Chest radiographs versus CT for the detection of rib fractures in children (DRIFT): a diagnostic accuracy observational study

Abstract

Background: Internationally, chest radiography is the standard investigation for identifying rib fractures in suspected physical abuse in infants. Several small observation studies in children have found that chest CT can provide greater accuracy than radiography for fracture detection, potentially aiding medicolegal proceedings in abuse cases; however, to our knowledge, this greater accuracy has not been comprehensively evaluated. We aimed to determine differences in rib fracture detection rates between post-mortem chest radiographs and chest CT images, using forensic autopsy as the reference standard.

Methods: In this retrospective diagnostic accuracy study, we searched the Great Ormond Street Hospital (London, UK) radiology information system for all children aged 0-16 years who had a post-mortem skeletal survey (ie, full-body radiography), CT, and full autopsy between Jan 1, 2012, and Jan 1, 2017, for a purpose of death investigation. Cases were excluded if the imaging was done for a reason other than a forensic investigation or if image quality was suboptimal. Radiologists were recruited as reporters on a voluntary basis via membership databases from international radiology and post-mortem imaging societies with no specific inclusion or exclusion criteria. Reporters were sent a set of chest radiographs on a password protected and encrypted USB flash drive or via a secure filesharing website and independently reported on the presence of rib fractures, fracture location, and the confidence level of their interpretation. They were masked to the clinical information of the images. 1 month later, the same reporters were sent CTs for the same cases in a random order and asked to report on the same features. The primary objective was to compare the accuracy of detection of rib fractures by use of post-mortem chest radiographs and CTs, with autopsy data as reference standard. Accuracy was assessed by comparison of diagnostic statistics, calculated using random-intercept multilevel logistic models with reporter and patient included as cross-classified random-effects.

Findings: 25 cases of children (aged 1 month to 7 years), with 136 rib fractures at autopsy with paired post-mortem chest radiographs and CTs, were selected for analysis. 38 radiologists were recruited as reporters from 23 international centres; 12 (32%) were consultants, median experience of 14·5 years (range 6-27), and 26 (68%) were registrars, median experience of 4 years (range 2-9). Across all radiologists, three times as many rib fractures were correctly detected by use of chest CTs compared with chest radiography (sensitivity 44·9% [95% CI 31·7-58·9] vs 13·5% [8·1-21·5]; difference 31·4% [23·3-37·8; p<0·001]). Sensitivity for detection on the correct rib was higher by use of CT than by use of radiography (62·4% [95% CI 44·9-77·1] vs 23·1% [12·9-37·8]; difference 39·3% [31·9-42·2; p<0·001]), as was diagnosis of a patient with any rib fracture or fractures (81·5% [75·8-86·0] vs 64·7% [57·3-71·4]; difference 16·7% [11·5-22·2; p<0·001]). Radiologist confidence was higher when using CT images than radiographs (highest confidence rating given on 3317 [63·6%] of 5218 fractures for CT vs 1518 [46·6%] of 3303 on radiographs) and was a predictor for accurate fracture detection.

Interpretation: Chest CT provides greater accuracy than conventional chest radiography for post-mortem rib fracture detection, irrespective of radiologist experience or fracture location, although both methods detected a substantial number of false positives. The diagnostic accuracy of CT should be studied further in live children ideally in a multicentre trial to assess the applicability of our results.

Funding: Great Ormond Street Children's Charity, Medical Research Council, Royal College of Radiologists, Research Councils UK, National Institute for Health Research.

Figure 1

Case availability and selection *No whole-body images were available, and hence no ribs could be analysed.

Figure 2

Scatter plots of sensitivity (A) and specificity (B) of CT versus chest radiography, by reporter job title Datapoints show the sensitivity and specificity of each reporter that completed both phases of the study (n=35), and those that only completed the chest radiography analysis (n=38). Estimates were derived from the random-effects of multilevel models. Graph B is on a reduced scale of 85–100%.

Figure 3

Scatter plots of sensitivity versus specificity for chest radiography (A) and CT (B) by reporter job title Datapoints show the sensitivity and specificity of each reporter that completed both phases of the study (n=35), and those that only completed the chest radiography analysis (n=38). Estimates were derived from the random-effects of multilevel models.