We identified a new taxonomy to describe Twitter use in health research with 6 categories. Many data elements discernible from a user's Twitter profile, especially demographics, have been underreported in the literature and can provide new opportunities to characterize the users whose data are analyzed in these studies. Twitter-based health research is a growing field funded by a diversity of organizations. Public health implications. Future work should develop standardized reporting guidelines for health researchers who use Twitter and policies that address privacy and ethical concerns in social media research.

Background. Researchers have used traditional databases to study public health for decades. Less is known about the use of social media data sources, such as Twitter, for this purpose. Objectives. To systematically review the use of Twitter in health research, define a taxonomy to describe Twitter use, and characterize the current state of Twitter in health research. Search methods. We performed a literature search in PubMed, Embase, Web of Science, Google Scholar, and CINAHL through September 2015. Selection criteria. We searched for peer-reviewed original research studies that primarily used Twitter for health research. Data collection and analysis. Two authors independently screened studies and abstracted data related to the approach to analysis of Twitter data, methodology used to study Twitter, and current state of Twitter research by evaluating time of publication, research topic, discussion of ethical concerns, and study funding source. Main results. Of 1110 unique health-related articles mentioning Twitter, 137 met eligibility criteria. The primary approaches for using Twitter in health research that constitute a new taxonomy were content analysis (56%; n = 77), surveillance (26%; n = 36), engagement (14%; n = 19), recruitment (7%; n = 9), intervention (7%; n = 9), and network analysis (4%; n = 5). These studies collectively analyzed more than 5 billion tweets primarily by using the Twitter application program interface. Of 38 potential data features describing tweets and Twitter users, 23 were reported in fewer than 4% of the articles. The Twitter-based studies in this review focused on a small subset of data elements including content analysis, geotags, and language. Most studies were published recently (33% in 2015). Public health (23%; n = 31) and infectious disease (20%; n = 28) were the research fields most commonly represented in the included studies. Approximately one third of the studies mentioned ethical board approval in their articles. Primary funding sources included federal (63%), university (13%), and foundation (6%). Conclusions. We identified a new taxonomy to describe Twitter use in health research with 6 categories. Many data elements discernible from a user’s Twitter profile, especially demographics, have been underreported in the literature and can provide new opportunities to characterize the users whose data are analyzed in these studies. Twitter-based health research is a growing field funded by a diversity of organizations. Public health implications. Future work should develop standardized reporting guidelines for health researchers who use Twitter and policies that address privacy and ethical concerns in social media research.

No abstract

Purpose: Hip fractures are a common cause of morbidity and mortality. Automatic identification and classification of hip fractures using deep learning may improve outcomes by reducing diagnostic errors and decreasing time to operation. Methods: Hip and pelvic radiographs from 1118 studies were reviewed and 3034 hips were labeled via bounding boxes and classified as normal, displaced femoral neck fracture, nondisplaced femoral neck fracture, intertrochanteric fracture, previous ORIF, or previous arthroplasty. A deep learning-based object detection model was trained to automate the placement of the bounding boxes. A Densely Connected Convolutional Neural Network (DenseNet) was trained on a subset of the bounding box images, and its performance evaluated on a held out test set and by comparison on a 100-image subset to two groups of human observers: fellowshiptrained radiologists and orthopaedists, and senior residents in emergency medicine, radiology, and orthopaedics. Results: The binary accuracy for fracture of our model was 93.8% (95% CI, 91.3-95.8%), with sensitivity of 92.7% (95% CI, 88.7-95.6%), and specificity 95.0% (95% CI, 91.5-97.3%). Multiclass classification accuracy was 90.4% (95% CI, 87.4-92.9%). When compared to human observers, our model achieved at least expert-level classification under all conditions. Additionally, when the model was used as an aid, human performance improved, with aided resident performance approximating unaided fellowship-trained expert performance. Conclusions: Our deep learning model identified and classified hip fractures with at least expert-level accuracy, and when used as an aid improved human performance, with aided resident performance approximating that of unaided fellowship-trained attendings.