Introduction
Paediatric and perinatal autopsy rates have declined over recent decades, leading to post-mortem cross-sectional imaging being proposed as a possible alternative or adjunctive approach [
1,
2]. Post-mortem magnetic resonance imaging (PMMR), when used in conjunction with other ancillary investigations (collectively known as minimally invasive autopsy [MIA]), has recently been shown to have a high diagnostic accuracy rate for cause of death or main diagnosis compared with traditional autopsy in fetuses, stillbirths and children [
3]. Whilst cardiac and neurological diagnoses account for the majority of causes of death in these age groups (12 and 28 % respectively [
3]), less is known about how well thoracic pathology can be diagnosed across a range of likely diagnoses, gestational age and childhood age ranges.
A systematic review of post-mortem imaging [
4] identified several previous studies [
5‐
8] which had attempted to investigate the accuracy of non-cardiac thoracic diagnosis, and reported a relatively good estimated pooled sensitivity of 82 % (95 % confidence intervals 49, 95 %) and specificity of 86 % (58, 96 %) for lung pathology. However, these studies were not fully blinded between reporting radiologists and pathologists, included highly selected small cohort groups (range, 6–37 subjects), with pooled data from only 73 fetuses and no children, giving wide confidence intervals. Breeze et al. [
9] independently reported a 2.5 % sensitivity and 87 % specificity for eight lung lesions in 30 fetuses, when assessed using PMMR, with a reported diagnostic accuracy of 83 %.
The aim of the MARIAS study was specifically to establish the diagnostic utility of less invasive autopsy, using PMMR, compared with traditional autopsy, in an unselected population for the major diagnosis or cause of death. These results have already been published [
3]. Here, we use the detailed MARIAS dataset for sub-analysis of the PMMR findings specifically for non-cardiac thoracic abnormalities, irrespective of whether the abnormalities contributed to the main diagnosis or cause of death.
Discussion
In this large prospective study, PMMR demonstrated poor diagnostic utility for detection of thoracic pathology in fetuses, newborns and children. Whilst the specificity was high at 86 %, due to the large number of normal cases, the sensitivity and positive predictive value were poor at 40 and 54 % respectively. Overall diagnostic accuracy was reasonable at 72 %. PMMR was most sensitive at detecting anatomical abnormalities such as lung or chest hypoplasia and post-mortem pleural effusions, but poor at detecting infection and pulmonary haemorrhage.
PMMR showed a 9.5 % apparent false-positive rate for thoracic imaging, which is in keeping with results from other body systems [
3]. Apparent false positives do not necessarily represent a significant failure of the minimally invasive autopsy model, since the vast majority (75 %) were overcalls of possible infection (consolidation/pneumonia) in children, which would lead to additional investigations and organ sampling for histopathological confirmation. There were several false positives where normal post-mortem accumulation of fluid in bodily cavities was identified but mis-interpreted as pathological pleural fluid, and most of the correct interpretations of lung hypoplasia were recorded in the context of skeletal dysplasia.
Our accuracy data in this large unselected population is lower than those previously published [
5‐
8], with lower sensitivity and overall concordance, but similar specificity. This is likely to be due to the difficulty in diagnosing infection, particularly in children, where such cases were not included in the several previous studies. Breeze et al. [
9] reported a reported a 62.5 % sensitivity (95 % CI = 29.0, 96.0) and 87 % specificity (95 % CI = 78.9, 100), for eight lung lesions over 30 fetuses. Our findings are similar, in that fetal lung hypoplasia was usually easily detected, but was also overcalled. Their series only had two false-positive lung diagnoses (2/30 = 6.7 %), and their PPV was 71.4 % (35.9, 91.8), NPV 87.0 % (67.9, 95.5), and overall concordance 83 % (66.4, 92.7 %), all greater than our data but with much wider confidence intervals. We postulate that lung PMMR is more difficult across a larger cohort of children when reported in a blinded fashion.
Infection (pneumonia or lung parenchymal consolidation) was by far both the commonest abnormality in our study, and also the most difficult to achieve correctly. There are several diagnostic difficulties highlighted by our study, and a small study of 44 children has previously identified similar difficulties in detecting lung parenchymal changes using whole-body post-mortem CT [
15]. In unventilated fetal lungs, infection was missed in the majority of cases, and infection was equally overcalled and missed in the paediatric population, with or without patchy signal changes in the lungs at post-mortem. We conclude that infection or pneumonia is currently difficult to identify at fetal PMMR, and that lung sampling should be performed in all cases where sepsis is suspected or possible. In older children, misinterpreting consolidation as normal post-mortem change, and vice versa, represents a real challenge in PMMR, recognised by other authors [
16]. Several features within the lungs may be interpreted as consolidation (as they would indicate such in life), whereas at autopsy the unequally distributed fluid accumulation within the lung parenchyma is a variant of normal post-mortem changes. We also failed to diagnose eight out of 12 cases of pulmonary haemorrhage, which were interpreted as ‘normal’ post-mortem changes. However, there are variable histological degrees of pulmonary congestion and intra-alveolar fluid in the majority of cases of infant death, regardless of cause. The high apparent false-negative rate (16.2 %; apparent pathologies present at autopsy which were not reported on PMMR) in this study also mainly relates to patchy but diffuse lung parenchymal change, i.e. pneumonia or haemorrhage. The majority of thoracic misses and overcalls in this study were in newborns and children, and mostly related to both overcalling and missing pneumonia.
A normal thoracic scan in fetuses predicts normal pathology in over 80 % of cases, probably because lung aeration does not present a confounding problem in differentiating between other causes of death, such as CDH or pulmonary hypoplasia, but this does not hold true for children. Our current practice is that unless a focal lung lesion is identified on PMMR, then standard lung histopathology should be performed in all childhood cases. It may be possible that in future this could be performed by percutaneous or endoscopic routes [
17], but the accuracy of this approach to detect patchy lung pathology, compared with open sampling, remains undetermined.
Lung abnormalities are notoriously difficult to detect using other types of post-mortem imaging. For instance, lung opacities on post-mortem radiographs have little correlation with histologically diagnostic pneumonia [
18]. Lung PMMR is unlikely to have higher diagnostic accuracy than thoracic MRI in life, and high-resolution computed tomography (CT) remains the mainstay of diagnostic imaging for parenchymal changes, including interstitial lung disease and more subtle pathology. Whilst infant and childhood lung MRI is improving, it is notoriously sequence-dependent and prone to respiratory and cardiac motion artefacts [
19]. The post-mortem state negates several of these issues, but clearly brings new diagnostic challenges.
The main strength of this study is the double-blinded study design and large prospective data collection of unselected cases, which allows categorical reporting of both conventional autopsy and PMMR data, in an independent way. By identifying small pleural effusions, and lung parenchymal changes more accurately, it is likely that correct interpretation of PMMR images will be easier in the future.
A limitation of this study was that these cases were all interpreted by both specialist paediatric radiologists in a tertiary centre and, as such, is unlikely to represent the situation in many other centres. It also highlights both the difficulty in interpreting cases correctly, as these are predominantly interpretation rather than observational errors, and therefore emphasises the need for specialist education in this field in order to maximise the diagnostic yield from both imaging and autopsy. We did not perform post-mortem CT in our cohort, the diagnostic accuracy of which in this setting remains to be evaluated. We also did not measure temperature during PMMR in this study, which may have had an impact on the reporting radiologists’ ability to correctly interpret PMMR images [
20]. We also did not evaluate more advanced MR techniques, such as T2 relaxometry of the lungs or diffusion-weighted imaging. Semi-quantitative measures of water redistribution may be able to differentiate pathological from non-pathological changes in the lungs, but such approaches require further investigation.
PMMR currently has relatively poor diagnostic detection rates for the commonest intra-thoracic pathologies identified at autopsy in fetuses and children, including infection and haemorrhage. The relatively high negative predictive value suggests that normal thoracic appearances at PMMR excludes the majority of important thoracic lesions at autopsy, and so could be useful in the context of minimally invasive autopsy for excluding non-cardiac thoracic abnormalities.
Acknowledgments
The scientific guarantor of this publication is Andrew M. Taylor. The authors of this manuscript declare no relationships with any companies, whose products or services may be related to the subject matter of the article. This study has received funding by the Policy Research Programme in the Department of Health (0550004). The views expressed are not necessarily those of the Department. The study was also supported by grants from the British Heart Foundation (CI/05/010). The MR facility at the UCL Centre for Cardiovascular Imaging was funded by the British Heart Foundation (CI/05/010). None of the funding bodies had any role in analysis of data, results or conclusions of the study. A.M.T. is supported by an NIHR Senior Research fellow award, NJS is supported by an NIHR Senior Investigator award, and O.A. and S.T. are supported by NIHR Clinician Scientist awards. L.S.C., N.J.S. and A.M.T. receive funding from the Great Ormond Street Hospital Children’s Charity. This work was undertaken at GOSH/ICH, UCLH/UCL including support from the NIHR GOSH Biomedical Research Centre. One of the authors has significant statistical expertise. Institutional Review Board approval was obtained. Written informed consent was obtained from all subjects (patients) in this study.
Some study subjects or cohorts have been previously reported by Thayyil et al. [
3]. Methodology: prospective, diagnostic study, multicentre study.
MaRIAS (Magnetic Resonance Imaging Autopsy Study) Collaborative group
Ms. Shea Addison (Research Assistant, UCL), Dr. Michael Ashworth (Consultant in Paediatric Pathology, GOSH) Dr. Alan Bainbridge (MR Physicist, UCL), Dr. Jocelyn Brookes (Consultant in Interventional Radiology, UCH), Prof. Lyn Chitty (Consultant in Fetal Medicine and Genetics, UCL), Dr. WK ‘Kling’ Chong (Consultant in Paediatric Neuroradiology, GOSH), Dr. Andrew Cook (Senior Lecturer in Cardiac Morphology, UCL), Dr. Enrico de Vita (MR Physicist, UCL), Dr. Roxana Gunny (Consultant in Paediatric Neuroradiology, GOSH), Dr. Brian Harding (Consultant in Paediatric Neuropathology, GOSH), Dr. Tom Jacques (Consultant in Paediatric Neuropathology, GOSH), Mr. Rod Jones (Research MR radiographer, UCL), Dr. Mark Lythgoe (Director, Centre for Advanced Biomedical Imaging, UCL), Dr. Marion Malone (Consultant in Paediatric pathology, GOSH), Wendy Norman (Research MR radiographer, UCL), Dr. Oystein Olsen (Consultant in Paediatric Chest and Abdomen Imaging, GOSH), Dr. Cathy Owens (Consultant in Paediatric Chest and Abdomen Imaging, GOSH), Dr. Amaka C. Offiah (Consultant in Paediatric Musculoskeletal Imaging, previously GOSH, currently Sheffield Children’s Hospital), Dr. Nikki Robertson (Reader and Consultant in Neonatology, UCH), Dr. Tony Risdon (Consultant in Paediatric Forensic Pathology, GOSH), Prof. Neil Sebire (Professor of Perinatal and Paediatric Developmental Pathology, GOSH), Dr. Rosemary Scott (Consultant in Perinatal pathology, UCH), Dr. Dawn Saunders (Consultant in Paediatric Neuroradiology, GOSH), Dr. Silvia Schievano (Senior Research Fellow in Medical Engineering, UCL), Ms. Angie Scales (Family liaison sister, GOSH), Prof. Andrew Taylor (Chief Investigator; Professor of Cardiovascular Imaging, UCL), Sudhin Thayyil (Senior Clinical Lecturer and Honorary Consultant in Neonatology, UCL), Angie Wade (Senior Lecturer in Medical Statistics, UCL).