[clinical] Add PatientID to the per-collection tables (whenever it is not already available) about etl HOT 7 CLOSED

So for every clinical collection I have seen there is a column that identifies the patient. In the clinical_meta data table, this column is indicated by the case_col column. Just query

SELECT collection_id, variable_label, variable_name FROM idc-dev-etl.clinical.clinical_meta where case_col= True

This column is sometimes PatientID, sometimes subjectId etc.

More problematic - sometimes the format of the PatientID is different in the clinical data than in the DICOM files. In some collections the letter case is different (ie all lower vs mix of upper & lower), or '-' gets replaced by '_' etc.

from etl.

Yes, I understand. But I think sometime it is possible there might be more substantiative differences (e.g., I think I saw somewhere where the spreadsheet would include suffix or number, instead of the full PatientID). I think it will be important to do the checks, confirm the values are matched with what we have in DICOM, and if not - add custom mapping, which hopefully should be feasible to define.

from etl.

@G-White-ISB I do not see this issue addressed in the current iteration of the tables. For example, in this table idc-dev-etl.idc_v9_clinical.lung_pet_ct_dx_clinical, there is tcia_case_id, but not PatientID.

Since this mapping will need to be curated manually, maybe we could add an additional attribute in this JSON https://github.com/fedorov/actcianable/blob/master/output/clinical_collections.json for this purpose? Of course, a complication is that the mapping will need to be (in the general case) from a column in a given sheet within a given file to PatientID. We could also introduce an additional mapping file, if this makes more sense. What are your thoughts?

I think without PatientID column in each of the clinical tables, this data will be pretty much useless, so I would rather postpone release until v10 than not have it in v9.

from etl.

tcia_case_id is the column I generated to match the case IDs as displayed in the IDC portal (and the TCIA portal). Do these case IDs match the PatientID in the dicom_all table?
I can rename tcia_case_id to PatientID.

Each table also retains the original case or patient identifier column. This identify of this column is recorded in the case_col column in the clinical_meta table:

SELECT table_name, variable_name FROM idc-dev-etl.idc_v9_clinical.clinical_meta where case_col=TRUE

from etl.

@G-White-ISB I do not see tcia_case_id column - I was expecting to find those for each collection_id using the query below - perhaps I am missing something.

SELECT collection_id, variable_name FROM `idc-dev-etl.idc_v9_clinical.clinical_meta_column` 
where variable_name like "%tcia%"

Also, we should not call it tcia_case_id - we will have data from sources other than TCIA, and also we should not add confusion between case and PatientID. I would suggest calling it dicom_PatientID. What do you think?

from etl.

These were changed to dicom_patient_id. Every clinical table has this column
SELECT collection_id, variable_name FROM idc-dev-etl.idc_v9_clinical.clinical_meta_column where variable_name like '%dicom_patient_id%' order by collection_id

from etl.

Excellent, thank you! And I confirmed that all of the 27 collections that have clinical data also have patient ID mapped.

WITH
  collections_with_ptid AS (
  SELECT
    DISTINCT(collection_id),
    variable_name
  FROM
    idc-dev-etl.idc_v9_clinical.clinical_meta_column
  WHERE
    variable_name LIKE '%dicom_patient_id%'
  ORDER BY
    collection_id)
SELECT
  DISTINCT(clinical_meta_column.collection_id)
FROM
  idc-dev-etl.idc_v9_clinical.clinical_meta_column AS clinical_meta_column
LEFT JOIN
  collections_with_ptid
ON
  clinical_meta_column.collection_id = collections_with_ptid.collection_id
WHERE
  clinical_meta_column.collection_id IS NOT NULL

from etl.