The databases searched in WoS include WoS core collection databases (including Science Citation Index Expanded, Social Sciences Citation Index, Arts and Humanities Citation Index and Emerging Sources Citation Index) and MEDLINE database. After pre-analysis and comparison, the determined search schema for searching the WoS core collection database and MEDLINE database was as follows: TS = (resilien^*) AND TS = (hospital^* OR medical OR health^* OR care) AND TS = (facility OR facilities OR asset^* OR “built environment” OR “building portfolio” OR lifeline OR equipment OR device^*). “TS” means the topic of an article, and “^*” refers to a fuzzy search. In addition, the determined search schema for searching the Scopus database was as follows: TITLE-ABS-KEY (resilien^*) AND TITLE-ABS-KEY (hospital^* OR medical OR health^* OR care) AND TITLE-ABS-KEY (facility OR facilities OR asset^* OR lifeline OR “built environment” OR “building portfolio” OR equipment OR device^*). Literature published before 2020 (including 2020) was searched from these three databases. The language is limited to English, and the type is confined to article. After data deduplication, 2,695 articles were obtained.

The retrieved results need to be reviewed to ensure that the selected articles meet the requirements for further analysis (28, 29). By reading the abstract, unrelated literature was excluded from detailed review and analysis according to the following criteria:

By further screening, 374 articles were eventually selected for scientometric analysis, among which there are 251 articles from the WoS core Collection database, 72 from Scopus database, and 51 from MEDLINE database. The period ranges from 2001 to 2020.

Figure 2 shows the distribution of HFR-related articles over time from 2000 to 2020, illustrating that HFR research was thriving. In 2000–2007 the number of published articles (8 articles in total) was relatively small. Preliminary explorations were conducted during this period. In 2008–2014, the number of articles fluctuated between 5 and 15 (64 articles). This period belongs to the period of slow growth. Between 2015 and 2020 is the period of rapid growth, in which the number of published articles grew rapidly from 27 to 93. The number of articles during this period is 302, accounting for 80.75% of the total number of articles (374).

Figure 3 shows the countries and institutions' co-occurrence network diagram of HFR research. The network contains 244 nodes and 491 links. Among which, 138 nodes represent countries, and 106 nodes display institutions. The larger the node, the more articles are published in the country or institution (34). The top five are the United States (174, accounting for 46.52% of the total), the United Kingdom (67, constituting 19.91% of the total), Australia (41, occupying 10.96% of the total), Italy (36, taking up 9.63% of the total), and China (35, amounting to 9.36% of the total). It is apparent to see that the United States is far ahead in this field. The complex links between nodes indicate a common phenomenon of transnational cooperation in this area. The thicker the link, the more collaborations and the closer the connection between the countries. The purple circle shows that the United States, the United Kingdom, Canada, and Spain have played a crucial role in international cooperation. In general, the distribution of institutions is consistent with the distribution of countries (regions) (35). The United States has the most active research institutions on HFR, University of Washington (6). It is followed by London School of Hygiene and Tropical Medicine (5) and Politeco di Torino (5), which come from the United Kingdom and Italy with high publication outputs, respectively. Overall, institutions conducting and publishing HFR research are relatively scattered.

The subject category co-occurrence network is shown in Figure 4. The network consists of 168 nodes and 382 links, suggesting that HFR-related research covers 168 subject categories, illustrating that the research is interdisciplinary. The more articles were published in a particular subject area, the larger the node (36). The top five subject categories are engineering (65); public, environmental, and occupational health (63); humans (33); environmental sciences and ecology (28); engineering and civil science (23). In Figure 4, the link between nodes indicates that two topics appear at the same time in the same article. The thicker the link, the greater the frequency. In Figure 4, nodes with high centrality are marked by purple rings (35). Some nodes with high centrality in co-occurring subject categories network are delivery of health care (0.83), ergonomics (0.78), industrial (0.77), and engineering (0.66), which demonstrate that these subject categories are the major turning nodes linking the HFR research in different phases (35).

According to the articles selected from the two databases, the co-occurrence network of keywords was generated, as shown in Figure 5. This network contains 270 nodes and 639 links, which shows 270 keywords. Node size indicates keyword frequency. The top five high-frequency keywords are “resilience” (frequency = 86), “hospital” (frequency = 63), “disaster” (frequency = 41), “healthcare” (frequency = 37), and “healthcare facility” (frequency = 31). The links between nodes are complicated. Multiple keywords frequently appear together in the same literature and are closely interrelated.

This study used the LLR algorithm to generate the literature co-cited network (Figure 6). This network consists of 197 nodes and 418 links. The modularity is 0.8646, indicating that these co-citation clusters can define the research areas of HFR (37). The mean silhouette is 0.8857 (>0.5), so the clustering result is reasonable (26). The silhouette of the major clusters discussed in the study is all over 0.9, reflecting the high homogeneity of the network (38). After excluding the small clusters with a small number of articles (the number <10), 7 clusters (#0 – #6) were finally identified in the literature co-cited network. The label of each cluster was determined by the citing literature in the cluster (26).

Furthermore, the timeline view (Figure 7) reflects the evolution of HFR knowledge, in which the co-citation outbreak of each cluster was shown explicitly. The nodes of high centrality are marked by purple circles. The larger the node, the more times the document is cited (38). The more highly cited articles in a cluster, the more important the cluster is. In recent years, no HFR literature with high citation frequency has been found; thus, citation frequency needs time to accumulate (38, 39). Given the impact on the evolution of HFR knowledge, the literature with high centrality and high citations should be paid attention to. Linking different clusters can also be a potential turning point (34). Furthermore, in different clusters, an article may be highly cited literature and highly citing literature. Seismic resilience of a hospital system, published by Cimellaro et al. (40), is an example. It is highly cited literature at #0 and the highly citing literature at #6. Such articles can also be a potential turning point.

On the basis of the main research content, seven clusters are classified into four knowledge domains (Numbered as KD1-KD4) as follows.

The knowledge domain KD1 focuses on the impact of extreme weather (heat waves, low temperatures, floods, and hurricanes) on healthcare facilities. Measures to reduce the negative effect of climate change are also discussed. Of all the extreme weather, scholars are mainly concerned with extreme heat due to serious climate change problems. This phenomenon can also be seen from the cited and citing references in this domain.

The most cited article in this domain was published by Lomas et al. (41), who proposed the renovation plans for the British hospital wards to cope with the elevated temperatures and achieve energy savings. In other highly cited articles, Short et al. and Lomas and Giridharan, worked on adapting hospital buildings for climate change with the application of data monitoring and building modeling technologies (42, 43). Hajat et al. and Nitschke et al. discussed the effects of heat waves on human health and the increased demand for healthcare facilities (44, 45). Morbidity, demand for ambulance services, demand for emergency medical services, hospitalization rates, and other indicators can be used to measure the impact on human health (14, 44, 45). Furthermore, indicators such as demand for ambulance services, hospitalization rates, and demand for emergency medical services can also be used to reflect the impact on health facilities (44, 45). The most citing article was published by Chand and Loosemore in 2016 (46), which discussed the vulnerability of healthcare facilities in geographical environment, built environment and organizational management under extreme weather. Six improvements were proposed to strengthen disaster management and improve HFR. Additionally, Curtis et al. (14) discussed the impact of climate change on the health and social care system given the healthcare services' provision and demand. Possible resilience enhancement measures, such as renovating hospitals to improve thermal comfort and increasing capacity for population risk identification and health awareness, are also discussed (14). Codjoe et al. analyzed the vulnerability of healthcare facilities and health services to flood and heat wave, and proposed measures to improve HFR in low-income areas (47). Extreme weather can cut off power supplies to healthcare facilities, which will disrupt the continuity of healthcare provision (48). Therefore, power outage problems have attracted the attention of researchers (49–51). A series of response measures were put forward, including preparing emergency generators (48, 50), better detection of power supply failures (49), the rise of distributed energy (52), and microgrids (53), etc.

From the timeline view (Figure 7), since 2008, the knowledge domain KD1 has begun to appear in highly influential articles. In 2009 and 2010, Costello et al. (54) and Hajat et al. (44) discussed the impact of climate change on health and the impact of surging demand on health care facilities. In 2012 and 2013, Lomas et al. (41), and Giridharan et al. (55) discussed the coping measures of hospital buildings and hospital spaces to elevated temperatures. In 2015, Li et al. discussed the impact of heat waves on morbidity and pointed out the need to establish data collection and monitoring systems to guide actions against climate change (56). In the same year, Watts et al. proposed a range of policy responses to climate change (57). Sendai Framework for Disaster Risk Reduction 2015-2030 was proposed in 2015, in which public health and climate change are important parts (58). It can be seen from the recent citing literature that people in recent years are paying more attention to the energy issues and sustainability issues induced by climate change (52). With the frequent threat of global climate change, the term “resilience” is increasingly discussed in healthcare facilities' operation areas around climate change adaptation and disaster risk reduction (59, 60). A series of resilience measures to climate change adaptation are studied in recent years, including enhanced energy supply management, establishing early warning systems to collect climate and morbidity information, ensuring environmental sustainability of healthcare facilities, etc. (14, 57, 61).

This knowledge domain focuses on the impact of wars and epidemics on healthcare systems and measures to improve resilience. In particular, of all the wars and epidemics, the Syrian war and the Ebola virus outbreak are the two major disasters discussed in HFR research.

Epidemic and war lead to a shortage of resources and a rise in mortality. Furthermore, these two disturbances can often be combined to disrupt the supply of healthcare services. During the war, the use of violence to destroy healthcare facilities is increasingly common (62, 63). The destruction of roads, the breakdown of communications, and the threat of death from the war made it difficult for healthcare facilities to obtain adequate resources timely and prevented patients from being able or afraid to go to hospitals for treatment (64–66). The scarcity of health care services and the increasing number of patients further promote the outbreak of epidemics and ultimately lead to the collapse of healthcare systems (65, 66). In this case, many scholars have studied how to improve HFR to ensure the accessibility of healthcare services during wars and epidemics. The most cited article in this domain is published by Kruk et al. in 2015, who pointed out that a resilient healthcare system should identify threats as quickly as possible, be multi-layered, and have broad coverage (67). Furthermore, a resilient healthcare system should have strong links and adequate communication with external organizations or countries (67, 68). During outbreaks of war and infectious diseases, the concept of resilient health systems has been studied in greater depth (69). People pay more attention to the positive role that individual facilities within the system can play (69). In highly cited articles, Gilson et al. (70) and Kieny et al. (71) also believe that giving full play to the role of leaders in the healthcare system, and strengthening the information exchange within and outside the healthcare system will be powerful measures. In the most citing literature, Jamal et al. (66) pointed out that providing physical and psychological support to employees, ensuring organizational flexibility, and establishing good collaboration and communication mechanisms are key points. In the highly citing literature published by Douedari et al. (72), strategic vision, participation, transparency, responsiveness, equity, effectiveness accountability were considered as critical elements for healthcare system governance. In addition, Fouad et al. (62) and Raven et al. (73) stated it was necessary to provide a safe working environment and appropriate incentives for front-line health workers (62, 73). These measures will help them relieve the psychological pressure, and subsequently the continuity of healthcare provision can be ensured to some extent (71). In addition, many cited articles (67, 68) and citing articles (62, 74) emphasized the necessity of establishing an information monitoring system, which is important for the prevention and control of epidemic outbreaks.

The timeline view (Figure 7) shows KD2 active since 2009. A highly cited paper in this knowledge domain was published by Rubenstein and Bittle (75) in 2010 which highlighted the importance of protecting healthcare facilities and medical staff in conflict and discussed the possible effective strategies. In 2014 and 2015, nodes with high centrality appeared. These two articles were published by Kieny et al. in 2014 (71) and Kruk et al. in 2015 (67), respectively. They all discussed the impact of the Ebola virus outbreak on healthcare systems and measures to improve resilience of systems. This may be related to the spread of Ebola virus from 2014 to 2015. The content of discussion in this knowledge domain does not evolve and change obviously.

In this domain, researchers applied various approaches to assess HFR and proposed several measures to improve resilience in the face of disasters (6). “Evaluation framework” consist of a number of resilience measures or indicators for resilience measurement (15, 76). “Functionality” is the most commonly used metrics for describing HFR (77). Particularly, the earthquake is the popular disaster discussed in this knowledge domain.

In the most cited literature, Jacques et al. (78) in 2014 used fault tree analysis to identify factors that affect critical hospital services, namely, non-structural component failures and external utility supply disruptions. In addition to the above two factors, Mitrani-Reiser et al. (79) and Kirsch et al. (10) also considered the damage to structural components, equipment and workers. Achour et al. (80) focused on the influence of public utilities, in which the impact of utility disruptions on healthcare facility was quantified. Some research utilized the modeling-based approach to assess HFR. For example, Cimellaro et al. (3) in 2011 and Cimellaro et al. (8) in 2016 proposed meta-models and discrete event simulation (DES) to assess resilience in the emergency department. In these studies, “patient waiting time” was used as a final measure of resilience (3, 8). Furthermore, the most citing article was published by Zhong et al. (15) in 2014, who developed a framework with eight areas to measure the resilience of hospitals to disasters. On this basis, factor analysis was conducted to extract four key capabilities (including emergency medical response capability, disaster management mechanisms, hospital infrastructural safety, and disaster resources) to measure HFR (15). In the same year, another framework was proposed by Zhong et al. (21) using literature review method. The key assessment areas are classified as robustness, redundancy, resourcefulness and rapidity. The functionality or performance of healthcare facilities is a commonly used resilience measure in many HFR assessment methods (77). In the highly cited articles, Cimellaro et al. (81) and Cimelaro et al. (40) quantified HFR by considering the loss and recovery phases of a healthcare facility's functionality. Furthermore, Khanmohammadi et al. (6) simulated the dynamic recovery process of hospital function after earthquakes. This model can help decision-makers assess the resilience of post-earthquake hospitals and determine the optimal use of available resources (to help hospitals recover their functions) (6). The citing article, Cimellaro et al. in 2019, used “patient waiting time” to measure the performance of a healthcare facility network in emergencies and proposed two resilience improvement strategies, namely, reallocating existing resources and building new emergency departments (82).

From the timeline view (Figure 7), the citation phenomenon appears among #0 “evaluation framework,” #2 “functionality,” #4 “earthquake,” which is sufficient to illustrate the intimate relationship between these three clusters. Among the three clusters, #2 “functionality” appeared the earliest among the top seven clusters, in which the citation outbreak appeared since 2003. The first widely cited article in this domain was published by Bruneau et al. (83) in 2003. In this study, “4R,” namely robustness, redundancy, resourcefulness, and rapidity, is used to summarize the concept of resilience for the first time (83). Since then, scholars have proposed many ways to evaluate HFR. Bruneau and Reinhorn (84) in 2007, Cimellaro et al. (81), and Cimellaro et al. (40) used probability functions to quantify the loss and recovery of performance/function in healthcare facilities. It is also common to identify key indicators and develop a framework model for assessment by combing the key capabilities of HFR (15, 21). Furthermore, scholars tend to evaluate HFR by modeling the interactions and behavioral changes of sub-systems within a facility, such as Cimellaro et al. (3) in 2011 and Cimellaro et al. (8) in 2016. With the deepening of the research, the research content gradually expanded from individual healthcare facilities to healthcare networks or systems. In 2010, Cimellaro et al. (40) assumed that the performance of healthcare networks was simply equal to the aggregation of individual facilities' performance. Based on an empirical study, Jacques et al. (78) in 2014 further discussed the positive effect of service sharing among hospitals on healthcare network performance. A recent citing article (85), which Hassan and Mahmoud published in 2020, established a framework model for assessing the resilience of post-earthquake healthcare systems. This model considered the quality and quantity of services provided by the hospitals, the demand and arrival rate of patients on the hospitals, and the interaction between hospitals and other infrastructures (85). HFR assessment studies are increasingly complex and profound.

This knowledge domain mainly discusses the application of technologies in the healthcare industry. With the help of these technologies, healthcare facilities can process large amounts of information quickly, and provide more efficient, higher-quality services to more patients in less time.

A common application area with technologies is optimizing the layout of healthcare facilities. For one thing, in resource-constrained settings, well-located facilities can greatly improve access to health services, ensuring accessibility at the community or city level while making full use of resources. The most cited article (86) by Ferguson et al. in 2016 used Geographic Information System (GIS) to identify critical and efficient paths for patients accessing health services. Healthcare facilities are prioritized at the sites with large population congregations and intersection of these paths (86, 87). For another, with the help of GIS, the location of healthcare facilities can avoid low-lying areas, flood and other disasters or reduce the probability of such events (47). Another application area discussed frequently is decision support for resource allocation. It is reflected in prioritizing the provision of resources to core functional components to ensure service provision, and facilitating the reasonable matching of resources and demands to speed up the recovery of facility performance. In an article published in 2010 by Paturas et al. (88), a Hospital Emergency Support Functions (HESF) model based on the personnel database was developed urgently to review all functions and staff in the hospital. The most suitable personnel will be assigned to key positions based on the importance of the function and the capacities of the people. Additionally, medical diagnosis and treatment is popular in recent years. The COVID-19 outbreak has boosted popularity of telemedicine technologies. As the citing article (18) by Haldane et al. in 2020 and the citing article (89) by Bhaskar et al. in 2020 pointed out, telemedicine can ensure physical distance between people, reduce the route of transmission of the virus, and provide health services continuously.

From the timeline view (Figure 7), the knowledge domain KD4 began to rise around 2010, which is a relatively new research topic. The development of technologies also takes time to accumulate. These may be the reasons why this domain has no large citation outbreak and high centrality nodes. In recent cited literature, COVID-19 is an unshirkable research content (18, 90). They illustrate the impact of epidemics outbreaks on the use of technologies in healthcare facilities. With the advent of the data age, the everchanging science and technology will make this field of knowledge active for a long time.

As shown in Figure 8, this framework depicts three critical research parts of HFR: disturbance scenario analysis, resilience assessment, and resilience enhancement. Their states are described as follows.

Reviewing the articles relating to HFR in 2000–2020, this study identifies the areas that are still requiring further studies. As Figure 8 shows, these areas are summarized into the knowledge gaps of HFR and set the directions for future research.

The first knowledge gap is related to understanding and monitoring the risks for uncertain adverse events. The resilience concept emphasizes the resistant capacity of the system to understand and prevent any possible hazards (including emerging risks) (98). In terms of the functionality of healthcare facilities, monitoring the risks of public health and the loopholes of the healthcare systems are essential for the healthcare facilities' disaster prevention (67). The traditional monitoring systems usually collected public health data from clinical cases, which are labor-intensive and time-consuming (120–123). This method is inefficient for dealing with the emerging risks, particularly the epidemics with a fast spread and long incubation period (103). Taking COVID-19 for example, in some countries without powerful healthcare support and warning mechanisms, an average delay of 29 days was observed from the onset of symptoms to the detection of the epidemic outbreak (124). This delay finally resulted in the widespread COVID-19 and huge pressure on healthcare systems. Thus, studying the hazard monitoring system for healthcare facilities is urgent. Specifically, monitoring the data of climate change (e.g., the rising temperature and possible floods) and exploring their impact on public health can be noted for the early identification and monitoring of public health events (56). Furthermore, given that the clues of risks are probably hiding in the public, how to use the data from social media for monitor the public's reactions is also worth further research, in which collection methods and data cleaning technologies are critical points.

The second knowledge gap is the modeling and simulation of system interdependency. A range of recent HFR studies discussed the impact of disturbances on healthcare facilities (14, 43, 80), but few studies can use the “structure-based” method to quantify the resilience level of healthcare facilities (112). The main reason for this limitation is that existing studies do not fully support modeling and simulation of system interdependency and cascading effects for healthcare facilities (112). In terms of system interdependence, the operational healthcare facility and its emergency management heavily depend on external lifeline services, internal equipment systems, healthcare organizations, and social units (80, 125). All these dimensions are highly interconnected and mutually interdependent. In disasters, the performance deterioration of healthcare facilities can be easily amplified because of interdependencies. Thus, for HFR research, future research would characterize the structure of complex healthcare facility systems and identify approaches to simulate this interdependent system.

The third knowledge gap is related to considering the cascading effects of healthcare facilities. How disturbances cascade through the interdependent systems has to be assessed to estimate the performance deterioration and mitigate the consequences of failures (125). Some studies proposed the methods of revealing cascade failures for interdependent critical infrastructures. For example, Utne et al. developed a cascade diagram to represent the cascading failures across the interdependent critical infrastructures under the accident scenario (126). Lam and Tai integrated network and fuzzy set theory to reveal the cascade effect from a disruption (127). Ouyang reviewed the approaches of modeling interdependent critical infrastructure systems, including empirical approaches, agent-based approaches, and network-based approaches (112). Those studies focused on the general essential infrastructure systems, such as the power grid, telecommunications, transportation, and water supply systems. However, in-depth research on modeling healthcare facility systems is still lacking. Furthermore, these proposed modeling methods for general infrastructure systems can also provide directions for HFR studies in the future.

The fourth knowledge gap is concerning the data collection for HFR modeling and simulation. All proposed resilience assessment approaches require a lot of relevant data, such as the healthcare facilities' system structures, organizational structures, operational data, emergency procedures, performance data, experts' experience, and historical events' data (112, 128). To collect these data is generally difficult because of a series of reasons. For example, much historical data on previous disturbances in healthcare facilities are incomplete and imprecise because of the backward information recording and preservation methods or awareness (129, 130). This limitation directly blocked the use of function-based approaches for empirical resilience assessment and data basis to validate other modeling and simulation methods (112). Furthermore, the healthcare facilities' operational data and their performance metrics are usually confidential because of safety issues or commercial secrets. Despite scholars and practitioners appealed to develop the data collection mechanism for events data in healthcare facilities, data collection or accidents reports for HFR studies have no standard. Thus, developing a unified database for monitoring and collecting the events data in healthcare facilities is essential. The following data standards and data analysis methods can be further researched for HFR studies.

The fifth knowledge gap is related to integration and co-simulation for HFR assessment. The previous review of resilience assessment approaches depicts that each approach has advantages and disadvantages. Function-based approaches can dynamically and directly reflect resilience capacities by the changing macro-level functionality/performance of healthcare facilities (40, 81). However, they fail to identify the critical components within healthcare facility systems that can significantly impact resilience because they did not consider the micro-level system structures, interdependences, and response behaviors (6, 128). Accordingly, the structure-based approaches only used topological indicators to represent resilience in a static way, which could not fully reveal resilience's “process-based” characteristics in a time scenario. To develop the strong points and avoid the weak points for HFR assessment studies, an open modeling framework to capture the macro-level performance indicators and the micro-level topological indicators is more desired for practical applications. This framework will involve integrating modeling and simulation methods for HFR, and specifically exploring the relationship between the micro-level topological indicators and the macro-level performance indicators in a mathematical way is necessary.

The sixth knowledge gap has to do with proactive adaptive management of healthcare facilities. Most existing studies give more efforts to existing function recovery, but little attention is devoted to recognizing the healthcare facilities as an adaptive system (92). After the disaster is completely gone, the system of healthcare facilities had been re-stabilized. In most cases, this stable state is different from the stable state before the disaster. This is due to the impact of disasters and measures implemented to ensure the functioning of healthcare facilities (12). Exploring differences between these two steady-states and their causes will be meaningful for discovering the pathways of strengthening future resilience in a system view (12). Additionally, tightly linking the recovery and learning experience to preparedness is crucial although often neglected once the function is recovered (12). Strengthening the resilience of healthcare facilities in an adaptive view is not only for improving the current system but is also better for response to future threats scenarios.