Possible improvements to GestationalAge

[bpow]

1. Age is represented using an ISO 8601 duration, but GestationalAge has key/value pairs for weeks & days.

   It would be more consistent to represent them both the same-- that is, GestationalAge of 33 weeks and 5 days could have iso8601duration of 'P33W5D', avoiding clients from having to deal with these ages as different formats.

2. It may also be important in some contexts to record the method by which gestational age was assessed (e.g., by LMP, ultrasound assessment, date of IVF embryo implantation).

[pnrobinson]

@bpow thanks! This is open for discussion! But the original motivation was that it would be better to avoid confusion of whether say, P33W5D, is postnatal or prenatal. By using a separate data structure, the chances of confusion are greatly reduced.
Whatever the decision on this, I would also support adding a way to record the way GA is assessed, as you suggest.

[bpow]

I'm sensitive to the potential concern about confusion between age and gestational age, but to me the same potential exists anytime there is a measurement with units and client code should be expected to read and interpret measurements according to values provided (either explicitly as a separate units field or implicitly because of the attribute that's being filled out).

If downstream code is just greping through the file for things that look like ISO 8601 durations and interpreting them all as post-birth ages (not recognizing that some are in Age objects and some in GestationalAge objects)), then that downstream code is broken! IMO there's only so much handholding a data model can do.

The alternative, that we have now, is that downstream code needs to have at least 3 different code paths for representing time elements-- Age as an ISO 8601 duration, Timestamp as an ISO 8601 date/time string, and GestationalAge as a weeks/days pair. Two of those are ISO 8601 standards at least...

[ielis]

... If downstream code is just greping through the file for things that look like ISO 8601 durations and interpreting them all as post-birth ages (not recognizing that some are in Age objects and some in GestationalAge objects)), then that downstream code is broken!

Greping through phenopackets sounds like a really bad idea™ to me as well, so it is good to hear that not being designed for greping is not a serious design flaw :) (... IMO there's only so much handholding a data model can do.).

I believe that the opposite is actually true - we should appreciate the type system protections that force us to handle the GestationalAge and Age in different code paths.

---

... but to me the same potential exists anytime there is a measurement with units and client code should be expected to read and interpret measurements according to values provided ...

In case of Age vs. GestationalAge, is impossible to ignore the unit since the values (ISO8601 duration or weeks & days) are inside the types. The only way to access the values is by using switch/match/if-else on the type and then access the value. This minimizes the risk of ignoring the units and misinterpretation of the measurements. I am not sure I can see how using ISO8601 for both GestationalAge and Age could make this better.. 😕

[bpow]

It doesn't seem like I'm getting much traction with this, maybe it would be worth discussing as part of the GA4GH revision process to see if there are others who are current or potential downstream users who, like me, see this as a bit of a "code smell" ("data structure smell"?).

I'll try one more time to see if I can better explain where I'm coming from.

GestationalAge is used as part of the disjunction of TimeElement. All of the other possible elements in this disjunction are either a categorical ontology-derived term (in ontology_class) or something incorporating ISO 8601 (either a timestamp or a duration). If client/application code is centralizing all of it's age handling/storage/validation around ISO8601, then we have to convert that into separate format just for the GestationalAge part.

In terms of potential confusion between gestational and postnatal age, there's already the distinction between

Postnatal:

timeElement:
  age:
    iso8601duration: "P22W5D"

(suggested) Gestational:

timeElement:
  gestational_age:
    iso8601duration: "P22W5D"

Could downstream code confuse those two? Maybe-- but the fact that the "key" used to reference them is distinct should be enough to mitigate that.