|Home | About | Journals | Submit | Contact Us | Français|
Of 57 case-control studies of antimicrobial resistance, matching was used in 23 (40%). Variables on which matching was done differed substantially across studies. Twelve (52%) justified the use of matching and 9 (39%) noted strengths or limitations of this approach. Analysis accounting for matching was performed only 52% of the time.
There is growing concern about antibiotic resistance.1, 2 Risk factors for antimicrobial resistance have been studied to guide interventions aimed at decreasing their emergence. These studies often employ a case-control study design and may utilize individual-level matching. Matching can be used to control for confounding and is particularly helpful when sample sizes are small, limiting the number of confounders that can be adjusted for in the analysis, or when there is insufficient overlap between case and control subjects (i.e., sibship).3, 4 However, it is unclear how often and why a matched case-control study is performed.
The purpose of this study was to investigate the role of matching in case-control studies evaluating risk factors for antimicrobial resistance. We assessed how often matching is utilized; what variables are matched on; whether authors justify the use of or comment on the strengths or limitations of matching; and how often analysis accounting for matching is performed.
We reviewed the table of contents of four infectious diseases and five general medicine journals we believed would most likely publish articles evaluating risk factors for antimicrobial resistance: Clinical Infectious Diseases (CID), Emerging Infectious Diseases (EID), Infection Control and Hospital Epidemiology (ICHE), American Journal of Infection Control, New England Journal of Medicine, Annals of Internal Medicine, Journal of the American Medical Association, Lancet, and British Medical Journal. All abstracts from 2000 to 2005 were reviewed to identify studies that used a case-control study design to evaluate an antimicrobial-resistant pathogen. Full text manuscripts were then read to identify studies that employed individual-level matching.
Two investigators independently reviewed the literature and assessed articles for inclusion (EC and IL). If opinions differed, two additional investigators evaluated the articles (EL and DL) and consensus was used to arrive at a final decision.
The following data were abstracted for each study: journal, year of publication, country of origin, study setting, number of case and control subjects, matched variables (i.e., age, sex, time at risk, and others); whether the use of matching was justified; whether the strengths or limitations of matching were noted; and whether analysis accounting for matching was used (i.e., McNemars test, sign test, paired t test, or conditional logistic regression).
Descriptive analyses were performed to determine how often matching was used and to describe the matched variables. We used Fisher's exact test to assess if matching differed across journal type (i.e., infectious diseases versus general medicine), the particular journal, or year of publication. Chi-square test for trend was used to evaluate the use of matching over time.
We identified 57 case-control studies in six journals evaluating risk factors for antimicrobial-resistant pathogens. A majority (96%) focused on bacterial pathogens. One study evaluated herpes simplex virus, and none evaluated fungal or mycobacterial pathogens.
Of the case-control studies, 23 (40%) from the following three journals utilized individual-level matching: CID = 14 (61%), ICHE = 6 (26%), and EID = 3 (13%). No general medicine journals contained articles meeting inclusion criteria.
Matching was not significantly associated with the journal type (p=0.51), a particular journal (p=0.15), or year of publication (p=0.93). There was no trend in the use matching over time (p =0.72).
The most commonly investigated pathogens were: methicillin resistant Staphylococcus aureus = 6 (26%), vancomycin resistant Enterococcus species = 5 (22%), resistant Pseudomonas aeruginosa = 4 (17%), and extended spectrum beta lactamase Escherichia coli or Klebsiella pneumoniae = 4 (17%). Only two studies were conducted outside the hospital setting.5, 6 Although most employed a 1:1 ratio of cases to controls [9 (39%)], several studies used more than one control for each case subject or multiple control groups. Most studies matched on more than one variable [19 (83%)]. The most commonly matched variables were: hospital ward = 11 (48%), date of admission = 8 (35%), time at risk = 8 (35%), age = 6 (26%), and sex = 5 (22%).
Twelve (52%) articles used statistical tests accounting for matching. Twelve articles (52%) justified the use of matching and 9 (39%) discussed the strengths or limitations of matching. Controlling for confounding was the primary justification and primary strength given.3, 4
In the nine journals we reviewed, individual-level matching was used in 40% of case-control studies evaluating risk factors for antimicrobial-resistant pathogens. Most matched on more than one variable, and these variables differed substantially across studies. Half the articles justified the use of matching and 39% acknowledged the strengths or limitations of matching. Only 52% used statistical tests accounting for matching.
A majority of the studies matched on more than one variable. Falagas and colleagues similarly reported that 76% of matched case-control studies evaluating post-operative infections used more than one matched variable.7 In their study, age was most commonly matched variable, followed by surgical procedure, gender, time at risk, and date of operation.7 Although age and sex have traditionally been the variables most often used for matching,4 we found that hospital ward, date of admission, and time at risk, were more commonly used.
Half the studies justified the use of matching and 39% commented on the strengths or limitations. Most reported that matching was used to control for confounding. The effect of potential confounders can be limited using several methods, including restriction, randomization, matching, or using stratified or multivariable analysis. Matching can be particularly helpful in studies with small sample sizes, where the number of patients may limit the number of confounders that can be adjusted for, or in studies where there may be insufficient overlap.3, 4 There are, however, limitations of matching. It can be logistically difficult to identify control patients, particularly if more than one matched variable is used; the matched variable cannot be assessed as a potential risk factor; and if case and control subjects are matched on variables related to the exposure, overmatching can occur.4 No articles commented on why matching was used over these other strategies.
Our study reported that only 52% of the studies used matched analysis. If matching is used in the study design, it is important to also use biostatistical tests accounting for matching in the analysis (i.e. McNemars test, sign test, paired t test, or conditional logistic regression). If these tests are not utilized, the benefits of matching done in the design of the study are lost (i.e., potential confounders are not controlled for).3, 4
There are several limitations of our study. We selected a representative sample rather than surveying all infectious diseases and general medicine journals. We believed the selected journals would most likely publish articles evaluating risk factors for antimicrobial resistance. Secondly, data abstraction was limited to the information in these articles. Authors may have been able to justify the use of or comment in further detail on the strengths and limitations of matching if contacted directly.
Matching can be used in case-control studies to adjust for potential confounders. However, matching has important limitations and other alternative methods are available to control for confounding. To better understand the role of matching in case-control studies evaluating risk factors for antimicrobial-resistant pathogens, it would be informative for future studies to justify why matching is being utilized over other available methods. Lastly, it is important for investigators that utilize matching in the study design to then account for matching in the analysis.
Financial Support: This work was supported by the Institutional National Research Service Award (T32 AI055435) (I.L.) and by the Mentored Patient Oriented Research Career Development Award (K23-AI-060887) of the NIH from the National Institute of Allergy and Infectious Diseases (D.R.L.).
Conflict of Interest: All authors report no conflict of interest relevant to this article.