Skip to Main Content

WATERS Crowdicity Community

Early Signal Detection on Patient Level Data During Data Capture Phase

“Nature” magazine reported that Google used 2.5 million patients to build an AI model for cardio pathology predictions. It takes years to collect and learn from such large datasets, but in the case of a pandemic lives are at stake and patients can’t wait. Trendalyze’s time series intelligence platform solves that problem. It provides APIs for rapid data ingestion from any device that can be enhanced to meet healthcare standards and regulations. Simultaneous data collection and pattern discovery in any dataset size facilitates monitoring and prediction of early signals that save lives. By enabling learning from small datasets at system deployment, we provide instant value and long-term extensibility. We identified this need in 2006 when the analysis of 50,000 breast cancer patients was hampered because 99% had unique comorbid conditions. Recent prototypes in Austria suggest that ingest-learn-monitor systems can be build on small data to detect SPO2 signals to move patients from home through to the ICU. Better still, the same signal can be identified from a home or a hospital device. Another example in “Nature” shows a reliable COVID-19 detection signal using data from 35 patients collected by a smart watch. These examples provide us with a blueprint for implementing quickly and maximizing the value of data capture both for early signal detection and for longitudinal research and reporting - a value proven go-to-market solution.

5 min Video of Capstone Project

Elevator Pitch

Google ingests and organizes the world web data via search. Trendalyze ingests and organizes medical device data via pattern search.

Challenge Goals

The project aims to develop an easy-to-use and robust data collection platform that delivers value on deployment and is also scalable and extensible to accommodate the emergence of new devices and data types. To meet this goal, we must keep in mind that the ultimate value of data is realized from the insights that are derived from it, and it is monetized by using these insights for decision support, monitoring, and predictive prevention. Also, the fastest growing health data comes from medical devices that generate granular time series data. Trendalyze provides APIs for data ingestion, a customizable data lake with a metadata layer that connects to analysis, visualization, and monitoring tools with special emphasis on streamed device data.

Feasibility

Trendalyze provides easy installation in the cloud (Azure, GCP, AWS, OCI) or on premise with out-of-the-box APIs to ingest streaming or batch data. The first week would consist of installation, data ingestion and processing. Users can start using our GUIs to configure metadata, visualize, analyze, catalogue and share meaningful data patterns and signals from the second day. Our interface offers query editors to post-process and enhance the data. The implementation of additional APIs would take place in the second week to monitor the patterns identified by the users and send real time alerts as the data is being ingested. Third-party and open source tools for data quality and data governance can easily be integrated in the entire process.

Design

Our platform is designed as a self-service platform for business professionals. Behind the scenes the platform performs complex data processing and pattern detection, which is typically done by IT consultants and data scientists. We leverage new algorithms via simple and intuitive UIs that are explainable and understandable by business professionals. We leverage open source technologies such as Hadoop, node.js and javascript which allows us to rapidly expand the UIs with new features and functions. From day-one users will get the data management, pattern detection, visualization and cataloging UIs. The APIs to ingest data are programmatically configurable, and UIs can easily be developed for the final system.

Innovation

Trendalyze uses a unique approach for pattern detection and monitoring in time series data based on Google like search algorithms. Our technology abbreviates time to market by not needing the model training and inference process that cost money and time. We increase the speed and accuracy of the learning process by making AI transparent and explainable to business experts. Users can visualize, analyze and extract insights as soon as the data is ingested into the platform, and apply monitoring to the stream of time series data generated by devices. Predictions generated by our innovative artificial logical networks are easier to understand, configure and explain compared to neural networks. We have 4 patents pending on the technology.

Flexibility & Scalability

Trendalyze is available in cloud on Azure, GCP, AWS, OCI or on premise on Linux. Our data storage is built on Hadoop for big data processing. The backend can accommodate various data schemas and is readily accessible for data ingestion via streaming and batch APIs. The data in the backend can be accessed for visualization and reporting with out-of-the-box Hadoop connectors. The end user UIs accessible from the browser can be extended via JavaScript. All the UI functionality is available as APIs for third party access. The platform can handle large volumes of data such as monitoring behaviors of 1,000,000 plus customers in real time and performing 15 million simultaneous pattern searches on high frequency streaming data.

Sustainability & Extensibility

We are built on open source technologies and can integrate with any systems via APIs. New data sources, new processing modules and new platforms can be connected via the APIs with minimal effort. The platform can be integrated into an existing continuous integration and deployment process or we can provide the means in the cloud. We have an onshore/offshore team that can provide managed services hosting and managing the Trendalyze platform in the cloud in the long term. The application health and performance will be monitored 24/7. Premium support team is also available to address any application specific issues.

Team & Collaboration

The team is led by Rado Kotorov who has 20+ years of experience in data management, BI and analytics across many verticals and has worked with executives around the world on how to use data and analytics to improve operations. The members of the team include product and project management expert, data integration expert, and UI/UX design expert with a team of data engineers and data scientists collaborating with our onshore/offshore support team. They have implemented complex solutions in hospitals like Mount Sinai, Mayo Clinic, VA, Beth Israel Medical Center, Lehigh Valley Health Network, and many more. The Trendalyze leadership team and its scientific advisors from top universities are listed here - https://trendalyze.com/about/

Additional Comments

Our innovation for building models using small data for remote patient monitoring (RPM) is discussed in “A Personalized Monitoring Model for Electrocardiogram (ECG) Signals: Diagnostic Accuracy Study” forthcoming in JMIR Biomedical Engineering and JMIR Theme Issue 2020:COVID-19 Special Issue (preprint is attached). We also attach a white paper “Scientific Approach of Visual Motif Discovery” which explains the relevance of the approach to patient data analysis including applications such as the NHS National Early Warning Score. Rado’s book “Data Driven Business Models for the Digital Economy” presents a customer use case on how RPM changes the healthcare business model from a discontinuous pay-per-visit to a 24/7 continuous subscription.

What Team(s) contributed to this Capstone Project?

Rado Kotorov has driven the design, the development, the R&D and the commercialization of the Trendalyze platform. Yoshiko Akai has been the chief product officer responsible for product development and testing of the entire Trendalyze platform. Jeff Hendrickson has been responsible for UI/UX and end user acceptance testing.


The team has worked together for many years. Prior to Trendalyze they have managed one of the leading and most scalable Business Intelligence (BI) platforms WebFOCUS, and its data management product suite Iway. They have overseen the development of reporting and decision support applications with large scale disparate data sources environments in over 65 countries.

If you are using patient data, are you using real patient data or mock data? Please use MOCK patient data only

Real data

Tagged users
edited on Nov 29, 2020 by Rado Kotorov
Rado Kotorov

Attach files

@

Rado Kotorov 6 months ago

Here is an HBR article about why small data is so important .
https://hbr.org/2020/02/small-data-can-play-a-big-role-in-ai

Reply 0

Rado Kotorov 6 months ago

Here is a Nature article on using time series data from a smart watch to detect early Covid-19 signals - https://www.nature.com/articles/s41551-020-00640-6

Reply 0

Rado Kotorov 6 months ago

In Chapter 3 we present a study about how digital technologies and remote patient monitoring is changing the healthcare business model. There is a rapid shift from the traditional pay-per visit discontinuous model to a 24/7 continuous subscription model. The book is availaible at Amazon and the publisher sitehttps://www.businessexpertpress.com/books/dat...igital-economy/.

Reply 0

Sophia B Liu 6 months ago

This idea has been advanced to the current phase

People's Choice Voting Extended

Reply 0

Sophia B Liu 6 months ago

This idea has been advanced to the next phase

People's Choice Voting Extended

Reply 0

Andrea Pitkus 5 months ago

Does your approach support lab testing and reporting?

Reply 0

Rado Kotorov 5 months ago

We support reporting and complex analysis. Not physical testing. But for example raspiratory activity can be tracked, analyzed and create alerts in our software. Since many new devices combine tests for respiration and temperature we are the perfect solution for them

Reply 0

Andrea Pitkus 5 months ago

Thank you. Was trying to understand if your approach provides a solution to the problem on the homepage and if so, how.

Reply 0

Rado Kotorov 5 months ago

Yes we do. Especially valuable for devices that collect continous data

Reply 0

Andrea Pitkus 5 months ago

Sounds like you don't allow for laboratory/IVD test data acquisition such as patient performed testing as COVID testing as you know is not continuous, but based upon a specimen in a "point in time" (also as reflected in the LOINC codes). Seems like your approach would be valuable for vitals, ventilator settings (i.e. O2sats) and other non lab observations though.

Reply 0

Rado Kotorov 5 months ago

we focus on continous, but batch data too

Reply 0

Rado Kotorov 5 months ago

The batch data correlates with continuous and some of the remote tests are continuous. The two approaches have to converge to drive accuracy and early detection

Reply 0

Rado Kotorov 5 months ago

We provided a screens shot in some of the documents how to track SPO2 in order to determine if a patient needs to be taken to hospital or put on a ventilator. Many more devices are being developet that track continuous measurments.

Reply 0

Rado Kotorov 5 months ago

the importance of embeded analytics for covid-diagnostic. https://www.bitpipe.com/data/document.do?res_...se%3A+185213%29

Reply 0

Share