A deep learning model for novel systemic biomarkers in photographs of the external eye: a retrospective study.

BACKGROUND: Photographs of the external eye were recently shown to reveal signs of diabetic retinal disease and elevated glycated haemoglobin. This study aimed to test the hypothesis that external eye photographs contain information about additional systemic medical conditions. METHODS: We developed a deep learning system (DLS) that takes external eye photographs as input and predicts systemic parameters, such as those related to the liver (albumin, aspartate aminotransferase [AST]); kidney (estimated glomerular filtration rate [eGFR], urine albumin-to-creatinine ratio [ACR]); bone or mineral (calcium); thyroid (thyroid stimulating hormone); and blood (haemoglobin, white blood cells [WBC], platelets). This DLS was trained using 123 130 images from 38 398 patients with diabetes undergoing diabetic eye screening in 11 sites across Los Angeles county, CA, USA. Evaluation focused on nine prespecified systemic parameters and leveraged three validation sets (A, B, C) spanning 25 510 patients with and without diabetes undergoing eye screening in three independent sites in Los Angeles county, CA, and the greater Atlanta area, GA, USA. We compared performance against baseline models incorporating available clinicodemographic variables (eg, age, sex, race and ethnicity, years with diabetes). FINDINGS: Relative to the baseline, the DLS achieved statistically significant superior performance at detecting AST >36·0 U/L, calcium <8·6 mg/dL, eGFR <60·0 mL/min/1·73 m2, haemoglobin <11·0 g/dL, platelets <150·0 × 103/µL, ACR ≥300 mg/g, and WBC <4·0 × 103/µL on validation set A (a population resembling the development datasets), with the area under the receiver operating characteristic curve (AUC) of the DLS exceeding that of the baseline by 5·3-19·9% (absolute differences in AUC). On validation sets B and C, with substantial patient population differences compared with the development datasets, the DLS outperformed the baseline for ACR ≥300·0 mg/g and haemoglobin <11·0 g/dL by 7·3-13·2%. INTERPRETATION: We found further evidence that external eye photographs contain biomarkers spanning multiple organ systems. Such biomarkers could enable accessible and non-invasive screening of disease. Further work is needed to understand the translational implications. FUNDING: Google.

Development of a Machine Learning Model for Sonographic Assessment of Gestational Age.

Importance: Fetal ultrasonography is essential for confirmation of gestational age (GA), and accurate GA assessment is important for providing appropriate care throughout pregnancy and for identifying complications, including fetal growth disorders. Derivation of GA from manual fetal biometry measurements (ie, head, abdomen, and femur) is operator dependent and time-consuming. Objective: To develop artificial intelligence (AI) models to estimate GA with higher accuracy and reliability, leveraging standard biometry images and fly-to ultrasonography videos. Design, Setting, and Participants: To improve GA estimates, this diagnostic study used AI to interpret standard plane ultrasonography images and fly-to ultrasonography videos, which are 5- to 10-second videos that can be automatically recorded as part of the standard of care before the still image is captured. Three AI models were developed and validated: (1) an image model using standard plane images, (2) a video model using fly-to videos, and (3) an ensemble model (combining both image and video models). The models were trained and evaluated on data from the Fetal Age Machine Learning Initiative (FAMLI) cohort, which included participants from 2 study sites at Chapel Hill, North Carolina (US), and Lusaka, Zambia. Participants were eligible to be part of this study if they received routine antenatal care at 1 of these sites, were aged 18 years or older, had a viable intrauterine singleton pregnancy, and could provide written consent. They were not eligible if they had known uterine or fetal abnormality, or had any other conditions that would make participation unsafe or complicate interpretation. Data analysis was performed from January to July 2022. Main Outcomes and Measures: The primary analysis outcome for GA was the mean difference in absolute error between the GA model estimate and the clinical standard estimate, with the ground truth GA extrapolated from the initial GA estimated at an initial examination. Results: Of the total cohort of 3842 participants, data were calculated for a test set of 404 participants with a mean (SD) age of 28.8 (5.6) years at enrollment. All models were statistically superior to standard fetal biometry-based GA estimates derived from images captured by expert sonographers. The ensemble model had the lowest mean absolute error compared with the clinical standard fetal biometry (mean [SD] difference, -1.51 [3.96] days; 95% CI, -1.90 to -1.10 days). All 3 models outperformed standard biometry by a more substantial margin on fetuses that were predicted to be small for their GA. Conclusions and Relevance: These findings suggest that AI models have the potential to empower trained operators to estimate GA with higher accuracy.

A mobile-optimized artificial intelligence system for gestational age and fetal malpresentation assessment.

Background: Fetal ultrasound is an important component of antenatal care, but shortage of adequately trained healthcare workers has limited its adoption in low-to-middle-income countries. This study investigated the use of artificial intelligence for fetal ultrasound in under-resourced settings. Methods: Blind sweep ultrasounds, consisting of six freehand ultrasound sweeps, were collected by sonographers in the USA and Zambia, and novice operators in Zambia. We developed artificial intelligence (AI) models that used blind sweeps to predict gestational age (GA) and fetal malpresentation. AI GA estimates and standard fetal biometry estimates were compared to a previously established ground truth, and evaluated for difference in absolute error. Fetal malpresentation (non-cephalic vs cephalic) was compared to sonographer assessment. On-device AI model run-times were benchmarked on Android mobile phones. Results: Here we show that GA estimation accuracy of the AI model is non-inferior to standard fetal biometry estimates (error difference -1.4 ± 4.5 days, 95% CI -1.8, -0.9, n = 406). Non-inferiority is maintained when blind sweeps are acquired by novice operators performing only two of six sweep motion types. Fetal malpresentation AUC-ROC is 0.977 (95% CI, 0.949, 1.00, n = 613), sonographers and novices have similar AUC-ROC. Software run-times on mobile phones for both diagnostic models are less than 3 s after completion of a sweep. Conclusions: The gestational age model is non-inferior to the clinical standard and the fetal malpresentation model has high AUC-ROCs across operators and devices. Our AI models are able to run on-device, without internet connectivity, and provide feedback scores to assist in upleveling the capabilities of lightly trained ultrasound operators in low resource settings.

Digital Applications Targeting Medication Safety in Ambulatory High-Risk CKD Patients: Randomized Controlled Clinical Trial.

BACKGROUND AND OBJECTIVES: Patients with CKD are at risk for adverse drug reactions, but effective community-based preventive programs remain elusive. In this study, we compared the effectiveness of two digital applications designed to improve outpatient medication safety. DESIGN, SETTING, PARTICIPANTS, & MEASUREMENTS: In a 1-year randomized controlled trial, 182 outpatients with advanced CKD were randomly assigned to receive a smartphone preloaded with either eKidneyCare (n=89) or MyMedRec (n=93). The experimental intervention, eKidneyCare, includes a medication feature that prompted patients to review medications monthly and report changes, additions, or medication problems to clinicians for reconciliation and early intervention. The active comparator was MyMedRec, a commercially available, standalone application for storing medication and other health information that can be shared with patients' providers. The primary outcome was the rate of medication discrepancy, defined as differences between the patient's reported history and the clinic's medication record, at exit. RESULTS: At exit, the eKidneyCare group had fewer total medication discrepancies compared with MyMedRec (median, 0.45; interquartile range, 0.33-0.63 versus 0.67; interquartile range, 0.40-1.00; P=0.001), and the change from baseline was 0.13±0.27 in eKidneyCare and 0.30±0.41 in MyMedRec (P=0.007). eKidneyCare use also reduced the severity of clinically relevant medication discrepancies in all categories, including those with the potential to cause serious harm (estimated rate ratio, 0.40; 95% confidence interval, 0.27 to 0.63). Usage data revealed that 72% of patients randomized to eKidneyCare completed one or more medication reviews per month, whereas only 30% of patients in the MyMedRec group (adjusted for dropouts) kept their medication profile on their phone. CONCLUSIONS: In patients who are high risk and have CKD, eKidneyCare significantly reduced the rate and severity of medication discrepancies, the proximal cause of medication errors, compared with the active comparator. CLINICAL TRIAL REGISTRY NAME AND REGISTRATION NUMBER: www.ClinicalTrials.gov, NCT:02905474.

Development and usability testing of HEARTPA♀N: protocol for a mixed methods strategy to develop an integrated smartphone and web-based intervention for women with cardiac pain.

INTRODUCTION: More women experience cardiac pain related to coronary artery disease and cardiac procedures compared with men. The overall goal of this programme of research is to develop an integrated smartphone and web-based intervention (HEARTPA♀N) to help women recognise and self-manage cardiac pain. METHODS AND ANALYSIS: This protocol outlines the mixed methods strategy used for the development of the HEARTPA♀N content/core feature set (phase 2A), usability testing (phase 2B) and evaluation with a pilot randomised controlled trial (RCT) (phase 3). We are using the individual and family self-management theory, mobile device functionality and pervasive information architecture of mHealth interventions, and following a sequential phased approach recommended by the Medical Research Council to develop HEARTPA♀N. The phase 3 pilot RCT will enable us to refine the prototype, inform the methodology and calculate the sample size for a larger multisite RCT (phase 4, future work). Patient partners have been actively involved in setting the HEARTPA♀N research agenda, including defining patient-reported outcome measures for the pilot RCT: pain and health-related quality of life (HRQoL). As such, the guidelines for Inclusion of Patient-Reported Outcomes in Clinical Trial Protocols (SPIRIT-PRO) are used to report the protocol for the pilot RCT (phase 3). Quantitative data (eg, demographic and clinical information) will be summarised using descriptive statistics (phases 2AB and 3) and a content analysis will be used to identify themes (phase 2AB). A process evaluation will be used to assess the feasibility of the implementation of the intervention and a preliminary efficacy evaluation will be undertaken focusing on the outcomes of pain and HRQoL (phase 3). ETHICS AND DISSEMINATION: Ethics approval was obtained from the University of Toronto (36415; 26 November 2018). We will disseminate knowledge of HEARTPA♀N through publication, conference presentation and national public forums (Café Scientifique), and through fact sheets, tweets and webinars. TRIAL REGISTRATION NUMBER: NCT03800082.

Development of a Wearable Cardiac Monitoring System for Behavioral Neurocardiac Training: A Usability Study.

BACKGROUND: Elevated blood pressure is one of the main risk factors for death globally. Behavioral neurocardiac training (BNT) is a complementary approach to blood pressure and stress management that is intended to exercise the autonomic reflexes, improve stress recovery, and lower blood pressure. BNT involves cognitive-behavioral therapy with a paced breathing technique and heart rate variability biofeedback. BNT is limited to in-clinic delivery and faces an accessibility barrier because of the need for clinical oversight and the use of complex monitoring tools. OBJECTIVE: The objective of this project was to design, develop, and evaluate a wearable electrocardiographic (ECG) sensor system for the delivery of BNT in a home setting. METHODS: The wearable sensor system, Beat, consists of an ECG sensor and a mobile app. It was developed iteratively using the principles of test-driven Agile development and user-centered design. A usability study was conducted at Toronto General Hospital to evaluate feasibility and user experience and identify areas of improvement. RESULTS: The Beat sensor was designed as a modular patch to be worn on the user's chest and uses standard ECG electrodes. It streams a single-lead ECG wirelessly to a mobile phone using Bluetooth Low Energy. The use of small, low-power electronics, a low device profile, and a tapered enclosure allowed for a device that can be unobtrusively worn under clothing. The sensor was designed to operate with a mobile app that guides users through the BNT exercises to train them to a slow-paced breathing technique for stress recovery. The BNT app uses the ECG captured by the sensor to provide heart rate variability biofeedback in the form of a real-time heart rate waveform to complement and reinforce the impact of the training. Usability testing (n=6) indicated that the overall response to the design and user experience of the system was perceived positively. All participants indicated that the system had a positive effect on stress management and that they would use it at home. Areas of improvement were identified, which focused primarily on the delivery of training and education on BNT through the app. CONCLUSIONS: The outcome of this project was a wearable sensor system to deliver BNT at home. The system has the potential to offer a complementary approach to blood pressure and stress management at home and reduce current accessibility barriers.