Heterogeneity in Systematic Reviews of Medical Imaging Diagnostic Test Accuracy Studies: A Systematic Review.

Importance: Systematic reviews of medical imaging diagnostic test accuracy (DTA) studies are affected by between-study heterogeneity due to a range of factors. Failure to appropriately assess the extent and causes of heterogeneity compromises the interpretability of systematic review findings. Objective: To assess how heterogeneity has been examined in medical imaging DTA studies. Evidence Review: The PubMed database was searched for systematic reviews of medical imaging DTA studies that performed a meta-analysis. The search was limited to the 40 journals with highest impact factor in the radiology, nuclear medicine, and medical imaging category in the InCites Journal Citation Reports of 2021 to reach a sample size of 200 to 300 included studies. Descriptive analysis was performed to characterize the imaging modality, target condition, type of meta-analysis model used, strategies for evaluating heterogeneity, and sources of heterogeneity identified. Multivariable logistic regression was performed to assess whether any factors were associated with at least 1 source of heterogeneity being identified in the included meta-analyses. Methodological quality evaluation was not performed. Data analysis occurred from October to December 2022. Findings: A total of 242 meta-analyses involving a median (range) of 987 (119-441 510) patients across a diverse range of disease categories and imaging modalities were included. The extent of heterogeneity was adequately described (ie, whether it was absent, low, moderate, or high) in 220 studies (91%) and was most commonly assessed using the I2 statistic (185 studies [76%]) and forest plots (181 studies [75%]). Heterogeneity was rated as moderate to high in 191 studies (79%). Of all included meta-analyses, 122 (50%) performed subgroup analysis and 87 (36%) performed meta-regression. Of the 242 studies assessed, 189 (78%) included 10 or more primary studies. Of these 189 studies, 60 (32%) did not perform meta-regression or subgroup analysis. Reasons for being unable to investigate sources of heterogeneity included inadequate reporting of primary study characteristics and a low number of included primary studies. Use of meta-regression was associated with identification of at least 1 source of variability (odds ratio, 1.90; 95% CI, 1.11-3.23; P = .02). Conclusions and Relevance: In this systematic review of assessment of heterogeneity in medical imaging DTA meta-analyses, most meta-analyses were impacted by a moderate to high level of heterogeneity, presenting interpretive challenges. These findings suggest that, despite the development and availability of more rigorous statistical models, heterogeneity appeared to be incomplete, inconsistently evaluated, or methodologically questionable in many cases, which lessened the interpretability of the analyses performed; comprehensive heterogeneity assessment should be addressed at the author level by improving personal familiarity with appropriate statistical methodology for assessing heterogeneity and involving biostatisticians and epidemiologists in study design, as well as at the editorial level, by mandating adherence to methodologic standards in primary DTA studies and DTA meta-analyses.

Vital sign measurements demonstrate terminal digit bias and boundary effects.

OBJECTIVE: The measurement and recording of vital signs may be impacted by biases, including preferences for even and round numbers. However, other biases, such as variation due to defined numerical boundaries (also known as boundary effects), may be present in vital signs data and have not yet been investigated in a medical setting. We aimed to assess vital signs data for such biases. These parameters are clinically significant as they influence care escalation. METHODS: Vital signs data (heart rate, respiratory rate, oxygen saturation and systolic blood pressure) were collected from a tertiary hospital electronic medical record over a 2-year period. These data were analysed using polynomial regression with additional terms to assess for underreporting of out-of-range observations and overreporting numbers with terminal digits of 0 (round numbers), 2 (even numbers) and 5. RESULTS: It was found that heart rate, oxygen saturation and systolic blood pressure demonstrated 'boundary effects', with values inside the 'normal' range disproportionately more likely to be recorded. Even number bias was observed in systolic heart rate, respiratory rate and blood pressure. Preference for multiples of 5 was observed for heart rate and blood pressure. Independent overrepresentation of multiples of 10 was demonstrated in heart rate data. CONCLUSION: Although often considered objective, vital signs data are affected by bias. These biases may impact the care patients receive. Additionally, it may have implications for creating and training machine learning models that utilise vital signs data.

Renal disease is not associated with delays in hyperacute stroke management in South Australia.

OBJECTIVE: The aims of the present study were to determine how renal disease is associated with the time to receive hyperacute stroke care. METHODS: The present study involved a 5-year cohort of all patients admitted to stroke units in South Australia. RESULTS: In those with pre-existing renal disease there were no significant differences in the time taken to receive a scan, thrombolysis or endovascular thrombectomy. CONCLUSIONS: The present study shows that in protocolised settings there were no significant delays in hyperacute stroke management for patients with renal disease.

Sensorineural hearing loss after cardiac surgery: a systematic review.

BACKGROUND: Sensorineural hearing loss (SNHL) may occur following cardiac surgery. Although preventing post-operative complications is vitally important in cardiac surgery, there are few guidelines regarding this issue. This review aimed to characterize SNHL after cardiac surgery. METHOD: This systematic review was registered on PROSPERO and conducted in accordance with PRISMA guidelines. A systematic search of the PubMed, Embase and Cochrane Library were conducted from inception. Eligibility determination, data extraction and methodological quality analysis were conducted in duplicate. RESULTS: There were 23 studies included in the review. In the adult population, there were six cohort studies, which included 36 cases of hearing loss in a total of 7135 patients (5.05 cases per 1000 operations). In seven cohort studies including paediatric patients, there were 88 cases of hearing loss in a total of 1342 operations. The majority of cases of hearing loss were mild in the adult population (56.6%). In the paediatric population 59.2% of hearing loss cases had moderate or worse hearing loss. The hearing loss most often affected the higher frequencies, over 6000 Hz. There have been studies indicating an association between hearing loss and extracorporeal circulation, but cases have also occurred without this intervention. CONCLUSION: SNHL is a rare but potentially serious complication after cardiac surgery. This hearing loss affects both paediatric and adult populations and may have significant long-term impacts. Further research is required, particularly with respect to the consideration of screening for SNHL in children after cardiac surgery.

Pre-stroke anticoagulation for atrial fibrillation in primary English speakers and non-primary English speakers: a multicentre retrospective cohort study.

BACKGROUND: Anticoagulation can prevent most strokes in individuals with atrial fibrillation (AF); however, many people presenting with stroke and known AF are not anticoagulated. Language barriers and poor health literacy have previously been associated with decreased patient medication adherence. The association between language barriers and initiation of anticoagulation therapy for AF is uncertain. AIMS: The aims of this study were to determine whether demographic factors, including non-English primary language, were (1) associated with not being initiated on anticoagulation for known AF prior to admission with stroke, and (2) associated with non-adherence to anticoagulation in the setting of known AF prior to admission with stroke. METHODS: A multicentre retrospective cohort study was conducted for consecutive individuals admitted to the three South Australian tertiary hospitals with stroke units over a 5-year period. RESULTS: There were 6829 individuals admitted with stroke. These cases included 5835 ischaemic stroke patients, 1333 of whom had pre-existing AF. Only 40.0% presenting with ischaemic stroke in the setting of known pre-existing AF were anticoagulated. When controlling for demographics, socioeconomic status and past medical history (including the components of the CHADS2VASC score and anticoagulation contraindications), having a primary language other than English was associated with a lower likelihood of having been commenced on anticoagulant for known pre-stroke AF (odds ratio: 0.52, 95% confidence interval: 0.36-0.77, P = 0.001), but was not associated with a differing likelihood of anticoagulation adherence. CONCLUSIONS: A significant proportion of patients with stroke have pre-existing unanticoagulated AF; these rates are substantially higher if the primary language is other than English. Targeted research and interventions to minimise evidence-treatment gaps in this cohort may significantly reduce stroke burden.

The Effectiveness of Self-supervised Pre-training for Multi-modal Endometriosis Classification*†.

Endometriosis is a debilitating condition affecting 5% to 10% of the women worldwide, where early detection and treatment are the best tools to manage the condition. Early detection can be done via surgery, but multi-modal medical imaging is preferable given the simpler and faster process. However, imaging-based endometriosis diagnosis is challenging as 1) there are few capable clinicians; and 2) it is characterised by small lesions unconfined to a specific location. These two issues challenge the development of endometriosis classifiers as the training datasets tend to be small and contain difficult samples, which leads to overfitting. Hence, it is important to consider generalisation techniques to mitigate this problem, particularly self-supervised pre-training methods that have shown outstanding results in computer vision and natural language processing applications. The main goal of this paper is to study the effectiveness of modern self-supervised pre-training techniques to overcome the two issues mentioned above for the classification of endometriosis from multi-modal imaging data. We also introduce a new masking image modelling self-supervised pre-training method that works with 3D multi-modal medical imaging. Furthermore, to the best of our knowledge, this paper presents the first endometriosis classifier, fine-tuned from the pre-trained model above, which works with multi-modal (i.e., T1 and T2) magnetic resonance imaging (MRI) data. Our results show that self-supervised pre-training improves endometriosis classification by as much as 31%, when compared with classifiers trained from scratch.

Machine learning and deep learning predictive models for long-term prognosis in patients with chronic obstructive pulmonary disease: a systematic review and meta-analysis.

BACKGROUND: Machine learning and deep learning models have been increasingly used to predict long-term disease progression in patients with chronic obstructive pulmonary disease (COPD). We aimed to summarise the performance of such prognostic models for COPD, compare their relative performances, and identify key research gaps. METHODS: We conducted a systematic review and meta-analysis to compare the performance of machine learning and deep learning prognostic models and identify pathways for future research. We searched PubMed, Embase, the Cochrane Library, ProQuest, Scopus, and Web of Science from database inception to April 6, 2023, for studies in English using machine learning or deep learning to predict patient outcomes at least 6 months after initial clinical presentation in those with COPD. We included studies comprising human adults aged 18-90 years and allowed for any input modalities. We reported area under the receiver operator characteristic curve (AUC) with 95% CI for predictions of mortality, exacerbation, and decline in forced expiratory volume in 1 s (FEV1). We reported the degree of interstudy heterogeneity using Cochran's Q test (significant heterogeneity was defined as p≤0·10 or I2>50%). Reporting quality was assessed using the TRIPOD checklist and a risk-of-bias assessment was done using the PROBAST checklist. This study was registered with PROSPERO (CRD42022323052). FINDINGS: We identified 3620 studies in the initial search. 18 studies were eligible, and, of these, 12 used conventional machine learning and six used deep learning models. Seven models analysed exacerbation risk, with only six reporting AUC and 95% CI on internal validation datasets (pooled AUC 0·77 [95% CI 0·69-0·85]) and there was significant heterogeneity (I2 97%, p<0·0001). 11 models analysed mortality risk, with only six reporting AUC and 95% CI on internal validation datasets (pooled AUC 0·77 [95% CI 0·74-0·80]) with significant degrees of heterogeneity (I2 60%, p=0·027). Two studies assessed decline in lung function and were unable to be pooled. Machine learning and deep learning models did not show significant improvement over pre-existing disease severity scores in predicting exacerbations (p=0·24). Three studies directly compared machine learning models against pre-existing severity scores for predicting mortality and pooled performance did not differ (p=0·57). Of the five studies that performed external validation, performance was worse than or equal to regression models. Incorrect handling of missing data, not reporting model uncertainty, and use of datasets that were too small relative to the number of predictive features included provided the largest risks of bias. INTERPRETATION: There is limited evidence that conventional machine learning and deep learning prognostic models demonstrate superior performance to pre-existing disease severity scores. More rigorous adherence to reporting guidelines would reduce the risk of bias in future studies and aid study reproducibility. FUNDING: None.

Surgery's Rosetta Stone: Natural language processing to predict discharge and readmission after general surgery.

BACKGROUND: This study aimed to examine the accuracy with which multiple natural language processing artificial intelligence models could predict discharge and readmissions after general surgery. METHODS: Natural language processing models were derived and validated to predict discharge within the next 48 hours and 7 days and readmission within 30 days (based on daily ward round notes and discharge summaries, respectively) for general surgery inpatients at 2 South Australian hospitals. Natural language processing models included logistic regression, artificial neural networks, and Bidirectional Encoder Representations from Transformers. RESULTS: For discharge prediction analyses, 14,690 admissions were included. For readmission prediction analyses, 12,457 patients were included. For prediction of discharge within 48 hours, derivation and validation data set area under the receiver operator characteristic curves were, respectively: 0.86 and 0.86 for Bidirectional Encoder Representations from Transformers, 0.82 and 0.81 for logistic regression, and 0.82 and 0.81 for artificial neural networks. For prediction of discharge within 7 days, derivation and validation data set area under the receiver operator characteristic curves were, respectively: 0.82 and 0.81 for Bidirectional Encoder Representations from Transformers, 0.75 and 0.72 for logistic regression, and 0.68 and 0.67 for artificial neural networks. For readmission prediction within 30 days, derivation and validation data set area under the receiver operator characteristic curves were, respectively: 0.55 and 0.59 for Bidirectional Encoder Representations from Transformers and 0.77 and 0.62 for logistic regression. CONCLUSION: Modern natural language processing models, particularly Bidirectional Encoder Representations from Transformers, can effectively and accurately identify general surgery patients who will be discharged in the next 48 hours. However, these approaches are less capable of identifying general surgery patients who will be discharged within the next 7 days or who will experience readmission within 30 days of discharge.

Opioid prescribing, pain, and hospital stay of general surgery patients with oxycodone allergies in South Australia.

BACKGROUND: The frequency of oxycodone adverse reactions, subsequent opioid prescription, effect on pain and patient care in general surgery patients are not well known. This study aimed to determine prevalence of documented oxycodone allergy and intolerances (independent variables) in a general surgical cohort, and association with prescribing other analgesics (particularly opioids), subjective pain scores, and length of hospital stay (dependent variables). METHODS: This retrospective cohort study included general surgery patients from two South Australian hospitals between April 2020 and March 2022. Multivariable logistic regression evaluated associations between previous oxycodone allergies and intolerances, prescription records, subjective pain scores, and length of hospital stay. RESULTS: Of 12 846 patients, 216 (1.7%) had oxycodone allergies, and 84 (0.7%) oxycodone intolerances. The 216 oxycodone allergy patients had lower odds of receiving oxycodone (OR 0.17, P < 0.001), higher odds of tramadol (OR 3.01, P < 0.001) and tapentadol (OR 2.87, P = 0.001), but 91 (42.3%) still received oxycodone and 19 (8.8%) morphine. The 84 with oxycodone intolerance patients had lower odds of receiving oxycodone (OR 0.23, P < 0.001), higher odds of fentanyl (OR 3.6, P < 0.001) and tramadol (OR 3.35, P < 0.001), but 42 (50%) still received oxycodone. Patients with oxycodone allergies and intolerances had higher odds of elevated subjective pain (OR 1.60, P = 0.013; OR 2.36, P = 0.002, respectively) and longer length of stay (OR 1.36, P = 0.038; OR 2.24, P = 0.002, respectively) than patients without these. CONCLUSIONS: General surgery patients with oxycodone allergies and intolerances are at greater risk of worse postoperative pain and longer length of stay, compared to patients without. Many still receive oxycodone, and other opioids that could cause cross-reactivity.

Neuroimaging Findings in Axenfeld-Rieger Syndrome: A Case Series.

Axenfeld-Rieger syndrome is an autosomal dominant condition associated with multisystemic features including developmental anomalies of the anterior segment of the eye. Single nucleotide and copy number variants in the paired-like homeodomain transcription factor 2 (PITX2) and forkhead box C1 (FOXC1) genes are associated with Axenfeld-Rieger syndrome as well as other CNS malformations. We determined the association between Axenfeld-Rieger syndrome and specific brain MR imaging neuroradiologic anomalies in cases with or without a genetic diagnosis. This case series included 8 individuals with pathogenic variants in FOXC1; 2, in PITX2; and 2 without a genetic diagnosis. The most common observation was vertebrobasilar artery dolichoectasia, with 46% prevalence. Other prevalent abnormalities included WM hyperintensities, cerebellar hypoplasia, and ventriculomegaly. Vertebrobasilar artery dolichoectasia and absent/hypoplastic olfactory bulbs were reported in >50% of individuals with FOXC1 variants compared with 0% of PITX2 variants. Notwithstanding the small sample size, neuroimaging abnormalities were more prevalent in individuals with FOXC1 variants compared those with PITX2 variants.

Systematic review of adherence to the standards for reporting of diagnostic accuracy studies (STARD) 2015 reporting guideline in cerebral aneurysm imaging diagnostic accuracy studies.

BACKGROUND: Diagnostic neuroimaging plays an essential role in guiding clinical decision-making in the management of patients with cerebral aneurysms. Imaging technologies for investigating cerebral aneurysms constantly evolve, and clinicians rely on the published literature to remain up to date. Reporting guidelines have been developed to standardise and strengthen the reporting of clinical evidence. Therefore, it is essential that radiological diagnostic accuracy studies adhere to such guidelines to ensure completeness of reporting. Incomplete reporting hampers the reader's ability to detect bias, determine generalisability of study results or replicate investigation parameters, detracting from the credibility and reliability of studies. OBJECTIVE: The purpose of this systematic review was to evaluate adherence to the Standards for Reporting of Diagnostic Accuracy Studies (STARD) 2015 reporting guideline amongst imaging diagnostic accuracy studies for cerebral aneurysms. METHODS: A systematic search for cerebral aneurysm imaging diagnostic accuracy studies was conducted. Journals were cross examined against the STARD 2015 checklist and their compliance with item numbers was recorded. RESULTS: The search yielded 66 articles. The mean number of STARD items reported was 24.2 ± 2.7 (71.2% ± 7.9%), with a range of 19 to 30 out of a maximum number of 34 items. CONCLUSION: Taken together, these results indicate that adherence to the STARD 2015 guideline in cerebral aneurysm imaging diagnostic accuracy studies was moderate. Measures to improve compliance include mandating STARD 2015 adherence in instructions to authors issued by journals.

Vital signs and medical emergency response (MER) activation predict in-hospital mortality in general surgery patients: a study of 15 969 admissions.

BACKGROUND: The applicability of the vital signs prompting medical emergency response (MER) activation has not previously been examined specifically in a large general surgical cohort. This study aimed to characterize the distribution, and predictive performance, of four vital signs selected based on Australian guidelines (oxygen saturation, respiratory rate, systolic blood pressure and heart rate); with those of the MER activation criteria. METHODS: A retrospective cohort study was conducted including patients admitted under general surgical services of two hospitals in South Australia over 2 years. Likelihood ratios for patients meeting MER activation criteria, or a vital sign in the most extreme 1% for general surgery inpatients (<0.5th percentile or > 99.5th percentile), were calculated to predict in-hospital mortality. RESULTS: 15 969 inpatient admissions were included comprising 2 254 617 total vital sign observations. The 0.5th and 99.5th centile for heart rate was 48 and 133, systolic blood pressure 85 and 184, respiratory rate 10 and 31, and oxygen saturations 89% and 100%, respectively. MER activation criteria with the highest positive likelihood ratio for in-hospital mortality were heart rate ≤ 39 (37.65, 95% CI 27.71-49.51), respiratory rate ≥ 31 (15.79, 95% CI 12.82-19.07), and respiratory rate ≤ 7 (10.53, 95% CI 6.79-14.84). These MER activation criteria likelihood ratios were similar to those derived when applying a threshold of the most extreme 1% of vital signs. CONCLUSIONS: This study demonstrated that vital signs within Australian guidelines, and escalation to MER activation, appropriately predict in-hospital mortality in a large cohort of patients admitted to general surgical services in South Australia.

Imaging evaluation of a proposed 3D generative model for MRI to CT translation in the lumbar spine.

BACKGROUND CONTEXT: A computed tomography (CT) and magnetic resonance imaging (MRI) are used routinely in the radiologic evaluation and surgical planning of patients with lumbar spine pathology, with the modalities being complimentary. We have developed a deep learning algorithm which can produce 3D lumbar spine CT images from MRI data alone. This has the potential to reduce radiation to the patient as well as burden on the health care system. PURPOSE: The purpose of this study is to evaluate the accuracy of the synthetic lumbar spine CT images produced using our deep learning model. STUDY DESIGN: A training set of 400 unpaired CTs and 400 unpaired MRI scans of the lumbar spine was used to train a supervised 3D cycle-Gan model. Evaluators performed a set of clinically relevant measurements on 20 matched synthetic CTs and true CTs. These measurements were then compared to assess the accuracy of the synthetic CTs. PATIENT SAMPLE: The evaluation data set consisted of 20 patients who had CT and MRI scans performed within a 30-day period of each other. All patient data was deidentified. Notable exclusions included artefact from patient motion, metallic implants or any intervention performed in the 30 day intervening period. OUTCOME MEASURES: The outcome measured was the mean difference in measurements performed by the group of evaluators between real CT and synthetic CTs in terms of absolute and relative error. METHODS: Data from the 20 MRI scans was supplied to our deep learning model which produced 20 "synthetic CT" scans. This formed the evaluation data set. Four clinical evaluators consisting of neurosurgeons and radiologists performed a set of 24 clinically relevant measurements on matched synthetic CT and true CTs in 20 patients. A test set of measurements were performed prior to commencing data collection to identify any significant interobserver variation in measurement technique. RESULTS: The measurements performed in the sagittal plane were all within 10% relative error with the majority within 5% relative error. The pedicle measurements performed in the axial plane were considerably less accurate with a relative error of up to 34%. CONCLUSIONS: The computer generated synthetic CTs demonstrated a high level of accuracy for the measurements performed in-plane to the original MRIs used for synthesis. The measurements performed on the axial reconstructed images were less accurate, attributable to the images being synthesized from nonvolumetric routine sagittal T1-weighted MRI sequences. It is hypothesized that if axial sequences or volumetric data were input into the algorithm these measurements would have improved accuracy.

Factors associated with delay to carotid endarterectomy for acute ischaemic stroke in South Australia: A multicentre retrospective cohort study.

BACKGROUND: The greatest benefits of carotid endarterectomy (CEA) accrue when performed within two weeks of acute ischaemic stroke (AIS) due to symptomatic carotid stenosis. Previous studies have identified multiple factors contributing to CEA delay. AIMS: To determine factors associated with delayed CEA in patients admitted to tertiary stroke centres within a major metropolitan region with AIS METHODS: In a retrospective cohort study, consecutive patients admitted to the tertiary hospitals with stroke units within South Australia (Lyell McEwin Hospital, Royal Adelaide Hospital and Flinders Medical Centre) between 2016 to 2020 were included. Univariable and multivariable logistic regression were used to identify individual factors associated with time from symptom onset to CEA of over two weeks. RESULTS: A total of 174 patients were included. The median time to CEA was 5 days (IQR 3-9.75). Delayed CEA beyond 14 days occurred in 28/174 (16%). Factors most associated with delayed CEA included presentation to a tertiary hospital without onsite Vascular Surgical Unit (OR 3.71, 95%CI 1.31-10.58), history of previous stroke (OR 3.38, 95% CI 1.11-9.84) and presenting NIHSS above 6 (OR 5.16, 95% CI 1.60-16.39). CONCLUSION: This study identified that presentation to a tertiary hospital without a Vascular Surgery Unit, history of previous stroke and presenting NIHSS above 6 were associated with delay to CEA in AIS patients in South Australia. Interventional studies aiming to improve the proportion of patients that receive CEA within 14 days are required.

Development and external validation of automated detection, classification, and localization of ankle fractures: inside the black box of a convolutional neural network (CNN).

PURPOSE: Convolutional neural networks (CNNs) are increasingly being developed for automated fracture detection in orthopaedic trauma surgery. Studies to date, however, are limited to providing classification based on the entire image-and only produce heatmaps for approximate fracture localization instead of delineating exact fracture morphology. Therefore, we aimed to answer (1) what is the performance of a CNN that detects, classifies, localizes, and segments an ankle fracture, and (2) would this be externally valid? METHODS: The training set included 326 isolated fibula fractures and 423 non-fracture radiographs. The Detectron2 implementation of the Mask R-CNN was trained with labelled and annotated radiographs. The internal validation (or 'test set') and external validation sets consisted of 300 and 334 radiographs, respectively. Consensus agreement between three experienced fellowship-trained trauma surgeons was defined as the ground truth label. Diagnostic accuracy and area under the receiver operator characteristic curve (AUC) were used to assess classification performance. The Intersection over Union (IoU) was used to quantify accuracy of the segmentation predictions by the CNN, where a value of 0.5 is generally considered an adequate segmentation. RESULTS: The final CNN was able to classify fibula fractures according to four classes (Danis-Weber A, B, C and No Fracture) with AUC values ranging from 0.93 to 0.99. Diagnostic accuracy was 89% on the test set with average sensitivity of 89% and specificity of 96%. External validity was 89-90% accurate on a set of radiographs from a different hospital. Accuracies/AUCs observed were 100/0.99 for the 'No Fracture' class, 92/0.99 for 'Weber B', 88/0.93 for 'Weber C', and 76/0.97 for 'Weber A'. For the fracture bounding box prediction by the CNN, a mean IoU of 0.65 (SD ± 0.16) was observed. The fracture segmentation predictions by the CNN resulted in a mean IoU of 0.47 (SD ± 0.17). CONCLUSIONS: This study presents a look into the 'black box' of CNNs and represents the first automated delineation (segmentation) of fracture lines on (ankle) radiographs. The AUC values presented in this paper indicate good discriminatory capability of the CNN and substantiate further study of CNNs in detecting and classifying ankle fractures. LEVEL OF EVIDENCE: II, Diagnostic imaging study.

The APOE E4 Allele Is Associated with Faster Rates of Neuroretinal Thinning in a Prospective Cohort Study of Suspect and Early Glaucoma.

Purpose: To investigate the association between the apolipoprotein E (APOE) E4 dementia-risk allele and prospective longitudinal retinal thinning in a cohort study of suspect and early manifest glaucoma. Design: Retrospective analysis of prospective cohort data. Participants: This study included all available eyes from participants recruited to the Progression Risk of Glaucoma: Relevant SNPs [single nucleotide polymorphisms] with Significant Association (PROGRESSA) study with genotyping data from which APOE genotypes could be determined. Methods: Apolipoprotein E alleles and genotypes were determined in PROGRESSA, and their distributions were compared with an age-matched and ancestrally matched normative cohort, the Blue Mountains Eye Study. Structural parameters of neuroretinal atrophy measured using spectral-domain OCT were compared within the PROGRESSA cohort on the basis of APOE E4 allele status. Main Outcome Measures: Longitudinal rates of thinning in the macular ganglion cell-inner plexiform layer (mGCIPL) complex and the peripapillary retinal nerve fiber layer (pRNFL). Results: Rates of mGCIPL complex thinning were faster in participants harboring ≥1 copies of the APOE E4 allele (ß = -0.13 µm/year; P ≤0.001). This finding was strongest in eyes affected by normal-tension glaucoma (NTG; ß = -0.20 µm/year; P = 0.003). Apolipoprotein E E4 allele carriers were also more likely to be lost to follow-up (P = 0.01) and to demonstrate a thinner average mGCIPL complex (70.9 µm vs. 71.9 µm; P = 0.011) and pRNFL (77.6 µm vs. 79.2 µm; P = 0.045) after a minimum of 3 years of monitoring. Conclusions: The APOE E4 allele was associated with faster rates of mCGIPL complex thinning, particularly in eyes with NTG. These results suggest that the APOE E4 allele may be a risk factor for retinal ganglion cell degeneration in glaucoma.

Small Study Effects in Diagnostic Imaging Accuracy: A Meta-Analysis.

Importance: Small study effects are the phenomena that studies with smaller sample sizes tend to report larger and more favorable effect estimates than studies with larger sample sizes. Objective: To evaluate the presence and extent of small study effects in diagnostic imaging accuracy meta-analyses. Data Sources: A search was conducted in the PubMed database for diagnostic imaging accuracy meta-analyses published between 2010 and 2019. Study Selection: Meta-analyses with 10 or more studies of medical imaging diagnostic accuracy, assessing a single imaging modality, and providing 2 × 2 contingency data were included. Studies that did not assess diagnostic accuracy of medical imaging techniques, compared 2 or more imaging modalities or different methods of 1 imaging modality, were cost analyses, used predictive or prognostic tests, did not provide individual patient data, or were network meta-analyses were excluded. Data Extraction and Synthesis: Data extraction was performed in accordance with the PRISMA guidelines. Main Outcomes and Measures: The diagnostic odds ratio (DOR) was calculated for each primary study using 2 × 2 contingency data. Regression analysis was used to examine the association between effect size estimate and precision across meta-analyses. Results: A total of 31 meta-analyses involving 668 primary studies and 80 206 patients were included. Fixed effects analysis produced a regression coefficient for the natural log of DOR against the SE of the natural log of DOR of 2.19 (95% CI, 1.49-2.90; P < .001), with computed tomography as the reference modality. Interaction test for modality and SE of the natural log of DOR did not depend on modality (Wald statistic P = .50). Taken together, this analysis found an inverse association between effect size estimate and precision that was independent of imaging modality. Of 26 meta-analyses that formally assessed for publication bias using funnel plots and statistical tests for funnel plot asymmetry, 21 found no evidence for such bias. Conclusions and Relevance: This meta-analysis found evidence of widespread prevalence of small study effects in the diagnostic imaging accuracy literature. One likely contributor to the observed effects is publication bias, which can undermine the results of many meta-analyses. Conventional methods for detecting funnel plot asymmetry conducted by included studies appeared to underestimate the presence of small study effects. Further studies are required to elucidate the various factors that contribute to small study effects.

Visualising and quantifying microvascular structure and function in patients with heart failure using optical coherence tomography.

Heart failure (HF) is characterised by abnormal conduit and resistance artery function in humans. Microvascular function in HF is less well characterised, due in part to the lack of tools to image these vessels in vivo. The skin microvasculature is a surrogate for systemic microvascular function and health and plays a key role in thermoregulation, which is dysfunctional in HF. We deployed a novel optical coherence tomography (OCT) technique to visualise and quantify microvascular structure and function in 10 subjects with HF and 10 age- and sex-matched controls. OCT images were obtained from the ventral aspect of the forearm, at baseline (33°C) and after 30 min of localised skin heating. At rest, OCT-derived microvascular density (20.3 ± 8.7%, P = 0.004), diameter (35.1 ± 6.0 µm, P = 0.006) and blood flow (82.9 ± 41.1 pl/s, P = 0.021) were significantly lower in HF than CON (27.2 ± 8.0%, 40.4 ± 5.8 µm, 110.8 ± 41.9 pl/s), whilst blood speed was not significantly lower (74.3 ± 11.0 µm/s vs. 81.3 ± 9.9 µm/s, P = 0.069). After local heating, the OCT-based density, diameter, blood speed and blood flow of HF patients were similar (all P > 0.05) to CON. Although abnormalities exist at rest which may reflect microvascular disease status, patients with HF retain the capacity to dilate cutaneous microvessels in response to localised heat stress. This is a novel in vivo human observation of microvascular dysfunction in HF, illustrating the feasibility of OCT to directly visualise and quantify microvascular responses to physiological stimuli in vivo. KEY POINTS: Microvessels in the skin are critical to human thermoregulation, which is compromised in participants with heart failure (HF). We have developed a powerful new non-invasive optical coherence tomography (OCT)-based approach for the study of microvascular structure and function in vivo. Our approach enabled us to observe and quantify abnormal resting microvascular function in participants with HF. Patients with HF were able to dilate skin microvessels in response to local heat stress, arguing against an underlying structural abnormality. This suggests that microvascular functional regulation is the primary abnormality in HF. OCT can be used to directly visualise and quantify microvascular responses to physiological stimuli in vivo.

Systematic Analysis of Publication Bias in Neurosurgery Meta-Analyses.

BACKGROUND: Statistically significant positive results are more likely to be published than negative or insignificant outcomes. This phenomenon, also termed publication bias, can skew the interpretation of meta-analyses. The widespread presence of publication bias in the biomedical literature has led to the development of various statistical approaches, such as the visual inspection of funnel plots, Begg test, and Egger test, to assess and account for it. OBJECTIVE: To determine how well publication bias is assessed for in meta-analyses of the neurosurgical literature. METHODS: A systematic search for meta-analyses from the top neurosurgery journals was conducted. Data relevant to the presence, assessment, and adjustments for publication bias were extracted. RESULTS: The search yielded 190 articles. Most of the articles (n = 108, 56.8%) were assessed for publication bias, of which 40 (37.0%) found evidence for publication bias whereas 61 (56.5%) did not. In the former case, only 11 (27.5%) made corrections for the bias using the trim-and-fill method, whereas 29 (72.5%) made no correction. Thus, 111 meta-analyses (58.4%) either did not assess for publication bias or, if assessed to be present, did not adjust for it. CONCLUSION: Taken together, these results indicate that publication bias remains largely unaccounted for in neurosurgical meta-analyses.