Simulation of the Impact of Programs for Prevention and Screening of Pediatric Abusive Head Trauma

Abstract

Primary prevention programs of pediatric abusive head trauma (PAHT) exist and early screening is proposed, but negative effects of mislabeling parents as abusers, an important issue, are not well documented. The aim of our study was to simulate the possible impact of programs for the primary prevention and screening of PAHT. We developed Markov models that simulate the life histories of PAHT with no intervention, with primary prevention program only, with screening program, and with both programs in a hypothetical cohort of 800,000 newborns in a high-income country. Screening program would be addressed to all families until children are 2 years old, during repeated consultations. Potential side effects for parents being mislabeled as abusers were supposed to increase the probability of PAHT and decrease participation in screening. Time horizon was 2 years with cycles of 15 days. Outcomes were number of deaths and abused children avoided. Uncertainty was specified with probability distributions. After 2 years, the median number of deaths avoided through primary prevention would vary from 6 (95% confidence interval [CI] 2–11) to 28 (95% CI 6–51) per 100,000 newborns. Screening could prevent up to 6 (95% CI 0–29) or cause up to 66 (95% CI 0–361) deaths per 100,000 children born alive. The impact of both programs was uncertain. Our model confirmed the potential benefits of primary prevention and documented the uncertainty associated with screening of PAHT.

Introduction

Pediatric Abusive Head Trauma (PAHT) results in significant morbidity in infants and toddlers and is one of the main causes of child abuse mortality.^1
–3 In the U.S. National Child Death Review Case Reporting System, PAHT accounted for 60% of physical-abuse deaths, and 30% of all child-abuse deaths from 2005 to 2009.⁴ Several programs to prevent abusive PAHT, in particular shaken baby syndrome, have been developed.^5

–8 Danger of violent infant shaking was explained to parents of newborns before the infant's discharge from the hospital and alternative responses to persistent infant crying were suggested.^5,9 Dias and associates⁵ and Altman and colleagues⁹ showed a decrease of PAHT incidence during the months following program introduction. A favorable cost-utility ratio of a prevention program based on the program of Dias and associates was estimated from a 5-year cohort of infants with PAHT admitted to a hospital in Auckland, New Zealand.¹⁰ However, sensitivity analyses did not test uncertainties on incidence, mortality, and program coverage. The estimation of incidence is challenging because of inconsistencies in the definition of PAHT, and variation in how definitions are applied.¹¹ The American Academy of Pediatrics proposed in 2012 a definition of abusive head trauma.¹¹

Several studies suggest that children could be victims of repeated episodes of PAHT before it is detected,¹² or that physicians miss cases of PAHT.¹³ For these reasons, several authors advocate systematic screening of PAHT, for example among infants presenting to the emergency department.^14,15 The understanding of how PAHT occurs has also improved. The role of infant crying has been recognized as the most common stimulus for PAHT.^16,17 Crying, especially when parents do not understand its meaning, can provoke negative, destructive, survival-endangering responses resulting in PAHT. PAHT could result from a failure of normal iterative interactions between infants and their parents.¹⁷ Theoretically, this abnormal interaction could be the target of screening, before any head trauma.

The impact of possible prevention or screening programs is difficult to estimate, in particular because of the lack of knowledge about the beginning of PAHT, and difficulties in measuring all cases of PAHT. Large and expensive cohorts would be necessary in view of PAHT incidence. Negative effects of mislabeling parents as abusers are yet not well documented, but several authors have raised the issue of the potential impact of such stigmatization.¹⁸ Consequently, a clinical study considering early screening of PAHT before children have serious consequences or when children have only abnormal interaction with their parents is not ethically acceptable. To overcome this difficulty, computer modeling has many advantages. The impact of strategies as a systematic early screening could be studied, whereas in reality such strategies could cause ethical problems. Cost of computer stimulation is very low and no expensive and time-consuming studies are necessary. Alternative strategies can be easily compared and; even if better knowledge about PAHT remains to be acquired, the contributions of each parameter or research hypothesis to uncertainty about the potential impact of programs can be estimated.¹⁹

We developed a Markov modeling of the evolution of PAHT, for four alternative strategies: no intervention, primary prevention program, screening program, and both primary prevention and screening programs. Our objective was to evaluate the possible impact, on a cohort of newborns, of primary prevention and screening programs targeted at PAHT.

Methods

Decision model

Our Markov model compared the evolution of PAHT with: no intervention, primary prevention alone, screening alone, and both primary prevention and screening. The simulated population was a cohort of 800,000 newborns, corresponding approximately to the annual number of newborns in France.²⁰ The model of natural evolution had five possible and mutually exclusive states (Fig. 1):

1. Normal interaction: Children are healthy with normal interaction between them and their parents.

2. Subclinical: Children have no clinical symptoms, but parents have negative interactions with their children.

3. Reversible clinical: Children have reversible, clinical, temporary consequences of PAHT, consecutive to a first shaking or another mechanism of inflicted trauma; these consequences can result in the child faint with spontaneous recovery.

4. Irreversible clinical: Children have irreversible consequences of PAHT without spontaneous recovery, consecutive to repetitive shaking or when first shaking was extremely violent, such as motor deficit with paralysis and spasticity, epilepsy.

5. Death: At each state child could die from PAHT or other causes.

FIG. 1.

Markov model of the impact of prevention and screening programs of pediatric abusive head trauma. Normal interaction state: child healthy with normal interaction with his/her parent(s); subclinical state: asymptomatic child with abnormal interaction; reversible/irreversible clinical state: child with reversible/irreversible consequences. Arrows: transitions between states; thick arrows: transitions increased by programs. NSS, no when subclinical state is considered; P, targeted by prevention; S, targeted by screening.

Because the existence of a subclinical state is not formally defined, we considered two evolutions of PAHT: a model with a subclinical state and a model without a subclinical state. Each Markov cycle was 15-days long.

Interventions

A primary prevention program aims to prevent PAHT before it ever occurs, that is, before the possible subclinical state, when parent and child have negative interactions, or if no subclinical state exists, before a reversible clinical state. The modeled primary prevention program would be offered to all parents with their consent. The proposed intervention would be based on current PAHT education interventions, such as the Dias or Altman program or in the first stage of the PURPLE program.^5,8,9 These programs were delivered during the postpartum hospital stay, immediately following delivery. For example, the PURPLE program used an educational package that includes a 10-min DVD and an 11-page booklet to explain to the parents that infant crying is normal, what are the appropriate responses to inconsolable crying, and the danger of shaking a baby. No side effects were considered.²¹

Screening aims at identifying PAHT when children have no symptoms or signs and are apparently in good health, which corresponds to the potential subclinical state and the reversible clinical state when the child has spontaneously recovered a normal state. A screening program would include all actions to identify early children who are victims of PAHT, and improve their health by avoiding severe consequences. It would include the initial screening test, a confirmation diagnostic strategy, and an intervention. Several authors have explored screening tests such as serum biomarkers,²² but no formal screening program currently exists. A hypothetical screening program was defined to be directed to all parents and children until the latter are 2 years old and had no irreversible consequences; it would be provided by family physicians or pediatricians during repeated follow-up consultations, monthly during the first 6 months of the child's life, and at 9, 12, 16, 20, and 24 months. If detection were positive, an education intervention would be proposed to stop PAHT evolution. Because the potential harm associated with mislabeling parents as child abusers has been emphasized,¹⁸ we modeled side effects for parents wrongly detected as abusers. Indeed, some parents could be identified as parents who have negative interactions with their children (subclinical state, reversible clinical state), whereas they have in fact normal interactions with their children (normal interaction state). These parents would be false-positive. The effects of this stigmatization might lead to an increase of parent's stress and of their anxiety.¹⁸ We also assumed that these parents could experience losing confidence in themselves, poor self-esteem, depression or paranoia, and social isolation. As these effects have been identified as risk factors of child abuse,^23,24 this stigmatization may place children at greater risk to become abused than before the screening. Thus, we assumed that potential side effects of screening could be to increase the probability of PAHT and to decrease future participation in the screening program.

Model parameters and assumptions

Whenever possible, transition probabilities were extracted from the literature (Table 1). Transition probability between normal and clinical states of PAHT varied over time. It was more important when the child was less than one year of age.^25,26 When considering a subclinical state, we chose two extreme scenarios: most in favor and most in disfavor of screening. The scenario most in favor of screening assumes that all parent/child pairs are in a subclinical state after the first Markov cycle, and remain in this state until children have clinical symptoms. In this scenario, no parents could be false-positive. Indeed, after screening intervention, some parents are in a normal interaction state but in the following cycle all these parents return to a subclinical state. The impact of simultaneous primary prevention and screening programs could not be estimated in this case. The scenario is most in disfavor of screening when transition probability to remain in a subclinical state is nil. Parents and children are in a subclinical state during only the 15 days before clinical manifestations of PAHT.

Table 1.

Parameters for Estimating the Impact of Primary Prevention and Screening of Pediatric Abusive Head Trauma

Parameters	Plausible values	Sources and references
Inc = transition probability between normal state and clinical states of PAHT	Three scenario with β distribution of annual incidence:
	<1 year old (α = 180; β = 1383041); >1 year old (α = 34.6; β = 1440530)	Literature²⁵
	<1 year old (α = 1071; β = 3205592.6); >1 year old (α = 8.8; β = 227434.9)	Literature^3,26
	<1 year old (α = 37;β = 66085); >1 year old (α = 8; β = 64573)	Literature²⁴
	Inc = 1-EXP(LN(1-annual incidence)^*(1/24))
p1: transition between normal and subclinical state	p1 = (1-pmortaBL) in the scenario most in favor of screening	Assumptions according to scenario in favor or disfavor of screening
	p1 = Inc in the scenario most in disfavor of screening
p2: transition between subclinical and reversible clinical state when subclinical state is considered, otherwise transition between normal and reversible state	p2 = x1^* X with x1 uniform distribution from 0 to 1	x1 randomly selected between extreme possible values
p4: transition between subclinical and irreversible clinical state when subclinical state is considered; otherwise transition between normal and irreversible state	p4 = X-p2	Computed from other parameters
pmorta1AHT: mortality rate of PAHT in subclinical state	pmorta1AHT = x2^(p2+p4) with x2 uniform distribution from 0 to 0.428: pmorta1AHT/(pmorta1AHT* + p2 + p4) = 0 to 0.3	Literature^{3, 27 –29}
X = p2+p4	X = Inc/(1+x2) when subclinical state has not been considered and in the scenario most in favor of screening	According to scenario
	X = (1-pmortaBL)/(1+x2) in the scenario most in disfavor of screening
pmorta2AHT: mortality rate of PAHT in reversible state	Uniform distribution from 0 to 0.3	Literature^3,27 –29
pmorta3AHT: mortality rate of PAHT in irreversible state	Uniform distribution from 0 to 0.3	Literature^3,27 –29
p3: transition between reversible clinical state and irreversible clinical state	Uniform distribution from 0 to (1-pmortaBL-pmorta2AHT)	Randomly selected between extreme possible values
adhP1: adherence to primary prevention	Uniform distribution from 0.6 to 1	Literature^5,9,30
eP1: effectiveness of primary prevention	Uniform distribution from 0.4 to 0.8	Literature^5,9
spec: specificity of screening strategy	Uniform distribution from 0.5 to 1	Assumption
sens1: sensitivity of screening strategy for subclinical state	Uniform distribution from 0.5 to 1	Assumption
sens2: sensitivity of screening strategy for reversible clinical state	Uniform distribution from sens1 to 1	Assumption
ad: adherence to screening when caregivers were never false-positive and when primary prevention is not proposed	Uniform distribution from 0 to 1	Randomly selected between extreme possible values
adFplus: adherence to screening when caregivers were at least once false-positive and when primary prevention is not proposed	Uniform distribution from 0 to ad	Assumption
eD: effectiveness of screening	Uniform distribution from 0 to 1	Randomly selected between extreme possible values
eSE: side effects of being false-positive during screening on probability of becoming abuser, likened to relative risk	Uniform distribution from 1 to (10 or maximum possible as sum of probabilities from normal state is 1)	Assumption
adP0: adherence to screening when caregivers were never false-positive and when they have not adhered to primary prevention	Uniform distribution from 0 to 1	Randomly selected between extreme possible values
adP0Fplus: adherence to screening when caregivers were at least once false-positive and when they have not adhered to primary prevention	Uniform distribution from 0 to adP0	Assumption
adP1: adherence to screening when caregivers were never false-positive and when they have adhered to primary prevention	Uniform distribution from adP0 to 1	Assumption
adP1Fplus: adherence to screening when caregivers were at least once false-positive and when they have adhered to primary prevention	Uniform distribution from 0 to adP1	Assumption

PAHT, pediatric abusive head trauma; pmortaBL, probability to die from causes other than PAHT.

For the two above-mentioned extreme situations and the model without a subclinical state, we have modeled three scenarios based on maximum, minimum, and median estimations of the incidence of PAHT in high-income countries, varying according to definitions used and populations.^3,25
–27 Shanahan and colleagues and Parrish and associates used the definition recommended by the Centers for Disease Control and Prevention in 2012.^11,25,27 Fujiwara and co-workers used the International Classification of Diseases, 10th edition codes.²⁶ Because Shanahan and colleagues estimated incidence only for children younger than one year old, we used the incidence estimation of Keenan and associates for children aged one and two years old.^3,27 Keenan and associates focused only on serious or fatal PAHT.³ The case-fatality of PAHT was estimated at around 20%,^3,28,29 sometimes less (6.7%), as in a study in the United States.³⁰ Thus, we had this parameter vary from 0 to 0.3. When neither data nor estimations were available, we assumed variation between extreme possible values. These values were zero, and the sum of transition probabilities from one state was 1. We assumed a probability of participation for a primary prevention program to be between 0.6 and 1^5,9,31 and effectiveness between 0.4 and 0.8.^5,9

Outcomes

Outcomes were measured after 2 years. Numbers of children in each state for 100,000 children born, the number of avoided or added deaths, and the difference of number of children in clinical states were reported.

Data analysis

We used Microsoft Office 2013's Excel (Microsoft, Redmond, WA) to run the model. Children moved from state to state during 49 cycles. For each scenario considering or not a subclinical state, we conducted probabilistic sensitivity analyses with three scenarios for transition probability between normal state and clinical states of PAHT. Beta distributions were used for transition probability between normal state and clinical states of PAHT and, when lack of knowledge did not allow estimating distributions, uniform distributions between extreme values. For the beta distributions, alpha and beta estimations were obtained by the method of moments. Random numbers bounded by these distribution assumptions were generated to calculate outputs for each cycle. We used 1000 simulations, where the analysis modeled uncertainty for all parameters simultaneously.

Results

Potential impact of primary prevention

Primary prevention programs would reduce mortality and morbidity (Table 2). Depending on the level of incidence, the application of primary prevention alone would avoid 6 (95% confidence interval [CI] 2–11) to 28 (95%CI 6–51) deaths per 100,000 children. The number of irreversible consequences of PAHT avoided would vary from 0 (95%CI 0–5) to 1 (95%CI 0–26) per 100,000 children. The number of children who would remain in a normal or subclinical state as a consequence of prevention would vary from 7 (95% CI 4–11) to 32 (95% CI 18–54) per 100,000 children.

Table 2.

Estimations of the Impact of Programs on Mortality and Morbidity by Pediatric Abusive Head Trauma (per 100,000 Children Born Alive, 95% Confidence Interval)

	Scenario without subclinical state			Scenario with a subclinical state lasting only 15 days			Scenario with the longest subclinical state
Outcome after 2 years	Low incidence hypothesis	Medium incidence hypothesis	High incidence hypothesis	Low incidence hypothesis	Medium incidence hypothesis	High incidence hypothesis	Low incidence hypothesis	Medium incidence hypothesis	High incidence hypothesis
Deaths of any cause
Without intervention	356 (345–538)	377 (355–381)	403 (355–429)	355 (345–358)	376 (350–380)	404 (356–426)	355 (344–357)	375 (356–379)	401(350–424)
Change with prevention	−7 (−11 – −2)	−16 (−26 – −6)	−28 (−51 – −6)	−6 (−11 – −2)	−16 (−26 – −4)	−28 (−50 – −7)	NE	NE	NE
Change with screening	14 (0–80)	35 (0–189)	66 (0–361)	12 (−1 – 77)	30 (−2 – 177)	59 (−3 – 352)	−1 (−7 – 0)	−3 (−16 – 0)	−6 (−29 – 0)
Change with prevention and screening	19 (−5–77)	50 (−14 – 191)	91 (−21 – 355)	16 (−6 – 76)	42 (−14 – 181)	85 (−21 – 375)	NE	NE	NE
Children with irreversible consequences of PAHT
Without intervention	1 (0–12)	1 (0–23)	3 (1–55)	1 (0–12)	1 (0–28)	2 (1–53)	1 (0–12)	1 (0–21)	2 (1–55)
Change with prevention	0 (−6 – 0)	0 (−11 – 0)	−1 (−26 – 0)	0 (−5 – 0)	0 (−13 – 0)	−1 (−24 – 0)	NE	NE	NE
Change with screening	1 (0–30)	2 (0–56)	5 (0–152)	1 (0–23)	2 (0–52)	4 (0–95)	0 (−2–0)	0 (−4 – 0)	0 (−11 – 0)
Change with prevention and screening	1 (0–31)	2 (0–60)	6 (−1 – 137)	1 (0–28)	2 (−2 – 55)	6 (−2 – 115)	NE	NE	NE
Children in normal state and potential subclinical state
Without intervention	99643	99622	99591	99644	99622	99591	99644	99623	99593
Change with prevention	7 (4–11)	17 (11–27)	32 (18–54)	7 (4–11)	17 (11–27)	31 (18–54)	NE	NE	NE
Change with screening	−17 (−85 – 0)	−43 (−208 – 0)	−79 (−404 – −1)	−15 (−81 – 1)	−40 (−198 – 3)	−75 (−379 – 3)	2 (0–7)	4 (0–17)	7 (0–32)
Change with prevention and screening	−24 (−84 – 6)	−59 (−207 – 15)	−115 (−397 – 22)	−21 (−81 – 7)	−53 (−192 – 15)	−104 (−415 – 25)	NE	NE	NE

NE, not estimable.

Potential impact of screening

The impact of screening was uncertain (Table 2). If being wrongly identified as an abuser would decrease participation in screening programs and increase the probability to become an abuser, screening could cause more cases of PAHT. Screening could prevent, after 2 years of evolution, up to 6 (95%CI 0–29) deaths per 100,000 children, or cause up to 66 (95%CI 0–361) deaths per 100,000 children. The number of children with irreversible consequences could be reduced to 0 (95%CI 0–11) per 100,000 children or increased to 5 (95%CI 0–152) per 100,000 children (Table 2).

Potential impact of associating primary prevention and screening

Simultaneous application of primary prevention and screening would not reduce mortality and morbidity with certainty (Table 2). In scenarios without any subclinical state, application of both primary prevention and screening could cause from 19 (95% CI −5 to 77) to 91 (95% CI 21–355) deaths per 100,000 children. When considering a subclinical state lasting only 15 days, from 16 (95% CI −6 to 76) to 85 (95% CI –21 to 375) deaths per 100,000 children could be caused. With a subclinical state lasting more than 15 days, screening would be more effective, because more parents would be potentially detected early, and fewer would be wrongly identified as abusers. Primary prevention would decrease the incidence of PAHT and increase the number of parents wrongly identified as abusers. The latter would less participate in subsequent rounds of screening and would be more likely to become abusers.

Discussion

We examined the potential impact of primary prevention and screening programs on PAHT in a hypothetical cohort from a high-income country, taking into account major uncertainties. We showed the potential benefit of a primary prevention program and the uncertainties concerning screening. We hypothesized that false identification of parents as abusers could either have no impact on the evolution of PAHT or increase the probability of becoming abusers. Under the latter hypothesis, a screening program could potentially increase mortality and morbidity. Very few studies have considered the effects of being wrongly identified as an abuser. When parents were victims of separation from their children because of false allegations of child abuse, they described the loss of the sense of being competent parents and a modification of their attachment to their child.³² These feelings would lead some parents to relations with a disorganized attachment with their child. Although the link between child maltreatment and disorganized attachment is unclear, maltreatment could be the cause of disorganized attachment and vice-versa.^33,34 Research on the potential effects of being wrongly identified as an abuser is very important to the consideration of screening programs.

We hypothesized in two models that a true subclinical state preceded the clinical state in which parents are acting out and become physically violent with their child. During this potential subclinical state, parents would go through stages of emotions such as accumulation of frustration or anger, especially because of a child's persistent crying.^7,17 Because of their incapacity to control and overcome these emotions, some parents could become violent. Parents could learn how to recognize and cope with these emotions through different stress techniques. For instance, the Perinatal Shaken Baby Syndrome Prevention Program uses coping strategies with an “anger thermometer.” Parents are reassured about the experience of anger that crying can cause and they identify strategies for dealing with anger.⁷ Screening of child-parent interaction difficulties should not result in a negative perception, but in positive parenting support measures, especially because many parents with these difficulties could never become authors of PAHT.

We often chose extreme limits for distributions of transition probabilities between stages leading to PAHT used in probabilistic sensitivity analyses. Indeed, few studies have documented the mid- and long-term outcomes of PAHT.^1,2,35,36 In these studies, the definition and qualification of injury varied, as well as initial severity, age at time of injury, follow-up periods, and outcome measures; attrition rates were often high. The use of a consensual definition, better follow-up of all victims, and measures of any possible consequences of PAHT are necessary.

We used available information about the incidence, based on the International Classification of Diseases. Several studies have documented the lack of accuracy of external cause coding in hospital records, in particular its low sensitivity.^37,38 A recent study estimated, however, the sensitivity and specificity respectively at 92% and 96%.³⁹ Consequently, the incidence of PAHT might be underestimated in our models, so that the benefits of a primary prevention program are underestimated. Similarly, we did not take into account the incidence variation during the first year of life by lack of precise estimations, but incidence is higher the younger the children and the benefits of the screening program would be underestimated.¹⁶

In our model, we have considered that without intervention, child health could deteriorate gradually but not improve. However, after the occurrence of PAHT, some improvement of child health over time could occur.⁴⁰ Consequences of the same PAHT would be also more important for younger children because of higher physiological vulnerability during early infancy.⁴¹

The effectiveness of primary prevention was considered constant over time but might decrease over time. The impact of this decrease is likely to be small, because of a decrease of the incidence over time. We considered that effectiveness of screening was constant over time. However, the effectiveness of screening could vary according to the state of detection and the age of the children. Similarly, we chose probability of screening participation constant over the time in each group (parents who were never false-positive and parents who were already at least once false-positive). But screening participation could change over the time.

We have considered a primary prevention program directed to all parents of newborns. A primary prevention program could be also only targeted for families with risk factors for PAHT. However, identification of high-risk families, before the occurrence of PAHT could be responsible for stigmatization of these families and a negative effect could be foreseen. Moreover, the predictive value of risk factors should be very strong, and should not imply missing opportunities to prevent occurrence of PAHT in other families. Current scientific knowledge is insufficient to define an adequate high-risk population. Actually, proposed primary prevention programs are systematic programs.^5,8,9 A screening program could be considered in a high-risk population seen in emergency departments; similarly, current knowledge is insufficient to define such a population. The rhythm at which screening tests should be repeated, who would be the program actors, and the type of intervention remain to be defined. We have chosen a decreasing frequency of screening tests because of age-specific incidence curves of PAHT.¹⁶ Actors who would screen should accept this procedure. Finally, no early intervention is well defined, but coping strategies to better react to infant crying could be adapted for parents who have negative interactions with their child.⁷

Following current recommendations that cost-effectiveness analysis should only be carried out for programs with a demonstrated effectiveness, we have not considered the economic impact of the modeled programs.⁴² Nevertheless, the cost-effectiveness ratio is important and should be estimated before implementation of prevention or screening programs. Further, the full cost of PAHT to society is not well documented today, precluding any full cost-effectiveness analysis. Medical costs has been studied,⁴³ but indirect costs of PAHT have never, to our knowledge, been considered. PAHT has an impact on educational pathways of affected children and, later, on their social integration and working capacity, in particular when they have severe physical or psychic disability.⁴³ These long-term consequences and their cost have not been sufficiently studied.

Conclusions

Our models confirm the relevance of primary prevention of PAHT. Before considering screening, future research should focus on describing the effects of wrongly identifying parents as abusers, and developing and evaluating tests to identify earlier children with PAHT.

Footnotes

Author Disclosure Statement

No competing financial interests exists.

References

Barlow

K.M.

, Thomson

, Johnson

, and Minns

R.A.

(2005). Late neurologic and cognitive sequelae of inflicted traumatic brain injury in infancy. Pediatrics, 116, e174–185.

Makaroff

K.L.

, and Putnam

F.W.

(2003). Outcomes of infants and children with inflicted traumatic brain injury. Dev. Med. Child Neurol., 45, 497–502.

Keenan

H.T.

, Runyan

D.K.

, Marshall

S.W.

, Nocera

M.A.

, Merten

D.F.

, and Sinal

S.H.

(2003). A population-based study of inflicted traumatic brain injury in young children. JAMA. 290, 621–626.

Palusci

V.J.

, and Covington

T.M.

(2014). Child maltreatment deaths in the U.S. National Child Death Review Case Reporting System. Child Abuse Negl., 38, 25–36.

Dias

M.S.

, Smith

, DeGuehery

, Mazur

, Li

, and Shaffer

M.L.

(2005). Preventing abusive head trauma among infants and young children: a hospital-based, parent education program. Pediatrics, 115, e470–e477.

Keenan

H.T.

, and Leventhal

J.M.

(2010). A case-control study to evaluate Utah's shaken baby prevention program. Acad. Pediatr., 10, 389–394.

Goulet

, Frappier

J.Y.

, Fortin

, Déziel

, Lampron

, and Boulanger

(2009). Development and evaluation of a shaken baby syndrome prevention program. J. Obstet. Gynecol. Neonatal Nurs., 38, 7–21.

Barr

R.G.

, Barr

, Fujiwara

, Conway

, Catherine

, and Brant

(2009). Do educational materials change knowledge and behaviour about crying and shaken baby syndrome? A randomized controlled trial. Can. Med. Assoc. J., 180, 727–733.

Altman

R.L.

, Canter

, Patrick

P.A.

, Daly

, Butt

N.K.

, and Brand

D.N.

(2011). Parent education by maternity nurses and prevention of abusive head trauma. Pediatrics, 128, e1164–e1172.

10.

Friedman

, Reed

, Sharplin

, and Kelly

(2012). Primary prevention of pediatric abusive head trauma: a cost audit and cost-utility analysis. Child Abuse Negl. 36, 760–770.

11.

Parks

S.E.

, Annest

J.L.

, Hill

H.A.

, and Karch

D.L.

(2012). Pediatric abusive head trauma: recommended definitions for public health surveillance and research. Centers of Disease Control and Prevention: Atlanta.

12.

Adamsbaum

, Grabar

, Mejean

, and Rey-Salmon

(2010). Abusive head trauma: judicial admissions highlight violent and repetitive shaking. Pediatrics, 126, 546–555.

13.

Jenny

, Hymel

K.P.

, Ritzen

, Reinert

S.E.

, and Hay

T.C.

(1999). Analysis of missed cases of abusive head trauma. JAMA. 281, 621–626.

14.

Campbell

K.A.

, Berger

R.P.

, Ettaro

, and Roberts

M.S.

(2007). Cost-effectiveness of head computed tomography in infants with possible inflicted traumatic brain injury. Pediatrics, 120, 295–304.

15.

Rangel

E.L.

, Cool

B.S.

, Bennett

B.L.

, Shebesta

, Ying

, and Falcone

R.A.

(2009). Eliminating disparity in evaluation for abuse in infants with head injury: use of a screening guideline. J. Pediatr. Surg., 44, 1229–1234; discussion 1234–1235.

16.

Barr

R.G.

, Trent

R.B.

, and Cross

(2006). Age-related incidence curve of hospitalized Shaken Baby Syndrome cases: convergent evidence for crying as a trigger to shaking. Child Abuse Negl. 30, 7–16.

17.

Barr

R.G.

(2012). Preventing abusive head trauma resulting from a failure of normal interaction between infants and their caregivers. Proc. Natl. Acad. Sci. U. S. A., 109, Suppl 2, 17294–17301.

18.

MacMillan

H.L.

, MacMillan

J.H.

, and Offord

D.R.

(1993). Periodic health examination, 1993 update: 1. Primary prevention of child maltreatment. The Canadian Task Force on the Periodic Health Examination. Can. Med. Assoc. J., 148, 151–163.

19.

Russell

L.B.

(2011). Exploring the Unknown and the unknowable with simulation models. Med. Decis. Making, 31, 521–523.

20.

Bellamy

, and Beaumel

(2013). Bilan Démographique 2012. INSEE première n°1429.

21.

Selph

S.S.

, Bougatsos

, Blazina

, and Nelson

H.D.

(2013). Behavioral interventions and counseling to prevent child abuse and neglect: a systematic review to update the U.S. Preventive Services Task Force recommendation. Ann. Intern. Med., 158, 179–190.

22.

Berger

R.P.

, Fromkin

, Rubin

, Snyder

, Richichi

, and Kochanek

(2015). Serum D-dimer concentrations are increased after pediatric traumatic brain injury. J. Pediatr., 166, 383–388.

23.

Dubowitz

, Kim

, Black

M.M.

, Weisbart

, Semiatin

and Magder

L.S.

(2011). Identifying children at high risk for a child maltreatment report. Child Abuse Negl. 35, 96–104.

24.

Kotch

J.B.

, Browne

D.S.

, RIngwalt

C.L.

, Stewart

P.W.

, Ruina

, Holt

, Lowman

, and Jung

J.W.

(1995). Risk of child abuse or neglect in a cohort of low-income children. Child Abuse Negl. 19, 1115–1130.

25.

Parrish

, Baldwin-Johnson

, Volz

, and Goldsmith

(2013). Abusive head trauma among children in Alaska: a population-based assessment. Int. J. Circumpolar Health, 72. doi: 10.3402/ijch.v72i0.21216, eCollection 2013.

26.

Fujiwara

, Barr

R.G.

, Brant

R.F.

, Rajabali

, and Pike

(2012). Using International Classification of Diseases, 10th edition, codes to estimate abusive head trauma in children. Am. J. Prev. Med., 43, 215–220.

27.

Shanahan

M.E.

, Zolotor

A.J.

, Parrish

J.W.

, Barr

R.G.

, and Runyan

D.K.

(2013). National, regional, and state abusive head trauma: application of the CDC algorithm. Pediatrics, 132, e1546–e1553.

28.

Kesler

, Dias

M.S.

, Shaffer

, Rottmund

, Cappos

, and Thomas

N.J.

(2008). Demographics of abusive head trauma in the Commonwealth of Pennsylvania. J. Neurosurg. Pediatr., 1, 351–356.

29.

Shein

S.L.

, Bell

M.J.

, Kochanek

P.M.

, Tyler-Kabara

E.C.

, Wisniewski

S.R.

, Feldman

, Makoroff

, Scribano

P.V.

, and Berger

R.P.

(2012). Risk factors for mortality in children with abusive head trauma. J. Pediatr., 161, 716–722.e1.

30.

Niederkrotenthaler

, Xu

, Parks

S.E.

, and Sugerman

D.E.

(2013). Descriptive factors of abusive head trauma in young children–United States, 2000–2009. Child Abuse Negl. 37, 446–455.

31.

Stewart

T.C.

, Polgar

, Gilliland

, Tanner

D.A.

, Girotti

M.J.

, Parry

, and Fraser

D.D.

(2011). Shaken baby syndrome and a triple-dose strategy for its prevention. J. Trauma, 71, 1801–1807.

32.

Zeman

(2005). Etiology of loss among parents falsely accused of abuse or neglect. J Loss Trauma, 10, 19–31.

33.

Rodriguez

C.M.

, and Tucker

M.C.

(2011). Behind the cycle of violence, beyond abuse history: a brief report on the association of parental attachment to physical child abuse potential. Violence Victims, 26, 246–256.

34.

Wilkins

(2012). Disorganised attachment indicates child maltreatment: how is this link useful for child protection social workers?. J. Soc. Work Pract., 26, 15–30.

35.

Stipanicic

, Nolin

, Fortin

, and Gobeil

M.F.

(2008). Comparative study of the cognitive sequelae of school-aged victims of Shaken Baby Syndrome. Child Abuse Negl. 32, 415–428.

36.

Talvik

, Männamaa

, Jüri

, Põder

, Hämarik

, Kool

, and Talvik

(2007). Outcome of infants with inflicted traumatic brain injury (shaken baby syndrome) in Estonia. Acta Paediatr. 96, 1164–1168.

37.

McKenzie

, Enraght-Moony

E.L.

, Walker

S.M.

, McClure

R.J.

, and Harrison

J.E.

(2009). Accuracy of external cause-of-injury coding in hospital records. Inj. Prev., 15, 60–64.

38.

Hooft

, Ronda

, Schaeffer

, Asnes

A.G.

, and Leventhal

J.M.

(2012). Identification of physical abuse cases in hospitalized children: Accuracy of International Classification of Diseases codes. J. Pediatr., 162, 80–85.

39.

Berger

R.P.

, Parks

, Fromkin

, Rubin

, and Pecora

P.J.

(2015). Assessing the accuracy of the International Classification of Diseases codes to identify abusive head trauma: a feasibility study. Inj. Prev., 21, e133–e137.

40.

Haviland

, and Russell

R.I.

(1997). Outcome after severe non-accidental head injury. Arch. Dis. Child., 77, 504–507.

41.

Anderson

, Spencer-Smith

, and Wood

(2011). Do children really recover better? Neurobehavioural plasticity after early brain insult. Brain J. Neurol., 134, 2197–2221.

42.

Drummond

M.F.

, Sculpher

M.J.

, Torrance

G.W.

, O'Brien

O.J.

, and Stoddart

G.L.

(2005). Methods for the Economic Evaluation of Health Care Programmes, 3rd ed. Oxford University Press: Oxford.

43.

Peterson

, Xu

, Florence

, Parks

S.E.

, Miller

T.R.

, Barr

R.G.

, Barr

, and Steinbeigle

(2014). The medical cost of abusive head trauma in the United States. Pediatrics, 134, 91–99.