Key Takeaways:
Athis year’s AAMI Exchange and Medical Device Software Development Summit, Nihon Kohden Digital Health Solutions (NKDHS) highlighted how continuous physiological data can be transformed into clinically meaningful insight through the Nihon Kohden Digital Health Platform (NKDHP). Designed to compute, stream, and store physiological data in real time, DHP enables hospitals to deploy predictive analytics at scale, regardless of complexity or clinical use case. The presentations emphasized that success in AI-driven care relies not only on algorithm accuracy but also on thoughtful clinical integration, interpretability, and the ability to deliver insights directly within the clinician’s workflow.

In the NKDHS research article, Bridging the Gap: Overview of Hospital-based System for Data Collection and Model Deployment, a real-time hospital-based streaming architecture was presented to address a key challenge in healthcare: the lack of scalable infrastructure to support clinical integration of AI. While predictive models continue to advance in accuracy, their clinical impact is often limited by fragmented data systems and the absence of platforms capable of ingesting and delivering continuous patient-centric physiological data at scale.

The platform described in the article was designed to ingest and store high-frequency waveform and vital sign data from bedside monitors, transform and normalize the data in real time, and stream structured outputs to predictive models and visualization systems. To validate the architecture’s versatility and performance, three distinct models were tested: a deep learning model for predicting cardiac arrest, a logistic regression model for detecting hemodynamic instability, and a rule-based respiratory risk score. Each model was deployed and tested using the DHP data replay engine, enabling evaluation of real-time compute performance on streaming data from 100 bedside monitors. The results confirmed consistency and reliability under varying operational loads.

Results demonstrated low-latency data delivery, stable algorithm execution, and consistent alignment between predicted outputs and clinical timeframes across all tests. These findings support the platform’s capacity to operationalize models at scale with differing computational requirements and data dependencies. By enabling the delivery of live, patient-centric to algorithms and distributing results to clinical applications, this streaming infrastructure lays the foundation for scalable AI deployment in hospital settings.

Predictive Models at the Bedside: Insights from AAMI Exchange
At the 2025 AAMI Exchange, Timothy Ruchti, PhD, and Abel Lin of Nihon Kohden Digital Health Solutions presented “AI at the Bedside: Tackling Development Challenges and Clinical Integration.” Their session examined the realities of deploying AI-driven predictive models in clinical settings, emphasizing that high model performance alone is not enough when true impact requires seamless clinical integration.

Ruchti and Lin stressed that predictive models trained on continuous physiological data, including waveforms and vital signs, must operate within a robust infrastructure that supports timely delivery of insights to the bedside. This perspective was reinforced by recent NKDHS research, which outlined the development of a real-time hospital-based streaming platform capable of ingesting and computing high-frequency data, executing predictive models, and streaming results directly to clinicians.

Overcoming the Last Mile: Takeaways from the Medical Device Software Development Summit
During the 2025 Medical Device Software Development Summit, Timothy Ruchti, PhD, delivered a presentation titled “Transforming Patient Care with Predictive Algorithms & Real-Time Physiological Data.” This session shifted focus to the challenges of large-scale deployment such as bias, model generalizability, and the infrastructure needed to support real-time delivery.

Ruchti presented recent findings indicating that predictive applications relying solely on physiological data from bedside monitors are more likely to generalize over time and across diverse clinical settings than those based on Electronic Medical Record variables, which may be surrogates for clinical suspicion.

These predictive models provide clinicians with information they would not otherwise have by effectively anticipating catastrophic events and contributing to proactive treatment decisions. Avoiding complications and events may lead to better outcomes and more efficient care.

Given this promise, ensuring that a reusable predictive analytics infrastructure is available becomes a strategic investment. Drawing on findings from NKDHS's recent research paper, Ruchti emphasized that platforms should be capable of ingesting physiological data from any device vendor, scale with the number of devices and algorithms and support both algorithm testing and deployment. These capabilities are essential to ensure that the resources expended to develop effective AI algorithms lead to improve clinical care and operational efficiency.

The presentation also addressed how predictive outputs must be transformed into meaningful clinical action. Rather than simply generating risk scores, the Digital Health Platform enables structured, patient-specific insights to be routed directly to clinicians as to bridge the gap between machine learning outputs and real-time care decisions.


Ready to see how the Digital Health Platform can transform your Research?

NKDHS’s Digital Health Platform turns unstructured data into predictive clinical insights by empowering research teams to move at the speed of data. Built for interoperability and scalability, it integrates with existing workflows to deliver value from day one. Explore the platform and discover how we’re helping healthcare systems move from algorithms to impact.

Learn more at: Digital Health Platform