Hoping this forum is still active

[Tzippy Shochat]

Hello all.

I hope this forum is still active.

I am a practical medical statistician with a recurring problem.

I am frequently asked to analyze Admission Days (Days from Admission to Release from hospitalization) as a dependent variable.

The problem is that a non-ignorable number of patients, who died, also have short admission days.

I tried convincing the researchers not to use Admission Days as an endpoint.

Is there a good way to analyze this?

Thanks

Tzippy Shochat

[Barry McDonald]

Hi Tzippy,

Yes, interesting. If we think of Admission Days as being a surrogate for “how bad is this case?”  then a really bad case could be represented by a really low number (acute illness, the patient died) or a really high number (took many days/weeks until release).   Perhaps you need to present/ model  figures for deaths and live releases separately.   Or if you have to present/model  them together then redefine the endpoint as live release and set deaths to a very large number (in place of infinity).

Regards, Barry

--
--
To post a new thread to MedStats, send email to MedS...@googlegroups.com .
MedStats' home page is http://groups.google.com/group/MedStats .
Rules: http://groups.google.com/group/MedStats/web/medstats-rules

---
You received this message because you are subscribed to the Google Groups "MedStats" group.
To unsubscribe from this group and stop receiving emails from it, send an email to medstats+u...@googlegroups.com.
To view this discussion on the web, visit https://groups.google.com/d/msgid/medstats/CAK2R97neyJZrChNOY4_73QRj0%3DhqEiiJ_cOATZqt%3DXWJ_H0bvw%40mail.gmail.com.

[Rich Ulrich]

This is parallel to questions I've considered before, though I have

not had to deal with hospital admissions and deaths.

It is useful and powerful to have a single degree of freedom as

an outcome variable.  But it is terrible to have an outcome with

a grievous confounding of two dimensions, like "good outcome"

and "quick death" - which is what you have. I wonder if there are

other confounding outcomes, such as, "transfer to hostel care or

home, to die."

When you have two dimensions to combine to one, it /is/

conceivable to devise a composite variable, or do scoring that

otherwise seems to reflect one underlying latent factor that could

be a range from "good" to "bad."  However, dying quickly might or

might not be a better outcome than dying slowly:  pain?  conscious?

 - I have doubts about /your/ finding a single score that will satisfy

many people, unless there is a very narrow problem at hand.

So: The simplest way to consider the two dimensions when one

features 'death' is to do separate analyses on two samples. 

"Living, with good outcome" provides a sample where the length

of stay sounds like a reasonable ordinal metric, from better to

worse:  I would further consider taking the logarithm, or some other

transformation to achieve something like equal intervals (because,

adding a week to "two days stay" is not an equal  interval to adding a

week to "two months".

You might try parallel analyses on the rest of the cases, but I think

I would ask my physicians to give case examples and opinions about

the other varieties of outcomes, for the cases in your own population.

If 'days' for these people does not scale in your head as a range from

better to worse, then you may be stuck with several categories of

outcome in order to have a fair outcome measure, whether this is for

doing tests in research or giving administrative reports.

Of course, a mathematician who does not do data analyses might

suggest that you dump several variables into a multivariate ANOVA

(Days; death; what-have-you as outcomes) and get confusing coefficients

for one or more equations, right from the start.  I might consider that

as a /summary/ step after I looked at the Dead and Alive separately,

if I had figured out that those results were highly parallel, so that a

summary could make sense.

Hope this helps.

Rich Ulrich

---

From: meds...@googlegroups.com <meds...@googlegroups.com> on behalf of Tzippy Shochat <tz.sh...@gmail.com>
Sent: Thursday, June 23, 2022 12:58 AM
To: meds...@googlegroups.com <meds...@googlegroups.com>

Subject: {MEDSTATS} Hoping this forum is still active

[Tzippy Shochat]

Thanks for the great answers. I found this article and am trying various approaches.

https://bmcmedresmethodol.biomedcentral.com/track/pdf/10.1186/1471-2288-11-144.pdf

Tzippy

‫בתאריך יום ה׳, 23 ביוני 2022 ב-8:58 מאת ‪Rich Ulrich‬‏ <‪rich-...@live.com‬‏>:‬

To view this discussion on the web, visit https://groups.google.com/d/msgid/medstats/BL0PR05MB4900E2EEC1BD9D4F723FFECA97B59%40BL0PR05MB4900.namprd05.prod.outlook.com.

[Prof Sada Nand Dwivedi]

Instead of considering admission days, you have to analyze the data considering time to event, i.e., admission to discharge, approach. The deaths etc will go as censor records, but their consideration is necessary for adjustments in the estimates average admission days.

Thanks

S.N. Dwivedi, Ph.D., FSMS

Professor of Biostatistics

International Centre for Health Research (ICHR)

RD Gardi Medical College

(A Unit of Ujjain Charitable Hospital and Research Centre)

Agar Road, Village Sursa

Ujjain-456006, Madhya Pradesh, India

and

President-Elect(until 2022) and President (2023-2024), Indian Society for Medical Statistics

Founder Vice-President, Society for Evidence Based Health Care India

&

Former Professor, Department of Biostatistics &

Adjunct Faculty, Clinical Epidemiology Unit

All India Institute of Medical Sciences

Ansari Nagar, New Delhi-110029, India           

Tel: 

      91-9810571956 

      91-9868397937

Other Emails:    dwiv...@hotmail.com

                         dwiv...@aiims.edu

                         dwiv...@yahoo.com

[Greg Snow]

The best way to analyze this type of data is with survival analysis (a brief overview is here: https://en.wikipedia.org/wiki/Survival_analysis).  This allows you to take into account the information on those that died as censored observations.  Within survival analysis are techniques for analyzing competing risks (death vs. discharge to another facility vs. discharge to home, etc.) which may give even more insight to your questions.

 

Survival analysis is often a full semester graduate level course, so if you are not already familiar with it, you should probably involve a statistician that is familiar with it to help you (or take the class).  This is not something that most people can spend an hour reading up on and then know enough to expand from regular regression models (you can learn the commands in the software in that time, but really understanding what you are doing, the assumptions being made, the possible pitfalls, etc. takes some real effort).

 

 

From: meds...@googlegroups.com <meds...@googlegroups.com> On Behalf Of Tzippy Shochat

Sent: Wednesday, June 22, 2022 10:58 PM
To: meds...@googlegroups.com
Subject: {MEDSTATS} Hoping this forum is still active

 

External Sender: Be aware! Read with care!

 

--

--
To post a new thread to MedStats, send email to MedS...@googlegroups.com .
MedStats' home page is http://groups.google.com/group/MedStats .
Rules: http://groups.google.com/group/MedStats/web/medstats-rules

---
You received this message because you are subscribed to the Google Groups "MedStats" group.
To unsubscribe from this group and stop receiving emails from it, send an email to medstats+u...@googlegroups.com.
To view this discussion on the web, visit https://groups.google.com/d/msgid/medstats/CAK2R97neyJZrChNOY4_73QRj0%3DhqEiiJ_cOATZqt%3DXWJ_H0bvw%40mail.gmail.com.

NOTICE: This e-mail is for the sole use of the intended recipient and may contain confidential and privileged information. If you are not the intended recipient, you are prohibited from reviewing, using, disclosing or distributing this e-mail or its contents. If you have received this e-mail in error, please contact the sender by reply e-mail and destroy all copies of this e-mail and its contents.

[John Whittington]

Tzippy,

You've had some interesting responses, most of which I would fully agree with.

However, one thing which doesn't seem to have been said is that the appropriate response presumably depends crucially on what one is actually interested in looking at.

In particular, if the question relates to 'resources' (bed occupancy, costs, adequacy of number of beds and supporting them with  appropriate staff etc.), then the 'raw' number of 'admission days' (including those who died) presumably IS what one is interested in - since it doesn't matter whether an individual's period of occupancy of a bed comes to an end because they recovered and went home', were transferred elsewhere or died.

As many have said or implied, the number of 'admission days' (including those who died) itself clearly clearly cannot be used as an index of such things as disease severity, success of treatment or even 'quality of care', since a short hospital stay can mean (amongst other things) 'rapid recovery' or 'rapid death' !

Kind Regards,
John

At 05:58 23/06/2022, Tzippy Shochat wrote:

Hello all.

I hope this forum is still active.

I am a practical medical statistician with a recurring problem.

I am frequently asked to analyze Admission Days (Days from Admission to Release from hospitalization) as a dependent variable.

The problem is that a non-ignorable number of patients, who died, also have short admission days.

I tried convincing the researchers not to use Admission Days as an endpoint.

Is there a good way to analyze this?

Thanks

Tzippy Shochat

--
--
To post a new thread to MedStats, send email to MedS...@googlegroups.com .
MedStats' home page is http://groups.google.com/group/MedStats .
Rules: http://groups.google.com/group/MedStats/web/medstats-rules

---
You received this message because you are subscribed to the Google Groups "MedStats" group.
To unsubscribe from this group and stop receiving emails from it, send an email to medstats+u...@googlegroups.com.
To view this discussion on the web, visit https://groups.google.com/d/msgid/medstats/CAK2R97neyJZrChNOY4_73QRj0%3DhqEiiJ_cOATZqt%3DXWJ_H0bvw%40mail.gmail.com .

John

----------------------------------------------------------------

Dr John Whittington,       Voice:    +44 (0) 1296 730225

Mediscience Services       Fax:      +44 (0) 1296 738893

Twyford Manor, Twyford,    E-mail:   Joh...@mediscience.co.uk

Buckingham  MK18 4EL, UK            

----------------------------------------------------------------