Skip to Main Content

Signal Detection in Patient Level Data

We are currently mining patient level data to identify signals about changes in patient conditions. It is very useful to detect early signals in order to be able to respond and move patients from the home to the hospital or from the ward to the ICU. The data is all time series data coming from medical devices. We ae trying to figure how to create a robust library of patterns that can be used for monitoring and prediction as it will help reduce the burden on medical personnel. The approach is very similar to the analysis and prediction on ECG  data for remote patient monitoring. Here is a link to our paper on ECG  predictive monitoring using small data sets - https://preprints.jmir.org/preprint/24388. Attached is also a PPT with a more general description of what and how we do it.

Public (7)
You will need to login to post a comment
Shawnnah Monterrey Nov 13, 2020

It would be great if this data could be correlated with the test results, our platform will host the test results.

Rado Kotorov Nov 13, 2020

What tests results do you have in mind. Can you provide details

Shawnnah Monterrey Nov 13, 2020

I'm thinking there is value in correlating your detection with the actual positive/negative test results, either to validate your algorithm or improve and refine (reduce false pos/neg).

Rado Kotorov Nov 13, 2020

I now understand. Our data is on patients who already have tested positive. We track the progression of the condition

AAG Nov 16, 2020

Is it possible to use Continuous Glucose Monitoring (CGM) technology to detect COVID-19 in an individual and monitor positivity for COVID-19?

Rado Kotorov Nov 16, 2020

that is a very interesting idea. I do not know the answer but it is worth considering

Rado Kotorov Nov 16, 2020

Innovation happens when you try ideas :)

Rado Kotorov Nov 16, 2020

This is how we see the data.

Rado Kotorov Nov 18, 2020

The pattern is very important as you can see on this screenshot single instances of SPO2 drop are not important, but a 30 min continuous decline signals the need for a respirator.

Rado Kotorov Nov 18, 2020

Here is the pitch deck for our proposed solution.

Rado Kotorov Nov 19, 2020

Nature just published a very interesting article about early COVID-19 detection using smart watch and anomaly detection in time series. Quite simple patterns can signal Covid-19 infection. Here is the link to the article - https://www.nature.com/articles/s41551-020-00640-6

Andrea Pitkus Dec 3, 2020

Trying to understand if there are preconditions met by other software. (In general) 1. How to collect AOEs, from ordering provider/patient/specimen collector 2. Integrate into app/LIS or other information source for patient to be married to results of IVD test device/system 3. All transmitted to public health (ELR) 4. All transmitted to HHS (may be met by 3).

Will your approach support all 4 or only portions of above for "tracking lab results" or would that be out of scope and your approach applied much further downstream or only focused on clinical/non lab data from medical devices as desceribed?

Rado Kotorov Dec 3, 2020

Results can be transmitted to us via APIs. Integration of disparate sources can be done within our platform via third party or open source integration tools. So our core value comes when we store the data, create metadata and analyze or monitor incoming streams of data.