Mimic-code: meaning of startdate/enddate in table prescriptions

Yes, that's the right interpretation, In your example metoprolol was prescribed for those four days - which means it was very likely administered to the patient (but not 100% as it's a prescription record, not an administration record).

I see, thanks a lot for your response!

Interesting question. What about this example below:

ROW_ID | HADM_ID | STARTDATE | ENDDATE | DRUG | DOSE & UNIT
------ | ------- | --------- | ------- | ---- | -----------
232390 | 100765 | 2126-03-12 | 2126-03-13 | Meropenem | 500 mg
232385 | 100765 | 2126-03-12 | 2126-03-13 | Meropenem | 500 mg
232391 | 100765 | 2126-03-12 | 2126-03-18 | Meropenem | 500 mg
232543 | 100765 | 2126-03-18 | 2126-03-19 | Meropenem | 500 mg

This example means that the patient takes three doses of Meropenem 500 mg between 12th March and 13th March? And after 14th March he takes only one dose?

Thanks!

Not necessarily, and this is a limitation of the medication table. You know that these medications were ordered with those start/end dates, but you don't know how many doses were administered. You also don't know the frequency - one or two of those orders could have been STAT (immediately), followed by a q12hr or something like that.

Hum... trick question.
There is any way to infer the ordered frequency of the medication based on prescription table?

For non-intravenous medications there's no easy answer. For IV medications you can look for the administration in the inputevents_mv or inputevents_cv tables.

We do hope to release more data in the future which covers these gaps.

Great! Thanks @alistairewj!
We are trying to replicate this paper experiments with MIMIC. I'll let you know when we finish.

Henrique,

Nice points here! Trying to figure out how you are doing inference of
proper dosage with planned data from prescriptions, when it is clear in
documentation/and paper that actually you do not know if that prescription
was administered or not to the patient (or how/when it was ). Without using
confirmation check-points (as far as I know they are not available in Mimic
) you are using treatment plans as labels to target your prediction of
proper usage, and this may sound inappropriate.

To do something pretty similar in our hospital EHR we are analyzing
outliers only for IV antibiotics , adjusting information by weight / renal
function (eGFR), using stepwise regressions considering timelines, and so
on ( here, you cannot avoid to look into the big tables on mimic) to follow
this approach!

Mimic was wonderful to help us understand how to model our dataset to
comeout with this kind of analyzes, but not as a second database for
reproduction purposes (perhaps in the future). Now our model is a reality
here as a nudge implemented in our homebrew EHR thanks to what we have
learned from mimic!

Best regards,
Tiago / your neighbor

Em sex, 2 de nov de 2018 às 10:44, Henrique Dias notifications@github.com
escreveu:

Great! Thanks @alistairewj https://github.com/alistairewj!
We are trying to replicate this paper
https://ieeexplore.ieee.org/document/8340108 experiments with MIMIC.
I'll let you know when we finish.

—
You are receiving this because you are subscribed to this thread.
Reply to this email directly, view it on GitHub
https://github.com/MIT-LCP/mimic-code/issues/384#issuecomment-435384437,
or mute the thread
https://github.com/notifications/unsubscribe-auth/AAt7Dd5JOdOiyXxEfnXR6cvbwkQo1j_Zks5urExDgaJpZM4RvSuK
.

>

Obrigado,
Tiago Vaz