Three Part Question
In [children presenting to the Emergency Department] are [rib fractures on plain radiographs] [specific for non-accidental injury]?
While reading an orthopaedic text you find a table that states rib fractures are highly specific for non-accidental injury. No papers are referenced and you wonder what evidence exists to support this statement.
Cochrane and Medline (including Medline corrections) 1966 –01/08/2003.
Cochrane:Rib fractures and non-accidental injury.
Medline [(Validated paediatric search filter (March 2003) for Ovid BestBETS ) AND (exp Child Abuse or non-accidental injury.mp or child abuse$.mp or deliberate injury.mp or exp. domestic violence or child abuse, sexual or exp. Munchausen syndrome by proxy or exp torture or domestic violence.mp or Munchausen syndrome by proxy.mp. or torture.mp or non-accidental injuries.mp) and (exp.rib fractures or rib fracture$.mp or posterior rib fracture$.mp or multiple rib fracture$.mp or bilateral rib fracture$.mp or exp. thoracic injuries or thoracic injury.mp or chostochondral junction injuries.mp)] limited to human and English Language
Cochrane:No relevant reviews found.
Medline:93 papers were identified, 86 were of insufficient quality for inclusion or irrelevant. One paper was subsequently excluded on critical appraisal due to flaws in case selection. Handsearch 2 further papers sufficient quality for inclusion.
|Author, date and country
||Study type (level of evidence)
|467 children referred to UK paediatrician for opinion as to cause of injuries. 425 were below 2 years of age.||Case series with non-independent reference standard. (3B)||Specificity LR +||100 (96-100)% 52||Case collected from personal case series and as such not reflective of a standard paediatric take.|
|Bulloch et al|
|39 children (all less than 12 months of age) identified as having rib fractures on basis of standard American radiology codes||Case series with non-independent reference standard. (3B)||Positive predictive value||82 (66.5–82.5)%||Each film reported by a single radiologist, with causality determined by case review involving 2 paediatricians.
|18 infants (less than 2 years) with documented rib fractures in a tertiary referral paediatric centre.||Case series with non-independent reference standard. (3B)||Positive predictive value||83 (58.6-96.4)%||Each child's case reviewed by multi-disciplinary team to determine whether child was abused or not.
|215 children (less than 3 years) with fractures.
Data collection from Emergency Medicine Department logs and
Child Abuse Register for that period||Case series with non-independent reference standard. (3B)||Specificity if all children in "unknown group" were abused||100 (94-100)%||Due to nature of data reporting, the only figure of value is that there were no children with rib fractures who had been abused, allowing a specificity to be calculated.|
|Specificity if all children in "unknown group" were not abused||100 (96-100)%|
|King et al|
|189 (<1 month to 13 years) children referred to an assessment team for investigation of abuse. 163 were less than 2 years of age||Case series with non-independent reference standard. (3B)||Positive predictive value||22(17-27)%||Not explicitly stated whether all children received a chest radiograph.
Wide age range|
|21 children (3 months to 15 years 4 months) admitted to paediatric emergency care facility with rib fractures between 01/1980 and 06/1985)||Case series with non-independent reference standard. (3B)||Positive predictive value||23(5-41)%||Small numbers.
Wide age range.|
|Merten et al|
|Initial recruitment of 904 infants and children with "strong clinical evidence of abuse". Only those with complete radiological examinations (n=494) were included in the analysis. Of those analysed, 419 were less than 5 years of age.||Case series with non-independent reference standard. (3B)||Age stratified positive predictive value. Age (years)/PPV. <1 (n=190)||10 (4-16)%||Sample included those children who had been sexually abused as well as neglected. It is unclear how many of these children were included in the analysis.|
|1-2 (n=101)||0.8 (0.62-0.98)%|
|2-5 (n=125)||0.2 (0.03-0.37)%|
|>5 (n=75)||0.1 (0.07-0.13)%|
|All ages||6 (4-8)%|
|25 infants (all less than 1 year) with rib fracture identified from large number of chest x-rays (exact number not specified). Data collected 1969 – 1979||Case series with non-independent reference standard. (3B)||Positive predictive value||24 (8-40)%||Data collected both from inpatients and outpatients.
All the studies appraised lack a gold standard method for the diagnosis of non-accidental injury. Even if such a tool existed, it would not be globally applicable as subtle cultural factors play a part in defining child abuse. Furthermore, the validation study of rib fractures, as a 'rule in' for non-accidental injury would ideally exclude the radiograph from the diagnostic gold standard. The chest radiograph is central to diagnostic protocols, and such purism would be hard to justify.
A further criticism of all the papers reviewed is a lack of explicit blinding between radiologists and investigating parties. Blinding in this circumstance would be limited to stating the child has either received a chest injury or is under investigation for non-accidental injury. The clinical information that accompanies a radiograph is an essential part of the ultimate report, as subtle differences in rib morphology may be variably interpreted according to the clinical scenario.
An ideal study and that feasible in the clinical setting may be difficult to reconcile. The purist would argue that we should aim for perfection, particularly in an area where accurate diagnosis is vital. To the authors' knowledge there is no globally applicable diagnostic tool in use for the diagnosis of non-accidental injury, nor a nationally accepted gold standard approach for diagnosing non-accidental injury. As such the comparators used by the various investigators are hard to criticise, but limits the comparability of the various results.
All the studies examine a pre-selected group of patients, and as such the values obtained can only be applied to those specific groups. A number of the studies have sufficient data to derive a two-by-two table from which likelihood ratios could be calculated, but the data is not presented in a way to allow these to be extracted.
The largest UK data set is from Carty, who presents a personal case series, which while large in number, is unreflective of the situation in the emergency department or paediatric assessment unit. The selection bias leads to an over estimation of the specificity of rib fractures in the general population presenting to the general physician. Rib fractures in older children result from sport, accidental falls and motor vehicle accidents. In Carty's series there are no accidental rib fractures despite the presence of children older than 2 years of age, which is not reflective of those physicians who work in the acute setting, particularly with children older than three years.
The small, well designed studies by Bulloch and Cadzow support the premise that rib fractures in a child less than 2 years are predictive of non-accidental injury. The data in both these studies examines children in whom a rib fracture is already present and as such cannot be used to derive specificity. Unlike the papers from Carty and Leventhal that calculate a specificity of 100%, their data would suggest a lower value as the numerator for the calculation of specificity is greater than zero. How much this would reduce the specificity is open to conjecture. The remaining papers council caution, calculating lower positive predictive values for radiographic rib fractures and non-accidental injury. King's paper, samples a large group of individuals, but is hampered by the wide age range of children assessed, as well as the lack of confirmation whether all the children received a chest x-ray.
The data supports the conclusion that rib fractures in children less than 3 years are highly specific for non-accidental injury, and as such grouping all children together irrelevant of age is counterproductive. As such, King tells us that rib fractures in children of any age are poor predictors of non-accidental injury. This is reflected in the paper by Merten, that shows a reduction in positive predictive value with increasing age, as well as a lower value when all ages are grouped together. Reanalysis of the data from King by age may have confirmed the figures of Carty, Bulloch, Cadzow and Leventhal . Thomas's paper again indicates a lower predictive value for rib fractures, but looked at children, who diagnoses at the time were felt not to be either pathological or accidental, but on re-reading the paper may well have been non-accidental in nature.
In any study, patient selection affects the applicability of the results, a point of particular relevance to diagnostic test studies. A study carried out in ventilated children in a tertiary referral central, is not applicable to a general practitioner presented with a report confirming a rib fracture in a four-year-old child. Taking this into account, along with the radiation issues already noted, it would be reasonable to look at two separate groups of children, ideally through a prospective trial.
The first group would be those children undergoing chest radiography, each child being subsequently investigated for non-accidental injury through a standard protocol. This would allow the calculation of the specificity of rib fracture in the general hospital population receiving chest radiography. Issues of consent for such a study may be difficult to resolve, as the consent form would explain the reason for the study (the diagnosis of non-accidental injury) and lead to an underestimation of the prevalence of non-accidental injury as abusers may not give consent for their child's inclusion.
The second study would be those children being investigated for abuse (sexual, physical and neglect), and ensure all receive a skeletal survey as part of their work-up. These children are pre-selected into a high-risk group, and such a study would allow the calculation of prevalence, specificity and sensitivity along with negative and positive likelihood ratios, which would further quantify the risk of abuse. It is recognised that forms of abuse are not exclusive, and such a study would allow the specificity of a variety of different fracture types to be calculated in the various forms of abuse.
Clinical Bottom Line
In children under the age of 3 years of age who are suspected of being abused, rib fractures should be considered as highly suggestive.
In the general population, the presence of rib fractures in a child under three years of age should be thoroughly investigated, with non-accidental injury being the most likely diagnosis until proved otherwise.
- Bestbets Paediatric filter [Online] Available at http://www.bestbets.org/links/bb_paed_filter_2003.pdf
- Carty H. Pierce A. Non-accidental injury: a retrospective analysis of a large cohort. European Radiology. 12(12):2919-25, 2002 Dec.
- Bulloch B. Schubert CJ. Brophy PD. Johnson N. Reed MH. Shapiro RA. Cause and clinical characteristics of rib fractures in infants. Pediatrics. 105(4):E48, 2000 Apr.
- Cadzow SP. Armstrong KL. Rib fractures in infants: red alert The clinical features, investigations and child protection outcomes. Journal of Paediatrics & Child Health. 36(4):322-6, 2000 Aug.
- Leventhal JM. Thomas SA. Rosenfield NS. Markowitz RI. Fractures in young children. Distinguishing child abuse from unintentional injuries. American Journal of Diseases of Children. 147(1):87-92, 1993 Jan.
- King J. Diefendorf D. Apthorp J. Negrete VF. Carlson M. Analysis of 429 fractures in 189 battered children. Journal of Pediatric Orthopedics. 8(5):585-9, 1988 Sep-Oct.
- Schweich P. Fleisher G. Rib fractures in children. Pediatric Emergency Care. 1(4):187-9, 1985 Dec.
- Merten DF. Radkowski MA. Leonidas JC. The abused child: a radiological reappraisal. Radiology. 146(2):377-81, 1983 Feb.
- Thomas PS. Rib fractures in infancy. Annales de Radiologie. 20(1):115-22, 1977 Jan-Feb.