Substituting drug for non-drug treatments reduces numerator and
denominator
8
$5000  $1 (500  1000) $3500
Floating Point Value 

 $0.7
(3000  2000)
5000
Hospital A receives $500  $0.7  3000  $2600
Hospital B receives $1000  $0.7  2000  $2400
9

Do hospitals increase “sheltered” drug expenditures by
substituting drug treatments for non-drug treatments?

This question is especially relevant because the share of
pharmaceutical expenditure as a percentage of total expenditure
is around 25% in Taiwan, much higher than that of OECD
countries.
 2010: Japan (20.3%) Korea (20.5%) U. S. (11.8%)
10
 Each condition can be treated with drug or non-drug treatments,
with associated service points: (α, β)
 Non-drug treatments for each condition vary in intensity,
represented as q(α, β), which is directly related to cost
 Hospitals choose the optimal cutoff for each condition:
• use non-drug treatments for patients requiring intensity lower
than cutoff point,
• use drug treatments for patients that would require more
intense non-drug treatment
 Derive the optimal cutoffs for different conditions under
• a FFS system, where all prices are known ex ante
• a GBS system, where only drug prices are known ex ante
11

We find a pure strategy Nash Equilibrium when hospitals are
reimbursed under the GBS

Shift from FFS to GBS causes hospitals to substitute drug for
non-drug treatments if the probability of lower ex post
reimbursements for (unsheltered) services increases and the
probability of higher ex post reimbursements decreases for more
conditions (Note: Sufficient but not necessary condition)

Expectation: GBS will cause expenditures on drug treatments
to increase and expenditures on non-drug treatments to decrease
12

Principal Data: longitudinal claims of a random sample of one
million NHI enrollees drawn from the 2005 claims data
 Extract those patients’ complete outpatient claims from 1997 to
2006.

Eliminate:
 Outpatient records for dental care, for traditional Chinese medical
treatments, or for patients treated at local clinics
 Patients receiving dialysis treatment and surgeries performed at
outpatient visits.
 Patients that suffered major illnesses.
13

Each claim has
 diagnoses of diseases, dates of treatment, the department that
provided the services
 detailed description of the hospital’s claim for reimbursement
 a unique identifier for the health provider

We aggregate the patient-level data to hospital department level

The final sample size is 411,840 observations of hospital
departments over the period 1997-2006 (though sample size
varies with dependent variable because we take logs)
14

GBS was implemented nationwide at once
 No obvious treatment and control groups

GBS has a larger impact on departments that have a higher
propensity to use drug treatments.

We use a department’s “drug ratio” as a basic measure of the
propensity to use drug treatments

A Department’s Drug Ratio in 2001
= Drug Expenditures/(Drug + Non-drug Expenditures)
15
0
50
100
150
200
250
Figure 2: Distribution of Drug Ratio in 2001
0
.2
.4
.6
Drug Ratio in 2001
.8
1
16
Average 2001 Drug Ratio by Department
Department Names
General
Orthopedics
Urology
Obstetrics and Gynecology
Ear, Nose and Throat
Neurosurgery
Gastroenterology
Ophthalmology
Neurology
Rheumatoid Immune Branch
Pediatrics
Internal Medicine
Renal Medicine
Cardiology
Family Medicine
Endocrinology
Dermatology
Thorax Internal
Psychiatrics
Sample Size Drug Ratio 2001
43,345
0.3751
30,721
0.4802
19,837
0.5443
28,323
0.5781
17,178
0.5781
12,140
0.5845
11,111
0.5902
19,548
0.6173
15,631
0.7151
5,618
0.7181
21,814
0.7521
46,974
0.7553
9,609
0.7600
11,708
0.7607
23,103
0.7662
8,397
0.7736
16,088
0.7753
10,571
0.7857
18,326
0.8381
BACK
17
,
Log(Ydjmt) = α(GBmt*DP2001d) + X’djmtβ + ϕd + τt+ γdj + εdjmt
• Log (Y): Log of outcome Y in department-type d in hospital j in
month m in year t
• DP2001d: department type d’s propensity to treat with drugs
•Xdjmt: Time-varying department characteristics (e.g. proportion of
elderly, male, diabetes, hypertension, arthritis, heart diseases,
psychological disorders); size of hospital where department located
•ϕd: Department-type fixed effects
•τt: Year fixed effects
•γdj: Hospital-department fixed effects
18

Coefficient of variation of drug ratio in 2001 in the 19 different
types of hospital departments
 CV=mean/std. dev.

Fixed effect coefficients for each type of department from the
regression using years <=2001
 Estimate: Drug ratio = dept type_FE + X + year_FE
 As above, X includes department characteristics, as well as time-
invariant hospital characteristics (owner type; region) because we
cannot control for hospital*department FE here.
 19 dept_FE will capture unobserved propensity to use drug for
each department
19
Dependent Variables
Average Drug Expenditure Per Case (NT$)
Mean
412.10
Std. Dev.
410.94
Average Non-drug Expenditure Per Case, at ex ante
prices (NT$)
251.87
413.42
20
300
400
500
600
700
Drug Expenditure
1996m1
1998m1
2000m1
2002m1
Year-Month
Scatter Plot
Fitted Value After GB
2004m1
2006m1
Fitted Value Before GB
21
log( Y jhmt )   ( GB jhmt  D jh ) 
2006
  (Year  D
t 1998
t
t
jh
) 11 X jhmt  12 H ht   jh   t   jht
GB=1 for months after July 2002
α: identifies lower bound of GB effect with Year×DDR2001 included
GBS Impact:
Calculate the percentage change of outcome due to GBS as
exp(αGBmt*DP2001d*mean of DP2001d) - 1
22
Log drug
expenditures
Log non-drug
expenditures
DP2001 = coefficient of variation of drug ratio1 in 2001 by department type
GB* DP2001
R-squared
GB Impact
0.013***
[7.78]
0.189
0.047
0.001
[0.42]
0.064
0.0004
DP2001 = fixed effect coefficients2 for each department type from the
regression using years <=2001
GB* DP2001
R-squared
GB Impact
0.060***
[9.22]
0.187
0.051
0.009
[0.62]
0.065
0.007
Observations3
411,840
393,021
23

Alternative calculations of drug ratio underlying DP2001

Mean reversion

IV estimation

Alternative samples (not in paper)
24
25
26
27
28

Do physicians lengthen prescriptions?
 Drug days

Do physicians switch among drugs? Look at expenditures on:
 Domestic drugs
 Most expensive domestic drug prescribed
 Imported drugs
 Most expensive imported drug prescribed
29
Log drug days
Log domestic drug
expenditure
Log max.
domestic drug
expenditure
Log imported
drug
expenditure
Log max.
imported drug
expenditure
DP2001 = coefficient of variation of drug ratio1 in 2001 by department type
GB* DP2001
R-squared
GB Impact
0.002*
[1.90]
0.415
0.006
0.012***
[6.05]
0.096
0.041
0.015***
[5.36]
0.133
0.048
0.012***
[6.53]
0.100
0.041
0.013***
[5.07]
0.130
0.044
DP2001 = fixed effect coefficients2 for each department type from the regression using years <=2001
GB* DP2001
R-squared
GB Impact
Observations3
0.010
[1.10]
0.408
0.009
412,522
0.056***
[5.08]
0.097
0.048
407,238
0.074***
[5.56]
0.137
0.064
376,812
0.056***
[5.97]
0.099
0.048
407,241
0.069***
[5.11]
0.133
0.059
376,823
30
Log drug days
Log domestic drug
expenditure
Log max.
domestic drug
expenditure
Log imported
drug
expenditure
Log max.
imported drug
expenditure
DP2001 = coefficient of variation of drug ratio1 in 2001 by department type
GB* DP2001
R-squared
GB Impact
0.002*
[1.90]
0.415
0.006
0.012***
[6.05]
0.096
0.041
0.015***
[5.36]
0.133
0.048
0.012***
[6.53]
0.100
0.041
0.013***
[5.07]
0.130
0.044
DP2001 = fixed effect coefficients2 for each department type from the regression using years <=2001
GB* DP2001
R-squared
GB Impact
Observations3
0.010
[1.10]
0.408
0.009
412,522
0.056***
[5.08]
0.097
0.048
407,238
0.074***
[5.56]
0.137
0.064
376,812
0.056***
[5.97]
0.099
0.048
407,241
0.069***
[5.11]
0.133
0.059
376,823
31
Log drug days
Log domestic drug
expenditure
Log max.
domestic drug
expenditure
Log imported
drug
expenditure
Log max.
imported drug
expenditure
DP2001 = coefficient of variation of drug ratio1 in 2001 by department type
GB* DP2001
R-squared
GB Impact
0.002*
[1.90]
0.415
0.006
0.012***
[6.05]
0.096
0.041
0.015***
[5.36]
0.133
0.048
0.012***
[6.53]
0.100
0.041
0.013***
[5.07]
0.130
0.044
DP2001 = fixed effect coefficients2 for each department type from the regression using years <=2001
GB* DP2001
R-squared
GB Impact
Observations3
0.010
[1.10]
0.408
0.009
412,522
0.056***
[5.08]
0.097
0.048
407,238
0.074***
[5.56]
0.137
0.064
376,812
0.056***
[5.97]
0.099
0.048
407,241
0.069***
[5.11]
0.133
0.059
376,823
32

GB system with a sheltered expenditure category will cause
providers to skew provision of health services towards the
sheltered spending, in this case drug treatments.
 The most conservative estimate: GB increases hospitals’ average
monthly drug expenditures by about 5%.
There is no evidence of physicians replacing non-drug
treatments with drug treatments
 There is no evidence of physicians lengthening prescriptions
 Instead, physicians may substitute more expensive (domestic and
imported) drugs for less expensive drugs.

33
Download