During January 2017–December 2018, EIP conducted candidemia surveillance in specific counties in 9 US states: California, Colorado, Georgia, Maryland, Minnesota, New Mexico, New York, Oregon, and Tennessee. EIP candidemia surveillance methods have been previously described [6]. Surveillance personnel used standardized case report forms to collect demographic and clinical data from medical and laboratory records.

A case of candidemia was defined as a blood culture positive for a Candida species collected from a resident of the surveillance area. Any blood cultures positive for a Candida species within 30 days of the initial positive culture from the same patient were considered part of the same case, including if more than 1 Candida species was detected. A positive Candida culture after the 30-day period was considered a new case in the same patient.

All isolates from initial blood cultures positive for Candida spp. are speciated by inpatient and outpatient clinical, reference, or commercial laboratories serving the surveillance population as part of routine practices. Some laboratories also conducted AFST on certain isolates depending on laboratory testing and clinician ordering practices. We restricted our analysis to the index candidemia case (defined as the first case per patient within the analytic period) for adult patients (aged ≥18 years) during 2017–2018, using species identification and AFST results provided by the laboratories.

Our analysis of candidemia treatment practices at 9 US surveillance sites found that most adult patients received an echinocandin rather than fluconazole as initial treatment for candidemia. However, a substantial proportion (29.6%) of patients initially received fluconazole, which resulted in potentially inappropriate treatment. More than one-half of patients started on fluconazole were later found to grow a non-albicans Candida species on initial blood cultures, and >10% of Candida isolates from patients who received fluconazole and whose specimens were tested for fluconazole susceptibility were resistant. Geographic area, but not diabetes, malignancy, and recent exposure to fluconazole (known risk factors for fluconazole resistance [13, 20]), was one of the strongest predictors of initial drug choice. This finding suggests that regional practices play a stronger role in initial candidemia therapy than patient-level factors and that clinician education regarding risk factors for drug-resistant candidemia may be needed.

Cirrhosis was the only medical condition significantly predictive of echinocandin over fluconazole as initial treatment. This might be due to several factors, including concerns for potential fluconazole-associated hepatotoxicity [21] or fungal infections as an emerging concern in patients with cirrhosis given the particularly poor outcomes of candidemia in this population [22, 23]. Patients with a history of recent hospitalization were also more likely to receive fluconazole than an echinocandin as initial candidemia treatment. Given these unexpected findings, further research is needed regarding how patient-level factors might influence clinician antifungal choices.

Geographic location was among the few independent predictors of whether a patient received an echinocandin as initial treatment, with patients at certain surveillance sites having over twice the odds of initially receiving an echinocandin compared with other sites. Differences in antifungal selection among sites are likely due to a combination of factors, including the local epidemiology of Candida spp. distributions, risk factors of the local population, antifungal susceptibility patterns, facility type, and clinician awareness and appropriate implementation of IDSA guidelines. Species and drug- resistance patterns are known to vary substantially by geography, facility, and even among units within the same facility [24, 25]. Therefore, variation in treatment patterns by surveillance site was not unexpected given the epidemiologic variety that likely results in a diversity of clinician practices across the United States; the impact of these factors on candidemia treatment decisions merits further research.

Approximately 25% of patients in our analysis first received antifungal treatment ≥3 days after the date of their first positive blood culture collection, which is consistent with previous research of candidemia treatment delays [26]. These possible treatment delays likely occurred because of the non-specific clinical presentation of candidemia, low sensitivity of blood cultures for candidemia detection, and the time needed to detect Candida in blood cultures [15, 27]. Furthermore, 1 in 4 patients received an antifungal drug on or during the 14 days before the specimen collection day of their first positive blood culture, suggesting that they may have received empiric therapy for suspected candidemia before Candida was detected. Although retrospective studies have suggested that empiric antifungal therapy might improve survival for select high-risk patients [28], a large, multicenter randomized control trial of empiric antifungal therapy in nonneutropenic critically ill patients did not find a significant difference in survival among patients treated empirically with micafungin versus placebo [29]. The appropriate role for empiric antifungal therapy for suspected candidemia remains controversial, and further investigation of this topic is needed [29]. Our findings underscore the continued need for the evaluation of culture-independent diagnostic tests to assist with earlier candidemia detection and treatment, as well as continued clinician education about candidemia in at-risk patients [15].

Our findings also suggest an opportunity for increased use of AFST because AFST results can be an important adjunct in clinical decision-making [30]. Although IDSA guidelines recommend fluconazole AFST for all bloodstream Candida isolates, nearly one-half of patients did not receive testing for fluconazole susceptibility. Compared with fluconazole AFST, echinocandin AFST was performed less frequently, which is expected because echinocandin AFST is not recommended for all bloodstream isolates [13]. However, contrary to IDSA guidelines, fewer than half of C glabrata or C parapsilosis isolates underwent echinocandin AFST, which is recommended because of concern for echinocandin resistance in these species. In our analysis, 39.1% of patients were switched from an echinocandin to fluconazole without receiving fluconazole AFST. Although AFST results are generally unavailable at the time treatment is initiated, the use of AFST, in conjunction with facility-specific antibiograms, may improve antifungal stewardship by facilitating the transition from echinocandins to fluconazole when appropriate [17, 31].

The proportion of patients with isolates that underwent AFST varied widely among surveillance sites. Although the capacity of laboratories in the United States to perform AFST has increased in recent years, the frequency of AFST performance and use in clinical decision-making might vary based on several factors, including cost, a lack of personnel trained in mycology, and reliance on off-site testing, which can substantially delay results (from ~3 days with on-site testing versus ~7 days for off-site testing) [14, 32]. Furthermore, a lack of reflexive AFST (ie, automatic testing versus requiring a clinician order) for Candida isolates, may further hinder timely receipt of appropriate treatment and the likelihood of clinicians to use AFST results for treatment decision-making. A recent analysis using National Healthcare Safety Network data found that reflexive AFST is lacking in approximately two-thirds of acute care hospitals [14]. As antifungal drug resistance continues to rise, efforts to increase local laboratory AFST capacity and integrate AFST into treatment practices are becoming critically important.

In our investigation, patients treated initially with an echinocandin were more likely to die during their candidemia-associated hospitalization than those treated with fluconazole. However, this finding is likely from confounding factors because critically ill patients are more likely to be treated with an echinocandin [33]. A randomized controlled trial, more robust for examining effects of treatment, found superior outcomes for patients who received an echinocandin as primary treatment for invasive candidiasis [12].

Our findings are subject to several limitations. Although EIP candidemia surveillance encompassed a source population of approximately 17.1 million, treatment practices observed in EIP surveillance sites may not reflect those throughout the country. EIP candidemia surveillance data are useful for analyzing broad national trends, but our analysis was limited by the lack of facility-level data needed to compare practices among facility types (eg, tertiary, academic, community hospitals), as well as a lack of data on setting of care (eg, ward, intensive care unit) and treating physician type (eg, infectious disease specialist) at the time of antifungal drug treatment initiation. Clinicians most likely choose candidemia treatments based on a combination of factors including local adoption of guidelines, drug cost and availability, ability to administer fluconazole orally, individual patient characteristics, facility type, and local epidemiology, which makes it difficult to generalize about treatment practices across the nation or to definitively state whether the treatment decisions were inappropriate. Further, EIP surveillance data lack information regarding why patients were switched from one drug class to another, how clinicians might use site-specific antibiograms or local candidemia species data, and information regarding the timing of when clinicians receive results of species testing and AFST. We noted that nearly one-half of patients started initially on fluconazole were then switched to an echinocandin, usually within several days; although the motivation driving this switch is unknown, we suspect that these changes could be due to treatment failure, receipt of AFST results, identification of additional species that warranted a new treatment plan, specialist consultation, or clinical worsening.

Nonetheless, our analysis of data from a large, population-based, and multisite candidemia surveillance system identified instances in which patients received potentially inappropriate candidemia treatment. Our findings highlight opportunities for closer adherence to IDSA guidelines on antifungal drug selection and AFST utilization, as well as the need for improved candidemia diagnostic tools. Further research regarding how clinicians choose antifungal drugs for candidemia and barriers to use of IDSA guidelines is also needed. Although treatment guidelines should never supplant clinician judgment when caring for individual patients, increased clinician education and stricter adherence to IDSA guidelines should be considered.