The GNUmed Report Generator
GNUmed offers two fundamentally different ways to search the database:
- across the EMR of the currently active patient
- across the entire medical database regardless of the active patient
The second approach is sometimes called data mining. GNUmed has a plugin called Reports to enable you to create database-wide reports.
The plugin is intended to generate simple reports. The powers of this tool do not go beyond what you can do from within PostgreSQL. However, one can enhance PostgreSQL with the "R" procedural language (or, in fact, any other one) in order to unleash considerable statistical powers right from within the SQL query.
For anything more sophisticated than that (say, post-processing the report results) one will have to turn to custom scripting, off-the-shelf report generators or data mining tools such as NetEpi.
Generating Reports
To generate a report from the database you need to run an SQL query. The query has to be typed into the Command (SQL) field of the Reports plugin. Then hit the button [Run]. The results will be shown in the list at the bottom. The columns of the list will correspond to the columns of the database table(s) you collect data from with the query. You may want to use the SQL AS column alias syntax to map database columns to convenient list column labels.
Here are a few things to know:
- you don't need to worry about leading or trailing whitespace/linefeeds
- your query may span several lines
- you don't need to end your query with a semicolon (";") - but you can
- you can paste query text from elsewhere via the clipboard
- you can drag a file onto the Reports plugin and GNUmed will interpret the file content as the query to run
- note that you need to drag the file onto an area of the plugin outside the actual query command field (this will be improved later)
- the plugin will artificially limit the results list to 1000 rows (one thousand) such as to somewhat safeguard against queries going berserk
- the queries are run in a read-only connection with the credentials of the user who logged on with this client
- no writing to the database is possible, not even as a side-effect of a
select my_writing_func()
query
- you might want to make your query return a column named
pk_patient
- this will enable you to double-click on any row in the results list which will activate the patient identified by the database ID found in
pk_patient
- if there is no such column nothing will happen (besides an error message being shown)
- note that you can, of course, make a query return an arbitrary number in a column named
pk_patient
in which case the corresponding patient will, indeed, be activated upon double-clicking a row but there may not, in fact, be any other meaningful correlation of the patient with that row
Visualizing reports
Hitting the [Visualize] button will let you select a column from the report results list from each of the x- and y-axis. The data in those columns is extracted from the report and sent to gnuplot for display.
Reusing report definitions
The Report field acts as a phrasewheel offering names of reports that were previously saved in the database. You can either type part of a name our part of a query (such as a table name) and select a report definition from the appearing dropdown match list. The corresponding query will be loaded from the database.
If you press [Save] the report definition will be saved in the database. If the report name is already known in the database the existing report definition will be overwritten. If not a new report definition will be created.
Hitting [Contribute] will email the report definition (name and query - nothing else) to the mailing list of the GNUmed community for all to share. This will happen anonymously. If you want to receive credit for it you'll have to actively claim it on the mailing list.
Note that report results are only preserved as long as the client instance they were generated in stays open. They will, however, survive changing the active patient.
The [Schema] button will take you to the GNUmed database schema documentation in our wiki for your reference.
Sample queries
How many believed-to-be-alive patients remain in the praxis database?
select count(*) from dem.identity
where deleted is FALSE / TRUE
where deceased is NULL / NOT NULL
List for me the (hopefully less than 1024) patients in this database:
select lastnames, firstnames, title, pk_identity AS pk_patient
from dem.v_basic_person
where dem.v_basic_person.lastnames is NOT NULL
order by lastnames, firstnames
List for me patients having a particular postal code
select number, street, dem.v_basic_person.lastnames, dem.v_basic_person.preferred, dem.v_basic_person.firstnames, suburb, urb, postcode, pk_identity as pk_patient
from
dem.v_basic_person
inner join
dem.v_pat_addresses
using (pk_identity)
where
LOWER(dem.v_pat_addresses.postcode) = 'inputDesiredPostalCodeHereInLowerCase'
order by
street, number
List for me the patients waitlisted (without a waiting_zone specied) for more than 14 days:
select lastnames, firstnames, title, comment, waiting_time_formatted, pk_identity as pk_patient
from
clin.v_waiting_list
where
waiting_time < '14 days'
and waiting_zone is NULL
A query that (new in gnumed_v9) can search for patients based on the diagnostic code
select *
from
dem.v_basic_person
inner join
clin.v_coded_item_narrative
using (pk_identity)
where
code = ...
and coding_system = ...
and soap_cat = ...
;
A query that would help by providing more fields (and sample values) that could be altered and used to find a patient when the standard patient search field did not permit a patient to be found, perhaps including the communication channels (phone numbers). Such queries could be
select *
from
dem.v_basic_person
inner join
dem.v_person_comms / dem.v_person_jobs / dem.v_external_ids4identity
using (pk_identity)
where
dem.v_person_comms.url = ... /
dem.v_person_jobs.l10n_occupation = ... /
dem.v_external_ids4identity.value = ...
;
A query that would fetch, from the inbox audit table, the messages deleted within the past 7 days, ordered by recency of last-modified
SELECT *
FROM audit.log_message_inbox
WHERE
fk_staff = <staff ID of provider>
AND
audit_action = 'DELETE'
AND
audit_when > (now() - '7 days'::interval)
ORDER BY
<audit_when / orig_when / modified_when> DESC
;
A query that could identify auto-created persons as might result from a data importer
SELECT * from dem.clin_ext_id_type, dem.identity where
dem.clin_ext_id_type.name = "lab autoimport fake person" WHERE
dem.clin_ext_id_type.fk_person = dem.identity.pk
;
There has been discussion offlist between Karsten and Jim "on theory of primary care" modeling levels of clinician diagnostic certainly. Once this would be captured in the encounters it would make for interesting queries to the effect of
The patients I would most worry about would be those who
- remain our responsibility (they did not abandon us)
--> last seen in the most recent 6 (?) months
- and have an active issue or episode of certainty of A or B or C that is
--> persisting over multiple encounters
>= 2 encounters if symptom(s) are "alarming" or "worsening"
>= 3 encounters if B or C
Many patients have a chronic single symptoms at level A, and maybe a
chronic symptom complex at level B, but – as long as their episode is
not worsening (or provided the patient's episodes are not becoming more
frequent which would be a separate clinically informative query) – then
it may be tolerable to optionally and by default omit such patients with
chronicity of > 6 or 9 months from inclusion in the result of a query if
the purpose is "who must I make sure I do not overlook a condition that I
should perhaps be diagnosing?"