iupui-soic / mimic-openmrs-cdss Goto Github PK

To avoid duplicates through distinct itemid, we are able to import fewer obs because concepts that have similar names in prescriptions, labevents, and chartevents have different itemid in d_items and d_labitems.

e.g. SELECT * FROM mimic4.d_labitems l, mimic4.d_items i where l.label LIKE i.label; produces Hemoglobin with 4 different itemid that should refer to the same concept.

One approach could be create a mapping CSV that links all d_labitems.itemid to d_items.itemid and creates only one concept, for which obs.concept_id is the same, even if the obs came from chartevents, labevents, prescription or elsewhere.

The jupyter dashboards project has been deprecated, and the recommended approach is to use Voila (https://voila.readthedocs.io/). The first pass should be to use Voila and render the notebooks as standalone application per patient, execution call made, when the CDSS tab is opened from each patient's dashboard. This will also simplify the multiuser, multi-kernel problem that our current architecture has with running and training models by multiple users for the same patient.

We need a mechanism to check for minor typographical differences in drug names so that they are not repeated in the concept or drug tables. Currently, spaces, multiple spaces, or lack of spaces is creating duplicate concept/drug names in the database.
Below is an example:

Only Numeric and Text concepts (as plaintext without coded answers) have been imported into OpenMRS. To have a complete list of chartevents imported to obs, all concepts have to be imported from d_items . Also to be accurate, Text concepts should be imported as Coded and their coded answers have to be imported based on values in the chartevents table.

When importing from MIMIC-IV, we were able to change the date of birth and death date with the anchor_age when creating patients in OpenMRS. However, the visit, encounter, obs and encounter_diagnosis entries include the MIMIC-IV dates without moving them to an approximate date.

Thus, any analysis needs to factor that if age is used for data related to obs or encounter.

We have noticed that the output of the tabs is not cleared after selecting a new risk. This means that the results of the previous tabs are still visible on the page, and the new results are displayed below them, creating multiple tab results one below the other.

This issue may cause confusion for users, as it may be difficult to distinguish between the results of different selections. To resolve this issue, clearing the output after selecting a new diagnosis. This would ensure that the results of the previous selection are not visible on the page, and the new results are displayed clearly.

This would greatly enhance the user experience and improve the overall usability of the CDSS Dashboard.

Currently the required version of OpenMRS is v1.5 and the API is v1.8.2. These old versions have a lot of vulnerabilities in their dependencies that are inherited because of the module. Upgrading to v2.x will likely remove these.

In the Statistical Test Tab of our dashboard, there are unnecessary spaces between the two tests (Shapiro-Wilk and Statistical Tests) that occupy a lot of space. This causes the dimensions of the dashboard to increase, making it difficult to see the results in one click and requiring scrolling down.

This issue may cause inconvenience to users who are in a hurry and need to quickly access the results of their statistical tests. Moreover, it may also cause readability issues for users with smaller screens.

To resolve this issue, removing the spaces between the two tests. This would make the dashboard more compact and user-friendly, allowing users to see the results of both tests without having to scroll down or expand the dashboard.

Since jupyter notebook deployments will change at each restart, externalizing the authentication token through a global property setting is necessary to show the notebook.
This will still mean that we can only run one dashboard per patient, which is not scalable. However, this will at least work for one patient at time.