After-hour trauma-radiograph interpretation in the emergency centre of a District Hospital

Introduction Plain radiographs remain a first-line trauma investigation. Most trauma radiographs worldwide are reported by junior doctors. This study assesses the accuracy of after-hour acute trauma radiograph reporting by emergency centre (EC) doctors in an African district hospital. Methods An institutional review board approved retrospective descriptive study over two consecutive weekends in February 2020. The radiologist report on the admission radiographs of adult trauma patients was compared with the initial EC interpretation. The accuracy, sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV) for EC interpretation were calculated with 95% confidence intervals (95%CI). The association between reporting accuracy and anatomical region, mechanism of injury, time of investigation, and the number of abnormalities per radiograph was assessed. Results 140 radiographs were included, of which 49 (35%) were abnormal. EC doctors recorded (95%CI) 77% (69-84%) accuracy, 38% (25-54%) sensitivity, 97% (91-99%) specificity, 86% (65-95%) PPV and 76% (71-80%) NPV. Performance was associated with the anatomical region (p=0.02), mechanism of injury (p=<0.01) time of day (p=0.04) and the number of abnormalities on the film (p=<0.01). The highest sensitivity was achieved in reports of the appendicular skeleton (42%) and in the setting of simple blunt trauma (62%). Overall accuracy was in line with the range (44%-99%) reported in the international literature. Discussion Accurate reporting of acute trauma radiographs is challenging. Key factors impact performance. Further training of junior doctors in this area of clinical practice is recommended. Future work should focus on assessing the impact of such training on reporting performance.


Introduction
Due to their affordability, availability, and relatively low ionizing radiation exposure, plain radiographs remain a crucial first-line trauma investigation, particularly in resource-limited settings. For the past fifty years, the global radiological workload has increased at a faster rate than the supply of radiologists, compounded by the introduction of new imaging techniques, such as computed tomography, magnetic resonance imaging and fluoroscopically guided interventions. There has thus been an increasing global shortage of radiologists, with most trauma radiographs worldwide being reported by junior, inexperienced doctors [1][2][3][4] .
Trauma care creates a "perfect storm " for medical errors, with high workload, unstable patients, incomplete histories, time-critical decisions, multi-tasking, interdisciplinary involvement, and junior staff working outside normal hours [5] . It is therefore inevitable that errors will be made and that some diagnoses missed [1] . Failure to detect fractures is the most common error in Accident and Emergency Units [6] . Approximately 10% of medical litigation cases in high-income countries involve imaging, of which nearly 80% relate to trauma [2] . The morbidity and mortality associated with missed fractures are well documented [7][8][9][10] .
The accuracy of trauma radiographs reported by junior doctors has been extensively researched in the last half century [ 4 , 11-23 ]. Previous studies have largely been performed in well-resourced environments and have included all levels of healthcare facilities. A variety of research methods have been utilized, yielding a wide range (32% -99%) of reporting accuracy for junior colleagues.
Despite most injuries worldwide occurring in low-and middleincome countries [24] , there has been very limited work on the accuracy of trauma radiograph reporting in such settings. The two studies to date have assessed fracture detection by junior South African (SA) doctors working in tertiary hospitals and have recorded accuracies of 68% [25] and 89% [26] , respectively. However, there has been no study in a district hospital setting in Africa.
Such a study is important, since a review of radiological resources in Southern, East, and West Africa reveals a broad network of district hospitals with plain radiographic equipment [27][28][29][30] . Furthermore, the District Health System (DHS) [25] has been adopted as the vehicle for delivery of comprehensive primary health care in many African countries.
The aim of this study was an audit of the accuracy of after-hour acute trauma-radiograph reporting by doctors in the emergency centre (EC) of a district hospital in Africa.

Methods
This retrospective descriptive study was conducted at a large SA district hospital, with more than 300 beds and a busy EC, treating more than 3000 patients monthly.
The EC is staffed by physicians with a wide range of experience, including interns, post-internship "community service " doctors, medical officers, registrars (trainee specialists), and qualified specialists. Each patient's initial management is based on the individual assessment of a junior EC doctor, who requests and interprets first-line investigations, including radiographs. Imaging requests are at the sole discretion of the examining physician and are informed by clinical findings and the mechanism of trauma. There is subsequent formal specialist-review of patient management.
The hospital has a fully digital imaging platform, with a filmless and paperless radiology department, but no after-hour radiologist reporting service. An electronic timestamp is generated for each investigation. All studies are available for clinical viewing at any hospital workstation immediately after completion. All images are stored on the institutional picture archiving and communication system (PACS), which thus represents a comprehensive imaging database. In addition, the full medical record for each patient is available digitally on the electronic medical record.
The study was conducted over two consecutive pre-COVID weekends in February 2020. For study purposes, a weekend extended from 16h00 Friday -08h00 Monday; day shift was from 08h00 -20h00, and night shift from 20h00 -08h00 the following day.
The admission radiographs of trauma patients aged 18 years or older, who were managed in the EC over the study weekends, were included. All follow-up and non-trauma radiographs, those performed on down-referred patients, and any examination on patients less than 18 years of age, were excluded.
Any injury caused by a gunshot or stab with a knife, broken bottle, or other sharp object was deemed penetrating. An injury was classified as blunt if due to a punch, kick, or blow with any blunt object such as a rock, plank, or concrete slab. Traffic-related injuries, characterised by their high-velocity and propensity for polytrauma, included motor vehicle accidents (MVA) and pedestrian vehicle accidents (PVA). Community assault, a distinct mechanism of injury specified in the referral details, involves polytrauma induced by whipping and direct blows with multiple different blunt objects. It infrequently involves penetrating injuries.
There was independent reporting of all study radiographs by two radiologists blinded to clinical details: a radiology consultant with 11 years' experience and a senior radiology registrar with 5 years' experience.
Images with technical limitations such as under-penetration, overpenetration, or inappropriate positioning were excluded.
Radiographs were assessed for fractures, dislocations, joint effusions, pneumomediastinum, pneumothorax, haemothorax, pulmonary opacification, widened mediastinum and pneumoperitoneum. All abnormal findings were subjectively stratified as "easily identified " or "subtle ". Results were captured on a customized, forced-choice reporting template. Discrepancies between radiologists were resolved by consensus.
Details of patients' age, gender, mechanism of injury (MOI), time of examination, and injuries detected by EC doctors were subsequently included on the spreadsheet.
The axial skeleton included the skull, facial bones, and the spine; the appendicular skeleton comprised the upper/lower limbs and pelvis. Chest and abdominal radiographs were assessed separately.
The study endpoint was the overall accuracy of initial EC doctor radiographic interpretation, with the radiologist report considered the "gold-standard ". For each radiograph, the final radiologist report was compared with the initial EC interpretation. Injuries were deemed detected if documented in the EC clerking notes or included in clinical details on subsequent imaging requests. Conversely, an undetected injury was neither specified in the initial EC notes nor on any additional imaging request.
The overall accuracy, sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV) for initial EC radiographic interpretation were calculated with 95% confidence intervals. The Fischer's Exact Test assessed the association between reporting accuracy and anatomical region, time of investigation, mechanism of injury and the number of abnormalities per radiograph. Undetected injuries were classified as "clinically significant " or "not clinically significant ", the former if non-detection impacted clinical management and potentially influenced patient outcome.
This study was approved by the Health Research Ethics Committee of Stellenbosch University and the National Health Research Database.
Reporting sensitivity and specificity tended to be associated with the site of injury. Performance was best for abnormalities of the appendicular skeleton, decreasing sequentially for the chest, axial skeleton and abdomen (p = 0.02).
Similarly, performance tended to be associated with the mechanism of injury, with blunt trauma achieving the highest, and community assault the lowest sensitivity and specificity, respectively (p < 0.01).
Night shift reporting tended to be more accurate than day shift (p = 0.04). Accuracy was also associated with the number of abnormalities on a single film. Images demonstrating one, two and three-or-more abnormalities were interpreted correctly in 48%, 26% and 0% of cases, respectively (p < 0.01).
False positives: There were six false positive findings: three fractures (right mandibular body, humeral shaft and metacarpal), two pneumothoraces and one haemothorax.
The missed fractures were not clinically significant since management was not impacted. In the case of the false positive right mandibular fracture, there was a true-positive contralateral mandibular fracture that required definitive CT imaging, which demonstrated a normal right mandible. The missed humeral fracture occurred in conjunction with a true-positive ulna fracture, which was appropriately managed with a back-slab and follow-up imaging. The metacarpal fracture was also managed conservatively.
One false positive pneumothorax was not clinically significant, as the patient had an ipsilateral haemothorax for which an intercostal drain was placed.
One false positive pneumothorax and the haemothorax were clinically significant as they were managed with intercostal drainage.
The 6 undetected abnormalities deemed insignificant included 2 haemothoraces with correctly identified pneumothoraces for which intercostal drains were appropriately sited, 2 unidentified pulmonary lacerations, and 2 undetected rib fractures required only conservative management.
One-third of spurious findings ( n = 2/6, 33%) were clinically significant, involving a single patient and resulting in unnecessary placement of an intercostal drain.

Discussion
To the best of our knowledge, this work represents the most detailed analysis of acute trauma radiograph reporting by junior doctors in any setting. It thus makes an important contribution to the broad body of existing literature on this subject. It is also the first study in the setting   Fig. 2. False negative. Note: Galeazzi fracture-dislocation of the right forearm, with fracture of the distal third of the radius and disruption of the distal radioulnar joint. The radial fracture was detected (solid white arrow), but the radio-ulnar dislocation (dashed white arrow) and ulnar styloid fracture (white arrow head) were missed. Deemed 'easily-detectable, clinically significant ". of the African district hospital, providing important new insights into an under-researched area of traumatology and emergency medicine. It documents key baseline data in this context. Firstly, our cohort had a higher proportion of abnormal radiographs (35%) than reported (8% -25%) [ 14 , 17 , 18 , 31 ] in other settings. Williams [22] and Vincent [18] have stressed the importance of this parameter when assessing reporting accuracy, noting its omission in most published work. They caution that studies involving a high proportion of normal radiographs may be biased towards improved outcomes if false negatives were simply recorded as a percentage of the total examinations. Our finding that normal radiographs were reported with almost complete accuracy (90/91, 99%) supports this contention. Furthermore, it has been suggested [31] that high proportions of normal trauma radiographs reflect inappropriate imaging requests and warrant more stringent referral criteria. Although the optimum ratio of normal to abnormal trauma radiographs is moot, the percentage of abnormal radiographs in our study arguably reflects conservative utilization of imaging in a resource constrained environment.
Secondly, almost half (22/49, 45%) the abnormal radiographs showed more than one pathological finding, a detail not routinely included in the body of literature. Our finding that reporting accuracy decreases with the number of abnormalities on a film (p < 0.01) is intuitive but highlights its importance. The challenge of making a correct and complete diagnosis in the multiply injured has been previously documented [ 32 , 33 ]. However, the so-called "satisfaction of search " [34] , whereby pathology on the same radiograph remains undetected after identification of an initial abnormality, is a well-recognised radiological pitfall that has not been highlighted in the EC literature to date.
Thirdly, approximately one-in-five (15/79, 20%) radiographic abnormalities were considered "subtle ", and thus required substantial radiological training and experience for identification. The importance of this detail, which is mostly not reported, has been stressed by Tachakra [16] and borne out by our finding that only 13% of "subtle " abnormalities were identified, compared to 69% of those deemed "easily identifiable ".
Fourthly, almost three-quarters of reporting errors (27/37, 73%) were considered clinically significant, which is substantially higher than the 6% -44% reported in previous studies [ 14 , 17 , 18 , 22 , 26 ]. This may reflect the complexity of trauma in our environment. South Africa is recognised as one of the most violent countries in the world, with an exceptionally high prevalence of interpersonal violence, as evidenced by almost sixty percent of our cohort being victims of penetrating injury or community assault [35] .
The finding that reporting accuracy is associated with anatomical region (p = 0.02), type of trauma (p = < 0.01) and time of day (p = 0.04) is important. It suggests caution should be exercised when comparing performance across studies while highlighting key variables that may impact outcomes. Our study nonetheless provides useful baseline data in the African context.
We have shown that injuries to the appendicular skeleton and chest are most common in our setting, together accounting for almost 90% of all trauma radiographs. We have also shown that reporting sensitivity is higher for these anatomical regions than for those less commonly injured (axial skeleton and abdomen). It could be that the sheer volume of appendicular skeleton and chest trauma hones reporting skills.
Additionally, we have shown that radiographs resulting from the more conventional blunt trauma tend to be reported with greater sensitivity than those from more complex mechanisms of injury, such as penetrating trauma, motor vehicle accidents and community assault. In their classic manuscript, Tachakra and Becket [16] defined four scenarios accounting for radiological errors by junior doctors in the accident and emergency unit. The first was insufficient care when reviewing the radiograph, with missed fractures at the site of maximum tenderness being a prime example. The second was unexpected radiological findings, such as fractures distant from the site of maximum tenderness. The third was radiological inexperience, whereby abnormalities were either identified, but considered normal variants, or not identified due to their subtlety. The fourth scenario was radiographic interpretation without clinical examination. The risks inherent at the time of staff changeover were highlighted. The scenario of staff at the start of a shift reviewing the radiographs of patients for whom they had assumed responsibility, but had not examined personally, was noted as particularly high risk. It is unclear to what extent this scenario contributed to the dis-parity in reporting accuracy between the day and night shift in our study.
The finding that night shift was more accurate than day shift reporting is certainly counter-intuitive. The explanation is not immediately apparent. The determinants of sensitivity in trauma reporting remain poorly understood and will require evaluation in future work. However, review of Table 1 will show that there was a higher proportion of normal studies performed in the night shift (52/75, 69%) than the day shift (39/65, 60%). This may have contributed to this finding.
Notwithstanding the limitations of comparing studies of radiological reporting performance, the 77% overall accuracy achieved by EC staff in this study is well within the 44% -99% range [ref] documented in the international literature for non-radiologist reporting of conventional trauma radiographs in the clinical environment [ 13 , 22 , 36 ].
A major strength of this study was its foundation on two robust digital databases, one being an electronic medical record and the other a picture archiving and communication system. This allowed comprehensive retrieval of imaging and clinical data. In addition, the study was designed to facilitate the most detailed analysis to date of overall accuracy of initial EC doctor radiographic interpretation. As such it reported parameters not included in previous manuscripts, such as sensitivity, specificity, PPV, NPV. It also included parameters inconsistently documented in the literature, such as the proportion of radiographs with any abnormality, more than one abnormality, and subtle abnormality. It also assessed the association between reporting accuracy and anatomical region, mechanism of injury and the time of day. Future studies in this domain could utilize this methodology.
The study was limited by retrospective data acquisition and a relatively small sample size. Additionally, it did not assess final patient management or outcomes. It is thus possible that radiological errors by junior doctors were corrected during subsequent consultant ward rounds or consultations.
This study has identified key factors that impact the accuracy of trauma radiograph reporting in our setting. It also underscores the challenge of such work and highlights the need for further training of junior doctors in this area of clinical practice. Future work in this domain should focus on assessing the impact of such training on the reporting accuracy.

Dissemination of Results
The results will be presented to the local health facilities and the local health authorities as well as the education committee involved in the training of medical students. The findings of this research will be presented at national and international emergency medicine conferences.

Author Contributions
Authors contributed as follow to the conception or design of the work; the acquisition, analysis, or interpretation of data for the work; and drafting the work or revising it critically for important intellectual content: YML contributed 50%; RDP 25%; SO 15%; FCH and SL contributed 5% each. All authors approved the version to be published and agreed to be accountable for all aspects of the work.

Declaration of Competing Interest
The authors declare no conflicts of interest.