dhis2-devs team mailing list archive

[Bob Jolliffe <bobjolliffe@xxxxxxxxx>]

On 3 October 2011 12:28, Bob Jolliffe <bobjolliffe@xxxxxxxxx> wrote:
> Greetings
>
> Back in the land of the little people and the bogs.  I've just
> returned from a week in Rwanda (which was much too short) where I had
> the opportunity to work with Arthur Heywood and some good people in
> the Rwanda MOH who have done a great job at redesigning the MOH
> collection instruments ie. the forms.
>
> Most of my week was spent on infrastructure stuff but towards the end
> of the week I had a (first) opportunity for me to work on the frontend
> of the dhis process.  This was a *really* worthwhile process for me
> and I learned a great deal.  Anyway, in the process of helping the
> implementors move from the form design to coding the dhis dataelements
> I am left with two interesting implementor design issues which I would
> appreciate some input on.
>
> The first relates to the attached picture file
> (split_dataelement.png).  A naive implementation of this form section,
> which I suspect might also be the most common, is to make the contents
> of the Diagnosis column (Mental Problems, Epilepsy, Diabetes etc) as
> the datelements.  And to create two categories (a Sex/Gender category
> and another category for new cases, old cases, referrals and deaths).
> The temptation to do this is of course very strong, because we can
> then use the automatic form layout facility to present an html form
> which looks like the paper form.  But to me at least, this seems
> clearly wrong ... in fact something of an evil.
>
> New and old cases have nothing to do with referrals and deaths (apples
> and oranges).  What is actually underlying a more nuanced reading of
> this form are two datelements for each of the diagnosis column eg.
> Epilepsy Cases and Epilepsy Outcomes.  I've tried to illustrate this
> by boxing them off in the table.  If I am reading this right then the
> use of auto form layout is positively driving the implementor to
> create both wrong dataelements and wrong categories.
>
> In fact a deeper look at this (since drawing the picture) seems to
> indicate that a better dataelement breakdown might even be "Chronic
> Disease and Mental Health Cases" and "Chronic Disease and Mental
> Health Outcomes" for which we have an additional category (dimension)
> for the diagnosis which would be a subset of a list of ICD10 codes and
> descriptions.  This breakdown has the combined effects of reducing the
> dataelement count as well as incorporating ICD codelists which could
> be useful for cross-matching against other data later.
>
> So as a consequence of this little bit of analysis, I have recommended
> to the implementors that they *really*, *really*, *really* should
> avoid auto-form layout completely as following the logic of form
> layout necessarily leads them to the evil of poorly constructed
> structural metadata.  The issue is illustrated in the attachment, but
> this pattern repeats itself on various parts of the form.  But I am
> surely not the only one to have observed the anti-pattern of
> structuring dataelements and categories on the basis of what we want
> the dataentry form to look like.  I would welcome some feedback from
> others on this before I lead them down this road.  Do we have any good
> experience from the field elsewhere on this subtle process of teasing
> out the actual dataelements and dimensions from their visual
> representation on forms?   From the many databases I have looked at,
> we certainly do have many examples of the naive, crude approach.

A slightly softened position .. its not auto form layout which is
necessarily evil.  With well structured dataelements, the auto layout
will produce reasonable forms.  But if we want to produce forms like
the attached snippet using auto form layout, we can, but at the cost
of making silly categories.

>
> For the moment the implementors are prepared to put in the effort of
> doing custom form design.  My suggestion has been to put the form
> layout on the back burner initially - this week - and focus on
> identifying correct, or at least sensible, dataelements and
> categories.   As a carrot, I have promised to provide some simple
> tooling to help make the generation of html tables and mapped cells a
> bit easier.  In the short term this won't be in dhis as we are too
> close to release.  I'm not even too sure how I'm going to do this yet,
> but I am prepared to do anything to prevent then doing the
> datastructuring wrong.
>
> Right.  That was conundruum number one.  Conundrum number two is not
> entirely unrelated ....
>
> We are all familiar with our current problem of agegroups, or more
> generally the issue of unique category option codes.  There are the
> usual categories like {"<5",">=5"} and {"under5","5-18",">=19"} etc.
> We have a longer term plan for tying these up with concepts but that
> is not targeted at the upcoming release.  But after 15 cups of coffee
> on Saturday afternoon it occurred to me we could have a reasonable
> workaround.  It seems there is nothing to prevent the implementors
> having a single age group category containing *all* the age groups.
> The main thing one loses by doing this is that is the auto-form layout
> would be a mess (theoretically they can grey out the unused options
> but that would lead to more grey than white on the form).  But in
> light of the discussion above regarding identifying dataelements and
> dimensions, we have decided to avoid the evil of the auto form layout
> anyway.  In which case, the implementor can, indeed he must, select
> the categoryoptioncombos to use individually anyway.  So he can
> constrain the agegroups to collect against in his design of the custom
> html form layout which seems adequate to me.
>
> So a question:  are there any unforeseen consequences I might have
> missed in creating a single age group category and using it like this?
>  I can't off hand think of any which are too serious, but I could be
> wrong.
>
> Final note - a strong motivation for trying to get this structural
> metadata sensible is that in Rwanda, at least in the medium term,
> dhis2 is likely to be acting as the authoritative health information
> metadata repository for the country.  This means that other 3rd party
> systems will be obliged to use metadata codelists provided by dhis to
> mediate exchange between quite a heterogenous mix of systems.  Given
> this requirement we have to somehow avoid inscribing the two dhis
> idiosyncracies described above into the metadata design.  I *think*
> the approach described will effectively work around the problems, but
> welcome some feedback.
>
> Regards
> Bob
>