Stream: implementers
Topic: Observation Statistics
Grahame Grieve (Dec 09 2016 at 01:34):
I have a question about Observation.$stats - what should happen to the return if there is no matching observations for the patient/codes/duration combination
Grahame Grieve (Dec 09 2016 at 01:35):
I also have a proposed answer: you return an observation with a single component, total count, with a value of 0
Abbie Watson (Dec 09 2016 at 01:39):
value: null would be more accurate, no?
Grahame Grieve (Dec 09 2016 at 01:40):
would it? why? and how would that look in a response?
Grahame Grieve (Dec 09 2016 at 11:45):
so I'm just finishing my first implementation of this off.
Grahame Grieve (Dec 09 2016 at 11:46):
there turns out to be a bit of a challenge for me.. security/consent on the resources... hmmm
Grahame Grieve (Dec 09 2016 at 11:47):
I presume that we can have consensus that if you can't see the data, it can't appear in the analysis
Grahame Grieve (Dec 09 2016 at 12:48):
so my implementation is live. You can check it out with this request:
Grahame Grieve (Dec 09 2016 at 12:49):
oops. This one:
http://fhir3.healthintersections.com.au/open/Observation/$stats?subject=Patient/123&code=29463-7&system=http://loinc.org&duration=100000¶ms=count;minimum;maximum;average;median;sum;std-dev;%2520;%2580;4-lower;4-upper;4-dev;5-1;5-2;5-3;5-4;skew;kurtosis;regression&include=true
Grahame Grieve (Dec 09 2016 at 12:50):
the underlying data is middle quartile WHO weight for 5-10 yr old, so it's not incredibly useful. If you want better data, you can post it yourself
Rob Hausam (Dec 09 2016 at 13:28):
thanks, Grahame - this looks great overall
but the documentation (current build) seems inconsistent and doesn't match entirely with the implementation - e.g. 'params' or 'statistic'?
Grahame Grieve (Dec 09 2016 at 19:06):
yes that's on my notes to fix after doing the implementation. which do we want it to be?
Grahame Grieve (Dec 09 2016 at 19:57):
Rob Hausam (Dec 09 2016 at 20:48):
statistic - in my opinion
Rob Hausam (Dec 09 2016 at 21:16):
agree generally with the details in GF#12452
regarding unit variations, I think that probably what you meant to say is "my server, if all the observations have the same units, I'll calculate with the units provided"?
other than that, what you've suggested and done is probably the best that we can get, provided that canonical units are identified and that translation to them is performed - I'm wondering if most implementers will support that?
if they do, that's great - but if not, then the fall back can be if the units are the same then calculate with that, and, if not, then don't calculate the statistics
Grahame Grieve (Dec 09 2016 at 21:19):
most will probably source their own observations, and the question won't arise.
Grahame Grieve (Dec 09 2016 at 21:20):
for those that do haev the question, well, UCUM logic is widely available now. I handle that internally, btw. You give me any UCUM unit, and I'll have the canonical form internally. which I store, for this analysis.
Last updated: Apr 12 2022 at 19:14 UTC
