This systematic review was prospectively registered with PROSPERO (https://www.crd.york.ac.uk/PROSPERO/view/CRD420251084048) and a systematic search of relevant databases was conducted following the Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) statement (16). The PICO framework was used to develop the search strategy based on Table 1. A systematic search of Embase, PubMed, and Web of Science was conducted between June and July 2025. Combinations of MeSH and Boolean terms were used and following full search string was used in all databases to search for appropriate articles: ((((((((pediatric) OR (child)) OR (paediatric)) OR (neonate)) OR (infant)) OR (adolescent)) AND ((mass casualty incident) OR (mass casualty event))) AND (preparedness)) NOT (review). Following the completion of the initial search, a manual search of the reference lists of the selected sources was conducted to ensure identification of any sources not apparent on our initial search.

Studies were required to satisfy the following inclusion criteria: (1) the study population comprised healthcare staff; (2) interventions encompassed educational simulations, drills, or didactic materials (e.g., lectures, resources) that incorporated a paediatric focus or element; (3) assessed outcomes pertained to preparedness, knowledge, or the effectiveness of the drills, as evidenced by observational assessment, quizzes, or qualitative survey feedback; and (4) all included studies were published in the English language.

Studies were excluded where (1) the study population did not comprise healthcare staff. Further exclusions were applied if (2) interventions did not involve human participants or demonstrably lacked a paediatric element. Studies were also excluded if (3) their primary focus was solely on the development of an exercise or if they presented no formal assessment. Finally, (4) reports classified as case studies were excluded.

Two researchers (EB and ZA) independently screened each article using Covidence systematic review software (Covidence, Melbourne, Australia). Initial screening involved reviewing titles and abstracts. If the contents remained unclear regarding inclusion criteria, the full article was screened. There were no conflicts during the study screening process and therefore conflicts did not need to be resolved through discussion.

Outcomes for which data were sought included measures of preparedness, knowledge, and the effectiveness of MCI or disaster preparedness exercises. Various formats of compatible assessment were acceptable for measuring these outcomes, such as quizzes, qualitative surveys, and feedback.

A data extraction template was designed and applied to the collected data, encompassing study type, design, aim, population, and key outcomes. This ensured the identification and collection of all results aligned with the review outcomes. All included studies were compatible with all specified outcomes, although a substantial degree of heterogeneity existed in the methods by which these outcomes were measured.

Given the expected heterogeneity of the results, a quantitative meta-analysis was not possible and a narrative synthesis was identified as the most appropriate analytical method. This approach involved the systematic organisation and categorisation of results, allowing the identification of overarching themes and patterns. Individual subgroup analyses were conducted following the categorisation of themes, allowing for deeper exploration of more nuanced ideas.

The Risk Of Bias In Non randomised Studies of Interventions Exposure (ROBINS E) tool (17) was deemed the most appropriate method for assessing the risk of bias, primarily due to the observational nature of the included papers and their investigation of exposure effects. This is because other tools such as ROBINS-I are generally used for non-randomised studies looking at the effects of treatments, drugs or planned programs. To ensure robustness, two reviewers (EB and ZA) independently assessed the risk of bias using the ROBINS E tool; any disagreements were resolved by discussion.

An initial search generated 223 results, with an additional four studies identified through hand searching reference lists of the included studies. After removing duplicates, 153 articles were screened, from which 125 studies were excluded. Of the 28 studies that remained, 11 studies were excluded after full-text reading for various reasons including wrong intervention, wrong outcomes and wrong study design. This left a total of 17 studies that were included in this systematic review and qualitatively analysed (18–34) (Figure 1).

Fourteen of the included studies were observational (82.4%), with the vast majority [12 studies (70.6%)] conducted in the United States (20, 22–29, 31–33). Studies largely adopted single centre designs [13 studies (76.5%)], with a predominance of highly specialised paediatric trauma centres (58.8%; Table 2). Participant sample sizes also displayed a large variation, with a median of 45.5 (range 10–337) and a mean of 75.9 across the 15 studies reporting participant numbers. Mixed healthcare teams working in acute and emergency settings made up most of the samples (82.4%), with only a relatively small minority deciding to focus on specific groups of healthcare workers (19, 22, 24, 29, 34). Training programs favoured active learning, with seven studies adopting a mixed methods approach combining simulations with didactic or online teaching. The exercise scenarios were variable, ranging from hyper realistic simulations to discussion based tabletop exercises, often tailored to the type of MCI most likely to be faced in the area, such as a road traffic accident (RTA) or a natural disaster. Studies were mostly conducted between 2010 and the early 2020s, reporting a wide range of outcomes, including knowledge, skills, confidence, and teamwork. Outcomes were measured using a variety of instruments, from self-perceived Likert scales to objective assessments and external evaluations, with six studies incorporating longitudinal designs to examine knowledge retention over time (19, 23–26, 34).

The duration of the MCI education programs varied significantly, from very short, repetitive drills to multi-hour workshops (Table 2). Three studies employed brief interventions, with Gross et al. (26) implementing weekly “disaster huddles”, averaging at just 7 min each and led to improvement in triage accuracy over the 26 weeks. Two studies conducted single day interventions, with both showing improvements in knowledge, preparedness, and specific skills, along with positive learner feedback (21, 32). These brief interventions are highlighted as feasible and cost-effective approaches to training. Moderate duration interventions, typically 2–5 h long, also consistently demonstrated improvements. For instance, Cicero et al. (22) found that a 2 h didactic course significantly increased knowledge, though participants preferred hands on drills. Other simulation based studies of similar duration led to sustained gains in knowledge and perceived preparedness for up to 6 months. Only one study fell into the extended category with a 6 h program (34). This comprehensive approach resulted in substantial and lasting gains in self-reported comfort and knowledge, which were sustained over 6 months.

Six studies in the review used a longitudinal design to assess the durability of training effects beyond the immediate post intervention period (19, 23–26, 34) (Table 3). The follow up intervals varied between studies, with assessment periods ranging from 2 weeks to 6 months and number of data collections ranging from 2 to 26. Outcomes measured varied from objective measures such as triage accuracy (23) or the donning of PPE (25), to subjective measures of perceived preparedness (19, 34). Despite this heterogeneity, the training interventions consistently led to improvements in both perceived and objective outcomes across all studies. However, lost to follow up was a common limitation. For example, one study reported a low retest rate, with only 14 out of 42 participants completing the follow up quiz (24). Notably, some studies were able to mitigate for this potential attrition bias, as a study on HAZMAT training showed no significant difference in scores between those who completed the retest and those who did not, suggesting the findings were not skewed (25). In general, the rates for participants who completed the educational programme to the end of the course, ranged from just over 33% to just over 96% (23–26, 34), with two studies not reporting completion rates (19).

A subgroup analysis based on the primary goals of the exercises was conducted to identify more nuanced ideas between studies (Table 4). Three main outcome categories were identified: purely triage, a combination of triage and management and other specific elements of MCI management. The majority of the studies (n = 9) elected to simultaneously assess the ability of staff to triage and manage paediatric patients during an MCI (19, 20, 22, 24, 28, 31–34), in comparison to three studies focusing on triage alone (23, 27, 32). A further five focused on another specific element of MCI management, such as Hewett et al. (25) who measured the ability to don PPE (18, 25, 26, 29, 30).

All outcomes generally focused on the acquisition of knowledge following exposure to exercises, but studies tended to approach this differently. For example, studies focusing purely on triage, such as the study performed by Cicero et al. (23) and Kenningham et al. (27) sought to demonstrate improvements in a single measurable skill, achieved through the assessment of the number of correctly triaged patients (23, 27). The same was found in studies focusing on specific elements of an MCI, such as a study by Hewett et al. (25) which evaluated skill retention in donning PPE in the case of a disaster (25). This contrasts with studies assessing triage alongside broader management skills, such as those by Bank and Khalil (19) and Burke et al. (20), who reported a more holistic sense of preparedness alongside objective measures, including the acquisition of “soft skills” like confidence in resource management and inter team communication (19, 20).

Additionally, the type of data collected varied greatly within each group. Objective, performance based tools were used across the three groups to assess outcomes. For example, Li et al. (28) and Delgado et al. (24) both measured the levels of correct triage of simulated patients (24, 25). Kenningham et al. (27) had participants self-assess their triage decisions against a correct answer key, while Cicero et al. (23) used a checklist-based tool for evaluation by an online assessor, supported by video recordings (23, 27). Another approach was employed by Toida et al. (33) who retrospectively reviewed triage tags and medical records following a drill. Knowledge based assessments were also frequently used. Delgado et al. (24) and Cicero et al. (22) both assessed triage knowledge through post-test quizzes (22, 24). Hewett et al. (25) included a multiple choice test on HAZMAT principles, and Tan et al. (32) evaluated JumpSTART triage skills using simulation based worksheets (25). Finally, self-reported scales were a popular method for gauging perceived competence. Wright et al. (34) and Opsahl et al. (31) utilised Likert scales for participants to self-assess their triage abilities (31, 34). Other studies, such as those by Tan et al. (32) and Bank and Khalil (19) incorporated follow up questionnaires for self-evaluation (19, 32).

Beyond individual performance, some studies adopted a broader approach to evaluate systemic or group level triage capabilities. Naru et al. (30), for instance, used non participant observation to analyse how teams adapted to MCI conditions and managed patient triage (30). Burke et al. (20) employed a mixed methods design, combining qualitative, and quantitative feedback with observer findings to assess systemic issues (20). Other studies indirectly assessed triage, such as Asenjo et al. (18), which measured the impact of a drill on real patients' waiting times in a paediatric emergency department (18).

A summary chart and the risk of bias (RoB) in the individual studies are presented in Figures 2A, B. Just over 88% of the studies had an overall “High” to “Very high” RoB, with the remaining 12% of studies also demonstrating some concerns (Figure 2A). The most prevalent source of “High” RoB was Domain 1 (Bias due to confounding), affecting 14 out of the 17 included studies (Figure 2B). This was primarily due to a lack of adjustment for key confounders such as variations in the training and prior experience of staff, reliance on self-reported outcomes and the absence of control groups in the studies. The highest rate of “Very high” RoB was found in Domain 5 (Bias due to missing data), impacting almost 30% of the studies (Figure 2B). Often, this was caused by the failure of the longitudinal studies included to perform complete case analysis or adequately adjust for missing data in follow up cohorts (20, 24). One notable example was a multi hospital study undertaken by Burke et al. (20), where a disproportionately high follow up response rate from a paediatric centre led to significant data gaps from other centres, introducing a substantial risk of bias (20).

Despite the low quality of the available evidence, this review does have several notable strengths. A comprehensive search strategy was employed to include a broad and diverse range of study designs and outcomes. This allowed for a detailed characterisation of the study characteristics, interventions, and outcomes. Furthermore, we conducted a subgroup analysis to highlight and explore the specific nuances of groups, which helped to uncover patterns that might otherwise have been missed.

However, as discussed, the overall quality of evidence in this field is poor, with significant bias present throughout. The large and heterogeneous study designs and outcomes prevented us from completing a meta-analysis. Additionally, the results relied heavily on self-reported outcome measures, which introduced a large amount of reporting bias. Several studies even noted the drawbacks of using scales like Likert scales due to this associated bias. A disproportionate number of the included studies were conducted in the United States, which introduces the possibility of geographical bias due to differences in healthcare systems. Geographical bias is recognised in academic publishing, which is dominated by Western Europe, North America, and the Global North. This can lead to disparity in representation and access to research, particularly from low- and middle-income countries or the Global South. However, strategies to diversify academic publishing and commitment by Journals and publishers will help tackle these inequalities (36, 37).

Restricting the retrieval of studies to only those reported in English introduces language bias, geographical bias, incomplete evidence base, reduced generalisability and potential over- and under-estimation of effects, limiting the usefulness of our study. Generalisability was further limited because most of the studies were conducted at a single site. Statistically, with only one study conducting a power calculation (25), the small sample sizes of many studies likely led to underpowered results. Finally, many of the studies only went as far as assessing triage accuracy. While an important component, several other techniques are required to handle MCIs with children. Therefore, drills incorporating whole hospital systems and their surrounding areas would likely improve preparedness, as other necessary measures, such as forming a “creche” for non-injured or very minorly injured children if they become separated from parents and safely discharging patients from wards to free up beds, could also be practised.

A mixture of quantitative and qualitative outcome assessment measures was used by the studies in this review. All outcomes assessed indicated positive improvements in outcomes compared to their baseline. However, due to a large amount of heterogeneity between results, a meta-analysis was not possible; therefore, standardisation of outcome measures is required to compare methods and assess outcomes. Knowledge based assessments, such as multiple choice pre- and post-test quizzes, are a simple and relatively inexpensive method of objective assessment utilised by several studies (22, 24, 25, 28). These have no risk of reporting bias, which methods such as Likert Scales are liable to (25, 32). However, as with any self-motivated assessment method, they are liable to attrition bias, especially when measuring long term knowledge retention (22, 24, 25). Qualitative data also forms an essential component of the data collection process due to its unique ability to provide deeper insight into the perceptions of the exercises. For example, the studies conducted by Burke et al. (20) and Li et al. (28) both revealed unique insights about “soft skills” such as communication during the exercises, something that was not captured in quantitative feedback (20, 28). Future studies adopting a standardised approach that incorporates qualitative and quantitative feedback are likely to provide more robust information to this field.

This review has highlighted several techniques that, if implemented, have the potential to improve future paediatric MCI preparedness. The findings indicate that active learning techniques, such as drills and simulations, are associated with increased participant confidence and a heightened sense of preparedness for paediatric MCI situations, which contrasts with passive learning approaches. The review also suggests that a “one size fits all” approach to preparedness drills may not be optimal. While full scale exercises using high fidelity mannequins show positive results, smaller, more frequent exercises, such as the “disaster huddles” described by Gross et al. (26), also appear promising (26). Furthermore, the involvement of multiple members of the MDT may also more accurately mimic the teamwork required during an MCI, leading to improved preparedness.

Several papers identified their approach to MCI preparedness training as being “cost effective” (8, 21, 24, 26) for various reasons linked to their exercise format. For example, Gross et al. (26) deemed the use of “disaster huddles” a “low effort, low time commitment” resolution to a more resource intensive full scale exercise (FSE). Chou et al. (21) also made a similar argument for functional exercises (21, 26). Delgado et al. (24) highlighted the flexibility and therefore cost effectiveness that comes with combining an asynchronous online learning module with an in person simulation (35). Li et al. (28) specifically designed a simulation that required few resources and used high fidelity mannequins instead of actors to help reduce costs associated with the exercise (28). Cost effectiveness is an important consideration for any type of training within healthcare systems and maximisation of benefit to users is essential. Whilst studies may deem themselves to be “cost effective”, the lack of direct comparison with other preparedness exercise types somewhat discredits the arguments of authors. Therefore, it is imperative to conduct further research, such as randomised control trials (RCTs) to definitively assess the cost effectiveness of the preparedness exercises.

To advance the evidence base, there is a clear need for more rigorous studies, particularly RCTs. Future research should aim to compare the outcomes of purely active learning interventions with mixed modality approaches to definitively assess differences in perceived preparedness and other relevant outcomes. Such studies should use standardised outcome assessment tools to enable comparisons of how effective different interventions are. Key areas to investigate include the cost effectiveness of various education types and their impact on long term skill retention and knowledge acquisition. To inform resource allocation and optimise training effectiveness, a cost benefit analysis comparing these different modalities would be highly beneficial, considering factors beyond immediate preparedness, such as logistical feasibility. While variation in exercises may be preferable, the development of a standardised MCI educational curriculum framework, outlining feasible training timelines and structures and defining measurable outcomes, could facilitate the formation of more efficient and comparable training initiatives. Implementing an iterative feedback loop, where insights from training evaluations and drills directly inform curriculum development, would foster continuous improvement and positive change in preparedness education. Such research will provide the causal evidence needed to optimise paediatric MCI preparedness training and ultimately improve outcomes for children in mass casualty events.