Background

Classic metaphyseal lesions (CMLs) are considered specific for child abuse, but the reliability of radiologists in distinguishing CMLs from metaphyseal variations is unclear.

Objective

To evaluate the diagnostic performance of pediatric and adult emergency radiologists in differentiating CMLs from metaphyseal variations in the knees.

Materials and methods

We retrospectively reviewed distal femur and proximal tibia radiographs in children under 1 year of age who underwent skeletal surveys for suspected child abuse. A consensus diagnosis for CMLs and metaphyseal variations—serving as the ground truth—was established by two pediatric radiologists. The CML group comprised children diagnosed with abuse and confirmed CMLs. The metaphyseal variation group included children not diagnosed with abuse, who exhibited metaphyseal variations and had either no fractures or only an isolated skull fracture. Radiographs were trimmed to exclude other injuries. Four pediatric and four adult radiologists reviewed anonymized studies and categorized each case as CML, metaphyseal variation, normal, or indeterminate, with confidence levels (high, moderate, low). We analyzed diagnoses with moderate or high confidence. Interobserver agreement was assessed using kappa statistics.

Results

There were 44 children with CMLs (10 initial, 7 follow-up, 27 initial and follow-up) and 22 with metaphyseal variations (10 initial, 7 follow-up, 5 initial and follow-up). Metaphyseal fragmentation was the most common variation, identified in 249 of 344 femurs (72.4%, 95% CI 67.3-77.0%) and 60 of 69 tibias (87.0%, 76.7-93.9%). Fragmentations were most frequently located in the posterior or medial metaphysis, or both, in 238 of 249 femurs (95.6%, 92.2-97.8%) and 60 of 69 tibias (87.0%, 76.7-93.9%).

In the CML group, 33 of 114 initial CML diagnoses (28.9%, 20.8-38.2%) were read on follow-up as either metaphyseal variation (n = 17) or normal (n = 16). In contrast, in the metaphyseal variation group, only one follow-up case was diagnosed as a CML; the remainder were diagnosed on follow-up as metaphyseal variation (n = 24).

Diagnostic performance for CML demonstrated high specificity (90.9%, 85.6-94.7%) and positive predictive value (95.6%, 93.0-97.5%), with moderate accuracy (79.3%, 75.9-82.4%), sensitivity (74.9%, 70.8-78.8%), and negative predictive value (57.6%, 51.5-63.5%). Interobserver agreement was substantial, with a mean kappa of 0.61 (range 0.45–0.84).

Conclusion

Radiologists demonstrated substantial agreement and high specificity in distinguishing CMLs from metaphyseal variations. Metaphyseal fragmentation was the most common variation and was uncommonly diagnosed as CML on follow-up.