The most award winning
healthcare information source.
TRUSTED FOR FOUR DECADES.
Abstracts & Commentary
Synopsis: While this analysis of the EORTC trials database concludes, by multivariate analysis, that the administration of azole antifungal agents increases the risk of subsequent bacteremia in febrile neutropenic patients with malignancy, the validity of the results are questioned.
Sources: Viscoli C, et al. Association between antifungal prophylaxis and rate of documented bacteremia in febrile neutropenic cancer patients. Clin Infect Dis. 2001;32:1532-1537; Wenzel RP, et al. Antifungal antibiotics and breakthrough bacteremias. Clin Infect Dis. 2001;32:1538-1539.
Viscoli and colleagues were intrigued by several reports that azole antifungal agents appeared to predispose neutropenic patients to developing bacteremia and by their own earlier observation that exposure to such agents was found to be an independent risk factor for bacteremia. Their position in the EORTC gave them access to a large uniform database containing 3080 records generated from similar trials that had been conducted during 1986 to 1994 to address the totally different question of which regimen was better for the empirical therapy of the febrile neutropenic patient. A diverse number of variables were included in the logistic regression model of the probability of detecting bacteremia with antifungal prophylaxis first being omitted and then included. This led to an improved model such that the odds ratio (OR) for receipt of absorbable antifungal agents for developing bacteremia was 1.41. Seven other factors gave higher ORs including shock, high temperature, and at least 1 day of granulocytopenia (see Table), while being older than 30 years of age yielded almost the same OR as absorbable antifungal agents. Factors that were less likely to be associated with bacteremia included hematological malignancies other than acute lymphoblastic leukemia (OR 0.64), granulocyte count < 0.1 ´ 109/L (OR 0.69), and antibacterial prophylaxis (OR 0.7). Hence, the results appear to support an association between the use of the absorbable azoles—ketoconazole, fluconazole, and itraconazole—and the occurrence of bacteremia suggesting that this may be clinically important.
Factors Associated with Bacteremia Detection
|Temperature ³ 40°C||4.95||3.14-7.81||< 0.0001|
|> 15 days of
(< 0.5 ´ 109 cells/L)
|> 10 days in hospital||2.07||1.49-2.90||< 0.0001|
|Treatment for other than
|1-15 days of granulocyto-
penia (< 0.5 ´ 109 cells/L)
|> 30 years of age||1.37||1.10-1.69||0.004|
The accompanying editorial by Wenzel and colleagues casts serious doubt upon this conclusion because of the failure to correct for confounding factors, the failure to estimate the risk for developing bacteremia that can be attributed to absorbable antifungal regimens, which turned out to be only 7% and, more importantly, the failure to offer any biologically plausible explanation for the phenomenon.
Comment by J. Peter Donnelly, Phd
Studies of this type are fashionable and it has become popular to try and distill evidence from large databases for or against an opinion, a long-held suspicion, an apparently common observation or, as in this case, an intriguing series of independent and unusual observations because PCs are ubiquitous, the necessary software is readily available, and it is relatively easy make use of existing databases. Granted, a professional statistician was intimately involved in Viscoli et al’s study, but such an undertaking would have been just too time-consuming to do a decade ago to be considered worthwhile. That aside, what is the reader supposed to make of this and similar reports and how should they be placed in a proper context? Wenzel et al have done us a great service in critically appraising this study from an epidemiological standpoint. Their highlighting of the inadequate correction of confounding variables and their estimation of attributable risk help in this regard. However, even without their guidance, an amateur epidemiologist like myself would have had serious intuitive reservations since it is indeed hard to even begin to advance a biological explanation for the observation. Viscoli et al did admit to recognizing that the use of antifungal prophylaxis might be a marker for another factor but did not venture to suggest any. Might it not be that the drugs in question were only given to those patients likely to be at greater risk of developing bacteremia anyway because the intense treatment they received induced more mucosal damage which is associated with certain types of bacteremia? We were also not informed which bacteria were involved in bacteraemia, but one would guess that the Gram-positive cocci predominated over the Gram-negative bacilli given the fact that half the patients had been given antibacterial prophylaxis and, further, that the viridans streptococci, which are known to be associated with oral mucositis, would have been among the most frequent isolates. Indeed, had mucositis been included in the model we might have seen that this rather than antifungal prophylaxis was a risk factor. There is just no substitute for there being a biologically plausible basis for any association found between variables no matter how statistically significant. The temptation to think otherwise should be firmly resisted.