dhis2-devs team mailing list archive

[Calle Hedberg <calle.hedberg@xxxxxxxxx>]

Hi,

I agree 100% with Marta's points. The practice of ensuring that all
catcombos (and attributecombos) are discrete break-downs of the DE is
generally advicable both because it is easier to avoid analytical mistakes
and because it's conceptually easier to explain and "internalise" for users.

That said, HISP-SA had been forced to create a number of exeptions to this
general rule:
- in order to generate stacked graphs for 90-90-90 monitoring, we had to
create derived data elements with non-discrete catcombos (so basically a
hack because of limitations in current DHIS2 reporting apps)
- for audit purposes, where facilities are revisited and the same data
collated & captured again, we also ended up using non-discrete catcombos
(original data, audit data, discrepancy between the two)
but those are special cases.

A common mistake related to attributeoptioncombos are when
attrributeoptioncombos are used to capture data from different partners for
the same orgunit - so e.g. the Ministry, MSF, PSI, and Red Cross are all
supporting the same facility. Then it is critical to ensure that the
different partners do not collate data for the same events/patients (aka
several partners "claim" the same patients because they do provide SOME
part of the total service delivery). So multiple partners provide services
to the same patients, it might be necessary to represent that using
different DEs, otherwise you end up doubling or tripling data (you do
remember the case for Botswana discovered by the New York Times a few years
ago, where PEPFAR claimed they supported ~16,000 ART patients in Botswana -
whereas they had actually only supported some workshops for the Ministry
staff providing ART services...)

The boundary between what's covered in the orghierarchy dimension and what
covered in the fact dimension also depend on what your ou reporting units
are. As an example, let us consider deaths from trauma in a conflict
situation. You might want to differentiate between
- deaths before anybody finds/attends the patient ("Trauma - dead on
site"). Those numbers are useful only for sensitisation etc.
- deaths while under EMRS care ("Trauma - dead on arrival"). Ditto use, but
also relevant to analyse EMRS equipment & staff skills + EMRS effectiveness
(no of ambulances, speed of transport)
- deaths while under the care of health personnel in a health facility
("Trauma - death"). Relevant for general sensitisation like the above, but
also for analysing quality of care, capacity, equipment/blood supplies, and
similar.
How many data elements you need for the above will depend on e.g. if EMRS
ambulances are separate reporting units - if yes, they will presumable
report on the first two types above.

Beyond the basic death due to trauma, you might then use catcombos to break
those down into other discrete categories e.g. Trauma due to gunshots, due
to IEDs, due to bombing, due to physical assault, etc etc. OR you might
want sub-divide deaths into patient characteristics (age, gender,
soldier/civilian, resident/refugee, etc). Most such break-downs can be
handled through catcombos as long as the category options are discrete - if
they are not, I would in general create separate Data elements.

As your DE granularity grows, you should also consider moving to case-based
data collection (possible even a full EMR system) with diagnosis,
procedures, outcomes etc - which enables more detailed data mining. You can
see that the small example above increasingly start looking like a list of
ICD-10 trauma codes.....

Regards
Calle


On Thu, 15 Nov 2018 at 14:25, Marta Vila <martavila@xxxxxxxxx> wrote:

> Hi Damien!
>
> there is no magic formula for all these config decisions, but one pretty
> generic good practice would be that the aggregation of your cat. combo
> options into a data element should be meaningful. As it is the value that
> users will get when they select the DE in data visualizer or pivot table
> without details.
>
> From your example: I would say that using two different data elements for
> Malaria and Malaria (deaths) is correct. If you put them under one unique
> Malaria DE with a category (cases/deaths), the total number will be mixing
> cases and deaths, which is most likely incorrect from an analysis point of
> view and very easy to misinterpret by the users.
>
> Another generic good practice advise is to design your configuration
> thinking of your desired output (analysis).
>
> Hope it helps!
> Marta
>
> On Wed, 14 Nov 2018 at 20:17, Damien Scarlett <
> Damien.Scarlett@xxxxxxxxxxxxxxxx> wrote:
>
>> Hi Calle,
>>
>>
>>
>> Thanks for your 2c worth (we say 20c from where I am from J). I may be
>> explaining a different perspective here even still being a part of the
>> wider MSF posse but we (the Brussels office) have some indicators (e.g. %
>> of Morbidity / Mortality) that we have created additional related Data
>> Elements that map to the disease e.g. we have different DEs for ‘
>> *Malaria’* & for ‘*Malaria (death)*’ to use for Mortality-type
>> indicators. Mind you this is mapping to the aggregate data capture.
>>
>>
>>
>> I’m unsure if there are any Best Practices for these types of indicators
>> & how to set them up (I would presume they are common) but if there is any
>> we would be happy to know them & how organisations have configured these.
>>
>> From reading the below it may be wise to create Cat. Combos instead of
>> additional DE’s but other people may have experience in which is best to
>> capture this & easier to manage over the long term.
>>
>>
>>
>> Have a good evening,
>>
>> *Damien *
>>
>>
>>
>> *From:* Dhis2-users <dhis2-users-bounces+damien.scarlett=
>> brussels.msf.org@xxxxxxxxxxxxxxxxxxx> *On Behalf Of *Calle Hedberg
>> *Sent:* Wednesday, 14 November 2018 6:02 PM
>> *To:* Jaime Bosque <Jaime.Bosque@xxxxxxxxxxxxxxxxx>
>> *Cc:* DHIS 2 Users list <dhis2-users@xxxxxxxxxxxxxxxxxxx>; dhis2-devs <
>> dhis2-devs@xxxxxxxxxxxxxxxxxxx>
>> *Subject:* Re: [Dhis2-users] [Dhis2-devs] Filtering indicators by OU
>>
>>
>>
>> Jaime and others,
>>
>>
>>
>> Using category combos (or attribute combos) are
>> fundamentally/conceptually the same as having multiple data elements,
>> because at the end of the day you will have the same number of physical
>> records in your database (e.g. the datavalue table). Or in other words, the
>> level of "granularity" in your data remains the same. (The main advantage
>> in using catcombos is a reduction in the amount of meta-data records you
>> maintain: so instead of creating and maintaining let us say 300 data
>> elements, you create and maintain 50 data elements with 2 gender catoptions
>> and 3 age catoptions (55 meta-data items) -> 50x2x3=300
>> dataelement&catoptioncombos variants. (Maintaining 55 items is presumably
>> easier than 300 - although for many users it's also more difficult to fully
>> grasp and I have seen a lot of databases where the catcombos have become
>> really messy over the years. Typical examples are multiple changes in e.g.
>> age groups over time).
>>
>>
>>
>> There are situations where you can use for instance orgunit groups to get
>> "ou-dependent" indicator values, but I would in general not recommend them
>> except when the relation between orgunit(type) and indicator is an inherent
>> and stable dimension of every indicator at a particular orgunit level. A
>> practical example:
>>
>> 1.
>>
>> If you collect data per hospital, and the hospital have 8 different wards
>> (1 maternity, 2 medical, 2 surgical, 1 paediatrics, 1 ICU, 1 orthopaedics),
>> and you need e.g. Bed Utilisation Rates for each ward type, you would
>> typically create one data element & catcombo set for each type:
>>
>> Inpatient days - maternity, inpatient days - medical, inpatients
>> separations - maternity, inpatient separations - medical, inpatient death -
>> maternity, inpatient death - medical, etc.
>>
>> 2.
>>
>> If you on the other hand collect the same data per hospital ward (i.e.
>> expand your orghierarchy to the sub-facility level), and these are grouped
>> per ward type, you need only
>>
>> inpatient days, inpatient separations, inpatient deaths
>>
>> and you would do analysis per ou group. NOTE, though, that such a model
>> means you cannot change a ward's orgunittype without messing up the
>> historic data analysis. So if a ward is changed from a medical to a
>> surgical ward, you would have to close the medical ou and open a surgical
>> ou.
>>
>>
>>
>> Another perspective on the same is to say that the higher granularity you
>> have in the geographic/administrative dimension (orghierarchy), the less
>> granularity tend to be required in the fact dimension (data
>> elements/indicators). When implementing something like option 2, though,
>> you must also consider aggregation of data - if for instance some of your
>> indicators are using higher level data (e.g. district totals) for some
>> denominators, you might end up struggling to get that.
>>
>>
>>
>> Jaime do not provide any specifics about exactly what A,B,C is and
>> whether each of them are firmly and permanently bonded to specific outypes
>> only so it is difficult to know if "transferring" some of the granularity
>> to the orgunit dimension is advisable (as indicated above, aggregated
>> analysis needs might also play a role). As a general rule, I would only
>> recommend such shifts if the artifact/event/measurement that the data
>> elements represent is firmly and permanently linked to orgunit types (as is
>> the case with ward-specific indicators) - and even then users have to
>> understand the inherent restrictions like limits to changing the outype
>> group over time. In the case of MSF type health units, it might mean that
>> e.g. an MSF unit grouped under "emergency trauma centre" cannot be kept as
>> the same orgunit if the unit changes from e.g. emergency trauma to
>> long-term rehab. (Not sure if the example is good, but hope you get my
>> drift).
>>
>>
>>
>> My 2c worth...
>>
>>
>>
>> Calle
>>
>>
>>
>> On Wed, 14 Nov 2018 at 16:44, Jaime Bosque <
>> Jaime.Bosque@xxxxxxxxxxxxxxxxx> wrote:
>>
>> Dear Barnabas,
>>
>>
>> Thanks a lot for your response. In the end what we were doing is not
>> exactly what you were saying (no Category combinations). So after checking
>> with a couple of colleagues here we realized that this was not a
>> possibility and so we are not merging the data elements in the production
>> environment.
>>
>>
>>
>> Kind regards,
>>
>>
>>
>> Jaime Bosque Torrecilla
>>
>> Applications Technician, eHealth & Operations Applications (‘Apps4OPS’)
>>
>> *Projects & IT Unit*
>> *Médecins Sans Frontières (MSF) Spain – Barcelona Office*
>>
>> Fixed: +34 935 213 048 – Skype: msfe-healthdatatech2
>>
>> Email: jaime.bosque@xxxxxxxxxxxxxxxxx – www.msf.org
>>
>>
>> ------------------------------
>>
>> *From:* Barnabas Akumba <akumbabarns@xxxxxxxxx>
>> *Sent:* 13 November 2018 17:46:26
>> *To:* Jaime Bosque
>> *Cc:* dhis2-users; dhis2-devs
>> *Subject:* Re: [Dhis2-devs] Filtering indicators by OU
>>
>>
>>
>> Hello Jaime,
>>
>>
>>
>> If I understand you correctly, you have three Data Elements that are all
>> disaggregation of a particular Data Element X.
>>
>> Your initial design was that you created each of the DEs (A, B, C) as
>> individual Data Elements.
>>
>> My understanding is that you merged them by creating a Category
>> combination with a Data Dimension Type "Disaggregation" A, B and C as
>> options and assigned to the Element Called X. If that is the case, during
>> data entry, the system will disaggregate the Data Element X into X(A), X(B)
>> and X(C).
>>
>> If you have gone this way, your analysis won't be an issue.
>>
>>
>>
>> Please verify and confirm if this is similar to what you want.
>>
>>
>>
>> Regards
>>
>>
>>
>> On Tue, Nov 13, 2018 at 5:34 PM Jaime Bosque <
>> Jaime.Bosque@xxxxxxxxxxxxxxxxx> wrote:
>>
>> Hello and thanks for the quick response... it seems it might have not
>> been really clear.
>>
>>
>>
>> I will try to put an example.
>>
>> We had 3 Data Elements (A,B and C) that have been merged in one (let's
>> call it X). The reason they have been merged is because A,B and C belong to
>> a Medical Service called Nutrition and as they could be filtered later by
>> that service (Nutrition A, Nutrition B and Nutrition C) it simplified the
>> data input and analysis.
>>
>>
>>
>> So, for example, now in the OU Nutrition A people will fill the Data
>> Element X. And when doing the analysis they will select Data Element X
>> filtered by service Nutrition A. Same for B and Same for C.
>>
>>
>>
>> These three indicators were used before separately in several indicators,
>> sometimes we needed A+B+C/100. In those cases we don't need to do anything
>> as they have already been merged and we can use X/100.
>>
>>
>>
>> However, and here is the issue. We have realized that when calculating
>> Deaths, the only valid deaths are those coming from Nutrition A and
>> Nutrition B. So before, we would have the indicator Nutrition A + Nutrition
>> B.
>>
>>
>>
>> Ideally, I would like to be able to create an indicator where Data
>> Elements can be filtered, so I could create an indicator like X (Nutrition
>> A) + X (Nutrition B).
>>
>>
>>
>> If this is not possible it obliges me to keep the old model with A, B and
>> C separated instead of the X. I hope it is clear... I will happily provide
>> more information if requested.
>>
>>
>> Thanks,
>>
>>
>>
>> Jaime Bosque Torrecilla
>>
>> Applications Technician, eHealth & Operations Applications (‘Apps4OPS’)
>>
>> *Projects & IT Unit*
>> *Médecins Sans Frontières (MSF) Spain – Barcelona Office*
>>
>> Fixed: +34 935 213 048 – Skype: msfe-healthdatatech2
>>
>> Email: jaime.bosque@xxxxxxxxxxxxxxxxx – www.msf.org
>>
>>
>> ------------------------------
>>
>> *From:* Barnabas Akumba <akumbabarns@xxxxxxxxx>
>> *Sent:* 13 November 2018 16:48:02
>> *To:* Jaime Bosque
>> *Cc:* dhis2-users; dhis2-devs
>> *Subject:* Re: [Dhis2-devs] Filtering indicators by OU
>>
>>
>>
>> Hello Jaime,
>>
>>
>>
>> Could you be more detailed? Let's understand your implementation (eg. of
>> DEs).
>>
>>
>>
>> Regards
>>
>>
>>
>> On Tue, Nov 13, 2018 at 4:44 PM Jaime Bosque <
>> Jaime.Bosque@xxxxxxxxxxxxxxxxx> wrote:
>>
>> Good afternoon. First email to the list looking for some help...
>>
>>
>>
>> We are revising our data model and we have come out with a simplification
>> on the number of Data Elements. For example, before we were using three
>> Data Elements that we have realized they can be joined and then filtered by
>> Organisation Unit when doing the analysis.
>>
>>
>>
>> This works wonder in data analysis, however we have realized that this
>> will not work for the indicators as there is no possible way of filtering
>> by OU in the indicators. Am I correct? Is there a way that this could be
>> done somehow?
>>
>>
>>
>> Thanks a lot,
>>
>>
>>
>> Jaime Bosque Torrecilla
>>
>> Applications Technician, eHealth & Operations Applications (‘Apps4OPS’)
>>
>> *Projects & IT Unit*
>> *Médecins Sans Frontières (MSF) Spain – Barcelona Office*
>>
>> Fixed: +34 935 213 048 – Skype: msfe-healthdatatech2
>>
>> Email: jaime.bosque@xxxxxxxxxxxxxxxxx – www.msf.org
>>
>>
>>
>> _______________________________________________
>> 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
>>
>>
>>
>>
>> --
>>
>>
>>
>> Barnabas AKUMBA
>>
>>
>>
>> *Mobile:* +2348036195778
>>
>> *Skype:* barnabas.akumba
>>
>>
>>
>>
>> --
>>
>>
>>
>> Barnabas AKUMBA
>>
>>
>>
>> *Mobile:* +2348036195778
>>
>> *Skype:* barnabas.akumba
>>
>> _______________________________________________
>> 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
>>
>>
>>
>>
>> --
>>
>> *******************************************
>>
>> Calle Hedberg
>>
>> 46D Alma Road, 7700 Rosebank, SOUTH AFRICA
>>
>> Tel/fax (home): +27-21-685-6472
>>
>> Cell: +27-82-853-5352
>>
>> Iridium SatPhone: +8816-315-19119
>>
>> Email: calle.hedberg@xxxxxxxxx
>>
>> Skype: calle_hedberg
>>
>> *******************************************
>>
>>
>> _______________________________________________
>> Mailing list: https://launchpad.net/~dhis2-users
>> Post to     : dhis2-users@xxxxxxxxxxxxxxxxxxx
>> Unsubscribe : https://launchpad.net/~dhis2-users
>> More help   : https://help.launchpad.net/ListHelp
>>
>

-- 

*******************************************

Calle Hedberg

46D Alma Road, 7700 Rosebank, SOUTH AFRICA

Tel/fax (home): +27-21-685-6472

Cell: +27-82-853-5352

Iridium SatPhone: +8816-315-19119

Email: calle.hedberg@xxxxxxxxx

Skype: calle_hedberg

*******************************************