The emergence of C. difficile as an infection in individuals without prior hospitalization, and presumed community acquisition, represents a concerning development in the ongoing emergence of this pathogen. As any prescription of an antimicrobial agent to a patient in an outpatient setting requires a careful weighing of risks and benefits, we performed a systematic review to quantify the risks associated with individual antibiotic classes and to identify areas of heterogeneity in such risk. Overall use of antibiotic agents is associated with a 3-fold increased risk of community-acquired CDI, but we also detected substantial variation in risk associated with different antimicrobial classes, with fluoroquinolones, CMCs, and clindamycin associated with the greatest enhancement of risk.
This study largely corroborates the associations found for hospital-associated CDI risk (10
). In keeping with many historic studies of CDI risk and outbreaks of the disease, clindamycin was found to have the strongest association with risk. One must note however, that clindamycin has not been associated with the greatest risk enhancement in every study (28
); variability in effects may be due to true biological heterogeneity of effect (e.g., variable strain susceptibility to clindamycin [29
], timing of inoculation relative to the end of antibiotic exposure), or it could be an artifact of the different methods used for outcome ascertainment (see below).
Our study found large effects for fluoroquinolones and CMCs. This could be expected given the broad spectrum of activity of these agents against intestinal microbes and the low susceptibility of Clostridium difficile
to these classes of antibiotics (30
). The risk associated with CMCs was highly variable across studies, in contrast to fluoroquinolones, which appeared to have more-consistent effects. This heterogeneity may be due to the greater activity of newer cephalosporins against anaerobic bacteria and Gram-negative bacilli (10
). In contrast, one study limited to patients with fluoroquinolone exposures (31
) found no differences in effect between levofloxacin, gatifloxacin, and moxifloxacin, notwithstanding the enhanced anti-anaerobic spectrum of the latter agents.
Our findings reaffirmed the finding of moderate effects for penicillins, macrolides, and sulfonamides and trimethoprim from a recent hospital-based study (28
); this is in contrast to other hospital-based studies that have noted large effects for penicillins (11
). The relatively low MICs for penicillins among common CDI strains (32
) could help explain the observed modestly elevated risk level for the penicillin class, such that the enhancement of CDI risk resulting from elimination of normal enteric flora is somewhat counterbalanced by anti-C. difficile
activity. These discrepancies may also result from wide variations in the antibiotic spectrum of penicillin subclasses (including broad-spectrum penicillins used more in the hospital setting such as piperacillin-tazobactam). Our meta-analysis noted that tetracycline antibiotics have little antibiotic-associated risk, which is in keeping with the only meta-analysis of inpatient CDI (10
Like any observational study, the findings of studies incorporated into this meta-analysis could have been biased by methodological flaws, including issues of control selection, misclassification of both outcomes and exposures, and residual confounding (33
). Our quality checklist attempted to assess the overall risk of these biases in each study; we outline some specific observations below. With respect to the definition of the population, two studies did not exclude patients exposed to hospital settings during the risk period, and as such may actually represent studies of community-onset but hospital-acquired disease (20
) while two studies were restricted to patients who received a C. difficile
assay, and as such, the controls did not represent the source population of cases (20
). With respect to ascertaining the outcome, all positive cases may have been subject to misclassification due to infection with another diarrhea-causing organism. Further, in two studies (17
), a lack of clinical detail meant that hospital-diagnosed cases with onset of symptoms ≥48 h after admission could not be separated from those with onset within 48 h. As such, unmeasured inpatient antibiotic exposures may have caused the disease outcome. Indeed, in the study of Dial et al. (17
), outpatient antibiotic exposures were detected in only 47.1% of the cases. Of the studies included in this meta-analysis study, only one (34
) considered the robustness of results to diagnostic suspicion bias by comparing effect sizes from clinical diagnoses to those with test-based confirmation; they found no significant differences in effect with the clinically diagnosed subgroup.
Other potential sources of bias in our meta-analysis could include a lack of consensus regarding the appropriate time window for identification of antibiotic exposure. As risk associated with antibiotic exposure decreases with increasing time, larger effect sizes are liable to be found in studies looking at the shorter time windows. Indeed, the study in our primary analyses with the shortest exposure window reported larger effect sizes for all antibiotics except tetracyclines (17
). In fact, the appropriate time window may differ between antibiotic classes due to differing antimicrobial effect duration (35
). In addition, simultaneous administration of multiple antimicrobial agents and confounding by indication (as individuals receiving antimicrobials may have underlying health conditions placing them at greater risk for CDI) may have biased results.
Finally, although we did not find evidence to suggest that our findings were influenced by publication bias, we did notice some selective reporting of antibiotic class exposures. Specifically, the smallest study meeting the inclusion criteria (36
) failed to report effect estimates for 3 of the 7 antibiotic classes (tetracyclines, macrolides, and sulfonamides and trimethoprim), and none of the studies reported on the impact of oxalizidenones, glycopeptides, carbapenems, or aminoglycosides.
In summary, and on the basis of the best available evidence, we found that the risk profiles for antimicrobial classes as risk factors for community-acquired CDI are similar to those described for health care-associated disease. In particular, antimicrobial classes with broad-spectrum, and potent anti-Gram-negative and/or antianaerobic bacterial activity, including cephalosporins, fluoroquinolones, and clindamycin, are most likely to cause CDI. In contrast, macrolides, penicillins, sulfonamides and trimethoprim, and particularly tetracyclines confer a lower risk of CDI. While community-acquired CDI remains fortunately less common than its health care-associated counterpart, we propose that CDI risk represents yet another factor that needs to be factored into the decision to prescribe antimicrobials (and the choice of antimicrobial) in the outpatient setting.