Abstract
OBJECTIVE. Early diagnosis of inflicted traumatic brain injury may reduce morbidity and mortality associated with repeated inflicted traumatic brain injuries. We undertook this study to estimate the cost-effectiveness of a policy of head computed tomography (CT) for inflicted traumatic brain injury in selected infants seen in an emergency department.
METHODS. We constructed Markov models to compare a policy of CT to no CT in an asymptomatic 5-week-old infant with either (1) unexplained scalp bruising or (2) a history of an apparent life-threatening event. Health states modeled were no inflicted traumatic brain injury, misdiagnosed inflicted traumatic brain injury, mild inflicted traumatic brain injury (diagnosed or undiagnosed), and severe and fatal inflicted traumatic brain injury. Infants with undiagnosed inflicted traumatic brain injury were at increased risk of repeat inflicted traumatic brain injury. We used available literature to estimate probabilities, costs, and outcomes. The models terminated at death or at 52 weeks of age. Outcomes considered were severe and fatal inflicted traumatic brain injury cases averted through early detection of mild inflicted traumatic brain injury.
DATA SOURCES. We conducted a literature review for estimates of inflicted traumatic brain injury incidence, outcome probabilities, and medical and societal costs. Wide ranges were set for sensitivity and Monte Carlo analyses.
RESULTS. From a medical payer perspective, head CT for inflicted traumatic brain injury in infants with unexplained scalp bruising saved money. Sensitivity analysis demonstrated costs less than $50000 per severe or fatal inflicted traumatic brain injury averted in scenarios in which initial inflicted traumatic brain injury prevalence was >3%. From a societal perspective, costs of child protection made head CT for inflicted traumatic brain injury more expensive.
CONCLUSIONS. From a medical payer perspective, our models demonstrate that CT for inflicted traumatic brain injury can be cost-effective and improve outcomes. The finding of higher societal cost reflects the substantial short-term costs of child protection. Our study supports a low medical threshold for CT screening and highlights the need for improved understanding of long-term costs and outcomes of child abuse.
In the United States, inflicted traumatic brain injury (iTBI) is the most common cause of trauma-related death in infants.1,2 Half of all survivors suffer long-term disability.3–7 Missed diagnosis at initial presentation is cited as a contributor to the high morbidity and mortality associated with iTBI.8 To avoid missed diagnosis of iTBI, child abuse pediatricians have recommended a low threshold for evaluating infants with selected clinical presentations using head computed tomography (CT).8,9
Understanding where to set this low threshold, however, is difficult.10 Many infants diagnosed with iTBI present without an acute trauma history, making mild iTBI difficult to distinguish from frequently occurring conditions, such as gastroesophageal reflux, acute gastroenteritis, or colic.3,7,11,12 The decision to evaluate an infant for possible iTBI balances the impact of missed iTBI against risks and costs of unnecessary imaging. Effects of this evaluation may also include subsequent medical, social, and legal interventions. We are not aware of any reports that have examined the cost-effectiveness of head CT in infants with selected clinical presentations for possible iTBI.
There are substantial barriers to clinical trials in radiologic imaging for possible iTBI. Computer modeling has been useful in settings where traditional experimental methodology is not possible. Markov models estimate costs and effects of a clinical decision using set probabilities of events unfolding as a consequence of that decision. Most likely probabilities define a “base-case” scenario; these probabilities can be tested over a determined range in sensitivity analyses.
We developed an exploratory Markov model of a clinical decision to obtain head CTs for possible iTBI in well-appearing infants presenting to the emergency department (ED) under various scenarios. Our objective was to evaluate the short-term cost-effectiveness of a policy of head CT for select infant cohorts seen in an ED. These results may inform future protocols for the evaluation of infants with possible iTBI and demonstrate the utility of computer modeling in exploring the complex medical and societal effects of decision-making in cases of suspected child abuse.
METHODS
Population
We examined a policy of CT for possible iTBI in 2 hypothetical populations of 5-week-old infants without a history of recent trauma and with normal neurologic examinations. Our first cohort was composed of infants presenting to an ED with a history of an apparent life-threatening event (ALTE). The second cohort was composed of infants with unexplained scalp bruising on ED examination. On the basis of a review of the literature, we assigned a prevalence of initial iTBI of 2.4% to infants presenting with ALTE and 24% to infants presenting with scalp bruising.13,14
Case Definition
Initial iTBI cases were defined by acute subdural hemorrhage detected on initial CT (Table 1). Subsequent iTBI was any repeat iTBI after initial iTBI. All iTBI events were classified as mild, severe, or fatal. Infants with mild iTBI had full neurologic recovery 12 months from iTBI; infants with severe iTBI had neurologic disability 12 months from iTBI. All initial iTBI in our asymptomatic cohorts was defined as mild. Subsequent iTBI could be mild, severe, or fatal.
iTBI Definitions
Outcomes
We evaluated costs per iTBI detected for all infants presenting with ALTE or scalp bruising. Clinical outcomes were dichotomized as (1) no or mild iTBI and (2) severe or fatal iTBI. Model results were reported as costs per severe or fatal iTBI events averted through CT diagnosis of initial iTBI.
Model Structure
We developed Markov models simulating patients from first ED presentation at 5 weeks through 52 weeks of age (Fig 1). Under a CT policy, each infant had imaging during ED evaluation; under a no-CT policy, infants were discharged without CT. Infants entered the model with no iTBI or initial iTBI on the basis of estimated iTBI prevalence in the clinical scenario being modeled. Four possible outcomes of initial CT were modeled. Infants could have no iTBI and normal CT (true-negative), no iTBI and abnormal CT (false-positive), iTBI and normal CT (false-negative) or iTBI and abnormal CT (true-positive). If no CT was obtained, infants could have no iTBI (true-negative) or undiagnosed iTBI (false-negative).
Model outline. a Baseline iTBI risk for children in protective placement (97%); elevated risk for children in nonprotective placement (3%).
We modeled consequences of the initial decision to CT or not CT with weekly cycles. Infants with undiagnosed iTBI had an increased risk of repeat iTBI. Infants with diagnosed iTBI entered the child protection system and were placed into protective or nonprotective care. Protective care resulted in a return to a baseline iTBI risk. Nonprotective care resulted in increased repeat iTBI risk. Repeat iTBI events could be mild, severe, or fatal. Severe and fatal iTBI events were always diagnosed appropriately; mild events could be diagnosed or undiagnosed. Infants who were undiagnosed remained at high risk of repeat iTBI. Infants with a false-positive CT received additional medical evaluation, but no child protection involvement. All models were developed with TreeAge Software (Pro Suite 2006 release 1.0 [www.treeage.com]).
Probabilities
We estimated the risk of repeat iTBI in infants with undiagnosed iTBI using studies citing radiologic evidence of previous intracranial injury in 30% (26%–45%) of children admitted with iTBI5,15–17 (Table 2). All repeat abuse was presumed to occur over the model year.18
Probabilities and Costs
In this model, protective placement after iTBI implied protection from future iTBI. Although maltreatment in foster care is higher than in the general population, injuries such as iTBI are rare.19–21 For our base case, we assumed that 3% of infants entered a nonprotective setting after diagnosis of iTBI, remaining at increased risk of repeat iTBI. Infants in protective placement had a risk of repeat iTBI comparable to the general population.
We estimated sensitivity and specificity of CT for initial iTBI by using published data; because of limited pediatric-specific estimates, wide margins were set for sensitivity analyses.16,22,23 Probability of correct diagnosis of repeat iTBI was set at 69% per iTBI event on the basis of the observation that 31% of children with iTBI are missed at the time of medical evaluation.8 Recognizing limitations of this single-study estimate, this probability was ranged widely for sensitivity analysis.
Several studies describing iTBI outcomes are hampered by exclusion of mild injuries,5,24 thereby overestimating case fatality rates. Institution-based case series exclude deaths out-of-hospital, underestimating case fatality rates.25–27 Studies using national health databases avoid these confounders. A review of all subdural hemorrhages in children in Wales identified 4 deaths among 27 cases of iTBI7; a review of Scottish children found 1 death of 19 cases of iTBI.28 A review of British and Irish cases found 18 fatalities of 106.29 Using these data, base-case estimate of fatality with repeat iTBI was set at 15% (5.3%–17%).
Estimates of neurologic disability after iTBI are limited by case selection, follow-up time, and definitions of disability. In the Welsh study, 15 (63%) of 24 iTBI survivors were described as profoundly disabled 1 year postinjury. This estimate is higher than findings of moderate to severe disability at 1 year among 19 (46%) of 41 infants admitted to North Carolina PICUs with iTBI.6 Because this estimate of disability correlated with health care utilization, it was set as our base-case estimate with a wide range for sensitivity analysis (29%–63%).
Foster care is more expensive than in-home support services, thus probability of out-of-home placement influenced societal costs. We estimated that 63% (48%–68%) of children with diagnosed iTBI would be placed out-of-home.3,12,30,31
Costs
All costs were adjusted to 2005 dollars by using the all-items consumer price index.32
Medical Costs
Costs for initial evaluations were taken from the Medicare Physician Fee Schedule (2005), using an average of global diagnostic modifiers for each CPT code.33 All infants had an ED visit (99284–99285); infants undergoing CT had an unsedated, noncontrast head CT (70450). Infants with misdiagnosed iTBI underwent skeletal survey (76062) at presentation, a follow-up skeletal survey, and outpatient consultation by a medical child protection team (99245).
Hospital charge data for ICU and non-ICU iTBI admissions to the Children's Hospital of Pittsburgh from 1995 to 1999 were used for acute medical charges associated with iTBI.34 Quartile data from this previously published analysis were used for our estimates. We assumed that acute medical charges would be proportional to the severity of brain injury being treated. Medical charges for initial iTBI in our asymptomatic 5-week-old infants were set as the first-quartile charges of all iTBI admissions. Medical charges for mild iTBI and severe iTBI were set as the second- and third-quartile charges, respectively. Medical charges associated with fatal iTBI were assumed to be equal to mild iTBI, but ranged from the minimum to the maximum of all medical charges for iTBI. All charges were adjusted by using a generic cost-to-charge ratio obtained by using economic data from children's hospitals on 1150 infants discharged in 2003 with a Clinical Classifications Software diagnosis of intracranial injury (0.45).35 Recognizing the limitations of our assumptions, acute costs were ranged widely for sensitivity analyses.
Chronic medical costs for survivors of iTBI were estimated by using Medicaid expenditures for children with multiple medical needs. Infants with no or mild iTBI (no disability) accumulated weekly medical costs comparable to all children ($8/week). Survivors with neurologic disability accumulated higher costs, equivalent to median weekly medical costs of children with cerebral palsy ($92/week).36
Societal Costs
A second model included additional costs to examine the societal impact of a medical decision to obtain CT for possible iTBI.37 At iTBI diagnosis (new or repeat), acute costs of child protective services, police investigation, and legal action ($2000 per event) were accumulated.37 Available sources do not stratify costs by abuse type and likely underestimate costs associated with more severe abuse such as iTBI.
Chronic societal costs after diagnosis of iTBI involve weekly costs of inhome or out-of-home placements for iTBI survivors. Inhome services were estimated to be $70/week by Courtney.38 Out-of-home placement costs were based on pooled data from states participating in a study on child welfare services ($470/week).39
Time Horizon
Markov models typically run until all members of the original cohort have died. We elected to terminate this exploratory model at death or at 52 weeks of age to limit unfounded assumptions about lifetime costs and effects related to radiation exposure, sequelae of iTBI, foster care placement, involvement in child protective services, or recurrent maltreatment.
Sensitivity Analysis
Model building required assumptions that oversimplify realities of clinical practice. Sensitivity analysis allowed examination of alternate scenarios not represented by base-case assumptions. One-way sensitivity analyses were performed with each variable over wide margins to test model logic. The 1- and 2-way sensitivity analyses were repeated by using literature-derived intervals for each variable, identifying those that substantially altered model outcomes.
Monte Carlo sensitivity analyses were run to simulate the “real world” variation of costs and probabilities between individual patients. In this analysis, each parameter was simultaneously varied using set probability distributions as the model ran through multiple iterations. Monte Carlo results were expressed as the proportion of patients with cost-effectiveness ratios falling below a cost-per-case-averted cut point. We ran Monte Carlo analyses for each clinical scenario and payer perspective in >10000 patients to describe the spectrum of cost-effectiveness ratios observed under the CT versus no-CT policies.
RESULTS
Base-Case Analysis
From a medical payer perspective, CT saved money in the bruising scenario ($3880 saved for every severe and fatal iTBI averted) (Table 3). CT cost $72744 per case averted in the ALTE scenario. When we included societal costs in the model, CT was $132701 and $209328 per case averted for bruising and ALTE scenarios, respectively.
Base-Case Results
Sensitivity Analysis: Medical Payer Perspective
We conducted sensitivity analyses to examine the impact of assumptions in our base-case scenario. In the bruising scenario, the finding of cost savings with a CT policy as robust to all assumptions except acute medical costs of iTBI, cost of CT, and probability of repeat iTBI. Obtaining a CT saved money if imaging cost less than $368, and the probability of repeat iTBI was >34%. Similarly, CT saved money if medical costs of acute initial iTBI were <35% of the costs associated with mild iTBI. In the ALTE scenario, cost-effectiveness of CT was sensitive to many assumptions, particularly those associated with acute medical costs of iTBI (Fig 2). In this model, CT costs influenced cost-effectiveness, as did the rate of repeat iTBI events and probability of repeat iTBI recognition.
Tornado diagram for cost-effectiveness of head CT in infants with history of ALTE. Bars represent the range of cost-effectiveness ratios observed when the model is run over the values of the variable of analysis. All variables that changed cost-effectiveness ratios by more than $10000 per severe or fatal iTBI averted are shown.
Variation of initial iTBI prevalence demonstrated that CT saved money when initial iTBI was above 16% (Fig 3). When iTBI prevalence fell below 3.3% and 1.8%, the incremental cost-effectiveness rose above $50000 and $100000 per case averted, respectively. When acute medical costs of initial iTBI rose to 50% of medical costs of mild iTBI, CT was no longer cost-saving. Cost-effectiveness ratios fell below $50000 and $100000 per case averted at iTBI prevalence levels of 4.3% and 2.1%, respectively. When acute medical costs of initial and mild iTBI were equal, costs fell below $100000 per case averted when initial iTBI prevalence rose above 3.5%.
Twelve-month cost-effectiveness of a policy of head CT according to prevalence and acute costs of initial iTBI. Assumptions regarding prevalence and acute medical costs of initial iTBI were explored with this 2-way sensitivity analysis. In our base case (solid line), we set the costs of acute medical care of initial iTBI as described in the text. For the 50% and 100% cases, acute costs of initial iTBI were set at 50% and 100% of the costs of subsequent mild iTBI, respectively. Negative incremental cost-effectiveness ratios occur when a policy of head CT saves money and prevents severe or fatal iTBI, as seen when iTBI prevalence exceeds 16% under our base-case assumptions.
Sensitivity Analysis: Societal Perspective
From a societal perspective, costs of foster care had the strongest influence on societal estimates of cost-effectiveness. In Missouri, with reported costs of $213 per week for foster care services, the cost-effectiveness ratio for CT for the bruising scenario was $73314 per case averted. In Delaware, with costs of $1402 per week for foster care, the cost-effectiveness ratio for the same scenario was $345945 per case averted.39
Monte Carlo Simulation
From a medical payer perspective, Monte Carlo simulation indicated that CT might be more efficient than reflected in our base-case analysis (Fig 4). For bruising, CT was cost-saving for 96% of 10000 patients, and cost-effectiveness ratios were under $50000 per case averted in all patients. For the ALTE scenario, CT saved money in 42% of patients modeled; cost-effectiveness ratios fell $100000 per case averted 98% of the time. When societal costs were included, 89% and 60% of all runs fell below $200000 per case averted in bruising and ALTE scenarios, respectively.
Monte Carlo sensitivity analysis. Each scenario and perspective was run with 10000 patients, each drawing cost and probability values from the ranges and distributions described in Table 2. Results are expressed as the percentage of runs (y-axis) with incremental cost-effectiveness ratios falling below a set threshold (x-axis).
DISCUSSION
Despite the common perception that child abuse is both medically and socially expensive, there has been little economic analysis in the field of child abuse.40 To our knowledge, this is the first study to consider the cost-effectiveness of medical decision-making related to diagnosis of child abuse. Our results indicate that, from a medical payer perspective, CT diagnosis of iTBI in asymptomatic infants is cost-saving or cost-effective among infants by preventing more severe subsequent iTBI. From a societal perspective, early diagnosis of iTBI is more expensive in this short-term analysis. These results reflect the high costs of child protection, cited by others both as evidence of the need for improved efficiency and accountability in a publicly funded welfare program,38,41,42 and as justification for increased spending in child abuse prevention programs.43–46 As a secondary prevention measure, CT screening prevents severe and fatal iTBI only by increasing the number of children in the child protection system over the model year. These results support the observation that efforts to prevent abuse, rather than improve detection, may prove more cost-effective from a societal perspective.
The higher societal costs also reflect our choice of outcome measures and time horizon. In cost-effectiveness analysis, a societal perspective measures the broad effects of health policy for a population.47 Traditionally, this perspective has improved cost-effectiveness of medical interventions. Most medical interventions generate primarily medical costs; societal benefits associated with a healthier population often outweigh added societal costs. Given the short time horizon of our model, we cannot account for potential societal benefits of early iTBI detection over subsequent years, which might include lowered educational costs and improved productivity.44 Although such benefits are anecdotally appealing, current understanding of long-term outcomes for survivors of mild iTBI and for children in the child welfare system is not sufficient for modeling purposes.
Our choice of outcome measures reflects a limited understanding of expected outcomes among infants in our patient cohorts. For health policy purposes, cost-effectiveness analysis is most useful when outcomes measures are comparable across populations and interventions, most commonly expressed as a quality-adjusted life-year.47 Previous work in quality of life among survivors of illness in infancy translates poorly to our model cohorts.48 Considerations of proxy response, caretaker transition, and recurrent maltreatment are not addressed in quality-of-life estimates for survivors of prematurity, central nervous system neoplasm, or meningitis.49–53 Moreover, such measures are derived for school-aged survivors and cannot account for developmental differences in quality-of-life domains among infants. Our short time horizon precludes consideration of long-term effects of radiation in infancy. Although radiation risk data are limited, 1 model estimated ∼1 radiation-related cancer fatality for every 1000 head CTs performed in infancy.54 Under these conditions, a CT policy for our patients results in 1 excess cancer death for every 24 (ALTE scenario) or 240 (bruising scenario) iTBI cases detected. Because of this uncertainty in expected outcomes, we chose to measure outcomes in more concrete (severe and fatal cases averted) but less broadly comparable (quality-adjusted life-year) terms.
Poor understanding of long-term outcomes, as well as ethical concerns, limits our willingness to suggest a cost threshold for interventions that prevent severe or fatal child abuse injury. Nonetheless, we believe that our cost-effectiveness estimates fall within acceptable cost boundaries of lifesaving interventions. Universal newborn sickle-cell screening costs $30000 per life-year saved; child car seats cost $73000 per life-year saved.55,56 Economic researchers suggest that an adult life is valued between 1.5 and 3 million dollars.57,58 Presuming a normal lifespan of 75 years after diagnosis of asymptomatic iTBI (with 3% discounting), a policy of head CT for a history of ALTE costs less than $7000 per healthy life-year saved from a societal perspective.
Any conclusions from our study must be viewed in light of assumptions made for this analysis. Our model simplifies the complexity of a medical workup for possible iTBI, but we believe that these simplifications do not substantially impact the outcomes of the analysis. The assumption that misdiagnosis of iTBI results in additional medical workup but not child protection intervention will be argued by physicians with memories of cases in which misdiagnosis led to complex interactions with child protection agencies.59 We do not dispute such anecdotes, but cannot estimate the frequency with which they occur. Such mistaken placements would increase societal costs associated with a CT policy. Although quality of life is impacted by such events, it does not affect the probability of severe or fatal iTBI.
Multiple approaches to evaluation of possible iTBI have been recommended, and it is possible to examine incremental cost-effectiveness ratios comparing CT with other methods. Currently, however, CT is both the test of choice and the gold standard for suspected iTBI in infants in the United States. CT is better than MRI for acute evaluation of intracranial hemorrhages,60 and MRI is rarely available for routine ED use. Skull fractures detected on plain radiograph are neither sensitive nor specific for iTBI. Ophthalmologic examination for retinal hemorrhage requires specialized training; moreover, a normal retinal examination does not exclude the presence of iTBI.11,24,26 Cases of iTBI with initially normal CT may occur, but are not represented in our model.61
Our study is limited by our estimates of acute medical costs. The assumption that acute medical costs are proportional to iTBI severity is rational but without clear evidence. Although several studies have examined medical charges in iTBI, we are unaware of any that have stratified charges by severity or clinical presentation. We addressed these limitations by setting wide and overlapping ranges for each charge type and by conducting 2-way sensitivity analyses with these variables. We relied on generic cost-to-charge ratios where specific cost data were unavailable, using characteristics most likely to affect cost-to-charge ratios (diagnosis, age, and hospital type) to improve the specificity of our estimates but cannot be certain that these reflect the true cost-to-charge ratio in our institution. Future economic studies should include medical cost data.
By definition, outcomes associated with undiagnosed iTBI are unknown. Assumptions that children with mild iTBI have full recovery may be optimistic, yet we assume that such children are unlikely to have significant excess medical costs in the first year of life. Estimates of the rate of reabuse in children with “missed” iTBI are based on observations that old intracranial injuries are present in one third of infants at the time of diagnosis with iTBI.5,15–17 This may underestimate repeat abuse, because small intracranial injuries can resolve or be obscured by subsequent injury; it may overestimate repeat abuse, because infants may experience single unrecognized episodes of iTBI without subsequent injury. We conducted sensitivity analyses over wide ranges for this variable to account for this uncertainty.
We set initial probabilities of iTBI using ALTE and scalp bruising cohorts based on published data that may not reflect experience elsewhere. We selected these scenarios because they are common presentations in pediatric EDs that create management uncertainty. We believe that physicians suspecting a 25% probability of iTBI obtain CTs for these patients; a decision not to image infants with scalp bruising likely reflects doubts regarding this prevalence estimate rather than concerns over cost-effectiveness. Using sensitivity analysis across a range of iTBI probabilities, however, our study demonstrates that CT diagnosis of asymptomatic iTBI is cost-saving from a medical perspective when iTBI prevalence is >16%. Short-term cost-effectiveness ratios of $50000 to $100000 per severe or fatal iTBI averted are observed at iTBI prevalence of 3.3% and 1.8%, respectively.
CONCLUSIONS
This study indicates that, for medical payers, CT screening is cost-saving in infants with unexplained scalp bruising and is cost-effective for infants with a history of ALTE. On the other hand, any iTBI carries high societal costs. Recognizing ethical concerns with conclusions suggesting that child abuse may be too costly to diagnose, we cannot identify a clear threshold above which iTBI diagnosis is “too expensive.” Future studies should examine long-term costs and effects of iTBI; policymakers should consider costs of iTBI in the funding of projects that may prevent iTBI. Our results may provide a framework for medical protocols on the evaluation of select infants in the ED and act as a springboard for multidisciplinary teams evaluating current practices and outcomes from the perspective of quality and efficiency.
Acknowledgments
Dr Campbell conducted this study during her clinical research fellowship funded by grant D55HP05156-01-00 from the Health Resources and Services Administration (Division of General Academic Pediatrics, Children's Hospital of Pittsburgh).
We thank Kenneth Smith, MD, for valued suggestions in preparation of this manuscript and the Division of General Academic Pediatrics at the Children's Hospital of Pittsburgh for support provided throughout the project.
Footnotes
- Accepted March 23, 2007.
- Address correspondence to Kristine A. Campbell, MD, MSc, Division of Safe and Healthy Families, Department of Pediatrics, University of Utah, 295 Chipeta Way, Box 581289, Salt Lake City, UT 84158. E-mail: kristine.campbell{at}hsc.utah.edu
The authors have indicated they have no financial relationships relevant to this article to disclose.
REFERENCES
- Copyright © 2007 by the American Academy of Pediatrics
In this issue
