Methodology

- Contents lists available at sciencedirect.com

Journal homepage: www.elsevier.com/locate/vhri

The Factors Associated With Direct Medical Costs in Patients With

Gastric Cancer: Quantile Regression Approach Compared With Gamma
Regression
Saeed Mohammadpour, PhD,1 Noureddin Niknam, PhD,2 Javad Javan-Noughabi, PhD,3,4 Mehdi Yousefi, PhD,4
Hosein Ebrahimipour, PhD,4 Hajar Haghighi, PhD,5 Farzaneh Kasraei, BSc,6 Mehdi Kargar, PhD,7 Tahere Sharifi, PhD8,9,*
1
Department of Health Economics, School of Management and Medical Information, Iran University of Medical Sciences, Tehran, Iran; 2Department of Health Services
Management, School of Management and Medical Information, Iran University of Medical Sciences, Tehran, Iran; 3Health Management and Economics Research
Center, Iran University of Medical Sciences, Tehran, Iran; 4Social Determinants of Health Research Center, Mashhad University of Medical Sciences, Mashhad, Iran;
5
Health Policy, Tehran University of Medical Sciences, Tehran, Iran; 6School of Health and Research Center for Health Sciences, Shiraz University of Medical Sciences,
Shiraz, Iran; 7Department of Health Promotion, Faculty of Health, Shiraz University of Medical Sciences, Shiraz, Iran; 8Department of Health Care Management,
School of Health, Tehran University of Medical Sciences, Tehran, Iran; 9Health Sciences Research Center, Torbat Heydariyeh University of Medical Sciences, Torbat
Heydariyeh, Iran.

A B S T R A C T

Objectives: Evaluating the costs of illness can provide evidence to improve performance at all levels of health organizations.
This study aimed to identify the relationship between the costs of diagnosing and treating patients with gastric cancer and
their explanatory variables, using quantile and gamma regressions and comparing the results of the two models.
Methods: This was a cross-sectional and descriptive-analytic study carried out in 2016. In total, 449 patients with gastric
cancer were selected at a hospital affiliated with Mashhad University of Medical Sciences. Direct costs and other variables
were collected from medical documents. Data were analyzed using the STATA 12 software, using quantile and gamma
regression analysis, and the results were compared.
Results: The highest average cost per patient was related to hospitalization costs in both metastatic (20 911 034 Iranian Rials)
and nonmetastatic patients (20 738 062 Iranian Rials). The lowest average cost was related to biopsy services in
nonmetastatic patients. The results of the study also showed that quantile regression is an appropriate substitute for gamma
regression and, in some cases, can provide more information for the analysis of disease costs. Based on the results of the
quantile regression, being a male and having a shorter stay had a positive effect on cost and the age of the patient had a
significantly negative effect.
Conclusions: Examining the cost of a common illness, such as gastric cancer, is an important economic tool for policy makers
and decision makers. It provides evidence-based decision making about resource allocation that they can use for future
planning and cost control.

Keywords: health expenditures, hospitals, stomach neoplasms.

VALUE IN HEALTH REGIONAL ISSUES. 2020; 21(C):127–132

Introduction country. Moreover, the age of outbreak of this highly expensive

and hard-to-treat disease has gone down. As a result, some
Cancer is the most important noncontagious disease with specialists use the term “cancer tsunami” to express the severity of
harmful effects in recent decades. Nowadays, 20 million people the financial burden and the number of human victims of this
live with cancer, and this number is expected to reach 30 million disease in Iran.1
by 2020. Cancer is known as the third leading cause of death in Gastrointestinal cancers are the second most prevalent group of
Iran after cardiovascular disease and trauma and accidents. With cancers worldwide, and gastric cancer is the fourth most common
the increasing life expectancy and development of the elderly cancer globally after lung, breast, and colon cancers; however, it is
population, the number of cancer cases has multiplied in the still known as the second agent of death. Based on statistics from

Conflict of interest: The authors have declared that they have no conflicts of interest.
* Address correspondence to: Tahere Sharifi, PhD, Department of Health Care Management, School of Health, Tehran University of Medical Sciences, Tehran, Iran.
Email: sharifit@razi.tums.ac.ir
2212-1099/$36.00 - see front matter ª 2019 ISPOR–The professional society for health economics and outcomes research. Published by Elsevier Inc.
https://doi.org/10.1016/j.vhri.2019.08.479
128 VALUE IN HEALTH REGIONAL ISSUES MAY 2020

the Iran Cancer Research Center, gastric cancer is the most care options for gastric cancer were extracted from the
prevalent cancer among men and the second most prevalent aforementioned guidelines under the observation of the oncology
among women in Iran. In contrast with the Western countries and radiologists, and the terminology for each of these care options
Japan, the incidence of gastric cancer in Iran is increasing.2,3 was designed identically based on the service codes and forms
According to reports from the World Health Organization (WHO), written in the aforementioned book of tariffs. Finally, a form sheet
the mortality rate of cancer is 127.7 people per every 100 000.4 consisting of all provided care options for diagnosing and treating
Gastric cancer can be treated in the short term or the long term, gastric cancer was extracted. The mentioned care options included
and its cost can be limited or high.5 It is estimated that its cost runs endoscopy and biopsy, surgery, radiography, CT scan, radiotherapy,
10 times higher than the cost of noncancerous patients.6 Because laboratory tests, chemotherapy, medicine, and hospital admission
patients have different personal characteristics, their disease sectors. The reliability of the form was affirmed by the five
diagnosis and treatment methods are diverse and, consequently, oncologists. Its validity was also assessed using the parallel
not even similar cases cost the same amount. Nevertheless, there method and the same sheet forms. This was done thanks to the
are some similarities between the mean costs of them.5 use of patients’ charts with the cooperation of two oncology
Health systems are seriously concerned about assessing and radiotherapists. The correlations among the collected data were
controlling costs. They use prediction models, which examine the then assessed by Kappa coefficient. The form sheet was evaluated,
process of a disease cost’s condition, as an indicator of patients’ revised, and completed by two specialists, who also completed
and health centers’ characteristics.7 statistical consults. The data were collected simultaneously by two
Statistical models are typically used in most of the economic health information technologists. These technologists were
studies that relate to the health fields. The main challenge is to trained practically and informed about how the form sheets had
estimate the mean healthcare costs and recognize the association been completed. The second phase was done to calculate the cost
between cost and modifying variables through regression models. of gastric cancer treatment and diagnosis. The data were collected
The data of disease costs indicate a considerable skew so that from the charts of 449 patients with gastric cancer who had been
generalized linear models provide a framework for linear referred to the hospital during the 2005 to 2015 period.
model-making of those variables, which are not distributed The same prices and tariffs, based on formal tariffs in the year
normally. In healthcare, generalized linear models do not include 2015, were used for all the charts to calculate direct medical costs.
the challenges of usual linear regression.8,9 The cost of medications came from the website www.daroobank.
Gamma regression is a generalized linear model used in most com and reflect the price of medications in the year 2015.
investigations for making models of disease expenses.10,11 It is Because the data were gathered out of the patients’ charts and the
used like ordinary least squares (OLS) for estimating the effect size information of prices were reported from the year 2015, there was
of descriptive variables about cost in which the coefficient of an no need to calculate the discount rate.
independent variable represents the average rate of cost change Frequency, mean, standard deviation, and the maximum are
for the data with positive skewness and variance inconsistency. the indices that were used to describe the costs.
Skewness and inconsistency are the two main characteristics of Potential factors describing the diagnosis and treatment of
disease cost data.12 Quantile regression is a method used in some gastric cancer costs were modeled using generalized linear and
studies for disease cost model-making.13,14 It estimates the change quantile regressions. A generalized linear regression is a series of
in a defined quantile of conditional distribution related to a statistical models for making relations between skewed dependent
dependent variable, which is associated to a change unit of an variable and independent variables.15 A gamma regression was
independent variable. This model provides no assumption about used in the present study. Quantile regression makes it possible to
data distribution and only produces some neutral estimates, even assess the association of independent variables with the quantiles
when lacking normal distribution or outliers. This approach can of dependent variables, even in the presence of outliers and
replace linear regression and provide a comprehensive landscape without any need for the normality of data.
of different factors that can affect disease costs.12 In this study, the dependent variable was the cost of
Regarding the importance of disease costs and related diagnosing and treating gastric cancer. In addition, the backward
determinants, the present investigation was conducted to investigate method was used to choose the explanatory factors of the model.
the relationship between the costs of diagnosing and treating gastric The assessment criterion for the model’s relevance was the Akaike
cancer and modifying variables by using quantile and gamma information criterion (AIC). This criterion strikes a balance
regressions and comparing the two models’ results with one another. between the model’s precision and its complexity. Its equation is
AIC=2k-2ln (L), where k and L represent, respectively, the number
of the statistical model’s parameters and the maximum likelihood
Methods for the estimated model.16,17
Finally, gender, age, insurance coverage, metastasis condition,
This is a cross-sectional, descriptive-analytical, and practical and the length of stay in the hospital were chosen as the
investigation, which was conducted using information gathered describing factors of disease costs. The achieved results were
from a hospital. It was done using a retrospective approach in compared with each other. STATA 12 was used to analyze the data
2016 from the charts of patients with gastric cancer admitted to and the significance level was supposed to be 0.05. The ethical
Mashhad city’s Omid Hospital between 2005 and 2015. Data considerations of the present study included the commitment of
collection was done in two phases. The first phase involved researchers to keep confidential the personal information of
designing a form sheet, which included diagnostic and treatment patients and to describe the goals of the study to the hospital’s
services for gastric cancer based on credible guidelines and staff who took part in gathering the data.
according to the literature, as well as according to the book of
tariffs in Iran (which is called The Relative Value of National Health
Care Services). In this phase, the most important clinical guidelines Results
about cancer treatment were noted through open interviews with
five oncology specialists who each had at least five years of As shown in Table 1, male and female patients made up,
experience working with treating cancer. Afterward, all necessary respectively, 76.2% and 23.8% of the studied population. The
 METHODOLOGY 129

Table 1. Demographic information.

Variable Abundance (%) Variable Abundance (%)

Sex Male 342 (76.2) Underlying disease None 376 (83.7)
Female 107 (23.8) Yes 73 (16.3)
Age, years 26-35 11 (2.4) Stage of disease None 14 (3.1)
36-45 29 (6.5) Yes 109 (24.3)
46-55 78 (17.4) Step 1 80 (17.8)
56-65 114 (25.4) Step 2 117 (26.1)
66-75 132 (29.4) Step 3 129 (28)
76-85 73 (16.3) Type of disease Step 4 117 (26)
86-95 8 (1.8) Unknown 332 (74)

majority of patients were between the ages of 66 and 75 (29.4%). By increasing the cost variable distribution quantile, the Beta
There were 117 patients in the metastatic group and 332 in the coefficient of constant also increased from 9.7 million to 115
nonmetastatic group. million. The magnitude of the estimated coefficient was due to the
According to the results, the highest average service delivery largeness of the measure unit of the dependent variable per Rial.
rate was for laboratory services with 29 services for a patient Because being female was considered the basis for the gender
without metastasis and 24 for a metastatic patient; the highest variable, in the first 50% of the data distribution, the male patients
average cost per patient was related to hospitalization costs in had a positive and significant effect on cost (b = 4 100 000; P = .01).
both the metastatic (20 911 034 Iranian Rials) and nonmetastatic The duration of hospital stay showed a positive and significant
(20 738 062 Iranian Rials) groups. correlation with cost in all quantiles of distribution of cost data.
The lowest average cost of receiving medical services was The results of the estimation of coefficients showed that the
related to biopsy services in the nonmetastatic group. In Table 2, length of stay in hospital in the higher quantiles had a greater
the mean and standard deviations of costs are differentiated impact on patient costs, so that the highest quantile (95%) had an
according to the type of services received for both the metastatic estimated coefficient of 3 082 353. Because the patient's
and nonmetastatic patients. nonmetastatic status was considered the baseline, being
The results of the quantile regression are shown in Table 3. The metastatic had a negative impact on cost, but this effect was not
width of the model's constant, which represents cost changes in statistically confirmed in any of the quantiles. No insurance
the event of the removal of explanatory variables, was meaningful coverage was given baseline status. Insurance coverage in any of
in fitted quantiles. the quantiles was not meaningful. Nevertheless, insurance

Table 2. The costs according to received services.

Service Metast Max Mean Std Service Metast Max Mean

Num endoscopy No 4 1 0.735 Num radiotherapy No 38 15.83
Yes 3 0.78 0.767 Yes 29 5.46
Endoscopy cost No 7 392 000 1 860 988 1 380 682 Radiotherapy cost No 9 856 000 1 435 036
Yes 5 544 000 1 452 376 1 450 350 Yes 2 552 000 480 615.4
Num biopsy No 1 0.02 0.122 Num test No 118 28.81
Yes 0 0 0 Yes 108 23.87
Biopsy cost No 4 488 000 38 963.86 368 746.8 Laboratory cost No 28 632 560 6 903 133
Yes 0 0 0 Yes 21 840 720 5 395 513
Surgery cost No 20 240 000 4 310 675 4 699 082 Num surgery No 3 0.5
Yes 20 416 000 1 357 607 3 489 108 Yes 4 0.17
Num gerafy No 33 2.04 2.691 Num chemotherapy No 79 4.89
Yes 7 1.69 1.523 Yes 20 5.44
Gerafy cost No 26 716 800 1 650 286 2 179 989 Chemotherapy cost No 27 874 880 811 073.7
Yes 5 667 200 1 352 808 1 186 502 Yes 4 299 680 817 527.5
Num scan No 3 0.32 0.555 Drug cost No 182 027 400 9 570 183
Yes 4 0.37 0.651 Yes 139 279 800 15 653 953
Scan cost No 14 406 480 491 419 1 500 388 Hospitality cost No 10 5267 600 20 911 034
Yes 14 967 040 635 525.5 1 995 176 Yes 88 372 800 20 738 062

Metast indicates metastasis; Num, number; Std, standard deviation.

Costs calculated based on Iranian Rial.
130 VALUE IN HEALTH REGIONAL ISSUES MAY 2020

Table 3. The estimation of quantile regression related to diagnosis and treatment expenses of gastric cancer versus descriptive
variables.

Explanatory 50th percentile 75th percentile 90th percentile 95th percentile

variables
B Confidence B Confidence B Confidence B Confidence
interval (95%) interval (95%) interval (95%) interval (95%)
Constant 9 700 000 (1 050 407 to 2.26E107 (10 900 000 to 5.51E107 (31 300 000 to 1.15E108 (56 000 000 to
18 300 000)† 34 200 000)* 78 800 000)* 174 000 000)*
Sex
Female Ref
Male 4 100 000 (968 571 to 2 329 412 (-1 879 931 to -766 666.7 (-9 363 316 to 764 705.9 (-20 700 000 to
7 231 429)† 6 538 755) 7 829 982) 22 200 000)
Insurance
No coverage Ref
Coverage 4 100 000 (-1 086 224 to 3 264 706 (-3 706 742 to 566 666.7 (-13 700 000 to -700 000 (-36 200 000 to
9 286 224) 10 200 000) 14 800 000) 34 800 000)
Metast
No metast Ref
Metast -2 000 000 (-4 955 398 to -6.35E105 (-4 608 011 to -1 600 000 (-9 713 394 to -2 700 000 (-22 900 000 to
955 397.9) 3 337 423) 6 513 394) 17 500 000)
Length of 2 200 000 (2 105 707 to 2 464 706 (2 337 954 to 2 766 667 (2 507 805 to 3 082 353 (2 437 309 to
stay 2 294 293)* 2 591 457)* 3 025 528)* 3 727 397)*
Age group
Age 26-39 Ref
Age 40-52 -3 100 000 (-10 300 000 to -1.28E107 (-22 500 000 to -3.97E107 (-59 500 000 to -9.40E107 (-143 000 000 to
4 118 508) -3 126 119)† -19 900 000)* -44 600 000)*
Age 53-65 -5 200 000 (-12 100 000 to -1.17E107 (-20 900 000 to -3.37E107 (-52 600 000 to -8.89E107 (-136 000 000 to
1 689 790) -2 391 519)† -14 800 000)‡ -41 800 000)*
Age 66-78 -6 900 000 (-13 800 000 to -1.55E107 (-24 700 000 -4.09E107 (-59 800 000 to -9.87E107 (-146 000 000 to
-32 075.46)† to -6 262 087)‡ -22 000 000)* -51 700 000)*
Age .79 -8 500 000 (-16 500 000 to -1.43E107 (-24 900 000 to -3.52E107 (-57 000 000 to -9.60E107 (-1 500 00 000 to
-546 585.4)† -3 561 769)‡ -13 300 000)‡ -41 600 000)*
Metast indicates metastasis.
*Significant at .001 level.
†
Significant at .05 level.
‡
Significant at .01 level.

coverage had a positive effect on the highest quantile (95%) and between the ages of 40 and 52 had a negative and significant
had a negative effect (b = -700 000; P = .969) on the expenses effect on cost in the quantiles of 70%, 90%, and 95%. Nevertheless,
variable in the lower quantiles. The results of the model with the increasing age of patients, significant levels were
estimation in different age groups of patients showed that being confirmed in all the distributed quantiles. In patients over 66 years

Table 4. The results of gamma regression fitting between diagnosis and treatment of gastric cancer and descriptive variables.

Total cost b Standard error Z P.z Confidence interval

Min Max
Sex 0.1059593 0.0401156 2.64 0.008 0.0273341 0.1845845
Insurance 0.1415995 0.0658542 2.15 0.032 0.0125277 0.2706713
Metast -0.0880848 0.0376956 -2.34 0.019 -0.1619668 -0.0142029
Length of stay 0.0434759 0.0013906 31.26 0 0.0407505 0.0462014
Age 40-52 years -0.2057819 0.0920339 -2.24 0.025 -0.386165 -0.0253989
Age 53-65 years -0.1965785 0.0885918 -2.22 0.026 -0.3702152 -0.0229419
Age 66-78 years -0.2821556 0.0882257 -3.2 0.001 -0.4550747 -0.1092365
Age .79 years -0.2971134 0.1022324 -2.91 0.004 -0.4974853 -0.0967415
Constant 16.84202 0.1113039 151.32 0 16.62387 17.06017
 METHODOLOGY 131

of age (the groups 66-78 years and more than 79 years old), age in every stage after the first stage of the disease, the
was an effective variable with a negative effect on the cost of pharmaceutical expenses account for the largest share of
diagnosing and treating gastric cancer. expenses.19
The gamma regression with logarithmic function had the According to Amani’s study, the first and third priorities were
lowest AIC and was selected as the appropriate model. As the cost of treatment for patients with drug use in the treatment
explained in Table 4, all the descriptive variables had a significant department and in the operating room.21 In the study by Mlcoch
effect on the cost of the disease. The patient's metastatic status et al, the highest costs were also related to drugs and equipment.23
(b = -0.0880848; P = .019) and age had a negative effect on the cost In almost all of the cost groups, the proportion of patients
of the disease. The patient's gender (b = 0.1059593; P = .008), without metastasis had higher costs than patients with
insurance coverage (b = 0.1415995; P = .032), and duration of stay metastasis, but in the surgery group, costs were distributed in an
in the hospital (b = 0.0434759; P = .000) all had positive effects on almost equal proportion among patients with metastasis or
the expenses of the disease. Also, the constant was confirmed in without metastasis (57% for nonmetastasis patients and 42% for
the model with a significant level of significance. patients with metastasis). The results of Izadi et al, however,
Table 4 shows the results of the fitting gamma regression showed that the average cost of treating patients in advanced
between the costs of diagnosing and treating gastric cancer and stages (metastasis: stage 3) is more than the cost of treating
descriptive variables. patients in the first stage of the disease. This confirms this study’s
finding. The study of Sarvepalli et al in 2018 illustrated that people
with advanced gastrointestinal cancers face a higher cost of
Discussion treatment owing to the use of more advanced therapies and more
services.6 Dvortsin and colleagues had a similar result in the study
This study aimed to determine the relationship between the of breast cancer costs in 2016.24 Based on the results of the
costs of diagnosing and treating patients with gastric cancer and quantile regression, in the first 50% of the data distribution, being
explanatory variables, using quantile and gamma regression a man had a positive and significant effect on cost, which may be
methods and comparing the results of the two models with each explained by the tendency of men to ignore presentations of the
other. This was accomplished in one of the special cancer hospitals disease and early referrals to the physician. This behavior happens
in Iran. in advanced stages of disease treatment and leads to higher cost.
The results of the study showed that quantile regression not The gamma regression model reported the same result. The
only can be a suitable substitute for gamma regression, but also, in results of the gamma regression in the study by Mlcoch et al,
some cases, can provide more information than gamma regression however, showed that the gender of the patient was not
for cost analysis. Regression, in addition to its effectiveness against significantly related to the cost of cystic fibrosis, which could be
outliers, allows us to examine the association between gastric due to differences in the studied population and the type of
cancer costs and explanatory variables. The use of the AIC disease in these two investigations.23
information criterion also confirms the properness of fitness of The quantile regression analysis results showed a positive and
quantile regression models. significant relationship between how long the patient stayed in
Lo et al have also demonstrated the superiority of the quantile the hospital and the cost in all cost-sharing quantiles. This was
regression model to traditional regression models, such as consistent with the results of the gamma regression. It makes
generalized linear regression, and recommend the use of quantile sense that with an increasing length of stay (LOS), the treatment
regression for future cost studies.12 costs also increased. Regarding the relationship between cost of
In the present study, men accounted for 76% of the patients disease and insurance coverage, however, the two regression
with gastric cancer and women for 24%, illustrating the higher models showed different results, so that in the gamma regression,
prevalence of gastric cancer among men. Similarly, in Nowruzi Nia insurance coverage had a positive effect on cost, but in the
et al, 74% of the patients were male.18 In the research of Izadi et al, quantile regression, the effect was not significant in any of the
65% of the population studied were male.19 According to a global quantiles, even in the lower quantiles. The insurance coverage had
study about the burden of diseases and risk factors in 2017, gastric a positive and, in the highest quantile (95%), a negative effect on
cancer is among the leading causes of death in men, whereas in the variable of costs. Ultimately, for the reasons given at the
women, it is the second leading cause of death.20 beginning of the discussion, the results of the quantile regression
In this study, the mean age of patients with gastric cancer was model were more reliable.
63 years. The average age in other Iranian studies has also been According to the results of the gamma regression, age had a
reported as between 62 and 65 years old.19,21 The studies negative effect on disease costs, but the estimation of the quantile
demonstrate that the onset of disease usually occurs in the fourth regression model in different patient age groups showed that
and fifth decades of life, and as the age increases, the prevalence being between the ages of 40 and 52 had a significantly negative
increases, with the highest prevalence occurring in the seventh effect on costs in the quantiles of 70%, 90%, and 95%, respectively.
decade of life. According to predictions made in the Japanese Nevertheless, with the increasing age of patients, significant levels
study by Kayko Haga et al, however, the incidence of gastric were confirmed in all of the quantiles of the cost distribution so
cancer and mortality rates will gradually increase in the future due that, in patients over 66 years of age (in the groups 66-78 years
to technological advances in diagnosing and treating gastric and more than 79 years old), age was a justifiable variable with a
cancer and the increase in life expectancies.22 negative effect on the cost of gastric cancer’s diagnosis and
According to the results obtained in this study, the lowest treatment. Moreover, the results of the gamma regression in the
expense was due to biopsy services. This component, as study by Mlcoch et al demonstrated that the patient's age had a
mentioned in the study of Izadi et al, is related in the early stages significantly negatively correlation with the cost of cystic fibrosis
of the disease to the patient's visit services and in the advanced disease. The highest cost was seen at the age of 16. In older people,
stages of the disease to laboratory testing.19 In the study by Amani the costs decreased, which is partly consistent with the current
et al, the cheapest expense was visiting services.21 According to results. Another study in Japan examined the cost of diagnostic
that study, drug expenses are the most important component of and therapeutic endoscopic treatment of gastric cancer patients in
treatment expenses for patients with the disease. In Izadi’s study, two groups: less than 80 years old and elderly (80 years or
132 VALUE IN HEALTH REGIONAL ISSUES MAY 2020

older).14 The results of the multiple linear regression showed that Sanandaj. Scientific Journal of Kurdistan University of Medical Sciences.
2013;18:1–8.
the expenses between the two groups were different. The costs
3. Akbarzadehbaghban A, Esmaeili M, Kimiafar K. Medical information
related to the elderly were significantly higher during management and assessment of direct costs of treatment of lung cancer.
hospitalization. Therefore, age has a significantly positive effect on Health Inf Manage. 2009;5:151–158.
the costs of these two groups, which is not consistent with the 4. Siegel RL, Miller KD, Jemal A. Cancer statistics, 2017. CA Cancer J Clin.
2017;67(1):7–30.
current results. The reason for this difference may be explained by 5. Jo C. Cost-of-illness studies: concepts, scopes, and methods. Clin Mol Hepatol.
the type of costs because the Japanese study only considered 2014;20(4):327–337.
endoscopic costs. 6. Sarvepalli S, Garg SK, Sarvepalli SS, et al. Hospital utilization in patients With
gastric cancer and factors affecting in-hospital mortality, length of stay, and
The limitations of this study include the lack of access to the costs. J Clin Gastroenterol. 2019;53(4):e157–e163.
electronic records of the patients, the absence of using the same 7. Gregori D, Petrinco M, Bo S, Desideri A, Merleti F, Pagano E.
and common literature and terminology among patient records, Regression models for analyzing costs and their determinants in health care:
an introductory review. Int J Qual Health Care. 2011;23(3):331–341.
and the lack of registration and registry systems for cancer 8. Malehi AS, Pourmotahari F, Angali KA. Statistical models for the analysis of
patients. skewed healthcare cost data: a simulation study. Health Econ Rev. 2015;5:11.
9. Pourhoseingholi A, Pourhoseingholi MA, Vahedi M, Moghimi-Dehkordi B,
Maserat AS, Zali MR. Relation between demographic factors and
hospitalization in patients with gastrointestinal disorders, using quantile
Conclusion regression analysis. East Afr J Public Health. 2009;6(suppl 1):45–47.
10. McCullagh P. Generalized Linear Models. New York: Routledge; 2018.
11. Ng VK, Cribbie RA. The gamma generalized linear model, log transformation,
According to the results, the highest costs are related to
and the robust Yuen-Welch test for analyzing group means with skewed and
hospitalization services and the lowest are for biopsy services. heteroscedastic data. Commun Stat Simul Comput. 2018;48(8):2269–2286.
Considering that gastric cancer is the most common type of cancer 12. Lo T, Parkinson L, Cunich M, Byles J. Factors associated with higher healthcare
in men in Iran and the second most common cancer in women costs in individuals living with arthritis: evidence from the quantile regres-
sion approach. Expert Rev Pharmacoecon Outcomes Res. 2015;15(5):833–841.
and that it imposes a great deal of healthcare costs on the health 13. Eckardt M, Brettschneider C, van den Bussche H, König HH. Analysis of health
system, studying the cost of disease is an important economic tool care costs in elderly patients with multiple chronic conditions using a finite
for politicians and policy makers to measure the economic burden mixture of generalized linear models. Health Econ. 2017;26(5):582–599.
14. Murata A, Muramatsu K, Ichimiya Y, Kubo T, Fujino Y, Matsuda S. Endoscopic
and identify the ways they can control and reduce the burden. submucosal dissection for gastric cancer in elderly Japanese patients: an
Without cost-benefit studies, it would be impossible to observational study of financial costs of treatment based on a national
conduct evidence-based decision making, which allows policy administrative database. J Dig Dis. 2014;15(2):62–70.
15. Lee SM, Choi Is, Han E, et al. Incremental treatment costs attributable to
makers to prioritize and decide on the allocation of limited overweight and obesity in patients with diabetes: quantile regression
resources throughout the health system. Thus, health policy approach. Obesity (Silver Spring). 2018;26(1):223–232.
makers and managers should consider disease cost studies to be 16. Lindsey JK, Jones B. Choosing among generalized linear models applied to
medical data. Stat Med. 1998;17(1):59–68.
not only a purely descriptive study, but also the basic economic
17. Acquah G, De H. Comparison of Akaike information criterion (AIC) and
study in the health field. As a result, the support of the authorities Bayesian information criterion (BIC) in selection of an asymmetric price
in defining and developing a strong information system seems relationship. J Dev Agric Econ. 2010;2:1–6.
18. Nourozinia F, Rasmi Y, Otarod M, Golizadeh M, Khadem-Ansari MH.
necessary, as well as research funding in this field.
Epidemiology and histopathology of gastric cancer in Urmia. Urmia Medical
Journal. 2013;24:170–175.
19. Izadi A, Sirizi M, Esmaeelpour S, Barouni M. Evaluating direct costs of gastric
Acknowledgments cancer treatment in Iran - case study in Kerman City in 2015. Asian Pac J
Cancer Prev. 2016;17(6):3007–3013.
20. Fitzmaurice C, Allen C, Barber RM, et al. Global, regional, and national cancer
The present study is the result of an investigation approved by the incidence, mortality, years of life lost, years lived with disability,
Research Chancellor of Mashhad University of Medical Sciences and disability-adjusted life-years for 32 cancer groups, 1990 to 2015:
(code 931179). The researchers want to thank all the professors of the a systematic analysis for the global burden of disease study. JAMA Oncol.
oncology-radiotherapy group at Mashhad University of Medical Sciences 2017;3(4):524–548.
21. Amani F, Sadrkabir M, Ahari SS, et al. Epidemiology of gastric cancer in
and the manager and staff of the medical records at Omid Hospital Cancer
Northwest Iran: 2003-2011. Middle East J Cancer. 2015;6:189–193.
Research Center. 22. Haga K, Matsumoto K, Kitazawa T, Seto K, Fujita S, Hasegawa T. Cost of illness
of the stomach cancer in Japan - a time trend and future projections. BMC
Health Serv Res. 2013;13:283.
REFERENCES 23. Mlcoch T, Klimes J, Fila L, et al. Cost-of-illness analysis and regression
modeling in cystic fibrosis: a retrospective prevalence-based study. Eur J
1. Stewart BW, Wild CP, eds. World cancer report 2014. Lyon, France: Interna- Health Econ. 2017;18(1):73–82.
tional Agency for Research on Cancer, World Health Organization; 2014. 24. Dvortsin E, Gout-Zwart J, Eijssen EL, van Brussel J, Postma MJ. Comparative
2. Veisani Y, Delpisheh A, Sayehmiri K, Rahimi E. Survival rate estimation and its cost-effectiveness of drugs in early versus late stages of cancer; review of the
associated factors in patients with stomach cancer in Tohid Hospital in literature and a case study in breast cancer. PLoS One. 2016;11(1):e0146551.