NPJ Digital Medicine最新文献

Remote, digital cognitive testing on an individual’s own device provides the opportunity to deploy previously understudied but promising cognitive paradigms in preclinical Alzheimer’s disease (AD). The Boston Remote Assessment for NeuroCognitive Health (BRANCH) captures a personalized learning curve for the same information presented over seven consecutive days. Here, we examined BRANCH multi-day learning curves (MDLCs) in 167 cognitively unimpaired older adults (age = 74.3 ± 7.5, 63% female) with different amyloid-β (A) and tau (T) biomarker profiles on positron emission tomography. MDLC scores decreased across ascending biomarker groups, with the A + T- group performing numerically worse (β = –0.24, 95%CI[–0.55,0.07], p = 0.128) and the A + T+ group performing significantly worse (β = –0.58, 95%CI[–1.06,–0.10], p = 0.018) than the A-T- group. Further, lower MDLC scores were associated with greater cortical thinning (β = 0.18, 95%CI[0.04,0.34], p = 0.013). Our results suggest that diminished MDLCs track with advanced AD pathophysiology, and demonstrate how a digital multi-day learning paradigm can provide novel insights about cognitive decline during preclinical AD.

The 12-lead electrocardiogram (ECG) is inexpensive and widely available. Whether conditions across the human disease landscape can be detected using the ECG is unclear. We developed a deep learning denoising autoencoder and systematically evaluated associations between ECG encodings and ~1,600 Phecode-based diseases in three datasets separate from model development, and meta-analyzed the results. The latent space ECG model identified associations with 645 prevalent and 606 incident Phecodes. Associations were most enriched in the circulatory (n = 140, 82% of category-specific Phecodes), respiratory (n = 53, 62%) and endocrine/metabolic (n = 73, 45%) categories, with additional associations across the phenome. The strongest ECG association was with hypertension (p < 2.2×10^-308). The ECG latent space model demonstrated more associations than models using standard ECG intervals, and offered favorable discrimination of prevalent disease compared to models comprising age, sex, and race. We further demonstrate how latent space models can be used to generate disease-specific ECG waveforms and facilitate individual disease profiling.

Cardiac wall motion abnormalities (WMA) are strong predictors of mortality, but current screening methods using Q waves from electrocardiograms (ECGs) have limited accuracy and vary across racial and ethnic groups. This study aimed to identify novel ECG features using deep learning to enhance WMA detection, referencing echocardiography as the gold standard. We collected ECG and echocardiogram data from 35,210 patients in California and labeled WMA using unstructured language parsing of echocardiographic reports. A deep neural network (ECG-WMA-Net) was trained and outperformed both expert ECG interpretation and Q-wave indices, achieving an AUROC of 0.781 (CI: 0.762–0.799). The model was externally validated in a diverse cohort from Georgia (n = 2338), with an AUC of 0.723 (CI: 0.685–0.757). Explainability analysis revealed significant contributions from QRS and T-wave regions. This deep learning approach improves WMA screening accuracy, potentially addressing physiological differences not captured by standard ECG-based methods.

The current approach to fetal anomaly screening is based on biometric measurements derived from individually selected ultrasound images. In this paper, we introduce a paradigm shift that attains human-level performance in biometric measurement by aggregating automatically extracted biometrics from every frame across an entire scan, with no need for operator intervention. We use a neural network to classify each frame of an ultrasound video recording. We then measure fetal biometrics in every frame where appropriate anatomy is visible. We use a Bayesian method to estimate the true value of each biometric from a large number of measurements and probabilistically reject outliers. We performed a retrospective experiment on 1457 recordings (comprising 48 million frames) of 20-week ultrasound scans, estimated fetal biometrics in those scans and compared our estimates to real-time manual measurements. Our method achieves human-level performance in estimating fetal biometrics and estimates well-calibrated credible intervals for the true biometric value.

This retrospective cohort study evaluates the impact of an AI-supported continuous glucose monitoring (CGM) mobile app (“January V2”) on glycemic control and weight management in 944 users, including healthy individuals and those with prediabetes or type 2 diabetes (T2D). The app, leveraging AI to personalize feedback, tracked users’ food intake, activity, and glucose responses over 14 days. Significant improvements in time in range (TIR) were observed, particularly in users with lower baseline TIR. Healthy users’ TIR increased from 74.7% to 85.5% (p < 0.0001), while T2D users’ TIR improved from 49.7% to 57.4% (p < 0.0004). Higher app engagement correlated with greater TIR improvements. Users also experienced an average weight reduction of 3.3 lbs over 33 days. These findings suggest that AI-enhanced digital health interventions can improve glycemic control and promote weight loss, particularly when users are actively engaged.

Using administrative claims and electronic health records for observational studies is common but challenging due to data limitations. Researchers rely on phenotype algorithms, requiring labor-intensive chart reviews for validation. This study investigates whether case adjudication using the previously introduced Knowledge-Enhanced Electronic Profile Review (KEEPER) system with large language models (LLMs) is feasible and could serve as a viable alternative to manual chart review. The task involves adjudicating cases identified by a phenotype algorithm, with KEEPER extracting predefined findings such as symptoms, comorbidities, and treatments from structured data. LLMs then evaluate KEEPER outputs to determine whether a patient truly qualifies as a case. We tested four LLMs including GPT-4, hosted locally to ensure privacy. Using zero-shot prompting and iterative prompt optimization, we found LLM performance, across ten diseases, varied by prompt and model, with sensitivities from 78 to 98% and specificities from 48 to 98%, indicating promise for automating phenotype evaluation.

Neonatal seizures require urgent treatment, but often go undetected without expert EEG monitoring. We have developed and validated a seizure detection model using retrospective EEG data from 332 neonates. A convolutional neural network was trained and tested on over 50,000 hours (n = 202) of annotated single-channel EEG containing 12,402 seizure events. This model was then validated on two independent multi-reviewer datasets (n = 51 and n = 79). Increasing data and model size improved performance: Matthews correlation coefficient (MCC) and Pearson’s correlation (r) increased by up to 50% (15%) with data (model) scaling. The largest model (21m parameters) achieved state-of-the-art on an open-access dataset (MCC = 0.764, r = 0.824, and AUC = 0.982). This model also attained expert-level performance on both validation sets, a first in this field, with no significant difference in inter-rater agreement when the model replaces an expert (∣Δκ∣ < 0.094, p > 0.05).

This systematic review examines four themes of device-based remote monitoring (DRM): technology, patient monitoring and support, integration of DRM into clinical care, and patient engagement, and their impact on hospital service use. We included randomized controlled trials (RCTs) until 2024 comparing hospital service use in DRM with usual care. Hospital service use decreased in DRM in 72% of the 116 included RCTs. Non-implantable devices were most commonly used to measure data, but showed a lower decrease in hospital service use than implanted or mobile devices (69% vs 89% and 76%). Providing 24/7 support for patients led to a decrease (81% of the studies). DRM replacing usual care, involving designated healthcare providers, and patient-performed data transmission led to a greater decrease in hospital service use. DRM has the potential to further reduce hospital service use with additional factors such as sufficient support, automated processes, and optimized care redesign.

Accurate prognostication guides optimal clinical management in skin cancer. Merkel cell carcinoma (MCC) is the most aggressive form of skin cancer that often presents in advanced stages and is associated with poor survival rates. There are no personalized prognostic tools in use in MCC. We employed explainability analysis to reveal new insights into mortality risk factors for this highly aggressive cancer. We then combined deep learning feature selection with a modified XGBoost framework, to develop a web-based prognostic tool for MCC termed ‘DeepMerkel’. DeepMerkel can make accurate personalised, time-dependent survival predictions for MCC from readily available clinical information. It demonstrated generalizability through high predictive performance in an international clinical cohort, out-performing current population-based prognostic staging systems. MCC and DeepMerkel provide the exemplar model of personalised machine learning prognostic tools in aggressive skin cancers.