Risk profiling of zoo and aquaria accidents: a historical and contextual analysis of severity by species and human factors

1 Introduction

Zoos, parks, and aquaria have evolved from historical menageries into modern institutions with central roles in biodiversity conservation, public education, and ex-situ animal’s management. Today, these facilities host a wide range of wild species under human care and receive hundreds of millions of visitors annually (Gusset and Dick, 2011; Miranda et al., 2023), making them important interfaces between human, animals, and environment. As such, zoos, parks, and aquaria represent complex socio-ecological systems in which animal welfare, human safety, and public health are closely interconnected (Spooner et al., 2023). Despite substantial advances in veterinary medicine, animal welfare science, enclosure design, and staff training, zoological facilities remain environments in which incidents and accidents (definition by the European Agency for Safety and Health at Work) may occur. These events arise from the simultaneous presence of animals, visitors, and personnel, as well as from direct and indirect interactions among them. Although most human-animal interaction in zoos, parks, and aquaria are managed safely, injuries and – in rare cases – fatalities have been reported, underscoring the need of systematic evaluation of risk factors associated.

From an occupational health perspective, zoo, park, and aquarium staff – including keepers, veterinarians, aquarists, trainers, educators, and any other workers – are routinely exposed to physical, biological, and zoonotic hazards, that can happen during daily activities such as feeding, cleaning, handling, and medical procedures. On the other hand, close proximity associated with training, husbandry, and veterinary activities is not behaviorally neutral for animals and may limit avoidance opportunities. Such interactions may apply cognitive and physical demands, and shape stress responses, particularly in species with complex behavioral needs. Professional training and adherence to standardized safety protocols are therefore essential components of public health and animal welfare in these settings. Visitors may also be exposed to risks linked to facility design, management policies, or misinterpretation of animal behavior, mainly in contexts where close-encounters and interactive experiences are permitted such as touching pools or petting areas (Casamitjana, 2004; Higgs et al., 2025). From an ethological perspective, visitors’ interactions – especially those involving children –may constitute unpredictable stimuli for animals housed in environments with restricted options to retreat. Such encounters, by amplifying defensive responses, raise concern not only for human safety but also for animal welfare and One Health outcomes. In this context, One Health refers to the integrated framework promoted by the World Health Organization (WHO), the Food and Agriculture Organization (FAO), and the World Organization for Animal Health (WOAH), which recognizes the interdependence of human health, animal welfare, and environmental conditions. Risks perception and safety awareness among zoo visitors are influenced by demographic factors, visiting patterns, sex, and age, which may affect compliance with safety instructions and behavioral guidelines (Ginsburg and Miller, 1982; Siew et al., 2018; Welsh et al., 2025). In addition, unauthorized access to animal enclosures by intruders, although uncommon, represents a distinct unpredictable source of risk. Zoo, parks, and aquaria have legal and ethical responsibilities to ensure the safety of staff, visitors, and animals, yet the extent to which animal-related injuries occur across different institutional contexts remains poorly characterized.

Although the number of zoos, parks, and aquaria is estimated to be globally in the thousands, yet no comprehensive international registry exists. Visitor numbers are similarly difficult to quantify. According to a recent estimate from AZA (Association of Zoos and Aquariums), global attendance reaches several hundred million annually, generating substantial employment and economic impact, although these figures do not encompass all facilities worldwide (AZA, 2025). Marked heterogeneity in governance, resources, species composition, enclosure design, and management practices across facilities may contribute to substantial variation in accident risk at national and international levels. Nevertheless, the scientific literature addressing human injuries caused by animals in zoos, parks, and aquaria remains fragmented and lacks comprehensive global perspective. The aim of the present study is to provide a systematic overview of documented animal-related accidents involving staff and visitors in zoos, parks, and aquaria worldwide and to characterize these events according to the type of individuals involved (staff, visitors, intruders), animal taxa, and context of occurrence. Moreover, we want to try to identify recurring patterns and knowledge gaps relevant to animal welfare, occupational safety, and One Health.

A better understanding of these events is essential for developing evidence-based strategies to reduce risk and improve safety in zoological facilities that house exotic and wild animals. The more appropriate the approach, the greater the support for an animal-centered management which raises public awareness and education, and aligns with the facilities’ broader objectives, such as biodiversity conservation and scientific research for reintroduction into the wild.

2 Material and methods

2.1 Data sources

This study employed a retrospective descriptive approach to identify and analyze documented animal-related accidents only occurring in zoos, parks, and aquaria worldwide, regardless of facility size or professional status, provided that animals were kept as part of a public zoological collection. Cases involving exotic animals kept as private pets, animal used in circuses, or animal housed in sanctuaries were excluded a priori, as these settings differ substantially from zoological institutions in terms of management objectives, human-animal interaction, regulatory oversight, and data comparability. The large number of excluded records reflects the breadth of the initial search rather than a comparative assessment of occident frequency across different human-animal systems. For the purpose of this study, an accident was defined as an event that resulted in physical injury or death of humans, in accordance with the definitions provided by the European Agency for Safety and Health at Work (EUROSTAT, 2025). We exclusively considered events resulting in physical injury or death to humans caused by animals housed under human care; injuries or death of animals themselves were not considered for the purpose of data inclusion. Cases were collected through a systematic search of publicly accessible sources, including peer-reviewed scientific literature, institutional reports, official press releases, and verified media coverage. This multi-source strategy was adopted due to absence of a centralized international reporting system for accidents occurring in zoological institutions. Unfortunately, data regarding zoos and aquariums accidents are collected mainly internally by facilities or regulatory bodies and are rarely consolidated in academic publications or – when found in scientific papers – too often they lack substantial information such as location, year, and context. Consequently, relevant information is often dispersed across gray literature (Rees, 2023), including newspapers, professional associations, and animal welfare organizations. Low-severity accidents are likely underreported, whereas accidents resulting in serious injuries, fatal outcomes, or involving children, are more consistently documented in public sources. To reduce reporting bias and enhance data reliability, online sources were independently identified and cross-checked by multiple operators. All the extracted information is provided in a downloadable Excel file (Supplementary Dataset S1), which contains two sheets: accidents and source.

2.2 Case selection, eligibility criteria, and data extraction

An adapted PRISMA (Preferred Reporting Items for Systematic Review and Meta-Analysis) framework was used to document the selection, screening, and inclusion of cases (Figure 1). The eligibility criteria included only physical accidents that occurred in zoos, parks, and aquaria worldwide and did not allow inference on comparative risk between zoos and other forms of animal use. Zoonosis or incidents involving exotic animals raised as private pets, animals used in circuses, or animals outside institutional zoological settings were excluded. The initial search yielded 11,375 records from online platforms, academic journals, newspapers, and media outlets (Supplementary Dataset 1, sheet “source”). After removal of duplicates and exclusion of records that did not meet eligibility criteria (n=10,328), a total of 1,047 unique accidents remained included in the final dataset. The included cases span more than two centuries, from 1815 to the present.

For each included case, the following variables were extracted when available (Supplementary Dataset 1, sheet “accidents”): temporal and geographical information (year, country, and state); habitat type (terrestrial, marine, or freshwater); the animal species or taxonomic group involved; the role of people involved as staff (keepers, vets, trainers, owner, divers, guides, and other employees), or visitor (general visitor, unauthorized individuals, and children); the number of victims per accident; the injury severity, classified using a five-level ordinal scale (0=threatening; 1= unstitched lacerations or affected finger/toes; 2= deep laceration; 3= fractures, serious injuries to internal organs; 4=fatal outcome); the context and description of the cause of the accident (Table 1) and the ID of the animal involved. The context category used are descriptive rather than judgement based. Terms such as “provoked” are employed to denote structured or intentional human demands, reduced predictability, or limited animal avoidance options (e.g., training, live encounters, intrusions), whereas “human error” refers to procedural or operational failures without intentional interaction (e.g., failure to secure barriers, entering enclosures alone, ignoring safety protocols). “Found dead” cases reflect the absence of witnesses or recording and do not constitute a distinct behavioral mechanism. The classification aligns with ethological concepts of control, predictability, and behavioral choice, which influence stress response and behavioral outcomes. Finally, we reported the sources documentation (one to three independent references per case) and date of last retrieval. For incidents involving elephants we mainly used cross-referencing with records from elephant.se (Koehl, 2025) and other web sources.

Table 1. Categories used to classify the contexts in which injurious events occurred, caused by exotic species housed under human care in zoos, parks and aquaria.

These categories are descriptive and follow occupational safety approaches commonly used in incident reporting and classification. The category “found dead” – associated with the highest severity level – reflects a classification choice, rather than causal mechanisms.

2.3 Ethical statement

This study was based exclusively on retrospective data obtained from publicly accessible sources. No original data were collected, and no interaction with human or animals were performed. No personal identifying information beyond that already in the public domain was reported.

2.4 Statistical analysis

Data were analyzed using descriptive and regression-bases statistical approaches. All analyses were conducted to characterize associations between injury severity and selected explanatory variables. We could not estimate absolute risk or causal effects, as exposure denominators were not available. The injury severity was analyzed using cumulative link models with a logit link function (Agresti, 2010; French and Shotwell, 2022), modeling severity as an ordinal response and estimating the association between predictors and the probability of observing outcomes at or above each severity threshold. Due to the large number of species represented in the dataset, we focused on the five most frequently involved taxa i.e., tiger, elephant, killer whale, lion, and bear. We grouped all remaining species into a single category (“other” species) to ensure model stability without intend to imply any biological similarity among taxa in the aggregated category. Therefore, the category “other”, despite being composed of different taxa in different environment and with different human-animal interaction patterns, nevertheless represents a statistical device and not an interpretative group. For transparency, results for this category are retained in tables and figures to document overall data structure, but no inferential or comparative conclusions are drowned from this group.

The regression model included the predictors species category, habitat type (terrestrial, marine, freshwater), human role (staff or visitor), and context. Cases with missing information for severity were excluded from the regression analysis (n=105). In addition to regression modeling, pairwise comparisons of severity distributions across species categories were conducted for descriptive purposes. These comparisons evaluated differences in the modeled severity distributions and were adjusted for multiple testing using Holm correction. Given the absence of exposure data, these comparisons were interpreted as differences in observed severity patterns rather than differences in underlying risk. All statistical analyses were performed using R (version 4.5.2). The ordinal package was used for cumulative link and partial proportional odds models, and the emmeans package was used for post hoc comparisons. Statistical significance was assessed at α = 0.05. The R scripts are available as Supplementary Data S2.

3 Results

A total of 1,047 animal-human interaction accidents were included in the dataset. Among the total accidents, five species were mostly involved: tiger (n=169), elephant (n=142), killer whale (n=123), lion (n=86), bear (n=60), accounting for 59.2% of the accidents. “Other” species accidents (n=427) accounted for 40.8%. Across all variables, injury outcomes were strongly skewed toward low-severity events; severity category 1 represented more than 90% of the incidents in most groups. Severe injuries (categories 3-4) were comparatively rare. Incidents occurred predominantly in terrestrial habitat (n=846) with smaller number for marine (n=186) and freshwater species and settings (n=33). Staff were involved in 686 accidents, whereas 367 incidents occurred to visitors. The most frequent contextual classifications were provoked interactions (n=454) and incidents attributes to human error (n=265). Relatively few cases were classified as animal escape (n=37) or unprovoked events (n=19). The majority of incidents originates in the Americas (n=579) and Europe (n=202), followed by Africa (n=149) and Asia (n=104).

3.1 Injury severity by species and habitat

Species-specific differences in injury severity were observed particularly at higher severity thresholds (Table 2). At lower severity levels, predicted probabilities were generally low and relatively similar across species. As the severity level increases, a species-specific pattern emerged, and differences in modeled probabilities became more pronounced across species categories.

Table 2. Predicted probability of severity outcomes for different species, with 95% CI.

For each species and severity level, the number of documentable incidents is reported. Missing numerical data (NaN) is 106 on the total 1,047 cases in the study.

Elephant-related incidents showed higher predicted probabilities at intermediate and high severity levels, including severity category 3 (0.28, 95% CI 0.20-0.35), and severity category 4 (0.47, 95% CI 0.37-0.56). Elevated predicted probabilities at the highest severity level were also observed for lions (0.44, 95% CI 0.33-0.55), tigers (0.38, 95% CI 0.30-0.46), and bears (0.32, 95% CI 0.20-0.44). Killer whale incidents were strongly associated with low-intermediate severity outcomes in the present dataset, despite the existence of high-profile severe cases reported in the literature, which are also included among the data sources used in this study. The observed pattern reflects the relative frequency distribution of recorded events rather than the absence of severe incidents. Incidents grouped as “other species” are shown for completeness; however, given the extreme taxonomic and functional heterogeneity of this category, these estimates are reported descriptively only and should not be interpreted as reflecting a coherent biological or risk-related pattern (Figure 2).

Habitat type was associated with differences in modeled injury severity. Marine habitats were associated with lower odds of higher-severity outcomes compared with terrestrial environments (0.43, 95% CI: 0.19–0.97; Figure 3). Freshwater habitats accounted for a small number of incidents and did not display a distinct severity profile.

3.2 Injury severity by context

Predicted severity outcomes differed meaningful across contextual categories (Figure 4). At lower severity thresholds, predicted probability estimates were generally low and relatively similar across contexts. At higher severity level, incident classified as human error or animal escape events showed the highest predicted probability profile, with comparable risk patterns. Unprovoked incidents display increasing predicted probabilities across moderate severity levels (categories 1-2). The category “found dead” – associated with the highest predicted probabilities at severity category 4 – reflects the classification structure rather than contextual risk.

3.3 Pairwise species comparisons

Pairwise comparisons of the modeled five-level severity distributions between species categories are presented in Table 3. Statistically significant differences were most evident at severity categories 1 and 4. Elephant-related incidents differed from killer whale and “other species” incidents, with lower modeled probabilities at severity category 1 and higher modeled probabilities at severity category 4. Pairwise contrasts involving the category “other” are reported for transparency but should not be interpreted inferentially, as this group does not represent a biologically coherent comparator. In this case too, the category “other” should be understood as taxa outside the five most representative species treated, as they are aggregated for purely statistical reasons. Killer whale incidents showed higher modeled probabilities at low severity levels and lower probabilities at fatal outcomes compared with large terrestrial mammals. Differences between elephants, lions and tigers were not statistically significant across severity levels, indicating similar modeled severity distributions for these species. Bears showed wider confidence intervals, reflecting the smaller number of recorded incidents in this category.

Table 3. Summary of species differences across severity levels.

Numbers show how much higher (+) or lower (–) each species is relative to another species. Significance levels are indicated by asterisks (p < 0.01***). Patterns reveal strongest differences at Severity 1 and Severity 4.

4 Discussion

This study provides a comprehensive assessment of over 1,000 incidents that occurred in zoos, parks, and aquaria globally over 200-year period. The results highlight clear patterns in incident severity, species risk, habitat influence, and human roles, offering practical insights for improving safety and facility management in zoological settings.

Across the whole dataset, most recorded incidents resulted in low-severity outcomes. This indicates that serious injuries and fatalities remain relatively rare in zoological settings. The pronounced skew toward minor injuries is consistent with the structured nature of modern zoological management and safety protocols. Nevertheless, severe outcomes were not randomly distributed. Instead, higher-severity incidents systematically clustered around specific combinations of species, environments, and human involvement, suggesting the value of severity-based risk profiling rather than reliance on incident counts alone. Species-specific differences in injury severity emerged most clearly at higher severity thresholds. Large-bodied terrestrial mammals, including elephants and large carnivores, exhibited higher predicted probabilities of severe and fatal outcomes than other taxa. These differences were not evident at lower severity levels, where predicted probabilities were broadly similar across species. The observed patterns are consistent with differences in the physical consequences of human–animal contact across taxa, although such mechanisms were not directly evaluated in this study. Staff members were involved in most accidents, especially in cases of human error (Barreiros and Haddad, 2016; Hosey and Melfi, 2014; Rosenthal and Xanten, 2012). Given that terrestrial settings and staff interactions represent the greatest exposure opportunities, their high frequency likely reflects activity differential exposure rather than intrinsically elevated risk per-interaction. On the other hand, the number of accidents among visitors may reflect the limited familiarity with exotic animals or with safety procedures compared with that of trained personnel (Hosey and Melfi, 2014); this is especially true for children, who can also suffer more serious injuries when attacked, precisely because they are smaller than adults.

Species such as tigers, elephants, and killer whales were disproportionally represented in high-severity cases (Anderson et al., 2016; Coyne, 2019; Koehl, 2025). Pairwise statistics suggest a severity gradient among the five most frequently species involved, with elephants, lions, and tigers (Nyhus et al., 2003; Schowe, 2022) occupying the upper range of predicted severity, while bear, and killer whale accidents tend to cluster at the lower end (Almunia, 2025; Sánchez-Hernández et al., 2019). Although elephants, lions, and tigers do not differ significantly from one another at any severity level (predicted severity statistically similar), they show higher estimates than other species with lower predicted severity. Instead, bears (brown bear, polar bear, giant panda) represent the rarest accident type, which is reflected in their wider confidence interval along with a lower estimate stability. Yet, as uncertainty increases with low sample size, being bear accidents small in numbers, these findings suggest caution in interpreting the exact position of bears within the severity gradient. Marine environments were associated with elevated severity in the context of provoked accidents, especially regarding the cetacean shows and trainer injuries (Anderson et al., 2016), which sometimes can be lethal. These results support that high-severity accidents may reflect factors intrinsic to the biology of the species. These species are characterized by complex cognition, complex social structure, and large spatial requirements, which may be constrained within captive management systems. For these animals (as for many other taxa not discussed here) when behavioral needs exceed the opportunity for choice, predictability, or avoidance, the likelihood of high-intensity responses during close human interaction may increase.

The human role was a central determinant of the observed patterns. Staff members were involved in most incidents, maybe reflecting routine occupational exposure during husbandry, training, and veterinary activities (Szleszkowski et al., 2017; Kelly et al., 2020; Hosey and Melfi, 2014). This involvement should not be interpreted as evidence of increased per-interaction risk, but rather as a statistical consequence of repeated and close exposure associated with routine occupational activities with animals. Instead, the comparatively low severity profile observed in public visiting aquaria may reflect the controlled contexts in which these encounters occur: visitors never enter the aquariums, and exposure occurs mainly at the touch tanks, where they can pet rays or feed fish (Casamitjana, 2004). Marine enclosures indeed typically restrict access to trained personnel, preventing unstructured interactions and reducing the potential for escalation and dampen injury severity (Barreiros and Haddad, 2016). Unfortunately, this interpretation remains provisional because we were unable to retrieve information on enclosure design, access restrictions, or emergency response protocols in the present dataset. Within visitor-related incidents, intruders were disproportionately represented among high-severity outcomes. Intrusions were often associated with impaired judgment or compromised decision-making capacity including unauthorized enclosure entry, violation of safety barriers, intoxication, or deliberate rule-breaking (Supplementary File S1). From the animals’ perspective, intrusion into enclosures or sudden close contact may trigger startle reactions and defensive behavior, particularly when retreat options are limited. These fatal outcomes resulting from unauthorized access underscore the importance of access control and boundary integrity (Hosey and Melfi, 2014; Hosey et al., 2023; Barreiros and Haddad, 2016) and highlight enclosure and interaction designs that restrict the animals’ ability to withdraw from a perceived threat. Although restricted access and supervisory practices should have prevented accidents to children, live experiences with small animals, such as otters or monkeys, have often posed serious problems for younger visitors. Notably, some visitor-related bite incidents resulted in the euthanasia of animals for rabies testing, often with negative results (Supplementary File S1), underscoring the welfare and ethical costs of such encounters for animals rather than disease transmission as a causal factor.

The classification of the behavioral context in which the accidents took place differentiated severity profiles further. Incidents involving human agency (human error, provoked interactions) and animal escapes showed higher predicted probabilities of severe outcomes than unprovoked or ambiguous cases. These categories are descriptive and follow occupational safety approaches commonly used in incident reporting and classification. While they do not establish causation, they provide a structured framework for identifying operational scenarios in which severe outcomes are more likely to occur. When “found dead” cases involved keepers, they were always related to human error, for example, when a caretaker entered a facility alone, emphasizing the need for enhanced operational protocols and post-incident investigation procedures. Far from making moral judgment, we have decided to classify trainer–animal incidents (as well as the accidents occurring during live experiences with exotics) as “provoked” in any case, based on ideas grounded in animal welfare science (Dawkins, 2006; Tallo-Parra et al., 2023). In fact, in captive environments, training procedures introduce structured human demands that shape animal behavior beyond their natural patterns. Although modern training techniques rely on positive reinforcement (Spiezio et al., 2015; 2017; Melfi et al., 2020), no interaction is neutral: any interplay requires close proximity, repeated performance, and reduced opportunities for the animal to avoid the stimulus. From a welfare perspective, this creates conditions that may elevate stress, frustration, or defensive responses. The statistical pattern in this study is consistent with that interpretation. Trainer-related incidents were more likely to involve higher severity levels than incidents involving non-staff categories, particularly for large-bodied terrestrial species and cetaceans. This aligns with the principle that animals under direct human control, especially in high-engagement contexts, face pressures and stimuli that increase the likelihood of injury outcomes. In contrast, routine husbandry behaviors associated with medical or health management —such as foot presentation in elephants or voluntary examination —showed lower predicted severity, reflecting lower conflict potential when tasks align with welfare care objectives. The case of cetaceans reinforces this distinction. Among the most involved categories, killer whales demonstrated comparatively high estimates at lower severity levels and wide uncertainty across levels 2–4, consistent with both limited sample size and recognized welfare constraints associated with marine mammal captivity (Lott and Williamson, 2017; Lauderdale et al., 2021). Unlike terrestrial mammals, cetaceans under professional care are often trained primarily for performance rather than for health-related husbandry or conservation. Because their enclosures cannot replicate natural ranging, social structure, or sensory environments, the welfare load for cetaceans is inherently higher (Anderson et al., 2016). Behavioral training in these conditions may increase unpredictability and reduce behavioral buffering capacity, contributing to elevated risk of adverse interactions.

From a management perspective, these findings suggest that the training associated with non-welfare objectives (e.g., entertainment, performance, public display) may introduce avoidable risk, particularly in species with complex social and cognitive needs. Risk mitigation should prioritize training restricted to essential husbandry and medical behaviors, reducing unnecessary proximity or behavioral demands. Facilities should implement species-specific welfare audits, focused on training load, enclosure complexity, behavioral choice, and social structure to evaluate whether current practices support or compromise welfare capacity. Moreover, worldwide institutions may consider policy frameworks that limit show-based training, particularly for cetaceans and other cognitively advanced species, given the combined welfare and safety implications reflected in the severity distribution. While this study offers one of the most extensive quantitative analyses of zoological incidents to date, potential underreporting limits direct comparability across historical periods due to changing reporting standards, especially in earlier centuries or in underregulated regions. The heterogeneity within the “other” category, have represented a limit for the inference in this work; however, a dedicated disaggregation of this group was beyond the scope of the present study, and will be addressed in future, with species-focused analyses. Finally, the severity grading has been based on secondary descriptions rather than standardized injury scales, and there is a lack of behavioral situational nuance, e.g., the exact condition of provocation.

Taken together, these results indicate that trainer–animal interactions are not simply behavioral exchanges but structured welfare exposures that shape risk outcomes. By integrating welfare science with injury data, the study supports a management approach in which training is guided by necessity, welfare benefit, and species capacity rather than tradition or entertainment value.

Generative AI statement

The author(s) declared that generative AI was not used in the creation of this manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher’s note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Supplementary material

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fetho.2026.1784527/full#supplementary-material

Supplementary material Supplementary material at the source

Supplementary material Supplementary material at the source

Footnotes

Edited by: Jesús A. Rivas, New Mexico Highlands University, United States

Reviewed by: HyeJin Kang, Seoul National University, Republic of Korea

Ji-yoon Kim, Animal Welfare Research under Human Care (UHC), Republic of Korea