Yeast colonisation is quite common in intensive care unit (ICU) populations. Up to 73% of patients have been reported to be colonised by yeast, predominantly by Candida albicans 1. Candida species are among the most commonly isolated microorganisms from the abdomen and urine in surgical patients with infections 2.

The development of fungal/yeast infections is a rapidly increasing health problem, especially in hospitalized patients and in patients with impaired host defences. In 1995, yeast was reported to be the fourth most common ICU-acquired infection in Europe, where it represented approximately 17% of all isolates 3. This percentage may be even higher now, although more recent data are lacking. Particularly in patients with peritonitis, Candida frequently can be cultured from the abdomen, with prevalences as high as 30% to 40% 4567. Systemic yeast infections are associated with high mortality, often more than 50% 8, with C. albicans as the predominant species responsible 9. Systemic fungal/yeast infections have become more common over the past two decades. Candida is the fourth leading cause of all nosocomial bloodstream infections in the US, accounting for up to 11% of all infections 10. As early as the 1980s, an increase in surgical yeast infections from 2.5/1,000 discharges to 5.6/1,000 discharges was observed 11. The incidence of candidemia increased to 9.8/1,000 ICU admissions among postoperative ICU patients in 1999 12. In another study, the incidence increased from 1.25/10,000 in 1999 to 3.06/10,000 patient-days per year in 2003 13. Invasive yeast infections are associated with high morbidity and mortality, and the cost of bloodstream Candida infection alone is already approaching $1 billion per year in the US 14.

Proper management of yeast infections is challenging because the diagnosis is often elusive. At present, laboratory tests can be inconclusive (blood cultures have a sensitivity of only 70% 15) and it is difficult to distinguish between colonisers and pathogens. Yeast is part of the physiological microbiological flora, thus positive cultures may merely reflect colonisation or environmental contamination instead of actual infection. On the other hand, the gold standard for the diagnosis of candidemia, blood culture, is not perfect. False-negative blood cultures, especially, are a problem because sensitivity is approximately 70% 15.

Given the high and increasing incidence of Candida infection, its major clinical impact, and the lack of tests for an early and accurate diagnosis, a prophylactic approach for high-risk patients might be beneficial. Previous reviews on this topic have analysed specific yeast prophylaxis regimens with either a single-drug antifungal prophylaxis (SAP) or a multi-drug regimen of selective decontamination of the digestive tract (SDD). There are no direct randomised comparisons between SAP and SDD treatments. Our aim is to review and compare the effectiveness of both therapeutic strategies on yeast colonisation, invasive yeast infection, candidemia, and in-hospital mortality.

Both methods of antifungal prophylaxis (SAP and SDD) reduced the odds of developing Candida colonisation, invasive infection, candidemia, and mortality to various degrees in critically ill patients. The present comparative meta-analysis of SAP and SDD antifungal prophylactic regimens allowed us to analyse the differences between both antifungal strategies on outcomes. This is important since no head-to-head comparison studies have been performed.

SDD was more effective in reducing yeast colonisation and infection than single-drug prophylaxis. The clinical importance of these effects on outcomes is illustrated by the fact that Candida colonisation or infection with an identical strain frequently precedes advanced (bloodstream) infection in non-neutropenic patients 2627. Furthermore, the intensity of Candida colonisation is an independent factor for the development of Candida infection 23. However, in a large prospective cohort study, prior colonisation was not associated with bloodstream infections 18. The exact role of colonisation in yeast-related disease should be elaborated further to fully appreciate the effects of prophylactic therapy.

The risk of developing yeast infection was reduced 3.2-fold (from 8.3% to 2.6%) by SDD, which was significantly more effective than SAP. Since most SAP regimens comprise absorbable (mostly intravenous) drugs and SDD contains enteral non-absorbable drugs, these results corroborate the hypothesis that the gastrointestinal tract is the primary source of yeast causing infection. Alterations in the host defence of critically ill patients (prior surgery, parenteral nutrition, hemodialysis, and mechanical ventilation) imply immunosuppression and/or a breach of the mucosal barrier 182328. This can lead to overgrowth of Candida species and increased microbial translocation 7. The difference in efficacy of SDD compared to SAP could possibly be explained by the fact that SDD reduces yeast load at the source of primary yeast colonisation (that is, the gastrointestinal tract).

The incidence of candidemia was significantly reduced by SAP, whereas the reduction in SDD studies was potentially relevant but not statistically significant. Furthermore, the meta-regression analysis revealed no significant difference between the pooled ORs of both types of prophylaxis, so no definite conclusions should be drawn. With a relatively high NNT of 38 for SAP and a relatively low incidence of candidemia of 3.8% in these critically ill patients, the prophylactic use of SAP in the general population of ICU patients does not seem justified, a reported high mortality rate of 25% to 60% 29 in patients with candidemia notwithstanding.

SDD led to a significant reduction in all-cause in-hospital mortality. The decrease in mortality rate may be due, at least in part, to the reduction of Candida infection by SDD, and antifungal prophylaxis indeed decreased mortality directly attributable to yeast. However, in the present review, SDD did not significantly reduce the rate of candidemia, as was shown in an earlier study 30. A previous meta-analysis in critically ill patients showed that pneumonia and Gram-negative as well as overall bloodstream infections were reduced by SDD. 1931, which would account for the decrease in mortality. The value of antifungal prophylaxis is likely to be much higher in individuals who receive antibacterial agents, and the reduced gastrointestinal microbial load may be a possible factor in the greater efficacy of SDD compared with SAP.

A systematic review of SDD in ICU patients found that mortality was reduced significantly in surgical patients only and not in medical patients. 19. In the present review, adjustment for the proportion of surgical patients changed the difference in effect between the two types of antifungal prophylaxis for the outcomes of colonisation and candidemia. The difference in effect of SAP compared to SDD became smaller, but the direction of the difference was stable in favour of SDD. These results show that the effectiveness of prophylaxis is reduced with an increased proportion of included surgical patients.

The results of this review generally confirm the conclusions of earlier reviews examining the separate effects on yeast infections of SDD 3032 or SAP 333435. However, these reviews have some shortcomings. The reviews on SDD/non-absorbable antifungal prophylaxis included pediatric studies 3032 or liver transplant studies 30, which were excluded in the present review. Previous reviews on systemic prophylaxis included studies that included fluconazole studies only 33, studies on surgical patients only 34, or studies that examined groups with non-absorbable prophylaxis as control groups 35. The present review is the first to examine the two methods of antifungal prophylaxis concurrently, showing the pros and cons of each regimen with respect to the outcomes of yeast colonisation, infection, candidemia, and hospital mortality.

The use of systemic (absorbable) drugs, like fluconazole, may have to be restricted for two reasons. First, antifungal drug use can initiate the possible emergence of azole-resistant strains. The present study did not investigate this aspect. Playford and colleagues 35 could find no significant increase of resistant strains in a review of absorbable antifungal prophylaxis. However, the pooled estimates on the emergence of resistant strains had wide CIs and could have suffered from insufficient power. Thus, no definite conclusions can be drawn from these observations. There is no compelling evidence to link prophylactic antifungal therapy with resistance, but it is quite clear that increased use of antifungal drugs has promoted the dissemination of azole-resistant fungi. This needs to be taken into consideration when considering the risk-benefit ratio for instituting widespread use of antifungal prophylaxis in critical care units. Second, systemic antifungal agents can have potential toxic effects. 35. Systemic drugs may be advised as prophylaxis only in patients with increased risk of developing Candida bloodstream infections 182328. SDD may have fewer systemic side effects and thus can be given to critically ill patients to prevent Candida colonisation and infection. However, the effect of SDD use on resistance patterns of yeast is still a matter of debate. SDD may decrease the emergence of antibiotic resistance 36, but an increase in pathological bacteria (enterococci and coagulase-negative staphylococci) is of concern 37. The matter of SDD-induced antifungal drug resistance is unclear.

Several factors have to be considered while interpreting the results of this review. First, most included studies had small sample sizes and the event rates of several outcomes were small. This means that individual studies had wide CIs, but even CIs around pooled ORs were still wide, and therefore the power for detecting clinically relevant differences for some outcomes was small. It means also that subgroup analyses have to be interpreted with care. Second, although the methodological quality of the studies was good on average, there is still room for improvement as only half of the studies applied blinding of treatment allocation and outcome assessment.

Third, studies used a wide variety of criteria for patient inclusion and exclusion. In addition, definitions of yeast colonisation and invasive yeast infection differed between studies as there is still no consensus on this subject. The value of the effect of prophylaxis on invasive yeast infection, especially, must be interpreted with scepticism because definitions vary between articles and it is debatable whether a mixed culture can be seen as evidence of an invasive infection. Unfortunately, not all articles published data on the culture results. The variation among definitions hampers the comparison of these outcomes between studies. In this review, we used widely accepted definitions and tried to redefine results if individual studies used other definitions. In particular, the definitions of prophylactic and pre-emptive treatment are often overlapping. Critically ill patients in the ICU are often already colonised or may be infected without being cultured, so prophylactic therapy can often be seen as pre-emptive. However, pre-emptive therapy will be given to a selected group of patients with a higher risk of yeast infections. There was one study that compared a pre-emptive strategy with a control group 38. In the analysis of our data, we made no distinction between prophylactic and pre-emptive treatment strategies. Despite these differences in populations and definitions, the results for most outcomes were relatively homogeneous across studies with accompanying values of the I² statistic that were low.

Fourth, our review, like any other review, may have suffered from publication or selective reporting bias. Studies with a 'negative' result may be less likely get published, and results from non-significant outcomes are less likely to be reported 3940. The number of these file drawer studies exceeded the number of included studies in the analyses with positive results, reducing the unreliability of the validity of the outcomes. However, 14 studies provided antifungal therapy but were excluded for not scoring yeast/fungi. These studies did not state whether culture results were negative or whether no specific cultures were performed. This may cause pooled results of included trials, which are too optimistic.

Both SAP and SDD antifungal prophylaxis strategies were effective in reducing yeast-associated disease across a range of critically ill patients. The reduction of yeast colonisation and infection was more pronounced in SDD studies compared with SAP, whereas candidemia was reduced foremost by SAP. SDD reduced all-cause in-hospital mortality, but both strategies reduced yeast-related mortality. Systemic drugs may be advised as prophylaxis in patients with a high risk of developing Candida bloodstream infections, whereas SDD may be given to critically ill patients to prevent Candida colonisation and infection.

• Antifungal prophylaxis reduces yeast-related morbidity and mortality.

• Selective decontamination of digestive tract (SDD) regimens are more effective than prophylactic regimens that include single antifungal drugs in reducing yeast colonisation, invasive yeast infection, and in-hospital mortality.

• The effectiveness of antifungal prophylaxis is inversely related to the proportion of included surgical patients.

• The incidence of candidemia is reduced by systemic antifungal prophylaxis but not by SDD.

ALI = acute lung injury; APACHE II = Acute Physiology and Chronic Health Evaluation II; ARDS = acute respiratory distress syndrome; CI = confidence interval; DF = degree of freedom; IQR = interquartile range; NNT = number needed to treat; OR = odds ratio; RCT = randomised controlled trial; SAP = single-drug antifungal prophylaxis; SDD = selective decontamination of the digestive tract; SICU = surgical intensive care unit.

The authors declare that they have no competing interests.

OvT participated in designing the study and in collecting and entering data. OvR and BL participated in designing the study. RW participated in collecting and entering data. JR participated in advising on statistical methodology. MB participated in designing the study and in advising on statistical methodology. All authors were responsible for critical analysis and interpretation of data. All authors read and approved the final manuscript.