Objective. Neonatal candidemia is an increasing cause of infant morbidity and mortality. We evaluated the current medical literature in an effort to critique the literature and to document the reported prevalences of end-organ damage after neonatal candidemia.
Methods. We analyzed all peer-reviewed articles of neonatal candidemia published in the English language; inclusion criteria included a cohort limited to all neonatal intensive care unit admissions or all episodes of candidemia in neonates. Articles that also incorporated older patients, did not define a numerator and a denominator for at least 1 form of end-organ damage, included patients from other reports, or did not include all episodes of candidemia in the source population were excluded from the analysis.
Results. Thirty-four articles reported episodes of candidemia and mortality; 21 articles reported prevalence for at least 1 form of end-organ damage. Only 4 (19%) of 21 articles reported prevalence for >4 forms of end-organ damage from the following list: endophthalmitis, meningitis, brain parenchyma invasion, endocarditis, renal abscesses, positive cultures from other normally sterile body fluids, or hepatosplenic abscesses. The median reported prevalence of endophthalmitis was 3% (interquartile range [IQR]: 0%–17%), of meningitis was 15% (IQR: 3%–23%), of brain abscess or ventriculitis was 4% (IQR: 3%–21%), of endocarditis was 5% (IQR: 0%–13%), of positive renal ultrasound was 5% (IQR: 0%–14%), and of positive urine culture was 61% (IQR: 40%–76%). The medical literature concerning end-organ evaluation after episodes of neonatal candidemia is heterogeneous and consists largely of single-center retrospective studies. Year that the data were collected and prevalence of neonates infected with Candida albicans were associated with observed heterogeneity.
Conclusions. Given the heterogeneity of the medical literature, precise estimates of the frequencies of end-organ damage are not possible and a prospective multicenter trial is warranted, but the data from the published literature suggest that the prevalence of neonates with end-organ damage not only is greater than 0 but also is high enough that until such a prospective trial is completed, end-organ studies should be considered before the conclusion of antifungal therapy.
Neonatal candidemia occurs in 4% to 15% of extremely low birth weight infants (birth weight <1000 g), and the 30-day mortality approaches 40%.1–6 Candidemia is also the source of considerable morbidity: invasion of the eyes can cause blindness, meningitis and brain abscesses can lead to severe neurologic impairment, fungal endocarditis often requires surgical intervention or prolonged therapy, and renal abscesses can cause renal failure.6–26
There has not been a large multicenter prospective study to outline the prevalence of end-organ damage in neonates with candidemia. There is no consensus of which organs to evaluate or how that information might be applied to clinical decision making. It is not surprising, therefore, that evaluation of neonates who experience candidemia is likely to be incomplete.27 In this study, we evaluate the medical literature of candidemia, assess the literature for heterogeneity using meta-analytic techniques, provide clinicians the distribution of the reported prevalence of end-organ damage after candidemia, suggest strategies for evaluation on the basis of the data, and discuss how such testing might alter clinical management.
We conducted a Medline review covering January 1966 through March 2002 with the search terms “Candida or candidemia or candidiasis” and limited our search to “neonates or infants or children.” We included articles that provided at least 1 estimate of the prevalence of Candida invasion (number of patients with end-organ damage/number of patients evaluated for end-organ damage) for 1 organ system, consisting of endophthalmitis, meningitis, brain parenchyma abscesses and ventriculitis, endocarditis, positive urine cultures, renal abscesses, and hepatosplenic abscesses. We included peer-reviewed articles that contained a well-described population of either all admissions to a neonatal intensive care unit (NICU) or all neonates from a NICU who had at least 1 positive blood culture for Candida.
Because the methods of ascertaining candidemia as the cause of death were disparate (eg, autopsy requirements, death within 1–7 days after most recent positive blood culture, oftentimes undefined), we relied on the interpretation by the principal investigator as to candidemia as the cause of death. We required positive cerebrospinal fluid (CSF) cultures for meningitis, radiographic evidence of abscesses or ventriculitis, and ophthalmologic examination conclusive for endophthalmitis. End-organ damage of the heart was determined by echocardiography findings consistent with endocarditis or large sessile mass of the wall of the myocardium (because formal criteria for the diagnosis of endocarditis, ie, Duke criteria, were not provided by most authors). Echogenic findings consistent with abscesses of the liver and spleen or kidney were required for hepatosplenic involvement and renal end-organ damage, respectively. We defined prospective articles to be studies in which the authors initiated a prospective study to ascertain risk factors related to candidemia or complications from candidemia for a presubscribed period of time. We defined articles that were retrospective analyses of prospectively acquired data as retrospective.
We limited our search to articles written in English. We excluded articles that included patients who were not neonates, for example, articles that (in addition to neonates) incorporated children from pediatric intensive care units, older children, or adults. We excluded articles that were case series based on end-organ damage, for example, articles that reviewed all cases of neonatal arthritis, abscesses, or Candida meningitis but did not provide data for all episodes of candidemia. We excluded neonates who had other known underlying immune compromise, such as human immunodeficiency virus or severe combined immunodeficiency, or those who received chemotherapy. When a complete accounting for the denominator and a numerator was not included for at least 1 organ system, we did not include the article. We also excluded several articles that included cases of candidemia, candiduria, and Candida meningitis and reported end-organ damage but did not link end-organ damage to candidemia.28 We excluded articles that had overlap of patients from previous articles, so as not to count the same patients multiple times.29–31
Additional information was found by following the reference citations from retrieved articles and by hand search. Two reviewers (D.K.B., J.R.) selected studies according to the criteria above.
We developed a template for data abstraction before review to increase reliability. Two reviewers (D.K.B., W.J.S.) independently abstracted the results. When differences occurred, the reviewers discussed the disparity and came to a consensus. On 1 occasion, a third reviewer (J.R.) settled the dispute. The following information was abstracted from each included study (Tables 1 and 2): first author, total number of patients, end-organ damage, type of study (prospective, retrospective, randomized), number of hospitals that comprised the source population, year that the study was conducted and published, number of patients who died secondary to candidemia, and species of Candida recorded.
Data Transformation and Analysis
We assigned a median year of data collection to each article (eg, an article that evaluated all patients with candidemia between January 1, 1981, and December 31, 1985, would have a median year of data collection of 1983). We then split the articles into 2 approximately equal-sized groups, an early group (before 1991) and a reference category, the late group. Prospective studies, studies conducted at >1 center, studies in which attributed mortality was <25%, and proportion of episodes attributed to Candida albicans <80% were the reference categories when compared with retrospective studies, single-center studies, mortality >25%, and C albicans proportions >80%, respectively (Table 2). We also evaluated year that the study was conducted, the proportion of infants who died, and the proportion of infants who were infected with C albicans as continuous variables.
We estimate prevalence in this text because it is possible for neonates to succumb to candidemia before end-organ evaluation. Prevalence estimates were summarized with the minimum, median, maximum, and interquartile range to illustrate the distribution of study findings without consideration of heterogeneity. These were specific to each form of end-organ damage. For example, if there were 5 studies that reported a numerator and a denominator for radiographic evidence of brain abscesses or ventriculitis and the studies that reported 0 of 4, 3 of 87, 1 of 23, 7 of 33, and 3 of 6 children had brain parenchyma invasion, then we would report the minimum as 0 (0 of 4), the median as 4% (1 of 23), and the maximum as 50% (3 of 6).
We then transformed the organ-specific prevalence estimates into their respective logits (natural log of the probability divided by 1 − probability) to take advantage of the properties of the logit (normal distribution and stable variance) transformation. We then tested for evidence of heterogeneity and publication bias. We used funnel plots to estimate asymmetry in the relationship between the logit and the precision of the logit estimate; we also used rank order correlation tests to quantify the association between the prevalence estimates and their standard errors.32–37 To address heterogeneity, we compared average prevalence estimates between groups of studies using random-effects meta-regression.38 The dependent variable was the logit of the prevalence of organ-specific estimates, and the independent variables were timing of study, type of study, number of centers, attributable mortality within study, and proportion of infections attributable to C albicans (Table 3).
The regressions were fit with random-effects weights (reciprocal sum of the estimated within-study and among-study variances) using restricted maximum likelihood estimation of the between-study variance to account for heterogeneity. For each independent variable, we compare studies and present 95% confidence intervals (CI) of the prevalence odds ratios (POR). The meta-regression techniques presented are analogous to logistic regression. However, because systematic reviews are primarily explanatory and observational, P values are not presented, and formal inferences are not drawn from the presented 95% CI.
Because several studies reported a prevalence of 0 for various types of end-organ damage, we conducted an informal sensitivity analysis of the assessment of heterogeneity. When transforming prevalence estimates to the logit, if the prevalence is 0, then the logit (natural log of prevalence divided by 1 minus the prevalence) becomes undefined and the study cannot be evaluated by meta-regression, although some previous investigators have advocated adding 0.5 to each numerator and denominator in such a scenario.39 In some of these studies, however, the sample size is so small that such a procedure would often substantially alter the estimated prevalences, sometimes enough to move a study from the bottom of the rank order to near the top. We therefore repeated the analysis by adding 0.5, 0.25, and 0.15 to the numerator and 0.5, 0.75, and 0.85 to the denominator, respectively, for each study that reported a prevalence of 0. This is equivalent to using a very diffuse β prior-probability distribution with N = 1 in generating the study-specific prevalence estimates.
Thirty-seven articles provided an identifiable cohort of neonates with positive blood cultures with no overlap from patients included in other articles. Only 21 of these (Table 1) provided prevalence of end-organ damage (complete presentation of numerator and denominator) for at least 1 organ system.6–26 Only 6 of the studies were prospective in design: 5 observational and 1 randomized placebo controlled trial. All but 2 of the studies were single-center studies (Table 2).
Of the 564 patients included in the 21 studies, 129 died secondary to candidiasis, according to the authors. In Table 3, we present the minimum, median, maximum and interquartile prevalence reported for each form of end-organ damage. There were substantial differences in prevalence within organ system reported between studies. The median prevalence for endophthalmitis was 3% (range: 0%–57%), for culture-positive meningitis was 16% (range: 0%–67%), for other central nervous system invasion such as ventriculitis and brain parenchyma abscesses was 4% (range: 0%–69%), for endocarditis was 5% (range: 0%–15%), for positive urine culture was 49% (range: 0%–80%), for renal abscesses was 5% (range: 0%–33%), and for hepatosplenic abscesses was 1% (range: 0%–3%).
We then assessed the potential for publication bias by examining funnel plots (not shown) for each of the end-organ prevalence measures and conducted the Begg and Mazumdar40 and Egger et al36 tests for funnel plot asymmetry. The P values for Begg tests were >.20 for each of the organ system prevalence evaluated and the Egger P values were .003 for meningitis, .19 for endophthalmitis and endocarditis, .46 for abdominal abscess, .67 for central nervous system involvement, and .99 for positive urine culture; but tests for funnel plot asymmetry traditionally have low power. We also evaluated the heterogeneity of prevalence estimates for each organ system by Cochran’s Q test. This test gave a P value of <.05 for most of the organ systems, an indication of substantial heterogeneity. Given these results, the wide range of prevalence values for each of the organ systems, and the differences in study design, overall summary estimates using meta-analytic methods were not considered appropriate.
In an effort to determine the cause of the observed heterogeneity, we conducted meta-regression analyses using the logit of the end-organ damage prevalences as the dependent variable and the study characteristics in Table 2 as the independent variables. We provide a detailed example of this analysis in Table 4 using meningitis as the end organ damage of interest. The calendar year that the study was conducted was associated with a lower prevalence of end-organ damage, with studies conducted in the 1970s and 1980s reporting higher prevalences of meningitis than studies conducted in the 1990s (POR: 2.18). Studies that reported a higher proportion of neonates infected with C albicans were also associated with a higher prevalence of meningitis (POR: 2.8).
We then repeated the analysis in the other organ systems, and the results were similar. Earlier time of study and higher proportion of C albicans species were also associated with higher prevalence of endophthalmitis (POR: 1.19 and 5.4, respectively). Sample sizes were small, but an earlier year of study was weakly associated with higher reported prevalence in the evaluation of ventriculitis and brain abscesses (POR: 1.17) and positive urine culture (POR: 1.06). Sample size constraints prohibited stable meta-regression using the proportion of Candida species as the independent variable and the logit of the prevalence of endocarditis, ventriculitis and brain abscess, or abdominal abscess as the dependent variable. Sample size also prohibited multivariable meta-regression.
For the variables most strongly associated with reported prevalence (median year of investigation and proportion of infections with C albicans), we repeated the analysis for these independent variables as continuous variables. The results were similar with the exception of endocarditis and median year of observation. In the case of meningitis, for example, a difference of 10 years in median year of observation (eg, 1986–1976) resulted in a POR of 1.81 (95% CI: 1.23–2.7), suggesting that over the last 30 years, reported prevalence of meningitis has decreased. The association between median year of observation and reported prevalence of endophthalmitis was similar to meningitis (POR: 5.5; 95% CI: 2.0–100) for 10 years’ difference in observation. The association between median year of observation and reported prevalence of positive urine culture (POR: 1.81; 95% CI: 0.67–4.76), positive echocardiogram (POR: 2.0; 95% CI: 0.13–31), and brain parenchyma involvement (POR: 5.0; 95% CI: 0.58–50) was not as strong as that observed in meningitis and endophthalmitis.
Reported prevalence was also associated with changes in the proportion of children infected with C albicans; an absolute increase in the proportion of C albicans infection of 10% was associated with a meningitis POR of 1.4 (95% CI: 1.0–1.9). This suggests that as the proportion of children infected with C albicans increases, so, too, does the reported prevalence of meningitis. This association between C albicans proportion of infection and increase in the reported prevalence of end-organ damage was also observed in endophthalmitis (POR: 1.59; 95% CI: :0.71–3. 6), positive urine culture (POR: 1.5; 95% CI: 1.0–2.1), and positive echocardiogram (POR: 2.0; 95% CI: 0.73–5.4). Limited sample size precluded similar analyses for abdominal ultrasound results. In the informal sensitivity analysis, the associations between independent variables (year of investigation and proportion of infections caused by C albicans) strongly associated with reported higher reported prevalence were again seen to be associated with higher reported prevalence of end-organ damage.
Observed Heterogeneity and Association With Selected Study Characteristics
Although a majority of authors reported the Candida species of infecting organism, rarely was species and episode of end-organ damage clearly linked in the articles. However, we found that the proportion of infants who were infected with C albicans was linked with an increase in the prevalence odds of end-organ damage: the higher the proportion of infants who were infected with C albicans, the higher the proportion of infants with meningitis, endophthalmitis, and so forth. The changing epidemiology of candidemia in many NICUs may explain some of the heterogeneity in the proportion of infants who experienced end-organ damage. Although Candida parapsilosis probably causes less mortality than C albicans,1,5 there are not enough data to support different end-organ damage evaluations on the basis of species.
The years that the participants were studied were also associated with observed heterogeneity in the reported prevalence of end-organ damage. An earlier median year of participant observation was associated with higher reported prevalence of end-organ damage. One possible explanation for the reduction in reported prevalence over time relates to publication bias. Investigators notice an unusually high prevalence of infectious complications and report their observations. Subsequent investigators also observe infectious complications (in this case, end-organ damage after candidemia), but the prevalence is not as high. A second possible explanation for the observed heterogeneity may be an increased awareness of the morbidity and mortality associated with candidemia and improved management. In the 1970s, as ventilator and other supportive technology made the survival for very premature infants possible, neonatal candidemia gained recognition as an infectious cause of morbidity and mortality in the nursery. By the 1990s, Candida species became the fourth most common cause of nosocomial bloodstream infection. During the 1980s and 1990s, clinicians became increasingly aware of prompt removal of the central catheter in nosocomial candidemia, aggressive and prolonged antifungal therapy—management that improved morbidity and mortality in immunocompromised patients.
The relationships observed between median year of publication and reported prevalence and proportion of infections with C albicans could be confounded. That is, over the past decade, the cause of candidemia is more frequently C parapsilosis rather than C albicans. Potential confounding between species, timing of study, and prevalence of end-organ damage could be addressed by multivariable meta-regression. However, the limited number of studies that reported both proportion of infections attributable to C albicans and the prevalence of specific types of end-organ damage precluded this analysis.
Study Characteristics With Little Observed Influence on the Reported Prevalence
We suspected that increased mortality might be associated with increased (reported prevalence of end-organ damage) morbidity, but increased mortality was not strongly associated with increased reported prevalence of end-organ damage. We also suspected that two aspects of study design—retrospective studies versus prospective studies and multicenter studies versus single-center studies—might be associated with increased reported prevalence. Neither prospective design nor multicenter design was associated with reported prevalence of end-organ damage. However, the sample sizes are extremely small; only 2 studies were conducted at >1 hospital, and only 6 were prospective.
Candida species are capable of invading all vital organs after candidemia, yet Rowen and Tate27 reported in a 1998 survey of neonatologists and infectious disease consultants that end-organ evaluation is inconsistent. In that survey, 88% of respondents obtained CSF after documented neonatal candidemia, whereas fewer obtained a urine culture (86%), an ophthalmologic examination (81%), an echocardiogram (66%), abdominal ultrasound (52%), or head ultrasound (19%). In this meta-analysis, we presented the distribution of the estimates of the reported prevalence of end-organ damage resulting from neonatal candidemia. For clinicians who care for only a few infants with candidemia each year, it is likely that they will not routinely care for patients with endophthalmitis or brain parenchyma abscesses secondary to candidemia. With the potential for infrequent clinical exposure and disparate reporting in a rapidly expanding fund of clinical knowledge, it is not surprising to find disparate practice styles in end-organ evaluation.
The data that we have reviewed are largely retrospective case series; because the overall quality of the data are suspect and there was evidence of heterogeneity, we cannot provide strong, evidence-based conclusions for end-organ evaluation. Despite the observed heterogeneity and an inability to provide a precise estimate of end-organ damage after neonatal candidemia, the probability of end-organ damage exceeds 0. Therefore, on the basis of the data that we have presented, application of studies completed in older children and adults, clinical experience, and the recommendations proposed by the Infectious Disease Society of America, we provide suggestions to the practicing clinician for end-organ evaluation.
We suggest routine examination of the CSF in all neonates with candidemia because the data have both prognostic and therapeutic significance. Accurate assessment of Candida meningitis requires acquisition of culture because CSF cultures can be positive despite normal cell counts. Although amphotericin B penetrates the CSF of the neonate more efficiently than the adult patient, doses of <1 mg/kg/d amphotericin B are not sufficient for the treatment of Candida meningitis.40–42 Although it is not the purpose of this article to outline antifungal management, in cases of documented Candida meningitis, the clinician should consider doses of amphotericin B of ≥1 mg/kg/d, the addition of 5-fluorocytosine, or both.
Obtaining a head ultrasound is also worthwhile to provide families prognostic information given the data that indicate that children who have had Candida ventriculitis or brain abscesses have a poor neurologic outcome compared with birth weight–matched control subjects.11,43 An ophthalmologic examination after candidemia is also probably warranted. There are data,22 albeit disputed,44,45 to indicate that candidemia increases the need for 1) surgical correction in retinopathy of prematurity and 2) vitrectomy, especially when there is a high fungal burden in the vitreous.46,47 We do not have the data to support medical management or surgical intervention but suggest ophthalmologic examination during antifungal therapy, as it may lead to consultation with a retinal specialist, surgery, combination therapy, or prolonged therapy.48
Positive findings on the abdominal ultrasound were unusual. Hepatosplenic lesions in adults are often managed with prolonged therapy.49 Imaging of the kidneys may be useful not only for decisions regarding length of therapy (and surgical decisions if urine outflow is completely obstructed) but also to evaluate the anatomy. We suggest obtaining such imaging sometime during the course of therapy if the clinician predetermines that length of therapy will be influenced by the result, or if mechanical obstruction is a concern.50
Fungal endocarditis is an indication for surgical intervention when anatomic and clinical conditions permit. Avoiding surgery in the premature neonate with endocarditis is (aside from anatomic concerns) supported by the pathophysiology—neonatal endocarditis is generally associated with central catheter blood stream infection and is therefore usually “right-sided.” When surgical intervention is not pursued in cases of fungal endocarditis, combination antifungal therapy with prolonged treatment and long-term suppression is generally recommended.51 These recommendations for fungal infective endocarditis are based on clinical series, animal models, and data applied from bacterial infective endocarditis. Fungal endocarditis (like other forms of end-organ invasion) is possible when the patient has only 1 positive blood culture.12,13 Because findings on echocardiogram can so dramatically affect care, we believe that an echocardiogram after candidemia should be considered.
The review that we present may be incomplete because not every available database was searched and authors were not contacted for unpublished data. For example, in a multicenter prospective cohort study,51 there were 35 cases of candidemia, yet end-organ damage was not reported, and we did not contact the authors for data regarding prevalence of end-organ damage. Nonetheless, there was no evidence of publication bias that would indicate a study selection bias and biased findings, and neither prospective study design nor multicenter study design was associated with increase POR of end-organ damage.
Different methodologies (retrospective cohort, case control, randomized trials, prospective observational studies, etc) each have their own limitations and drawbacks. When an investigator attempts to combine data from several different sources and apply those data to the clinical arena, the conclusions may be substantially flawed. This is especially a concern when the investigator suggests a precise estimate for the therapy or outcome under review. In this review, we do not claim to provide a uniform or precise estimate for any of the outcomes under study, but we believe that the prevalence of end-organ damage, based on the best available data in the medical literature, is >0 and is high enough to warrant testing.
Ultimately, however, the question of end-organ damage prevalence and its evaluation will be answered by a multicenter prospective trial design. This analysis is a logical step between single-center retrospective studies that dominate the neonatal candidiasis literature and a future prospective collaborative effort.
Dr Benjamin received support from the National Institute of Child Health and Human Development (R03HD42940-01).
- Received September 25, 2002.
- Accepted February 27, 2003.
- Address correspondence to Daniel K. Benjamin, Jr, MD, MPH, PhD, Duke Clinical Research Institute, Box 17969, Durham, NC 27715. E-mail:
- ↵Stoll BJ, Hansen N, Fanaroff AA, et al. Late-onset sepsis in very low birth weight neonates: the experience of the NICHD Neonatal Research Network. Pediatrics.2002;110 :285– 291
- Baley JE, Kliegman RM, Fanaroff AA. Disseminated fungal infections in very low-birth-weight infants: clinical manifestations and epidemiology. Pediatrics.1984;73 :144– 152
- Baley JE, Silverman RA. Systemic candidiasis: cutaneous manifestations in low birth weight infants. Pediatrics.1988;82 :211– 215
- ↵Benjamin DK Jr, Ross K, McKinney RE Jr, Benjamin DK, Auten R, Fisher RG. When to suspect fungal infection in neonates: a clinical comparison of Candida albicans and Candida parapsilosis fungemia with coagulase-negative staphylococcal bacteremia. Pediatrics.2000;106 :712– 718
- Johnson DE, Thompson TR, Green TP, Ferrieri P. Systemic candidiasis in very low-birth-weight infants (less than 1,500 grams). Pediatrics.1984;73 :138– 143
- ↵Mittal M, Dhanireddy R, Higgins RD. Candida sepsis and association with retinopathy of prematurity. Pediatrics.1998;101 :654– 657
- Sigmundsdottir G, Christensson B, Bjorklund LJ, Hakansson K, Pehrson C, Larsson L. Urine D-arabinitol/L-arabinitol ratio in diagnosis of invasive candidiasis in newborn infants. J Clin Microbiol.2000;38 :3039– 3042
- Smith H, Congdon P. Neonatal systemic candidiasis. Arch Dis Child.1985;60 :365– 369
- ↵Makhoul IR, Kassis I, Smolkin T, Tamir A, Sujov P. Review of 49 neonates with acquired fungal sepsis: further characterization. Pediatrics.2001;107 :61– 66
- ↵Egger M, Schneider M, Smith GD. Spurious precision? Meta-analysis of observational studies. Br Med J.1998;316 :140– 144
- ↵Fleiss JL. Statistical Methods for Rates and Proportions. 2nd ed. New York, NY: John Wiley & Sons; 1981:64
- ↵Karlowicz MG, Giannone PJ, Pestian J, Morrow AL, Shults J. Does candidemia predict threshold retinopathy of prematurity in extremely low birth weight (≤1000 g) neonates? Pediatrics.2000;105 :1036– 1040
- Copyright © 2003 by the American Academy of Pediatrics