Identification of Abusive Head Trauma in High-Risk Infants: A Cost-Effectiveness Analysis

Abstract

Objectives:

To evaluate the cost-effectiveness of abusive head trauma detection strategies in emergency department (ED) settings with and without rapid magnetic resonance imaging (rMRI) availability.

Study design:

A Markov decision model estimated outcomes in well-appearing infants with high-risk chief complaints. In an ED without rMRI, we considered 3 strategies: clinical judgment, universal head computed tomography (CT), or Pittsburgh Infant Brain Injury Score (PIBIS) with CT. In an ED with rMRI for brain availability, we considered additional strategies: universal rMRI, universal rMRI with CT, PIBIS with rMRI, and PIBIS with rMRI followed by CT . Correct diagnosis eliminated future risk; missed AHT led to re-injury risk with associated poor outcomes. Cohorts were followed for one year from a healthcare perspective. One-way and probabilistic sensitivity analyses were performed. Main outcomes evaluated in this study were AHT correctly identified and incremental cost per quality-adjusted life-year.

Results:

Without rMRI availability, PIBIS followed by CT was the most cost-effective strategy. Results were sensitive to variation of CT-induced cancer parameters and AHT prevalence. When rMRI was available, universal rMRI followed by confirmatory CT cost $25,791 to gain one additional quality-adjusted life-year compared with PIBIS followed by rMRI with confirmatory CT. In both models, clinical judgement was less effective than alternative strategies.

Conclusions:

By applying CT to a more targeted population, PIBIS decreases radiation exposure and is more effective for AHT identification compared with clinical judgment. When rMRI is available, universal rMRI with CT is more effective than PIBIS and is economically favorable.

Abusive head trauma is the leading cause of fatal traumatic brain injury in infants.^1,2 One third of children with AHT are initially misdiagnosed, contributing to increased morbidity and mortality.^3–6 Diagnosis can be challenging as there often is no reported history of trauma and presenting symptoms are nonspecific.^7,8 The standard criterion for diagnosis is abnormal head computed tomography (CT), an imaging modality associated with significant radiation exposure, particularly for young infants.^9–11 Balancing a desire to have a low-threshold to evaluate infants for AHT with a desire to minimize radiation exposure poses a clinical challenge.

The Pittsburgh Infant Brain Injury Score (PIBIS) is a validated tool to identify infants in the emergency department (ED) at high risk of AHT and most likely to benefit from head CT.¹² PIBIS offers improved ability to identify at-risk infants and, with this, introduces the potential for increased imaging overall. Rapid magnetic resonance imaging (rMRI) of the brain has emerged as an alternative to CT for identification of AHT without the risk of radiation or sedation.^13–15 rMRI has the potential to increase cost and has limited availability. The optimal application of PIBIS and selection of imaging modality to optimize medical costs, radiation exposure, and clinical outcomes is not established.

The decision to incorporate PIBIS into clinical decision making must be weighed against the effectiveness and cost of traditional detection strategies, factoring in differences in effectiveness, cost, availability of imaging modalities, and the risk of radiation-induced cancer. We used decision modeling techniques to address these issues and evaluate the cost-effectiveness of different strategies to identify infants with AHT.

Methods

We performed a cost-utility analysis to compare outcomes, costs, and cost-effectiveness of strategies to identify AHT in a hypothetical cohort of 1,000 infants presenting to the ED with high-risk chief complaints. A decision-analytic Markov model simulated transitions between health states. The decision model was programmed in TreeAge Pro 2016 (TreeAge Software, Inc, Williamstown, MA).

Study Setting and Population

We considered two scenarios: an ED in which CT is available and rMRI is not, and an ED having both CTH and rMRI readily available. Our base case was that of a well-appearing 4-month-old infant presenting to the ED with a high-risk chief complaint. High-risk chief complaints included: vomiting without diarrhea, fussiness, seizure or spell, brief resolved unexplained event, feeding difficulties, or nonspecific complaint.^4,10,12 Infants were assumed to have no stated history of trauma.

PIBIS Screen

PIBIS is a validated clinical prediction rule for infants <12 months of age presenting to an ED with a high-risk chief complaint. It is designed to identify infants most likely to benefit from neuroimaging to evaluate for AHT. AHT risk is assessed using a 5-point scoring system. Two points are assigned for bruising, and one point for each of the following: age ≥3 months, head circumference >85^th percentile, or hemoglobin <11.2 g/dL.¹²

Model Design

The decision model included seven health states: 1) well, 2) AHT diagnosed and treated, 3) missed AHT, 4) well following missed AHT, 5) recurrent AHT, 6) severe neurologic disability, and 7) death. Infants begin the model either with or without AHT. All infants have a baseline risk of death and disability. Infants diagnosed with AHT incur costs of medical treatment and return to a well state. Because the model considers a select population of well-appearing infants, correctly diagnosed infants at initial presentation are assumed to attain full recovery. Costs of medical treatment were equally applied to both true and false positive diagnoses of AHT. Infants with missed AHT have an increased risk of death, disability, and reinjury. Reinjured infants represent to the ED. Those who are diagnosed incur costs of medical treatment and return to a well state, and those with recurrent missed AHT remain at risk for death, disability, and reinjury. Transition between health states is shown in Figure 1. We used a one-year time horizon and tracked disutilities for long-term outcomes as outlined below. Disutility was defined as decrease in quality of life and/or length of life associated with a particular event or health state.¹⁶

Figure 1.

Model schematic. Infants begin the model in the ED with a high-risk chief complaint and with or without AHT. After undergoing strategy-specific AHT screening, they are either correctly or incorrectly diagnosed. Infants without AHT who screen negative return to the well state. Those without AHT who screen positive incur costs of care and return to the well state. Correctly diagnosed infants with AHT incur the same costs of care and return to the well state. Infants with missed AHT have an increased risk of death and disability and can progress to the well state after AHT; in this health state infants are well but remain at risk for recurrent AHT for several cycles. Infants who experience recurrent AHT represent to the ED where they are either correctly or incorrectly diagnosed. All infants have a baseline risk of death (not shown).

ED, emergency department; AHT, abusive head trauma

For a CT-only ED, we considered 3 strategies for identifying AHT: clinical judgment, PIBIS with head CT (PIBIS+CTH), and universal head CT. In the clinical judgment strategy, we assumed imaging was at the discretion of the physician and, based on practice patterns of the past 30 years, a sensitivity of 70% was assigned.^3,4 A specificity of 95% for this strategy was assumed. For PIBIS+CT, all infants received a PIBIS score. Those with a score of 2 or higher underwent CTH. In universal CTH, all infants with high-risk chief complaints underwent CTH. In an rMRI-capable ED, we considered the impact of rMRI in evaluation of an identical hypothetical infant cohort. Four strategies were added to those considered in the CT-only ED model: PIBIS with rMRI for infants with PIBIS score ≥2 (PIBIS+rMRI); PIBIS with rMRI for infants with PIBIS score ≥2 followed by confirmatory CT for abnormal or equivocal rMRI (PIBIS+rMRI+CT); universal rMRI; and universal rMRI followed by confirmatory CT for abnormal or equivocal rMRI (universal rMRI+CT). Strategies combining rMRI and CT were based on previously published studies.^13,14,17,18 Hypothetical rMRI-only strategies were analyzed in keeping with International Society for Pharmacoeconomics and Outcomes Research recommendations to consider all plausible strategies.¹⁹

Input parameters for probabilities, costs, and outcomes are presented in Table 1. For each category, we included an estimated 95% probability range. Probabilities of outcomes from undiagnosed AHT were derived from published literature (Table 1), with ranges accounting for variation among sources. We included risk for radiation-induced cancer from head CT. A baseline risk of neurologic disability was estimated. Costs include direct medical costs of ED visits, detection strategies, hospitalization, and medical treatment. The analysis took a healthcare perspective, and thus indirect costs were not included in the model. All-cause mortality was estimated using U.S. National Center for Health Statistics life tables.²⁰

Table 1.

Model Inputs: Baseline Parameter Values and Ranges

Description	Point Estimate (Range)
Probabilities
Risk of AHT12,41,45,46,48	3% (0-4%)
Risk of recurrent AHT3–5	39% (28-53%)
CTH13,17,18,60
Sensitivity	99% (95-100%)
Specificity	98% (94-100%)
rMRI Sensitivity13,14,18
Sensitivity	98% (95-99%)
Specificity	91% (80-99%)
PIBIS Score ≥212
Sensitivity	93% (74-100%)
Specificity	53% (42-64%)
Clinical judgement3,4,17,53
Sensitivity, first ED visit	70% (69-75%)
Sensitivity, second ED visit	92% (85-99%)
Specificity	95% (90-100%)
Risk of radiation-induced cancer, CTH10,11,39,40	0.1% (0.02-0.2%)
Risk of death, baseline20	0.54%
Risk of death, AHT misdiagnosis25	10% (5-20%)
Risk of disability, baseline	0.11% (0.1-0.12%)
Costs a
ED visit61	560 (448-672)
CTH62	117 (94-250)
rMRI62	232 (186-360)
Complete blood count63	12 (10-14)
Hospitalization, AHT64	21,995 (17,596-26,394)
Severe disability, first year of life55,65–68	5,824 (824-10,824)
Utilities and Disutilities b
Well, infant27,28	0.95
AHT27,28	0.88 (0.65-0.97)
Recurrent AHT27,28	0.51 (0.39-0.63)
Severe neurologic disability27,28	0.59 (0.36-0.83)
Radiation induced cancer, disutility26	9.9 (8.3-11.5)
Death in infancy, disutility20	30.98
Discount rate24	0.03

^aCosts are in 2016 US dollars.

^bDisutility values are lifetime quality adjusted life years lost.

AHT, abusive head trauma; CTH, computed tomography of the head; rMRI, rapid brain magnetic resonance imaging; PIBIS, Pittsburgh infant brain injury score; ED, emergency department

All costs were adjusted to 2016 US dollars based on the medical cost component of the Consumer Price Index.²¹ Imaging costs included costs of performing the test and interpretation by a radiologist. We assumed a willingness to pay of $100,000 per quality adjusted life year (QALY) gained, a commonly cited benchmark for the US healthcare system.²²

Health state utilities were assigned a value of 0-1, with 0 equivalent to death and 1 representing perfect health.²³ The disutility of radiation induced cancer and infant mortality were factored as lifetime disutilities. All costs and utilities, including QALYs lost due to infant mortality, were discounted at 3% per year, as recommended for cost-effectiveness analysis design.²⁴ QALY loss was derived from the literature.^25–28

Outcome Measures

The primary outcomes evaluated were effectiveness (AHT cases correctly identified), cost, and cost-effectiveness (cost per case identified) for each strategy. Strategies were ranked by cost then compared in terms of cost, effectiveness, and incremental cost-effectiveness ratio (ICER: additional cost in dollars per event or disutility cost averted). Secondary outcomes included hospitalizations, deaths, and QALYs lost. In the cost-effectiveness calculation, effectiveness was tracked as a disutility, representing lost quality and duration of life from AHT, diagnostic strategies, medical management, and death. A secondary analysis of cost-per-case was performed.

Sensitivity Analyses

We conducted one-way sensitivity analyses to determine if varying any single parameter across its listed range (Table I) substantially changed results. Probabilistic sensitivity analyses (using 1,000 simulated event combinations, simultaneously varying all parameter values over distributions) were performed to estimate uncertainties in the primary and secondary outcomes resulting from that variation. Distributions were chosen to reflect the level of certainty and the characteristics of the parameter range and methodological standards. β distributions were used for probabilities and quality adjustments; γ distributions were used for costs. Threshold analyses were performed to determine the point at which changes to input parameters resulted in differing strategies being preferred. A structural sensitivity analysis was performed to test the assumption of full recovery after correct diagnosis of AHT. In this analysis, correctly diagnosed infants with AHT experienced loss of QALYs after accruing medical costs of treatment, varied over a range of values.

Results

CT-only ED model

In the base-case analysis, clinical judgement was the least expensive and the least effective strategy (with a cost of $1,237, and a loss of 0.482 QALYs). PIBIS+CT was preferred, costing an additional $17,722/QALY gained (Table 2). Universal CT was more effective and more costly than PIBIS+CTH, exceeding the $100,000/QALY willingness to pay threshold, indicating that the added cost outweighed added effectiveness. Comparative clinical outcomes in a hypothetical population are shown in Table 3 (available at www.jpeds.com).

Table 2.

Results of cost-effectiveness analyses

	Cost ($)	Incremental Cost ($)	Effectiveness (QALY)	Incremental Effectiveness (QALY)	ICER ($/QALY)
CTH-only Emergency Department
Clinical judgement	$1,237	-	−0.482	-	-
PIBIS+CTHa	$1,561	$324	−0.464	0.018	$17,722
Universal CTH	$1,865	$304	−0.462	0.002	$161,238
rMRI-capable Emergency Department
Clinical judgement	$1,237	-	−0.482	-	-
PIBIS+rMRI+CTH	$1,437	$199	−0.461	0.021	$9,476
PIBIS+CTH	$1,561	$124	−0.464	−0.003	Dominatedb
Universal rMRI+CTH	$1,597	$160	−0.455	0.006	$25,791
Universal CTH	$1,865	$268	−0.462	−0.007	Dominated
PIBIS+rMRI only	$2,384	$787	−0.458	−0.004	Dominated
Universal rMRI only	$3,611	$2,015	−0.451	0.004	$473,842

^aBold text: Favored strategy at a $100,000 per quality adjusted life year threshold.

^bA dominated strategy is more costly and less effective than other strategies.

QALY, quality adjusted life years; ICER, incremental cost-effectiveness ratio; PIBIS, Pittsburgh infant brain injury score; CTH, computed tomography of the head; rMRI, rapid brain magnetic resonance imaging.

Table 3.

Outcomes by strategy in a population of 1,000 infants with high-risk chief complaints, of whom 30 have AHT

Strategy	AHT Cases, n	Correctly diagnosed AHT, % (n)	Missed AHT, % (n)	Recurrent AHT, % (n)	False positive AHT, % (n)	Radiation-induced cancer, %
Clinical judgement	30	70 (21)	30.0 (9)	17.1 (5)	<0.1 (0)	0.01
PIBIS+CTH	30	92 (28)	7.9 (2)	4.5 (1)	0.1 (1)	0.05
Universal CTH	30	99 (30)	1.0 (0)	<0.01 (0)	2.0 (19)	0.1
PIBIS+rMRI only	30	92 (28)	7.9 (2)	4.5 (1)	5.0 (44)	0
PIBIS+rMRI+CTH	30	91 (27)	8.8 (3)	5.1 (2)	0.2 (2)	0.01
Universal rMRI only	30	99 (30)	1.0 (0)	<0.01 (0)	9.5 (92)	0
Universal rMRI+CTH	30	98 (29)	2.0 (1)	<0.01 (0)	0.2 (2)	0.01

AHT, abusive head trauma; PIBIS, Pittsburgh infant brain injury score; CTH, computed tomography of the head; rMRI, rapid brain magnetic resonance imaging

In one-way sensitivity analyses, results were impacted by changes in several key variables, including radiation-induced cancer disutility and risk, and AHT risk (Table 4; available at www.jpeds.com). Results were not sensitive to variation in costs associated with ED evaluation, neuroimaging, or transient quality of life parameters (Figure 2; available at www.jpeds.com). Threshold analyses demonstrated that universal CT would be favored if the risk and disutility of radiation-induced cancer were lower or if AHT risk was higher (Table 4). Clinical judgement was favored when the risk of AHT was <0.9%. Structural sensitivity analysis did not change overall model outcomes. Results of the cost-per-case analysis are shown in Table 5 (available at www.jpeds.com).

Table 4.

One-way sensitivity analysis results

			Preferred Strategy
Variable	Base-case	Threshold	Below threshold	Above threshold
CTH-only ED
Radiation-induced cancer
Risk	0.1%	0.077%	Universal CTH	PIBIS+CTH
Disutility	9.9 QALY	7.7 QALY	Universal CTH	PIBIS+CTH
CTH specificity	98%	99.1%	Universal CTH	PIBIS+CTH
Risk of AHT	3%	3.5%	PIBIS+CTH	Universal CTH
Risk of AHT	3%	0.9%	Clinical judgement	PIBIS+CTH
rMRI-capable ED
rMRI specificity	91%	99.3%	Universal rMRI+CTH	Universal rMRI only
PIBIS sensitivity	93%	98.0%	Universal rMRI+CTH	PIBIS+rMRI+CTH
Risk of AHT	3%	0.8%	PIBIS+rMRI+CTH	Universal rMRI+CTH

ED, emergency department; QALY, quality adjusted life years; CTH, computed tomography of the head; AHT, abusive head trauma; PIBIS, Pittsburgh infant brain injury score; rMRI, rapid magnetic resonance imaging of the brain

Figure 2.

One-way sensitivity analysis of utility values tested across a range of plausible values. The impact on the ICER for PIBIS+CT compared with the clinical judgement strategy in CT-only ED model is shown on the x-axis. Changing the value of the lifetime disutility associated with radiation-induced cancer shifts the ICER by nearly $3,000/QALY but does not cause PIBIS+CT to become less costly than CT. Changing other utility values had a smaller impact. QALY, quality adjusted life years; AHT, abusive head trauma; PIBIS, Pittsburgh infant brain injury score; CT, computed tomography; ED, emergency department; ICER, incremental cost-effectiveness ratio

Table 5.

Cost-per-case analyses in a population of 1,000 infants, of whom 30 have AHT

Strategy	Total cost per patient, $	AHT correctly diagnosed, n	Cost per case correctly diagnosed, $	Recurrent AHT, n	Cost per recurrent AHT averted, $
Clinical judgement	$1,236	21	$58,857	5	$49,440
PIBIS+CTH	$1,560	28	$55,714	1	$53,793
Universal CTH	$1,865	30	$62,167	0	$62,167
PIBIS+rMRI only	$2,384	28	$85,143	1	$82,207
PIBIS+rMRI+CTH	$1,436	27	$53,185	2	$51,286
Universal rMRI only	$3,611	30	$120,367	0	$120,367
Universal rMRI+CTH	$1,596	29	$55,034	0	$53,200

AHT, abusive head trauma; ICER, incremental cost-effectiveness ratio; PIBIS, Pittsburgh infant brain injury score; CTH, computed tomography of the head; rMRI, rapid brain magnetic resonance imaging

Probabilistic sensitivity analysis results are summarized as acceptability curves, showing the likelihood that strategies are favored over a range of willingness to pay (or acceptability) thresholds, as shown in Figure 3. PIBIS+CT remained the preferred strategy from a willingness to pay of $20,000 to $200,000/QALY. At a willingness to pay of $100,000/QALY, PIBIS+CT was favored 64% of the time (Figure 3, top panel).

Figure 3.

Probabilistic sensitivity analysis. Results are shown as a cost-effectiveness acceptability curve. The y-axis shows the likelihood that strategies would be considered cost-effective for a range of cost-effectiveness willingness to pay thresholds (x-axis).

ED, emergency department; QALY, quality adjusted life year; CT, computed tomography, PIBIS, Pittsburgh infant brain injury score, rMRI, rapid magnetic resonance imaging

rMRI-capable ED model

With the addition of rMRI strategies, clinical judgement remained the least expensive strategy. PIBIS+rMRI+CT was more effective and cost $9,476/QALY gained. Universal rMRI+CT was the favored strategy, costing an additional $25,791/QALY gained (Table 2). Universal rMRI alone cost >$400,000/QALY. All other strategies were less effective and more costly. In one-way sensitivity analyses, results were sensitive to rMRI specificity, PIBIS sensitivity, and AHT risk (Table 3). Probabilistic sensitivity analysis indicated that at a threshold of $100,000/QALY, universal rMRI+CT was favored 79% of the time (Figure 3, bottom panel).

Discussion

We found that applying PIBIS to identify AHT in a CT-only ED setting was more cost-effective than either clinical judgment or universal CT. When rMRI was available, PIBIS was again a cost-effective option but universal rMRI with CT for abnormal or equivocal findings was preferred. Universal rMRI+CT was more expensive but more effective than PIBIS, adding QALYs at economically reasonable rates.

One of the strengths of this study is the consideration of two ED settings. In keeping with the “as low as reasonably achievable” principle, radiation exposure must be minimized and alternative means of diagnosis sought when possible.²⁹ rMRI is suggested as an alternative to CT, but its availability remains limited. More than 90% of children seeking emergency medical care are evaluated in non-specialized EDs, many of which do not have rMRI capabilities.^30,31 Thus, strategies to identify AHT in EDs with only CT available must be evaluated.

We found PIBIS+CT was preferred for AHT prevalence up to 3.3%. An effective clinical decision rule has strong predictive power, changes physician decision making, and has minimal implementation barriers.³⁴ PIBIS uses simple scoring criteria and is practical to implement.¹² Moreover, PIBIS provides an objective rationale for pursuing imaging, as opposed to prior recommendations of awareness or a high index of suspicion.^35,36 Campbell et al found a cost-savings advantage of using CT to identify AHT, compared with clinical judgment, was present when AHT prevalence was >1.8%.²⁵ Our model adds to these findings by offering a strategy to apply CT to a more targeted population, increasing the yield and decreasing infant radiation exposure.

We selected a PIBIS score of 2 as the evaluation threshold in our base case. The PIBIS study authors do not make recommendations on the optimal application of PIBIS, instead publishing sensitivity and specificity by score. Our sensitivity analysis demonstrated that a PIBIS sensitivity >98% would make PIBIS+rMRI+CT the preferred strategy in the rMRI capable ED model. A PIBIS score of 2 has a sensitivity of 93% and specificity of 53%. A PIBIS score of 1 has 99% sensitivity and 12% specificity.¹² Due to the marked drop in specificity, it is unlikely that a threshold score of 1 offers an advantage.

In our CT-only ED model, sensitivity analyses indicated that both the radiation-induced cancer risk and disutility substantially affected results, suggesting that the cost-effectiveness of universal CT is sensitive to potential radiation effects, an area of uncertainty. We used a radiation-induced cancer risk of 1 in 1000. Others have suggested that the risk is as low as 1 in 3,000-10,000.^8,37–40 Our use of what may be a high value for radiation risk reflects caution in the analysis. Despite this, PIBIS+CT was the preferred strategy in the CT-only ED.

The utility and application of rMRI in AHT continues to be studied in multiple US children’s hospitals.^{12,13,15,17,32,33} To our knowledge, there is no prior evaluation of rMRI cost-effectiveness compared with CT in these patients. When rMRI was available, universal rMRI followed by CT was favored. Without the risk of radiation-induced cancer, the upfront costs of imaging all infants with high-risk chief complaints are outweighed by the prevention of recurrent abusive injuries and fatalities. Although universally imaging a population of infants may seem radical, in our model, this $232 test⁴¹ significantly decreased the morbidity and mortality associated with the clinical judgment strategy. Universal rMRI+CT is more sensitive than PIBIS and decreases radiation risks compared with CT for all.

Many of the identified high-risk chief complaints suggest potential neurologic pathology. In the PIBIS validation cohort of patients, 9% of those with neuroimaging abnormalities had atraumatic findings, including hydrocephalus, tumors, and stroke.¹² AHT is one of many potential diagnoses for infants presenting with neurologic symptoms. It is not unreasonable to obtain neuroimaging when potential explanations include diagnoses that must be managed emergently and for which the consequences of misdiagnosis are deterioration and death.

There are several limitations to this study. AHT prevalence in this population is unknown. Population-based reports published more than a decade ago indicate an incidence of approximately 1 in 3000 infants, but this rate primarily reflects fatal or severe AHT.^1,42–45 More recent studies of non-fatal AHT, based on the Centers for Disease Control definition, suggest that AHT incidence may be higher.^45,46 Infants presenting to the ED for vague or neurologic complaints are a select group. Studies reporting AHT prevalence among infants presenting with apparent life-threatening events are retrospective, single institution studies^47–49 performed prior to terminology change from “apparent life-threatening events” to “brief resolved unexplained events”.⁵⁰ AHT prevalence in the PIBIS validation was >10%, but this is an overestimate, because 100% of patients diagnosed with AHT during the study period were enrolled whereas controls were enrolled selectively.¹² In the rMRI-capable ED model, universal rMRI+CT is preferred for an AHT prevalence down to 0.8%.

Similarly, recurrent AHT risk in missed infants is unknown and outcomes in children with missed AHT who are never diagnosed are difficult to measure. Data were derived from studies of children diagnosed with AHT for whom missed opportunities for diagnosis were retrospectively identified.^3–5 To address these uncertainties, wide ranges for risks, costs, and utilities associated with missed AHT outcomes were tested in sensitivity analyses and did not impact model outcomes.

Clinical judgment parameters were based on available literature. Rates of missed AHT were derived from studies of US children’s hospitals and may not reflect rates in non-specialized EDs.^{3,4,6,51–53} The assumed 95% specificity for clinical judgment biases results in favor of this strategy. Despite this, clinical judgment was less effective than alternative strategies in all analyses. Varying the sensitivity of clinical judgement across the range of tested values did not change model outcomes.

We found a range of values reported for rMRI sensitivity and specificity for AHT, with more recent studies reflecting improved specificity.^13–15 In these studies, abnormal or equivocal rMRI findings were compared with CT imaging. Lindberg et al compared rMRI with CT in pediatric patients with known trauma, showed that rMRI was less sensitive for linear, nondepressed skull fractures and was able to identify traumatic brain injury in 5 patients not identified with CTH.¹⁸ More data are needed to fully define the optimal application of rMRI for AHT.

We assumed that all infants correctly diagnosed with AHT would attain full recovery, including a subset of those diagnosed after a recurrent AHT episode. We attempted to account for this by: limiting our cohort to well-appearing infants, including the potential for long-term disability for those with missed AHT, including a 10% mortality risk among infants with repeat injury, and performing a sensitivity analysis in which correctly diagnosed infants did not return to their previous well state but experienced a decreased quality of life. AHT comprises a wide spectrum of severity. Inherent to the challenge of identification is that severely injured children may appear well. Reinjured children experience escalating morbidity and mortality.^4,5,54 AHT sequela go beyond medical treatment and physical healing. Children who experience abuse may go on to develop epilepsy, visual impairment, cognitive, behavioral, mood, and sleep disorders.^55–57 Added consideration of disease complexity, diagnostic challenges, and costs (in both quality of life and medical expenses) of missed AHT further supports the need to improve current practice.

We used a one-year time horizon as the selected clinical scenario is unique to infancy. Because of this, the long-term negative impacts of death and radiation-induced cancer were accounted for using discounted lifetime QALYs lost. We did not model lifetime negative impacts of AHT, biasing the analysis against strategies that minimize missed AHT.

We adapted utility values from the Glasgow Outcome Scale-Extended Pediatric utility weights. Infant health state utilities, particularly for victims of abuse, are poorly defined and understudied.⁵⁸ It is possible that an older child’s experience with head injury or physical abuse is different from that of an infant. For this reason, selected utility values were varied over wide ranges. When lifetime disutility of radiation-induced cancer was <8.6 QALYs, universal head CT was preferred in the CT-only model. Varying remaining utility values did not change favored strategies (Figure 2).

We did not account for the societal impact of improved AHT identification with subsequent enlistment of police and social services. The short-term impact of these costs can be substantial and the long-term societal economic impact of AHT on educational expenses, economic contributions, and healthcare expenditures are difficult to quantify. Previous studies noted that the societal perspective weighs acute costs more heavily and could suggest an ethically-concerning conclusion that AHT may be too expensive to diagnose.¹⁷ This study focused on novel decision-making tools in the emergency setting, and how can we effectively and efficiently evaluate an infant with symptoms of neurologic pathology from the perspective of clinicians and health systems.

We used a healthcare perspective, which does not consider the perspectives of individual hospitals, providers, and patients. Although neuroimaging enhanced the more cost-effective strategies in these models, ED length of stay associated with additional testing and radiologic image interpretation could be affected. Data from validation of the PIBIS score suggested that imaging frequency with PIBIS would not increase significantly beyond current practice.¹² Finally, we did not evaluate the perspective of families and caregivers, the costs of missed or lost employment, the impact on siblings in the home, or the aftereffects of a child abuse investigation, confirmed or not, on individuals and relationships.^59,60

Our findings suggest that more sensitive detection strategies can improve diagnostic accuracy and decrease costs. In an ED setting with only CT available, PIBIS offers a more cost-effective identification strategy than clinical judgment. In an ED with rMRI availability, universal rMRI+CT is more effective than PIBIS and is economically favorable.

Abbreviations

CT

computed tomography

PIBIS

Pittsburgh Infant Brain Injury Score

ED

emergency department

rMRI

rapid magnetic resonance imaging

QALY

quality-adjusted life-years

ICER

incremental cost-effectiveness ratio