🚨 Pulse of Innovation 🚨

Breaking news in the healthcare AI

Oracle Rolling out EHR with built-in AI tools

Electronic Health Records (EHR) have been integral part of modern medicine and the wealth of data it generates enables research, quality improvement and many more vital aspects of the healthcare. Use of AI is inevitable in healthcare and integration of AI tools in the EHR is critical.

The Epic Systems leads the US EHR market with 39% share and Oracle’s Cerner is second with 25% market share. The integration of third party applications in the the EHR is burdensome as it requires several quality and safety assessment for data integrity. Therefore several Health AI applications have not been able to integrate in the current leading EHR systems.

With Oracle (acquired Cerner for $28B in 2022) rolling out EHR with built-in AI tools several applications will be readily available for use and it will likely make third party integration of AI tools in the existing system faster and more secured.

The new Oracle EHR is not only able to integrate Natural Language Processing (NLP) and Generative AI seamlessly it will power the physician experience to be more time-effective with automatic summarization and voice enable search through medical records. You will be able aske questions to the EHR system and it will provide the answers, take notes, take orders verbally and execute several tasks behind the scene such as summarizing test results, building a physician encounter summary and critically involving the patient physician through out the process. The system likely available in June 2025.

🧑🏼‍🔬 Bench to Bedside👨🏽‍🔬

Developments in healthcare AI research and innovations

PRISM: Patient Records Interpretation for Semantic clinical trial Matching system using large language models.

The PRISM study proposes an innovative clinical trial matching system aimed at automating the alignment of patients with suitable clinical trials using large language models (LLMs). The system, leveraging unstructured electronic health records (EHRs) and advanced natural language processing, aims to tackle the complexity of determining trial eligibility, a traditionally labor-intensive process. The core of this approach, “OncoLLM,” a fine-tuned LLM model, demonstrates enhanced accuracy in patient-trial matching, surpassing other LLMs such as GPT-3.5 and showing performance comparable to medical professionals. PRISM employs a structured pipeline, enabling bi-directional matching—both identifying trials for patients and patient cohorts for trials.

Key Aspects of the Study

1. Pipeline Design: PRISM incorporates an end-to-end pipeline that processes trial inclusion/exclusion criteria alongside unstructured EHRs. This involves decomposing trial criteria into binary questions for compatibility checks with patient records, creating a scalable and interpretable model.

2. Performance and Model Validation: OncoLLM, a proprietary model fine-tuned specifically for oncology, achieves an accuracy rate comparable to that of GPT-4 but at significantly reduced computational costs. Its ranking algorithm consistently places relevant trials within the top-three for patient eligibility.

3. Scalability and Privacy: Unlike other proprietary models, OncoLLM is deployable within institutional infrastructures, preserving patient data privacy. The system shows potential for reducing manual EHR review times, a known bottleneck in clinical trial enrollment.

Drawbacks and Limitations

While the PRISM system shows promising results, it still encounters challenges in certain areas. The model’s reliance on unstructured data can lead to gaps, particularly where crucial information, such as laboratory values, is stored in structured formats. Additionally, embedding-based retrievers may fall short in handling the nuanced nature of medical data. Lastly, the study highlights the necessity for robust annotation processes and standardization in criteria assessment, as variability among annotators presents challenges in achieving consistent results. Future iterations might benefit from integrating both structured and unstructured data sources and enhancing retrievers to improve accuracy and reliability in clinical applications.

 Nature Digital Article

• Walkfit is a cutting-edge walking stick attachment designed to assess fall risk in elderly users by employing AI and pressure sensors. Developed by Rahi Shah and Hriday Boriawala, 12th-grade students from Poddar International’s robotics lab, the device transforms any standard walking stick into a smart tool for health monitoring.

• Walkfit consists of a circuit fitted with six sensors—two on the grip and four at the base. These sensors continuously track the pressure exerted by the user and wirelessly connect to a laptop via Bluetooth. By generating real-time heat maps, Walkfit illustrates pressure distribution across the stick, showing the user’s reliance on it for balance and support. The heat maps use a gradient system, with blue indicating areas of minimal pressure and red showing areas of high pressure.

• Through data from force-sensitive resistors and gyroscopic sensors, Walkfit accurately measures the user’s weight supported by the stick and evaluates their risk of falling. This approach not only highlights an individual’s dependency level but also predicts potential fall risks, causes of unstable gait thereby offering a powerful tool to improve safety and stability for elderly users.

  

  Source: The Better India Article

• The founders hope to develop an app to help caregivers and healthcare providers insight into risks of falling and effectiveness of treatment. The device is tentatively priced at 60 USD and has been tested on 50 geriatric patients in old age homes.

Advances in Ethical Use of AI

ICMR Sets Ethical Guidelines for AI in Healthcare
The Indian Council of Medical Research established ethical guidelines for AI in healthcare, covering patient consent and data security principles.

ARPA-H Launches AI Degradation Program
The Advanced Research Projects Agency for Health introduced a program focusing on auto-correcting AI-enabled tools that become misaligned with their training data over time.

Emory University Launches Empathetic AI for Health Institute
Emory University introduced the Emory Empathetic AI for Health Institute, aiming to improve health equity and patient outcomes through AI and machine learning.

AI-Driven Risk Prediction Models for Minority Patients
Researchers at Emory AI Health are developing dedicated AI-based risk-prediction models for minority patients to address health disparities.