Stream: Orders and Observation WG
Topic: ReferenceRange value set - type/meaning
Rob Hausam (Dec 13 2018 at 11:33):
Moving this discussion https://chat.fhir.org/#narrow/stream/48-terminology/subject/ReferenceRange.20value.20set.20-.20type.2Fmeaning to the OO stream. I agree generally with the previous comments about needing something for 'critical' and this being too tight. In the current build "normal" has the definition "Within two standard deviations from the mean and accounting for 95.45% of the relevant control population." - that's even tighter than before and 95.45% just seems wrong. @Eric Haas?
Eric Haas (Dec 13 2018 at 13:09):
Propose changes to normal is to say ‘is usually defined as’
Eric Haas (Dec 13 2018 at 16:03):
actually there are several issues here.
- The committee voted to loosen the binding to 'preferred' ( which is 'example ' my book ) so you can use what you want as an implementer. And so we could just do nothing.
- Any change to the definition of normal would be breaking- I don't know how to deal with that.
- What is the definition of normal then. "not abnormal"? "determined by the producer of the results. typically based on a statistical value like within two standard deviations from the mean..."
- Adding a non-critical value is non-breaking and target for R5. as long as there is a decent definition provided for what it means.
Rob Hausam (Dec 13 2018 at 18:51):
I hadn't picked up that the definition of "normal" changed from "Based on 95th percentile for the relevant control population." in the May 2018 R4 ballot to "Within two standard deviations from the mean and accounting for 95.45% of the relevant control population." in the September ballot. I don't know why that changed, and more importantly, I'm questioning if 95.45% is correct (in any sense that I know of). For one thing, it seems far too precise to be able to be a correct value. If it's not the correct and intended value, then it would be good if we can fix it now as a technical correction and get it in for R4 (if that's possible).
Rob Hausam (Dec 13 2018 at 19:21):
OK, 95.45% isn't wrong, as it's the more precise value for the population within 2 standard deviations of the mean.
Rob Hausam (Dec 13 2018 at 19:26):
So, as Eric said, if we want to add a 'critical' range or change the definition of 'normal' or anything else we would deal with that in R5.
Grahame Grieve (Dec 13 2018 at 19:58):
procedurally, this value set is not a required or extensible binding, so it's not normative
Grahame Grieve (Dec 13 2018 at 19:59):
breaking change, or change right now, is procedurally ok. And I think that the definition of normal is commonly done based in confidence intervals, but this is not proscriptively true, so the definition could be fixed by the addition of one or 2 qualifying words
Eric Haas (Dec 13 2018 at 21:00):
the change between may and sept was based on a ballot comment.
"Values expected for a member of the relevant control population being measured. For example a healthy person. Typically each results producer such as a laboratory has specific normal ranges and they are usually defined as within two standard deviations from the mean and account for 95.45% of the relevant control population."
Eric Haas (Dec 13 2018 at 21:01):
how about that definition and can we preapply and vote next week ( if we have a call?)
Grahame Grieve (Dec 13 2018 at 21:07):
that's a better definition, sure. Though
For example a healthy person
doesn't seem to add value
Eric Haas (Dec 13 2018 at 22:57):
"Values expected for a member of the relevant control population being measured. Typically each results producer such as a laboratory has specific normal ranges and they are usually defined as within two standard deviations from the mean and account for 95.45% of this population."
Eric Haas (Dec 13 2018 at 22:57):
what steps do we need to take to update now. Get tentative approval from OO chairs?
Eric Haas (Dec 13 2018 at 23:16):
This a nonsubstantive change, but is not a technical correction right?
Grahame Grieve (Dec 13 2018 at 23:54):
yes
Eric Haas (Dec 14 2018 at 00:31):
do I have your permission to preapply this now?
Grahame Grieve (Dec 14 2018 at 00:38):
I'd be happier if I'd seen some co-chair endorsement
Grahame Grieve (Dec 14 2018 at 00:39):
but you can pre-apply and make a PR. I can approve it once we see co-chair endorsement
Rob Hausam (Dec 14 2018 at 03:55):
You have my endorsement.
Grahame Grieve (Dec 14 2018 at 03:59):
is that enough? See emails with co-chairs
Rob Hausam (Dec 14 2018 at 04:04):
hard to know what constitutes enough - pre-apply and formally vote later if that's possible
John L Allen (Dec 14 2018 at 08:59):
In regards to the term for critical reference range, which technically is 'non-critical'. Calling a reference range 'non-critical' might be confusing, as some might view it as the normal reference range vs the critical reference range. The synonym for 'critical' in this context is 'panic', and I feel this term might be better. In other words, values inside the range are 'non-critical', values outside are 'critical', but none of the values (inside or outside) are 'panic' by a strict definition of the word. Some may cause one to panic, but the values themselves aren't 'panic'.
In regards to the statistics of the term 'normal' (ie population/distribution based value), I'm not a statistics guy. But I was under the impression that the reference ranges were chosen to accompany 95% of the healthy population (2.5 percentile to 97.5 percentile). The standard deviation of 2.0 in this case is rounded up from the actual standard deviation of around 1.96 that better fits the 95% span. Either using 95% or a SD of 2 seems to me to be totally arbitrary, so I think definition of normal should err on side of being more general. This may 'break' the definition, but I don't think it's a clinically significant break for anyone (esp if previous definition is '95th percentile', which is not very accurate for any circumstance).
Another issue with reference ranges are that they don't account for the variations by pregnancy trimester. They can account for the different reference ranges between people of various super-specific ethnic groups ("Manley Hot Springs" vs "Mentasta Lake"), but this is something I would never use clinically. There are only a handful of race/ethnicity based variations that I would use in practice (eg GFR/creatinine formulas, and maybe LLN for WBC). And trimester based ranges are fairly significant (eg a recent EMR I came across did not base its interpretation flags on trimester specific reference ranges, and a third trimester pregnant women had her serum alk phos value marked as more than 2x the ULN despite it being normal for her subpopulation.)
Eric Haas (Dec 14 2018 at 14:22):
@John L Allen we have deferred discussion of panic or critical range meaning to the next FHIR version and proposed loosening up the description of normal as discussed above to squeeze in right now. it does not say normal has to be within 2 sds just that is one way is is applied.
Eric Haas (Dec 14 2018 at 14:24):
@Rob Hausam @Grahame Grieve I will create the PR if we get one more approval from the co chairs thats a majority and i personally think thats enough to preapply now.
Rob Hausam (Dec 14 2018 at 17:29):
So you did make the PR with the latest wording from the discussion with Hans - I see it on GH pending Grahame's response.
Grahame Grieve (Dec 14 2018 at 20:14):
applied now
Eric Haas (Dec 14 2018 at 22:23):
yes
John L Allen (Dec 15 2018 at 01:39):
And another unsolicited observation/question...is it safe to consider these reference range type values mutually exclusive? Currently, cardinality for 'type' is "0..1". I can think of a few instances in which a reference range should have multiple types. If 'panic' is implemented, then a trough drug level (ie type 'pre') could be paired with either 'recommended' or 'panic'. Or a 'postmenopausal' range could be either 'normal' or 'recommended'.
Eric Haas (Dec 15 2018 at 01:57):
create 2 reference ranges.
Eric Haas (Dec 15 2018 at 01:57):
one for each type
John L Allen (Dec 15 2018 at 04:39):
@Eric Haas But the single reference range is only meaningful when you combine both types. It would be analogous to having type values of 'male', 'female', 'pediatric', 'adult'. A reference range might have types 'adult' and 'female'. Another range might have 'adult' and 'male'. Another 'pediatric' and 'female'.
Eric Haas (Dec 15 2018 at 18:02):
I misunderstand your earlier question. there is only a single type yes. This valueset is not an exhaustive list of possible values, we attempted to provide the most commonly used concepts. If you wish to implement 'panic drug level' then you would define that concept locally and 'extend' the valueset in your implementation. Another option is to represent this range or set of ranges as text which is common way for implementers (at least in the lab space ) to define relatively complex range interpretations. As to making the type 0..* I think this would add a layer of complexity to the interpretation that would outweigh any benefits. For example, what is the difference between a trough level and a recommended trough level ... panic trough and a panic. In addition there is an interpretation field that represents the Panic concept.
John L Allen (Dec 16 2018 at 01:41):
@Eric Haas I agree that it would probably not be worth it in terms of the added complexity. Most panic levels don't seem to apply to troughs anyway I suppose, but rather to random levels. The only real world example I can think of at the moment (with multiple types) is "pregnancy" and "panic" combination used for a certain urine protein level, which is different than the "pregnancy" and "recommended" value; there is no panic level for non-pregnant individuals.
But I do think 'panic' should be added to the official list rather than relegated to an extension, as panic reference ranges are very common. There are panic/critical highs/lows in the interpretation, but the reference ranges should reflect how this interpretation was reached.
Grahame Grieve (Dec 16 2018 at 06:18):
Eric created a task for this; will probably appear in R5 build sometime early next year, and you'll be able to use it in R4 once it's defined
Andrea Pitkus, PhD, MLS(ASCP)CM, CSM (Jan 28 2019 at 17:23):
Joining thread late. Each performing laboratory is required to determine their own reference ranges, including high, low, panic/critical levels per CLIA law (in US) as part of their test system validation. The same result performed in two different laboratories will likely have different reference ranges, panics, etc. I've personally seen the same VALUE flagged/marked differently based on where the test is performed in the same health system too. One was marked hi and the other abnormal due to different reference ranges and flags. Reference ranges cannot be compared across performing laboratories or used with results of other performing laboratories. see pasted image
It is also correct that peak, trough and random test result values will have different reference ranges, so the context is vital to avoid patient harm. Regarding Urine microalbumin levels during pregnancy, they can be an indicator of preeclampsia which can be a clinical emergency impacting the life of the child and/or the mother. Obviously, this does not apply to male patients or females that are not pregnant (and of child bearing age. ) Context is important.
Last updated: Apr 12 2022 at 19:14 UTC
