Best practice to capture a blank/missing obs_datetime

[Stephen Po-Chedley]

Hi Everyone,

I have some observations that do not necesarily have a date (obs_datetime). Specifically, the observation is a CD4 Count that is recorded in a patient file (on their initiation of ART therapy). Often times, the date recorded in the patient file is missing. 

OpenMRS defaults this value to the encounter_datetime. This isn't exactly right - we know the value couldn't have been after this encounter_datetime, but the encounter_datetime isn't actually correct. To complicate matters, we use this concept with other encounter types (e.g. lab encounter) in which the obs_datetime is accurate. So we trust the obs_datetime associated with the CD4 Count concept with some encounter types, but not others. 

We currently record the date using a separate CD4 date concept (for ART Initiation), though this requires a fair amount of logic to check to pair these two observations (the numeric CD4 count and the date value of that CD4 count) for specific encounter types (ART Initiation, but not lab encounters). 

I don't think the obs_datetime for the CD4 count can be left blank, which is the most straightforward/accurate way to represent this situation. I'm wondering if anyone else has experience this problem or how to best deal with it? 

Thanks,

Stephen

[Burke Mamlin]

It wouldn't really make sense to record a CD4 by itself without any context.  Perhaps the problem is that you are using the "CD4 Count" concept to record "CD4 Count at Initiation of ART".

You could create a "CD4 Count at Initiation of ART" concept.  If you make another flavor of CD4 Count (i.e., numeric with same range), then you could store you CD4 Count there without a date.  Well, there's a date, but it clearly becomes the date that the "CD4 Count at Initiation of ART" was recorded instead of the implied date at which the patient's CD4 Count was drawn.  If you have a "Date of ART Initiation" or "Date of ART Initiation CD4 Count", then you can record these separately or place them into an obs group.

Cheers,

-Burke

--
OpenMRS Implementers: http://go.openmrs.org/implementers
Post: implem...@openmrs.org
Unsubscribe: implementers...@openmrs.org
Manage your OpenMRS subscriptions at https://id.openmrs.org/

[Hannan, Terry J (DHHS)]

Burke, this discussion raises an issue for myself even that I have thought about for years and we did not solve well in my original work on the Hopkins system. Though I am not actually implementing anywhere at the moment. As Stephen states sometimes there is no available date or accurate date for a procedure e.g. appendicectomy/biopsy/chest X Ray.

This is because the patient often has no prior documentation and cannot recall the exact dates.

Do you have a mechanism for documenting (date) this in OpenMRS?

Terry

 

Dr Terry J. Hannan MBBS;FRACP;FACHI;FACMI
Consultant Physician
Clinical Associate Professor  School of Human Health Sciences, University of Tasmania Department of Medicine, Launceston General Hospital
Charles Street Launceston 7250

Moderator: http://www.ghdonline.org/

Ph. 61 3 6348 7578
Mob. 0417 144 881
Fax 61 3 6348 7577
Email terry....@dhhs.tas.gov.au

Skype: thehannans

To unsubscribe from this group and stop receiving emails from it, send an email to implementers...@openmrs.org.

---

CONFIDENTIALITY NOTICE AND DISCLAIMER
The information in this transmission may be confidential and/or protected by legal professional privilege, and is intended only for the person or persons to whom it is addressed. If you are not such a person, you are warned that any disclosure, copying or dissemination of the information is unauthorised. If you have received the transmission in error, please immediately contact this office by telephone, fax or email, to inform us of the error and to enable arrangements to be made for the destruction of the transmission, or its return at our cost. No liability is accepted for any unauthorised use of the information contained in this transmission.

[Burke Mamlin]

Observations are all timestamped.  If you observe something, but don't know when you observed it, then you can (1) estimate the date, (2) interpret that particular observation's timestamp as date collected instead of physiologic date, or (3) model your data differently.  For example, we'd suggest modeling historic data differently than prospective data – i.e., "History of Surgery = appendectomy" and optional "Date of Surgery" in an obs.

Cheers,

-Burke

[Hannan, Terry J (DHHS)]

Thank you it has given me strong ideas for my current rudimentary summary record

Sent from my iPhone Terry Hannan

[Michael Seaton]

Thanks Burke,

We can definitely use separate concepts here (and this is what we have done in the past) if that is the only solution.  One issue with this is that it introduces more complexity into the data model and requires others to know that CD4 Count-based calculations need to look at multiple concepts to get the full picture.  Simply graphing the CD4 count concept over time will miss out on this value, if we record it under the "CD4 count on ART initiation date" concept.  Naive report authors are also likely to miss this out without good documentation.

Ultimately, this is still a CD4 count, it is just that the date is only known to a certain level of precision.  We know that it is the most recent value as of date X.  And it isn't without context - it is contained within a particular encounter that has a particular encounter date.  Is this so different from Obs exceptions, just with the date of the Obs rather than the value of the Obs?  Didn't we have plans to introduce Partial or Estimated Dates more broadly in the future?

I know neither of these is available currently, but just wondering if your response is based on the limitations of our current data model, or would be considered the best way to do things regardless.

Also, what if the CD4 date just isn't known or is illegible?  What is the best practice here?

Thanks,
Mike

-----Original Message-----
From: Burke Mamlin <bu...@openmrs.org>
Reply-to: <implem...@openmrs.org>
To: implem...@openmrs.org <implem...@openmrs.org>
Subject: Re: Best practice to capture a blank/missing obs_datetime

[Andrew Kanter]

As long as there is an actual value, then I think this makes sense. What was being discussed in the past was what to do when you want to document no value (missing), and the presence of an actual record with a null value allowed users to bypass validation rules which required an actual value to be entered. Seems like we are on OK ground here, however.

Andy

 

--------------------
Andrew S. Kanter, MD MPH FACMI

Asst. Prof. of Clinical Biomedical Informatics and Clinical Epidemiology

Columbia University
Email: andrew...@dbmi.columbia.edu
Mobile: +1 (646) 469-2421
Office: +1 (212) 305-4842
Skype: akanter-ippnw
Yahoo: andy_kanter

[Burke Mamlin]

How do you graph something over time when it doesn't have a known timestamp?  Using a separate concept would allow the CD4 Count to be included in a graph, but only by a feature that understands how to put a "CD4 Count at Initiation of ART" correctly on the graph.

Introducing estimated dates for observations would still require that you pick a date, assuming you want to continue using any existing code.  Otherwise, all code touching observations would need to be refactored to no longer assume that observations have a precise date and we would need a predictable behavior in handling all of the downstream side effects – e.g., returning the most recent CD4 Count given data like: 200 (March 2014), 500 (3-March-2014).  It's not impossible, but I would guess that the complexity of broadly supporting estimated observation dates would be far greater than the complexity of workarounds for the few cases where observations don't have a known (or acceptable proxy for) "effective" time.

-Burke

[Andrew Kanter]

Has this not come up before? We typically document a full date and then set the estimated? flag. I have seen UIs where there is a mask for estimated dates, but how this gets started in the DB is an issue. Setting 1/1/xx or x/15/xx or 1/xx/xx creates distortions of the statistics. Randomly assigning the month and year for unknowns helps with that, but it is a real pain when trying to do calculations. See early discussion about Date of Birth :)

Andy

 

--------------------
Andrew S. Kanter, MD MPH FACMI

Asst. Prof. of Clinical Biomedical Informatics and Clinical Epidemiology

Columbia University
Email: andrew...@dbmi.columbia.edu
Mobile: +1 (646) 469-2421
Office: +1 (212) 305-4842
Skype: akanter-ippnw
Yahoo: andy_kanter