Deconstructing resuscitation training for healthcare providers: a protocol for a component network meta-analysis

Abstract

Introduction

The necessity of enhancing resuscitation training has been encouraged by The International Liaison Committee on Resuscitation and the American Heart Association to reduce mortality, disability and healthcare costs. Resuscitation training is a complicated approach that encompasses various components and their mixture. It is essential to identify the most effective of these components and their combinations, to measure the corresponding effect size and to understand which participant groups may enjoy the greatest advantage.

Methods and analysis

We will systematically search 12 databases and two clinical trial registries for randomised controlled trials (RCTs) that examine different resuscitation training methods from inception to April 2025. The analysis will be carried out using the standard network meta-analysis and component network meta-analysis models. Resuscitation skills of staff will be the primary outcome of this analysis. Paired reviewers will independently screen and extract data. A consensus will be sought with the principal investigators to resolve any disagreements that cannot be achieved through regular meetings. Each intervention in each RCT will be decomposed according to its constituent components, such as delivery method, interactivity, teamwork, digitalisation and type of simulator. The analysis will be conducted using the frequentist and bayesian approach in the R environment. RoB V.2.0 and Confidence in Network Meta-Analysis will, respectively, be used to assess the risk of bias and the certainty of the evidence.

Ethics and dissemination

As we will use only aggregated secondary data without individual identities, ethical approval is not required. Results of this review will be shared through a peer-reviewed publication and presentation of papers at any relevant conferences.

PROSPERO registration number

CRD42024532878

STRENGTHS AND LIMITATIONS OF THIS STUDY.

• Our network meta-analysis (NMA) will estimate the ranking of treatment efficacies of numerous possible competing comparisons even when no actual trials have directly examined them.

• Additive and interactive effects of combined resuscitation training will be estimated through a component NMA.

• Beside calculating the relative effect size for each constituent component, the component NMA can potentially produce new treatments through unique mixtures of available individual components.

• A large number of studies is required to estimate the incremental effect size for each resuscitation training component.

• We will be unable to estimate the relative efficacies of various resuscitation training components if NMA assumptions are violated.

Introduction

Resuscitation refers to a lifesaving intervention for patients who are experiencing a critical situation related a decrement in their clinical condition.¹ It is estimated that over 500 000 cardiac arrest cases resulted in millions of people needing emergency assistance every year.² The incidence of in-hospital cardiac arrest (IHCA) is estimated to range 1–10 per 1000 admissions.³,⁷ However, the quality of cardiopulmonary resuscitation (CPR) during IHCA is deemed to be inappropriate with standards recommended by international protocols.⁸,¹² Hence, there should be further efforts to improve the competence of healthcare providers in performing resuscitation.

The International Liaison Committee on Resuscitation (ILCOR) and the American Heart Association (AHA) recognise how essential it is to improve the quality of resuscitation training⁶¹³,¹⁶ in order to depress mortality and disability rates and cost burdens.¹⁷,²⁰ In the last few decades, several studies have evaluated the effects of various resuscitation training approaches. Based on the ILCOR’s extensive series of systematic reviews, The International Consensus on Cardiopulmonary Resuscitation and Emergency Cardiovascular Care Science With Treatment Recommendations summarised numerous instructional designs, which could potentially improve training outcomes.^{21 22} However, the question as to which training method is the most effective remains unresolved.

Resuscitation training involves complex interventions composed of various components with significant variabilities.^{114 23},²⁶ This has resulted in large numbers of distinct interventions that have been compared in randomised controlled trials (RCTs). Prior multicomponent interventions in psychotherapies estimated their treatment efficacies from 63 (26) to 94 (25) arm comparisons. Similarly, in resuscitation training, we observed more than 100 arms will be potentially compared against unique components. Therefore, it is infeasible to evaluate the corresponding vast number of comparisons among all of these interventions through RCTs.²⁷

In the past few decades, investigators have applied qualitative synthesis techniques such as systematic reviews to evaluate the effectiveness of resuscitation training.¹⁴²⁸,⁴⁰ Nonetheless, this method may produce a high possibility of bias and is inadequate for decision-making.²⁷ Some researchers are still dependent on traditional pairwise meta-analyses to estimate each treatment’s efficacy.^{1 14 24 25} This approach compares only two interventions at a time and may thus provide only partial conclusions, while it often assumes that all active interventions are the same, despite differences in terms of their components,⁴¹ leading to excess heterogeneity. A component network meta-analysis (NMA; CNMA) is a new approach that combines information from trials comparing different combinations of the same set of basic components and can be used to identify the best-performing components and thus inform clinical practice.^{42 43} However, to date, no CNMA has been conducted to compare the effectiveness of various resuscitation training methods, with results published in the literature. We aim (a) to examine whether any resuscitation training method will affect training outcomes, (b) to identify the most effective components within these training programmes and (c) to determine for whom these are most effective. These findings are expected to enhance our understanding of how training components can be customised to address the various needs of resuscitation trainees in various contexts and for different purposes.

Methods

This systematic review and CNMA will be carried out according to the Preferred Reporting Items for Systematic Review and Meta-Analysis Protocols extension for network meta-analyses,⁴⁴ and study protocols.⁴⁵ Registration for this study was submitted to PROSPERO (CRD42024532878). Any amendments in our protocol will be recorded in our registry.

Eligibility criteria

This review will assess full papers from peer-reviewed articles as well as grey literature, including theses and dissertations from inception to April 2025. To avoid potential impacts of non-English studies on conclusions of this systematic review,⁴⁶ we will adhere to Cochrane guidelines⁴⁷ by including both English and non-English publications. Therefore, our study should produce more comprehensive and precise evidence.^{48 49} The patient/problem, intervention, comparison, outcome framework will be used to determine our inclusion and exclusion criteria.⁵⁰

Types of populations

This study will include all healthcare professionals who have undergone various resuscitation training programmes. Healthcare providers consist of medical staff, nurses, midwives, other allied healthcare workers who joined structured resuscitation training (SRT). In line with a previous study,⁵¹ we will also consider healthcare-related students as part of our population of interest. On the other hand, lay responders, such as security personnel, and school or university students will be excluded from this review.

We will accentuate previous clinical experience as the main value that differentiates staff from students. Hence, we will classify medical or nursing residents as staff. On the other hand, the student category will consist of undergraduate students, medical interns and prelicensure attendees. There are several studies that inconsistently categorised staff and students. Therefore, our study will employ the following approaches: (a) if one article clearly mentions its population as healthcare-related staff or students, then it will be included and (b) if one article does not exactly define its population criteria, we will classify them based on the characteristics of participants’ data. A study will be included if its participants were healthcare staff and/or students and will be excluded if it involved bystanders.

Types of interventions and comparators

Our interest is to investigate the effectiveness of any variant of resuscitation training methods. This review will adopt prior SRT definitions as a predefined and structured resuscitation curriculum, which include basic life support (BLS), advanced cardiac life support (ACLS), advanced life support (ALS), advanced trauma life support (ATLS), neonatal resuscitation programmes (NRPs), and paediatric advanced life support (PALS) as both initial and recertification training.¹⁴ Consequently, we will omit single-procedure training such as automated external defibrillation, intubation, chest compression (CC) only and drug administration training. Considering different intentions between simulation-based research as a training methodology and as an investigative methodology,⁵² we will examine trials that are classified as investigative methodology in a separate meta-analysis. By pursuing this approach, we aim to ensure that our findings will closely reflect the nature of real-world resuscitation events.

We will conceptualise resuscitation training components based on the International Consensus on Cardiopulmonary Resuscitation and Emergency by International Liaison Committee on Resuscitation (ILCOR) 2020.^{21 22} To increase our precision, we will elaborate several ILCOR²⁶³⁴,^{40 53} and non-ILCOR³⁰,^{3351 54} systematic reviews to formulate our instructional design component definitions and examples. Those components will categorise delivery methods, interactivity, ratios, digital aids, types of simulators, human feedback and booster practice. Online supplemental appendix 1 shows possible instructional design components for resuscitation training and their definitions. Studies will be included if they comprise at least one or more combination, and there should be at least one component different between the intervention and comparator groups.

We will consider no intervention, standard training and placebo effect training to be passive comparators. Because of the same clinical meaning, conventional and traditional training will be included in the standard training group. In a preliminary scanning of the literature, we found that several authors technically defined the standard training group in different ways. For example, Campbell, Barozzino, Farrugia and Sgro⁵⁵ defined standard training as face to face+interactive + group+video + low-fidelity manikins+debriefing. Meanwhile Cherry, Williams, George and Ali⁵⁶ used face to face+interactive + group+low-fidelity manikins+debriefing to describe their study. Similarly, but not the same, face to face+interactive + group+high-fidelity manikin features were used as standard training by Low et al⁵⁷ We conceive that there will be some variations of standard training. If a randomised trial does not clearly specify its traditional training arm, we will assume that it has face-to-face, instructor-led, group, didactic lecture, low-fidelity and mass-learning features. Table 1 shows how different resuscitation training methods and controls can be conceptualised from combinations of various potential training components.

Table 1. Conceptualisation of different resuscitation training methods according to potential combination of components.

Group	Possible component combinations
No intervention	NA
Standard training	ff+il+gr+dl+lfm+ml
Self-learning	sin±vr±vid±fd±(lfm/hfm) ±ca±con
Online learning	ol+sin+(gr/ind) ±dl±vr±id±ca±con+ mL
Booster training	(ff±ol)+(int/sin)+(gr/ind)±vr±fd+(lfm/hfm)±ca+bt
Virtual reality	(ff±ol)+(int/sin)+(gr/ind) +vr±vid±ca±con±pp+(ml/sl)
Individual training	(ff±ol)+(il/int)+ind±vr±vid±(lfm/hfm) ±ca±fd±con±pp+(ml/sl)
Rapid cycle deliberate practice	ff+int+gr±dl±vid±fd+(lm/lfm/hfm) ±ca±(deb/con) ±bt±pp+dp
Debriefing	ff+(il/int)+(gr/ind) ±dl±vid±fd+(lm/lfm/hfm) ±ca+deb±bt±pp±ml
Teamwork training	(ff±ol)+(il/int)+gr± dL±vid±fd+(lfm/hfm) ±ca±(deb/con)±pp+(ml/sl)
Simulation-based training	ff+(int/sin)+(gr/ind) ±vid±fd±(lm/lfm/hfm) ±ca±(deb/con)±bt±pp±ml
Life model training	(ff±ol)+(il/int/sin)+(gr/ind) ±dl±vid±fd+lm±ca±(deb/con) ±bt±pp+(ml/sl)
High-fidelity manikins	(ff±ol)+(il/int/sin)+(gr/ind) ±dl±vid±fd+hfm±ca±(deb/con) ±bt±pp+(ml/sl)
Video training	(ff±ol)+(il/int/sin)+(gr/ind) ±dl±vr+vid±(lm/lfm/hfm) ±ca±(deb/con)+(ml/sl)
Hybrid learning	ff+ol+(il/int)+(gr/ind) ±dl±vr±vid±fd+(lfm/hfm) ±ca±(deb/con)±bt±pp+(ml/sl)
Interactive learning	(ff±ol)+int+(gr/ind) ±dl±vr±vid±fd+(lm/lfm/hfm) ±ca±(deb/con) ±bt±pp+(ml/sl)
Feedback device	(ff±ol)+(il/int/sin)+(gr/ind) ±dl±vid+fd+(lfm/hfm)±ca±(deb/con) ±bt±pp+(ml/sl)
Cognitive aid	(ff±ol)+(il/int/sin)+(gr/ind) ±dl±vid±fd+(lfm/hfm)+ca±(deb/con) ±bt±pp+(ml/sl)
Group training	(ff±ol)+(il/int)+gr± dL±vr±vid±fd+(lm/lfm/hfm) ±ca±(deb/con) ±bt±pp±(ml/sl/dp)

bt, booster training; ca, cognitive aid; con, consultation; deb, debriefing; dl, didactic lecture; dp, deliberate practice; fd, feedback device; ff, face to face; gr, group; hfm, high fidelity manikins; hy, hybrid; il, instructor-led; ind, individual; int, interactive; lfm, low fidelity manikins; lm, live model; ml, massed learning; NA, Not applicable; ol, on-line; pp, pre-course preparation; sl, spaced learning; vid, video; vr, virtual reality.

For studies claiming to follow AHA, European and Canadian resuscitation standards, we consider that they will minimally provide group learning, face-to-face, interactive, video materials, plus didactic learning and simulations using low-realism manikins.⁵⁸,⁶⁰ If a training combines individual and group sessions, we will classify it as group training. Self-instructed training definitions should be done in whole sessions. Therefore, if a training mixes self-instruction with instructor-led or interactive components, we will conservatively define the training as instructor-led or interactive. We will consider a null component if the training component is delivered during and after the outcome assessment.

Intended outcomes

Primary outcomes

We predefine our main outcomes in accordance with ILCOR’s recommendations on outcome selection.⁶¹ In terms of timeframe measurements, we will adopt the ILCOR 2020 classification as outcomes at (a) course conclusion, (b) between course conclusion and 1 year and (c) after 1 year²². In this review, behavioural performance outcomes will be our pivotal interest as described here.

Skill performance

Skill performance (SP) describes the ability of trainees to perform various tasks in a resuscitation procedure, such as pulse check, positioning, CC, ventilation and drug administration. We will include actual scores from any measurement instruments such as The KidSIM Pediatric Resuscitation Assessment Tool,⁶² The Simulation Team Assessment Tool,⁶³ The Reliability of the Resuscitation Assessment Tool,⁶⁴ Clinical Performance Tool⁶⁵ and computerised simulation performance assessment tools.⁶⁶

Percentage compliance with the correct CC depth

Percentage compliance with the correct CC depth (PCwCCCD) refers to the proportion of total CCs, which adheres to CC depth recommendations of 5–6 cm.^{67 68 68} For paediatrics spanning from birth to puberty, the ILCOR recommends one-third anteroposterior chest diameter (approximately 1.5 inches (4 cm) in infants to 2 inches (5 cm) in children for the correct compression depth.⁶⁹

Percentage compliance with the correct CC rate

The PCwCCCR is defined as the percentage of CCs meeting the correct compression rate of 100–120/min in adults, paediatrics and infants.⁶⁷,⁶⁹

Overall excellence CC

Overall excellence CC (OECC) assesses the percentage of simultaneous compliance with depth and rate.

Percentage compliance with chest recoil

The percentage compliance with chest recoil (PCwCR) expresses the percentage of total CCs, which provides sufficient time for the chest to fully return to the prior condition before the next CC.⁷⁰

Secondary outcomes

Knowledge

Knowledge provides information related to a trainee’s understanding and their ability to recall information of resuscitation protocols. We will collect data from any instrument that evaluates theoretical skills, such as multiple‐choice questions developed based on national or international resuscitation guidelines such as the UK Resuscitation Council, AHA, ERC and ILCOR. Additionally, knowledge assessments conducted during scenario-based training, such as algorithm application and action prioritisation, will be considered part of the resuscitation knowledge domain.

Teamwork skill

Unlike the SP, teamwork skill (TS) specifies non-procedural skills, which focus on assessing a trainee’s ability to work cohesively in a team. It may consist of leadership, task management and communication skills.

Time to task completion

Time to task completion (TtTC) is defined as the amount of time needed to complete a certain task in a resuscitation simulation. This outcome will be explored in detail such as: (a) time to intubation; (b) time to first CC; (c) time to activated emergency medical system, (d) time to defibrillation; (e) time to first positive pressure ventilation; etc. We will use seconds as the measurement unit for procedures.

Mean correct CC depth

The mean correct CC depth (MCCCD) provides information on the average of CC depths during a resuscitation procedure, which is expressed in centimetres.

Mean correct CC rate

The mean correct CC rate (MCCCR) is defined as the average of CCs in beat per minutes during a resuscitation attempt.

Trainee perceptions

These include any beliefs, ideas or imaginings as a result of perceiving the resuscitation process. This may include TPs on training quality, participant’s improvements, etc.

Confidence

We will collect data from any tools or questionnaires, which assess beliefs or feelings of being confident in performing resuscitation procedures.

Satisfaction

Some studies describe trainees’ pleasant feelings when they obtain an experience they desired.

Cost-effectiveness

If a study reports how well resources were used in resuscitation training to achieve maximum results, we will extract such data. In case studies reporting costs using different currencies, we will convert those into US dollars.

Patient-reported outcomes

Patient-reported outcomes (PROs) refer to any report of a patient’s health condition including survival rate, burden (eg, length of stay), quality of life, and other given information.⁴⁷

Study design

Our review will solely include RCTs that execute comparisons between an active intervention and another eligible intervention or control. Randomisation can be done at either the individual or group level. For clustered RCTs, we will adjust SEs to account for design effects using a standard formula.⁷¹ Additionally, this review will include both parallel group and crossover designs. If feasible, we will collect data prior to the washout period to reduce any carryover effects in crossover trials.^{47 72 73} In case a study only reports a relative treatment estimate, we will use the following approach: (a) if the authors report the carryover test, we will include this trial only if there is no evidence of a carryover effect and (b) if the trial declares no carryover test, we will include that study and will consider downgrading its risk of bias assessment.

Search strategy

We will perform an extensive literature search over twelve databases, which include nine Western databases (CENTRAL, EMBASE, MEDLINE, ProQuest, Scopus, Web of Science, CINAHL, IEEE Explore and ERIC) and three Chinese databases (Wanfang, Airity and CNKI) from inception to April 2025. The sample of keyword term applied is listed in online supplemental appendix 2. We will also search the grey literature in the International Clinical Trials Registry Platform and ClinicalTrials.gov. In addition, papers retrieved from hand-searching of specific publishers will also be included, since that can significantly increase the number of articles based on our prior experience.⁷⁴ Thus, we will search nine publishers and related journals (Taylor and Francis, Cambridge Core, Wiley, Science Direct, Oxford Academic, Sage, Springer Link, Resuscitation and Resuscitation Plus). Finally, we will also systematically collect reference lists of potential included studies.

Study identification and selection

Checking and screening duplicate titles and abstracts will be carried out by two independent reviewers using Rayyan.⁷⁵ Afterward, a full-text review will be conducted by two independent authors. Any discrepancies that occur will be carefully examined and resolved through consensus with the principal investigator. Every justification for excluding a paper will be recorded in a document and will be reported in our final report. If we find the same data published in multiple journals, we will choose the report that has the best clarity. In the event of both published and unpublished data, we will prioritise proceeding with the peer-reviewed document. If two articles are obtained from a single study, they will still be included as long as they report different outcomes. We will prioritise the treatment component, which is explicated in the most comprehensive report.

Data collection and integrity checks

Data extraction

Prior to formal data extraction, a minimum of 30 included articles will be used to design a tailor-made data extraction form. For every record, two independent assessors will investigate and extract data into long-format data, involving publication metadata (eg, author, title and year), study design (eg, parallel, cross-over, cluster RCT, two-arm or multiarm parallel trial), dropout rate, type of analysis (intention-to-treat (ITT) or per-protocol (PP)), topics (BLS, ACLS, PALS, ATLS and NRP), setting (intrahospital and extrahospital) and population (monoprofessional or multiprofessional). Characteristics of trainees will encompass gender (male or female), age, number of participants, duration of resuscitation training, time since last resuscitation training, experience in a real resuscitation and refresher training.

For SP, knowledge, TS, TtTC, TCC, MCCCD, MCCCR, TP, confidence, satisfaction and CE, we will collect means and SDs, and/or effect estimates such as mean differences (MDs) along with their 95% CIs and p values. The presentation of data as medians or other spread measurements will be changed to means and SDs using an established formula.⁷⁶ In case only figures are presented (rather than numerical data within the text), data will be extracted using PlotDigitizer to estimate the length (in pixels) of the axes for calibration, and then the lengths in pixels of the data points of interest will be determined.⁷⁷ If this proves infeasible, we will substitute missing SDs from studies reporting SDs.⁷⁸ The effect of this approach will be examined through a sensitivity analysis. Moreover, for binary outcomes such as PCwCCCD, PCqCCCR, OECC, PCwCR, CtG and PRO, our interest is collecting incidence rates and/or effect measures such as ORs or risk ratios.

Identification of components

The principal investigator will evolve the primary classification of treatment arms and components according to a prior published review.⁷⁹ Furthermore, a pilot data extraction step will be executed by two independent assessors using 40 available RCTs. This step will focus on identifying characteristics of arms, the number of resuscitation training components and an operational definition of each included training component. Following an iterative process, we will extensively revise the training component classifications and definitions through consensus. We will also adopt some recommendations from the ILCOR,^{20 21 80 81} which may justify most of our training taxonomy and provide extra component features such as mass learning, spaced learning, deliberate practice and mastery learning.^{34 36 40} Consequently, we will revise the protocol to formulate the ultimate version of training components, as described in online supplemental appendix 1. All these changes will be applied through panel discussions and consultations with a Vice-Chair of ILCOR’s Task Force on Education, Implementation and Teams. All piloted notes will be extracted again to meet the new data format and refined-definitions.

Two independent reviewers will extract the characteristics of the intervention components for each record. A specifically designed training programme will be held to assure the ability of all review members. This programme is planned to run 72 hours over 1 month. Afterward, an inter-rater agreement test will be carried out for all participants. Reviewers who achieve a very good performance will be eligible to serve as independent data extractors in our study.^{82 83} All of the data will be saved in a computerised and password-protected database, which can only be accessed by review team members. The data will also be securely stored by LIFESAVERS lab members. All study data will only be used for teaching and research purposes and will not be given to third parties.

Statistical analysis plan

Our analytical process methodology will follow previous guidelines where sequential techniques will be applied to investigate arm- and component-level training effects on the outcomes of interest.^{42 43 84}

Summary measures

A MD will be used if a continuous outcome for all studies reported results used the same scale (eg, TtTC, MCCCDR, CE). For measures using multiple scales (eg, knowledge, SP, TS, TP, confidence and satisfaction),⁸⁵ we will use a standardised MD. For studies which report SP as ORs, we will convert ORs to effect sizes using a reported formula.⁸⁶ A 95% CI will be reported to measure the level of imprecision alongside effect estimates.⁸⁷ Finally, for outcomes of binary data (eg, PCwCCCD, PCwCCCR, OECC, PCwCR, CtG and PRO), pooled estimates will be reported as ORs. Once data pooling is not possible, the outcome will be described descriptively.

Network meta-analysis

We plan to perform a frequentist graph theoretical NMA, which will be conducted using the R package ‘netmeta’.⁸⁸,⁹⁰ We plan to select the random-effect model as we expect that studies will differ both methodologically and clinically (between-study variability).^{1 24 26} To visualise treatment comparisons derived from direct evidence, a network plot will be used for each outcome of interest.⁹¹ The p score, a frequentist counterpart to the surface under the cumulative ranking curve, will be used to estimate the average certainty in ranking competing treatments.^{92 93} Of note, an NMA at the treatment level will only be possible if the network is connected.

Component-level NMA

We will perform a CNMA to measure the effects of various components and their combinations.^{42 84} The effects of individual treatment components can be summed to offer a more-robust estimate of combined treatment effects, rather than relying solely on studies of direct combinations. These combinations can be modelled as additive or as interactive effects, whether synergistic or antagonistic.⁹⁴ The CNMA can also be used even if the network is disconnected at the treatment level, which would render a standard NMA infeasible.⁹⁵,⁹⁷

The CNMA presumes that the effects of individual treatment components are additive when the components are combined.⁴² Based on our clinical expertise, we will evaluate whether it is likely that the components of a treatment combination work through different mechanisms.^{42 43} This evaluation will guide our decision as to whether to consider their effects as additive.⁹⁸ To assess the additivity assumption, we will also compare treatment estimates from the standard NMA and additive CNMA models.⁹⁹ We will compare how well the additive and interaction models fit the data using model fit Q statistics measures.^{42 95 96}

Transitivity, heterogeneity and inconsistency

We will first visually check with box plots whether effect modifiers (gender and age) are equally distributed across treatment comparisons. Consistency, the agreement between direct and indirect evidence, will be evaluated using a global assessment, that is, a design-by-treatment interaction model¹⁰⁰ and a local assessment through a node-splitting approach.^{41 101} A check of inconsistency will also be appraised at the component level if the data permit.¹⁰² We will evaluate heterogeneity using τ.² We will further compare estimated values with their empirical distribution for binary^{103 104} and continuous¹⁰⁴ outcomes with sufficient data.^{1 24 105}

Subgroup analyses

If enough data are available, we will conduct subgroup analyses to assess various potential sources of heterogeneity/inconsistency. These sources include topics (BLS/ACLS/ATLS/PALS/NRP),^{23 24} population (staff/student),²³ teamwork (multi/monodisciplinary),²⁴ prior training (yes/no), gender (male/female), setting (hospital/non-hospital), country (developing/developed country), study design (parallel/cluster)²³ and type of analysis (PP/ITT). If the source of heterogeneity cannot be assessed, we will consider that it can decrease the ability of the model to provide significant conclusions in a diverse population.

Risk of bias assessment and certainty of the evidence

We will assess the risk of bias for every individual study utilising the updated Cochrane Risk of Bias (RoB V.2.0) tool for RCTs.¹⁰⁶ Two independent assessors will check the likelihood of bias in various domains, including (a) randomisation procedures (selection bias); (b) adherence to the intervention protocol (performance bias); (c) missing data (attrition bias); (d) outcome measurement (detection bias) and (e) selective reporting (reporting bias). Each study will be classified as having low, high or some concerns regarding bias. Any disagreement will be resolved through discussion or involving a third reviewer for arbitration if necessary. The R package and Shiny web app for visualising risk-of-bias assessments will be used to provide RoB data visualisation after consensus among reviewers is achieved.¹⁰⁷

The certainty of the evidence from the main outcomes will be assessed using Confidence in Network Meta-Analysis (CINeMA) and will be presented in a summary of findings table.¹⁰⁸,¹¹⁰ CINeMA was specifically established for NMAs and has four grades of overall confidence rating for each pairwise comparison: ‘high’, ‘moderate’, ‘low’ or ‘very low’. Two independent reviewers will work to assess (a) within-study bias, (b) reporting bias, (c) indirectness, (d) imprecision, (e) heterogeneity and (f) incoherence.¹⁰⁹ A table generated by CINeMA software will be used to describe each domain and overall confidence rating of each treatment comparison.¹¹⁰

Publication bias and small study effect

When there are more than 10 studies, a comparison-adjusted funnel plot will be drawn to assess publication bias and small study effects. We will explore extreme study effects (ie, outliers) with the forward search algorithm using the R package ‘NMAoutlier’.^{111 112}

Sensitivity analysis

A sensitivity analysis should be administered to evaluate the robustness of the overall findings and their impacts on clinical decisions.⁴⁷ When data are sufficient, we will consider (a) excluding studies with imputed missing data; (b) excluding studies with samples that include fewer than 20 participants in all arms; (c) excluding studies with a high risk of bias; (d) excluding studies containing components with unexpected components and (e) incorporating important two-way information from the main analysis. We will use a panelised regression model in a Bayesian setting to compare numerous possible two-way interactions among components by including all potential interaction terms in the model and applying shrinkage to their coefficients. Coefficients of interaction terms with weaker evidence in the data will be shrunk more aggressively, while those with stronger evidence will be shrunk less.⁹⁷

A series of sensitivity analyses will be performed to evaluate the robustness of the overall findings and their impacts on clinical decisions.⁴⁷ If data permits, we will (a) exclude studies with imputed missing data; (b) exclude studies with a total sample size of less than 20; (c) exclude studies with a high risk of bias; (d) exclude studies containing unexpected components and (e) incorporate important interactions in the main analysis. Based on our expertise, we believe that interactive learning which utilises a high-fidelity manikin, feedback device and interactive learning may produce synergistic effects on our main outcome of interest. In contrast, interactions among self-instructed simulations and online training approaches will work antagonistically.

Ethics and disseminations

Results of this work will be published in peer-reviewed journals and disseminated both electronically and in print. Moreover, findings will be presented as abstracts and/or personal communications at national and international conferences. This project does not encompass primary data collection from humans, as it is based on secondary analyses of already collected anonymised datasets. Thus, it is considered exempt from ethical clearance review. Both national and international regulations on patient privacy will strictly be followed.

Patient and public involvement

This study acknowledges the critical role of patient and public involvement in improving the relevance, accessibility and overall impact of resuscitation training programmes. While patients were not directly involved in the development of this protocol, future phases of the study will actively engage key stakeholders, including resuscitation trainers, guideline developers and frontline healthcare providers. Their contributions will inform the development and refinement of training content, delivery methods and dissemination strategies to ensure the intervention is practical, contextually appropriate and aligned with the needs of end users.

Data availability statement

No data are available.