OBJECTIVES: Federated learning (FL) has gained popularity in clinical research in recent years to facilitate privacy-preserving collaboration. Structured data, one of the most prevalent forms of clinical data, has experienced significant growth in volume concurrently, notably with the widespread adoption of electronic health records in clinical practice. This review examines FL applications on structured medical data, identifies contemporary limitations, and discusses potential innovations. MATERIALS AND METHODS: We searched 5 databases, SCOPUS, MEDLINE, Web of Science, Embase, and CINAHL, to identify articles that applied FL to structured medical data and reported results following the PRISMA guidelines. Each selected publication was evaluated from 3 primary perspectives, including data quality, modeling strategies, and FL frameworks. RESULTS: Out of the 1193 papers screened, 34 met the inclusion criteria, with each article consisting of one or more studies that used FL to handle structured clinical/medical data. Of these, 24 utilized data acquired from electronic health records, with clinical predictions and association studies being the most common clinical research tasks that FL was applied to. Only one article exclusively explored the vertical FL setting, while the remaining 33 explored the horizontal FL setting, with only 14 discussing comparisons between single-site (local) and FL (global) analysis. CONCLUSIONS: The existing FL applications on structured medical data lack sufficient evaluations of clinically meaningful benefits, particularly when compared to single-site analyses. Therefore, it is crucial for future FL applications to prioritize clinical motivations and develop designs and methodologies that can effectively support and aid clinical practice and research.

Clinical applications that involves risk stratification of patients often predict the likely occurrence of an event such as the development of a complication as well as the time to the next event. Existing approaches that provide individual survival distribution across time often use baseline measurements for the risk prediction of an event occurring, and is unable to estimate the time to next event. Further, they do not deal with the complex and heterogeneous information from past visit records, and hence do not handle repeated events which are common in chronic diseases. We address the above limitations by designing a recurrent temporal time process network that incorporates longitudinal visit information to increase the accuracy of risk predictions for clinical events and to estimate the time to next event. Our proposed solution utilizes a recurrent neural network to learn a latent representation of patient visit history. This representation enables us to approximate the conditional intensity function of the temporal point process, from which we derive the survival probability function to obtain the risk of an event occurring and to estimate the time to next event. We demonstrate our approach on two real-world healthcare datasets and show that the proposed approach is able to achieve a significant performance improvement over state-of-the-art methods.

The American College of Radiology (ACR) has guidelines on appropriate ordering of Magnetic Resonance Imaging (MRI) brain scans. MRI requests are currently manually reviewed by radiologists to ensure compliance to these guidelines. In this paper, we implemented a stacked recurrent neural network (RNN) utilizing a bidirectional long short-term memory (Bi-LSTM) sequence with BioWordVec, a biomedical word embedding vector that uses word representations from a lexicon developed from medical publications, to develop an automated classification system for request audit. To overcome the problems of interpretation by black-box models, the RNN is integrated with a model agnostic explainer LIME (Local Interpretable Model-Agnostic Explanations) to provide explainable support for clinicians in the healthcare environment. The performance of the RNN is compared with a Random Forest (RF) algorithm that utilizes the bag-of-words concept. The RNN was trained and validated on 2470 rows of different patient free-text orders and tested on a separate 2711 orders, producing an accuracy of 82.51% and a ROC value of 0.89 which was either comparable to or surpassing RF both in performance and usability. The use of deep learning with explainable LIME in this study provided a good use case towards an augmented decision-making framework in healthcare.

Understanding a power wheelchair users mobility characteristics is critical because mobility is an important factor for social participation and quality of life of an individual. Although power wheelchairs can improve the mobility for people with disabilities, research has shown that power wheelchair users tend to live an inactive lifestyle. A sedentary lifestyle exposes wheelchair users to a greater risk of secondary health issues, such as cardiovascular diseases, obesity, diabetes, etc. Therefore, it is critical to assess wheelchair users mobility to ensure that they maintain an active lifestyle. However, existing health tracking applications are not suitable for power wheelchair users. They either require sensors to be installed on the wheels of a wheelchair (hence bringing installation and maintenance burdens) or are designed for able individuals by detecting the users steps, whose characteristics are significantly different from the dynamics of a power wheelchair. Furthermore, data captured by the inertial sensors (e.g., accelerometer or gyroscope) demonstrates a wide variety of patterns owing to different terrains on which the wheelchair travels. In this study, we propose to use the accelerometer in a smartphone for data collection, and employ mathematics and physics techniques to process and transform the raw data so that patterns intrinsic to wheelchair maneuvers are revealed. Based on the processed data, we developed a learning-based approach to analyze wheelchair users mobility by leveraging such patterns. We have conducted a sequence of experiments to evaluate the proposed approach. Experimental results showed that our approach correctly recognized all the bouts (segments of continuous movement), and achieved accurate measurements on bout maneuvering time and maximum period of continuous movement, which are critical indicators of a wheelchair users mobility.

Sixteen conventional heart beat variability (HRV) parameters and eight vital signs have shown promise in the prediction of cardiac arrest within 72 hours. Besides these 24 parameters, we proposed adding two new features for cardiac arrest prediction, which are approximate entropy (ApEn) and sample entropy (SpEn). ApEn and SpEn are nonlinear HRV parameters capable of characterizing heart conditions. These two entropies were derived from electrocardiography recordings and combined with the existing 24 features to form feature combinations. The experiments were conducted by using linear kernel Support Vector Machine classification technique to investigate the effects of using ApEn, SpEn together with 24 parameters on cardiac arrest prediction. The dimensionality reduction approach, Principal Component Analysis, was applied to suppress the dimensionality. Results reveal that the prediction performance of adding ApEn and SpEn to the 24 parameters is improved significantly compared to using the 24 parameters only. Dimension reduction has additional positive effects on improving the prediction results.

Fast and accurate risk stratification is essential in the emergency department (ED) as it allows clinicians to identify chest pain patients who are at high risk of cardiac complications and require intensive monitoring and early intervention. In this paper, we present a novel intelligent scoring system using heart rate variability, 12-lead electrocardiogram (ECG), and vital signs where a hybrid sampling-based ensemble learning strategy is proposed to handle data imbalance. The experiments were conducted on a dataset consisting of 564 chest pain patients recruited at the ED of a tertiary hospital. The proposed ensemble-based scoring system was compared with established scoring methods such as the modified early warning score and the thrombolysis in myocardial infarction score, and showed its effectiveness in predicting acute cardiac complications within 72 h in terms of the receiver operation characteristic analysis.

Traditional risk score prediction is based on vital signs and clinical assessment. In this paper, we present an intelligent scoring system for the prediction of cardiac arrest within 72 h. The patient population is represented by a set of feature vectors, from which risk scores are derived based on geometric distance calculation and support vector machine. Each feature vector is a combination of heart rate variability (HRV) parameters and vital signs. Performance evaluation is conducted on the leave-one-out cross-validation framework, and receiver operating characteristic, sensitivity, specificity, positive predictive value, and negative predictive value are reported. Experimental results reveal that the proposed scoring system not only achieves satisfactory performance on determining the risk of cardiac arrest within 72 h but also has the ability to generate continuous risk scores rather than a simple binary decision by a traditional classifier. Furthermore, the proposed scoring system works well for both balanced and imbalanced datasets, and the combination of HRV parameters and vital signs shows superiority in prediction to using HRV parameters only or vital signs only.