Decision making, evidence, audit, and education: case study of antibiotic prescribing in general practiceCommentary: What can we learn from narratives of implementing evidence?

Editor,

We were interested in the study by Lipman and Price into the prescribing
of trimethoprim for urinary tract infections (1). We have recently carried
out a study using the General Practice Research Database (GPRD),
investigating resistance associated with various antibiotics used to treat
young women with a diagnosis of a urinary tract infection.

The GPRD
contains the anonymised patient records of over 4 million people resident
throughout the United Kingdom and has been described elsewhere (2). Our
study included all women aged 15-44 years with a record of UTI or
cystitis between 1992 and 1998 and an accompanying prescription for an
antibiotic. Patients were selected to have at least six months data prior
to the diagnosis and no record of cystitis/UTI during this period. Only
the first infection in each patient in the study period was included. All
patients with a diagnosis plus a prescription were followed for 28 days
and the need for any further prescriptions of an antibiotic were taken as
a failure of treatment. Patients with a first infection and a prescription
for trimethoprim were analysed based on whether 3, 5 or 7 days of
treatment were prescribed. A Cox regression analysis was carried out,
adjusting for age, comparing the treatment durations and the results are
presented in Table 1.

There were 76,629 first prescriptions of
trimethoprim associated with a diagnosis of UTI in the study period where
the duration of the prescription or the number of tablets of 200mg
trimethoprim were recorded. Only 8.2% of prescriptions were for 3 days,
60.0% were for 5 days and 31.8% were for 7 days. The percentage of all
trimethoprim prescriptions written in any one year for 3 days increased
from 6.3% in 1992 to 14.6% in 1998. As can be seen from table 1, there was
a 1.4% difference in the percentage of patients being prescribed a second
course of antibiotics within 14 days of the initial prescription i.e. a
NNT of 71 patients. However, if an adjustment is made for the length of
the prescription there is no significant difference between relapse rates.

This study was retrospective and patients were not randomised between
treatments. It is possible that the duration of treatment reflected
symptom severity rather than the prescribing preferences of individual
GPs. Nevertheless, we would suggest 3 days of treatment with trimethoprim
to be as effective as 5 or 7 days of treatment and we would support this
as the preferred option for uncomplicated UTIs in young women.

 Table 1. Percentage of treatment failures (+95% confidence intervals)
in women treated for urinary tract infections with 3, 5 and 7 days of 
200mg trimethoprim.

____________________________________________________________
                     3 days	    5 days	     7 days
____________________________________________________________
Percentage failure   
within 14 days of
start of initial
prescription	    11.6             10.5            10.2
                  (10.8,12.4)    (10.3,10.8)      (9.9,10.6)
Percentage failure
within 14 days of
end of initial
prescription	    12.8            12.4            12.9
                 (12.0,13.7)	 (12.1,12.7)     (12.5,13.4)
____________________________________________________________	

References.

1) Lipman T, Price D. Decision making, evidence, audit and education: case
study of antibiotic prescribing in general practice. BMJ 2000; 320: 1114-8

2) Lawrenson RA, Williams T, Farmer RD. Clinical information for
research; the use of general practice databases. Journal of Public Health
1999; 21(3): 299-304

Ross Lawrenson

Professor of Primary Health Care

John Logie

Research Fellow

Postgraduate Medical School,
University of Surrey,
Stirling Road,
Guildford,
Surrey
GU2 7DJ

Sir

Drs Lipman and Price provide an interesting account of how difficult
life can be when you start applying modern standards of analysis to the
scientific data used to support elderly medicinal products.

There is, however, one aspect of this conflict they have overlooked.
Trimethoprim, licensed as it was at a time of very differing standards in
clinical and financial management, has a recommended dose schedule that
does not include a 3-day treatment regimen.

Health care professionals are encouraged to look at the Summary of
Product Characteristics for a product before prescribing it (in a previous
incarnation the SPC was known as the Data Sheet). If they do this for
trimethoprim they will see it is recommended that "treatment should
continue for at least one week..."* or similar.

In adopting a practice protocol recommending a 3-day course of
treatment for uncomplicated UTI, the authors are suggesting their
colleagues use a medicinal product outside the terms of its product
licence. Legally, the manufacturer's liability is limited and the
prescriber is using the product on their own responsibility. The
situation is similar to using an unlicensed product on "named patient" or
"compassionate supply" basis.

I am not suggesting that the authors' partnership is not encouraging
best practice. The problem is that the British legal system may not see
it the same way. A treatment protocol with a seven-day course of
treatment might prove cheaper in the long term.

This is not the only instance of a medicine widely used in a manner
not covered by its product licence. There are no sound commercial reasons
why manufacturers should go to the trouble and expense of changing their
licences. These are mostly patent-expired products with small sales and
even smaller profits.

As the authors discovered the supporting data is not robust. A
company trying to change its trimethoprim licence to include 3-day
treatment for simple UTI is unlikely to succeed anyway.

Yours

Jay Herbert

* Monotrim Summary of Product Characteristics 1997: Solvay Healthcare
Limited.

David Syme's point that we can use computerised formularies to ensure
consistency of prescribing is well made. Since our paper was written we
have had PRODIGY installed in EMIS and (provided we first code "urinary
tract infection") are given a first choice of 3 days of trimethoprim.
However our original question was not only to do with consistency, it was
about clinical effectiveness, and we asked the question: do 3 day courses
of trimethoprim in women with uncomplicated lower urinary tract infection
lead to higher second consultation rates? That is why our first step was
to critically appraise the evidence and to search for more when we found
it flawed. Of course we discovered that no evidence exists that can
reliably answer our question so, given that we still had to make a
decision one way or the other, audited prescribing, investigations and
outcomes as described in the paper in order to give us more information to
help us decide whether or not to continue with 3 day courses.

David Taylor draws a different conclusion than ours - that because
there was a (non-significant) trend towards higher second consultation
rates in patients given a 3 day course, he would change back to 5 days. As
he points out, the confidence intervals were wide so, had this been a
randomised controlled trial, a better interpretation would have been that
the study was underpowered and that no valid conclusion could be drawn.

However this was not a randomised controlled trial. It was a
retrospective analysis of routinely recorded data together with a
narrative of the practice's decision making and implementation processes.
We relied upon our memory and our knowledge of how the practice functions
in writing the narrative. Our intention was to present a realistic picture
of our experiences and describe the context in which we made our
decisions. This has the advantage that the picture which emerged may be
more "true to life" than in a formal study and, in particular, avoids the
"Hawthorne effect" in which the process of being observed changes the
behaviour of the participants. But it has disadvantages too, in that we
could not retrospectively investigate the questions that it raised.

For example we formed the impression that some doctors (and perhaps
patients) did not fully trust the 3 day regime. Why for example did 10 out
of 49 patients in the 3 day group whose urine culture grew trimethoprim
sensitive organisms return for a second visit, at which only one was
prescribed a further course of antibiotics? Why was there such a highly
significant difference in second consultation rates among patients who had
urine culture performed at the first consultation compared to those who
did not (15% compared to 4%)? This mistrust and the behaviour of the GPs
("We'll send off a urine and book you an appointment for next week")
rather than a difference in effectiveness may explain the differences in
second consultation rates.

Narratives and other descriptive studies can enumerate measurable
outcomes, give us new information which may help to inform our decisions,
and identify areas for further research. They are not substitutes for
properly conducted randomised controlled trials, nor for appropriate
qualitative studies, and are biased by the viewpoint of the authors.
However they illuminate the context of everyday practice in a way that
more formal methods cannot, because of the fact that such studies
inevitably change the practice they are trying to investigate. I would be
interested to hear any views on the methodological issues raised by
narrative studies, in particular how they can be critically appraised for
validity and applicability.

Toby Lipman