Analysis of Stroke Detection during the COVID-19 Pandemic Using Natural Language Processing of Radiology Reports

BACKGROUND AND PURPOSE:The coronavirus disease 2019 (COVID-19) pandemic has led to decreases in neuroimaging volume. Our aim was to quantify the change in acute or subacute ischemic strokes detected on CT or MR imaging during the pandemic using natural language processing of radiology reports.MATERIALS AND METHODS:We retrospectively analyzed 32,555 radiology reports from brain CTs and MRIs from a comprehensive stroke center, performed from March 1 to April 30 each year from 2017 to 2020, involving 20,414 unique patients. To detect acute or subacute ischemic stroke in free-text reports, we trained a random forest natural language processing classifier using 1987 randomly sampled radiology reports with manual annotation. Natural language processing classifier generalizability was evaluated using 1974 imaging reports from an external dataset.RESULTS:The natural language processing classifier achieved a 5-fold cross-validation classification accuracy of 0.97 and an F1 score of 0.74, with a slight underestimation (−5%) of actual numbers of acute or subacute ischemic strokes in cross-validation. Importantly, cross-validation performance stratified by year was similar. Applying the classifier to the complete study cohort, we found an estimated 24% decrease in patients with acute or subacute ischemic strokes reported on CT or MR imaging from March to April 2020 compared with the average from those months in 2017–2019. Among patients with stroke-related order indications, the estimated proportion who underwent neuroimaging with acute or subacute ischemic stroke detection significantly increased from 16% during 2017–2019 to 21% in 2020 (P = .01). The natural language processing classifier performed worse on external data.CONCLUSIONS:Acute or subacute ischemic stroke cases detected by neuroimaging decreased during the COVID-19 pandemic, though a higher proportion of studies ordered for stroke were positive for acute or subacute ischemic strokes. Natural language processing approaches can help automatically track acute or subacute ischemic stroke numbers for epidemiologic studies, though local classifier training is important due to radiologist reporting style differences.

There is much concern regarding the impact of the coronavirus disease 2019 (COVID-19) pandemic on the quality of stroke care, including issues with hospital capacity, clinical resource re-allocation, and the safety of patients and clinicians.^1,2 Previous reports have shown that there have been substantial decreases in stroke neuroimaging volume during the pandemic.^3,4 In addition, acute ischemic infarcts have been found on neuroimaging studies in many hospitalized patients with COVID-19, though the causal relationship is unclear.^5,6 Studies like these and other epidemiologic analyses usually rely on the creation of manually curated databases, in which identification of cases can be time-consuming and difficult to update in real-time. One way to facilitate such research is to use natural language processing (NLP), which has shown utility for automated analysis of radiology report data.⁷ NLP algorithms have been developed previously for the classification of neuroradiology reports for the presence of ischemic stroke findings and acute ischemic stroke subtypes.^8,9 Thus, NLP has the potential to facilitate COVID-19 research.In this study, we developed an NLP machine learning model that classifies radiology reports for the presence or absence of acute or subacute ischemic stroke (ASIS), as opposed to chronic stroke. We used this model to quantify the change in ASIS detected on all CT or MR imaging studies performed at a large comprehensive stroke center during the COVID-19 pandemic in the United States. We also evaluated NLP model generalizability and different training strategies using a sample of radiology reports from a second stroke center.     

Can MRI differentiate between atypical cartilaginous tumors and high-grade chondrosarcoma? A systematic review

Background and purpose — Adequate staging of chondroid tumors at diagnosis is important as it determines both treatment and outcome. This systematic review provides an overview of MRI criteria used to differentiate between atypical cartilaginous tumors (ACT) and high-grade chondrosarcoma (HGCS).Patients and methods — For this systematic review PubMed and Embase were searched, from inception of the databases to July 12, 2018. All original articles describing MRI characteristics of pathologically proven primary central chondrosarcoma and ACT were included. A quality appraisal of the included papers was performed. Data on MRI characteristics and histological grade were extracted by 2 reviewers. Meta-analysis was performed if possible. The study is registered with PROSPERO, CRD42018067959.Results — Our search identified 2,132 unique records, of which 14 studies were included. 239 ACT and 140 HGCS were identified. The quality assessment showed great variability in consensus criteria used for both pathologic and radiologic diagnosis. Due to substantial heterogeneity we refrained from pooling the results in a meta-analysis and reported non-statistical syntheses. Loss of entrapped fatty marrow, cortical breakthrough, and extraosseous soft tissue expansion appeared to be present more often in HGCS compared with ACT.Interpretation — This systematic review provides an overview of MRI characteristics used to differentiate between ACT and HGCS. Future studies are needed to develop and assess more reliable imaging methods and/or features to differentiate ACT from HGCS.

The incidence of chondrosarcoma of bone appears to have been increasing during the last decade and is now reported to be the most common primary malignant bone tumor in several countries (Thorkildsen et al. 2018, van Praag et al. 2018). Conventional chondrosarcoma is the most common subtype of chondrosarcoma. Other subtypes of chondrosarcoma (e.g., juxtacortical, mesenchymal, or secondary chondrosarcoma) are rare and show different radiologic appearance and clinical behavior (Bindiganavile et al. 2015).Conventional chondrosarcoma is classified into the histological grades 1 (currently known as atypical cartilaginous tumor [ACT]), 2, and 3. The metastatic potential, and therefore the disease-specific survival, correlates with the histological grade (Fletcher et al. 2013, Laitinen et al. 2018, Thorkildsen et al. 2018). ACTs rarely metastasize and are therefore reclassified as an intermediate type of tumor, not a malignancy (Fletcher et al. 2013). Due to the increase in patients undergoing MRI examinations for joint-related complaints, the incidental detection of ACT has increased substantially (van Praag et al. 2018).With the increasing incidence of ACT, clear radiologic criteria to differentiate ACT from high-grade chondrosarcoma (i.e., grades 2 and 3) become more and more important. Adequate staging of chondroid tumors at diagnosis is important as it determines both treatment and prognosis. High-grade chondrosarcomas behave aggressively. Between 10% and 30% of grade 2 and about 70% of grade 3 chondrosarcomas metastasize (Evans et al. 1977). Hence, high-grade chondrosarcoma (HGCS) requires wide en bloc resection with free surgical margins. In contrast, ACTs are intermediate tumors and can be treated either with intralesional curettage and local adjuvant or nonoperatively with regular follow-up when located in the long bones (Deckers et al. 2016).Due to the heterogenous composition of chondroid tumors, diagnostic biopsy is unreliable in assessing the genuine histological grade and malignant potential of chondrosarcomas (Laitinen et al. 2018). Therefore, physicians need to rely on imaging and clinical findings (e.g., pain is more common in HGCS) to differentiate ACT from HGCS. Imaging evaluation of cartilaginous and other bone tumors is generally based on multimodal assessment including at least conventional radiography and MRI (Nascimento et al. 2014).During the most recent decades research has focused mainly on differentiating enchondroma from chondrosarcoma (Choi et al. 2013, Douis et al. 2014, Crim et al. 2015, Lisson et al. 2018). New insights have shown that both enchondroma and ACT located in the long bones can be observed without treatment (Deckers et al. 2016, Sampath Kumar et al. 2016, Chung et al. 2018). These insights make the differentiation between ACT and HGCS clinically relevant. Currently, literature on differentiating ACT from HGCS is sparse and clear radiologic criteria are lacking. Therefore, we performed a systematic review to provide an overview of MRI characteristics used to date to differentiate between ACT and HGCS.     

Jumping in and Out of the Dirty Water… Learning from Stories while Doing Social Science

ABSTRACT

The pursuit of knowledge surrounding health-related issues during disasters, emergencies, and crises, can be delicate and challenging. Social scientists use a host of research methods to design and execute studies with the goal of making intellectual contributions. During extended field work following Hurricane Harvey in the Greater Houston area, our team collected data – interviews, observations, and private social media – from citizens, emergency responders, and volunteer rescuers. Yet sometimes the data collected, analyzed, and reported in published findings is only part of the research story. The researchers’ experiences, both in the field as well as their past, can serve as personal-sensemaking devices. Integrating these stories can help scientists build trust and collect meaningful data, well beyond what is anticipated. In this essay, I share such examples, related to dirty water: temporarily health-compromised individuals, and responders doing double duty. Below the surface, there are many more opportunities for health communication to make an impact in times of crisis.     

Mind the gap: Teachers’ conceptions of student-staff partnership and its potential to enhance educational quality

Abstract

Introduction: Student-staff partnerships as a concept to improve medical education have received a growing amount of attention. Such partnerships are collaborations in which students and teachers seek to improve education by each adding their unique contribution to decision-making and implementation processes. Although previous research has demonstrated that students are favourable to this concept, teachers remain hesitant. The present study investigated teachers’ conceptions of student-staff partnerships and of the prerequisites that are necessary to render such partnerships successful and enhance educational quality.

Method: We conducted semi-structured interviews with 14 course coordinators who lead course design teams and also teach in 4 bachelor health programmes, using Bovill and Bulley’s levels of student participation as sensitising concepts during data analysis.

Results: The results pointed to three different conceptions of student-staff partnerships existing among teachers: Teachers teach and students study; teachers teach and value students’ feedback; and teachers and students co-create. The prerequisites for effective co-creation teachers identified were: Teachers must be open to involve students and create dialogues; students must be motivated and have good communication skills; the organisation must be supportive; and teachers should have the final say.

Conclusion: We conclude that teachers’ conceptions are consistent with Bovill and Bulley’s levels of student participation. Under certain conditions, teachers are willing to co-create and reach the highest levels of student participation.     

Building on the success of anti-vascular endothelial growth factor therapy: a vision for the next decade

This article aims to identify key opportunities for improvement in the diagnosis and treatment of retinal disease, and describe recent innovations that will potentially facilitate improved outcomes with existing intravitreal vascular endothelial growth factor (VEGF) therapies and lay the groundwork for new treatment approaches. The review begins with a summary of the key discoveries that led to the development of anti-VEGF therapies and briefly reviews their impact on clinical practice. Opportunities for improvements in diagnosis, real-world outcomes with existing therapies, long-acting therapeutics and personalised health care are discussed, as well as the need to identify new targets for therapeutic intervention. Low-cost, remote patient screening and monitoring using artificial intelligence (AI)-based technologies can help improve diagnosis rates and enable remote disease monitoring with minimal patient burden. AI-based tools can be applied to generate patient-level prognostic data and predict individual treatment needs, reducing the time needed to optimise a patient’s treatment regimen. Long-acting therapeutics can help improve visual outcomes by reducing the treatment burden. When paired with AI-generated prognoses, long-acting therapeutics enable the possibility of vision loss prevention. Dual-acting drugs may help improve efficacy and/or durability beyond what is possible with anti-VEGF agents alone. Recent developments and ongoing innovations will help build upon the success of anti-VEGF therapies to further reduce vision loss owing to retinal disease while lowering the overall burden of care.Subject terms: Retinal diseases, Quality of life     

Twelve tips for developing and implementing a medical education Twitter chat

Abstract

Live discussions on the social media site Twitter or Twitter chats are gaining popularity as powerful tools for engaging a broad audience in an interactive discussion. Medical education, in particular, is experiencing an increase in the use of this modality to support informal learning, as a means to encourage collaboration and share best practices, and as a platform for large-scale mentorship. Despite this growth in popularity, there are limited data to guide medical educators on the fundamentals of organizing a Twitter chat. In this Twelve Tips article, we discuss strategies relevant to potential Twitter chat organizers. We have arranged the tips chronologically, beginning with a discussion of initial considerations when planning and formulating a chat topic and publicizing the chat to potentially interested people and groups, followed by practical considerations while hosting the chat, and finally strategies for evaluating and extending a Twitter chat’s impact.     

Minority migrant men’s attitudes toward female genital mutilation: Developing strategies to engage men

Abstract

This article explores minority migrant men's attitudes towards female genital mutilation (FGM), and how these attitudes can be used to develop strategies to engage men in the eradication of FGM. Based on interviews and focus group discussions, the article finds that men's attitudes can be enabling, disabling or neutral: the identification of and variations between these need to be taken into account when developing strategies to engage men in the eradication of FGM. There is currently a window of opportunity for involving minority migrant men in the prevention of FGM and in the challenging of a minority migrant gender regime.     

Improvements in balance reaction impairments following reactive balance training in individuals with sub-acute stroke: A prospective cohort study with historical control

ABSTRACT

Background

Reactive balance training (RBT) has been previously found to reduce fall risk in individuals with sub-acute stroke; however, our understanding of the effects of RBT on specific balance impairments is lacking.     

Humans and machines: Moving towards a more symbiotic approach to learning clinical reasoning

Abstract

Artificial intelligence is a growing phenomenon that is driving major changes to how we deliver healthcare. One of its most significant and challenging contributions is likely to be in diagnosis. Artificial intelligence is challenging the physician’s exclusive role in diagnosis and in some areas, its diagnostic accuracy exceeds that of humans. We argue that we urgently need to consider how we will incorporate AI into our teaching of clinical reasoning in the undergraduate curriculum; students need to successfully navigate the benefits and potential issues of new and developing approaches to AI in clinical diagnosis. We offer a pedagogical framework for this challenging change to our curriculum.