Assistive AI in Lung Cancer Screening: A Retrospective Multinational Study in the United States and Japan

Purpose To evaluate the impact of an artificial intelligence (AI) assistant for lung cancer screening on multinational clinical workflows. Materials and Methods An AI assistant for lung cancer screening was evaluated on two retrospective randomized multireader multicase studies where 627 (141 cancer-positive cases) low-dose chest CT cases were each read twice (with and without AI assistance) by experienced thoracic radiologists (six U.S.-based or six Japan-based radiologists), resulting in a total of 7524 interpretations. Positive cases were defined as those within 2 years before a pathology-confirmed lung cancer diagnosis. Negative cases were defined as those without any subsequent cancer diagnosis for at least 2 years and were enriched for a spectrum of diverse nodules. The studies measured the readers' level of suspicion (on a 0-100 scale), country-specific screening system scoring categories, and management recommendations. Evaluation metrics included the area under the receiver operating characteristic curve (AUC) for level of suspicion and sensitivity and specificity of recall recommendations. Results With AI assistance, the radiologists' AUC increased by 0.023 (0.70 to 0.72; P = .02) for the U.S. study and by 0.023 (0.93 to 0.96; P = .18) for the Japan study. Scoring system specificity for actionable findings increased 5.5% (57% to 63%; P < .001) for the U.S. study and 6.7% (23% to 30%; P < .001) for the Japan study. There was no evidence of a difference in corresponding sensitivity between unassisted and AI-assisted reads for the U.S. (67.3% to 67.5%; P = .88) and Japan (98% to 100%; P > .99) studies. Corresponding stand-alone AI AUC system performance was 0.75 (95% CI: 0.70, 0.81) and 0.88 (95% CI: 0.78, 0.97) for the U.S.- and Japan-based datasets, respectively. Conclusion The concurrent AI interface improved lung cancer screening specificity in both U.S.- and Japan-based reader studies, meriting further study in additional international screening environments. Keywords: Assistive Artificial Intelligence, Lung Cancer Screening, CT Supplemental material is available for this article. Published under a CC BY 4.0 license.

Understanding metric-related pitfalls in image analysis validation

Validation metrics are key for tracking scientific progress and bridging the current chasm between artificial intelligence research and its translation into practice. However, increasing evidence shows that, particularly in image analysis, metrics are often chosen inadequately. Although taking into account the individual strengths, weaknesses and limitations of validation metrics is a critical prerequisite to making educated choices, the relevant knowledge is currently scattered and poorly accessible to individual researchers. Based on a multistage Delphi process conducted by a multidisciplinary expert consortium as well as extensive community feedback, the present work provides a reliable and comprehensive common point of access to information on pitfalls related to validation metrics in image analysis. Although focused on biomedical image analysis, the addressed pitfalls generalize across application domains and are categorized according to a newly created, domain-agnostic taxonomy. The work serves to enhance global comprehension of a key topic in image analysis validation.

Metrics reloaded: recommendations for image analysis validation

Increasing evidence shows that flaws in machine learning (ML) algorithm validation are an underestimated global problem. In biomedical image analysis, chosen performance metrics often do not reflect the domain interest, and thus fail to adequately measure scientific progress and hinder translation of ML techniques into practice. To overcome this, we created Metrics Reloaded, a comprehensive framework guiding researchers in the problem-aware selection of metrics. Developed by a large international consortium in a multistage Delphi process, it is based on the novel concept of a problem fingerprint-a structured representation of the given problem that captures all aspects that are relevant for metric selection, from the domain interest to the properties of the target structure(s), dataset and algorithm output. On the basis of the problem fingerprint, users are guided through the process of choosing and applying appropriate validation metrics while being made aware of potential pitfalls. Metrics Reloaded targets image analysis problems that can be interpreted as classification tasks at image, object or pixel level, namely image-level classification, object detection, semantic segmentation and instance segmentation tasks. To improve the user experience, we implemented the framework in the Metrics Reloaded online tool. Following the convergence of ML methodology across application domains, Metrics Reloaded fosters the convergence of validation methodology. Its applicability is demonstrated for various biomedical use cases.

Evaluation of RURS elbow guard in the management of thumb-sucking habit in children with and without intellectual disability

BACKGROUND

Digit sucking is a common oral habit among many children, which involves placing the thumb/finger into the mouth, which can cause malocclusion in mixed and permanent dentition.

AIM

To evaluate the efficacy of the RURS elbow guard in the management of thumb-sucking habits in children with intellectual disabilities and those without in terms of the mean duration of appliance therapy. The secondary objective was to compare the RURS elbow guard with an intraoral crib in healthy children (without intellectual disabilities) to manage the habit of thumb-sucking.

DESIGN

Children with intellectual disabilities and those without between the age of 4 and 16 years were included in the study and categorised into three groups, namely group I (50 normal children; intraoral crib appliance), group II (50 normal children; RURS elbow guard) and group III (50 children with intellectual disabilities; RURS elbow guard).

RESULTS

The mean duration of appliance therapy for groups I, II and III were 200.20 ± 20.43 days, 204.34 ± 20.56 days, and 218.43 ± 15.66 days, respectively (p < 0.001). The differences in the mean duration between group I and group II were statistically non-significant, with statistically significant differences between group I and group III (p < 0.001) and between group II and group III (p < 0.05).

CONCLUSION

The RURS elbow guard was found to be an efficient appliance in treating thumb-sucking among children who had mild to moderate intellectual disabilities. RURS elbow guard was equally efficient as an intraoral crib appliance in managing thumb-sucking habits in children without intellectual disabilities.

An intentional approach to managing bias in general purpose embedding models

Advances in machine learning for health care have brought concerns about bias from the research community; specifically, the introduction, perpetuation, or exacerbation of care disparities. Reinforcing these concerns is the finding that medical images often reveal signals about sensitive attributes in ways that are hard to pinpoint by both algorithms and people. This finding raises a question about how to best design general purpose pretrained embeddings (GPPEs, defined as embeddings meant to support a broad array of use cases) for building downstream models that are free from particular types of bias. The downstream model should be carefully evaluated for bias, and audited and improved as appropriate. However, in our view, well intentioned attempts to prevent the upstream components-GPPEs-from learning sensitive attributes can have unintended consequences on the downstream models. Despite producing a veneer of technical neutrality, the resultant end-to-end system might still be biased or poorly performing. We present reasons, by building on previously published data, to support the reasoning that GPPEs should ideally contain as much information as the original data contain, and highlight the perils of trying to remove sensitive attributes from a GPPE. We also emphasise that downstream prediction models trained for specific tasks and settings, whether developed using GPPEs or not, should be carefully designed and evaluated to avoid bias that makes models vulnerable to issues such as distributional shift. These evaluations should be done by a diverse team, including social scientists, on a diverse cohort representing the full breadth of the patient population for which the final model is intended.

Enhancing the reliability and accuracy of AI-enabled diagnosis via complementarity-driven deferral to clinicians

Predictive artificial intelligence (AI) systems based on deep learning have been shown to achieve expert-level identification of diseases in multiple medical imaging settings, but can make errors in cases accurately diagnosed by clinicians and vice versa. We developed Complementarity-Driven Deferral to Clinical Workflow (CoDoC), a system that can learn to decide between the opinion of a predictive AI model and a clinical workflow. CoDoC enhances accuracy relative to clinician-only or AI-only baselines in clinical workflows that screen for breast cancer or tuberculosis (TB). For breast cancer screening, compared to double reading with arbitration in a screening program in the UK, CoDoC reduced false positives by 25% at the same false-negative rate, while achieving a 66% reduction in clinician workload. For TB triaging, compared to standalone AI and clinical workflows, CoDoC achieved a 5-15% reduction in false positives at the same false-negative rate for three of five commercially available predictive AI systems. To facilitate the deployment of CoDoC in novel futuristic clinical settings, we present results showing that CoDoC's performance gains are sustained across several axes of variation (imaging modality, clinical setting and predictive AI system) and discuss the limitations of our evaluation and where further validation would be needed. We provide an open-source implementation to encourage further research and application.

Clinical outcome of anterior cruciate ligament reconstruction with modified transtibial and anteromedial portal

Arthroscopic ACL reconstruction is the current standard care of treatment for anterior cruciate ligament (ACL) injuries. Modified transtibial (mTT) and anteromedial portal (AMP) techniques aim at the anatomical placement of femoral tunnel. Controversy existed in the literature with regard to the outcome of these techniques. Hence, we designed a retrospective comparative study to analyse the clinical and functional outcomes of mTT and AMP techniques. We hypothesized that there would be no difference between the clinical and functional outcomes in mTT and AMP techniques. This retrospective observational study was conducted in consecutive patients who underwent arthroscopic ACL reconstruction using semitendinosus-gracilis (STG) quadrupled graft in our tertiary care centre with a minimum follow-up of two years. Out of 69 patients, 37 had undergone ACL reconstruction by mTT technique and remaining by AMP technique. All the patients were assessed clinically by anterior drawer, Lachman's, pivot shift and single-legged hop test. Lysholm Knee Scoring Scale and International Knee Documentation Committee (IKDC) subjective knee evaluation score were used for the functional status. Knee instability was assessed objectively by KT-1000 arthrometer. There was no statistically significant difference in baseline demographic characteristics between mTT and AMP groups. At the end of 2 years, no statistically significant difference was noted in the anterior drawer and Lachman's test. Though not significant, IKDC scores and Lysholm's scores showed a better outcome in the AMP group when compared to the mTT group. AMP group showed significantly better outcome with KT-1000 arthrometer. Based on the results obtained, we presume that overall both mTT and AMP have similar functional outcome. However, as AMP technique offers significantly improved subjective rotational stability on pivot shift test, better hop limb symmetry index and KT 1000 readings compared to mTT, we suggest AMP over mTT.

Comparative in vitro evaluation of remaining dentine thickness following instrumentation with hand and rotary endodontic files during pulpectomy in primary molars: a systematic review

PURPOSE

The aim was to compare the remaining dentine thickness (RDT) following instrumentation with hand and rotary endodontic files during pulpectomy in primary molars. Research question was 'Is there any difference between the remaining dentine thickness following instrumentation with hand and rotary endodontic files during pulpectomy in primary molars?'.

METHODS

Electronic Databases like MEDLINE PubMed, Cochrane Library, EBSCOhost, Google scholar and grey literature were searched between January 1, 2006 and August 31, 2022 for in vitro and ex vivo studies that compared hand and rotary endodontic instrumentation to evaluate the RDT in primary molars. Articles published in English or which could be translated into English were searched. Two reviewers independently selected studies, extracted data, assessed risk of bias using the Revised, validated version of MINORS criteria.

RESULTS

Twelve studies were included in qualitative analysis. All included studies showed low risk of bias. Six studies showed more RDT with rotary instrumentation compared to manual instrumentation. Whereas, five studies showed variable results for RDT with manual and rotary instrumentation at different levels of root canals. One study showed no significant difference between manual and rotary instrumentation. In view of methodological heterogeneity of the findings, a meta-analysis was not conducted.

CONCLUSION

High quality of evidence based on low risk of bias was found in all the included studies. Statistically, rotary instrumentation showed more RDT than manual instrumentation according to majority of studies. Despite the shortcomings of this systematic review, it is possible to infer that the use of rotary instrumentation provides more RDT and thus there is considerable conservation of tooth structure.

Development of a Machine Learning Model for Sonographic Assessment of Gestational Age

IMPORTANCE

Fetal ultrasonography is essential for confirmation of gestational age (GA), and accurate GA assessment is important for providing appropriate care throughout pregnancy and for identifying complications, including fetal growth disorders. Derivation of GA from manual fetal biometry measurements (ie, head, abdomen, and femur) is operator dependent and time-consuming.

OBJECTIVE

To develop artificial intelligence (AI) models to estimate GA with higher accuracy and reliability, leveraging standard biometry images and fly-to ultrasonography videos.

DESIGN, SETTING, AND PARTICIPANTS

To improve GA estimates, this diagnostic study used AI to interpret standard plane ultrasonography images and fly-to ultrasonography videos, which are 5- to 10-second videos that can be automatically recorded as part of the standard of care before the still image is captured. Three AI models were developed and validated: (1) an image model using standard plane images, (2) a video model using fly-to videos, and (3) an ensemble model (combining both image and video models). The models were trained and evaluated on data from the Fetal Age Machine Learning Initiative (FAMLI) cohort, which included participants from 2 study sites at Chapel Hill, North Carolina (US), and Lusaka, Zambia. Participants were eligible to be part of this study if they received routine antenatal care at 1 of these sites, were aged 18 years or older, had a viable intrauterine singleton pregnancy, and could provide written consent. They were not eligible if they had known uterine or fetal abnormality, or had any other conditions that would make participation unsafe or complicate interpretation. Data analysis was performed from January to July 2022.

MAIN OUTCOMES AND MEASURES

The primary analysis outcome for GA was the mean difference in absolute error between the GA model estimate and the clinical standard estimate, with the ground truth GA extrapolated from the initial GA estimated at an initial examination.

RESULTS

Of the total cohort of 3842 participants, data were calculated for a test set of 404 participants with a mean (SD) age of 28.8 (5.6) years at enrollment. All models were statistically superior to standard fetal biometry-based GA estimates derived from images captured by expert sonographers. The ensemble model had the lowest mean absolute error compared with the clinical standard fetal biometry (mean [SD] difference, -1.51 [3.96] days; 95% CI, -1.90 to -1.10 days). All 3 models outperformed standard biometry by a more substantial margin on fetuses that were predicted to be small for their GA.

CONCLUSIONS AND RELEVANCE

These findings suggest that AI models have the potential to empower trained operators to estimate GA with higher accuracy.

Deep Learning Detection of Active Pulmonary Tuberculosis at Chest Radiography Matched the Clinical Performance of Radiologists

Background The World Health Organization (WHO) recommends chest radiography to facilitate tuberculosis (TB) screening. However, chest radiograph interpretation expertise remains limited in many regions. Purpose To develop a deep learning system (DLS) to detect active pulmonary TB on chest radiographs and compare its performance to that of radiologists. Materials and Methods A DLS was trained and tested using retrospective chest radiographs (acquired between 1996 and 2020) from 10 countries. To improve generalization, large-scale chest radiograph pretraining, attention pooling, and semisupervised learning ("noisy-student") were incorporated. The DLS was evaluated in a four-country test set (China, India, the United States, and Zambia) and in a mining population in South Africa, with positive TB confirmed with microbiological tests or nucleic acid amplification testing (NAAT). The performance of the DLS was compared with that of 14 radiologists. The authors studied the efficacy of the DLS compared with that of nine radiologists using the Obuchowski-Rockette-Hillis procedure. Given WHO targets of 90% sensitivity and 70% specificity, the operating point of the DLS (0.45) was prespecified to favor sensitivity. Results A total of 165 754 images in 22 284 subjects (mean age, 45 years; 21% female) were used for model development and testing. In the four-country test set (1236 subjects, 17% with active TB), the receiver operating characteristic (ROC) curve of the DLS was higher than those for all nine India-based radiologists, with an area under the ROC curve of 0.89 (95% CI: 0.87, 0.91). Compared with these radiologists, at the prespecified operating point, the DLS sensitivity was higher (88% vs 75%, P < .001) and specificity was noninferior (79% vs 84%, P = .004). Trends were similar within other patient subgroups, in the South Africa data set, and across various TB-specific chest radiograph findings. In simulations, the use of the DLS to identify likely TB-positive chest radiographs for NAAT confirmation reduced the cost by 40%-80% per TB-positive patient detected. Conclusion A deep learning method was found to be noninferior to radiologists for the determination of active tuberculosis on digital chest radiographs. © RSNA, 2022 Online supplemental material is available for this article. See also the editorial by van Ginneken in this issue.

Bifocal Stabilisation of Acute Acromioclavicular Joint Dislocation using Suture Anchor and Temporary K-Wires: A Retrospective Analysis

Introduction

The acromioclavicular joint (ACJ) is a major link connecting the upper limb to the torso. The acromioclavicular and coracoclavicular (CC) ligaments help in stabilising the joint. We feel it is prudent to address both these ligament injuries, to achieve optimum result. This study was undertaken to analyse the results of a simple frugal surgical technique we used to deal with this injury considering stabilisation for both these ligaments.

Materials and methods

In this retrospective study, skeletally mature patients with Type III, IV or V ACJ dislocations who underwent open reduction and stabilisation of the joint with temporary K-wires, repair of the capsule and augmentation of CC ligaments with suture anchors were included. Clinico-radiological and functional outcome was evaluated. Functional assessment of the upper limb was analysed using the Disabilities of Arm, Shoulder, and Hand Score (DASH), Constant shoulder score (CSS) and Oxford shoulder score (OSS).

Results

Clinical and radiological evaluation of the 32 patients who had completed two years from the index surgery, was done. Out of the 37 patients included initially, five were lost in follow-up. Majority of the subjects included were males and type V was the most common injury. Mean pre-operative CC distance on the affected side was 13.92±4.94mm. In the immediate post-operative radiograph, it was 7.63±2.08mm and in the final follow- up was 9.36±2.75mm. Measurements were taken by two independent investigators and inter, and intra-observer reliability were analysed by Interclass correlation coefficient. Excellent functional outcome was noted despite the 1.81±1.50mm average loss of correction. At final follow-up, mean DASH score was 4.67±4.18, Oxford shoulder score was 44.06±2.44 and Constant shoulder score was 86.37±5.81. The severity of the injury had no significant effect on the functional outcome post our method of stabilisation and rehabilitation.

Conclusion

Bifocal fixation restores the multidirectional stability of the disrupted ACJ. Adequate radiological reduction, good functional outcome and simplicity of execution make this technique an undemanding one for use in regular practice.

Effects of transplantation-related immunosuppression on co-existent neuroendocrine tumours

BACKGROUND

Here we detail our experience of managing patients found to have a neuroendocrine neoplasm (NEN) whilst on immunosuppression for a transplanted organ.

AIM

We aimed to quantify the behaviour of NENs under solid-organ transplant-related immunosuppression.

DESIGN

This was an observational, retrospective case series.

METHODS

Ten patients were identified from a prospectively kept database. Three were excluded.

RESULTS

Four patients received a liver, two a kidney, and one a heart transplant. All but one received calcineurin-based immunosuppression. NENs were found in five patients post-transplant: one had surgery for transverse colonic neuroendocrine carcinoma NEC (pT4N1M0, Ki67 60%), was cancer-free after four years; one had cold biopsy of duodenal NEN (pT1N0M0, Ki67 2%), cancer-free at four months; one 7 mm pancreatic NEN (pT1N0M0), untreated and stable for seven years; one small-bowel NEN with mesenteric metastasis (pTxNxM1), alive four years after diagnosis; and one untreated small-bowel NEN with mesenteric metastasis, stable at 1 year after liver transplantation. Two NENs were discovered pre-transplant, one pancreatic NEN (pT1N0M0, Ki67 5%), remains untreated and stable at three years. One gastric NEN (type 3, pT1bN0M0, Ki67 2%) remains stable without treatment for two years.

CONCLUSIONS

NENs demonstrate indolent behaviour in the presence of transplant-related immunosuppression.

A mobile-optimized artificial intelligence system for gestational age and fetal malpresentation assessment

Background

Fetal ultrasound is an important component of antenatal care, but shortage of adequately trained healthcare workers has limited its adoption in low-to-middle-income countries. This study investigated the use of artificial intelligence for fetal ultrasound in under-resourced settings.

Methods

Blind sweep ultrasounds, consisting of six freehand ultrasound sweeps, were collected by sonographers in the USA and Zambia, and novice operators in Zambia. We developed artificial intelligence (AI) models that used blind sweeps to predict gestational age (GA) and fetal malpresentation. AI GA estimates and standard fetal biometry estimates were compared to a previously established ground truth, and evaluated for difference in absolute error. Fetal malpresentation (non-cephalic vs cephalic) was compared to sonographer assessment. On-device AI model run-times were benchmarked on Android mobile phones.

Results

Here we show that GA estimation accuracy of the AI model is non-inferior to standard fetal biometry estimates (error difference -1.4 ± 4.5 days, 95% CI -1.8, -0.9, n = 406). Non-inferiority is maintained when blind sweeps are acquired by novice operators performing only two of six sweep motion types. Fetal malpresentation AUC-ROC is 0.977 (95% CI, 0.949, 1.00, n = 613), sonographers and novices have similar AUC-ROC. Software run-times on mobile phones for both diagnostic models are less than 3 s after completion of a sweep.

Conclusions

The gestational age model is non-inferior to the clinical standard and the fetal malpresentation model has high AUC-ROCs across operators and devices. Our AI models are able to run on-device, without internet connectivity, and provide feedback scores to assist in upleveling the capabilities of lightly trained ultrasound operators in low resource settings.

Evaluation of dentinal crack propagation, amount of gutta percha remaining and time required during removal of gutta percha using two different rotary instruments and hand instruments - An In vitro study

Background

One of the most important goal of non surgical endodontic retreatment is the successful removal of gutta percha and sealers from the root canal system. A variety of techniques have been recommended for retreatment procedures for the removal of gutta-percha and sealers with or without the help of adjunctive chemical solvents, by using stainless steel hand files or nickel-titanium rotary files, gate glidden burs, heated instruments, ultrasonic instruments, and lasers 1,3. The current advancements in the design of NiTi instruments have proved efficacious in the removal of filling materials from the root canal wall and various studies have also confirmed their cleaning ability and efficacy 4,5. Nevertheless, the use of rotary instrumentation can lead to the formation of dentinal cracks in the root canal dentin. Many researchers have reported the incidence of crack formation and propagation after the procedure with manual, rotary and reciprocating instruments. The behavior of rotary instruments in the generation of defects have been the point of greatest interest during many years 6. These dentinal cracks can be defined as defects with a complete crack line extending from inner root canal space up to the outer surface of the root when the tensile stress in the root canal wall exceeds the tensile stress of dentin 7.

Aim

This study was conducted to investigate and to compare the amount of dentinal microcracks formation with various new instrumentation methods and conventional hand filing method. It also looks into amount of gutta-percha removed after retreatment from the canal and the time required for all the instrumentation technique.

Methodology

Sixty extracted human maxillary first molars with curved roots were mounted on addition silicone impression material incorporated in an aluminium hollow block, then instrumented using step-back preparation with 35 size K files. Obturation was done using gutta-percha with AH plus sealer. These were stored for 14 days and divided into three groups Mani GPR, Endostar Re Endo and H file and were subjected to retreatment procedures. Retreatment was considered complete when no filling material was observed on the canal wall and the canal was smooth and free of visible debris. The samples were examined under scanning electron microscope and the number of cracks were calculated. The percentage of root canal filling material and time taken was recorded.

Statistical Analysis

The data obtained were analyzed by using descriptive statistics, ANOVA (Analysis of Variance), chi-square test and Scheffe's post hoc test through SPSS for window (version 22.0).

Result

All the techniques showed similar amount of crack propagation, with no statistical difference between the group. Retreatment done using H Files required more time and removed less material. The coronal third showed less amount of gutta-percha remnants than the apical third in all groups.

Conclusion

All the groups showed a similar amount of crack propagation. Less number of cracks were observed in the coronal one third and more amount of cracks were found at the apical third. Endostar RE Endo rotary instrument proved to be most effective and least time-consuming. Hedstrom Files required more time and removed less material.

Adverse neonatal outcome in twin pregnancy complicated by small-for-gestational age: twin vs singleton reference charts

OBJECTIVE

The use of twin-specific vs singleton growth charts in the assessment of twin pregnancy has been controversial. The aim of this study was to assess whether a diagnosis of small-for-gestational age (SGA) made using twin-specific estimated-fetal-weight (EFW) and birth-weight (BW) charts is associated more strongly with adverse neonatal outcomes in twin pregnancies, compared with when the diagnosis is made using singleton charts.

METHODS

This was a cohort study of twin pregnancies delivered at St George's Hospital, London, between January 2007 and May 2020. Twin pregnancies complicated by intrauterine death of one or both twins, fetal aneuploidy or major abnormality, twin-twin transfusion syndrome or twin anemia-polycythemia sequence and those delivered before 32 weeks' gestation, were excluded. SGA was defined as EFW or BW below the 10^th centile, and was assessed using both twin-specific and singleton EFW and BW charts. The main study outcome was composite adverse neonatal outcome. Mixed-effects logistic regression analysis with random pregnancy-level intercepts was used to test the association between SGA classified using the different charts and adverse neonatal outcome.

RESULTS

A total of 1329 twin pregnancies were identified, of which 913 (1826 infants) were included in the analysis. Of these pregnancies, 723 (79.2%) were dichorionic and 190 (20.8%) were monochorionic. Using the singleton charts, 33.3% and 35.7% of pregnancies were classified as SGA based on EFW and BW, respectively. The corresponding values were 5.9% and 5.6% when using the twin-specific charts. Classification as SGA based on EFW using the twin charts was associated significantly with composite adverse neonatal outcome (odds ratio (OR), 4.78 (95% CI, 1.47-14.7); P = 0.007), as compared with classification as appropriate-for-gestational age (AGA). However, classification as SGA based on EFW using the singleton standard was not associated significantly with composite adverse neonatal outcome (OR, 1.36 (95% CI, 0.63-2.88); P = 0.424). Classification as SGA based on EFW using twin-specific standards provided a significantly better model fit than did using the singleton standard (likelihood ratio test, P < 0.001). When twin-specific charts were used, classification as SGA based on BW was associated significantly with a 9.3 times increased odds of composite adverse neonatal outcome (OR, 9.27 (95% CI, 2.86-30.0); P < 0.001). Neonates classified as SGA according to the singleton BW standard but not according to the twin-specific BW standards had a significantly lower rate of composite adverse neonatal outcome than did AGA twins (OR, 0.24 (95% CI, 0.07-0.66); P = 0.009).

CONCLUSIONS

The singleton charts classified one-third of twins as SGA, both prenatally and postnatally. Infants classified as SGA according to the twin-specific charts, but not those classified as SGA according to the singleton charts, had a significantly increased risk of adverse neonatal outcome compared with infants classified as AGA. This study provides further evidence that twin-specific charts perform better than do singleton charts in the prediction of adverse neonatal outcome in twin pregnancies. The use of these charts may reduce misclassification of twins as SGA and improve identification of those that are truly growth restricted. © 2021 International Society of Ultrasound in Obstetrics and Gynecology.

Plantar pressure distribution profile of type 2 diabetes mellitus with diabetic foot syndrome: A hospital-based observational study

INTRODUCTION: The prevalence of diabetic foot syndrome is increasing in the Indian population. It is a triad of neurological, vascular, and biomechanical changes due to long term diabetes mellitus (DM). Altered plantar pressure distribution is a risk factor for developing diabetic foot ulcers. The purpose of this study is to evaluate the altered plantar pressure distribution in diabetic peripheral neuropathy individuals with peripheral vascular disease in comparison with diabetic peripheral neuropathy and non-diabetic neuropathy. Therefore, the objective of the study is to evaluate the plantar pressure distribution and parameters in type 2 DM (T2DM) with diabetic foot syndrome.METHODS: In this study, 60 participants with T2DM were recruited in either one of the three groups: 20 Diabetic non-neuropathy (DNN), 20 Diabetic peripheral neuropathy (DPN), and 20 Diabetic peripheral neuropathy with the peripheral arterial disease (DNPAD). Then we compared the plantar pressure parameters like maximum plantar pressure, pressure-time integral, fore foot-hind foot ratio, and total contact area were measured using a WinTrack plantar pressure system.RESULTS: The diabetic peripheral neuropathy with peripheral vascular disease group shown a significant difference in all the  plantar pressure parameters measured in comparison with the diabetic peripheral neuropathy group. There was increased Maximum plantar pressure, pressure-time integral and forefoot hindfoot ratio, and reduced total contact area of the foot (p < 0.05).CONCLUSION: For patients, with combined peripheral neuropathy and peripheral vascular disease have increased plantar pressure distribution, are at higher risks of developing neuro-ischemic foot, which further leads to diabetic foot ulcers.

Over 60 years of age as an independent prognostic factor of in-hospital mortality among COVID-19 patients: a cohort study in an Iranian high-incidence area

Objectives

COVID-19 continues to cause devastation throughout the world. Various factors influence the perioperative course and prognosis of COVID-19. This study aims to collate the independent prognostic factors among hospitalised COVID-19 patients in east Iran.

Study design

In this cohort study, all patients with a confirmed diagnosis of COVID-19 between 19 February 2020 and 1 August 2020 who were admitted to nine public hospitals of South Khorasan province, Iran, were enrolled.

Methods

Univariate analysis (chi-square [χ²], and Mann–Whitney U test) and multiple logistic regression were performed.

Results

This study included 1290 participants; 676 patients (52.4%) were male. A total of 1189 (92.2%) recovered, and 101 (7.8%) died. The results show that in-hospital mortality increases with advanced age (the optimal cut-off point = 62 years). The following three variables were shown to have the most significant role in in-hospital mortality: age >60 years (odds ratio [OR] = 8.01, 95% confidence interval [CI] 4.8–13.35), shortness of breath (OR = 2.65, 95% CI: 1.4–69.17) and atypical radiological manifestations in a chest X-ray on admission (OR = 2.16, 95% CI: 1.3–28.64). In the univariate analysis, associated comorbidities, such as cardiovascular diseases, influenced the in-hospital mortality rate, while the same could not be replicated in the multiple variable analysis.

Conclusions

This study revealed the potential predictors of COVID-19 and highlighted the need to be cautious with advanced age and heightened clinical symptoms at the time of admission.

Well-differentiated gastroenteropancreatic G3 NET: findings from a large single centre cohort

Neuroendocrine neoplasms are known to have heterogeneous biological behavior. G3 neuroendocrine tumours (NET G3) are characterized by well-differentiated morphology and Ki67 > 20%. The prognosis of this disease is understood to be intermediate between NET G2 and neuroendocrine carcinoma (NEC). Clinical management of NET G3 is challenging due to limited data to inform treatment strategies. We describe clinical characteristics, treatment, and outcomes in a large single centre cohort of patients with gastroenteropancreatic NET G3. Data was reviewed from 26 cases managed at Queen Elizabeth Hospital, Birmingham, UK, from 2012 to 2019. Most commonly the site of the primary tumour was unknown and majority of cases with identifiable primaries originated in the GI tract. Majority of cases demonstrated somatostatin receptor avidity. Median Ki67 was 30%, and most cases had stage IV disease at diagnosis. Treatment options included surgery, somatostatin analogs (SSA), and chemotherapy with either platinum-based or temozolomide-based regimens. Estimated progression free survival was 4 months following initiation of SSA and 3 months following initiation of chemotherapy. Disease control was observed following treatment in 5/11 patients treated with chemotherapy. Estimated median survival was 19 months; estimated 1 year survival was 60% and estimated 2 year survival was 13%. NET G3 is a heterogeneous group of tumours and patients which commonly have advanced disease at presentation. Prognosis is typically poor, though select cases may respond to treatment with SSA and/or chemotherapy. Further study is needed to compare efficacy of different treatment strategies for this disease.

Deep learning for distinguishing normal versus abnormal chest radiographs and generalization to two unseen diseases tuberculosis and COVID-19

Chest radiography (CXR) is the most widely-used thoracic clinical imaging modality and is crucial for guiding the management of cardiothoracic conditions. The detection of specific CXR findings has been the main focus of several artificial intelligence (AI) systems. However, the wide range of possible CXR abnormalities makes it impractical to detect every possible condition by building multiple separate systems, each of which detects one or more pre-specified conditions. In this work, we developed and evaluated an AI system to classify CXRs as normal or abnormal. For training and tuning the system, we used a de-identified dataset of 248,445 patients from a multi-city hospital network in India. To assess generalizability, we evaluated our system using 6 international datasets from India, China, and the United States. Of these datasets, 4 focused on diseases that the AI was not trained to detect: 2 datasets with tuberculosis and 2 datasets with coronavirus disease 2019. Our results suggest that the AI system trained using a large dataset containing a diverse array of CXR abnormalities generalizes to new patient populations and unseen diseases. In a simulated workflow where the AI system prioritized abnormal cases, the turnaround time for abnormal cases reduced by 7-28%. These results represent an important step towards evaluating whether AI can be safely used to flag cases in a general setting where previously unseen abnormalities exist. Lastly, to facilitate the continued development of AI models for CXR, we release our collected labels for the publicly available dataset.

Twin chorionicity-specific population birth-weight charts adjusted for estimated fetal weight

OBJECTIVES

To construct chorionicity-specific birth-weight reference charts for dichorionic diamniotic (DCDA) and monochorionic diamniotic (MCDA) twin pregnancies, incorporating estimated-fetal-weight (EFW) data in order to adjust for the relationship between suboptimal growth and preterm delivery. An additional aim was to determine if the inclusion of complicated twin pregnancies impacts on the reference charts produced.

METHODS

The inclusion criteria for this retrospective cohort study were twin pregnancy of known DCDA or MCDA chorionicity, known pregnancy outcome, last ultrasound scan within 14 days before birth and delivery between 25 and 38 weeks' gestation (Analysis A). An analysis was also conducted excluding pregnancies with complications recorded (Analysis B). Previously published twin EFW reference ranges were used in the analysis. A joint statistical model for EFW and observed birth weight for each pregnancy was created in order to estimate population birth-weight reference ranges corresponding to the distribution expected if all pregnancies delivered at any given gestational age. It was not assumed that the median EFW was equal to birth weight for any given gestational age. The models were fitted using a Bayesian approach.

RESULTS

We retrieved data on 1664 twin pregnancies, of which 707 DCDA and 241 MCDA pregnancies met the inclusion criteria. In Analysis A, the estimated population median birth weight was similar to the median EFW at around 27 weeks' gestation but fell below the EFW values with increasing gestation, being 156 g lower in both DCDA and MCDA pregnancies at 35 weeks; this finding was confirmed by direct comparison of the last EFW and birth-weight values in each pregnancy. When the analysis was repeated after excluding complicated twin pregnancies (Analysis B), compared with Analysis A, there was very little difference in the median birth-weight results obtained across gestation. The largest absolute difference between Analyses A and B for DCDA twins was at 31, 32 and 33 weeks, with a 9-g lower median birth weight in Analysis A compared with Analysis B. The largest absolute difference for MCDA twins was greater than that for DCDA twins, with a 21-g lower median birth weight at 25 weeks in Analysis A compared with Analysis B.

CONCLUSIONS

We have established population chorionicity-specific birth-weight reference charts for DCDA and MCDA twin pregnancies, corresponding to the range expected were all pregnancies to deliver at any given gestational age. In this population of twins, the median birth weight was consistently lower than that reported for singletons, and there was variation in the median birth weight at different gestational ages according to chorionicity. © 2021 The Authors. Ultrasound in Obstetrics & Gynecology published by John Wiley & Sons Ltd on behalf of International Society of Ultrasound in Obstetrics and Gynecology. - Legal Statement: This is an open access article under the terms of the Creative Commons Attribution-NonCommercial-NoDerivs License, which permits use and distribution in any medium, provided the original work is properly cited, the use is non-commercial and no modifications or adaptations are made.

Improving reference standards for validation of AI-based radiography

Objective:

Demonstrate the importance of combining multiple readers' opinions, in a context-aware manner, when establishing the reference standard for validation of artificial intelligence (AI) applications for, e.g. chest radiographs. By comparing individual readers, majority vote of a panel, and panel-based discussion, we identify methods which maximize interobserver agreement and label reproducibility.

Methods:

1100 frontal chest radiographs were evaluated for 6 findings: airspace opacity, cardiomegaly, pulmonary edema, fracture, nodules, and pneumothorax. Each image was reviewed by six radiologists, first individually and then via asynchronous adjudication (web-based discussion) in two panels of three readers to resolve disagreements within each panel. We quantified the reproducibility of each method by measuring interreader agreement.

Results:

Panel-based majority vote improved agreement relative to individual readers for all findings. Most disagreements were resolved with two rounds of adjudication, which further improved reproducibility for some findings, particularly reducing misses. Improvements varied across finding categories, with adjudication improving agreement for cardiomegaly, fractures, and pneumothorax.

Conclusion:

The likelihood of interreader agreement, even within panels of US board-certified radiologists, must be considered before reads can be used as a reference standard for validation of proposed AI tools. Agreement and, by extension, reproducibility can be improved by applying majority vote, maximum sensitivity, or asynchronous adjudication for different findings, which supports the development of higher quality clinical research.

Advances in knowledge:

A panel of three experts is a common technique for establishing reference standards when ground truth is not available for use in AI validation. The manner in which differing opinions are resolved is shown to be important, and has not been previously explored.

Developing a reporting guideline for artificial intelligence-centred diagnostic test accuracy studies: the STARD-AI protocol

INTRODUCTION

Standards for Reporting of Diagnostic Accuracy Study (STARD) was developed to improve the completeness and transparency of reporting in studies investigating diagnostic test accuracy. However, its current form, STARD 2015 does not address the issues and challenges raised by artificial intelligence (AI)-centred interventions. As such, we propose an AI-specific version of the STARD checklist (STARD-AI), which focuses on the reporting of AI diagnostic test accuracy studies. This paper describes the methods that will be used to develop STARD-AI.

METHODS AND ANALYSIS

The development of the STARD-AI checklist can be distilled into six stages. (1) A project organisation phase has been undertaken, during which a Project Team and a Steering Committee were established; (2) An item generation process has been completed following a literature review, a patient and public involvement and engagement exercise and an online scoping survey of international experts; (3) A three-round modified Delphi consensus methodology is underway, which will culminate in a teleconference consensus meeting of experts; (4) Thereafter, the Project Team will draft the initial STARD-AI checklist and the accompanying documents; (5) A piloting phase among expert users will be undertaken to identify items which are either unclear or missing. This process, consisting of surveys and semistructured interviews, will contribute towards the explanation and elaboration document and (6) On finalisation of the manuscripts, the group's efforts turn towards an organised dissemination and implementation strategy to maximise end-user adoption.

ETHICS AND DISSEMINATION

Ethical approval has been granted by the Joint Research Compliance Office at Imperial College London (reference number: 19IC5679). A dissemination strategy will be aimed towards five groups of stakeholders: (1) academia, (2) policy, (3) guidelines and regulation, (4) industry and (5) public and non-specific stakeholders. We anticipate that dissemination will take place in Q3 of 2021.

Association of UGT1A6 gene polymorphism with clinical outcome in pediatric epileptic patients on sodium valproate monotherapy

Pediatric epilepsy comprises chronic neurological disorders characterized by recurrent seizures. Sodium valproate is one of the common antiseizure medications used for treatment. Glucuronide conjugation is the major metabolic pathway of sodium valproate, carried out by the enzyme uridine 5'-diphosphate (UDP) glucuronosyl transferase (UGT) whose gene polymorphisms may alter the clinical outcome. The objective of this study was to assess the association between UGT1A6 genetic polymorphism and clinical outcome in terms of efficacy and tolerability in pediatric epileptic patients on sodium valproate monotherapy. Pediatric epileptic patients (n=65) aged 2-18 years receiving sodium valproate monotherapy for the past one month were included. Genetic polymorphism patterns of UGT1A6 (T19G, A541G, A552C) were evaluated by PCR-RFLP. Clinical outcome was seizure control during the 6 months observation period. Tolerability was measured by estimating the hepatic, renal, and other lab parameters. Out of 65 patients, TT (40%), TG (57%), and GG (3%) patterns were observed in UGT1A6 (T19G) gene, AA (51%), AG (40%), and GG (9%) in (A541G) gene, and AA (43%), AC (43%), and CC (14%) in (A552C) gene. No statistical difference in clinical outcome was found for different UGT1A6 genetic polymorphism patterns. We concluded that different patterns of UGT1A6 genetic polymorphism were not associated with the clinical outcome of sodium valproate in terms of efficacy and tolerability. Sodium valproate was well-tolerated among pediatric patients with epilepsy and can be used as an effective antiseizure medication.

769 Analysing the Change – Outcomes and Benefits of Clinic-Based Removal of Renal Transplant Stents Using A Disposable Cystoscope

Introduction

Literature is chock-full of data regarding the “when” of ureteric stent removal after renal transplantation. We have attempted to address the “who”, “where” and “how” components.

Method

The Isiris disposable scope was used to remove renal allograft stents from 383 patients in the Transplant Clinic from June 2018 to April 2020. An advanced nurse practitioner was trained in the procedure. The learning curve, incidence of complications, benefits and cost implications were studied, and compared with the cohort having stent removal with a traditional flexible cystoscope in theatres.

Results

There were 14 failures necessitating theatre removal. The transplant-to-stent-removal interval was significantly lower in the clinic cohort with a mean of 38.80 days (95%CI 37.26–40.34) to 46.55 days in theatres (95%CI 43.47–49.62). 11 patients had urgent bedside stent removal. The service was delivered independently by the nurse for 5.3% of the patients in June 2018 and progressed to over 80% by April 2019. Estimated net gain was £919/patient.

Conclusions

Moving transplant ureteric stent removals from a resource intensive all-day process in theatres to a one-stop event in the clinic is a safe and economical model that can streamline flow in patient pathway and inculcate new skills in other members of the multidisciplinary team.

Impact of acute kidney injury on in-hospital outcomes among patients hospitalized with acute heart failure: a propensity score matched analysis

Background

Acute kidney injury (AKI) is frequently present in patients admitted for acute heart failure (AHF). Several studies have evaluated the mortality risk and have concluded poor prognosis in any patient with AKI admitted for AHF. For the most part, the additional morbidity and mortality burden in AHF patients with AKI has been attributed to the concomitant comorbidities, and/or interventions.

Purpose

We sought to determine the impact of acute kidney injury (AKI) on in-hospital outcomes in patients presenting with acute heart failure (AHF). We identified isolated AKI patients after excluding other concomitant diagnoses and procedures, which may contribute to an increased risk of mortality and morbidity.

Methods

Data from the National Inpatient Sample (2012- 14) were used to identify patients with the principal diagnosis of AHF and the concomitant secondary diagnosis of AKI. Propensity score matching was performed on 30 baseline variables to identify a matched cohort. The outcome of interest was in-hospital mortality. We further evaluated in-hospital procedures and complications.

Results

Of 1,470,450 patients admitted with AHF, 24.3% had AKI. After propensity matching a matched cohort of 356,940 patients was identified. In this matched group, the AKI group had significantly higher in-hospital mortality (3.8% vs 1.7%, p&lt;0.001). Complications such as sepsis and cardiac arrest were higher in the AKI group. Similarly, in-hospital procedures including CABG, mechanical ventilation and IABP were performed more in the AKI group. AHF patients with AKI had longer in-hospital stay of ∼1.7 days.

Conclusions

In a propensity score-matched cohort of AHF with and without AKI, the risk of in-hospital mortality was &gt;2-fold in the AKI group. Healthcare utilization and burden of complications were higher in the AKI group.

Funding Acknowledgement

Type of funding source: None

Impact of prolotherapy in temporomandibular joint disorder: a quality of life assessment

Chronic pain and functional impairment associated with temporomandibular joint (TMJ) disorders (TMD) considerably reduce oral health-related quality of life (OHRQoL). In the present study we have assessed the influence of prolotherapy in patients with TMD by the subjective measurement of QoL using the Oral Health Impact Profile-14 (OHIP-14). Twenty-five patients diagnosed with TMD (mean (range) age 38 (18 - 70) years) were included. They had all undergone dextrose prolotherapy to the TMJ at regular time intervals (four times at intervals of two weeks) using the method suggested by Hemwall-Hackett. They were asked to answer the OHIP-14 questionnaire before and two years after prolotherapy. Seven domains of OHRQoL were rated on a 5-point Likert scale from 0 (never) to 4 (very often). Domain scores and total OHIP-14 scores were compared using inferential statistics (chi squared and Wilcoxon signed rank tests). Prolotherapy was effective over time, as all the domains' mean scores decreased considerably after treatment. The total mean score before prolotherapy was 21.20, which was extensively reduced to 13.08 after prolotherapy (p=0.001). There was statistically significant improvement in all domains, including functional limitation, physical pain, psychological discomfort, physical disability, psychological disability, social disability, and handicap (p≤0.005 in all cases). We concluded that prolotherapy has a promising role in the improvement of OHRQoL of patients with TMD, and its beneficial effects persist at least two years after treatment.

Consensus guidelines for interventional cardiology services delivery during covid-19 pandemic in Australia and new Zealand

The global coronavirus disease (COVID-19) pandemic poses an unprecedented stress on healthcare systems internationally. These Health system-wide demands call for efficient utilisation of resources at this time in a fair, consistent, ethical and efficient manner would improve our ability to treat patients. Excellent co-operation between hospital units (especially intensive care unit [ICU], emergency department [ED] and cardiology) is critical in ensuring optimal patient outcomes. The purpose of this document is to provide practical guidelines for the effective use of interventional cardiology services in Australia and New Zealand. The document will be updated regularly as new evidence and knowledge is gained with time. Goals Considerations.

International Nosocomial Infection Control Consortium (INICC) report, data summary of 45 countries for 2012-2017: Device-associated module

BACKGROUND

We report the results of International Nosocomial Infection Control Consortium (INICC) surveillance study from January 2012 to December 2017 in 523 intensive care units (ICUs) in 45 countries from Latin America, Europe, Eastern Mediterranean, Southeast Asia, and Western Pacific.

METHODS

During the 6-year study period, prospective data from 532,483 ICU patients hospitalized in 242 hospitals, for an aggregate of 2,197,304 patient days, were collected through the INICC Surveillance Online System (ISOS). The Centers for Disease Control and Prevention-National Healthcare Safety Network (CDC-NHSN) definitions for device-associated health care-associated infection (DA-HAI) were applied.

RESULTS

Although device use in INICC ICUs was similar to that reported from CDC-NHSN ICUs, DA-HAI rates were higher in the INICC ICUs: in the medical-surgical ICUs, the pooled central line-associated bloodstream infection rate was higher (5.05 vs 0.8 per 1,000 central line-days); the ventilator-associated pneumonia rate was also higher (14.1 vs 0.9 per 1,000 ventilator-days,), as well as the rate of catheter-associated urinary tract infection (5.1 vs 1.7 per 1,000 catheter-days). From blood cultures samples, frequencies of resistance, such as of Pseudomonas aeruginosa to piperacillin-tazobactam (33.0% vs 18.3%), were also higher.

CONCLUSIONS

Despite a significant trend toward the reduction in INICC ICUs, DA-HAI rates are still much higher compared with CDC-NHSN's ICUs representing the developed world. It is INICC's main goal to provide basic and cost-effective resources, through the INICC Surveillance Online System to tackle the burden of DA-HAIs effectively.

International evaluation of an AI system for breast cancer screening

Screening mammography aims to identify breast cancer at earlier stages of the disease, when treatment can be more successful1. Despite the existence of screening programmes worldwide, the interpretation of mammograms is affected by high rates of false positives and false negatives2. Here we present an artificial intelligence (AI) system that is capable of surpassing human experts in breast cancer prediction. To assess its performance in the clinical setting, we curated a large representative dataset from the UK and a large enriched dataset from the USA. We show an absolute reduction of 5.7% and 1.2% (USA and UK) in false positives and 9.4% and 2.7% in false negatives. We provide evidence of the ability of the system to generalize from the UK to the USA. In an independent study of six radiologists, the AI system outperformed all of the human readers: the area under the receiver operating characteristic curve (AUC-ROC) for the AI system was greater than the AUC-ROC for the average radiologist by an absolute margin of 11.5%. We ran a simulation in which the AI system participated in the double-reading process that is used in the UK, and found that the AI system maintained non-inferior performance and reduced the workload of the second reader by 88%. This robust assessment of the AI system paves the way for clinical trials to improve the accuracy and efficiency of breast cancer screening.

Chest Radiograph Interpretation with Deep Learning Models: Assessment with Radiologist-adjudicated Reference Standards and Population-adjusted Evaluation

BackgroundDeep learning has the potential to augment the use of chest radiography in clinical radiology, but challenges include poor generalizability, spectrum bias, and difficulty comparing across studies.PurposeTo develop and evaluate deep learning models for chest radiograph interpretation by using radiologist-adjudicated reference standards.Materials and MethodsDeep learning models were developed to detect four findings (pneumothorax, opacity, nodule or mass, and fracture) on frontal chest radiographs. This retrospective study used two data sets. Data set 1 (DS1) consisted of 759 611 images from a multicity hospital network and ChestX-ray14 is a publicly available data set with 112 120 images. Natural language processing and expert review of a subset of images provided labels for 657 954 training images. Test sets consisted of 1818 and 1962 images from DS1 and ChestX-ray14, respectively. Reference standards were defined by radiologist-adjudicated image review. Performance was evaluated by area under the receiver operating characteristic curve analysis, sensitivity, specificity, and positive predictive value. Four radiologists reviewed test set images for performance comparison. Inverse probability weighting was applied to DS1 to account for positive radiograph enrichment and estimate population-level performance.ResultsIn DS1, population-adjusted areas under the receiver operating characteristic curve for pneumothorax, nodule or mass, airspace opacity, and fracture were, respectively, 0.95 (95% confidence interval [CI]: 0.91, 0.99), 0.72 (95% CI: 0.66, 0.77), 0.91 (95% CI: 0.88, 0.93), and 0.86 (95% CI: 0.79, 0.92). With ChestX-ray14, areas under the receiver operating characteristic curve were 0.94 (95% CI: 0.93, 0.96), 0.91 (95% CI: 0.89, 0.93), 0.94 (95% CI: 0.93, 0.95), and 0.81 (95% CI: 0.75, 0.86), respectively.ConclusionExpert-level models for detecting clinically relevant chest radiograph findings were developed for this study by using adjudicated reference standards and with population-level performance estimation. Radiologist-adjudicated labels for 2412 ChestX-ray14 validation set images and 1962 test set images are provided.© RSNA, 2019Online supplemental material is available for this article.See also the editorial by Chang in this issue.

P4988Beta-blockers in diabetes with stable coronary heart disease - a cause for concern?

Background

Beta-blocker (BB) therapy is a cornerstone for the treatment of coronary heart disease (CHD). The evidence of the benefit from long-term BB therapy in diabetic patients with stable CHD is scarce. This meta-analysis summarises the evidence relating to the BB therapy in diabetic patients with stable CHD.

Methods

A meta-analysis was performed according to PRISMA and MOOSE guidelines for reporting of systematic reviews of observational studies. PubMed, Embase, and Cochrane central were searched and two authors independently screened studies for eligibility. The quality of studies was assessed with the Newcastle Ottawa scale. The primary outcome of interest was all-cause mortality, cardiovascular (CV) mortality and major adverse cardiovascular events (MACE) in diabetic patients with and without BB therapy. A generic inverse variance model was used to pool the odds ratio or hazards ratio from included studies to calculate the overall effect estimate. The significance threshold was set at p-value <0.05. Heterogeneity was assessed by I2.

Results

Four non-randomized studies with 9,515 participants were selected for the analyses. Four studies were post-hoc analyses of randomised controlled trials, and 1 article was an analysis of a nationally representative survey. In a fixed effects model, BB therapy in diabetic patients with stable CHD was found to be associated with increased risk of CV mortality, and MACE (27%, and 32% respectively; p-value <0.05) and was not associated with a reduction in all-cause mortality (HR 1.12; 95% CI 0.94–1.33; p-value =0.22).

Conclusion

BB therapy in diabetic patients with stable CHD appears to be linked to higher mortality. Large randomised trials are needed in this population to confirm these findings.

Acknowledgement/Funding

None

P6535Atrial fibrillation and its impact in hospitalised cancer patients

Background

Atrial fibrillation (AF) is associated with increased morbidity and mortality. However, its impact on outcomes of cancer patients is largely unknown. We sought to determine the prevalence and effect of AF on in-hospital outcomes of patients admitted with AF in the United States.

Methods

We obtained and analyzed data from Nationwide Inpatient Sample (NIS 2002- 2014). The study population included all adults (>18 years old) with the discharge diagnosis of cancer and the associated secondary diagnosis of AF. The associations of AF with in-hospital complications, procedures, discharge outcomes, and mortality, were analyzed after adjusting for potential confounders using logistic regression analyses.

Results

During the study period, 12,410,290 patients were admitted with cancer; of which, 1,013,735 (8.2%) had AF. Patients with AF were likely to be male (51.1% vs 42.7%), white (86.3% vs 73.6%), and had significantly higher rates comorbidities including hypertension (59.6% vs 44.6%), diabetes (20.9% vs 15.8%), heart failure (23.8% vs 4.5%), and stroke (1.3% vs 0.7%). Similarly, in-hospital complications (including infections, venous thromboembolism, gastrointestinal bleeding, myocardial infarction) and hospital procedures (including cardiac catheterization, intubation, blood transfusion, percutaneous coronary intervention) were found at a significantly higher rate in cancer patients with AF compared to without AF (p-value for all <0.0001). After adjusting for all confounding factors, cancer patients in the AF group had higher odds of in-hospital death (odds ratio, 1.26; 95% confidence interval, 1.23–1.28; P<0.0001). The presence of AF conferred on average three days of extra inpatient stay in patients with cancer. Mortality in the cancer patients admitted with a concomitant diagnosis of AF has gradually increased from 7.2% to 9.7% over the 13-year study period.

Table 1. Multivariate model showing Atrial fibrillation as an independent risk factor of inpatient mortality in cancer patients Unadjusted univariate model Multivariate model adjusted for age, sex, race and potential confounders* Odds ratio (95% confidence interval P-value Odds ratio (95% confidence interval P-value Discharged alive Reference Reference In-hospital mortality 2.08 (2.07, 2.10) <0.0001 1.26 (1.23, 1.28) <0.0001 *Confounders adjusted for include all clinical comorbidities and in-hospital complications and procedures.

Conclusions

In patients with cancer, AF confers significant in-hospital burden by increasing the hospital stay by 3 days. AF is also associated with significantly higher rates of in-hospital procedures and complications in cancer patients and is an independent risk factor of in-hospital mortality.

Acknowledgement/Funding

None

P1571Atrial fibrillation is an independent predictor of mortality in hospitalized cancer patients

Background

Atrial fibrillation (AF) is prevalent in patients with cancer. We performed a study to evaluate the outcomes associated with hospitalised cancer patients with AF.

Methods

The Nationwide Inpatient Sample was used to identify patients with a cancer diagnosis, who were found to have atrial fibrillation from the years 2002–2014. Descriptive statistics for mortality were calculated using univariate and multivariate model for each cancer, and we identified the type of cancer associated with the highest inpatient mortality. We also calculated the percentage of hospital mortality attributable to AF.

Results

12,410,290 (nationwide estimate) patients with a cancer diagnosis were identified, and 8.2% of them had AF. In a multivariate adjusted model for various relevant comorbidities, age, gender and race, AF was also found to be an independent risk factor for higher in-hospital mortality in cancer patients, odds ratio (OR) 1.25 (95% CI 1.23, 1.28, p<0.0001). Colon cancer was found to be associated with the worst outcomes and the highest mortality. In cancer patients, the incident percentage of hospital mortality attributed to AF was 46.7%, population attributable mortality risk of was 7.17%, and number needed to harm was 21 (p<0.0001).

Conclusion

AF conferred significant morbidity and was found to be an independent risk factor for increased mortality in hospitalised patients with cancer. Colon cancer was found to have the strongest association of worst outcomes in hospitalized patients with AF.

Acknowledgement/Funding

None

P1574Predictors of mortality in hospitalized atrial fibrillation patients with cancer

Background

Recent reports indicate an important interplay between Atrial fibrillation (AF) and cancer. There is little information regarding the outcomes of these patients. Hence, we performed a study to identify predictors of in-hospital mortality to help guide goals of care discussions.

Methods

The Nationwide Inpatient Sample was used to identify patients with a diagnosis of cancer, who were found to have AF from 2002–2014. Trend rate, patients' and hospital characteristics along with in-hospital complications and predictors of in-hospital mortality were assessed. Backward stepwise elimination technique was used to fit the multivariate regression model.

Results

Over the 13-year study period, 12,410,290 (national estimate) patients with a cancer diagnosis were identified. 1,013,735 had AF, and 10.2% of the AF patients with cancer died while hospitalised. A variety of comorbidities, in-hospital procedures and in-hospital complications increased the odds of in-hospital mortality in these patients. Also, weekend admissions, elective admissions, and rural hospitals in comparison to urban teaching and non-teaching hospitals were associated with higher in-hospital mortality.

Conclusion

Stroke, myocardial infarction, pulmonary embolism, deep venous thrombosis, acute kidney injury, congestive heart failure, sepsis, and cardiogenic shock are most significant predictors of in-hospital mortality in AF patients with cancer.

Acknowledgement/Funding

None

End-to-end lung cancer screening with three-dimensional deep learning on low-dose chest computed tomography

With an estimated 160,000 deaths in 2018, lung cancer is the most common cause of cancer death in the United States¹. Lung cancer screening using low-dose computed tomography has been shown to reduce mortality by 20-43% and is now included in US screening guidelines^1-6. Existing challenges include inter-grader variability and high false-positive and false-negative rates^7-10. We propose a deep learning algorithm that uses a patient's current and prior computed tomography volumes to predict the risk of lung cancer. Our model achieves a state-of-the-art performance (94.4% area under the curve) on 6,716 National Lung Cancer Screening Trial cases, and performs similarly on an independent clinical validation set of 1,139 cases. We conducted two reader studies. When prior computed tomography imaging was not available, our model outperformed all six radiologists with absolute reductions of 11% in false positives and 5% in false negatives. Where prior computed tomography imaging was available, the model performance was on-par with the same radiologists. This creates an opportunity to optimize the screening process via computer assistance and automation. While the vast majority of patients remain unscreened, we show the potential for deep learning models to increase the accuracy, consistency and adoption of lung cancer screening worldwide.