JAC Advance Access originally published online on July 26, 2006
Journal of Antimicrobial Chemotherapy 2006 58(3):489-491; doi:10.1093/jac/dkl300
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Editorial |
Back to the basics: epidemiology in the Journal of Antimicrobial Chemotherapy
Journal of Antimicrobial Chemotherapy, 11 The Wharf 16 Bridge Street, Birmingham B1 2JS, UK
*Corresponding author. Department of Social Medicine, University of Bristol, Canynge Hall, Whiteladies Road, Bristol BS8 2PR, UK. Tel: +44-117-9287242; Fax: +44-117-9287202; E-mail: abigail.fraser{at}bristol.ac.uk
Abstract
Good epidemiological methods and appropriate statistical analysis are cornerstones of any valid, reliable and publishable epidemiological study. This editorial summarizes epidemiological studies published in the Journal of Antimicrobial Chemotherapy (12 issues, May 2005--April 2006) with respect to study objective, type and design. A significant proportion of these studies started off with a single study group. Drawing on this finding, various methodological aspects of choosing a sampling frame and of sampling methods are reviewed.
Keywords: epidemiological studies , sampling frames , sampling methods , methodologies , statistics
Good epidemiological methods and appropriate statistical analysis are the cornerstones of any valid, reliable and publishable clinical study. Studies that were published in 12 issues of the Journal of Antimicrobial Chemotherapy (May 2005–April 2006) and that used mainly epidemiological methods are summarized with respect to study objective, type and design in Table 1. A variety of study designs was used by JAC authors: ecological studies; analysis of multiple panel data to examine trends over time; cross-sectional data; case–control, cohort and record-linkage studies; and clinical trials. However a significant proportion of the studies published in the Journal start off with a single study population, be it HIV patients receiving a given treatment regimen, Escherichia coli isolates from patients with urinary tract infections or general practitioners in a given country.
|
Single group studies can be further divided into descriptive studies and studies that examine associations by utilizing internal comparisons that arise from within the study population. These single group studies should be differentiated from randomized controlled trials, case–control studies or cohort studies which include multiple study groups from the outset as an integral part of their design. The multiple groups studies have their own basic issues that require sound approach and execution and will not be addressed in this editorial.
Although low on the ‘ladder of evidence’, single group studies are important. They describe the world as it is. For example, what are the antimicrobial practices in other medical centres or other countries? Did a certain resistant strain spread to new locations? Are prescribing practices or treatment outcomes associated with patient characteristics? Some single group studies are interesting in that they describe a unique setting, while others are conducted in ‘typical’ settings and therefore their results shed light on more general phenomena.
Therefore, a central issue with regard to single group studies is that of the generalization of results to a larger population or, in other words, their external validity. This is particularly pronounced when the study is conducted in a ‘typical’ setting. To enhance the external validity and improve publishability we would like to emphasize the importance of choosing the correct sampling frame and sampling method and reporting them in detail.
Sampling frame. The sampling frame is a list—real or theoretical—of the population from which a sample is drawn. The underlying assumption of any study is that the sample is representative of the population defined by the sampling frame. This allows results obtained from sample data to be generalized to the larger population. Whether the observation units of a study are hospitals, clinics, patients or microbes, the sampling frame should be clear to readers as it has direct bearing on the generalization of study results.
Sampling methods. Samples may be obtained by random or non-random (such as convenience) sampling procedures. Random sampling procedures ensure that each member of the sampling frame has an equal chance of being included in the sample. Non-random sampling methods are more widely used but reduce our confidence that the sample does indeed represent the population from which it was drawn. Therefore, when non-random sampling methods are used, descriptive data should be provided on relevant aspects of the population and the sample in order to increase the reader's confidence that the sample is truly representative. Results of studies using non-random sampling should be generalized with caution.
Sampling can be a two-stage process. For example, a convenience sample of GP practices in a given region may be sampled in the first stage. The second stage may include random sampling of all registered patients or involve identifying all registered patients with a given condition in a defined time frame. In both cases we would be more concerned about the implications of the first stage of sampling, i.e. the convenience sample of GP practices. We would wonder whether these practices are representative of all GP practices in the region, or in other words whether results could be generalized to the whole region. Providing comparative data on key characteristics of the sampling frame for the first stage (GP practices in region) and the sample, such as GP practice size, setting (urban versus rural) and mean age of patients, would increase our confidence in the study results.
The sampling frame and sampling methods should be explicitly recorded in the Methods section, while implications for the generalization of results should be addressed as part of the Discussion.
Sound research methods are a prerequisite for obtaining valid and unbiased scientific results. Appropriate statistical analysis and clear reporting of research are no less important. One such example is including measures of uncertainty, e.g. confidence intervals for reported estimates, be they estimates of relative risk obtained from comparisons or descriptive in nature such as prevalence. Many of the methods appear in the BMJ publications Statistics at Square One1 and Statistics with Confidence.2
While the issues raised here are some of the basic building blocks of many studies, they are not exhaustive. The use of reliable and validated methods of measurement, diagnosis and microbial characterization is just as important to obtaining unbiased results. Thus the complete gamut of good methodology is required in order to add new and important information to existing knowledge. However, the points raised should be addressed at the start of any research since any biases or problems introduced at the planning stage of a study cannot be corrected later.
Transparency declarations
None to declare.
References
1 In Swinscow TDV and Campbell MJ (Eds.). Statistics at Square One, 10th edn (2002) (BMJ Publishing Group, London) ISBN 0 7279 1552 5.
2 In Altman DG, Machin D, Bryant TN (Eds.), et al. Statistics with Confidence, 2nd edn (2000) (BMJ Publishing Group, London) ISBN 0 7279 1375 5.
CiteULike 
Connotea 
Del.icio.us What's this?
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||