GIS-Enabled Multi-Criteria Assessment for Hospital Site Suitability: A Case Study of Tehran
1. Introduction
Rapid urban expansion and population growth drive the demand for essential services, particularly healthcare [1]. Access to these services significantly enhances the quality of urban living. Among these services, hospitals play a pivotal role, making the selection of their locations of paramount importance [2,3]. The unequal distribution of hospitals within cities poses a significant challenge, hindering residents’ access to healthcare services [4] especially in developing nations due to urban growth, resource constraints, and construction expenses [5]. The selection of suitable healthcare locations involves the consideration of multiple, often interconnected criteria [6]. Therefore, an efficient protocol for site selection is imperative [3], relying on Multi-Criteria Decision-Making (MCDM) techniques [2,7]. MCDM is the preferred approach for addressing the complexity of hospital location requirements [8], aiding in identifying the best choice when criteria conflict [9,10].
Geographic Information Systems (GIS) play a crucial role in incorporating geographical considerations into location selection processes, thereby enhancing their effectiveness [11]. Serving as a decision support system, GIS complements the decision-making system, MCDM, collectively bolstering spatial analysis capabilities on different sources of data [11,12]. Accordingly, health-based information systems serves as a versatile analytical tool across various domains [13,14,15,16,17], contributing to increased precision while concurrently reducing errors, time, and costs [18,19,20,21]. In the realm of hospital location analysis, common techniques include the Analytical Hierarchy Process (AHP) and GIS-based models, often adopting hybrid approaches [22]. AHP frequently serves as the cornerstone of research in this field [7,23], often combined with GIS, Fuzzy AHP, or other strategies [24,25,26].
New methods and hybrid methodologies are constantly being used and implemented in other fields, and research has always been conducted on the suitability of hospital sites. New research offers new methodologies to address the shortcomings of previous studies or to introduce new topics. The present study aims to provide a comprehensive survey and comparison between conventional hospital site suitability methodologies and less studied methods. The previous methodologies mainly include three stages: providing spatial information of the criteria, criteria weighting, and combining criteria. The first stage is always done by GIS. The third stage is done by Weighted Linear Combination (WLC) in most cases. The second stage has been done by different weighting methods and has a high potential for improvement and updating. Also, the most important stage in an MCDM methodology is the weighting process. In the present study, a conventional methodology including GIS, weighting process and criteria combination is selected, and the goal is to improve this methodology by using different weighting methods. To achieve this goal, three weighting methods, including AHP, Best-Worst Method (BWM), and Step-wise Weight Assessment Ratio Analysis (SWARA) were used, and by a sensitivity analysis approach the most stable and reliable methodology was determined. In the following, the advantages and disadvantages of the weighting methods used in the present study are explained and compared.
The AHP weighting method is the most conventional and widely used weighting method in previous research in terms of hospital site suitability and other applications. This method is simple and efficient but has some shortcomings. The high number of pairwise comparisons and an inability to consider the relationships between criteria are the shortcomings of this method. The high amount of pairwise comparisons can lead to miscalculations [27]. In addition to the issues mentioned, the rank reversal problem is another shortcoming of this method [17]. Basically, to deal with these shortcomings, the Analytic Network Process (ANP) method has been introduced during development of the AHP method [28,29]. However, the design of the network structure, its implementation, and the very high pairwise comparisons of the ANP method are its shortcomings.
The new weighting method of BWM greatly reduces the number of pairwise comparisons compared to the AHP and ANP methods. The BWM method has been quickly adopted in various applications and the desirability of its performance has been determined [30,31]. This method significantly reduced pairwise comparisons and increased the ease of implementation and the consistency of the results, and consequently increased the reliability of the results [32].
Although the SWARA method was introduced before the BWM method, it has received less attention in research related to site suitability and in combination with GIS. Of course, new studies [33,34] have used this method in different applications and in combination with GIS. However, it has not been considered in hospital site suitability applications. SWARA can be successfully used instead of other MCDM and weighting methods, such as AHP and ANP [17]. The SWARA method offers distinct advantages by quantifying disparities in criteria importance and considering the perspectives of the decision-making panel [35,36]. This method greatly reduces the number of pairwise comparisons compared to other methods such as AHP, ANP, and even BWM. When the number of criteria increases, this feature is an advantage to the decision-maker. In addition to increasing the accuracy and reliability of the results, this significant reduction of pairwise comparisons also improves the ease of implementation [33,34].
The aim of this study is to determine potential hospital sites within the boundaries of the study area in Tehran, Iran, with the help of GIS-based MCDM (AHP, BWM, and SWARA). In this study, three hybrid methodologies—GIS-based AHP-WLC, GIS-based BWM-WLC, and GIS-based SWARA-WLC—were employed to assess hospital site suitability. The objectives of the study are listed as follows: (1) determination of criteria that are relevant in hospital site suitability and preparation of their spatial layers; (2) determination of criteria weights based on AHP, BWM, and SWARA methods; (3) determination of suitable areas in terms of hospital site suitability based on AHP, BWM, and SWARA methods; (4) analysis of the performance of each methodology in terms of hospital site suitability; (5) performing sensitivity analysis to survey the stability of the results obtained from different methodologies, and determination of the most stable method.
As can be seen in Section 2, most studies of hospital site suitability used AHP in their proposed methodologies. Also, some of them used BWM, but almost no research has used SWARA in its methodology. Therefore, the first contribution of the present study is the use of the SWARA method in terms of hospital site suitability. In addition, studies that compare different methods are always needed, hence, the second contribution of the present study is the surveying, evaluation and comparison of three different weighting methods in term of hospital site suitability, which has not been considered in previous research. Finally, for the third contribution, the present study has used a feasible and appropriate sensitivity analysis process in order to evaluate the stability of the results of the different methods, which has not been used in previous related works.
The rest of this paper includes the following sections. Related works are presented in the Section 2. The materials and methods used are presented in the Section 3. The experimental results are presented in the Section 4. Finally, the discussion and conclusion are presented in the Section 5 and Section 6, respectively.
2. Related Works
Numerous researchers have put forth distinctive hospital site selection approaches. Hospital site suitability can be investigated as an MCDM problem [2], an optimization problem [37], and as a classification or regression problem [38,39]. In terms of hospital site suitability or hospital site selection, usually the MCDM method is used. Hybrid methods include several MCDM methods, and the combination of MCDM methods with GIS has usually been given more attention in previous research. In the following, some of the more relevant studies are presented.
Yazdi et al. [40] employed a hybrid method for ranking nine candidate sites for hospital site selection. They used BWM in combination with Pythagorean fuzzy numbers in order to determine the criteria weights, and used the Evaluation by an Area-based Method of Ranking (EAMR) for hospital candidate site ranking. Zandi et al. [4] introduced a hybrid methodology for evaluating hospital site suitability in Tehran, incorporating GIS, Criteria Importance Through Inter-criteria Correlation (CRITIC), Shannon Entropy (SE), Dempster-Shafer Theory (DST), and Order Weighting Average (OWA). At first, they selected 10 hospital candidate sites and calculated the criteria values for each site. In the next step, they calculated the criteria weights by two objective weighting methods (CRITIC and SE) without need for experts’ opinions. Then, they fused the results of the two objective weighting methods by DST. Finally, they performed candidate site ranking by the OWA method, but they did not perform a sensitivity analysis on the candidate sites ranking.
Aydin & Seker [41] devised a hybrid approach that integrates Delphi, BWM and fuzzy Technique for Order of Preference by Similarity to Ideal Solution (TOPSIS). They used the Delphi method for determining the criteria and candidate site, BWM for criteria weighting, and fuzzy TOPSIS for candidate sites ranking. They also performed a sensitivity analysis on the candidate sites ranking. Based on their study, the proposed methodology was stable and results were reliable. Boyacı & Şişman [42] proposed an innovative methodology by integrating fuzzy AHP, GIS, and TOPSIS in their approach. They used GIS for preparing criteria layers, fuzzy AHP for criteria weighting, and TOPSIS for candidate sites ranking. Halder et al. [24], in order to assess hospital site suitability in India, proposed a hybrid approach including GIS for preparing criteria layers, AHP for criteria weighting, and WLC for criteria layer integration. They also suggested four suitable sites for building a new hospital. Tripathi et al. [3] performed a comparison of GIS-based AHP and fuzzy AHP in order to assess hospital site suitability in India. Based on their study, fuzzy AHP may be more suitable for hospital site suitability. Other research, such as that of Şahin et al. [23] and Vahidnia et al. [43] used AHP, while Lin & Tsai [44] used ANP and TOPSIS, and Kumar et al. [45] used fuzzy extended elimination and choice expressing reality (ELECTRE) in order to assess hospital site suitability. Table 1 shows a summary of the previous research. In previous methodologies, GIS was primarily used for preparing criteria information, and the AHP method was used to weigh the criteria. The BWM Weighting method has been relatively underutilized in research, and the SWARA method was not employed in the reviewed studies.
5. Discussion
Based on the suitability maps generated using the AHP, BWM, and SWARA weighting methods, the study designates areas within the study region as ‘very low suitability’, ‘low suitability’, ‘moderate suitability’, ‘high suitability’, and ‘very high suitability’ for the construction of new hospitals, with varying percentages ranging from 10–11% to 26–29%.
Our research findings reveal significant similarities in the calculated criteria weights between the AHP and BWM methods, with the weights nearly equal. In contrast, studies conducted by Ajrina et al. [65], Sahraei et al. [66], and Tan et al. [67], did not observe such pronounced similarities. Both methods consistently assign notably higher weights to the first three criteria (distance from existing hospitals, distance from main roads, distance from green spaces) and considerably lower weights to the last three criteria (distance from petrol stations, distance from residential areas, and population density). This similarity may be attributed to the use of the same preference scales in both methods. Notably, the congruence in results across the two methods aligns logically, given that the same experts were involved in the assessment.
A comparative analysis between the BWM and AHP methods indicates that the BWM method outperforms AHP, primarily due to the significantly reduced number of pairwise comparisons. Specifically, the BWM method necessitated only 19 pairwise comparisons, while the AHP method required 55 pairwise comparisons. This preference for BWM aligns with previous research, which has also favored BWM over AHP [68].
The calculated criteria weights derived from the AHP and BWM methods significantly differ from those obtained through the SWARA method. However, the ranking of criteria weights in SWARA closely aligns with that of AHP and BWM. In the SWARA method, weight differences among criteria are minimal, suggesting that all criteria carry significant importance in the decision-making process. In contrast, both AHP and BWM methods assign considerably higher weight to some criteria while diminishing the importance of others. Sharma et al. [68] found that the fuzzy SWARA method outperforms AHP and BWM methods, which is consistent with our findings in the present study. Our results, based on the hospital suitability map created using the SWARA weighting method, reinforce the superior performance of SWARA over AHP and BWM, primarily due to the simpler calculations required by SWARA (only n−1, or 10, pairwise comparisons).
It is noteworthy that SWARA demands fewer input data (expert opinions and pairwise comparisons) compared to AHP and BWM, resulting in significantly reduced complexity in its calculations. The implementation of SWARA is notably more straightforward than the other two methods, leading to a lower probability of calculation errors and more reliable results.
Furthermore, the high correlation observed among the weights and site suitability maps derived from all three methods underscores the suitability of SWARA as a viable alternative to AHP and BWM. Across all three weighting methods, distance from existing hospitals consistently emerges with the highest weight, aligning with previous studies by Tripathi et al. [3] and Dutta et al. [69]. It is worth noting that this contrasts with findings in studies like Halder et al. [24] and Zandi et al. [4], where distance from existing hospitals was assigned a notably lower weight. Other studies, such as Boyacı & Şişman [42], Halder et al. [24], Zandi et al. [4], Zolfani et al. [70], and Beshr et al. [26], have identified different criteria, such as distance to main roads, distance to residential areas, distance from green spaces, distance to industrial areas, and population density, as the most influential factors in hospital site suitability. In our study, distance from main roads, consistent with Halder et al. [24] and Zandi et al. [4], emerges as an important criterion in hospital site suitability. Interestingly, in our study, distance from green spaces and distance from industrial areas are assigned the lowest weights among criteria, contrary to Zandi et al. [4], which assigns a very high weight to distance from green spaces.
In the present study, a sensitivity analysis process based on the OAT approach was performed. The OAT results showed that GIS-based SWARA-WLC is the most stable model for hospital site suitability when compared with GIS-based AHP-WLC and GIS-based BWM-WLC. The amount of change in suitability class of pixels in the study area in GIS-based SWARA-WLC was much lower than in the other two methods. However, the change in the suitability class of pixels due to the change in the weights of the distance from existing hospitals and distance from main roads criteria in GIS-based SWARA-WLC was also high, and almost equivalent to the other two methods. In general, the stability of GIS-based SWARA-WLC was the highest, followed by GIS-based BWM-WLC. GIS-based AHP-WLC was unstable when changing the weights of all criteria.
Hospitals serve as crucial pillars in delivering healthcare services to communities. The World Health Organization recommends having 3.5 hospital beds per 1000 people, but developing countries typically fall short of this standard [45]. Tehran exhibits an improved situation with 2.8 hospital beds per 1000 people in 2020 [71]. Nevertheless, the distribution of hospitals and hospital beds within the city remains suboptimal. The studied area, encompassing Districts 21 and 22 of Tehran, houses just three hospitals, resulting in approximately 0.45 hospital beds per 1000 people. Given the trajectory of urban development, population growth, and the potential for natural disasters and epidemics, it becomes imperative to enhance the existing healthcare system and augment the number of hospitals and hospital beds.
Limitations
This study exhibits a few notable limitations. Firstly, the omission of key criteria, such as land cost and vulnerability to natural disasters, both crucial in hospital location decisions [3,4,43], was due to the unavailability of their respective spatial data layers. Additionally, the spatial distribution of patients and diseases, a highly influential criterion, was not included in this research due to data unavailability. Future research endeavors should consider incorporating these essential criteria to enhance the comprehensiveness of the assessment.
In recent years, data-driven weighting methods, which do not rely on pairwise comparisons and expert opinions, have gained traction in various applications, including hospital site suitability [2,4]. While addressing intricate issues like hospital site suitability typically involves the input of experts, it is advisable for future research to explore a comparative analysis between knowledge-driven methods, like those utilized in this study, and data-driven methods such as CRITIC [72], Robustness, Correlation, and Standard Deviation (ROCOSD) [73], and Method based on the Removal Effects of Criteria (MEREC) [74].
6. Conclusions
This study introduces a robust scientific framework that leverages GIS in conjunction with various MCDM weighting methods to create a hospital suitability map for Tehran. In our analysis, we employed three distinct MCDM weighting methods: AHP, BWM, and SWARA, for the development of site suitability maps. Notably, these methods have been relatively underutilized in the context of hospital site suitability assessments. Additionally, our study utilized WLC as the decision rule to integrate the spatial layers of criteria. Based on the sensitivity analysis, GIS-based SWARA-WLC was the most suitable and stable model for hospital site suitability in the study area. Moreover, the proposed method addresses hospital site suitability by considering criteria selected from previous research. Accordingly, it can be applied to other regions and enhanced with additional criteria. Thus, researchers can incorporate new criteria into this methodology based on the requirements of their study area, ultimately enhancing the quality of decision making.
To achieve the average hospital bed availability in Tehran, the studied area requires an additional 2206 beds alongside the existing facilities. Establishing a hospital with 200 beds would provide a marginal improvement, resulting in approximately 0.67 hospital beds per 1000 people. However, this falls short of the ideal distribution of hospital resources.
Future research could benefit from exploring alternative MCDM decision rules to potentially enhance the accuracy of the hospital site suitability map. The proposed methodology presented here holds promise for expansion to other cities in Iran in future research endeavors, facilitating a broader understanding of hospital location suitability across the country. Future studies should consider data related to hospitals and the economic status of people living nearby in order to better understand the suitability of the proposed methodology for hospital site suitability in addressing the needs of the community.
