1. Background
Model and Data
Results
Conclusion
Inpatient Payment Schemes and Cost Efficiency
Evidence from Swiss Public Hospitals
Stefan Meyer
Department of Health Economics, University of Basel
13 September 2013
Swiss Health Economic Workshop, Lucerne
Stefan Meyer Inpatient Payment Schemes and Cost Efficiency
3. Background
Model and Data
Results
Conclusion
Inefficiencies in Healthcare
Swiss healthcare expenditure has risen by 46% over the last
decade, reaching CHF 62.5 billion (10.9% of GDP) in 2010
(BFS, 2010).
A significant part of healthcare costs arise due to inefficient
allocation of resources and inefficient production of goods
and services (OECD, 2010).
A recent US study by the WHO (2010) estimates total costs
of inefficiencies to be in a range of $ 600 - 850 billion per year
(≈ 30% of total health spending).
Stefan Meyer Inpatient Payment Schemes and Cost Efficiency
4. Background
Model and Data
Results
Conclusion
Inefficiencies in Healthcare
Swiss healthcare expenditure has risen by 46% over the last
decade, reaching CHF 62.5 billion (10.9% of GDP) in 2010
(BFS, 2010).
A significant part of healthcare costs arise due to inefficient
allocation of resources and inefficient production of goods
and services (OECD, 2010).
A recent US study by the WHO (2010) estimates total costs
of inefficiencies to be in a range of $ 600 - 850 billion per year
(≈ 30% of total health spending).
Stefan Meyer Inpatient Payment Schemes and Cost Efficiency
5. Background
Model and Data
Results
Conclusion
Inefficiencies in Healthcare
Swiss healthcare expenditure has risen by 46% over the last
decade, reaching CHF 62.5 billion (10.9% of GDP) in 2010
(BFS, 2010).
A significant part of healthcare costs arise due to inefficient
allocation of resources and inefficient production of goods
and services (OECD, 2010).
A recent US study by the WHO (2010) estimates total costs
of inefficiencies to be in a range of $ 600 - 850 billion per year
(≈ 30% of total health spending).
Stefan Meyer Inpatient Payment Schemes and Cost Efficiency
6. Background
Model and Data
Results
Conclusion
Inefficiency and Payment Schemes
Well-known factors like income, technological progress, and
demography have contributed to the continuing upward
trend.
Still, as these forces are hardly controllable, policies aimed at
reducing inefficiencies have become essential.
Prospective reimbursement schemes can serve as controlling
instruments for lowering cost inefficiency.
Stefan Meyer Inpatient Payment Schemes and Cost Efficiency
7. Background
Model and Data
Results
Conclusion
Inefficiency and Payment Schemes
Well-known factors like income, technological progress, and
demography have contributed to the continuing upward
trend.
Still, as these forces are hardly controllable, policies aimed at
reducing inefficiencies have become essential.
Prospective reimbursement schemes can serve as controlling
instruments for lowering cost inefficiency.
Stefan Meyer Inpatient Payment Schemes and Cost Efficiency
8. Background
Model and Data
Results
Conclusion
Inefficiency and Payment Schemes
Well-known factors like income, technological progress, and
demography have contributed to the continuing upward
trend.
Still, as these forces are hardly controllable, policies aimed at
reducing inefficiencies have become essential.
Prospective reimbursement schemes can serve as controlling
instruments for lowering cost inefficiency.
Stefan Meyer Inpatient Payment Schemes and Cost Efficiency
9. Background
Model and Data
Results
Conclusion
Inpatient Sector in Switzerland I
A substantial part of healthcare resources is allocated to the
inpatient sector, amounting to CHF 21.7 billion (34.9%) in
2009 (OECD, 2012).
Small efficiency gains in this sector may lower total
healthcare costs considerably.
Saved resources can be reallocated to socially beneficial
areas of healthcare.
The average length of stay (ALOS) in Swiss hospitals
remains high room for improving cost efficiency.
Stefan Meyer Inpatient Payment Schemes and Cost Efficiency
10. Background
Model and Data
Results
Conclusion
Inpatient Sector in Switzerland I
A substantial part of healthcare resources is allocated to the
inpatient sector, amounting to CHF 21.7 billion (34.9%) in
2009 (OECD, 2012).
Small efficiency gains in this sector may lower total
healthcare costs considerably.
Saved resources can be reallocated to socially beneficial
areas of healthcare.
The average length of stay (ALOS) in Swiss hospitals
remains high room for improving cost efficiency.
Stefan Meyer Inpatient Payment Schemes and Cost Efficiency
11. Background
Model and Data
Results
Conclusion
Inpatient Sector in Switzerland I
A substantial part of healthcare resources is allocated to the
inpatient sector, amounting to CHF 21.7 billion (34.9%) in
2009 (OECD, 2012).
Small efficiency gains in this sector may lower total
healthcare costs considerably.
Saved resources can be reallocated to socially beneficial
areas of healthcare.
The average length of stay (ALOS) in Swiss hospitals
remains high room for improving cost efficiency.
Stefan Meyer Inpatient Payment Schemes and Cost Efficiency
12. Background
Model and Data
Results
Conclusion
Inpatient Sector in Switzerland I
A substantial part of healthcare resources is allocated to the
inpatient sector, amounting to CHF 21.7 billion (34.9%) in
2009 (OECD, 2012).
Small efficiency gains in this sector may lower total
healthcare costs considerably.
Saved resources can be reallocated to socially beneficial
areas of healthcare.
The average length of stay (ALOS) in Swiss hospitals
remains high room for improving cost efficiency.
Stefan Meyer Inpatient Payment Schemes and Cost Efficiency
14. Background
Model and Data
Results
Conclusion
Inpatient Sector in Switzerland II
Until 2011: Considerable differences in inpatient
reimbursement among hospitals and local healthcare areas
(cantons)
As of 2012: Implementation of cased-based payment
(SwissDRGs) in public and private hospitals all over
Switzerland
Stefan Meyer Inpatient Payment Schemes and Cost Efficiency
15. Background
Model and Data
Results
Conclusion
4 Payment Systems in Question I
Revenue per case (Ri ) under 4 payment schemes:
per diem: flat payment per patient day
Rdiem
i = pt · ti
APDRG: flat payment per case (according to a national DRG
catalogue)
RDRG
i = CW k · ¯P
Stefan Meyer Inpatient Payment Schemes and Cost Efficiency
16. Background
Model and Data
Results
Conclusion
4 Payment Systems in Question II
DCB: flat payment per case (according to the hospital
department involved)
RDCB
i = Bk
PLT (hybrid): flat payment per case + flat payment per
patient day (both depending on the hospital department
involved)
RPLT
i = Bk + pk
t · ti
PLT: Prozess-Leistungs-Tarifierung, DCB: Department case-based payments (Abteilungspauschalen). We treat the
patient pathway system (MIPP, Cantonal Hospital of Aarau) as DCB, since the two systems do not differ in the
way they offer financial incentives to the hospital.
Stefan Meyer Inpatient Payment Schemes and Cost Efficiency
18. Background
Model and Data
Results
Conclusion
Research Outline I
I aim to...
estimate cost (in)efficiency (CE) scores for Swiss public
hospitals by means of Stochastic Frontier Analysis (SFA),
show that flat payment schemes decrease inefficiency (c.p.).
analyse the consequences of heteroscedasticity (to be
discussed later on).
Stefan Meyer Inpatient Payment Schemes and Cost Efficiency
19. Background
Model and Data
Results
Conclusion
Research Outline I
I aim to...
estimate cost (in)efficiency (CE) scores for Swiss public
hospitals by means of Stochastic Frontier Analysis (SFA),
show that flat payment schemes decrease inefficiency (c.p.).
analyse the consequences of heteroscedasticity (to be
discussed later on).
Stefan Meyer Inpatient Payment Schemes and Cost Efficiency
20. Background
Model and Data
Results
Conclusion
Research Outline I
I aim to...
estimate cost (in)efficiency (CE) scores for Swiss public
hospitals by means of Stochastic Frontier Analysis (SFA),
show that flat payment schemes decrease inefficiency (c.p.).
analyse the consequences of heteroscedasticity (to be
discussed later on).
Stefan Meyer Inpatient Payment Schemes and Cost Efficiency
21. Background
Model and Data
Results
Conclusion
Research Outline II
I take advantage of the fact that...
until 2011, health insurers employed 4 different inpatient
payment schemes in 26 cantons.
the problem of unobserved heterogeneity across observation
areas is alleviated, as we only focus on one country.
the public sector (cantons) reimburses hospitals on the same
basis (object financing = global budget, deficit guarantee).
Stefan Meyer Inpatient Payment Schemes and Cost Efficiency
22. Background
Model and Data
Results
Conclusion
Research Outline II
I take advantage of the fact that...
until 2011, health insurers employed 4 different inpatient
payment schemes in 26 cantons.
the problem of unobserved heterogeneity across observation
areas is alleviated, as we only focus on one country.
the public sector (cantons) reimburses hospitals on the same
basis (object financing = global budget, deficit guarantee).
Stefan Meyer Inpatient Payment Schemes and Cost Efficiency
23. Background
Model and Data
Results
Conclusion
Research Outline II
I take advantage of the fact that...
until 2011, health insurers employed 4 different inpatient
payment schemes in 26 cantons.
the problem of unobserved heterogeneity across observation
areas is alleviated, as we only focus on one country.
the public sector (cantons) reimburses hospitals on the same
basis (object financing = global budget, deficit guarantee).
Stefan Meyer Inpatient Payment Schemes and Cost Efficiency
24. Background
Model and Data
Results
Conclusion
Econometric Model: Cobb Douglas Cost Frontier
ln TCit = α +
m
βm ln ym
it +
n
βn ln pn
it +
k
βksk
it + vit + uit
εit
TCit Total costs of hospital i at time t
ym
it Outputs
pn
it Factor prices
sk
it Hospital characteristics
vit Two-sided random noise component
uit Non-negative inefficiency component
Stefan Meyer Inpatient Payment Schemes and Cost Efficiency
25. Background
Model and Data
Results
Conclusion
Econometric Model: Inefficiency Term
Battese and Coelli (1995) propose a model in which
inefficiency effects are assumed to be a function of
firm-specific variables and time:
uit = zitδ + wit, uit ∼ N+(zitδ, σ2
u)
zit
Vector of hospital-specific and regional variables (payment
scheme and other variables)
wit Random noise component (truncated normal distributed)
δ Vector of parameters to be estimated
Stefan Meyer Inpatient Payment Schemes and Cost Efficiency
27. Background
Model and Data
Results
Conclusion
Data
BFS data on 122 public and publicly-financed hospitals in
Switzerland
Period: 2004 - 2009 (T = 6)
Information on hospital category, teaching, inputs, inpatient
and outpatient outputs, and costs
5 observations per hospital on average, total sample size of
606 (inclusion criterion: ≥ 2 obs., ≥750 inpatient days)
Data covers a total of 4.9 million cases (≈ 58% of all
hospital cases between 2004 and 2009)
Stefan Meyer Inpatient Payment Schemes and Cost Efficiency
28. Background
Model and Data
Results
Conclusion
Despriptive Statistics I (N = 606)
Variable Description Mean SD
Cost frontier variables
TC Total costs (k CHF) 130 010.4 195 921.7
CASES Number of CMI-adjusted discharges 8 283.1 10 109.3
OUTP Revenue from outpatients (k CHF) 24 746.3 37 935.3
PL Price of labour (k CHF) 100.1 14.4
PK Price of capital (k CHF) 170.6 76.9
SERVICE Number of hospital services 36.2 18.0
INTERN Number of advanced training
categories*
20.5 27.9
*FMH Weiterbildungskategorien
Stefan Meyer Inpatient Payment Schemes and Cost Efficiency
29. Background
Model and Data
Results
Conclusion
Despriptive Statistics II (N = 606)
Variable Description Mean SD
Inefficiency variables
APDRG APDRG-based payment system* 0.27
PLT Hybrid payment system* 0.54
DCB Department level case-based
system*
0.06
OCC Bed occupancy rate 0.88 0.10
BEDS Acute care beds per 1,000 residents 2.95 0.70
PHYS GPs specialists per 1,000 residents 2.00 0.57
MCARE Managed Care contracts / total
number of MHI contracts
0.17 0.11
POP Population density (Residents / sq
km)
536.10 1 062.30
*Dummy variable
Stefan Meyer Inpatient Payment Schemes and Cost Efficiency
30. Background
Model and Data
Results
Conclusion
Heteroscedasticity
SFA are likely to suffer from severe heteroscedasticity
(HEC), especially if the hospitals are of different size.
Coefficients are biased in SFA models if HEC is overlooked
(frontier is changed when dispersion increases; s. Caudill et
al., 1995).
Heteroscedastic uit: Large hospitals have more“under their
control”, efficiency gets more important with size large
hospitals have much more in common than their small
counterparts
Heteroscedastic vit: Greater impact of random shocks on
small units (“Law of large numbers”)
Stefan Meyer Inpatient Payment Schemes and Cost Efficiency
31. Background
Model and Data
Results
Conclusion
Heteroscedasticity
SFA are likely to suffer from severe heteroscedasticity
(HEC), especially if the hospitals are of different size.
Coefficients are biased in SFA models if HEC is overlooked
(frontier is changed when dispersion increases; s. Caudill et
al., 1995).
Heteroscedastic uit: Large hospitals have more“under their
control”, efficiency gets more important with size large
hospitals have much more in common than their small
counterparts
Heteroscedastic vit: Greater impact of random shocks on
small units (“Law of large numbers”)
Stefan Meyer Inpatient Payment Schemes and Cost Efficiency
32. Background
Model and Data
Results
Conclusion
Heteroscedasticity
SFA are likely to suffer from severe heteroscedasticity
(HEC), especially if the hospitals are of different size.
Coefficients are biased in SFA models if HEC is overlooked
(frontier is changed when dispersion increases; s. Caudill et
al., 1995).
Heteroscedastic uit: Large hospitals have more“under their
control”, efficiency gets more important with size large
hospitals have much more in common than their small
counterparts
Heteroscedastic vit: Greater impact of random shocks on
small units (“Law of large numbers”)
Stefan Meyer Inpatient Payment Schemes and Cost Efficiency
33. Background
Model and Data
Results
Conclusion
Specifying HEC
Three different specifications:
(1) No HEC
(2) HEC in uit
(3) HEC in both uit and vit
Assumption: σ2
uit = exp(CASESitη) and
σ2
vit = exp(CASESitφ), where η and φ are parameters to be
estimated.
Stefan Meyer Inpatient Payment Schemes and Cost Efficiency
34. Background
Model and Data
Results
Conclusion
HEC: Findings
1 Both σ2
uit and σ2
vit are negatively correlated with hospital size
(p 0.01), indicating that variation of uit and vit is more
distinct among smaller institutions.
2 Coefficients (and standard errors) slightly change when HEC
is taken into acount.
Stefan Meyer Inpatient Payment Schemes and Cost Efficiency
35. Background
Model and Data
Results
Conclusion
SFA Results
ln TC σ2
u/ σ2
v σ2
ui / σ2
v σ2
ui / σ2
vi
ln CASES 0.830*** (0.015) 0.882*** (0.013) 0.864*** (0.014)
ln OUTP 0.117*** (0.014) 0.098*** (0.010) 0.102*** (0.012)
ln PL 0.026 (0.072) 0.057 (0.037) 0.000 (0.039)
ln PK 0.271*** (0.027) 0.249*** (0.016) 0.258*** (0.020)
ln SERVICE 0.027* (0.014) 0.001* (0.000) 0.001* (0.000)
ln INTERN 0.029*** (0.007) 0.002*** (0.000) 0.002*** (0.000)
YEAR 0.014*** (0.004) 0.014*** (0.004) 0.007 (0.005)
APDRG =0.407** (0.171) =0.298*** (0.081) =0.289*** (0.072)
PLT =0.136 (0.084) =0.080* (0.043) =0.092** (0.040)
DCB =0.931 (0.810) =0.561** (0.237) =0.593* (0.338)
OCC =1.754** (0.869) =1.108*** (0.339) =1.239*** (0.359)
BEDS =0.414** (0.180) =0.246*** (0.067) =0.256*** (0.066)
PHYS 0.330** (0.156) 0.103** (0.050) 0.131*** (0.046)
MCARE =0.997 (0.775) =0.396* (0.239) =0.457** (0.229)
POP 0.000 (0.000) 0.000** (0.000) 0.000** (0.000)
YEAR 0.097 (0.067) 0.043** (0.020) 0.058*** (0.018)
Note: N = 606; Cluster robust standard errors are given in parentheses; *p 0.10, **p 0.05, ***p 0.01;
The sign of the efficiency variables are to be read as effects on inefficiency; Both constants not shown
Stefan Meyer Inpatient Payment Schemes and Cost Efficiency
36. Background
Model and Data
Results
Conclusion
Main Findings I
Compared to per diem, all three systems are associated with
decreased inefficiency.
This finding is most pronounced among hospitals that apply
flat payment schemes (APDRG, DCB).
High bed density in acute care is associated with enhanced
CE (highly competitive markets little market / bargaining
power)
Inefficient hospitals are likely found in population-dense
areas with a high penetration of GPs and specialists (
proxy measure of demand; s. Chirikos, 1998).
Stefan Meyer Inpatient Payment Schemes and Cost Efficiency
37. Background
Model and Data
Results
Conclusion
Main Findings I
Compared to per diem, all three systems are associated with
decreased inefficiency.
This finding is most pronounced among hospitals that apply
flat payment schemes (APDRG, DCB).
High bed density in acute care is associated with enhanced
CE (highly competitive markets little market / bargaining
power)
Inefficient hospitals are likely found in population-dense
areas with a high penetration of GPs and specialists (
proxy measure of demand; s. Chirikos, 1998).
Stefan Meyer Inpatient Payment Schemes and Cost Efficiency
38. Background
Model and Data
Results
Conclusion
Main Findings I
Compared to per diem, all three systems are associated with
decreased inefficiency.
This finding is most pronounced among hospitals that apply
flat payment schemes (APDRG, DCB).
High bed density in acute care is associated with enhanced
CE (highly competitive markets little market / bargaining
power)
Inefficient hospitals are likely found in population-dense
areas with a high penetration of GPs and specialists (
proxy measure of demand; s. Chirikos, 1998).
Stefan Meyer Inpatient Payment Schemes and Cost Efficiency
39. Background
Model and Data
Results
Conclusion
Main Findings II
As found in an earlier study by Rosko (2004), managed care
penetration is positively correlated with CE (Insurers may
place financial pressure on hospitals when prices are being
negotiated).
Mean cost inefficiency increased significantly from 7.8% in
2004 to 12.3% in 2009.
CE scores only changed marginally when the panel was
almost balanced (from 2006).
Stefan Meyer Inpatient Payment Schemes and Cost Efficiency
40. Background
Model and Data
Results
Conclusion
Main Findings II
As found in an earlier study by Rosko (2004), managed care
penetration is positively correlated with CE (Insurers may
place financial pressure on hospitals when prices are being
negotiated).
Mean cost inefficiency increased significantly from 7.8% in
2004 to 12.3% in 2009.
CE scores only changed marginally when the panel was
almost balanced (from 2006).
Stefan Meyer Inpatient Payment Schemes and Cost Efficiency
42. Background
Model and Data
Results
Conclusion
Conclusion
Positive correlation of flat payment and CE in Swiss public
hospitals (controlling for hospital characteristics, the market
environment, and time trends).
Findings are in line with the basic economic theory on the
financial incentives of payment schemes (Ellis and McGuire,
1986).
Considerable differences in mean inefficiencies across
hospital categories (maximum gap of about 15% between per
diem and DLCB hospitals in 2009).
In SFA, heteroscedastic error terms (uit and vit) lead to
biased coefficients and standard errors.
Stefan Meyer Inpatient Payment Schemes and Cost Efficiency
43. Background
Model and Data
Results
Conclusion
Conclusion
Positive correlation of flat payment and CE in Swiss public
hospitals (controlling for hospital characteristics, the market
environment, and time trends).
Findings are in line with the basic economic theory on the
financial incentives of payment schemes (Ellis and McGuire,
1986).
Considerable differences in mean inefficiencies across
hospital categories (maximum gap of about 15% between per
diem and DLCB hospitals in 2009).
In SFA, heteroscedastic error terms (uit and vit) lead to
biased coefficients and standard errors.
Stefan Meyer Inpatient Payment Schemes and Cost Efficiency
44. Background
Model and Data
Results
Conclusion
Conclusion
Positive correlation of flat payment and CE in Swiss public
hospitals (controlling for hospital characteristics, the market
environment, and time trends).
Findings are in line with the basic economic theory on the
financial incentives of payment schemes (Ellis and McGuire,
1986).
Considerable differences in mean inefficiencies across
hospital categories (maximum gap of about 15% between per
diem and DLCB hospitals in 2009).
In SFA, heteroscedastic error terms (uit and vit) lead to
biased coefficients and standard errors.
Stefan Meyer Inpatient Payment Schemes and Cost Efficiency
45. Background
Model and Data
Results
Conclusion
Limitations
I do not account for quality differences among hospitals
(evidence suggests that ignoring them may not be a serious
problem; s. Mutter et al., 2008).
Unobservable differences among cantons biased
coefficients of the payment scheme variables.
Potential endogeneity in the outputs and the factor prices
as a major drawback in SFA (especially the price proxies may
reflect the hospitals’ choices about the average skill-mix and
the amount and mix of capital; s. Zuckermann et al., 1994).
Small sample of Swiss public hospitals unclear whether
these findings can be applied to private hospitals as well (or
even other healthcare systems).
Stefan Meyer Inpatient Payment Schemes and Cost Efficiency
46. Background
Model and Data
Results
Conclusion
Limitations
I do not account for quality differences among hospitals
(evidence suggests that ignoring them may not be a serious
problem; s. Mutter et al., 2008).
Unobservable differences among cantons biased
coefficients of the payment scheme variables.
Potential endogeneity in the outputs and the factor prices
as a major drawback in SFA (especially the price proxies may
reflect the hospitals’ choices about the average skill-mix and
the amount and mix of capital; s. Zuckermann et al., 1994).
Small sample of Swiss public hospitals unclear whether
these findings can be applied to private hospitals as well (or
even other healthcare systems).
Stefan Meyer Inpatient Payment Schemes and Cost Efficiency
47. Background
Model and Data
Results
Conclusion
Limitations
I do not account for quality differences among hospitals
(evidence suggests that ignoring them may not be a serious
problem; s. Mutter et al., 2008).
Unobservable differences among cantons biased
coefficients of the payment scheme variables.
Potential endogeneity in the outputs and the factor prices
as a major drawback in SFA (especially the price proxies may
reflect the hospitals’ choices about the average skill-mix and
the amount and mix of capital; s. Zuckermann et al., 1994).
Small sample of Swiss public hospitals unclear whether
these findings can be applied to private hospitals as well (or
even other healthcare systems).
Stefan Meyer Inpatient Payment Schemes and Cost Efficiency
48. Background
Model and Data
Results
Conclusion
Limitations
I do not account for quality differences among hospitals
(evidence suggests that ignoring them may not be a serious
problem; s. Mutter et al., 2008).
Unobservable differences among cantons biased
coefficients of the payment scheme variables.
Potential endogeneity in the outputs and the factor prices
as a major drawback in SFA (especially the price proxies may
reflect the hospitals’ choices about the average skill-mix and
the amount and mix of capital; s. Zuckermann et al., 1994).
Small sample of Swiss public hospitals unclear whether
these findings can be applied to private hospitals as well (or
even other healthcare systems).
Stefan Meyer Inpatient Payment Schemes and Cost Efficiency