E-drug: TBnet (00230) Q&A: Improving compliance (cont'd)
-------------------------------------------------------------------------
Several recent e-drug comments have noted the mixed reviews that directly
observed
therapy (DOT) for tuberculosis has gotten. In that regard, several points
are
noteworthy.
First is the common-sense point that bad execution (also known as poor
proficiency)
can make DOT offensive, badly organized, or otherwise off-putting to
patients, thus
making it ineffective -- not because it is a maneuver that cannot work
under any
circumstances, but because it is a maneuver that cannot work unless it is
executed
proficiently. (A witch-doctor's incantations do not work under any
circumstances,
whether or not proficiently executed; penicillin for T. pallidum can work
when
administered proficiently, but not when administered in improficient ways
that do
not provide enough drug in a timely enough manner to permit it to work.)
Second is the technical question of how one evaluates proficiency of
execution of
key interventions. This question is crucial for meta-analysis, e.g., as
done by
the Cochrane Collaboration, and the utility of 'evidence-based medicine'.
In
intention-to-treat analysis, which is widely regarded as the 'gold
standard' policy
for evaluating clinical trials, issues of proficiency of execution are, by
policy,
ignored. That is done because of the fear of bias creeping in when a
post-randomization variable is used as covariate information in analysis.
But
ignoring everyting that happens after randomization also opens the door to
foolishness, e.g., intention-to-treat analysis showing post-operative wound
infections in the drug group in a surgery vs drug trial for coronary
arterial
disease, when some patients randomized to the drug later decided to have
surgery
instead. It is certainly true that patients may cross over for reasons
related to
disease severity, and thus be a biased sample of the whole trial
population, but it
is foolish to pretend that it never happened.
At the Limburg Compliance Symposium, held at the Belgian University of
Limburg in
1995, Prof. Sir David Cox commented on some of the limits of
intention-to-treat
analysis, and the need to take into account in one of the analyses one does
the
proficiency of treatment execution, e.g., reliably measured patient
compliance with
protocol-specified drug regimens. He recalled an agricultural trial in
which he
had been involved in his younger days, in which a newly synthesized
fertilizer was
being tested in fields randomized between the test agent and a control
agent. The
test agent stimulated stunningly luxuriant growth of the crop, so much so
that it
attracted a huge flock of gulls that descended upon the test plots and ate
all the
crop, leaving the control plots alone. The intention-to-treat analysis of
course
concluded that the test agent was less effective than the control agent.
But, as
he pointed out, it was not a useful conclusion to relay, unedited, to the
agricultural chemists. (Unfortunately, Cox did not include the story in
his
discussion of the symposium papers, but one can read his comments, along
with those
of Bradley Efron, plus the papers on various aspects of patient
noncompliance in
drug trials, in vol 17, no. 3 -- Feb 15, 1998-- of Statistics in Medicine.)
Thus, the methods by which DOT is evaluated, and the extent to which they
include
reliable ascertainments of proficiency, are important details in evaluating
DOT,
and in evaluating the evaluations of DOT. A formulaic approach that
ignores
proficiency is not a reliable basis for judgment.
John Urquhart, MD, FRCP(Edin)
Professor of Pharmaco-epidemiology, Maastricht University, Maastricht, NL
Chief Scientist, AARDEX Ltd/APREX Corp, Zug CH & Union City, CA, USA
Professor of Biopharmaceutical Sciences, UCSF, San Francisco
home office: 975 Hamilton Ave, Palo Alto,CA 94301 USA
email: urquhart@ix.netcom.com
--
Send mail for the `E-Drug' conference to `e-drug@usa.healthnet.org'.
Mail administrative requests to `majordomo@usa.healthnet.org'.
For additional assistance, send mail to: `owner-e-drug@usa.healthnet.org'.