Dear all,
Regarding the sensors data what is the best approach to include those in omop? is there a reference guide?
Thanks,
Nuno
Hello Nuno
I’m Jin Choi from Ajou University, South Korea.
Unfortunately, until now, there is no standard methodology for inserting sensor data to OMOP CDM.
Since signal waveforms are not suitable for SQL structure, people usually insert only metadata in CDM (data path, Hz, other information about patients…etc.)
By the way, If you post some sample data, we can discuss more.
There isn’t a single standard approach to supporting signal data in an OMOP CDM. However our group is implementing a proof of concept approach using the MIMIC-4 Waveform Database (WFDB) signal data into an OMOP CDM in the form of summarized features (ex Lead II R-wave amplitude, heart rate, respiratory rate, etc) as periodic representative measurements within the OMOP CDM. The waveform itself is not stored in the OMOP CDM, but the representative measurement is and referenced back to the source signal in a few ways so investigators can use “meta” data for queries, then pull the signals if they wish to pursue signal processing to produce alternative features.
Hi Jin Choi and Manlik Kwong,
thank you for your reply. We are using time series recorded from wearables.
In this case we will also consider the use of metadata to link to the raw data and save some standard metrics as suggested in https://github.com/OHDSI/MIMIC/tree/master/waveforms/Documents
Thank you
I’ll be working in this area of normalizing wearables into an OMOP CDM as well. For some signals like ECG, where the wearable acquires the signal there isn’t a good LOINC code to represent where on the body the electrode was placed - finger (Oura ring), Fitbit/Apple watch (wrist), etc. A custom vocabulary may be needed in the short term while proposal for new codes make its way through the standards bodies.
I am also doing some work to evaluate data like HealthKit and SensorKit and how it might be standardized.