OBJECTIVES: Because physicians may have difficulty distinguishing accidental fractures from those that are caused by abuse, abusive fractures may be at risk for delayed recognition; therefore, the primary objective of this study was to determine how frequently abusive fractures were missed by physicians during previous examinations. A secondary objective was to determine clinical predictors that are associated with unrecognized abuse.
METHODS: Children who were younger than 3 years and presented to a large academic children's hospital from January 1993 to December 2007 and received a diagnosis of abusive fractures by a multidisciplinary child protective team were included in this retrospective review. The main outcome measures included the proportion of children who had abusive fractures and had at least 1 previous physician visit with diagnosis of abuse not identified and predictors that were independently associated with missed abuse.
RESULTS: Of 258 patients with abusive fractures, 54 (20.9%) had at least 1 previous physician visit at which abuse was missed. The median time to correct diagnosis from the first visit was 8 days (minimum: 1; maximum: 160). Independent predictors of missed abuse were male gender, extremity versus axially located fracture, and presentation to a primary care setting versus pediatric emergency department or to a general versus pediatric emergency department.
CONCLUSIONS: One fifth of children with abuse-related fractures are missed during the initial medical visit. In particular, boys who present to a primary care or a general emergency department setting with an extremity fracture are at a particularly high risk for delayed diagnosis.
Although fractures are a common presenting finding in child abuse,1,2 clinicians may have difficulty differentiating abuse-related fractures from those that are caused by accidental trauma2–4; however, this distinction is crucial because of its impact on consequences for the child.5,6 Whereas accidental injuries carry only their inherent risks, repeat injury occurs in 35% of all abuse cases, and 5% to 10% of patients will die if there is no intervention.7
Despite the serious risks associated with delayed recognition of abusive fractures, patient assessment for this diagnosis is often suboptimal.1–4,8 One study found that of 100 children who were younger than 3 years and presented to an emergency department (ED) with long bone fractures, 31 had indicators suggestive of abuse but only 1 was referred to child protection services for additional assessment.2 Banaskiewitz et al3 demonstrated that in infants who were younger than 1 year, the possibility of abuse was underestimated by ED clinicians in ∼28% of cases when compared with a retrospective diagnosis by a child protection team pediatrician. Moreover, research conducted in a pediatric ED demonstrated that 42% of charts reviewed did not have adequate documentation to explain the cause of the fractures, and inflicted injuries were therefore not adequately ruled out.1
This evidence suggests that abusive fractures are likely at risk for escaping detection or delayed recognition; however, the frequency with which this occurs remains unknown. The primary objective of this study was to determine the proportion of abuse-related fractures that were missed at previous physician encounters. The clinical factors that may have contributed to the reasons for the diagnostic delay were also examined.
Children who were younger than 3 years,2,9–11 had abusive fractures that occurred from January 1993 to December 2007, and were referred to a multidisciplinary hospital-based Suspected Child Abuse and Neglect (SCAN) team at the Toronto Hospital for Sick Children (HSC) were included. HSC SCAN consists of specialty pediatricians, psychologists, social workers, and nurse practitioners. Members of HSC SCAN team are the only child abuse specialists in the Greater Toronto Area and are involved in the assessment of most cases of suspected abuse in that area. The HSC SCAN team's assessment results in a classification of these fractures as abusive, indeterminate, or accidental. The study sample included only cases for which the first physician visit was primarily for an isolated fracture. Cases were excluded when the child's clinical presentation was predominantly consistent with some other type of trauma, medical records were inaccessible, only metaphyseal corner chip fractures (usually asymptomatic) were present, or the cause of the fracture was indeterminate or accidental.
Fractures were determined to be abusive when at least 1 of the following criteria was met2,6,12: (1) confession of intentional injury by an adult caregiver; (2) inconsistent/inadequate history provided; (3) inappropriate delay in seeking medical care; (4) associated inadequately explained injuries; (5) in the absence of bone disease, presence of fractures uncommon for accidental injury and frequently reported in abusive injury (eg, metaphyseal limb fractures, posterior rib fractures not caused by birth trauma)6,13,14; and (6) witness to abuse came forward.
A case was considered “recognized” when a referral to local child protection authorities was made the first time the child presented to a physician with the index fracture(s). This is in contrast to “missed” when the child had at least 1 physician encounter for the index fracture(s) before the visit when the abuse was confirmed. In all missed cases, the signs and symptoms compatible with a fracture and/or a radiograph diagnosis were present at the initial visit, but the possibility of abuse was not raised. Thereafter, ≥1 of the following occurred: (1) the child improved clinically but experienced repeat trauma and the HSC SCAN team found the previous fracture(s) abusive; (2) recognition of red flags and referral to the SCAN team at a routine follow-up for the index fracture(s) led to recognition of abuse; (3) the child's continued symptoms resulted in repeat unscheduled visits and a referral to the SCAN team with recognition of the index fracture(s) as abuse-related; (4) the index radiographs initially read as normal by the primary treating physician were found by a radiologist to have a fracture that required a repeat visit, when the suspicion for abuse was raised; (5) the perpetrator later confessed or a witness came forward; and/or (6) abuse was suspected in a sibling and review of the patient's fractures yielded abuse as the cause. The determination of missed versus recognized cases was made independent of the knowledge of potential predictors.
Because specific income of the family was not available, this was estimated on the basis of median income of families in a given postal code.15 On the basis of the 2006 Ontario median household income of $60455, median income was then additionally classified as low (<$45341.25), middle ($45341.25—$90682.50), or high (>$90682.50).16 Income classification was used as a surrogate measure of socioeconomic status.17
Social concerns were defined as any primary caregiver who had ≥1 of the following: young single parent (younger than 20 years and no live-in partner at the time of the child's evaluation); previous contact with child protection services; or history of incarceration, substance abuse problem, living in group housing (eg, shelter), or domestic violence. A positive skeletal survey was defined as additional fractures other than the index fracture(s). In a primary care office, children are assessed by a family physician or pediatrician. In general EDs that serve all ages, children are seen by ED physicians.
Case Selection, Data Collection, and Review
Once the HSC SCAN team has reviewed a case, referral information is entered into a database. This database was searched for eligible patients, and study-specific information of identified cases was collected from original patient records (Fig 1). Information collected by 2 research assistants (Ms Ravichandiran and Dr Bejuk) who were trained in the methods of chart abstraction included relevant patient and family demographics, social history, history of present illness, details of the child's injury(ies), subsequent clinical course, and details from previous visits related to the index fracture(s). For missed cases, the clinical data from the initial physician visit(s) before the visit when abuse was diagnosed were reviewed by 1 SCAN physician (Dr Al-Harthy), who was masked to the final SCAN opinion and the purpose of the study, to ascertain the presence of indicators of abuse that should have led to a referral to a child protective team at that visit.
An a priori defined list of potential predictors that were independent of the outcome of a missed diagnosis of abuse was selected by 3 expert members of the HSC SCAN team1,2,6,10,12,18,19 and later modified in accordance with the available data. For example, although race3,8,12 has been strongly associated with referrals to child protective teams, this information is not collected by the reviewing HSC SCAN team. The final list of predictors is detailed in Table 1. Some of the variables used routinely in ascertaining abuse could not be considered as predictors because they are not independent of the outcome.
After data collection was complete, information on each patient was reviewed for accuracy and completeness by a pediatric ED physician (Dr Boutis) in collaboration with the 2 research assistants. Missing data were imputed by inserting the respective median (categorical) or mean (continuous data) value from the group data into blank cells.20 Permission for this research was obtained from our research ethics board.
The sample size was calculated by using the methods by Hsieh21 and the following parameters were used: α = .05, and β = .20, estimated proportion of missed abusive fractures of 20%,22 and an odds ratio (OR) of 2.023 of missed abuse corresponding to an increase of 1 SD from the mean value of a covariate.21 In this study, there are multiple covariates and a possibility of some unknown correlation between covariates. Thus, a conservative value of ρ = .5 was estimated, and the adjusted minimal total sample size is therefore 182.
A univariate analysis was used to assess whether a particular variable was associated with the outcome variable of interest, missed case of abusive fracture (Table 1). For the latter, Pearson χ2 test was used for categorical values and independent Student's t test for continuous variables. Independent variables with P ≤ .20 and any relevant interaction and confounding terms were entered into a multivariate logistic regression model using the forward selection method (Table 2). 24 Approximately 14 missed cases per variable were entered into the model, meeting the minimal criteria of 10 events per variable to minimize overfitting of the data.25–27 Wald and Likelihood ratio testing were then used to iteratively remove noncontributory variables from the model.24 Goodness of fit of the final model to the data was tested by using the Hosmer-Lemeshow test. A receiver operating characteristic curve was plotted to check the predictive ability of the model. Odds of a case being missed for a given variable were reported with respective 95% confidence intervals (CIs). All analyses were performed by using SPSS 13 for Windows (SAS Institute, Cary, NC).
This study included 258 eligible patients with abusive fractures (Fig 1). A comparison of characteristics of missed and recognized cases is detailed in Tables 1 and 2. Of the 258 patients, 54 (20.9% [95% CI: 15.8–26.0]) had at least 1 previous physician visit at which abuse was missed. Of the 145 children with an abusive extremity fracture, 41 (28.3% [95% CI 20.8–35.8]) were “missed.” From the initial visit for the index fracture(s), the median delay of the diagnosis of abuse was 8 days (minimum: 1; maximum: 160), and the median number of physician visits was 1 (minimum: 1; maximum: 3). Of the children who re-presented for medical care after the abusive cause of the fracture was missed, 9 (16.7%) presented with new abusive injuries. In 7 of these cases, there was a different fracture; 1 child had serious abdominal injuries, and another had serious head trauma that resulted in death.
Incorrect interpretation of the radiograph findings by the physician resulted in 18 (33.3%) missed cases (Fig 2), 7 of which were skull fractures. In 7 (13.0%), the initial imaging series was incomplete and the abuse-related fracture was therefore not seen. This subgroup returned to an ED because of persistence of symptoms, more extensive imaging was performed, the fracture was detected, and a referral to the SCAN team was made. The exact reasons that the remaining 29 cases were missed are not certain because of a lack of available data; however, inadequate screening or accepting implausible mechanisms may have contributed to missing these cases. SCAN documentation revealed that these children had risk factors for abuse: 25 (86.2%) of 29 were nonambulatory; in 26 (90.0%) of 29, parental report of mechanism did not explain injuries; and 14 (48.3%) of 29 had social concerns. Furthermore, review of the initial visit records demonstrated that 13 (50.0%) of 26 had incomplete documentation of the preceding events or possible related risk factors for abuse.
The univariate analysis demonstrated that 3 variables were found to be significantly associated with a missed diagnosis of abuse: male gender, initial presentation to a nonpediatric ED, and an extremity fracture (Table 1). The probability of missing this diagnosis for each predictor after adjustment for all significant predictors is summarized in Table 3. No statistically significant interaction terms or confounding variables were identified in this analysis. In the resultant model, the Hosmer-Lemeshow goodness-of-fit test did not reject the null hypothesis of good fit (P = .718), and the predictive ability of the model is good (area under the curve: 0.841). Applying this model predicts that if all 3 factors were present, then the probability that an abusive fracture would be missed is 50%.
Sixteen charts were missing and unavailable for review, and if we assume that all were recognized abuse, then the proportion of missed abuse would decrease only to 54 (19.1%) of 274. Of the 139 initial visits that occurred outside HSC, 3 of 45 of the missed and 15 of 94 of the recognized first physician visit documents were not available for detailed review. In 2 (0.8%) of the 258 cases, the initial clinical setting could not be determined. In 16 (6.2%) cases, it was uncertain whether an injury was reported. Forty (15.5%) had living status of parents unavailable. Finally, a postal code was not recorded for 49 (19.0%) cases. Sensitivity analyses with and without imputed data for these missing variables were performed and did not reveal any significant differences; therefore, only unimputed results are presented.
This study is the first to report the frequency of delayed recognition of abusive fractures in children. One fifth of children with abusive fractures were missed at initial physician visits, which is comparable to that reported for other types of abuse12,19; however, we do not know how many cases of abusive fractures are never detected. We also found that boys, children who present to a nonpediatric ED or a primary care setting, and/or those with an extremity fracture seem to be at the highest risk of the abusive etiology of the fracture escaping of detection by a physician at an initial visit.
In ∼17% of missed abuse cases, children sustained repeat injuries between their initial visit and their eventual diagnosis of abuse; previously missed fractures that led to serious abusive injuries were also found by Oral et al.28 The skeletal survey that was performed during subsequent visits may have a major impact on the correct diagnosis. In this study, two thirds of patients had healing fractures identified on the survey, and this is higher than that reported previously.5,22 This highlights the importance of having a low threshold to consider a skeletal survey for children who may be at risk for abuse5,14,22 before dismissing the fractures as accidental.
In the 54 missed cases, approximately one third of the fractures were not detected on the initial radiographs by front-line physicians in a country where immediate radiology interpretations are not routine practice in the ED or office setting. Pediatricians have limited skills in the recognition of fractures on radiographs.29 This is true particularly of skull fractures,30 and identification of this type of fracture, especially in very young infants, may prompt the physician to assess for other maltreatment risk factors.31 This study suggests that front-line physicians should strongly consider consulting a radiologist when the presence of a fracture may lead to increased suspicion of abuse.
In our study, abuse was more likely to be missed when a child presented to a general ED or primary care setting. These results support those by Trokel et al,23 who found lower rates of abuse in patients who had traumatic brain injury or femur fracture and presented to general hospitals compared with their pediatric counterparts. This could suggest that abuse may be missed in these settings. Clinicians who work in these areas may lack expertise in the recognition of abuse-related fractures despite the presence of indicators for abuse.1,2,32 This research supports the need for quality improvement programs at general hospitals and primary care settings.
Children with extremity shaft fractures caused by abuse were also found to be at increased risk for having physicians attribute their injuries to accidental causes. Although extremity fractures are the most common skeletal injuries that occur in abused children,2 radiology literature demonstrates that these injuries also have the lowest specificity for abuse.14 No fracture on its own can distinguish an accidental from a nonaccidental trauma,31 but the likelihood of abuse increases when there is a fracture in a nonambulatory child and when the fracture type includes the femur or humerus in infants who are younger than 18 months.31,33,34 Indeed, in this study, these types of fractures in nonambulatory children were commonly seen in cases for which abuse was missed. In addition, many of the missed extremity fractures had associated risk factors for abuse that were not adequately screened for at the initial physician visit; therefore, the possibility of abuse should be carefully considered for children with extremity fractures, and associated risk factors should be excluded.
An abuse-related fracture was almost twice as likely to be missed in a boy versus a girl. Although the reason for this is unclear, injuries in general occur more often in boys,35 which may bias a clinician in assuming that the cause of a fracture is accidental.
This study has limitations that warrant consideration. This was a retrospective study with its inherent limitations, such as missing data, and thus absent data may have biased predictor variable results. Although our case classification was based on current available standards for the diagnosis of abuse, there may have been ascertainment errors. Children with abusive fractures that were never referred to the SCAN team and were assumed to be accidental were not included in this review; however, given that ED records are often incomplete,1 a retrospective assessment by the child protection team of all of the nonreferred cases would have resulted in only speculative assignments of cause. Finally, although most cases of abusive fractures are seen by our SCAN team, some of the less complex cases may not have been seen. This introduces the potential for referral bias, and it may result in an overestimation of the proportion of cases that are missed at an initial physician visit; however, child abuse is underrecognized,12 and there is also the possibility that we are underestimating the proportion of cases missed.
Our results suggest that a considerable proportion of abuse-related pediatric fractures are missed during the initial visit. We can make the following suggestions that may facilitate the diagnoses of abusive fractures. A detailed review of the mechanism and screening for other risk factors of abuse should be included in the initial assessment of a young child with fractures. Children who are nonambulatory are at especially high risk, and consultation with the child protection team in these cases is often appropriate. Clinicians should have a low threshold to perform a skeletal survey in potentially vulnerable populations, and a radiologist's review of any imaging that may change suspicion for abuse is recommended. Finally, appropriate targeted education or practice guidelines may help in achieving better outcomes in clinical settings that are susceptible to missing abusive fractures.
This research was supported by a grant from the Canadian Hospitals Injury Reporting and Prevention Program (CHIRPP).
We acknowledge the efforts of Dr S. Walter and Mr A.O. Odueyungbo for statistical expertise and critical review of the analysis.
- Accepted July 28, 2009.
- Address correspondence to Kathy Boutis, MD, MSc, 555 University Ave, Toronto, ON, M5G 1X8, Canada. E-mail:
Financial Disclosure: The authors have indicated they have no financial relationships relevant to this article to disclose.
What's Known on This Subject:
Patient assessment by physicians of children who are at risk for abuse is suboptimal, and, therefore, abusive fractures are at risk for escaping detection or delayed recognition. It is unknown, however, how often this occurs.
What This Study Adds:
Approximately 20% of abusive fractures were missed at initial physician visits. Boys who present to a nonpediatric ED with an extremity fracture seem to be at highest risk of the abusive etiology of the fracture escaping of detection by a physician.
- ↵Johnson CF, Oral R. Diagnosis and Management of Child Abuse of Children. Columbus, OH: Ohio State University Print Shop; 1999
- ↵Worlock P, Stower M, Barbor P. Patterns of fractures in accidental and non-accidental injuries in children: a comparative study. Br Med J (Clin Res Ed).1986;293 (6539):100– 102
- ↵Kleinman PK. Diagnostic Imaging of Child Abuse. 2nd ed. St Louis, MO: Mosby; 1998
- ↵Statistics Canada. Census Tract Profiles, 2006 Census. Ottawa, Ontario, Canada; 2006. Available at: www12.statcan.ca/english/census06/data/profiles/ct/Index.cfm?Lang=E. Accessed August 8, 2008
- ↵Statistics Canada. Study: Income Inequality and Redistribution. Ottawa, Ontario, Canada; 2007. Available at: www.statcan.gc.ca/daily-quotidien/070511/dq070511b-eng.htm. Accessed September 12, 2008
- ↵Heisz A. Income inequality and redistribution; 1976 to 2004. Analytical Studies Branch Research Paper Series.2007;(298):1– 58
- ↵Trokel M, Wadimmba A, Griffith J, Sege R. Variation in the diagnosis of child abuse in severely injured infants [pubished correction appears in Pediatrics. 2006;118(3):1324]. Pediatrics.2006;117 (3):722– 728
- ↵Kleinbaum DG, Kupper LL, Muller KE, Nizam A. Applied Regression Analysis and Other Multivariable Methods. 3rd ed. Pacific Grove, CA: Brooks/Cole Publishing Company; 1998
- ↵Ryan LM, DePiero AD, Sadow KB, et al. Recognition and management of pediatric fractures by pediatric residents. Pediatrics.2004;114 (6):1530– 1533
- ↵Kemp AM, Dunstan F, Harrison S, et al. Patterns of skeletal fractures in child abuse: systematic review. BMJ.2008;337 :a1518
- ↵Leventhal JM, Martin KD, Asnes AG. Incidence of fractures attributable to abuse in young hospitalized children: results from analysis of a United States database. Pediatrics.2008;122 (3):599– 604
- ↵Cramer KE, Scherl SA. Orthopedic Surgery Essentials: Pediatrics. Philadelphia, PA: Lippincott Williams & Wilkins; 2003
- Copyright © 2010 by the American Academy of Pediatrics