Original Research

Digital Health
Volume 9: 1–13
Applying frontier approach to measure © The Author(s) 2023

the financial efficiency of hospitals

Article reuse guidelines:
sagepub.com/journals-permissions
DOI: 10.1177/20552076231162987
journals.sagepub.com/home/dhj

Jih-Shong Wu

Abstract

Objectives: The growth in global healthcare capacity has led to increased healthcare costs and a deterioration in the finances
of universal health insurance. Hospitals must consider how to improve financial efficiency and service quality in order to
survive and operate sustainably.
Methods: This study applies data envelopment analysis (DEA) and stochastic frontier analysis (SFA) to measure the financial
efficiency of hospitals and to identify the factors and business strategies to improve profitability.
Results: The findings and recommendations show that (1) the DEA and SFA methods are similar and have reference values;
(2) financial efficiency should be improved by reducing medical costs; (3) the quality of medical staff should be improved and
manpower reduced; and (4) information, computerisation, and human intelligence in healthcare and management should
be enhanced.
Conclusions: In terms of practical applications, this study recommends the promotion of smart healthcare to improve the
efficiency and quality of healthcare services, as well as the introduction of artificial intelligence and big data analysis to opti-
mise the use of healthcare manpower. Electronic medical records can be used to reduce the wastage of resources and
labour costs, a medication management system can be established, and changes to the procurement system can be
made to reduce inventory and improve the efficiency of medical equipment use. It is hoped that this study will provide ref-
erence materials and applications for healthcare organisations to improve their operational efficiency and strategies.

Keywords

Financial efficiency, hospital, health care, data envelopment analysis, stochastic frontier analysis
Submission date: 28 October 2022; Acceptance date: 23 February 2023

Introduction order to fund universal healthcare and meet the growth in

demand.3,4 It is also common for the government to
As a result of increasing life expectancy and the capacity of assume responsibility and relief when hospitals are in finan-
elderly care, healthcare systems around the world are facing cial crisis. In particular, when public hospitals run into
a twofold problem of increasing demand and supply costs.1 financial deficits, they are heavily subsidised by the govern-
Recent statistics from the United Kingdom suggest that the ment. However, this also undermines the policies that
aging population alone is likely to increase demand for
healthcare by 3.3% over the next 15 years.2 As people’s
incomes rise and medical technology advances, so do
College of General Education, Chihlee University of Technology, New Taipei,
expectations of a high-quality and affordable healthcare
Taiwan
system. The aging population is also increasing the
Corresponding author:
demand for healthcare, all of which have prompted govern- Jih-Shong Wu, No. 313, Sec. 1, Wenhua Rd., Banqiao Dist., New Taipei City
ments to increase taxes, reduce other local spending, and 220305, Taiwan.
promote more efficient ways of producing healthcare in Email: [email protected]

Creative Commons Non Commercial CC BY-NC: This article is distributed under the terms of the Creative Commons Attribution-NonCommercial
4.0 License (https://creativecommons.org/licenses/by-nc/4.0/) which permits non-commercial use, reproduction and distribution of the work
without further permission provided the original work is attributed as specified on the SAGE and Open Access page (https://us.sagepub.com/en-us/nam/
open-access-at-sage).
2 DIGITAL HEALTH

hospitals want to reform in order to improve the efficiency In study of hospital efficiency using the DEA approach.
of their services.5 Agrawal and Conway6 suggested that Nunamaker15 was the first to apply DEA to the medical
healthcare should have the threefold aim of better health, field, where the cost of hospitalisation was the output and
better care, and lower costs. These are essential components the three inputs were the elderly and children, female
and goals for reforming healthcare policy. It is a fundamen- patients, and other days of hospitalisation. Seventeen hospi-
tal strategy that can be used to save the healthcare system tals were analysed for the relative efficiency of care. The
and make it sustainable and efficient. Cleverley7 and study found that over 60% of the hospitals were inefficient.
Gapenski8 have shown that the operational efficiency of a Sherman16 was the second to apply DEA to the evaluation
hospital can affect its financial performance. The higher of hospital efficiency. The number of doctors and surgeons,
the operating efficiency, the higher the expected profitabil- the number of hours worked by allied staff, the cost of
ity, which in turn allows hospitals to make profits to invest medical materials, and the total number of beds in the
in equipment and manpower, and increases the competitive- year were inputs, and the number of days in the hospital
ness of hospitals in the healthcare market. This will in turn for insured elderly people, the number of days in the hos-
enable the hospital to earn profits to invest in hospital equip- pital for uninsured elderly people, the number of nurse trai-
ment and manpower, and increase its competitiveness in the nees, and the number of inpatient trainees were outputs.
healthcare market. Stefko et al.17 used DEA to measure the regional efficiency
According to the statistics on the numbers of public/ of healthcare institutions in Slovakia between 2008 and
private hospitals published by the Ministry of Health and 2015. The study selected five inputs: number of beds,
Welfare of Taiwan9 between 1989 and 2020 (see number of medical staff, number of medical equipment,
Figure 1). In the 25 years from 1995 to 2020, the number number of magnetic resonance (MR) equipment, and
of public hospitals has decreased by 14 and the number number of computed tomography (CT) equipment. It had
of private hospitals has decreased by 200 (a 34% decrease). two outputs: number of beds used and average time of
In addition, the tightening of health insurance premiums in care. The results of the study showed that a gradual increase
Taiwan in recent years has led to a number of policies and in the number of MRs, CTs, and medical equipment to the
measures to reduce expenditure. For example, the institutio- inputs did not have a significant impact on the technical effi-
nalised impact of the increased chance of premium rejection ciency of the health facility. Lin et al.18 used DEA to
from health insurance and the rationalisation of outpatient measure the operational efficiency of 19 medical centres/
services (a reduction in the number of outpatient clinics) hospitals in Taiwan. The five inputs used were total beds,
has left incorporated hospitals, which do not have govern- total physicians, total equipment, net fixed assets, and emer-
ment assistance, with uncertainty and the risk of closure gency transfers for hospital stays longer than 48 hours. The
of medical services. In the face of the strong competitive- outputs included surplus or deficit in funding, length of
ness of public hospitals with inherent advantages and gov- stay, total relative value units of outpatient services,
ernment subsidies, how to improve the financial efficiency inpatient services, self-funded income, etc. The results of
of non-profit proprietary hospitals (NPPHs) has become an the study showed that the technical efficiency of medical
important factor for the survival and development of centres/hospitals was 96%, the pure technical efficiency
hospitals. was 99.1%, and the scale efficiency (SE) was 96.8%, with
Over the past 50 years, the methods used to measure effi- private medical centres outperforming public medical
ciency can be roughly divided into mathematical program- centres/hospitals.
ming methods and econometric methods. Data envelopment In the study of hospital efficiency using the SFA meth-
analysis (DEA) is the most common method used in mathem- odology, Rosko19 measured the impact of environmental
atical programming to estimate production frontiers. stress on hospital inefficiency in 616 hospitals between
Stochastic frontier analysis (SFA) is the most representative 1990 and 1999. The findings showed that the average inef-
method used in econometric methods to estimate production ficiency rate decreased from 14.35% in 1990 to 11.42% in
frontiers.10 DEA is recognised as a powerful tool for perform- 1998, and increased to 11.78% in 1999. Goudarzi et al.20
ance analysis and benchmarking, and evaluations are used used SFA to evaluate the efficiency of Kermanshah
across a wide range of industries and functional areas, includ- University of Medical Sciences Hospital between 2005
ing healthcare.11–13 Simar and Wilson14 advocated that DEA and 2011. Outputs included number of admissions. Inputs
is the most convenient and effective tool for measuring effi- included numbers of doctors, nursing staff, active beds,
ciency in hospitals, making it one of the most commonly and outpatient admissions. The study showed that the
used methods for measuring efficiency in hospitals. The dif- average efficiency of hospitals was 63%, leaving 37%
ference between DEA and SFA is that DEA is mostly used room for improvement. Furthermore, it was suggested
to examine the relative efficiency of each study object. that reducing hospital manpower would reduce hospital
SFA, on the other hand, explores the absolute efficiency costs and improve hospital performance. Goodarzi et al.21
and interrelationship between the impact factors of inputs used both DEA and SFA to measure the efficiency of hos-
and production (cost) efficiency. pitals of Kermanshah University of Medical Sciences
Wu 3

Figure 1. Numbers of public and private hospitals in Taiwan from 1989 to 2020.

during the period of 2001–2007. The same results were number of days in the hospital. This study found that
obtained for DEA and SFA. The technical efficiency of when hospital efficiency is measured, non-financial data is
the hospital was 95%, and staff costs accounted for the often used to measure the productivity of hospitals. The
majority of fixed costs. The reduction of redundant choice of inputs and outputs is crucial to the estimation of
human resources would reduce hospital costs and health- hospital efficiency. In general, inputs should include all
care costs. necessary resources, while outputs are the main objective
In general, measuring the efficiency of hospitals by the to be managed. Some studies suggest that the main inputs
number of inputs and outputs is very difficult and complex. for efficiency assessment of hospital costs are labour of
In recent years, both DEA and SFA methods have been medical staff and nurses and other operating costs (including
used to assess the performance of hospitals and other health- the fast-growing costs of medical supplies and drugs),13,24,25
care providers. DEA is a non-parametric technique that uses while the outputs are mainly medical income.
linear programming to measure frontier production function. Today, however, the healthcare industry is faced with
DEA measures the relative efficiency of multiple inputs and increasing costs for medical labour, medical drugs, and
outputs and is often used in not-for-profit organisations. medical materials. In many countries, healthcare systems
SFA is a parametric technique, a theory of econometric and hospitals are facing problems with operating costs
models, and the theories of microeconomics, which requires and financial difficulties. Therefore, this study applies
the use of a combination of data (panel data) to measure pro- both DEA and SFA methods and collects data from
duction (cost) function with regard to hypothesis or statistical NPPHs to measure the financial efficiency of hospitals, as
tests. Therefore, SFA is often applied in the evaluation of the well as the differences between the empirical findings of
efficiency of profit-making organisations. The DEA approach the two methods. This study aims to measure the financial
measures the efficiency of other hospitals using the observed and operational efficiency of hospitals in order to explore
best practitioners relative to all hospitals, but it does not the important factors influencing the financial efficiency
provide any statistical tests to confirm the results. In terms of hospitals, and to propose relevant recommendations
of measuring the efficiency of hospitals, the SFA method and improvement strategies to enhance the financial effi-
uses a constructive production function or cost function to ciency and quality of healthcare in hospitals.
measure and rank hospitals.22 SFA can be used to detect dif-
ferences in inefficiencies between hospitals.19
In addition, Fazria and Dhamanti23 examined a total of Methodology
586 past research articles from 23 countries and found that Efficiency is the measurement of the operational perform-
the five most common input factors used to assess the effi- ance of a decision making unit (DMU) and the scope for
ciency of hospitals were number of beds, medical staff, non- improvement of the DMU’s operations. Efficiency is a
medical staff, medical technicians, and operating costs. The matter of measuring the best efficiency of input and
most common output factors were the number of inpatient output under the goal of maximum output or minimum
admissions, number of operations, number of emergency cost. Farrell26 was one of the first to explore efficiency
visits, number of outpatient attendances, cost of care, and measurement. He cited the work of Debreu27 and
4 DIGITAL HEALTH

Koopmans28 to define a simple measure of efficiency that and Xij denotes the ith input quantity of the jth DMU. All
could deal with a situation where there are multiple the weighting coefficients ur and vi must be positive and
inputs. He believed that a firm’s efficiency consists of two must not be zero. Finally, ɛ is set to a very small positive
components: technical efficiency, which represents a value, often 10−4 or 10−6 in practice, and is referred to as
firm’s ability to achieve maximum output with a given the non-Archimedean number.29
mix of inputs, and allocative efficiency, which reflects the Input-oriented BCC model. The input-oriented BCC
optimum input ratio for a firm at a constant relative price. model has an additional variable u0 compared to the CCR
The combination of these two components can be used to model. It represents the return to scale model, as in Coelli et al.29:
measure a firm’s total economic efficiency. Farrell26 sug-
s
u Y − u0
gested that efficiency could be measured either non- Max hj = r=1
m r rj (2)
parametrically or parametrically, and his proposed effi- i=1 vi Xij
ciency measure became a pioneer of the frontier
s
u Y − u0
approach.29 The non-parametric method, DEA, is the s.t. r=1
m r rj ≤ 1, j = 1, · · · , n
most commonly used method, while the parametric i=1 vi Xij

method, SFA, is the most representative, and both ur , vi ≥ ε ≥ 0, r = 1, · · · , s, i = 1, · · · , m.

methods are often used to measure the efficiency of hospi-
tals in various ways.
As this will produce an infinite set of solutions when
actually solved, the equation is converted to a linear pro-
DEA method gramming model, that is, the denominator is set to 1 to
facilitate the solution, as in Coelli et al.29:
Since Farrell26 proposed a non-parametric approach to
measure technical efficiency, only a few researchers have 
s
Max hj = ur Yrj − u0 (3)
considered it. It was not until Charnes, Cooper, and
r=1
Rhodes (CCR)30 proposed a mathematical model of DEA
based on Farrell’s26 concept of technical efficiency that 
m

the theoretical status of the non-parametric approach to effi- s.t. vi Xij = 1

i=1
ciency was established. The CCR model assumes
output-oriented production and constant returns to scale 
s 
m
ur Yrj − vi Xij − u0 ≤ 1, j = 1, · · · , n
(CRS). Subsequently, Banker, Charnes, and Cooper r=1 i=1
(BCC)31 removed the restriction on CRS from the CCR
model and proposed that their BCC model assume variable ur , vi ≥ ε ≥ 0, r = 1, · · · , s, i = 1, · · · , m,
returns to scale (VRS), that is, a partial increase in inputs where, the return to scale can be seen from u0 as follows:
does not result in a relative increase in output. These two When u0 = 0, it represents CRS;
models are now recognised by academics as the most influ- When u0 > 0, it represents decreasing returns to scale
ential in the field of DEA.32 Total economic efficiency in (DRS); and
the input-oriented considers the reduction of input costs When u0 < 0, it represents increasing returns to scale (IRS).
and is therefore also known as cost efficiency.
Output-oriented aggregate economic efficiency considers
the increase in revenue from the output and is therefore
also known as revenue efficiency. SFA method
Input-oriented CCR model. The mathematical program-
The SFA method was developed by Aigner et al.33 and
ming for the input-oriented CCR model is as follows.
Meeusen and van den Broeck.34 The main concept is to
Assume that a DMU has s outputs and m inputs, with a
connect the most efficient input and output combinations
total of n DMUs, and hj denotes the efficiency index of a
of the firm to be evaluated into a productivity frontier and
particular evaluated DMU among the n DMUs29:

s to assume that not all firms evaluated are efficient. Only
ur Yrj firms at the production frontier of efficiency are technically
Max hj =
r=1 m (1)
i=1 vi Xij
efficient. This study uses the Cobb–Douglas stochastic

s frontier production functions proposed by Battese and
ur Yrj Coelli35 for continuous data across time. The Cobb–
s.t.
r=1
m ≤ 1, j = 1, · · · , n
i=1 vi Xij Douglas stochastic frontier model takes the form29:

ur , vi ≥ ε ≥ 0, r = 1, · · · , s, i = 1, · · · , m. lnYit = β0 + βj lnXit + vit − uit , i = 1, · · · N, t
j
where, hj denotes the relative efficiency value of the jth
DMU, Yrj denotes the rth output quantity of the jth DMU, = 1, · · · T (4)
Wu 5

where, Yit denotes the appropriate type produced by the ith determine efficiency, but we have no way of knowing
hospital at time t, Xit denotes the ith vendor input at time t, β whether the hospital is losing money and going out of
denotes the coefficient of input for each of the production business.
functions, and vit represents the symmetric interference This study uses both DEA and SFA to measure effi-
term of the random variation of the production function. ciency analysis. In the DEA method, one output factor,
Here, vit ∼ iddN(0, σ 2v ) and uit are independent variables, GMI, and four input factors, MPE, DC, MMC, and R &
with uit representing the technical inefficiency effects, the D, were selected as the data for analysis. The SFA
only non-negative random variable, these technical effi- method was also used to construct the Cobb–Douglas sto-
ciency predictions are between zero and one, with a value chastic frontier function model with the same data as the
of one indicating full technical efficiency.29 DEA, and to investigate the difference between general
and religious NPPHs, which are funded by religious
groups, and to measure whether there is a difference in
Data and empirical model their operational efficiency. The statistics of the variables
in this study are shown in Table 1.
Samples and data sources
The sample for this study was collected from 53 NPPHs
from the Ministry of Health and Welfare (MOHW)9 in
Taiwan from 2005 to 2020, excluding incomplete data
from the previous two years, data from hospitals that are Empirical model of DEA
closed down, and missing data. The data for this study This study used the CRS and VRS input-oriented DEA
was obtained from 37 hospitals over a 14-year sample models to estimate the technical efficiency of each hospital.
period from 2007 to 2020, with a total sample size of The SE (= CRS/VRS) was obtained by subtracting the CCR
518. The financial statements of hospitals in Taiwan are from the BCC. The study data were used to estimate the
audited by an accountant and approved by the MOHW technical efficiency of 37 hospitals based on the DEA meth-
before being made public. The financial statements odology over a period of 14 years using one output, GMI,
contain information on three main items: medical revenues, and four inputs: MPE, DC, MMC, and R & D. The com-
medical costs, and operating expenses. Medical revenues puter software used for calculation was DEAP Version
include emergency and outpatient services; medical costs 2.1, which is freely available from Coelli.36
include personnel costs, drug costs (DCs), medical device
costs, and depreciation; and operating costs include market-
ing, management, and research and development costs
incurred in the provision of medical services.9
In this study, one output was selected, gross medical Empirical model of SFA
income (GMI), and four inputs were selected: medical per-
This study relies on the Cobb–Douglas stochastic frontier
sonnel expenses (MPEs), DCs/expenses, medical material
model proposed by Aigner et al.33 and Battese & Coelli35
costs (MMCs), and research and development (R & D)
uses the same data as the DEA approach to develop an
costs. As the sample period for this study is 14 years, the
SFA model to measure inefficiency indicators in hospitals.
2020 price index is used as the base period for the conver-
The output term (GMI) and input terms (MPE, DC, MMC,
sion of deflators. The Taiwan dollar is converted into US
and R&D) of the DMU are substituted in Equation (4),
dollars based on the average exchange rate against the US
which can be converted to Equation (5). Calculations
dollar for each year.
were made using the Frontier Version 4.1 computer soft-
ware, which is available free of charge from Coelli.37 The
Variables empirical model is:
This study focuses on the financial characteristics of hospi- ln (GMIit ) =β0 + β1 ln (MPEit ) + β2 ln(DCit )
tals, selecting input and output financial data for efficiency + β3 ln (MMCit ) + β4 ln (R&Dit ) + Vit − Uit
analysis, and examining the financial performance of hospi-
(5)
tals in terms of operating costs in order to reduce costs and
improve operational strategies. The use of financial state- where, i is the hospital code, where i = 1, 2, …, N; t is the
ment data provides a direct insight into the profitability time, where t = 1, 2, …, T; Vit – Uit is the composite error,
and expense profile of a hospital, as profitability is what which is a combination of random shock and production
determines the viability of a business, that is, whether a hos- inefficiency; Vit is the random error term of hospital i’s
pital can generate sufficient remuneration to continue to bilateral allocation in period t; and Uit is the inefficiency
operate. This study differs from previous studies that used factor of hospital i’s unilateral allocation in period t and
the numbers of medical staff, beds, and patients to is a non-negative truncated normative allocation.29
6 DIGITAL HEALTH

Table 1. Descriptive statistics of variables (Unit: USD$1000).

Items Variables Samples Mean Std. Min. Max. Median

Output Gross medical income 518 179,025 321,343 2747 2,366,808 77,535

Inputs Medical Personnel Expenses 518 77,973 145,431 1474 1,060,231 32,882

Drug costs 518 37,106 68,207 394 513,631 12,355

Medical material costs 518 19,903 35,388 15 286,434 7594

R & D costs 518 6136 15,178 2 125,865 1828

Results and discussion (six times), Hospital A1 (five times), Hospital A15 (four
times), Hospital A14 and A31 (three times each), Hospital
Annual average efficiency analysis and discussion A10 and A35 (two times each), and Hospital A18 (one
Table 2 and Figure 2 present the estimated DEA and SFA time). The more times a hospital is peer-referenced, the
annual average efficiency values for 37 hospitals from 2007 more relatively efficient it is and the more robust it is mea-
to 2020 (a total sample period of 14 years). In terms of the sured to be. This allows other hospitals to review the input
DEA results, 10 hospitals with a technical efficiency of 1 in and output reference values of peer hospitals of their own
terms of CRS are relatively efficient hospitals. In terms of size to improve their own operational efficiency.
pure technical efficiency for VRS, 16 hospitals have an effi- The SFA estimates in Table 2 show that the hospitals
ciency value of 1 and are relatively efficient hospitals. In with the highest-ranked efficiency of 0.98 for the sample
terms of SE, 14 hospitals have an SE of 1, which indicates period were A14, A18, A19, and A30 (four hospitals in
that these hospitals have the best operational performance total). The hospitals with the second-highest efficiency
and capacity at the most appropriate scale of production. In of 0.96 were A10, A15, A16, and A22 (four hospitals in
the RTS column, 26 hospitals are shown as ‘DRS’, which total). The third-highest-ranked hospitals with efficiency
means that the hospital is in a state of decreasing scale and values of 0.95 were A4, A17, A24, and A35 (four hospi-
the input resources should be suitably reduced. Two hospitals, tals in total). This means that these 12 hospitals are per-
A13 and A16, show ‘IRS’, which means that the hospital is forming relatively well. The average SFA efficiency
experiencing incremental growth in scale and that additional value of 0.91 shows that NPPHs have excellent financial
inputs could increase the scale of operation. The BCC VRS efficiency, but there is still room for improvement at 9%
model is a measure of pure technical efficiency for hospitals, overall.
which prevents the scale of production from affecting the In addition, Table 2 and Figure 2 show that there is only
DEA efficiency measure. Table 2 shows that the pure tech- a very small difference between the average efficiency
nical efficiency values (VRSTE) for BCC are greater than values of DEA and SFA. Among the individual hospitals,
the technical efficiency values (CRSTE) for CCR, which indi- the best DEA efficiency was 1. Using the SFA, there
cates that, with the exception of the hospitals with a technical were eight hospitals in common that had better efficiencies
efficiency of 1 for CCR, the efficiency performance of all hos- of 0.95 or more. This indicates that the DEA and SFA find-
pitals is affected by the scale of production or production ings are both very similar and of good reference value.
technology. There are 16 hospitals with an efficiency of 1 However, in terms of efficiency and ranking, SFA takes
in the input-oriented BCC model. This figure is seven more the absolute efficiency of all hospitals in the sample,
than the nine hospitals using the CCR model, which indicates while DEA takes the relative efficiency of each hospital.
that the seven hospitals are purely technically efficient due to Therefore, SFA is more accurate than DEA. Furthermore,
economies of scale. In addition, the total annual average effi- after examining separately whether the hospitals established
ciencies for CRS, VRS, and SE are 0.91, 0.96, and 0.95, were general or religious NPPHs, the differences in average
respectively. This indicates that the technical efficiency, efficiency values were not significant, suggesting that the
pure technical efficiency, and SE of all hospitals are good, financial efficiency of NPPHs did not differ depending on
but there is still room for improvement at 9%, 4%, and 5%, the nature of their establishment.
respectively. Among the sample hospitals in the reference
peer group, the hospitals that could provide the reference
target and reference frequency were Hospital A24 (a total Analysis and discussion of DEA slack variables
of twenty times), Hospital A7 (thirteen times), Hospital The DEA slack variable represents the amount of input that
A22 (eight times), Hospital A5 (seven times), Hospital A30 should be reduced or the amount of output that should be
Wu 7

Table 2. Estimates of the average annual efficiencies of DEA and SFA for each hospital.

DEA efficiency SFA efficiency

Reference peer Number of

Firms (DMUs) CRSTE VRSTE SE RTS hospitals references Efficiency sort

A1 1.00 1.00 1.00 – – 5 0.93 5

A2 0.84 0.86 0.98 DRS A5、A24、A30 0.89 9

A3 0.83 0.94 0.89 DRS A7、A15、A22、A24 0.93 5

A4 0.83 0.94 0.88 DRS A7、A24 0.95 3

A5 1.00 1.00 1.00 – – 7 0.91 7

A6 0.87 0.92 0.95 DRS A5、A7、A22、 0.80 17

A24、A30

A7 0.98 1.00 0.98 DRS – 13 0.94 4

A8 0.97 0.98 0.99 DRS A1、A24、A31 0.93 5

A9 1.00 1.00 1.00 – – 0.87 11

A10 0.86 1.00 0.86 DRS – 2 0.96 2

A11 0.80 0.85 0.94 DRS A7、A15、A22、A24 0.91 7

A12 0.84 0.96 0.88 DRS A7、A10、A14、 0.91 7

A15、A24

A13 0.88 0.89 0.99 IRS A1、A22、A24、A30 0.82 14

A14 1.00 1.00 1.00 – – 3 0.98 1

A15 0.90 1.00 0.90 DRS – 4 0.96 2

A16 0.93 0.94 0.99 IRS A22、A24、A30 0.96 2

A17 0.77 1.00 0.77 DRS – 0.95 3

A18 0.83 1.00 0.83 DRS – 1 0.98 1

A19 0.92 0.98 0.94 DRS A1、A7、A24、A35 0.98 1

A20 0.97 1.00 0.97 DRS – 0.72 18

A21 0.96 0.99 0.97 DRS A5、A7、A24 0.93 5

A22 1.00 1.00 1.00 – – 8 0.96 2

A23 0.82 0.92 0.90 DRS A5、A7、A22、A24 0.88 10

A24 1.00 1.00 1.00 – – 20 0.95 3

A25 0.82 0.84 0.97 DRS A5、A7、A22、A24 0.90 8

(continued)
8 DIGITAL HEALTH

Table 2. Continued.
DEA efficiency SFA efficiency

Reference peer Number of

Firms (DMUs) CRSTE VRSTE SE RTS hospitals references Efficiency sort

A26 0.96 0.98 0.98 DRS A5、A7、A24 0.92 6

A27 1.00 1.00 1.00 – – 0.93 5

A28 0.83 0.93 0.89 DRS A15、A18、A24 0.81 16

A29 0.86 0.98 0.87 DRS – 0.94 4

A30 1.00 1.00 1.00 – – 6 0.98 1

A31 1.00 1.00 1.00 – – 3 0.85 12

A32 0.86 0.87 0.99 DRS A24、A31 0.94 4

A33 0.86 0.86 1.00 DRS A5、A24、A30 0.82 15

A34 0.93 0.93 1.00 DRS A1、A24、A31 0.94 4

A35 0.86 1.00 0.86 DRS – 2 0.95 3

A36 0.82 0.89 0.92 DRS A7、A10、A14、 0.92 6

A24、A35

A37 0.92 0.93 0.99 DRS A1、A7、A14、 0.83 13

A22、A30

Average value for religiously 0.90 0.96 0.94 0.91

incorporated hospitals.

Average value for general corporate 0.91 0.95 0.96 0.91

hospitals.

Mean 0.91 0.96 0.95 0.91

Note: 1. CRSTE = technical efficiency from CRS DEA.

2. VRSTE = technical efficiency from VRS DEA.
3. SE = scale efficiency = CRSTE/VRSTE.
DEA: data envelopment analysis; SFA: stochastic frontier analysis; DMU: decision making unit.

increased in order to achieve the same resource efficiency the high costs of medical materials and research and devel-
as the relatively inefficient hospitals. Slack variable ana- opment costs.
lysis can be used to provide a direction and magnitude
of improvement for inefficient hospitals for both reference
Analysis and discussion of the SFA method
and application. In Table 3, the slack variables for ineffi-
cient hospitals show that three hospitals had to reduce estimation results
their medical staff costs by an average of US$513,509, In this study, the SFA method was used to generate the
two hospitals had to reduce their medical product costs results using the maximum likelihood estimation method,
by an average of US$104,778, 12 hospitals had to as shown in Table 4. Table 4 shows that the inefficiency
reduce their MMCs by an average of US$414,382, and indices for hospital inputs such as MPE, DC, and MMC,
14 hospitals had to reduce their research and development which are all relevant to the GMI of the hospital, have a
costs by an average of US$571,691. This study used the highly significant positive correlation while R & D has a
slack variables to identify inefficiencies in output due to significant positive correlation. This means that expenditure
Wu 9

Figure 2. Distributions of the average annual efficiencies of DEA and SFA for each hospital.
DEA: data envelopment analysis; SFA: stochastic frontier analysis.

on these inputs should be reduced first, particularly the The third recommendation is the application of AI to the
higher estimates of medical staff costs (estimate = 0.4569) database of imaging medicine and diagnostic accuracy.
and pharmaceutical costs (estimate = 0.3332), followed by With the help of medical robots, the load on medical man-
a reduction in the cost of medical materials (estimate = power can be reduced and efficiency improved, and overall
0.1121). medical staff costs reduced. In addition, according to data
There are large and persistent differences in healthcare collected by this study, the cost of medical manpower in
expenditure and outcomes. In the United Kingdom, for private hospitals in Taiwan accounts for 51.59% of the
example, long-term records show that 10% of hospitals total cost of medical services. Second, the remaining 50%
with the lowest risk of death due to health service variation of the total cost of care is spent on medical equipment
are 21% lower than the U.K. average.1 First, the healthcare and consumable pharmaceutical and MMCs. Therefore,
system should provide better doctors and nurses to reduce many studies suggest that the efficiency of healthcare
the number of less-skilled medical staff, and to streamline, should be improved and maintained by finding and combin-
professionalise, and make more efficient the work and ing suppliers who can provide low-cost and high-quality
medical practice by informing the practice, tests, and con- medical equipment and supplies.40–42 This study suggests
sultations. Table 2 uses the second-most efficient hospital that medical devices and drug management systems
in the SFA (number A10) as an example. Hospital A10 should be enhanced to reduce costs and that a knowledge
uses artificial intelligence (AI) to predict when patients management system of physician usage patterns should
will arrive for emergency care and the number of patients be developed that incorporates big data to adopt specific
at peak times to ensure that medical staff and supplies are purchasing strategies to reduce inventory. It may even be
sufficient to meet the demand and uses AI to optimise possible to review how medical supplies are actually used
staff scheduling and bed arrangements for a smoother emer- by healthcare professionals to improve the efficiency of
gency process.38 Second, the construction and updating of investment and use of medical equipment.38
electronic medical records and standards can reduce the
cost of space and time wasted on paper medical records.
In hospital number A10 and A15, for example, electronic Conclusions
medical records can be used to reduce the waste of paper This study measured 37 NPPHs with a total sample size of
resources and can be integrated with AI programmes to 518 for the period of 2007 to 2020. The methodology was
control the safety of patients’ medication. When doctors based on the CRS and VRS input-oriented DEA models,
prescribe too high a dose of medication, the computer and the Cobb–Douglas stochastic frontier model, in order
system will alert them to avoid medical errors and to measure the inefficiency of a hospital. The difference
enhance the professionalism and efficiency of care.38,39 between the DEA and SFA methods is that DEA measures
10 DIGITAL HEALTH

Table 3. Slack variables for inefficient hospitals (Units: US$1000). Table 3. Continued.
Medical Medical
Medical Medical
Firms personnel Drug material
Firms personnel Drug material
(DMUs) expenses costs costs R&D
(DMUs) expenses costs costs R&D

A28 12,195.501 0 0 3757.842

A1 0 0 0 0
A29 279.679 0 1284.246 1594.805
A2 0 0 1859.351 135.187
A30 0 0 0 0
A3 0 0 0 112.648
A31 0 0 0 0
A4 6524.644 0 142.32 3033.677
A32 0 963.081 1180.383 168.292
A5 0 0 0 0
A33 0 0 3332.017 2241.767
A6 0 0 0 0
A34 0 0 261.647 3445.979
A7 0 0 0 0
A35 0 0 0 0
A8 0 2913.704 463.587 0
A36 0 0 0 0
A9 0 0 0 0
A37 0 0 0 0
A10 0 0 0 0
Number of 3 2 12 14
A11 0 0 0 1064.951
DMUs
with
A12 0 0 0 0
slack
variable
A13 0 0 83.371 0
Mean 513.509 104.778 414.382 571.691
A14 0 0 0 0
DMU: decision making unit.
A15 0 0 0 0

A16 0 0 919.433 387.758

technical inefficiency based on the actual output being less
A17 0 0 0 0 than the frontier output. Meanwhile, SFA uses production
functions and estimates the values of individual parameters
A18 0 0 0 0 by means of the best approximation, and ignores the effect
of random factors on output. The results also do not have
A19 0 0 1227.286 0 the same efficiency of 1 for multiple DMUs as in the case
of DEA. However, DEA measures multiple outputs and
A20 0 0 0 0
inputs and eliminates the need to construct production func-
A21 0 0 26.64 1342.318 tions to estimate the ease of efficiency, thus making the use
of DEA methods more common, extensive, and convenient
A22 0 0 0 0 than SFA methods for studying hospital efficiency.
The empirical results of this study gave the following
A23 0 0 0 640.119 findings. (1) The overall operating efficiency of the
NPPHs was excellent and reached at least 91% on
A24 0 0 0 0 average, with room for improvement of approximately
5% to 9%. (2) The average efficiency values of DEA and
A25 0 0 0 1033.925
SFA differed only marginally. (3) When we compared
A26 0 0 4551.851 2193.31 general and religious NPPHs, we found that there was
little difference in the average efficiencies, which indicates
A27 0 0 0 0 that there is no difference in operational efficiency accord-
ing to how the hospitals are established. (4) The results of
(continued) the SFA show that the inefficiency indices of three input
Wu 11

Table 4. Estimates of the stochastic frontier function.

Variable description Coefficient Estimate Standard error t-ratio

Constant β0 2.2577*** 0.1403 16.0916***

lnMPEit β1 0.4569*** 0.0382 11.9593***

lnDCit β2 0.3332*** 0.0288 11.5823***

lnMMCjt β3 0.1121*** 0.0133 8.4051***

lnR&Djt β4 0.0223** 0.0100 2.2396**

σ 2s = σ 2u + σ 2v σ 2s 0.0408*** 0.0055 7.4511***

r = σ 2u / σ 2s R 0.3719*** 0.0854 4.3537***

log likelihood function = 189.7692

Note: 1. ***, **, and * represent significance at the 1%, 5%, and 10% levels, respectively.
2. σ 2u and σ 2v represent the inefficiency error variance and random error variance, respectively; σ 2s is the total variance; r is the proportion of inefficiency error
variance (σ 2u ) to total variance (σ 2s ).

factors, namely, medical personnel costs, DCs, and MMCs, be encouraged in order to enhance medical technology and
have a highly significant positive correlation, and according efficiency. In recent years, the vision for healthcare in
to the data obtained from this study, these three factors also Taiwan has been to promote ‘smart healthcare’ to
account for almost all of the total medical costs. Hospitals improve the efficiency and quality of medical services. In
should actively reduce their expenditure in these three terms of practical applications, AI will be actively imple-
areas to significantly improve efficiency. (5) The DEA mented and combined with big data analysis to predict
and SFA methods are similar and have reference values. the number of patients arriving at hospitals, in order to opti-
Based on the findings of this study, the following strat- mise manpower and the quantity of medical drugs and
egies for improving healthcare are recommended. (1) materials, thereby smoothing out healthcare services and
Medical personnel costs can be reduced appropriately to processes. Secondly, the use of electronic medical records
enhance financial efficiency. However, in order to avoid stored in the cloud reduces the wastage of resources and
excessive reduction in manpower and resulting in overload- eliminates the need for huge storage space and the process-
ing of medical staff, it is recommended that the quality of ing and labour costs required to access the records. In add-
the overall medical staff should be improved and that infor- ition, medication management systems can be developed to
mation on the practice, tests, and consultations should be reduce prescription errors and integrate with purchasing
streamlined to make work and medical practice more strategies to reduce inventory. The actual frequency of
streamlined, professional, and efficient. (2) Informative use of medical devices can even be reviewed to improve
electronic medical records and standards should be con- the efficiency of investment and use of medical equipment.
structed and updated, which can reduce the space and It is hoped that the results of this study can be used as a ref-
time costs wasted on paper medical records. (3) AI can be erence for healthcare organisations to improve their busi-
applied to the database and diagnostic accuracy of ness strategies and efficiency.
imaging medicine. Medical robots can be used to reduce
the load on medical manpower and increase efficiency.
This is a new trend in the future of healthcare. (4) In the Acknowledgements: The free software ‘Frontier Version 4.1’,
and ‘DEAP Version 2.1’ can be used to estimate the equations,
area of pharmaceuticals, it is proposed to strengthen cost
which was kindly provided by Professor Coelli.33,34
management and implement stock control and prescription
management, and to reduce costs and improve financial per-
formance by purchasing directly from pharmaceutical com- Declaration of conflicting interests: The author declared no
panies and reviewing the use of pharmaceuticals. (5) The potential conflicts of interest with respect to the research,
cost of medical devices should be appropriately reduced authorship, and/or publication of this article.
to enhance efficiency, and suppliers that can provide
low-cost, high-quality medical devices and related supplies
should be identified and integrated. (6) R & D in the field of Funding: The author received no financial support for the
combined imaging medicine and intelligent robotics should research, authorship, and/or publication of this article.
12 DIGITAL HEALTH

Ethical approval: Does not apply. This article does not contain 15. Nunamaker TR. Measuring routine nursing service efficiency:
any studies with human participants or animals performed by a comparison of cost per patient day and data envelopment
any of the authors. analysis models. Health Serv Res 1983; 18: 183–208.
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1068745/
pdf/hsresearch00522-0075.pdf.
Guarantor: JSW. 16. Sherman HD. Hospital efficiency measurement and evalu-
ation: empirical test of a new technique. Med Care 1984;
22: 922–938. https://www.jstor.org/stable/3764490.
ORCID iD: Jih-Shong Wu https://orcid.org/0000-0001-5136- 17. Stefko R, Gavurova B and Kocisova K. Healthcare efficiency
2739 assessment using DEA analysis in the Slovak republic. Health
Econ Rev 2018; 8: 1–6.
18. Lin CS, Chiu CM, Huang YC, et al. Evaluating the oper-
References ational efficiency and quality of tertiary hospitals in Taiwan:
1. Lee T, Propper C and Stoye G. Medical labour supply and the the application of the EBITDA indicator to the DEA
production of healthcare. Fisc Stud 2019; 40: 621–661. method and TOBIT regression. Healthcare 2022; 10: 58.
2. Charlesworth A, Firth Z, Gerschlick B, et al. Future pressures 19. Rosko MD. Performance of U.S. teaching hospitals: a panel
on the NHS and social care. In: Charlesworth A and Johnson analysis of cost inefficiency. Health Care Manag Sci 2004;
P (eds) Securing the future: funding health and social care to 7: 7–16.
the 2030s, London: Institute for Fiscal Studies, 2018, pp.1–2. 20. Goudarzi R, RjabiGilan N, Ghasemi SR, et al. Efficiency meas-
3. Kiadaliri AA, Jafari M and Gerdtham UG. Frontier-based urement using econometric stochastic frontier analysis (SFA)
techniques in measuring hospital efficiency in Iran: a system- method, case study: hospitals of Kermanshah University of
atic review and meta-regression analysis. BMC Health Serv Medical Sciences. Journal Kermanshah University Medical
Res 2013; 13: 1–11. Science 2014; 17: 666–672.
4. Yaisawarng S. Performance measurement and resource allo- 21. Goodarzi G, Nasab MHI, Mehr NJ, et al. Hospital performance
cation. In: Fox KJ (ed) Efficiency in the public sector. assessment of Lorestan University of Medical Sciences. Payesh
Studies in productivity and efficiency. Boston, MA: 2012; 11: 309–315. http://payeshjournal.ir/article-1-453-en.html.
Springer, 2002; 1, pp. 61–68. 22. Rezaei S, Zandian H, Baniasadi A, et al. Measuring the efficiency
5. Berger M, Sommersguter-Reichmann M and Czypionka T. of a hospital based on the econometric stochastic frontier analysis
Determinants of soft budget constraints: how public debt (SFA) method. Electron Physician 2016; 8: 2025–2029.
affects hospital performance in Austria. Soc Sci Med 2020; 23. Fazria NF and Dhamanti I. A literature review on the identi-
249: 1–11. fication of variables for measuring hospital efficiency in the
6. Agrawal S and Conway PH. Aligning emergency care with data envelopment analysis (DEA). Unnes Journal of Public
the triple aim: opportunities and future directions after health- Health 2021; 10: 1–15.
care reform. Healthcare 2014; 2: 184–189. 24. O’Neill L, Rauner M, Heidenberger K, et al. A cross-national
7. Cleverley WO. Essentials of health care finance. comparison and taxonomy of DEA-based hospital efficiency
Gaithersburg: Aspen Publishers, Inc., 1992. studies. Socioecon Plann Sci 2008; 42: 158–189.
8. Gapenski LC. Understanding health care financial manage- 25. Ozcan YA. Health care benchmarking and performance
ment. Arlingtin: AUPHA Press, 1996. evaluation: an assessment using data envelopment analysis
9. Ministry of Health and Welfare of Taiwan. Number of (DEA). New York: Springer, 2008. https://link.springer.com/
medical institutions. Retrieved from https://dep.mohw.gov. book/10.1007/978-0-387-75448-2
tw/dos/cp-5301-62356-113.html (accessed 2 April 2022). 26. Farrell DJ. The measurement of productive efficiency.
10. Lovell CK. Production frontiers and productive efficiency. In: Journal of the Royal Statistical Society 1957; 120: 253–290.
Fried HO, Lovell CK and Schmidt S.S (eds) The measurement 27. Debreu G. The coefficient of resource utilization.
of productive efficiency-techniques and applications. Oxford: Econometric 1951; 19: 273–292.
Oxford University Press, 1993; pp. 3–67. https://www. 28. Koopmans TC. Chapter III: an analysis of production as an effi-
researchgate.net/publication/239060004. cient combination of activities, activity analysis of production
11. Araújo C, Barros CP and Wanke P. Efficiency determinants and allocation, Owles commission for research in economics.
and capacity issues in Brazilian for-profit hospitals. Health Monograph. New York: Wiley. 1951, Issue 13; pp. 33–97.
Care Manag Sci 2014; 17: 126–138. https://cowles.yale.edu/sites/default/files/files/pub/mon/m13-
12. de Almeida Botega L, Andrade MV and Guedes GR. all.pdf
Brazilian hospitals’ performance: an assessment of the 29. Coelli TJ, Prasada Rao DS O’, et al. An introduction to efficiency
unified health system (SUS). Health Care Manag Sci 2020; and productivity analysis. 2nd ed. New York: Spinger, 2005.
23: 443–452. 30. Charnes A, Cooper WW and Rhodes E. Measuring the efficiency
13. Kohl S, Schoenfelder J, Fügener A, et al. The use of data of decision-making units. Eur J Oper Res 1978; 2: 429–444.
envelopment analysis (DEA) in healthcare with a focus on 31. Banker RD, Charnes A and Cooper WW. Some models for
hospitals. Health Care Manag Sci 2019; 22: 245–286. estimating technical and scale inefficiencies in data envelop-
14. Simar L and Wilson PW. Sensitivity analysis of efficiency ment analysis. Manage Sci 1984; 30: 1078–1092.
scores: how to bootstrap in nonparametric frontier models. 32. Seiford LM. Data envelopment analysis: the evolution of the
Manage Sci 1998; 44: 49–61. https://pubsonline.informs. state of the art (1978-1995). Journal of Productivity Analysis
org/doi/abs/10.1287/mnsc.44.1.49. 1996; 7: 99–137. https://www.jstor.org/stable/41770796.
Wu 13

33. Aigner D, Lovell CAK and Schmidt P. Formulation and 38. Chen FY. The president of the first western hospital in northern
estimation of stochastic frontier production function models. Taiwan: smart medical care should be patient-centered and
J Econom 1977; 6: 21–37. reduce the burden on colleagues. Common Wealth Magazine
34. Meeusen W and van den Broeck J. Efficiency estimation from 2020; 709. https://futurecity.cw.com.tw/article/1708 (accessed
Cobb-Douglas production function with composed error. Int 15 January 2023).
Econ Rev (Philadelphia) 1977; 18: 435–444. 39. Wan NS. 24-h intelligent monitoring in the emergency room,
35. Battese GE and Coelli TJ. Prediction of firm-level technical simulated hospitalization rate control bed. Business Today, 22
efficiencies with a generalized frontier production function February 2022. Available at: https://www.businesstoday.com.
and panel data. J Econom 1988; 38: 387–399. tw/article/category/80394/post/202202220048/ (accessed 15
36. Coelli TJ. A guide to DEAP Version 2.1: A data envelopment ana- January 2023).
lysis (computer) program. CEPA Working Paper 96/8, 40. Kowalski JC. Needed: a strategic approach to supply chain
Department of Econometrics, University of New England, management. Healthc Financ Manage 2009; 63: 90–98.
Armidale, Australia. 1996a. https://www.owlnet.rice.edu/∼econ PMID: 19526824.
380/DEAP.PDF. 41. Sener Z and Dursun M. A fuzzy decision making approach for
37. Coelli TJ. A guide to FRONTIER Version 4.1: A computer program supplier selection in healthcare industry. International Scholarly
for stochastic frontier production and cost function estimation. and Scientific Research & Innovation 2014; 8: 597–600.
CEPA Working Paper 96/7, Department of Econometrics, 42. Alshahrani S, Rahman S and Chan C. Hospital-supplier integra-
University of New England, Armidale, Australia. 1996b. https:// tion and hospital performance: evidence from Saudi Arabia. The
iranarze.ir/wp-content/uploads/2017/07/7209-English-IranArze.pdf International Journal of Logistics Management 2018; 29: 22–45.