Control of blood glucose is essential for diabetes management. Current digital therapeutic approaches for subjects with Type 1 diabetes mellitus (T1DM) such as the artificial pancreas and insulin bolus calculators leverage machine learning techniques for predicting subcutaneous glucose for improved control. Deep learning has recently been applied in healthcare and medical research to achieve state-of-the-art results in a range of tasks including disease diagnosis, and patient state prediction among others. In this work, we present a deep learning model that is capable of forecasting glucose levels with leading accuracy for simulated patient cases (RMSE = 9.38±0.71 [mg/dL] over a 30-minute horizon, RMSE = 18.87±2.25 [mg/dL] over a 60-minute horizon) and real patient cases (RMSE = 21.07±2.35 [mg/dL] for 30-minute, RMSE = 33.27±4.79\% for 60-minute). In addition, the model provides competitive performance in providing effective prediction horizon ( $PH_{eff}$ ) with minimal time lag both in a simulated patient dataset ( $PH_{eff}$ = 29.0±0.7 for 30-min and $PH_{eff}$ = 49.8±2.9 for 60-min) and in a real patient dataset ( $PH_{eff}$ = 19.3±3.1 for 30-min and $PH_{eff}$ = 29.3±9.4 for 60-min). This approach is evaluated on a dataset of 10 simulated cases generated from the UVa/Padova simulator and a clinical dataset of 10 real cases each containing glucose readings, insulin bolus, and meal (carbohydrate) data. Performance of the recurrent convolutional neural network is benchmarked against four algorithms. The proposed algorithm is implemented on an Android mobile phone, with an execution time of 6ms on a phone compared to an execution time of 780ms on a laptop.

Blood glucose (BG) prediction is essential to the success of glycemic control in type 1 diabetes (T1D) management. Empowered by the recent development of the Internet of Medical Things (IoMT), continuous glucose monitoring (CGM) and deep learning technologies have been demonstrated to achieve the state of the art in BG prediction. However, it is challenging to implement such algorithms in actual clinical settings to provide persistent decision support due to the high demand for computational resources, while smartphone-based implementations are limited by short battery life and require users to carry the device. In this work, we propose a new deep learning model using an attention-based evidential recurrent neural network and design an IoMT-enabled wearable device to implement the embedded model, which comprises a low-cost and low-power system on a chip to perform Bluetooth connectivity and edge computing for real-time BG prediction and predictive hypoglycemia detection. In addition, we developed a smartphone app to visualize BG trajectories and predictions, and desktop and cloud platforms to backup data and fine-tune models. The embedded model was evaluated on three clinical data sets including 47 T1D subjects. The proposed model achieved superior performance of root mean square error (RMSE), mean absolute error, and glucose-specific RMSE, and obtained the best accuracy for hypoglycemia detection when compared with a group of machine learning baseline methods. Moreover, we performed hardware-in-the-loop in silico trials with ten virtual T1D adults to test the whole IoMT system with predictive low-glucose management, which significantly reduced hypoglycemia and improved BG control.

Blood glucose prediction algorithms are key tools in the development of decision support systems and closed-loop insulin delivery systems for blood glucose control in diabetes. Deep learning models have provided leading results among machine learning algorithms to date in glucose prediction. However these models typically require large amounts of data to obtain best personalised glucose prediction results. Multitask learning facilitates an approach for leveraging data from multiple subjects while still learning accurate personalised models. In this work we present results comparing the effectiveness of multitask learning over sequential transfer learning, and learning only on subject-specific data with neural network and support vector regression. The multitask learning approach shows consistent leading performance in predictive metrics at both short-term and long-term prediction horizons. We obtain a predictive accuracy (RMSE) of 18.8 $\pm$ 2.3, 25.3 $\pm$ 2.9, 31.8 $\pm$ 3.9, 41.2 $\pm$ 4.5, 47.2 $\pm$ 4.6 mg/dL at 30, 45, 60, 90, and 120 min prediction horizons respectively, with at least 93% clinically acceptable predictions using the Clarke Error Grid (EGA) at each prediction horizon. We also identify relevant prior information such as glycaemic variability that can be incorporated to improve predictive performance at long-term prediction horizons. Furthermore, we show consistent performance - $\leq$ 5% change in both RMSE and EGA (Zone A) - in rare cases of adverse glycaemic events with 1-6 weeks of training data. In conclusion, a multitask approach can allow for deploying personalised models even with significantly less subject-specific data without compromising performance.

Current artificial pancreas (AP) systems are hybrid closed-loop systems that require manual meal announcements to manage postprandial glucose control effectively. This poses a cognitive burden and challenge to users with T1D since this relies on frequent user engagement to maintain tight glucose control. In order to move towards fully automated closed-loop glucose control, we propose an algorithm based on a deep learning framework that performs multitask quantile regression, for both meal detection and carbohydrate estimation. Our proposed method is evaluated in silico on 10 adult subjects from the UVa/Padova simulator with a Bio-inspired Artificial Pancreas (BiAP) control algorithm over a 2 month period. Three different configurations of the AP are evaluated -BiAP without meal announcement (BiAP-NMA), BiAP with meal announcement (BiAP-MA), and BiAP with meal detection (BiAP-MD). We present results showing an improvement of BiAP-MD over BiAP-NMA, demonstrating 144.5 ± 6.8 mg/dL mean blood glucose level (−4.4 mg/dL, $p<$ 0.01) and 77.8 ± 6.3% mean time between 70 and 180 mg/dL (+3.9%, $p<$ 0.001). This improvement in control is realised without a significant increase in mean in hypoglycaemia (+0.1%, $p=$ 0.4). In terms of detection of meals and snacks, the proposed method on average achieves 93% precision and 76% recall with a detection delay time of 38 ± 15 min (92% precision, 92% recall, and 37 min detection time for meals only). Furthermore, BiAP-MD handles hypoglycaemia better than BiAP-MA based on CVGA assessment with fewer control errors (10% vs. 20%). This study suggests that multitask quantile regression can improve the capability of AP systems for postprandial glucose control without increasing hypoglycaemia.

Blood-related infections are a significant concern in healthcare. They can lead to serious medical complications and even death if not promptly diagnosed and treated. Throughout time, medical research has sought to identify clinical factors and strategies to improve the management of these conditions. The increasing adoption of electronic health records has led to a wealth of electronically available medical information and predictive models have emerged as invaluable tools. This manuscript offers a detailed survey of machine-learning techniques used for the diagnosis and prognosis of bacteraemia, bloodstream infections, and sepsis shedding light on their efficacy, potential limitations, and the intricacies of their integration into clinical practice. This study presents a comprehensive analysis derived from a thorough search across prominent databases, namely EMBASE, Google Scholar, PubMed, Scopus, and Web of Science, spanning from their inception dates to October 25, 2023. Eligibility assessment was conducted independently by investigators, with inclusion criteria encompassing peer-reviewed articles and pertinent non-peer-reviewed literature. Clinical and technical data were meticulously extracted and integrated into a registry, facilitating a holistic examination of the subject matter. To maintain currency and comprehensiveness, readers are encouraged to contribute manuscript suggestions and/or reports for integration into this living registry. While machine learning (ML) models exhibit promise in advanced disease stages such as sepsis, early stages remain underexplored due to data limitations. Biochemical markers emerge as pivotal predictors during early stages such as bacteraemia, or bloodstream infections, while vital signs assume significance in sepsis prognosis. Integrating temporal trend information into conventional machine learning models appears to enhance performance. Unfortunately, sequential deep learning models face challenges, showing minimal performance improvements and significant drops in external datasets, potentially due to learning missing patterns within the scarce data available rather than understanding disease dynamics. Real-life implementation receives limited attention, as meeting design requirements proves challenging within existing healthcare infrastructure. The data collected in an event-based fashion during clinical practice is insufficient to fully harness the potential of these data-hungry models. Despite limitations, opportunities abound in leveraging flexible models and exploiting real-time non-invasive data collection technologies such as wearable devices or microneedles. Addressing research gaps in early disease stages, harnessing patient history data often underused, and embracing continual diagnostics beyond treatment initiation are crucial for improving healthcare decision-making support and adoption across the entire management pathway. This comprehensive survey illuminates the landscape of ML applications in blood-related infection management, offering insights for future research and clinical practice. Implementing clinical ML-based clinical decision support systems requires balancing research with practical considerations. Current methodologies often lead to complex models lacking transparency and practical validation. Integration into healthcare systems faces regulatory, privacy, and trust challenges. Clear presentations and adherence to standards are essential to boost confidence in machine learning models for real-world healthcare applications.

Close vital signs monitoring is crucial for the clinical management of patients with dengue. We investigated performance of a non-invasive wearable utilising photoplethysmography (PPG), to provide real-time risk prediction in hospitalised individuals. We performed a prospective observational clinical study in Vietnam between January 2020 and October 2022: 153 patients were included in analyses, providing 1353 h of PPG data. Using a multi-modal transformer approach, 10-min PPG waveform segments and basic clinical data (age, sex, clinical features on admission) were used as features to continuously forecast clinical state 2 h ahead. Prediction of low-risk states (17,939/80,843; 22.1%), defined by NEWS2 and mSOFA < 6, was associated with an area under the precision-recall curve of 0.67 and an area under the receiver operator curve of 0.83. Implementation of such interventions could provide cost-effective triage and clinical care in dengue, offering opportunities for safe ambulatory patient management.