[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

Re: st: longitudinal data with non-detectable measurements

From   "Joseph Coveney" <[email protected]>
To   "Statalist" <[email protected]>
Subject   Re: st: longitudinal data with non-detectable measurements
Date   Tue, 13 May 2008 22:10:53 +0900

Leny Mathew wrote:

Thank you very much for your suggestions, Joseph and Nick. 'intcens'
looks to be a very promising option. There seems to be quite a bit of
this in environmental and chemistry literature compared to medicine.


My understanding is that clinical laboratory tests are typically set up to
be used such that a result's falling below the limit of quantitation (LOQ)
doesn't matter, so you never really hear much about its being a problem in
medicine.  For example, in many circumstances the clinical laboratory test
is used in the manner of a "limit test" where a value above (or below) some
threshold or limit is considered to indicate health (or disease).  If the
threshold for some analyte has been established as, say, 5 units per
milliliter, and a patient's result is below an LOQ of, say, 0.05 units per
milliliter, then the physician can still make a determination regardless of
how far below the LOQ the true value is.  This holds even for clinical
laboratory tests that have a "normal range" (two threshholds, a test result
between which indicates health and beyond which indicates disease), so that
a right-censored value of "too numerous to count" as well as left-censored
one of "none detected" are still medically interpretable.

I don't know how widely among the basic medical sciences left censoring is
fretted over, but the problem gets attention from pharmacologists and
pharmacokineticists.  As one example, how to handle the inevitable below-LOQ
assay results in a pharmacokinetics study is a perennial topic on the
PharmPK listserve ( ).  You'll see in that list's
archives at least one participant advocating doing away with LOQ altogether,
dismissing it as an unnecessary criterion for acceptance of the data:
instrumental error can be accommodated explicitly as Gaussian even in
fitting models of responses that are nonegative or strictly positive in
theory, so that all data would be used as-reported by the assay apparatus,
including negative values for concentrations of a drug in blood serum that
result when instrumental error becomes salient as drug concentration falls
to low levels.

Joseph Coveney

*   For searches and help try:

© Copyright 1996–2024 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index