Presentation, pattern, and natural course of severe symptoms, and role of antibiotics and antibiotic resistance among patients presenting with suspected uncomplicated urinary tract infection in primary care: observational studyBMJ 2010; 340 doi: https://doi.org/10.1136/bmj.b5633 (Published 05 February 2010) Cite this as: BMJ 2010;340:b5633
All rapid responses
With great interest we read the article of Little et al..(1) We agree
with the authors that learning about the extent to which patient
characteristics modify the effectiveness of antibiotics is very relevant.
However, although this issue is put forward twice in the Introduction
section of the paper, the analysis does not appear to have addressed it.
We are currently performing the Amsterdam Cystitis / Urinary Tract
Infection Study (ACUTIS)(2), in which we investigate which combinations of
diagnostic variables best help to diagnose uncomplicated UTI. ACUTIS is
similar to the diagnostic study within which Little et al.’s study was
nested.(3) In addition, we aim to identify patients who are most likely to
benefit from antibiotic treatment, based on their clinical profiles by
studying treatment-profile interaction.
Little et al.’s description of their modelling strategy raises some
questions. Was the aim to learn about underlying pathophysiologic
mechanisms and to describe factors that independently (and perhaps
causally) have an effect on the course of UTI? Or were the authors more
interested in their direct clinical application and did they intend to
present the selection of an efficient set of variables to predict the
course of UTI irrespective of causality? These questions require different
modelling strategies.(4) Apart from these issues, the variable selection
strategy seems to be unusual and calls for further elucidation. In the
description of their analysis, Little et al write “we entered variables
significant in univariate analysis (…) into multivariate analyses and
retained them if they were significant; all the univariate variables were
then tested in the model and any further significant variables retained”.
Were the univariate variables tested simultaneously or one by one? Were
the variables that were insignificant in the first multivariate step re-
entered at later stages? It is known that variables may show different
confounding patterns depending on which other variables are in a model. In
addition, the analysis might have gained transparency by presenting
evidence on the normality of the symptom severity data (measured on 7
point scales). Finally, we note that only 511/843 patients could be fully
analysed. We wonder if the analyses presented under the Response bias
heading were all univariate and whether the authors considered using
weighting techniques to account for any (multivariate) selective loss to
follow-up.(5,6) Could it, for example, be possible that young women with a
good response to treatment were less likely to return their diaries?
In much primary-care based research, investigators tend to adopt an
approach that closely resembles daily practice when designing their study.
This may lead to the incorporation of a physician-based routine that still
lacks a sound basis in medical evidence. Therefore, in contrast to Little
et al., our study (ACUTIS) uses the patient-reported inclusion criterion
‘painful and/or frequent micturition’, rather than relying on the
criterion of ‘suspected urinary tract infection’, which is open to
physicians’ interpretations and opens the door to arbitrary selection of
patients. The more ambiguous the definition of a study’s clinical domain,
the more difficult the applicability of the results will be. A final
example of meticulousness (deliberately avoiding resemblance to daily
practice) serving study validity, and therefore its applicability, is that
in ACUTIS we assure accuracy of urine cultures by keeping urine samples
refrigerated during transport until being cultured, whereas Little et al.
chose to transport them as in routine practice thereby risking
overestimation of UTI prevalence.
(1) Little P, Merriman R, Turner S, Rumsby K, Warner G, Lowes JA et
al. Presentation, pattern, and natural course of severe symptoms, and role
of antibiotics and antibiotic resistance among patients presenting with
suspected uncomplicated urinary tract infection in primary care:
observational study. BMJ 2010; 340:b5633.
(2) Knottnerus BJ, Bindels PJ, Geerlings SE, Moll van Charante EP,
ter Riet G. Optimizing the diagnostic work-up of acute uncomplicated
urinary tract infections. BMC Fam Pract 2008; 9:64.
(3) Little P, Turner S, Rumsby K, Warner G, Moore M, Lowes JA et al.
Dipsticks and diagnostic algorithms in urinary tract infection:
development and validation, randomised trial, economic analysis,
observational cohort and qualitative study. Health Technol Assess 2009;
(4) Clayton D, Hills M. Statistical models in epidemiology. 1998 ed.
New York: Oxford University Press; 1993.
(5) Alonso A, Segui-Gomez M, de Irala J, Sanchez-Villegas A, Beunza
JJ, Martinez-Gonzalez MA. Predictors of follow-up and assessment of
selection bias from dropouts using inverse probability weighting in a
cohort of university graduates. Eur J Epidemiol 2006; 21(5):351-358.
(6) Hernan MA, Hernandez-Diaz S, Robins JM. A structural approach to
selection bias. Epidemiology 2004; 15(5):615-625.
Competing interests: No competing interests