Earlier this year, Tom Stanis was puzzling through what was next for his startup Story Health, which helps providers care for people with heart failure. The company had some big-name customers and plans to expand, but it last raised money in 2022. Stanis saw two options: shake more cash out of a stingy venture capital market, or sell.

Armed with $275 million in fresh funding and a built-in customer base, artificial intelligence company Innovaccer made the answer easy. It gobbled up Story Health for an undisclosed mix of equity and cash in September. 

Story Health is the fourth Innovaccer acquisition in about a year as it aims to become the default AI platform for health systems. CEO Abhinav Shashank plans to rapidly expand and to “accelerate that development through M&A,” he told STAT.

Innovaccer’s shopping spree is just one example of a trend playing out in digital health: big, well-funded companies with momentum are snapping up smaller players.

Continue to STAT+ to read the full story…

© Adobe

About a year ago, I logged into MyChart the day before my annual wellness visit and saw a new consent form. My doctor wanted to record our visit so artificial intelligence could create notes and update my medical record. I agreed. Just hours after my visit ended, a comprehensive, accurate visit summary appeared in MyChart.

As a career clinical trialist, I immediately wondered: What if this tool were used for trial visits? Could it enable a breakthrough in research or patient engagement? How will clinical research adapt to what could be a “new normal” in medical documentation?

Driven by a crisis of clinician burnout, ambient digital scribes (ADS) like the one my doctor used are being rapidly adopted. Studies suggest these digital scribes can reduce documentation burden, improve clinician efficiency, and potentially enhance the patient experience. By digitizing conversations in real time, these scribes open possibilities beyond routine care, particularly in clinical research and trials.

Continue to STAT+ to read the full story…

© Adobe

npj Digital Medicine, Published online: 08 October 2025; doi:10.1038/s41746-025-01939-x

Systematic review: digital biomarkers of fatigue in chronic diseases

npj Digital Medicine, Published online: 06 October 2025; doi:10.1038/s41746-025-01978-4

Digital twin models for predicting venetoclax and azacitidine-induced neutropenia in patients with acute myeloid leukemia

Nature, Published online: 08 October 2025; doi:10.1038/s41586-025-09603-w

MetaGraph enables scalable indexing of large sets of DNA, RNA or protein sequences using annotated de Bruijn graphs.

Nature, Published online: 08 October 2025; doi:10.1038/d41586-025-03195-1

Susumu Kitagawa, Richard Robson and Omar Yaghi pioneered the creation of metal–organic frameworks, which can capture and store molecules such as carbon dioxide.

Nature, Published online: 08 October 2025; doi:10.1038/d41586-025-03219-w

MetaGraph compresses vast data archives into a search engine for scientists, opening up new frontiers of biological discovery.

Nature, Published online: 07 October 2025; doi:10.1038/d41586-025-03287-y

Regulatory T cells, which help to dampen inflammation, are being used in clinical trials against ailments such as rheumatoid arthritis.

Nature, Published online: 07 October 2025; doi:10.1038/d41586-025-03196-0

Scientists, research institutions, funders, libraries and publishers must all improve software practices.

Nature, Published online: 06 October 2025; doi:10.1038/d41586-025-03286-z

Meet the winners of this year’s Nobel Prize in Physiology or Medicine. Plus, creative hobbies could slow brain ageing at the molecular level and a visit to a sewage-filtering wetland.

Nature, Published online: 06 October 2025; doi:10.1038/d41586-025-03193-3

Mary Brunkow, Fred Ramsdell and Shimon Sakaguchi discovered cells that protect the body from autoimmune diseases.

Background: With 28%-35% of individuals aged 65 years and older experiencing incidents of falling, falls are the second leading cause of unintentional injury–related deaths globally. Limited availability of clinical staff often impedes the timely detection and prevention of potential falls. Advances in artificial intelligence (AI) could complement existing fall risk assessment and help better allocate nursing care resources. Yet, many studies are based on small datasets from a single institution, which can restrict the generalizability of the model, and do not investigate important aspects in AI model development, such as fairness across demographic groups. Objective: This study aimed to provide a comprehensive empirical evaluation of the potential of AI in nursing care, focusing on the case of fall risk prediction. To account for demographic and contextual differences in fall incidences, we analyze data from a university and a geriatric hospital in Germany. To the best of our knowledge, these are the largest fall risk prediction datasets to date with heterogeneous data distributions. We focus on 3 key objectives. First, does AI help in improving fall risk prediction? Second, how can AI models be trained safely across different hospitals? Finally, are these models fair? Methods: This study used 2 datasets for fall risk prediction: one from a university hospital with 931,726 participants, 10,442 of whom experienced falls, and another from a geriatric hospital with 12,773 participants, 1728 of whom have fallen. State-of-the-art AI models were trained with 3 approaches, including 2 decentralized learning paradigms. First, separate models were trained on data from each hospital; second, models were retrained on the respective other dataset; and federated learning (FL) was applied to both datasets. The performance of these models was compared with the rule-based systems as implemented in clinical practice for fall risk prediction. Additional analyses were conducted to test for model fairness. Results: Our findings demonstrate that AI models consistently outperform rule-based systems across all experimental setups, with the area under the receiver operating characteristic curve of 0.735 (90% CI 0.727-0.744) for the geriatric hospital, and 0.926 (90% CI 0.924-0.928) for the university hospital. FL did not improve the fall risk prediction in this setting. Our fairness analysis ruled out disparities in model performance between different sex groups, but we found fairness infringements across age groups. Conclusions: This study demonstrates that AI models consistently outperform traditional rule-based systems across heterogeneous datasets in predicting fall risk. However, it also reveals the challenges related to demographic shifts and label distribution imbalances, which limited the FL models’ ability to generalize. While the fairness analysis indicated fair results across sex subgroups, age-related disparities emerged. Addressing data imbalances and ensuring broader representation across demographic groups will be crucial for developing more fair and generalizable models.

Background: New media have become vital sources of cancer-related health information. However, concerns about the quality of that information persist. Objective: This study aims to identify characteristics of studies considering cancer-related information on new media (including social media and artificial intelligence chatbots); analyze patterns in information quality across different platforms, cancer types, and evaluation tools; and synthesize the quality levels of the information. Methods: We systematically searched PubMed, Web of Science, Scopus, and Medline databases for peer-reviewed studies published in English between 2014 and 2023. The validity of the included studies was assessed based on risk of bias, reporting quality, and ethical approval, using the Joanna Briggs Institute Critical Appraisal and the STROBE (Strengthening the Reporting of Observational Studies in Epidemiology) checklists. Features of platforms, cancer types, evaluation tools, and trends were summarized. Ordinal logistic regression was used to estimate the associations between the conclusion of quality assessments and study features. A random-effects meta-analysis of proportions was conducted to synthesize the overall levels of information quality and corresponding 95% CIs for each assessment indicator. Results: A total of 75 studies were included, encompassing 297,519 posts related to 17 cancer types across 15 media platforms. Studies focusing on video-based media (odds ratio [OR] 0.02, 95% CI 0.01-0.12), rare cancers (OR 0.32, 95% CI 0.16-0.65), and combined cancer types (OR 0.04, 95% CI 0.01-0.14) were statistically less likely to yield higher quality conclusions compared to those on text-based media and common cancers. The pooled estimates reported moderate overall quality (DISCERN 43.58, 95% CI 37.80-49.35; Global Quality Score 49.91, 95% CI 43.31-56.50), moderate technical quality (Journal of American Medical Association Benchmark Criteria 46.13, 95% CI 38.87-53.39; Health on the Net Foundation Code of Conduct 49.68, 95% CI 19.68-79.68), moderate-high understandability (Patient Education Material Assessment Tool for Understandability 66.92, 95% CI 59.86-73.99), moderate-low actionability (Patient Education Materials Assessment Tool for Actionability 37.24, 95% CI 18.08-58.68; usefulness 48.86, 95% CI 26.24-71.48), and moderate-low completeness (34.22, 95% CI 27.96-40.48). Furthermore, 27.15% (95% CI 21.36-33.35) of posts contained misinformation, 21.15% (95% CI 8.96-36.50) contained harmful information, and 12.46% (95% CI 7.52-17.39) contained commercial bias. Publication bias was detected only in misinformation studies (Egger test: bias –5.67, 95% CI –9.63 to –1.71; P=.006), with high heterogeneity across most outcomes (I²>75%). Conclusions: Meta-analysis results revealed that the overall quality of cancer-related information on social media and artificial intelligence chatbots was moderate, with relatively higher scores for understandability but lower scores for actionability and completeness. A notable proportion of content contained misleading, harmful, or commercially biased information, posing potential risks to users. To support informed decision-making in cancer care, it is essential to improve the quality of information delivered through these media platforms. Trial Registration: PROSPERO CRD420251058032; https://www.crd.york.ac.uk/PROSPERO/view/CRD420251058032

Background: Cardiovascular disease (CVD) remains the leading cause of death worldwide, yet many web-based sources on cardiovascular (CV) health are inaccessible. Large language models (LLMs) are increasingly used for health-related inquiries and offer an opportunity to produce accessible and scalable CV health information. However, because these models are trained on heterogeneous data, including unverified user-generated content, the quality and reliability of food and nutrition information on CVD prevention remain uncertain. Recent studies have examined LLM use in various health care applications, but their effectiveness for providing nutrition information remains understudied. Although retrieval-augmented generation (RAG) frameworks have been shown to enhance LLM consistency and accuracy, their use in delivering nutrition information for CVD prevention requires further evaluation. Objective: To evaluate the effectiveness of off-the-shelf and RAG-enhanced LLMs in delivering guideline-adherent nutrition information for CVD prevention, we assessed 3 off-the-shelf models (ChatGPT-4o, Perplexity, and Llama 3-70B) and a Llama 3-70B+RAG model. Methods: We curated 30 nutrition questions that comprehensively addressed CVD prevention. These were approved by a registered dietitian providing preventive cardiology services at an academic medical center and were posed 3 times to each model. We developed a 15,074-word knowledge bank incorporating the American Heart Association’s 2021 dietary guidelines and related website content to enhance Meta’s Llama 3-70B model using RAG. The model received this and a few-shot prompt as context, included citations in a Context Source section, and used vector similarity to align responses with guideline content, with the temperature parameter set to 0.5 to enhance consistency. Model responses were evaluated by 3 expert reviewers against benchmark CV guidelines for appropriateness, reliability, readability, harm, and guideline adherence. Mean scores were compared using ANOVA, with statistical significance set at P<.05. interrater agreement was measured using the cohen coefficient and readability estimated flesch-kincaid score. results: llama model scored higher than perplexity gpt-4o models on reliability appropriateness guideline adherence showed no harm.>70%; P<.001 indicated high reviewer agreement. conclusions: the llama model outperformed off-the-shelf models across all measures with no evidence of harm although responses were less readable due to technical language. scored lower on and produced some harmful responses. these findings highlight limitations demonstrate that rag system integration can enhance llm performance in delivering evidence-based dietary information.>

Public health is undergoing profound transformation driven by data from the global health sector and related fields. To address systemic health disparities, scholars and practitioners are increasingly applying a data equity lens, an approach that has become even more urgent as the United States faces the erosion of public health data infrastructure. This paper summarizes insights from an April 2024 convening by the Yale School of Public Health—The Role of Data in Public Health Equity and Innovation—with intersectoral stakeholders from academia, government (local, state, and federal), healthcare, and private industry. The convening included keynote presentations and roundtables regarding the depiction of social determinants of health (SDOH) in data; effects of artificial intelligence (AI) on health data equity; and community-based models for data, providing a framework for cross-cutting discussions. Through a narrative synthesis, themes were identified and synthesized from systematically gathered information from presentations and roundtables. This process led to a set of actionable, cross-cutting recommendations to guide inclusive and impactful data practices for policymakers, public health professionals, and health innovators across diverse contexts: (1) Enable big data and interoperability connecting SDOH and health outcomes; (2) Include diverse, non-technical voices in AI and health discussions; (3) Fund research on data equity and AI in health sciences; (4) Modernize Health Insurance Portability and Accountability Act (HIPAA) with new guidelines for AI and big data; and (5) Research and conceptual frameworks are needed to elucidate interconnections between data equity and health equity.

Nature Medicine, Published online: 06 October 2025; doi:10.1038/s41591-025-03983-2

This Review summarizes recent technical advancements in generative AI, outlines how new models might improve healthcare and discusses validation approaches—using lessons from recent successes and failures in the field.

Guest author: Or Hillel, Green Lamp

AI systems aren’t experimental anymore, they’re embedded in everyday decisions that affect millions. Yet as these models stretch into important spaces like real-time supply chain routing, medical diagnostics, and financial markets, something as simple as a stealthy data shift or an undetected anomaly can flip confident automation into costly breakdown or public embarrassment.

This isn’t just a problem for data scientists or machine learning engineers. Today, product managers, compliance officers, and business leaders are realising that AI’s value doesn’t just hinge on building a high-performing model, but on deeply understanding how, why, and when these models behave the way they do once exposed to the messiness of the real world.

Enter AI observability, a discipline that’s no longer an optional add-on, but a daily reality for teams committed to reliable, defensible, and scalable AI-driven products.

The best AI observability tools in 2025

1. Logz.io

Logz.io stands out in the AI observability landscape by providing an open, cloud-native platform tailored for the complexities of modern ML and AI systems. Its architecture fuses telemetry, logs, metrics, and traces into one actionable interface, empowering teams to visualize and analyse every stage of the AI lifecycle.

Key features include:

• AI-driven root cause analysis: Automated anomaly detection and intelligent guided troubleshooting accelerate issue resolution. The embedded AI Agent is able to surface trends, detect problems proactively, and provide explanations in natural language.
• Extensive integration: Logz.io seamlessly connects with major cloud providers, container orchestration, and popular ML frameworks. The flexibility ensures observability for hybrid and multi-cloud models without friction.
• Workflow enhancements: The platform’s interactive workflows promote faster investigation by guiding even junior engineers toward effective troubleshooting.
• Cost optimisation: Intelligent data management tools allow teams to optimise monitoring costs and prioritise valuable business insights.

2. Datadog

Datadog has evolved from a classic infrastructure monitoring tool into a powerhouse for AI observability in the enterprise. The platform harnesses an integrated stack of telemetry capture, real-time analytics, and ML-specific dashboards that provide both high-level and granular perspectives in the entire AI lifecycle.

Key features include:

• Comprehensive telemetry: Captures logs, traces, metrics, and model performance, enabling anomaly detection and quick identification of bottlenecks in both training and deployment.
• Machine learning monitoring: Specialised tools track data drift, prediction bias, and resource consumption at inference. Alerts and dashboards are tailored for model-centric use cases.
• Unified interface: Engineers, data scientists, and SREs all operate from shared dashboards, streamlining cross-team troubleshooting and collaboration.
• Rapid integration: Datadog supports dozens of AI and data science platforms, TensorFlow, PyTorch, MLflow, Kubeflow, and more, out of the box.

3. EdenAI

EdenAI addresses the needs of enterprises using multiple AI providers with a vendor-agnostic observability platform. The tool aggregates telemetry streams, monitors AI service health, and offers a unified response centre, regardless of the origin of the models, APIs, or data.

Key features include:

• Centralised dashboards: Monitor all AI models, APIs, and endpoints from a single pane of glass, ideal for organisations mixing public APIs, private models, and open-source services.
• Cross-platform drift and anomaly detection: AI-driven monitoring illuminates data drift, latency, and performance issues wherever AI is consumed or deployed.
• Automated auditing: Built-in logs and reporting features make it easy to satisfy regulatory requirements and support enterprise governance.
• Vendor-agnostic integration: Fast onboarding for new models, with connectors to major AI cloud services and on-premises deployments.

4. Dynatrace

Dynatrace has long been known for autonomous DevOps monitoring, and its AI observability features in 2025 carry that innovation into the AI realm. The platform’s core is the Davis® AI engine, which continuously analyses system health, model performance, and end-to-end dependencies throughout your ML pipelines.

Key features include:

• Autonomous anomaly detection: Davis® proactively identifies model drift, data pipeline snags, and abnormal behaviour in layers, from code to inference.
• Topology mapping: Visualizes relationships between services, models, data sources, and infrastructure, making it easy to trace the impact of changes or search for root causes.
• Predictive analytics: Helps anticipate incidents before they impact end-users by correlating macro system signals with fine-grained ML metrics.
• Scale and integration: Connects directly with leading cloud and MLOps platforms for seamless, low-touch monitoring at enterprise scale.

5. WhyLabs

WhyLabs has a data-centric approach to AI observability that centres on transparency, quantitative rigor, and proactive detection of risk in ML operations. The platform is built for organisations that want to govern and monitor the entire AI lifecycle, from raw data ingestion to live model predictions.

Key features include:

• Pipeline monitoring: Tracks data quality, schema changes, and feature drift in real-time, enabling early alerts for issues that could undermine model accuracy.
• Model performance dashboards: Visualize changes in predictive quality, bias, and rare event distribution in all deployed models.
• Rich telemetry integration: Supports monitoring for both structured and unstructured data types, reflecting the variety present in modern ML ecosystems.
• Collaborative workflows: Allows teams to annotate, triage, and resolve anomalies with a unified interface and pre-defined incident playbooks.

The real-world impact of AI observability

What does it look like in practice when an organisation gets AI observability right?

Enabling proactive incident response

In a hospital using AI for radiology triage, an unexpected equipment firmware update subtly shifts the pixel values of incoming images. Without observability, this shift goes undetected, producing subtly degraded diagnoses. With observability, the shift triggers alerts, and the team retrains the model or adjusts preprocessing, avoiding patient harm.

Preventing bias and drift

A fintech company notices a sudden, unexplained dip in loan approval rates for a specific demographic. Deep observability enables rapid investigation, diagnosis of data drift due to shifts in an upstream data partner, and quick mitigation, ensuring fairness and compliance.

Supporting human-AI collaboration

Customer support uses AI to recommend ticket responses. Observability-powered dashboards flag when auto-generated advice is leading to longer ticket resolution times for one product line. Teams use this to retrain the model, improving both customer satisfaction and business outcomes.

Choosing the right AI observability tool: Alignment, scale, and ecosystem

Selecting the best observability platform for AI depends on alignment with your organisation’s size, complexity, and goals. Consider:

• Breadth and depth of telemetry coverage
• Level of automation and intelligence provided
• Developer experience, onboarding, and ease of integrating with your stack
• Regulatory and compliance features for auditability
• Ecosystem fit, including support for your preferred cloud, frameworks, and workflows

Investing in the right observability platform is foundational for a resilient, auditable, and high-velocity AI practice in 2025 and beyond.

Guest author: Or Hillel, Green Lamp

Image source: Unsplash

The post 5 best AI observability tools in 2025 appeared first on AI News.