Child abuse is a leading cause of morbidity and mortality among young children. A report from the U.S. Department of Health and Human Services estimated that 1840 children died of abuse and neglect in 2019, representing a population rate of 2.5 deaths per 100,000 children [
1]. Radiologists play a crucial role in diagnosing child abuse, as fractures are the second most common finding of child abuse after bruising and other cutaneous signs [
2]. One characteristic injury in fractures due to child abuse is classic metaphyseal lesions (CML) [
3,
4]. Since the first description in 1953 [
5], scientific evidence has accumulated on the specificity of CML for the diagnosis of child abuse, including observed associations of CML with other severe injuries from abuse and findings from histopathologic investigations [
3,
4,
6–
13]. It is critical to detect CML, because missing these lesions can result in a failure to diagnose abuse and thus continued inflicted trauma or even death [
14].
Despite the foregoing evidence, some physicians involved in medicolegal work assert that CML result from vitamin D deficiency and rickets and are not traumatic fractures, thus challenging the specificity of CML for child abuse [
15,
16]. These practitioners claim that the existing literature is incorrect, having implications for medical management and medicolegal cases [
17]. Their arguments provide an alternative diagnosis in cases of suspected child abuse, leading to the possibility that juries may give less consideration to radiographic findings or even that prior convictions for child abuse could be overturned, potentially returning children to dangerous life-threatening environments [
18–
20].
Suboptimal vitamin D status due to insufficient dietary intake and/or insufficient sun exposure is common; therefore, children could have concurrent abuse injury and vitamin D deficiency [
21]. However, most children with biochemical evidence of vitamin D deficiency do not have radiographic signs of rickets [
22]. Investigations evaluating patients with vitamin D deficiency [
23] and those with rickets [
24] for the presence of fractures have not shown evidence of CML in these patients. Another study showed no evidence of rickets in 46 infants with fatal abusive head trauma and CML [
12].
Distinctive radiographic findings have been described that appear unique for both rickets and CML [
25]. Specific radiographic changes in rickets include loss of the zone of provisional calcification, splaying of the metaphyses, fraying of the metaphyses, and widening of the physes [
25]. These findings contrast to those of CML, which have corner or bucket-handle patterns with normal mineralization and physes [
25]. Although rickets and CML have different radiographic findings [
25], no studies to our knowledge have directly evaluated the diagnostic performance of radiologists in differentiating rickets and CML on radiographs. Such studies are challenging to perform. For example, rachitic changes on radiographs of infants and young children with low vitamin D levels are uncommon [
22]. In addition, radiographs obtained to evaluate for rickets are different from radiographs obtained to evaluate for suspected child abuse. Children with suspected rickets typically undergo radiography centered on the knee [
23], whereas children with suspected child abuse typically undergo standard leg radiography that may not include the entire knee in a single view [
26]. In addition, vitamin D levels are not routinely measured in the evaluation of children with suspected child abuse. To address this gap, we conducted this multicenter study to evaluate the diagnostic performance of radiologists in differentiating rickets and CML on radiographs.
Methods
This retrospective HIPAA-compliant multicenter study was performed at seven tertiary care children's hospitals. Institutional review board approval was received at each participating institution with a waiver of the requirement for written informed consent.
Patient Selection
At each site, a fellowship-trained pediatric radiologist (M.B.M. with 9, C.M.P. with 5, T.C. with 14, S.P. with 23, R.H.J. with 3, E.S.R. with 32, and M. Fadell with 12 years of posttraining experience) who did not participate in the later image evaluation searched the local radiology information system (RIS) and electronic medical record (EMR) from January 2007 to December 2018 to identify bilateral leg radiographs of children younger than 2 years that showed evidence of rickets or knee CML. To identify children with rickets, the RIS was searched for radiology reports containing the term “rickets” and any of the following terms: “knee radiograph,” “long leg study,” or “skeletal survey radiographs.” To identify children with CML, the RIS was searched for radiology reports containing the term “skeletal survey” and any of the following terms as a descriptor of the distal femur or proximal tibia: “CML,” “classic metaphyseal lesion,” “corner fracture,” or “bucket handle fracture.”
Vitamin D (25-hydroxyvitamin D) levels were identified from the EMR. Children were excluded if they did not have an available vitamin D level or if the only available vitamin D level was measured more than 90 days after the knee radiographs were obtained (reflecting the time period during which rachitic changes may persist [
27]. Children with suspected rickets were excluded if the vitamin D level was 20 ng/mL or higher or if the radio-graph showed a Thacher rickets severity score (RSS) of 0 (i.e., no evidence of rickets) (see later, Radiograph Selection and Preparation). Children with suspected CML were excluded if the local child abuse pediatrician's report indicated a negative or indeterminate child abuse diagnosis. Demographic information and underlying medical conditions were extracted from the EMR. Children in both groups were further excluded if born before 37 weeks' gestation or if they had skeletal dysplasia, renal or hepatic insufficiency, or hypophosphatemic rickets or had been treated with steroid or anticonvulsive therapy. Children were also excluded if the radiographs could not be successfully edited to present a standardized depiction of the knee (see later, Radiograph Selection and Preparation). These exclusions resulted in a final study sample that comprised a group of children with vitamin D levels less than 20 ng/mL and evidence of rickets on knee radio-graphs obtained within 90 days of vitamin D measurement and a group of children with a diagnosis of child abuse who underwent a skeletal survey showing knee CML and had an available vitamin D measurement within 90 days of the skeletal survey.
The Supplemental Methods (available in the
online supplement) describe recording of laboratory data for all children and additional details about medical record review for additional injuries in children with CML.
Radiograph Selection and Preparation
For each patient, a single set of radiographs was selected for subsequent evaluation. Radiographs obtained closest in time to recording of the low vitamin D level were preferred for this purpose. Follow-up radiographs obtained 90 days or less after the vitamin D measurement were used if the initial radiographs did not meet the selection criteria.
As most radiographs of children younger than 2 years who have low vitamin D levels are expected to be normal [
13,
14], the pediatric radiologist at each site who conducted the patient selection process reviewed radiographs being considered for selection for patients with rickets and assigned an RSS of 0 (normal) to 10 (severe) [
28]. Only radiographs assigned RSS 1 or greater (i.e., radiographic evidence of rickets) were eligible for selection.
For potentially eligible radiographs, the DICOM images were anonymized and sent to a central site (Riley Hospital for Children at IU Health). A fellowship-trained pediatric radiologist at the central site (B.K., 24 years of posttraining experience), who was not involved in patient selection or subsequent image evaluation, reviewed the submitted radiographs to assess whether the radio-graphs could be edited to present a standardized depiction of the knee for children in both groups and assigned a randomized number. The editing process entailed cropping the image to a standardized FOV centered on the knee and zooming the resulting image to a standardized size. Separate edited right and left anteroposterior projections were prepared for each knee, presenting the entirety of the given knee without overlying cast materials (
Fig. 1). Radiographs that could not be edited were ineligible for selection. At the central site, a nonauthor radiologic technologist under the supervision of the pediatric radiologist (B.K.) edited the selected radiographs using a DICOM editor (Sante, version 7.7.3, Santesoft). The resulting edited DICOM images were uploaded to an online platform for subsequent review by blinded radiologists across study sites.
Radiograph Review
Eight radiologists not involved in the patient or image selection independently reviewed the edited radiographs of each patient (separate images of right and left knees) using a DICOM viewer (MicroDicom DICOM Viewer, version 3.8.1, MicroDicom) on a diagnostic high-resolution screen. The reviewers were blinded to clinical information, and the radiographs were randomly presented. These radiologists were two less experienced nonpediatric radiologists (J.L., an emergency radiologist with 1 year of posttraining experience; J.E., a general radiologist with 3 years of posttraining experience), two more experienced non-pediatric radiologists (R.W.C., a musculoskeletal radiologist with 12 years of posttraining experience; C.K.S., an emergency radiologist with 9 years of posttraining experience), two less experienced pediatric radiologists (M.F-A. with 4 and M.P.M. with 3 years of posttraining experience), and two more experienced pediatric radiologists (M.M. and M.R.W., both with 15 years of posttraining experience) from four different centers. The radiologists were informed that all children had either rickets or CML and that children with CML included a combination of those with normal or low vitamin D levels, but they did not know the number of children in each group.
Although all children had a diagnosis of either only rickets or only CML, the radiologists had the option of interpreting each set of radiographs as normal or as showing rickets, CML, or both rickets and CML. The radiologists rated their confidence in their interpretation using a 3-point scale (low, moderate, or high confidence). If recording a diagnosis of rickets, the radiologists also assessed radiographs for symmetry, loss of zone of provisional calcification, cupping, fraying, and widening of the physis. If recording a diagnosis of CML, the radiologists also assessed radio-graphs for corner fracture, corner deformity, bucket-handle fracture (in any stage of healing, including presence of thin, thick, or endochondral density filling the gap) [
29], subphyseal lucency, deformed corner, metaphyseal irregularity, and subperiosteal new bone formation. Criteria for these signs are shown in Table S1 (available in the
online supplement). If recording that both CML and rickets were present, the radiologist recorded the presence of findings associated with either diagnosis. Before the interpretations, the radiologists were provided with an educational pdf containing examples of the radiographic signs.
Statistical Analysis
The statistical analysis is described in Supplemental Methods.
Results
Patient Sample
Across sites, the initial search identified 121 children with possible rickets and knee radiographs. Of these, children were excluded for the following reasons: vitamin D level unavailable (
n = 13), vitamin D level measured more than 90 days from the date of knee radiograph (
n = 1), vitamin D level 20 ng/mL or greater (
n = 13), RSS of 0 (
n = 6), prematurity (
n = 1), and inability to successfully edit knee radiographs (
n = 17). Across sites, the initial search identified 188 children with CML in the distal femur and/or proximal tibia. Of these, 111 children were excluded for the following reasons: vitamin D level unavailable (
n = 83), vitamin D level measured more than 90 days from date of skeletal survey (
n = 5), myelomeningocele (
n = 1), prematurity (
n = 6), negative or indeterminate diagnosis of child abuse by the local child abuse pediatrician (
n = 6), and inability to successfully edit knee radiographs (
n = 10). After these exclusions, the final study groups included 70 children with rickets and 77 children with child abuse who had knee CML.
Figure 2 summarizes the selection process. The number of children from each institution in the final sample is summarized in Table S2 (available in the
online supplement).
The children with rickets (27 girls, 43 boys) had a mean vitamin D level of 8.7 ng/mL (range, 2.8–18.6 ng/mL). The radiographs selected for evaluation were the initial radiographs in 66 cases and follow-up radiographs in four cases. Among the children with CML (37 girls, 40 boys), 53 (68.8%; 28 girls, 25 boys) had normal vitamin D levels (mean, 40.2 ng/mL; range, 30–132 ng/mL). The initial radiographs were selected in 47 cases and follow-up radiographs in six. Twenty-four of the 77 (31.2%) children with CML (nine girls, 15 boys) had low vitamin D levels (mean, 21.5 ng/mL; range, 7–29 ng/mL; 20–29 ng/mL in 16; < 20 ng/mL in eight). The initial radiographs were selected in 23 of these cases and follow-up radiographs in one case.
Children with CML were significantly (
p < .001) younger (mean, 3.7 months; range 0.7–18.3 months) than those with rickets (mean, 14.2 months; range 2.8–23.8 months) (
Table 1). Among the 77 children with CML, 69 (89.6%) were younger than 6 months, and three (3.9%) were older than 1 year. In comparison, among the 70 children with rickets, four (5.7%) were younger than 6 months, and 46 (65.7%) were older than 1 year. Among the 24 children with CML and low vitamin D levels, 23 (95.8%) were younger than 6 months.
The frequencies of abnormal laboratory values in children with rickets, CML with normal vitamin D levels, and CML with low vitamin D levels were 96.9%, 0.0%, and 38.1% for alkaline phosphatase; 74.6%, 0.0%, and 19.0% for phosphorus; 43.9%, 3.8%, and 20.0% for calcium; and 89.1%, 33.3%, and 33.3% for parathyroid hormone (
Table 2). These percentages were significantly different between the two CML groups and the rickets groups for all laboratory values and between the two CML groups for all laboratory values other than parathyroid hormone (all
p < .05). The mean interval from recording of the vitamin D levels to knee radiography was 6.6 days (range, 0–70 days) among children with rickets and 5.8 days (range, 0–81 days) among children with CML (
p = .35).
Injuries in Children With Classic Metaphyseal Lesions
Injuries in addition to knee CML are summarized in
Table 3. All children with CML had other signs of traumatic injuries at physical examination or imaging. Sixty of the 77 (77.9%) children had signs of injuries at physical examination, most commonly skin bruising (44/77 [57.1%]). Seventy-one of 77 (92.2%) children had fractures other than the knee CML; 51 (66.2%) had more than four other fractures. The mean number of other fractures was eight (range, 1–32 fractures). Twenty-four of the 77 (31.2%) children had posterior rib fractures, and 52 (67.5%) had CML in locations other than the knees. Eighteen of the 77 (23.4%) children had abusive head trauma, most commonly (15/16 [93.8%]) subdural hematoma. Other findings included hypoxic ischemic changes (
n = 5), brain contusion (
n = 2), and punctate parenchymal hemosiderin deposition (
n = 1).
Among the 24 children with CML and low vitamin D levels, 11 (45.8%) had skin bruising and five (20.8%) had abusive head trauma (all subdural hematoma). Twenty-two of the 24 (91.7%) had fractures in addition to knee CML (mean, 10.8 fractures per child), and 19 (79.2%) had more than four other fractures. Seventeen of the 24 (70.8%) children had posterior rib fractures.
Rib fractures were significantly (p = .02) more common in children with CML and low vitamin D levels (17/24 [70.8%]) than in children with CML and normal vitamin D levels (22/53 [41.5%]). The frequencies of injuries at physical evaluation, of abusive head injuries, and of fractures other than rib fractures were not significantly different between children with CML with normal and those with low vitamin D levels (all p > .05).
Interobserver Agreement
Table S3 (available in the
online supplement) summarizes agreement on interpretations at varying confidence levels and in various radiologist subsets. Agreement among radiologists was substantial to almost perfect for interpretations at any confidence level for rickets (κ = 0.84 [95% CI, 0.81–0.86]; absolute agreement, 91.9% [95% CI, 90.8–92.9%]) and for CML (κ = 0.77 [95% CI, 0.74–0.80]; absolute agreement, 88.5% [95% CI, 87.1–89.8%]). Agreement among radiologists was higher in terms of interpretations with moderate and high confidence for rickets (κ = 0.92 [95% CI, 0.91–0.94]; absolute agreement, 96.4% [95% CI, 95.7–97.0%]) and for CML (κ = 0.89 [95% CI, 0.87–0.92]; absolute agreement, 95.0% [95% CI, 93.6–96.4%]). Agreement among subgroups of nonpediatric, pediatric, less experienced, and more experienced radiologists was substantial to almost perfect.
Interpretations With Any Confidence
Table S4 (available in the
online supplement) summarizes sensitivity, specificity, PPV, NPV, and accuracy for each radiologist for interpretations at any confidence level. For the diagnosis of rickets, the model-based estimates combining data from all readers yielded sensitivity of 95.6% (95% CI, 89.8–98.1%), specificity of 91.3% (95% CI, 83.4–95.7%), PPV of 92.5% (95% CI, 82.2–97.1%), and NPV of 86.8% (95% CI, 74.2–93.7%). For the diagnosis of CML, model-based estimates yielded sensitivity of 91.3% (95% CI, 83.4–95.7%), specificity of 95.6% (95% CI, 89.8–98.1%), PPV of 97.8% (95% CI, 52.4–99.9%), and NPV of 82.0% (95% CI, 67.8–90.8%). Model-based estimated accuracy for the diagnosis of rickets or CML was 93.1% (range, 87.7–96.2%) and was not significantly different between pediatric and nonpediatric radiologists (
p = .37) or between less experienced and more experienced radiologists (
p = .86).
Interpretations With Moderate or High Confidence
Table S5 (available in the
online supplement) summarizes sensitivity, specificity, PPV, NPV, and accuracy for each radiologist for interpretations with moderate or high confidence. For a diagnosis of rickets, the model-based estimates combining data from all readers yielded sensitivity of 97.0% (95% CI, 92.7–98.8%), specificity of 95.1% (95% CI, 87.4–98.2%), PPV of 98.3% (95% CI, 96.1–99.3%), and NPV of 92.6% (95% CI, 80.3–97.5%). Three of 494 (0.6%) interpretations of CML with moderate or high confidence were false-positive, each by a single radiologist, and two of the children were older than 1 year. For a diagnosis of CML, model-based estimates yielded sensitivity of 95.1% (95% CI, 87.4–98.2%), specificity of 97.0% (95% CI, 92.7–98.8%), PPV of 99.3% (95% CI, 97.1–99.8%), and NPV of 90.0% (95% CI, 80.6–95.1%). Eight of 504 (1.6%) interpretations of rickets with moderate or high confidence were false-positive, in two children each by a single radiologist and in three children each by two radiologists.
Model-based estimated accuracy for the diagnosis of rickets or CML with moderate or high confidence was 96.0% (95% CI, 91.6–98.1%) and was not significantly different between pediatric and nonpediatric radiologists (p = .20) or between less and more experienced radiologists (p = .57). Across radiologists, the interpretation of combined CML and rickets was made with moderate or high confidence in a mean of 1.4 children (range, 0–4 children). However, only one radiologist made the interpretation of combined CML and rickets with moderate or high confidence in a child with CML and a low vitamin D level.
Figures 1A and
1B show a representative child with rickets, and
Figures 1C and
1D show a representative child with CML. In both cases, all eight radiologists made correct interpretations with moderate or high confidence.
Figure 3 shows false-positive interpretations of CML, and Figure S1 (available in the
online supplement), false-positive interpretations of rickets.
Radiographic Signs of Rickets and Classic Metaphyseal Lesions
Agreement on radiographic signs identified with any confidence (Table S6, available in the
online supplement) was moderate to almost perfect for symmetry (κ = 0.65 [95% CI, 0.61–0.68]), loss of metaphyseal zone of provisional calcification (κ = 0.83 [95% CI, 0.80–0.86]), cupping (κ = 0.57 [95% CI, 0.52–0.63]), fraying (κ = 0.84 [95% CI, 0.81–0.86]), widening of the physis (κ = 0.65 [95% CI, 0.60–0.69]), corner fracture (κ = 0.50 [95% CI, 0.44–0.56]), and subperiosteal new bone formation (κ = 0.75 [95% CI, 0.70–0.80]) and fair for subphyseal lucency (κ = 0.28 [95% CI, 0.24–0.33]), deformed corner (κ = 0.26 [95% CI, 0.20–0.33]), and metaphyseal irregularity (κ = 0.30 [95% CI, 0.25–0.35]). In comparison with agreement on radiographic signs with any confidence, agreement was higher for radiographic signs identified with moderate or high confidence (Table S7, available in the
online supplement). Agreement on the presence of radiographic signs among subsets of nonpediatric, pediatric, less experienced, and more experienced radiologists is summarized in Tables S6 and S7.
For radiographic signs identified with any confidence, loss of metaphyseal zone of provisional calcification, cupping, fraying, and widening of the physis were significantly more frequent for rickets than for CML (all
p < .001), being detected in less than 4% of children with CML (
Table 4). Corner fracture, bucket-handle fracture, subphyseal lucency, deformed corner, metaphyseal irregularity, and subperiosteal new bone formation were significantly more frequent in children with CML than in those with rickets (all
p < .001), being detected in less than 4% of children with rickets (
Table 4). No radiographic signs exhibited a significant difference in frequency between children with CML and normal vitamin D and those with low levels (all
p > .05) (
Table 5).
A multivariable analysis classification tree of radiographic signs identified with any confidence did not yield a simplified combination of signs that had a highest rate of correct interpretations. However, when the classification tree was applied to radiographic signs identified with moderate or high confidence, a simplified combination of either loss of metaphyseal zone of provisional calcification or presence of fraying versus the absence of both signs yielded the highest rate of correct interpretations. When data from all readers were pooled, the use of this combination of signs resulted in correct interpretations in 97.4% (491/504) of cases of rickets and 98.0% (484/494) of cases of CML.
Discussion
To our knowledge, this is the largest study evaluating radiologists' interpretations of radiographic knee abnormalities in children younger than 2 years with nutritional rickets (n = 70) or CML (n = 77). Radiologists rendered interpretations of rickets and CML with substantial to almost perfect agreement. Across radiologists, the model-based estimated sensitivity and specificity for CML were 91.3% and 95.6% and for rickets were 95.6% and 91.3% with accuracy of 93.1% for both diagnoses. Diagnostic performance improved when moderate or high confidence was required for interpretation; model-based estimated sensitivity and specificity were 95.1% and 97.0% for CML and 97.0% and 95.1% for rickets with accuracy of 96.0% for both diagnoses. Furthermore, the frequency of false-positive moderate- or high-confidence interpretations of CML in children with rickets was only 0.6%. These findings counter claims that radiologists cannot reliably differentiate rickets and CML and commonly misinterpret rachitic changes as representing CML.
Age was significantly different between children with rickets and those with CML; 86.8% of children with CML were younger than 6 months, whereas only 3.9% of children with CML were older than 1 year. In comparison, 5.7% of children with rickets were younger than 6 months. This low frequency of rickets in children younger than 6 months argues against a potential view that CML in infants within this age range are a consequence of nutritional rickets [
16].
The results of this study support the existence of previously reported distinct radiographic signs of rickets and CML [
25]. Various recognized radiographic signs of rickets (loss of metaphyseal zone of provisional calcification, fraying, cupping, widening of the physis) were significantly more common in children with rickets and were uncommon (≤ 4%) in children with CML. Similarly, various radiographic signs of CML that are specific for child abuse (corner and bucket-handle fractures) and signs that indicate fracture healing (subphyseal lucency, deformed corner, metaphyseal irregularity, and subperiosteal new bone formation) were significantly more common in children with CML and were uncommon (≤ 4%) in children with rickets. Agreement was moderate to almost perfect for all assessed radiographic signs, except for fair agreement for subphyseal lucency, deformed corner, and metaphyseal irregularity, which are considered less specific signs of CML. The frequencies of the radiographic signs of CML were not significantly different between children with CML with normal and those with low vitamin D levels. Use of a multivariable analysis classification tree for radiographic signs identified with moderate or high confidence revealed that either loss of metaphyseal zone of provisional calcification or presence of fraying yielded the highest frequency of correct interpretations. Loss of metaphyseal zone of provisional calcification has been considered among the most distinctive findings of rickets [
25].
A total of 31.2% of children with CML had low vitamin D levels (either insufficiency or deficiency). This result is not surprising in light of results of an epidemiologic study of vitamin D insufficiency and deficiency in infants and toddlers [
22]. Nonetheless, the frequency of injuries found at physical examination or of abusive head trauma was not significantly different between the two groups of children with CML. Except for rib fractures, which were significantly more common in children with CML and low vitamin D levels, frequencies of other fractures did not vary significantly between the two groups of children with CML. Alkaline phosphatase, phosphate, calcium, and parathyroid hormone levels were also not significantly different between these groups. In comparison, these laboratory levels were all significantly higher in children with rickets than in those with CML.
Our findings do not indicate the presence of any reliable radio-graphic signs of low vitamin D levels in children with CML. An interpretation of a combination of CML and rickets with moderate or high confidence was provided by only one radiologist in only one of 24 children with CML and low vitamin D levels. We advise that radiologists should be cautious when interpreting radio-graphs as showing both rickets and CML. If the presence of both diagnoses is suspected on radiographs in the absence of other evidence to support these diagnoses, it may be appropriate to render an indeterminate interpretation rather than to interpret the study as showing both diagnoses.
Some have argued that in children with low vitamin D levels, CML are related to rickets and not to abuse [
15,
16]. However, the characteristic radiographic signs of vitamin D insufficiency or deficiency and of CML are not the same [
30]. A prior study showed similar frequencies of low vitamin D levels in abused children and in nonabused children with accidental trauma [
31]. In addition, a study of fractures in children younger than 2 years with rickets showed that all fractures occurred in mobile children at least 8 months old with obvious radiographic features of rickets, including widespread metaphyseal fraying or cupping [
24]. In that study, none of these children had more than four fractures, and none of the fractures or clinical findings were typical of abuse. Fractures in these children included transverse long bone fractures and lateral and anterolateral rib fractures. In comparison, all children in our study who had CML and low vitamin D levels were younger than 6 months, all had evidence of other injuries, 70.8% had posterior rib fractures, and 79.2% had more than four fractures.
Our study had limitations. Pathologic analysis was unavailable as the reference standard for diagnoses of rickets and CML. These diagnoses were based on clinical evaluations including history, examination, laboratory data, imaging findings, and reports by local child abuse pediatricians. It is possible that the presence of knee CML influenced pediatricians' child abuse diagnoses. However, all children with CML had other injuries, suggesting that the diagnoses were not solely based on the presence of knee CML. In addition, we did not include normal knee radiographs, which could have lowered the observed diagnostic performances. We also evaluated knee radiographs only. In practice, patients with rickets commonly also undergo wrist radiography, and children evaluated for child abuse undergo a radiographic series and, in many cases, neuroimaging. Inclusion of radiographs of other regions, and possibly of clinical and laboratory information, may have improved radiologists' performance in differentiating rickets and CML. There was also potential selection bias related to inclusion only of children with CML who had available vitamin D levels. Finally, we cannot completely exclude the possibility that some children with rickets had also been abused.
In conclusion, less and more experienced pediatric and non-pediatric radiologists had high diagnostic performance in differentiating rickets and CML, regardless of the presence of vitamin D deficiency, with few false-positive interpretations for these diagnoses. The presence of radiographic signs of both rickets and CML should result in an indeterminate interpretation. The knowledge that CML mostly occur in children younger than 6 months and are unusual in children older than 1 year may assist interpretations. The results of this multicenter study confirm the presence of distinctive radiographic signs of rickets and CML and show that radiologists can reliably differentiate these two entities.
Acknowledgments
We thank Wendy Territo for editing the radiographs of the knees and Meenakshisundaram (Sundar) Paramasivam for writing a script for editing DICOM tags to optimize review by the radiologists.