Understanding phenotypic overlaps using CO
Last updated
Last updated
This example shows how to understand code overlaps. For example to check if a cohort has any unwanted disease codes in it.
Make sure to read about How to define a cohort in Atlas and about Standard and non-standard codes before continuing further.
In this example, we are tasked with making case and control cohorts about Asthma patients. The case cohort must include the codes and subcodes of ICD8 493, ICD9 493, ICD10 J45, and ICD10 J46. The control cohort must exclude these codes.
According to the instructions here we have made a case cohort including the codes listed above (Asthma_EL_case) and a control cohort (Asthma_EL_Control_2). Now we will check if the control cohort has filtered the codes out of the cohort as intended.
Following the instructions here, upload the cohorts you want to examine into Cohort Operations Tool. It's a good idea to upload the case-cohort into CO as well so you can compare them and spot mistakes more efficiently.
Open CodeWAS and upload the cohorts into Cohorts to compare.
In this example we are only using ICD codes, so from Analysis settings we can exclude other codes. You can also filter using registries if your cohorts are also filtered by them.
To ease the process of inspecting the codes included in the control cohort, we will select the precision of ICD codes to 3 and the precision of other codes to 1 from Advanced options. This will make all the sub-codes fall under one code for easy inspection. For example the codes J45.1, J45.2, J45.3.. etc. will all be under the code J45. The codes with the precision of 1 are selected to not be shown in high detail because our cohorts don't use them.
We will also exclude ICDO3 codes and include only the symptom diagnoses codes by selecting FG_CODE1 from the top right corner.
Click Run CodeWAS analysis & download results. It will run CodeWAS and download the results into a .html file into /home/ivm/Downloads/.
Open the CodeWAS file and click Excel. This will download a .xslx file to the same address as mentioned in the earlier step.
Open the .xslx file with LibreOffice. Using the search function, search for the codes you have tried to filter out of the control cohort. The lines of interest in this example are underlined by the blue line.
From the example Excel, we can see that the filtering works for ICD9 493 and ICD10 J45 codes since the column n_controls_yes shows 0 cases for the codes (highlighted by the green arrow).
Filtering of the ICD8 493 code could be checked, since there are 5 cases of it in n_controls_yes column.
We must improve the control cohort's filtering in some way. All of these steps are good to be repeated with every major filtering change until you are satisfied with your cohorts.
Recommended topics:
