dhis2-devs team mailing list archive

[Bob Jolliffe <bobjolliffe@xxxxxxxxx>]

2009/6/3 Ola Hodne Titlestad <olati@xxxxxxxxxx>:
> On Wed, Jun 3, 2009 at 12:04 PM, Bob Jolliffe <bobjolliffe@xxxxxxxxx> wrote:
>>
>> 2009/6/3 Ola Hodne Titlestad <olati@xxxxxxxxxx>:
>> > Hi Abyot,
>> >
>> > If you read my summary e-mails just before the skype conference you will
>> > see
>> > that my suggestion was NOT to have a different type of data element, and
>> > I
>> > understood from the skype chat that we agreed on the same. What we
>> > talked
>> > about was to possibly make a separation in the user interface to avoid
>> > confusing the users, but in the background use the same DataElement
>> > object,
>> > but I am not sure that will always be needed as there are lot of overlap
>> > between routine and CHIS data elements.
>> >
>> > As you say, if we want to easily reuse datasets and data entry forms
>> > functionality we need to use the DataElement object also for client data
>> > elements. And of course we want to reuse what Murid has implemented
>> > regarding option lists for pre-defined values for data elements.
>> >
>> > The separation comes in DataValue as the PatientDataValue will need
>> > other
>> > properties than the (routine) DataValue.
>>
>> Agreed.  But what would these properties be exactly?  Two options
>> which have surfaced are:
>> 1. an additional patientID attribute; or
>> 2. no additional attribute - association of patient as a "source"
>>
>> The first is most obvious and perhaps simplest.  And I suspect I am
>> the only one crazy enough to see any merit in exploring the second.
>
> patientID yes, but probably also a DataSetID as we need to keep track (and
> separation) of the encounters/visits (instances of a dataset, "a filled
> form") in a more efficient way than we do in DataValue now.
> At least this is how its done in 1.4 Patient module and also for Survey type
> data.
>
>
>>
>> So I'm guessing Abyot will make a PatientDataElement with something
>> like a patientID.
>
> Data elements do not have any direct reference to its source, so this should
> not be necessary. It is the datavalue that keeps this reference and which
> again is controlled by the dataset.

Sorry typo - I meant PatientDataValue ..


> We would in stead need a maintain Patients/Clients in a separate object, and
> pherhaps in a hierarchy (family, village). Lars also liked the idea of
> implementing the source object here, and I am open to that. After all that
> is why we created the source, to have diffeent types of sources to register
> data, not only orgunits.
> The peirod handling might also be different here as we always work on dates
> since these data are snapshots in time and not aggregtated over a certain
> period.
>
> Calle might have some useful input to how patient values are different from
> routine, apart from the security aspect we already discussed some weeks
> back.
>
>>
>> What else?  Do we need a concept like an encounter (or visit) to which
>> a date would be tied?  Or can something be done with a PeriodType?
>
> If we are going to reuse the DHIS concepts of data element, dataset, data
> entry form and data value then the dataset is the key here.
> In many ways routine datasets and "client" datasets are very similar, and "a
> filled form", or what might be called an instance of a dataset, contains
> values linked to data elements for a given period and a given source. Client
> encounters, rows from a register book, are also like that; a client name,
> multiple data elements (columns in the book) with values, and a date. After
> all its the final row of the register book, the total row aggregating all
> the encounters that gives the routine values for a monthly routine dataset.
> This example also illustrates how data elements overlap between client data
> and routine data, routine data are simply the total of "all clients" for the
> month. (This is not the case in survey audit type of datasets e.g. with
> maternal detah audits, but for standard CHIS it is mostly the case)
>
> If we keep track of the DatasetID in a ClientDataValue object we can the
> easily get an ecounter by querying for a client + a date + a dataset, or a
> list of all encounters (within a certain programme (dataset)) by querying
> for a client + a dataset. Of course it would be possible to get all data
> elements belonging to a dataset without directly referencing datasetid in
> datavalue, we do that today for dataset reports. Again, we need to check
> this with Calle or others, but I think client data are different in the way
> that a data element can exist in multiple datasets AND be registered for all
> of them for the same client and date, because the same data elements in
> different datasets might have different meanings and values. For routine
> this is not the case, that is why we di not keep a reference to dataset in
> datavalue, it is enough to use data element to describe the meaning of the
> data.
>
> So each encounter will be a data entry form, and its metadata will be
> controlled through the dataset object, similar to how its done for routine
> data. In the dataset object we need to specify what kind of data that is
> going to be registered,e.g. aggregated, disaggregated, survey(or routine,
> client, survey). Semi-permanent is then included in routine which is a bit
> confusing, that is why I prefer aggregated. Anyway, the principle is the
> same.

OK.  This makes sense.  A register object (for want of a better term)
would be a specialisation of dataset.

> Datasets could be made even more dynamic, as we discussed on Skype, by
> adding other attributes like a set of header data elements and a set of
> footer data elements. These will be based on the same type of data elements,
> but stored or treated in a different way (in data entry and data
> value).Exactly how I am not sure, but we should look in detail at how 1.4
> treats header data elements.

Trying to piece together what a register might look like in xml:

<register name="Immunization register">
    <dataset name="header" >
       <dataelement name="????" >
             <datavalue source="[clinicID]" period="???" value="34" />
       </dataelement>
       <dataelement name="????" >
             <datavalue source="[clinicID]" period="???" value="34" />
       </dataelement>
    </dataset>

   <patientdataset name="immunization data" />
        <dataelement name="???">
             <patientdatavalue source="[patientID1]" value="36"
date="01/01/2010" /> <!-- should date be done with a period type?
             <patientdatavalue source="[patientID2]" value="43"
date="01/01/2010" />
             <patientdatavalue source="[patientID3]" value="35"
date="01/01/2010" />
             <patientdatavalue source="[patientID4]" value="22"
date="01/01/2010" />
        </dataelement>
        <dataelement name="???">
             <patientdatavalue source="[patientID1]" value="36"
date="01/01/2010" />
             <patientdatavalue source="[patientID2]" value="43"
date="01/01/2010" />
             <patientdatavalue source="[patientID3]" value="35"
date="01/01/2010" />
             <patientdatavalue source="[patientID4]" value="22"
date="01/01/2010" />
        </dataelement>
   </patientdataelement>

</register>

While typing the above it occurred to me that header AND footer are
probably not necessary for representing a register.  What we really
need is a set of dataelements associated with the register and a set
associated with register rows.  Whether the elements in the former are
eventually rendered in the header or the footer is probably a
presentation issue which could be determined by, for example, the name
or ID of the dataelement.

Also the above patientdataset is grouped on the dataelement axis.
Could also be grouped on source/patientID, making it closer in analogy
to rows in a register.  Though deriving the latter from the former is
a simple enough transformation.

Abyot, returning to your original question, I don't know if having a
dataelement classification is necessary.  If the dataelements are
always members of a dataset (at least one or only one ..) then
probably not.  But I think you are right that it is only as you hammer
out the detail that the truth might emerge ...

Cheers
Bob

> Ola
> --------
>
>
>
>
>
>
>
>
>
>>
>> Regards
>> Bob
>>
>> > And we also talked about the need to extend the DataSet object to
>> > include
>> > more properties that makes datasets more flexible and dynamic as we need
>> > them for CIS and also for survey data.
>> >
>> > So here I guess we all agree, there is no need to come up with a
>> > separate
>> > PatientDataElement.
>> >
>> >
>> > best regards,
>> > Ola Hodne Titlestad
>> > HISP
>> > University of Oslo
>> >
>> >
>> > On Wed, Jun 3, 2009 at 11:26 AM, Abyot Gizaw <abyota@xxxxxxxxx> wrote:
>> >>
>> >> Hi All,
>> >>
>> >> Couldn't really convince myself as to the need to keep a separate track
>> >> of
>> >> dataelements called patientdataelement. I just did an implementation
>> >> for
>> >> patientdataelement ... but when giving it a thought about linking it
>> >> with
>> >> some custom and predefiend values, then I see that one already in place
>> >> by
>> >> Murod for the routine dataelements. And if we are going to have a case
>> >> of
>> >> like recording multiple values for a single patient dataelement, then
>> >> we
>> >> also will redo all the compex task of linking with options, categories
>> >> and
>> >> their combinations, which is again in place for the routine
>> >> dataelements.
>> >>
>> >> If the need to separate the two - routine and patient is only for the
>> >> purpose of managment, then I think it will be better if we could
>> >> introduce
>> >> an attribute called 'classification' for dataelements. With this
>> >> attribue we
>> >> can classify our dataelements like - Routine, Patient, Header,
>> >> Footer,...
>> >>
>> >> Any input will be appreciated.
>> >>
>> >> Thank you
>> >> Abyot.
>> >>
>> >> _______________________________________________
>> >> Mailing list: https://launchpad.net/~dhis2-devs
>> >> Post to     : dhis2-devs@xxxxxxxxxxxxxxxxxxx
>> >> Unsubscribe : https://launchpad.net/~dhis2-devs
>> >> More help   : https://help.launchpad.net/ListHelp
>> >>
>> >
>> >
>> > _______________________________________________
>> > Mailing list: https://launchpad.net/~dhis2-devs
>> > Post to     : dhis2-devs@xxxxxxxxxxxxxxxxxxx
>> > Unsubscribe : https://launchpad.net/~dhis2-devs
>> > More help   : https://help.launchpad.net/ListHelp
>> >
>> >
>
>