Comparing the Costs Associated with Anxiolytic Agents for Reducing Distress in Children Undergoing Laceration Repair in the Emergency Department

Abstract

Objectives

The use of procedural medications for anxiolysis during laceration repair in children can mitigate distress, facilitate procedures, and avoid physical restraint. However, the optimal medication for use in the pediatric emergency department is unclear. This study compares hospital length of stay (LOS) and costs for 4 different procedural medications for pediatric laceration repair from the perspective of the Canadian health care system.

Methods

We used an individual-level state-transition model to compare (i) intranasal midazolam (INM), (ii) inhaled nitrous oxide (N₂O), (iii) intranasal dexmedetomidine, and (iv) intravenous ketamine (IVK). Model inputs were informed by literature and expert consultations. LOS included time before, during, and after laceration repair, adjusting for adequacy of interventions and adverse events. Healthcare costs included personnel, equipment, and medication costs. We evaluated the drivers of model uncertainty using the TRansparent Uncertainty aSsessmenT (TRUST) tool and value of information analyses.

Results

The estimated LOS for N₂O, INM, IVK, and intranasal dexmedetomidine were 78.34, 82.13, 103.37, and 114.55 minutes with the per-person estimated costs of $176.7, $122.79, $167.16, and $165.07, respectively. These results were limited by the low quality of available evidence for model inputs. The key parameters contributing to decision uncertainty were procedure time for N₂O and IVK, with an expected value of $5.31 and $0.44 per person, respectively.

Conclusion

INM had the lowest expected healthcare cost, whereas children receiving N₂O had the shortest expected LOS. Decision uncertainty was driven by the lack of relevant studies examining the length of procedure with N₂O and IVK use.

The Bottom Line.

This is the first comprehensive evaluation of 4 procedural medications for anxiolysis during laceration repair in the pediatric emergency department from the perspective of the Canadian healthcare system: intranasal midazolam, inhaled nitrous oxide, intranasal dexmedetomidine, and intravenous ketamine. Costs were largely driven by personnel requirements, with intranasal midazolam being the least expensive due to lower staffing needs. Limited available evidence supporting simulations led to substantial decision uncertainty. We identified key model uncertainties and conducted value of information analyses, estimating the maximum value of future research at $484,816, particularly regarding procedure time with inhaled nitrous oxide and intravenous ketamine.

1. Introduction

1.1. Background

Laceration repairs account for approximately half of all procedures in the pediatric emergency department (ED).¹ These procedures are often painful and distressing for children, leading to difficulties in completing the repair,² increased procedural pain for subsequent repairs,³ and long-term avoidance of medical care.⁴ National agencies, including the Canadian Pediatric Society, endorse anxiolysis or procedural sedation to reduce the use of physical restraint, pain, and anxiety during laceration repairs.⁵ However, procedural medications remain underutilized,⁶ with significant practice variation, contributing to stress and fearfulness in children visiting the pediatric ED.⁵

Anxiolysis and procedural sedation aim to reduce distress during laceration repair without requiring general anesthesia, meaning that the patient is awake and responsive to verbal stimuli.⁷ Common medications used during laceration repairs include intranasal midazolam (INM),⁸ inhaled nitrous oxide (N₂O),⁹ and intravenous ketamine (IVK).¹⁰ IVK is a conventional procedural sedation drug, but is used frequently because of its favorable safety and efficacy profile.11, 12, 13, 14 The Canadian Pediatric Society has recommended more research for ketamine for anxiolysis and minor procedures.⁵ Furthermore, a new anxiolytic, intranasal dexmedetomidine (IND), has also recently demonstrated benefit and outperformed INM for dental procedures, with preliminary evidence suggesting its efficacy in laceration repair.15, 16, 17

1.2. Importance

Although the efficacy of commonly used procedural medications in terms of pain and distress management has been comprehensively evaluated,^9,10,17, 18, 19, 20, 21, 22, 23, 24, 25, 26 their wider impact on hospital length of stay (LOS) and healthcare costs from a Canadian healthcare payer perspective has been less frequently examined. In fact, we only identified one study that aimed to evaluate the cost-effectiveness of sedation medications in the pediatric EDs, but it did not focus on laceration repairs and compared intravenous midazolam with intravenous propofol,²⁷ an agent that is rarely used, in isolation, for laceration repair.¹⁰

Understanding the expected hospital LOS is crucial for enhancing patient care and service efficiency²⁸ as overcrowding of pediatric EDs and long wait times lead to poor quality of care and threaten patient safety.²⁹ Although studies have considered time to anxiolysis onset, procedure and recovery times as primary outcomes, and hospital LOS as a secondary outcome,^18,19,21, 22, 23, 24, 25, 26 a comprehensive evaluation of medications for laceration repair in pediatric EDs, which account for wider health service impacts on health resources and costs, has not been performed.

1.3. Goals of This Investigation

This primary objective was to undertake an economic evaluation of four procedural medications (INM, IND, N₂O, and IVK) used for anxiolysis during laceration repair, accounting for hospital LOS and healthcare costs from the perspective of the Canadian health service provider. This analysis aimed to provide evidence for practice guideline development at an institutional level to inform the optimal medication choice for anxiolysis in pediatric EDs. Additionally, a previous systematic review identified substantial heterogeneity and uncertainty in the evidence of relative effectiveness.¹⁰ We, therefore, also aimed to identify key areas of future research to address these uncertainties using the TRansparent Uncertainty aSsessmenT (TRUST) tool³⁰ and a value of information (VoI) analysis.

2. Methods

2.1. Study Design

This economic evaluation compared the LOS and any healthcare costs incurred in pediatric EDs for children undergoing laceration repair. We estimated hospital LOS as it influences resource consumption and assumed no difference in long-term health outcomes among these interventions. We compared the costs of 3 commonly used drugs, INM, IVK, and N₂O,¹⁰ and 1 emerging novel agent, IND.³¹ A health economic analysis plan was not developed for this analysis.

The 3 drug interventions (ie, excluding N₂O) can be delivered by multiple routes.¹⁰ We selected administration routes based on published evidence of their usage in clinical practice. For example, INM is more common than oral midazolam in the pediatric ED.³² Despite being initially approved for intravenous use, intranasal dexmedetomidine has been widely used in pediatric sedation.^1,31,33 For ketamine, intravenous administration is more common as the intranasal route has poor absorption and requires large volumes of nasal spray that is often unacceptable to children.^34,35

2.2. Study Setting

The reference population was children aged 1 to 13 years requiring minimal anxiolysis for a simple laceration repair in the pediatric ED. We chose the doses identified by clinical experts (N.P. and S.A.) and literature: INM 0.4 mg/kg, IVK 1 mg/kg, 50% N₂O/50% oxygen, and IND 3 mcg/kg.^5,10,31 Our model inputs were informed by published articles that matched as closely as possible to these doses and patient populations. We excluded articles using multiple procedural medications for a single procedure, but included procedures supported by topical analgesia.

2.3. Model Structure

We developed an individual-level state-transition model (or microsimulation model)³⁶ to determine the LOS (in minutes) and the cost (in Canadian dollars [CAD]) for the studied medications. Simulated patients can move through 4 distinct states during a laceration repair: pre-procedure, procedure, post-procedure, and discharge (Fig 1), defined as follows:

• •Pre-procedure: the time between the administration of the first dose and start of the laceration repair.
• •Procedure: the time between the first suture in the laceration repair and tying the final suture.
• •Post-procedure: the time between tying the final suture and discharge from the hospital.
• •Discharge: patients are discharged from the hospital; they incur no costs and cannot be readmitted.

Figure 1.

A pictorial representation of the individual-level 4-state transition model used to estimate the in-hospital LOS due to the laceration repair and costs for 4 anxiolytic agents. Dashed arrows indicate that the event of interest extends the LOS in the given state. LOS, length of stay.

The model used a 1-minute cycle length with constant transition probabilities. We assumed ED LOS could be prolonged by a set of “events”:

• 1.Patients requiring an extra dose to achieve adequate anxiolysis (E1), which extends the pre-procedure state.
• 2.Patients being inadequately sedated and requiring restraint during the laceration repair (E2), which extends the procedure length.
• 3.Patients experiencing adverse events (AEs) from the anxiolysis (E3), which extends the post-procedure time. The AEs included emesis/vomiting, ataxia/unsteadiness/dizziness, and agitation/inconsolability, suggested by clinical experts (N.P. and S.A.) and literature review.¹⁰ Note that other potential serious AEs are observed very infrequently and, therefore, were not considered in this study.

2.4. Model Parameters

The Supplementary Appendix 1 details the assumptions and sources of the estimates for the model parameters. Our model inputs include (i) the probability of experiencing the events, (ii) the baseline LOS in each state, (iii) the prolonged LOS because of the key events, and (iv) the costs associated with laceration repair. When model inputs were unavailable in the literature, eg, the prolonged LOS because of AEs, we elicited information from clinical experts (N.P. and S.A.). Probabilistic distributions were elicited to understand the uncertainty in the model inputs.³⁷

2.5. Costs

The cost of laceration repair includes the cost of (i) medications, (ii) equipment for medication administration, (iii) personnel for administering repair and child care, and (iv) miscellaneous and administrative expenses. The personnel costs were assumed to be dependent on the LOS in each state. We assumed that 1 nurse was present at the bedside in all 3 states. An additional physician was assigned to the pre-procedure state for N₂O and IVK. In the procedure state, INM, IND, and N₂O require one physician, whereas IVK requires two physicians. Additionally, the cost of N₂O was assumed to vary based on the LOS in the pre-procedure and procedure states as it must be continuously available until the laceration repair is completed. All the costs are adjusted to 2022 CAD using the Consumer Price Index.³⁸

2.6. Analyses

To clarify the terms used in this analysis, we provide definitions of key terms in the Supplementary Appendix 1.

2.6.1. Microsimulation model

We simulated $s=$ 10,000 patient trajectories with our microsimulation model.³⁹ The number of patients was determined through convergence testing.⁴⁰ For each procedural medication, we initially determined the individual’s transition probability by generating whether the individual experienced any key events. The transition probabilities were then calculated, with lower transition probabilities for individuals who experienced events. The observed LOS in each state was then generated separately for every patient based on these transition probabilities, and the LOS across all three states was summed up to estimate the individual’s total hospital LOS.

2.6.2. Probabilistic sensitivity analysis

To account for parameter uncertainty, we undertook a probabilistic sensitivity analysis with 4000 simulations.⁴¹ The number of simulations was also selected to ensure that the length of the 95% CI (assuming normality) is ≤5% of the point estimate of primary cost and hospital LOS outcomes.⁴⁰ We reported the average and 95% credible interval (CrIs) of the hospital LOS and costs across 4000 simulations, which represents variation from parameter uncertainty.

2.6.3. Bias analysis

We undertook an uncertainty assessment using the TRUST tool for the impact of our model assumptions on the results.³⁰ Uncertainty can be attributed to four sources: (i) stochastic and parameter uncertainty, (ii) uncertainty due to the model structure, (iii) uncertainty in the distribution of the model inputs, and (iv) the quality of evidence used to inform the model.³⁰ First, stochastic uncertainty arises from subjects simulated in the microsimulation model and can be thought of as unexplained variability.⁴¹ Parameter uncertainty considers statistical uncertainty in the parameter estimates. Second, uncertainty arises from the model structure, which simplifies the actual relationships between the events and outcomes in a pediatric ED. Third, uncertainty regarding model inputs reflects the accuracy of distributions used to capture the uncertainty in our model inputs. In particular, the use of a specific parametric assumption and distributions to describe the uncertainty might not reflect the true uncertainty of the events. Finally, uncertainty arises from the quality of evidence used in the model. For example, the heterogeneity of the population across studies from which data were extracted and the lack of evidence for several model inputs.

We identified key assumptions for these uncertainty sources and evaluated their impact using a 5-point scale, in which 1 indicates low impact and 5 indicates major impact. We then discussed how we aimed to alleviate the impact of uncertainty and disclosed key limitations.

2.6.4. Value of information analysis

To explore the parametric uncertainty further and determine the potential value of undertaking future research on these medications, we performed a VoI analysis.⁴² We computed the Expected Value of Perfect Information (EVPI),⁴³ which provides the maximum potential value of eliminating uncertainty in our decision.⁴⁴ We also computed the Expected Value of Partial Perfect Information (EVPPI), which provides the maximum value of a study targeting a key parameter or set of parameters.⁴² The VoI analyses in this study aim to understand the impact of parameter uncertainty on the total costs, assuming that decision-makers are aiming to minimize costs.

EVPI and EVPPI are estimated at the individual level but should be scaled to the population level for research prioritization.⁴⁵ The population-level EVPI and EVPPI were calculated by multiplying the individual-level EVPI and EVPPI by the total number of patients experiencing laceration repair in the Canadian pediatric EDs. This number was estimated based on the number of children, aged from 0 to 14 years, visiting hospitals with injuries in 2018 or 2019 in Canada and the prevalence of lacerations in all pediatric injury care in EDs.^46,47 Population-level EVPI and EVPPI were discounted at 1.5% per year over a 10-year time horizon, assuming constant laceration repair numbers.⁴⁸

We followed the Consolidated Health Economic Evaluation Reporting Standards Value of Information (CHEERS-VOI) checklist⁴⁹ to report this study (Supplementary Appendix 1). All the analyses were conducted using R (version 4.3.1; R Foundation for Statistical Computing). This analysis was conducted with approval from the Hospital for Sick Children Research Ethics Board (1000080354).

3. Results

3.1. Hospital Length of Stay

The total hospital LOS for N₂O, INM, IVK, and IND were 78.34 (28.47, 150.14), 82.13 (65.04, 103.01), 103.37 (36.97, 194.53), and 114.55 (54.71, 203.3) minutes, respectively. This ranking is visually depicted in Figure 2 through density curves.

Figure 2.

The total hospital LOS and the cost required to complete the procedure of 4 interventions: INM, N₂O, IVK, or IND. IND, intranasal dexmedetomidine. INM, intranasal midazolam; IVK, intravenous ketamine; LOS, length of stay; N₂O, inhaled nitrous oxide.

The state-specific LOS is also shown in Table 1 and illustrated in Figure 3 using density curves. N₂O had the average shortest pre-procedure and post-procedure LOS, 8.54 and 50.57 minutes, respectively, whereas IVK has the shortest procedure time, which is 10.25 minutes. Notably, substantial uncertainty remains.

Table 1.

Mean and 95% CrIs of pre-procedure, procedure, post-procedure, total hospital LOS, and total cost per patient.

Anxiolytic	INM	N2O	IVK	IND
Pre-procedure (min)	15.13 (13.2, 17.23)	8.54 (1, 21.33)	10.58 (4.93, 16.8)	43.22 (29.43, 56.84)
Procedure (min)	15.01 (11.65, 18.4)	19.22 (1.9, 49.2)	10.25 (1.62, 21.53)	18.4 (14.35, 22.92)
Post-procedure (min)	51.98 (35.53, 72.35)	50.57 (13.2, 114.02)	82.55 (15.61, 168.63)	52.93 (1.03, 138.27)
Total LOS (min)	82.13 (65.04, 103.01)	78.34 (28.47, 150.14)	103.37 (36.97, 194.53)	114.55 (54.71, 203.3)
Anxiolytics (CAD)a	3.83	27.76 (4.31, 58.86)	5.6	11.07
Equipment (CAD)a	6.23	4	5.42	6.23
Personnel (CAD)	90.21 (74.11, 106.83)	116.11 (34.42, 220.17)	140.77 (67.68, 228.28)	120.17 (76.71, 185.03)
Misc (CAD)	22.51 (17.48, 27.6)	28.83 (2.84, 73.8)	15.37 (2.44, 32.3)	27.6 (21.52, 34.38)
Total cost (CAD)	122.79 (102.55, 142.6)	176.7 (53.31, 357.3)	167.16 (81.21, 263.32)	165.07 (116.57, 230.94)

CAD, Canadian dollar; CrIs, credible interval; IND, intranasal dexmedetomidine; INM, intranasal midazolam; IVK, intravenous ketamine; LOS, length of stay; N₂O, inhaled nitrous oxide.

^aThe anxiolytic costs associated with INM, IVK, and IND, as well as equipment expenses were assumed to be constant across individuals and simulations in this study. Conversely, the cost of N₂O varies depending on the individual length of stay in both the pre-procedure and procedure stages.

Figure 3.

Distribution of hospital LOS of 4 interventions at pre-procedure, procedure, and post-procedure. LOS, length of stay.

3.2. Costs

The per-person cost estimates are shown in Table 1. Among the four procedural medications, INM ($122.79) had the lowest cost. The cost of IND, IVK, and N₂O was $165.07, 167.16, and $176.7, respectively.

Figure 2 depicts the distribution of cost per patient given that they received either INM, N₂O, IVK, or IND. The dispersion of the expected cost for N₂O, IVK, and IND was wider than that of INM, which indicates higher uncertainty in their cost estimates. The probability that INM, N₂O, IVK, and IND had the lowest cost among all 4 interventions was 0.544, 0.287, 0.143, and 0.03, respectively.

3.3. Evaluation of the Uncertainty

Table 2 displays our TRUST analysis, evaluating the sources of uncertainty in key categories. The stochastic uncertainty is impacted by the number of simulations.⁴¹ Convergence exploration of the microsimulation model indicates minimal impact of this uncertainty on the results. Similarly, for the parameter uncertainty, a probabilistic sensitivity analysis with 4000 simulations ensures that the population-mean cost is estimated within 5% of the mean value.⁴⁰

Table 2.

The results from the transparent uncertainty assessment tool, which aims to identify key factors driving uncertainty, imprecision, and bias uncertainty in our model. Our assessment of the impact of these factors on the model results presented in this manuscript is evaluated using a 5-point scale, in which 1 indicates minimal impact and 5 indicates maximum impact.

Source of uncertainty	Imprecision, bias, and indirectness	Rating
Stochastic and parameter uncertainty	1. The use of simulations to explore these uncertainties. Our model used simulations to understand individual-level and parameter uncertainty, with more simulations providing more accuracy.	1
Model structure	2. Our transition model uses the Markov “memoryless” property. Constant transition probabilities in our model assume that the probability of transitioning out of a state does not depend on the patient’s previous length of stay in the state, which is a simplification of reality.	1
	3. Using “onset of sedation” as a proxy for the length of the pre-procedure state The pre-procedure time of some anxiolytics was not available in the literature; thus, the outcome “onset of sedation,” which does not consider the time taken to prepare or administer the intervention, was used as the proxy.	2.5
	4. Only modeling common adverse events. Our model did not consider rare adverse events from the different anxiolytics, which could lead to biased results if these rare adverse events differ substantially across interventions.	2
	5. Ignoring differences in the LOS across anxiolytics for some key events. Our model assumed that there were no differences in the additional LOS due to inadequate sedation or the adverse events across the anxiolytics.	3
Model inputs	6. Estimating the probabilistic distribution of the LOS using the Method of Momentsa Only the mean and standard error for the model parameters were available in the literature, meaning that the probabilistic distribution may not reflect the true uncertainty.	2.5
Quality of evidence	7. Heterogeneous population Substantial heterogeneity in the patient population was seen in the literature.	4
	8. Alternative statistics used in reporting We aimed to extract the mean, standard deviation, and sample size from studies, but often, only the median, range, and/or quantiles were reported.	2.5
	9. Studies had small sample sizes. Many of the available studies enrolled fewer than 50 patients.	2.5
	10. Lack of evidence Information on key model parameters was not available in the literature.	5
	11. Using elicitation to inform model inputs Data for some model parameters were not available and were elicited from clinical experts.	2.5
	12. Cost data were extracted from different years. Costs were extracted from a range of sources across multiple years and then adjusted to 2022 Canadian dollars using the CPI.	1.5

CPI, Consumer Price Index; LOS, length of stay.

^aThe Method of Moments assumes that the parameters of a distribution can be estimated by “matching” the mean and standard deviation between the available data and the distribution of interest.

The uncertainty due to the model structure can be split into 4 key categories: the models used to estimate transition probabilities, the specification of states in the model, the inclusion of key events, and the assumption of no difference in the LOS across medications for key events. These assumptions impact the model’s ability to represent reality.⁴¹ These assumptions have an average rating of 2.13, with the largest impact associated with the assumption that the additional LOS because of inadequate anxiolysis and AEs did not vary across the different interventions.

The uncertainty from the model input distributions arises as the selected uncertainty distributions were not specified in the literature. We had to match the mean and standard error and specify the uncertainty distribution ourselves (Supplementary Appendix 1). This approach may not reflect enough of the true uncertainty in the LOS and, thus, bias the estimate of hospital LOS.

Finally, the uncertainty related to the quality of evidence had the highest impact on the outcomes, with an average rate of 3. Although this uncertainty was reduced by incorporating evidence from multiple sources, some model inputs lacked evidence because studies failed to meet the prespecified inclusion or exclusion criteria.

3.4. Value of Information

The overall EVPI indicated the maximum potential value of reducing parameter uncertainty at $11.85 per person. This implies that the maximum cost of a research study in this population should be below $484,816 to ensure it is considered worthwhile. Future research should focus on the procedure time for N₂O (EVPPI: $5.31) and IVK (EVPPI: $0.44), within a financial budget constrained by the corresponding population-level EVPPI of $217,407 and $18,088, respectively. We also calculated EVPPI for the combination of the procedure time for N₂O and IVK, which yielded a population EVPPI (per-person EVPPI) of $231,834 ($5.67).

4. Limitations

A major limitation of this study is the availability of evidence. We are aware of other anxiolytics, but we only included 4 comparators in this study based on our previous systematic review.¹⁰ We faced challenges in matching the patient population, dose, and route of administration from literature. The availability of evidence on INM likely contributes to the smallest CrIs for LOS and cost estimates, whereas only a few small studies report LOS for IVK and N₂O. In general, a pool of very few studies threatens the validity of the evidence because the heterogeneity is often estimated inadequately.^50,51

Limited evidence prevents our simulation model from addressing more specific research questions. For example, because of limited evidence, individual-level baseline characteristics, including age at admission, were not included in our model. Therefore, we are unable to provide more personalized recommendations based on age or other drivers of anxiety. Another limitation was the lack of patient-oriented effectiveness measures and patient engagement. The measurement of health utilities is more methodologically challenging for children.⁵² Many studies considered anxiety scales, but the heterogeneity of these scales precluded any formal analysis of the interventions’ effectiveness in reducing anxiety.¹⁰ A range of long-term impacts of poorly managed procedural distress have been reported, including long-term avoidance of medical care.^3,4 However, because of lack of data, we could not include these long-term effects, possibly leading to an incomplete summary of the interventions’ benefits. Future research should investigate these potential long-term effects.

In addition, our model did not account for variations in ED patient volume based on time of day or day of the week, nor did it consider healthcare resource capacity, which could confound our findings. Future research could use a more complex model to incorporate patient volume fluctuations and better understand their impact on LOS.

Finally, this study was designed to inform the Canadian pediatric practice. Although physician time was deliberately excluded from the post-procedure state to reflect typical clinical workflow in Canada, it may be part of the workflow in other jurisdictions. Researchers in other settings could adapt our approach to address context-specific questions.

5. Discussion

We recognize that the intricacies of clinical practice can influence decision-making. For example, the location of the repair may modify the findings. If the laceration is underneath the face mask, N₂O may theoretically be less viable, although in practice, the mask can be lifted for suturing and then reapplied. For longer repairs, such as layered closures, ketamine may be used for deeper sedation, especially in younger patients. Furthermore, INM was used more frequently¹ and this familiarity and breadth of evidence could also make INM more preferable. In such cases, additional economic evaluation is likely required. In addition, some jurisdictions or EDs may be unable to choose from all the studied medications in our analysis. In this case, the decision-makers should focus on the available agents in their setting and select the optimal intervention from among them.

We conducted the first comprehensive evaluation, to our knowledge, of 4 procedural medications (INM, IND, N₂O, and IVK) for anxiolysis during laceration repair in the pediatric ED. Costs were largely driven by personnel requirements. INM had the lowest cost, as it required fewer staff. Similarly, although IND had the longest pre-procedure time, it required only minimal staffing (1 nurse), resulting in the second-lowest cost estimate. Although N₂O had the shortest hospital LOS, its prolonged procedure time and the need for physician involvement made it the most expensive option. Notably, except for INM, the other 3 comparators had wide CrIs for total cost estimates, leading to substantial decision uncertainty in scenarios in which INM is not available. We identified key model uncertainties and VoI analyses highlighted the need for future research on procedure time for N₂O and IVK.

Author Contributions

N.T., A.H., and P.P. conceived the study and designed the work. N.T. and A.H. collected the data, whereas N.T. and Y.L. analyzed and interpreted the data. N.T. and Y.L. drafted the manuscript, and Y.L., P.P., S.A., D.C., N.P., and A.H. critically revised it. All authors reviewed and approved the final version for submission.

Funding and Support

This study was supported by the Government of Canada’s New Frontiers in Research Fund (NFRFE-2020-00749). A.H. is supported by a Canada Research Chair in Statistical Trial Design. Y.L. is supported by a grant from the Canadian Institutes of Health Research (grant 463252).

Ethical Considerations

This study was conducted with the approval of the Hospital for Sick Children Research Ethics Board (1000080354).

Conflict of Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.