Utilization of Intensive Care Interventions in Critically Ill Patients With Candidemia Versus Bacteremia: A Multicenter Retrospective Cohort Study

Abstract

Objectives:

Candida species are one of the most common causes of ICU-onset bloodstream infection (BSI). Yet, there is no robust guidance on when to initiate empiric antifungals for ICU patients suspected of BSI. We compared patients with ICU-onset candidemia vs. ICU-onset bacteremia to determine which patients may benefit from empiric antifungals.

Design:

We compared characteristics between patients with bacteremia vs. candidemia at time of culture. We determined risk factors for death and constructed a multivariable regression to determine if candidemia is an independent risk factor for death.

Setting/Patients:

We included ICU patients from a hospital system in Houston, Texas (n = 8 hospitals) with a BSI from 2016 to 2023 and validated our findings using ICU patients with a BSI in the publicly available Medical Information Mart for Intensive Care (MIMIC)-IV cohort.

Interventions:

None.

Measurements and Main Results:

Of 1509 total patients in our primary cohort, 290 (19.2%) had candidemia and 1219 (80.8%) had bacteremia. Patients with candidemia were more likely to be on invasive mechanical ventilation (72.8% vs. 52.8%; p < 0.001), vasopressors (39.3% vs. 24.0%; p < 0.001), and continuous renal replacement therapy (15.2% vs. 9.4%; p = 0.006) at the time of culture. They contracted infection later in ICU stay than bacteremia patients (7.2 vs. 5.0 d; p < 0.001) and were more likely to die within 30 days of culture (unadjusted odds ratio [OR], 1.62; 95% CI, 1.25–2.09). After adjusting for ICU interventions (invasive mechanical ventilation, vasopressors, and continuous renal replacement therapy) and baseline parameters, candidemia was not independently associated with mortality compared with bacteremia in our primary cohort (OR, 1.21; 95% CI, 0.92–1.60) but was in the MIMIC-IV cohort (OR, 1.48; 95% CI, 1.003–2.17).

Conclusions:

Patients with significant ICU resource utilization are at increased risk for candidemia. Our data suggest that when initiating empiric antibiotics in patients requiring high-resource ICU care, empiric antifungal therapy should be considered.

KEY POINTS.

Question: We compared patients with ICU-onset candidemia vs. bacteremia to better understand which patients might benefit from empiric antifungals in addition to empiric antibacterials.

Findings: Patients with candidemia were more likely to be on invasive mechanical ventilation, vasopressors, and continuous renal replacement therapy at the time of positive culture than patients with bacteremia. They contracted infection later in ICU stay and were more likely to die within 30 days, although candidemia was not independently associated with mortality in our primary cohort.

Meanings: These findings suggest that clinicians should consider initiating empiric antifungals in addition to antibacterials in patients requiring high-resource ICU care.

Candida species are one of the most common causes of bloodstream infections (BSIs) in the United States (1), and Candida BSI (candidemia) is associated with significant mortality, morbidity, and financial burden (2–5). Candidemia is particularly prevalent in critically ill patients. Indeed, in a 2022 nationwide study of BSIs, Candida species were the second most common cause of ICU-onset BSI (6). Additionally, outcomes among ICU patients with candidemia are poor, with a 30-day mortality rate between 40% and 50% (7–10). These challenges in treating candidemia underscore the importance of understanding clinical characteristics in critically ill patients who acquire candidemia.

Among patients with suspected sepsis, guidelines strongly recommend that antibacterials be administered promptly, before the causative organism is known (11). However, despite the mortality risk associated with candidemia, the 2021 Surviving Sepsis Campaign guidelines offered only a weak recommendation for antifungal initiation in patients at high risk of candidemia (11). This is because it is difficult to know which patients have candidemia before blood cultures result, which may take days (12). For example, the 2016 Empirical Antifungal Treatment in ICUs (EMPIRICUS) trial found no survival benefit for antifungal administration in patients with ICU-acquired sepsis and potential candidemia (13). Additionally, there have been several clinical scoring systems developed to help clinicians predict which ICU patients are at risk of candidemia or invasive infection, yet these models are not widely adopted by clinicians across the United States (14–17). Overall, there is no clinical standard for which patients should receive empiric antifungals because there is no clear method to select high-risk patients before definitive diagnosis. However, if we could better identify patients at high risk of candidemia, empiric antifungals for suspected BSI may prove beneficial within that population.

Previous studies characterizing risk factors for candidemia often compared patients with Candida infection to Candida-free controls (2, 18–26). However, due to the potential similarities in clinical presentation between patients who develop candidemia and bacteremia, comparing candidemia patients to bacteremia patients may provide more granular insight into the risk factors for development of the former among the critically ill. Few studies have compared risk factors for candidemia vs. bacteremia (27–29), particularly in the ICU setting (30–32). We therefore compared candidemia with bacteremia in ICU patients to identify key differences and determine which patients are most at risk of candidemia and therefore most likely to benefit from the addition of empiric antifungals to standard antibacterial therapy.

MATERIALS AND METHODS

Patient Population

This study was approved by the Houston Methodist Institutional Review Board (IRB PRO No. 00037862, “Sepsis and Bacteremia Study,” approved February 16, 2024) with a waiver of consent. All study procedures were followed in accordance with the Houston Methodist IRB and the Helsinki Declaration of 1975. We included datasets from two different patient populations. The first cohort included patients at a large hospital system in Houston, TX, with a BSI that occurred either during ICU stay or within 6 hours before ICU admission between June 1, 2016, and June 17, 2023. The second cohort was extracted from Medical Information Mart for Intensive Care (MIMIC)-IV, Version 2.2, a publicly available, de-identified dataset that includes patients admitted to the ICU or emergency department at the Beth Israel Deaconess Medical Center (Boston, MA) between 2008 and 2022 (33). From MIMIC-IV, we included only ICU patients with a BSI. We removed patients with greater than 40% of variables missing and imputed values for the remainder of the patients using the multiple imputation chain equation method.

Data Abstraction

We abstracted demographic and clinical variables, including invasive mechanical ventilation (IMV) status, vasopressor receipt, and continuous renal replacement therapy (CRRT) status. These variables were assessed in the 48 hours before blood culture collection.

Definitions

Among ICU patients who developed BSI during ICU admission, patients who developed candidemia were included in the Candida group, and patients who developed bacteremia were included in the bacteria group. Non-Candida fungal BSIs were excluded. Skin commensal organisms, as defined by the U.S. Centers for Disease Control and Prevention, were excluded from analysis (34). If patients had multiple BSIs during ICU stay, only the first was considered in our analysis. Polymicrobial BSIs were excluded if the culture grew both Candida and bacteria. ICU interventions were defined as IMV, vasopressors, and CRRT.

Statistical Analysis

Data were analyzed using R, Version 2024.04.01 (Posit Software, PBC, Boston, MA). We compared patients with Candida vs. bacterial BSI. Separately, to determine risk factors for mortality we compared patients who survived vs. those who died at 30 days in the hospital. Initial univariable analyses for statistical associations were conducted using Pearson chi-square test for categorical variables, Student t test for continuous variables, and Mann-Whitney U test for ordinal variables. Statistical significance was set at p = 0.05. We used univariable logistic regression to determine odds ratios (ORs) for 30-day mortality. We created a multivariable logistic regression to determine whether candidemia was associated with increased odds of 30-day mortality compared with bacteremia, after adjusting for key baseline characteristics selected a priori (age, race, gender, Elixhauser Comorbidity Index [ECI]) and ICU interventions. Post hoc sample size calculations using the “pwr” package in R (Champely S. 2023. pwr: Basic Functions for Power Analysis. R package version 1.3-0; https://github.com/heliosdrm/pwr) showed our study had a power of 95.5%.

RESULTS

Study Population

In our primary dataset, we captured 26,163 unique bloodstream cultures (Fig. 1), of which 2,738 (10.5%) were from ICU. After exclusions (see above), we included 1509 patients in our primary cohort. The mean age was 64.4 years (± 14.4 yr), and 42.0% of patients were female (Table 1). Six hundred fifty-one patients died (43.1%) within 30 days of BSI onset. The MIMIC-IV dataset captured 3659 unique BSIs (Supplemental Fig. 1, https://links.lww.com/CCM/H839), and we ultimately included 730 patients in our MIMIC cohort; patient characteristics are described in Supplemental Table 1 (https://links.lww.com/CCM/H839).

Figure 1.

Total number of bloodstream infections (BSIs) collected 2016–2023, inclusion and exclusion criteria, and ultimate number of ICU BSIs included in analysis, primary cohort. ID = identification.

TABLE 1.

Patient Demographics and Comparison Between Bacteremia and Candidemia Patients, Primary Cohort

Patient Characteristic	Total (n = 1509)	Bacteria (n = 1219)	Candida (n = 290)	p
Age, yr, mean (sd)	64.4 (14.4)	64.0 (14.5)	65.8 (13.9)	0.05
Gender, n (%)
Female	634 (42.0)	513 (42.1)	121 (41.7)	0.97
Male	875 (58.0)	706 (57.9)	169 (58.3)
Race, n (%)
White	998 (66.1)	823 (67.5)	175 (60.3)	0.03
Asian	83 (5.5)	60 (4.9)	23 (7.9)
Black	355 (23.5)	274 (22.5)	81 (27.9)
Other/unknown	73 (4.8)	62 (5.1)	11 (3.8)
Hispanic or Latino, n (%)	329 (21.8)	270 (22.1)	59 (20.3)	0.70
Body mass index, kg/m2, mean (sd)	29.1 (8.6)	29.0 (8.5)	29.2 (8.8)	0.80
30-d mortality, n (%)	651 (43.1)	498 (40.9)	153 (52.8)	< 0.001
Invasive mechanical ventilation, n (%)	855 (56.7)	644 (52.8)	211 (72.8)	< 0.001
Vasopressor, n (%)	406 (26.9)	292 (24.0)	114 (39.3)	< 0.001
Renal replacement therapy, n (%)	467 (30.9)	349 (28.6)	118 (40.7)	< 0.001
Continuous renal replacement therapy, n (%)	158 (10.5)	114 (9.4)	44 (15.2)	0.006
Hemodialysis, n (%)	375 (24.9)	285 (23.4)	90 (31.0)	0.009
Sequential Organ Failure Assessment score, median (IQR)	8 (5–11)	8 (5–11)	9 (6–12)	< 0.001
Elixhauser Comorbidity Index, median (IQR)	15 (6–26)	15 (6–26)	16 (6–25)	0.53
Glasgow Coma Scale, median (IQR)	15 (10–15)	15 (11–15)	11 (8–15)	< 0.001
Blood urea nitrogen, mg/dL, mean (sd)	49.1 (32.9)	47.1 (31.7)	57.2 (36.5)	< 0.001
Creatinine, mg/dL, mean (sd)	2.3 (2.1)	2.2 (2.1)	2.5 (1.9)	0.06
Lactic acid, mmol/L, mean (sd)	4.3 (4.1)	4.3 (4.0)	4.6 (4.5)	0.18
WBC, k/μL, mean (sd)	16.1 (16.3)	15.4 (16.9)	18.9 (13.3)	< 0.001
Hemoglobin, g/dL, mean (sd)	10.1 (2.2)	10.2 (2.3)	9.7 (2.0)	< 0.001
Platelet, k/μL, mean (sd)	192 (126)	191 (125)	196 (130)	0.59
Uncomplicated DM, n (%)	669 (44.3)	536 (44.0)	133 (45.9)	0.61
Complicated DM, n (%)	505 (33.5)	406 (33.3)	99 (34.1)	0.85

DM = diabetes mellitus, IQR = interquartile range.

Candidemia vs. Bacteremia: Patient Characteristics

Of 1509 patients in our primary cohort, 290 (19.2%) had candidemia and 1219 (80.8%) had bacteremia (Table 1). Patients with candidemia were older than those with bacteremia (65.8 ± 13.9 vs. 64.0 ± 14.5 yr; p = 0.05). There were more Black (27.9% vs. 22.5%) and Asian (7.9% vs. 4.9%) patients in the candidemia group and relatively fewer White patients (60.3% vs. 67.5%; p = 0.03). There was no difference in gender (41.7% vs. 42.1% female; p = 0.97) or median ECI (16 vs. 15; p = 0.53) between groups. Patients with candidemia were more acutely ill at time of culture, as evidenced by lower median Glasgow Coma Scale (GCS; 11 vs. 15; p < 0.001) and mean hemoglobin (9.7 vs. 10.2; p < 0.001), in addition to higher median Sequential Organ Failure Assessment (SOFA) score (9 vs. 8; p < 0.001) and mean WBC count (18.9 vs. 15.4; p < 0.001) than bacteremia patients. There was a trend toward higher mean lactic acid among candidemia patients (4.6 vs. 4.3; p = 0.18).

We then examined differences in intensive care interventions between groups. At time of culture, candidemia patients were more likely to be on IMV (72.8% vs. 52.8%; p < 0.001), receiving vasopressors (39.3% vs. 24.0%; p < 0.001), or receiving CRRT (15.2% vs. 9.4%; p = 0.006) than patients with bacteremia. When ICU interventions (IMV, vasopressor use, and CRRT) were totaled, more candidemia patients were on greater than or equal to 2 ICU interventions than bacteremia patients (36.9% vs. 21.2%; p < 0.001) (Fig. 2). Patients developed candidemia later in their ICU stay compared with bacteremia patients (median, 2.8 d [interquartile range (IQR), 0.1–8.8 d] vs. 0.8 d [IQR, 0.06–6.5 d]; p < 0.001) (Fig. 3). To investigate whether the difference in ICU intervention utilization between candidemia and bacteremia patients was explained primarily by BSI timing, we divided our cohort into BSI less than 48 hours from ICU admission (“early”) and after 48 hours (“late”). When the cohort was split into early and late groups, patients with candidemia in each group were overall still more acutely ill and utilizing more ICU interventions (Supplemental Tables 3 and 4, https://links.lww.com/CCM/H839), indicating that timing of BSI onset did not modify the relationship between ICU interventions and BSI type.

Figure 2.

Distribution of ICU interventions grouped by bloodstream infection (BSI) type, primary cohort. ICU interventions include invasive mechanical ventilation, administration or one or more vasopressor, and continuous renal replacement therapy.

Figure 3.

Violin plot depicting days from ICU admission to culture, primary cohort. ICU admission = day 0.

Of the 730 patients in the MIMIC-IV cohort, 165 (22.6%) had candidemia and 565 (77.4%) had bacteremia (Supplemental Table 1, https://links.lww.com/CCM/H839). Like the primary cohort, candidemia patients from MIMIC had clinical scores and laboratory values reflective of a higher acuity of illness than bacteremia patients. Additionally, in this cohort, candidemia patients had a higher mean lactic acid (3.5 vs. 2.8; p = 0.02). At time of culture, candidemia patients were more likely than bacteremia patients to be on IMV (54.5% vs. 43.5%; p = 0.02), vasopressors (30.9% vs. 18.1%; p < 0.001), or CRRT (21.8% vs. 9.6%; p < 0.001) and were more likely to be on greater than or equal to 2 interventions (35.6% vs. 16.5%; p < 0.001; [Supplemental Fig. 2, https://links.lww.com/CCM/H839]). Patients developed candidemia later than bacteremia patients (median, 7.2 d [IQR, 2.9–18.3 d] vs. median 5.0 d [IQR, 1.3–12.5 d]; p = 0.001; [Supplemental Fig. 3, https://links.lww.com/CCM/H839]). As with the primary cohort, patients with candidemia were overall more acutely ill even when divided into early and late time groups (Supplemental Tables 5 and 6, https://links.lww.com/CCM/H839).

Outcomes

Patients with candidemia were more likely to die: 52.8% (153/290) of candidemia patients in the primary cohort died within 30 days compared with 40.9% (498/1219) of bacteremia patients (p < 0.001) (Fig. 4). Among patients who were not mechanically ventilated at the time of culture, candidemia patients were more likely to be mechanically ventilated after BSI onset (OR, 1.80; 95% CI, 1.12–2.92). There was no difference between the two groups in ultimate need for CRRT after BSI (OR, 1.12; 95% CI, 0.80–1.54).

Figure 4.

Kaplan-Meier survival curve of 30-d mortality for candidemia and bacteremia patients, primary cohort. Day 0 represents the date of positive culture.

In our MIMIC cohort, 50.9% (84/165) of candidemia patients died within 30 days, while 35.4% (200/565) of bacteremia patients died in that period (p < 0.001; [Supplemental Fig. 4, https://links.lww.com/CCM/H839]). There was no difference in ultimate need for mechanical ventilation (OR, 1.28; 95% CI, 0.77–2.12) or CRRT (OR, 1.02; 95% CI, 0.56–1.78) between the two groups.

Risk Factors for Mortality

In univariable logistic regression, candidemia was associated with an increased risk of 30-day mortality compared with bacteremia (OR, 1.62; 95% CI, 1.25–2.09) (Table 2). Within our primary cohort, many of the clinical factors associated with contracting candidemia were also associated with 30-day mortality. IMV (OR, 2.47; 95% CI, 2.00–3.07), vasopressor use (OR, 2.73; 95% CI, 2.16–3.45), and CRRT (OR, 2.36; 95% CI, 1.68–3.32) were all associated with increased probability of death in a univariable logistic regression. Elevated lactic acid (OR, 1.20; 95% CI, 1.16–1.24) was also associated with increased probability of 30-day mortality, as was lower GCS (OR, 0.89; 95% CI, 0.87–0.92) and higher SOFA score (OR, 1.28; 95% CI, 1.24–1.32). After adjusting for baseline characteristics and ICU interventions of interest (as a marker of acuity of illness), candidemia was associated with a nonstatistically significant trend toward an increased risk of death compared with bacteremia (adjusted OR [aOR], 1.21; 95% CI, 0.92–1.60) in our primary cohort. Increased age (aOR, 1.02; 95% CI, 1.01–1.03), IMV (aOR, 2.10; 95% CI, 1.67–2.64), vasopressor (aOR, 1.98; 95% CI, 1.52–2.57), and CRRT (aOR, 1.59; 95% CI, 1.09–2.33) were independent predictors of 30-day mortality; full model results are shown in Table 2.

TABLE 2.

Risk Factors for 30-Day Mortality Among 1509 Patients With Bloodstream Infection, Primary Cohort

Patient Characteristic	Unadjusted OR (95% CI)	Adjusted OR (95% CI)
Bloodstream infection type
Bacteria	Reference	Reference
Candida	1.62 (1.25–2.09)	1.21 (0.92–1.60)
Age	1.02 (1.01–1.03)	1.02 (1.01–1.03)
Race
White	Reference	Reference
Asian	1.38 (0.88–2.17)	1.37 (0.85–2.22)
Black	0.95 (0.74–1.22)	1.03 (0.79–1.34)
Other/unknown	1.39 (0.86–2.23)	1.45 (0.88–2.34)
Ethnicity
Non-Hispanic	Reference
Hispanic	1.31 (1.02–1.67)
Gender
Male	Reference	Reference
Female	1.09 (0.88–1.34)	1.21 (0.97–1.51)
Body mass index	1.02 (1.01–1.03)
Invasive mechanical ventilation	2.47 (2.00–3.07)	2.10 (1.67–2.64)
Vasopressor	2.73 (2.16–3.45)	1.98 (1.52–2.57)
Renal replacement therapy	1.31 (1.05–1.63)
Continuous renal replacement therapy	2.36 (1.68–3.32)	1.59 (1.09–2.33)
Hemodialysis	1.08 (0.85–1.36)
Sequential Organ Failure Assessment score	1.28 (1.24–1.32)
Elixhauser Comorbidity Index	0.99 (0.99–1.00)	0.99 (0.99–1.00)
Glasgow Coma Scale	0.89 (0.87–0.92)
Blood urea nitrogen, mg/dL	1.01 (1.01–1.01)
Creatinine, mg/dL	1.04 (0.99–1.09)
Lactic acid, mmol/L	1.20 (1.16–1.24)
WBC, k/μL	1.00 (1.00–1.01)
Hemoglobin, g/dL	0.99 (0.95–1.04)
Platelet, k/μL	1.00 (1.00–1.00)
Uncomplicated DM	0.99 (0.81–1.22)
Complicated DM	0.97 (0.78–1.20)

DM = diabetes mellitus, OR = odds ratio.

There were similar risk factors for death in MIMIC-IV, including candidemia (OR, 1.89; 95% CI, 1.33–2.69; [Supplemental Table 2, https://links.lww.com/CCM/H839]). In this cohort, after adjusting for the same covariates, candidemia was associated with mortality compared with bacteremia (OR, 1.48; 95% CI, 1.003–2.17). Increased age (aOR, 1.02; 95% CI, 1.01–1.03), IMV (aOR, 1.98; 95% CI, 1.42–2.75), vasopressor (aOR, 2.45; 95% CI, 1.64–3.67), and ECI (aOR, 1.04; 95% CI, 1.02–1.06) were also independent predictors of 30-day mortality; full results are shown in Supplemental Table 2 (https://links.lww.com/CCM/H839).

DISCUSSION

In both cohorts, a high proportion of ICU-onset BSIs (approximately 20%) were candidemia. Compared with bacteremia, patients who developed candidemia were more likely to be receiving ICU interventions (regardless of early vs. late infection), developed BSI later in ICU stay, and had higher 30-day mortality. Our findings highlight that candidemia is common in ICUs, has a high prevalence among those utilizing ICU interventions, and is associated with poor outcomes.

Our findings on the prevalence and associated mortality rate of candidemia are consistent with other similar studies. In our primary cohort, 19.2% of BSIs were due to Candida species. This number broadly aligns with a 2022 study of approximately 150,000 ICU patients, in which 16% of ICU-onset BSIs were due to Candida species (2). Additionally, the candidemia 30-day mortality rate was 52.8% and 50.9% in the primary and MIMIC cohorts, respectively. Other studies focused on ICU patients observed similarly high 30-day mortality rates from 40% to almost 50%, although our two cohorts were on the higher end of that range (7–10).

In both cohorts, candidemia patients were more likely than bacteremia patients to be receiving ICU interventions at time of positive culture: they were more likely to be receiving IMV, vasopressors, or CRRT. Typical risk factors for candidemia include Candida colonization, central venous catheter (CVC) placement, use of total parenteral nutrition (TPN), or recent abdominal surgery (2, 22, 35). However, we observed a clear association between IMV, vasopressors, and CRRT status and development of candidemia. Given the heightened risk of candidemia among those receiving ICU interventions, these patient populations might benefit from receiving empiric antifungals in addition to antibiotics if BSI is suspected. Our findings align with the “Candida” score developed in 2006: in a comparison of patients colonized with Candida vs. those with invasive candidiasis, clinical symptoms of severe sepsis, including hypotension, could help predict Candida infection (17, 36). Besides hypotension, which was included in the “Candida” score, ICU interventions are largely absent from other clinical prediction tools (14–16), suggesting that these risk factors may be overlooked when clinicians are deciding to initiate antifungal therapy.

Multivariable analyses showed differing results regarding candidemia’s impact on mortality: an association was found in the MIMIC cohort but not in the primary cohort. Existing data suggest that candidemia does have an independent effect on mortality, with recent attributable mortality estimated at 18–27% (2, 26), and older studies identifying candidemia as an independent risk factor for mortality (37, 38). Although candidemia was not associated with mortality in our primary cohort in multivariable analysis, we saw a trend toward increased mortality among the candidemia group. Further, the fact that candidemia and bacteremia mortality hazards did not separate until day 3 suggests some independent effect (Fig. 4). In observational studies, delayed hazard separation may indicate that the independent variable directly influenced the outcome, whereas distinct hazards from day 0 indicates that the difference is more likely related to preexisting conditions (39).

Our study differs from much of the previous work investigating risk factors for candidemia because we compared ICU patients who contracted candidemia to those who contracted bacteremia, not those who remained Candida-free (2, 18). Few other studies in the past decade have compared candidemia patients to bacteremia patients, and most of those studies were comparing different patient populations than our own—such as internal medicine ward or transplant patients—or were investigating the use of various biomarkers as diagnostic tools (27–30). Still, one study of transplant recipients in 2023 found that candidemia patients were more likely to receive CRRT (27), and another from 2017 found that candidemia patients contracted infection later than bacteremia patients (30).

The Extended Prevalence of Infection in ICU Study (EPIC II) and Epidemiology and Outcome of Hospital-acquired Bacteremia (EUROBACT) studies, conducted in 2007 and 2010–2011, respectively, are most similar to the present study regarding patient populations and outcomes measured. Both were multisite, international prospective studies comparing candidemia to bacteremia patients in the ICU. Both EPIC II and EUROBACT had fewer cases of monomicrobial candidemia than our study (61 and 93) (31, 32). Neither found a difference in mortality rates among ICU patients with candidemia vs. bacteremia, nor were ICU patients with candidemia more likely to be on IMV or require vasopressors (EUROBACT captured hypotension). Additionally, candidemia patients were not more acutely ill than their bacteremia counterparts (31, 32). In EUROBACT, candidemia patients were more often in renal failure at the time of diagnosis, although CRRT status was unspecified (32). These contrasts to our results—no difference in mortality, IMV status, or hypotension—suggest the epidemiology of candidemia may be shifting and becoming more distinct from bacteremia, with the infection burden residing increasingly with the sickest patients in ICU. The shift toward increased ICU intervention utilization among candidemia patients is in line with current research on the changing epidemiology of candidemia. Indeed, patients who contract candidemia are becoming increasingly older and more acutely ill (40).

Our study has several limitations. First, candidemia is difficult to diagnose. Although blood culture remains the gold standard in diagnosis, cultures may miss invasive candidiasis up to 50% of the time (41). This limitation certainly affects our study, but this is a challenge in candidemia diagnosis and research more broadly. The primary analysis was conducted on a cohort from a single health system. Despite this, the system includes both a quaternary referral hospital and several community hospitals, and our inclusion of patients from MIMIC-IV increases generalizability. We did not perform a priori sample size analysis, but post hoc calculations showed a power of 95.5%. Our dataset did not include risk factors for candidemia mentioned above: colonization, CVC, TPN, or abdominal surgery (2, 22, 35). While these more “traditional” candidemia risk factors may further refine a clinician’s pretest probability of candidemia, our findings suggest simply that the presence and number of ICU interventions are important for clinicians to consider. Also, while IMV, CRRT, and vasopressor usage are not considered “traditional” risk factors for candidemia, other research has shown that these interventions do increase the risk of candidemia (42). However, our study does add novelty to the field by comparing the utilization of these ICU interventions to bacteremia patients. Additionally, the association between candidemia and mortality may be confounded. Although we attempted to adjust for confounders with multivariable analysis, there may be additional unmeasured confounders. Last, we did not include data on reason for ICU admission, ICU type, prior antibiotic use, or empiric treatment of suspected BSI. It is possible that the higher unadjusted mortality among patients with candidemia is due at least in part to less use of empiric antifungals than empiric antibacterial agents, and future research should examine empiric antifungal use and its impact on mortality.

CONCLUSIONS

The sickest patients in the ICU—those receiving IMV, vasopressors, and CRRT—are at a markedly higher risk of contracting candidemia. Although our study is unable to make specific statements on how clinicians should empirically prescribe antifungals, it does suggest that clinicians strongly consider starting empiric antifungals in addition to antibiotics in high-ICU utilizers with suspected BSI. Future research should focus on prevention and early diagnosis of candidemia among the sickest ICU patients.

ACKNOWLEDGMENTS

We thank Joanne Park at Houston Methodist for figure creation and editing.

Supplementary Material

ccm-54-246-s001.docx ^{(498.6KB, docx)}