Dual-energy X-ray absorptiometry (DXA) is underutilized to measure bone mineral density (BMD) and evaluate fracture risk. We present an automated tool to identify fractures, predict BMD, and evaluate fracture risk using plain radiographs. The tool performance is evaluated on 5164 and 18175 patients with pelvis/lumbar spine radiographs and Hologic DXA. The model is well calibrated with minimal bias in the hip (slope = 0.982, calibration-in-the-large = −0.003) and the lumbar spine BMD (slope = 0.978, calibration-in-the-large = 0.003). The area under the precision-recall curve and accuracy are 0.89 and 91.7% for hip osteoporosis, 0.89 and 86.2% for spine osteoporosis, 0.83 and 95.0% for high 10-year major fracture risk, and 0.96 and 90.0% for high hip fracture risk. The tool classifies 5206 (84.8%) patients with 95% positive or negative predictive value for osteoporosis, compared to 3008 DXA conducted at the same study period. This automated tool may help identify high-risk patients for osteoporosis.

How to produce expressive molecular representations is a fundamental challenge in artificial intelligence-driven drug discovery. Graph neural network (GNN) has emerged as a powerful technique for modeling molecular data. However, previous supervised approaches usually suffer from the scarcity of labeled data and poor generalization capability. Here, we propose a novel molecular pre-training graph-based deep learning framework, named MPG, that learns molecular representations from large-scale unlabeled molecules. In MPG, we proposed a powerful GNN for modelling molecular graph named MolGNet, and designed an effective self-supervised strategy for pre-training the model at both the node and graph-level. After pre-training on 11 million unlabeled molecules, we revealed that MolGNet can capture valuable chemical insights to produce interpretable representation. The pre-trained MolGNet can be fine-tuned with just one additional output layer to create state-of-the-art models for a wide range of drug discovery tasks, including molecular properties prediction, drug-drug interaction and drug-target interaction, on 14 benchmark datasets. The pre-trained MolGNet in MPG has the potential to become an advanced molecular encoder in the drug discovery pipeline.

Background: Internet hospitals in China are in great demand due to limited and unevenly distributed health care resources, lack of family doctors, increased burdens of chronic diseases, and rapid growth of the aged population. The COVID-19 epidemic catalyzed the expansion of online health care services. In recent years, internet hospitals have been rapidly developed. Ping An Good Doctor is the largest, national online medical entry point in China and is a widely used platform providing online health care services. Objective: This study aims to give a comprehensive description of the characteristics of the online consultations and inquisitions in Ping An Good Doctor. The analyses tried to answer the following questions: (1) What are the characteristics of the consultations in Ping An Good Doctor in terms of department and disease profiles? (2) Who uses the online health services most frequently? and (3) How is the user experience of the online consultations of Ping An Good Doctor? Methods: A total of 35.3 million consultations and inquisitions over the course of 1 year were analyzed with respect to the distributions of departments and diseases, user profiles, and consulting behaviors. Results: The geographical distribution of the usage of Ping An Good Doctor showed that Shandong (18.4%), Yunnan (15.6%), Shaanxi (7.2%), and Guangdong (5.5%) were the provinces that used it the most; they accounted for 46.6% of the total consultations and inquisitions. In terms of department distribution, we found that gynecology and obstetrics (19.2%), dermatology (17.0%), and pediatrics (14.4%) were the top three departments in Ping An Good Doctor. The disease distribution analysis showed that, except for nondisease-specific consultations, acute upper respiratory infection (AURI) (4.1%), pregnancy (2.8%), and dermatitis (2.4%) were the most frequently consulted diseases. In terms of user profiles, females (60.4%) from 19 to 35 years of age were most likely to seek consultations online, in general. The user behavior analyses showed that the peak times of day for online consultations occurred at 10 AM, 3 PM, and 9 PM. Regarding user experience, 93.0% of users gave full marks following their consultations. For some disease-related health problems, such as AURI, dermatitis, and eczema, the feedback scores were above average. Conclusions: The prevalence of internet hospitals, such as Ping An Good Doctor, illustrated the great demand for online health care services that can go beyond geographical limitations. Our analyses showed that nondisease-specific issues and moderate health problems were much more frequently consulted about than severe clinical conditions. This indicated that internet hospitals played the role of the family doctor, which helped to relieve the stress placed on offline hospitals and facilitated people’s lives. In addition, good user experiences, especially regarding disease-related inquisitions, suggested that online health services can help solve health problems. With support from the government and acceptance by the public, online health care services could develop at a fast pace and greatly benefit people’s daily lives.

Background: Images from magnetic resonance imaging (MRI) are crucial unstructured data for prognostic evaluation in nasopharyngeal carcinoma (NPC). We developed and validated a prognostic system based on the MRI features and clinical data of locoregionally advanced NPC (LA-NPC) patients to distinguish low-risk patients with LA-NPC for whom concurrent chemoradiotherapy (CCRT) is sufficient. Methods: This multicenter, retrospective study included 3444 patients with LA-NPC from January 1, 2010, to January 31, 2017. A 3-dimensional convolutional neural network was used to learn the image features from pretreatment MRI images. An eXtreme Gradient Boosting model was trained with the MRI features and clinical data to assign an overall score to each patient. Comprehensive evaluations were implemented to assess the performance of the predictive system. We applied the overall score to distinguish high-risk patients from low-risk patients. The clinical benefit of induction chemotherapy (IC) was analyzed in each risk group by survival curves. Results: We constructed a prognostic system displaying a concordance index of 0.776 (95% confidence interval [CI] = 0.746 to 0.806) for the internal validation cohort and 0.757 (95% CI = 0.695 to 0.819), 0.719 (95% CI = 0.650 to 0.789), and 0.746 (95% CI = 0.699 to 0.793) for the 3 external validation cohorts, which presented a statistically significant improvement compared with the conventional TNM staging system. In the high-risk group, patients who received induction chemotherapy plus CCRT had better outcomes than patients who received CCRT alone, whereas there was no statistically significant difference in the low-risk group. Conclusions: The proposed framework can capture more complex and heterogeneous information to predict the prognosis of patients with LA-NPC and potentially contribute to clinical decision making.

Objectives: As a life-threatening condition, sepsis is one of the major public health issues worldwide. Early prediction can improve sepsis outcomes with appropriate interventions. With the PhysioNet/Computing in Cardiology Challenge 2019, we aimed to develop and validate a machine learning algorithm with high prediction performance and clinical interpretability for prediction of sepsis onset during critical care in real-time. Design: Retrospective observational cohort study. Setting: The dataset was collected from three ICUs in three different U.S. hospitals. Two of them were publicly available for model development (offline) and one was used for testing (online). Patients: Forty-thousand three-hundred thirty-six ICU patients from the two model development databases and 24,819 from the test database. There are up to 40 hourly-recorded clinical variables for each ICU stay. The Sepsis-3 criteria were used to confirm sepsis onset. Interventions: None. Measurements and Main Results: Three-hundred twelve features were constructed hourly as the input of our proposed Time-phAsed machine learning model for Sepsis Prediction. Time-phAsed machine learning model for Sepsis Prediction first estimates the likelihood of sepsis onset for each hour of an ICU stay in the following 6 hours, and then makes a binary prediction with three time-phased cutoff values. On the internal validation set, the utility score (official challenge measurement) achieved by Time-phAsed machine learning model for Sepsis Prediction was 0.430. On the test set, the utility score reached was 0.354. Furthermore, Time-phAsed machine learning model for Sepsis Prediction provides an intuitive way to illustrate the impact of the input features on the outcome prediction, which makes it clinically interpretable. Conclusions: The proposed Time-phAsed machine learning model for Sepsis Prediction model is accurate and interpretable for real-time prediction of sepsis onset in critical care, which holds great potential for further evaluation in prospective studies.

Background: Fraud, Waste, and Abuse (FWA) detection is a significant yet challenging problem in the health insurance industry. An essential step in FWA detection is to check whether the medication is clinically reasonable with respect to the diagnosis. Currently, human experts with sufficient medical knowledge are required to perform this task. To reduce the cost, insurance inspectors tend to build an intelligent system to detect suspicious claims with inappropriate diagnoses/medications automatically. Objective: The aim of this study was to develop an automated method for making use of a medical knowledge graph to identify clinically suspected claims for FWA detection. Methods: First, we identified the medical knowledge that is required to assess the clinical rationality of the claims. We then searched for data sources that contain information to build such knowledge. In this study, we focused on Chinese medical knowledge. Second, we constructed a medical knowledge graph using unstructured knowledge. We used a deep learning–based method to extract the entities and relationships from the knowledge sources and developed a multilevel similarity matching approach to conduct the entity linking. To guarantee the quality of the medical knowledge graph, we involved human experts to review the entity and relationships with lower confidence. These reviewed results could be used to further improve the machine-learning models. Finally, we developed the rules to identify the suspected claims by reasoning according to the medical knowledge graph. Results: We collected 185,796 drug labels from the China Food and Drug Administration, 3390 types of disease information from medical textbooks (eg, symptoms, diagnosis, treatment, and prognosis), and information from 5272 examinations as the knowledge sources. The final medical knowledge graph includes 1,616,549 nodes and 5,963,444 edges. We designed three knowledge graph reasoning rules to identify three kinds of inappropriate diagnosis/medications. The experimental results showed that the medical knowledge graph helps to detect 70% of the suspected claims. Conclusions: The medical knowledge graph–based method successfully identified suspected cases of FWA (such as fraud diagnosis, excess prescription, and irrational prescription) from the claim documents, which helped to improve the efficiency of claim processing.