The studies used were examined on the basis of comparable characteristics describing their extent, scope and results. A summary of the main characteristics of public hospital studies using the SFA method is shown in Table 1, according to the presentation method of Alatawi et al. (2019) based on PRISMA guidelines. The table shows the measured size, the number of units, the type of units, the area where the research was conducted, the investigation period, the model used and the result over the time period, while the most common inputs and outputs in the SFA efficiency studies are shown in Table 2.
Table 1
Characteristics of hospital studies (individual or regional) using the SFA method.
Author(s)
|
Measuredsize
|
Number of units
|
Type of units
|
Region – Affiliation
|
Period
|
Model
|
Result
|
Zhivanetal. (2012).
|
TE
|
1544
|
Urbanhospitals
|
USA
|
2006
|
SFA
|
85%
|
Barrosetal. (2013).
|
TE
|
51
|
districthospitals
|
Portugal
|
1997–2008
|
latentclass SFA model
|
88%
|
Kiadalirietal. (2013).
|
TE
|
29 studiesmeta-analysis
|
hospitals
|
Iran
|
1996–2006
|
Metaanalysis
|
84%
|
Kinfu (2013)
|
TE
|
52
|
Regions
|
South Africa
|
2010
|
Productionbasedmodel
|
80%
|
Votápková& Št’astná (2013)
|
Costeff
|
99
|
General hospitals
|
Chech
|
2001–2008
|
Cobb-Douglas
|
72%
|
Goudarzietal. (2014)
|
TE
|
12
|
teachinghospitals
|
Tehran
|
1999–2011
|
SFA
|
61 to 71%
|
Izon&Pardini (2015)
|
TE
|
30
|
Safety-netHospitals
|
United States
|
2005–2013
|
SFA
|
85%
|
Kinfu&Sawhney (2015)
|
TE
|
28 Regions
|
Childbirth in institutions
|
India
|
2007–2008
|
Spatial SFA
|
70%
|
Xuetal. (2015)
|
TE
|
51
|
Tetiartyhospitals
|
China
|
2009–2011
|
SFA
|
87–84%
|
Atilgan (2016)
|
TE
|
429
|
Regionalhospitals
|
Turkey
|
2012–2014
|
Battese and Coelli (1995)
|
Unintended
|
Bastianetal. (2016)
|
TE
|
59 (+ 69 clinics)
|
Military
|
USA
|
2011–2013
|
SFA
|
78%
|
Rezaeietal. (2016)
|
TE
|
12
|
hospitals
|
Kurdistan
|
2007–2014
|
SFA
|
63 to 68%
|
Hamidi (2016)
|
TE
|
22
|
Governmenthospitals
|
Palestine
|
2006–2012
|
SFA
|
55% to 455
|
Choietal. (2017).
|
TE
|
1471
|
Severaltypes
|
USA
|
2001–2011
|
Translogfunction
|
49 to 41%
|
Prossetal. (2018).
|
TE
|
1394
|
Strokecare
|
Germany
|
2006–2013
|
Bayesianmodel
|
73%
|
Roskoetal. (2018).
|
TE
|
37 States
|
Severaltypes
|
USA
|
2006–2010
|
Cobb-Douglas
|
High 96 low 69, av. 82
|
Mahdiyanetal. (2019).
|
TE
|
490
|
Hospital
|
Iran
|
2001–2017
|
Metaanalysis
|
80%
|
Jiang&Andrews(2020)
|
TE &costeff
|
20 Regions
|
secondaryand tertiary
|
NewZealand
|
2010–2017
|
Cobb–Douglasform
|
86%
|
Yesilyurt&Selamzade(2020)
|
TE
|
-
|
hospitals
|
CIS (f. USSR)
|
2010–2015
|
LeastSquares& MLS
|
Several
|
Corderoetal. 2021.
|
TE
|
22
|
Twotypes of hospitals
|
Panama
|
2005–2015
|
SFA
|
82%
|
Table 2
Selection of most common inputs-outputs in SFA efficiency studies of health units.
|
Inputs
|
|
Outputs
|
Researchers
|
Beds
|
Doctors
|
Nursingstaff
|
Administrative staff
|
Financialdetails (others)
|
Other
|
hospitalisationdays
|
Inpatients
|
Outpatients
|
Zhivanetal. (2012).
|
|
X
|
X
|
|
|
Buildings -equipment
|
|
Χ
|
Χ
|
Barrosetal. (2013).
|
X
|
X
|
|
|
X
|
Duration of stay
|
|
X
|
X
|
Kiadalirietal. (2013).
|
X
|
X
|
X
|
X
|
X
|
|
Χ
|
|
Χ
|
Kinfu, (2013)
|
X
|
X
|
|
|
X
|
Normal indexes
|
|
|
|
Votápková&Š-astná (2013)
|
X
|
X
|
X
|
|
X
|
Inpatientcost
|
|
X
|
|
Goudarzietal. (2014)
|
Χ
|
Χ
|
Χ
|
Χ
|
|
|
|
X
|
|
Izon&Pardini (2015)
|
X
|
|
|
|
X
|
|
|
|
X
|
Kinfu&Sawhney (2015)
|
|
|
|
|
|
Facilities *3
|
|
|
|
Xuetal. (2015)
|
X
|
X
|
|
|
X
|
Serviceoutput
|
|
|
|
Hamidi (2016)
|
X
|
X
|
X
|
X
|
|
|
|
X
|
X
|
Atilgan (2016)
|
Χ
|
Χ
|
Χ
|
Χ
|
|
Inhabitants
|
|
|
X
|
Rezaeietal. (2016)
|
Χ
|
Χ
|
Χ
|
Χ
|
|
|
|
X
|
|
Bastianetal. (2016)
|
X
|
X
|
X
|
|
X
|
Inputs-outputsSet
|
|
X
|
X
|
Choietal. (2017).
|
|
|
|
|
X
|
|
Χ
|
Χ
|
X
|
Prossetal. (2018).
|
Χ
|
Χ
|
Χ
|
|
|
Ratio *2
|
|
|
|
Roskoetal. (2018).
|
Χ
|
|
|
|
X
|
Proportions
|
X
|
X
|
X
|
Mahdiyanetal. (2019).
|
Χ
|
Χ
|
Χ
|
|
|
Meta-analysis
|
Χ
|
Χ
|
|
Jiang&Andrews (2020)
|
|
Χ
|
Χ
|
Χ
|
Χ
|
Intermediate*
|
|
|
X
|
Yesilyurt&Selamzade (2020)
|
Χ
|
Χ
|
Χ
|
|
|
Healthindicators 4*
|
|
|
|
Corderoetal. 2021.
|
Χ
|
Χ
|
Χ
|
Χ
|
|
|
|
Χ
|
|
*employed medical and estimated outsourced medical
*2 30-day mortality ratio and 30-day readmissions
*3 Output: Institutional delivery rate (%), Inputs: Number of facilities per 1000 sq km, Facilities having Medical Officer/Obstetrician/Gynecologist (%), Facilities received untied funding in previous financial year, Facilities with selected basic amenities (%).
4* Life expectancy, mortality rate, tuberculosis rate (Health system study)
5* Healthcare Effectiveness Data and Information Set (HEDIS)
The most frequently used inputs and outputs arise. However, each study uses different inputs and outputs, i.e. different model for assessing efficiency. This shows that the results are not comparable to each other.
Figure 3 shows both the initial value and the final value of efficiency, with results from the measurement of efficiency over an extended period of time, as well as the average of this change or the lump sum value of efficiency.
The study of the efficiency of hospitals based on the processing of the results of the bibliographic review is further analysed at the discussion stage. In figure 5 we can see a Piechart produced by KNIME software for the models used in the studies involved in this meta-analysis. Ιn figure 6 we can see a Piechart for the measured sizes used in the studies that participated in this meta-analysis.
Table 3: Correlation coefficients for pair correlation
|
Measured size
|
Number of units
|
Type of units
|
Region- Affiliation
|
Start Year
|
End Year
|
Model
|
Lower Results
|
Upper Results
|
Measured size
|
1
|
-0,1522
|
0,2515
|
0,235
|
0,1207
|
0,2347
|
0,1349
|
0,1585
|
0,1488
|
Number of Units
|
-0,1522
|
1
|
-0,242
|
-0,242
|
-0,0719
|
-0,1361
|
0,3381
|
-0,1974
|
-0,2143
|
Type of units
|
0,2515
|
0,0738
|
1
|
0,3614
|
0,3294
|
0,4034
|
0,1549
|
-0,693
|
-0,2777
|
Region- Affiliation
|
0,235
|
-0,242
|
0,3614
|
1
|
0,4252
|
0,5798
|
0,1997
|
-0,909
|
-0,097
|
Start Year
|
0,1207
|
-0,0719
|
0,3294
|
0,4252
|
1
|
0,4398
|
0,1967
|
-0,1857
|
-0,2131
|
End Year
|
0,2347
|
-0,1361
|
0,4034
|
0,5798
|
0,4395
|
1
|
0,1388
|
-0,1554
|
-0,1614
|
Model
|
0,1349
|
0,3381
|
0,1949
|
0,1997
|
0,1967
|
0,1388
|
1
|
-0,2707
|
-0,3007
|
Lower Results
|
0,1585
|
-0,1974
|
-0,2683
|
-0,0909
|
-0,1857
|
-0,1554
|
-0,2707
|
1
|
0,9941
|
Upper Results
|
0,1488
|
-0,2143
|
-0,2777
|
-0,092
|
-0,213
|
-0,1614
|
-0,3007
|
0,9941
|
1
|
All the results are further analyzed at the discussion stage. Efficiency values are only moderately correlated with the models used. These values hardly correlate at all with the measured size, number of units, type of units, region-affiliation, start year, end year.
Table 4 shows the statistical analysis for all the sizes used in the survey. Quantities and percentages of the total sizes used are calculated.
Table 4: Statistics on sizes used in research.
Size
|
Quantity
|
Percentage
|
|
Size
|
Quantity
|
Percentage
|
TE
|
18
|
90%
|
|
Region
|
|
|
Costeff
|
1
|
5%
|
|
USA
|
5
|
25%
|
TE &Costeff
|
1
|
5%
|
|
Iran
|
2
|
10%
|
|
|
|
|
Portugal
|
1
|
5%
|
Number of units
|
|
|
|
South Africa
|
1
|
5%
|
0-50
|
4
|
20%
|
|
Chech
|
1
|
5%
|
25/ 250
|
5
|
25%
|
|
Tehran
|
1
|
5%
|
250-750
|
5
|
25%
|
|
India
|
1
|
5%
|
750-1250
|
1
|
5%
|
|
China
|
1
|
5%
|
>1250
|
5
|
25%
|
|
Turkey
|
1
|
5%
|
|
|
|
|
Kurdistan
|
1
|
5%
|
Type of units
|
|
|
|
Palestine
|
1
|
5%
|
hospitals
|
3
|
15%
|
|
Germany
|
1
|
5%
|
Regions
|
2
|
10%
|
|
NewZealand
|
1
|
5%
|
Tetiartyhospitals
|
2
|
10%
|
|
CIS (f. USSR)
|
1
|
5%
|
Severaltypes
|
2
|
10%
|
|
Panama
|
1
|
5%
|
Urbanhospitals
|
1
|
5%
|
|
|
|
|
Districthospitals
|
1
|
5%
|
|
Period
|
|
|
General hospitals
|
1
|
5%
|
|
'90s-’00s
|
1
|
5%
|
teachinghospitals
|
1
|
5%
|
|
'90s-’10s
|
1
|
5%
|
Safety-netHospitals
|
1
|
5%
|
|
'00s
|
2
|
10%
|
Childbirth in institutions
|
1
|
5%
|
|
'00s-’10s
|
11
|
55%
|
Military
|
1
|
5%
|
|
‘10s
|
4
|
20%
|
Governmenthospitals
|
1
|
5%
|
|
|
|
|
Strokecare
|
1
|
5%
|
|
Result
|
|
|
Twotypes of hospitals
|
1
|
5%
|
|
80-90%
|
7
|
39%
|
|
|
|
|
70–80%
|
5
|
28%
|
Model
|
|
|
|
60-70%
|
4
|
22%
|
SFA
|
8
|
40%
|
|
50-60%
|
1
|
6%
|
Cobb-Douglas
|
3
|
15%
|
|
40-50%
|
1
|
6%
|
latentclass SFA model
|
1
|
5%
|
|
|
|
|
Metaanalysis
|
1
|
5%
|
|
Changes
|
|
|
Productionbasedmodel
|
1
|
5%
|
|
Positive
|
2
|
40%
|
Spatial SFA
|
1
|
5%
|
|
Negative
|
3
|
60%
|
Battese and Coelli (1995)
|
1
|
5%
|
|
Neutral
|
0
|
0%
|
Translogfunction
|
1
|
5%
|
|
|
|
|
Bayesianmodel
|
1
|
5%
|
|
|
|
|
Metaanalysis
|
1
|
5%
|
|
|
|
|
LeastSquares& MLS
|
1
|
5%
|
|
|
|
|
The statistical analysis of the figures is carried out in the discussion section.
From the analysis of the statistics of sizes, we observe that the TE measurement dominates the SFA studies with 90% and participates in a combined percentage of an additional 5%. The number of units studied is divided almost equally between very small, small and medium-sized hospitals and very large hospitals. Large hospitals are practically non-existent in the studies. Among the types of hospitals, regional, tertiary and hospitals of several types prevail. The wide range of types of hospitals that appear in the overview is remarkable. In the model used, the SFA characterisation predominates (40%), followed by the Cobb-Douglas model (15%).
In most of the health efficiency studies that participated in our survey using SFA, health policy suggestions emerged to improve hospital inefficiencies. Such proposals for the implementation of policy measures are:
Zhivan and Diana (2012) document a correlation between the inefficiency of a hospital and the decision to introduce Health Information Technologies (HIT). They propose state subsidy to hospitals depending on their efficiency in order for HIT to be implemented in them.
Kinfu (2013) examines outputs in relation to inputs by health region. At the same time he includes social indicators in the assessment of efficiency and proposes a reallocation of resources between different levels of health care and an alternative funding strategy for health services. He also highlights the importance of tackling HIV infection and female illiteracy in the areas affected, in order to ensure better use of resources.
Kiadaliri et al. (2013) found a number of policy decision measures to improve resource efficiency. Such measures are the establishment of a Management Board in hospitals, the implementation of performance-based budgeting and the establishment of a central information system and maintenance management.
Votápková and Št’astná (2013) find that the educational character and the increased participation of older people reduce the efficiency of hospitals. They propose complementing the economic analysis with satisfaction surveys on the quality of care.
Goudarzi et al. (2014) focuses on changing processes from management to the correct selection of hospital managers based on specific qualifications in order to perform their work effectively. Findings of this study warrants policy-makers and top management in tums to consider steps to improve the financial management of the university hospitals. The proposals for policy measures are as follows:
- Interventions to improve the quality of management in the hospitals studied could help to improve efficiency.
- Hospital managers should have sufficient knowledge of the costs and financial management of hospitals
- Implementation of process improvement tools taking into account quality of life and patient satisfaction
Kinfu and Sawhney (2015) propose avoidance of competitiveness between health regions and ensurance of full capacity in hospital units in order to eliminate sources of inefficiency. The importance of allocating new resources and raising the standard of living of the population is very important for efficiency.
Izon and Pardini (2015) find that the antidote to efficiency reduction is orientation towards profit maximisation strategies that do not require efficiency improvements.
Xu et al. (2015) propose research and scientific analysis to formulate a public health system in scientific terms and to establish standards of efficiency for hospitals with parallel reforms.
Hamidi (2016) focuses on increasing patient health provision, reducing hospital inputs, and changing the organisation of procedures.
More specifically, he proposes:
- Orientation of expenditure towards the most cost-effective interventions
- Changing hospitals’ organisation and procedures, aiming to reduce inputs
- Reduction of the remaining staff by transfer to Health Centres
Atilgan (2016) analyses the new evaluation system of Turkish hospitals where he uses SFA studies to renew or terminate managers’ contracts. He makes a number of findings that have an impact on the nature of the models used in the measurement but also on the size of hospitals such as:
- He indicates that the optimal size of hospitals is 250-400 beds and therefore dictates the conversion of hospitals to average size
- Improving the efficiency of educational hospitals, which is lagging behind the rest
- SFA models strongly influence hospital scores but not their classification and should be considered in SFA efficiency studies
Rezaei et al. (2016) suggest that implemention and designing of training programs based on time planning and measurement planning should be encouraged, in order to use resources effectively.
Bastian et al. (2016) emphasize the need to transfer best practices (business processes, resources and services) between units in order to increase efficiency.
Choi et al. (2017) partially confirm “the Baumol phenomenon”, namely that wage increase disrupts productivity or efficiency growth, and confirm Baumol and Bowen (1966), in that it is difficult to achieve efficiency gains in a labour-intensive industry. They also highlight the contradictory results of previous efficiency studies among different categories of hospitals. For policy makers, they propose three points regarding the process of measuring hospital efficiency:
• Use of structural variables explaining structural measures in procedures and results.
• Use of procedural measures that include all activities of health care instruments.
• Development of quality measures to be used in hospitals and measurement of intermediate substantial results (e.g. not only mortality index).
Pross et al. (2018) measure hospital inefficiency with emphasis on quality of care and calculation of spatial patterns of inefficiency. They delineate the inefficiency models in health policy-making by including poor quality and mortality factors based on spatial classification in the measurement of technical efficiency.
Rosko et al. (2018) suggest improving processes or best human resource practices, i.e. improvements in recruitment, selection, training or motivation processes.
Mahdiyan et al. (2019) emphasise the importance of the impacts of choice in measuring inputs and outputs on health (such as using the period of stay index as an output that can encourage hospitals to accept patients with simple diseases). For this reason, great importance should be attached to identifying inputs and outputs of the hospital that are commonly used to measure the hospital's efficiency in an accurate manner.
Jiang and Andrews (2020) justify solely and specifically in the service of local demand for healthcare in the event that resources are limited, increasing outpatient treatments (which lower the provision of healthcare services such as discharges earlier than expected or delaying treatment for emergencies (causing outcomes that distort overall effective management).
Yesilyurt and Selamzade (2020) emphasize the importance of the lack of efficiency studies and the disagreements at political and strategic level between groups of states on the issues of measuring efficiency. According to them, the implementation of neoliberal policies in the countries of the Commonwealth of Independent States and the homogenization of policies can contribute to increased efficiency.
Cordero et al. (2021) appear sceptical about exploiting efficiency studies especially in developing countries due to lack of evidence and limitations. They propose that they should not be used in strong policy recommendations such as closing down hospitals or reducing investment in low-performing institutions.
Figure 9 illustrates the measures proposed to be taken because of implementing a policy to improve efficiency by category. According to a top-down approach, the categories in which the measures are classified are:
- Central policy measures, such as interventions in social actors.
- Measures related to management, such as the determination of efficiency issues, the actions of the Hospital Administrations, the measurement of efficiency with SFA and the actions related to efficiency.
- Measures on operational issues, such as measures on expenditure, staff allocation and miscellaneous.