Creation and Assessment of a Clinical Predictive Model for Candidemia in Patients with Candiduria

Abstract

Background

Candidemia is the most common clinical presentation of invasive candidiasis and is a major cause of morbidity and mortality. Candiduria is a predictor for candidemia; however, patient characteristics that are associated with concurrent candidemia in the setting of candiduria are unclear. Identifying these characteristics could aid in the early detection of systemic disease.

Methods

We performed a retrospective cohort analysis of hospitalized patients with candiduria at our institution over a 13-year period. Our evaluation of patient characteristics included demographics, comorbidities, medications, procedures, devices, vital signs, and laboratory values. We developed a multivariable logistic model to identify factors associated with candidemia in patients with candiduria.

Results

We identified 4,240 patients with candiduria, 263 (6.2%) of whom had candidemia. Predictors for increased risk of candidemia with candiduria included hospitalizations greater than 12 days, central venous catheter, parenteral nutrition, hematological and gynecological malignancy, and receipt of β-lactam/β-lactamase inhibitors. Vital signs and laboratory values associated with candidemia included elevated heart rate, temperature, and creatinine, along with neutropenia and neutrophilia. Factors that demonstrated a decreased risk of candidemia included diabetes mellitus, gastrostomy, and urinary catheter with antibiotic use. The c-statistic was 0.741 (95% CI, 0.710 – 0.772).

Conclusion

We identified a set of clinical characteristics that can predict the presence of candidemia with candiduria.

Introduction

Candiduria is a frequent finding amongst hospitalized patients with a global incidence rate of 27.4 per 1, 000 patient admissions^{1, 2}. The presence of yeast in the urine commonly represents benign colonization of the urinary tract or perineum, but can also be a sentinel of a hematogenous infection^3–5. Candidemia, the most common form of invasive candidiasis, is associated with an attributable mortality rate of 38–49% and is responsible for prolonged hospital stays and high healthcare expenditure^6–9.

Colonization with Candida is an independent risk factor for nosocomial candidemia^{8, 10–12}. Researchers have constructed clinical prediction tools to assess Candida colonization and identify patients at risk for invasive disease, with the Candida colonization index (CI) being one of the most widely studied^{11, 13}. Later studies found candiduria to correlate to a CI ≥ 0.5, the threshold value that places patients at higher risk for infection versus colonization^{11, 14}. Yet, the decision to initiate or forgo treatment when Candida species is detected in the urine remains controversial³. In most patients, candiduria is not a precipitant of candidemia, but multivariate analyses have confirmed it to be a risk factor for invasive disease^{4, 5, 15}. Clinical characteristics that are associated with candidemia in the setting of established candiduria have not been fully explored^{5, 15}. Identification of these patient characteristics would help clinicians to differentiate patients with candiduria from those with concomitant disease and to determine an appropriate treatment course.

To accomplish this goal, we performed a retrospective cohort analysis of patients with candiduria over a 13-year period. Our goal was to identify relevant patient comorbidities that could accurately predict the presence of candidemia in the setting of candiduria.

Methods

Cohort Construction

We collected data from patients admitted between January 2002 and January 2015 to Barnes Jewish Hospital, a 1,323-bed tertiary care hospital in St. Louis, MO. We confirm that the ethical policies of the journal have been adhered to and that the Washington University in St. Louis Institutional Review Board approved the study with a waiver of consent. All hospitalized patients with candiduria, defined as the presence of candida or yeast in at least one urine culture, were included in the study. Patients with urine cultures positive for non-candida yeast were removed from the study.

Data extracted from medical records included Candida species, duration of hospitalization, demographics, vital signs, in-hospital procedures and surgeries, cancer, laboratory values (absolute neutrophils, absolute lymphocytes, white blood cells, alanine aminotransferase, aspartate aminotransferase, amylase, lipase, total and direct bilirubin, creatinine, platelets, hemoglobin) and medications ordered within 90 days leading up to a positive urine culture. We defined the heart rate and temperature as the maximum heart rate 48 hours prior to or 24 hours after a positive urine culture was collected, along with the maximum temperature 24 hours prior to or after a positive urine culture, respectively. We defined neutropenia as an absolute neutrophil count (ANC) less than 1,500/mm³and neutrophilia as an ANC greater than 9,500/mm³ based on our institution’s upper threshold value¹⁶.

We defined comorbidities based on ICD-9 codes and included Elixhauser comorbidities along with diabetes, organ transplants, and cancers as previously defined¹⁷. Additional factors evaluated for predisposition of candidemia included recent use of central venous catheter (CVC), recent use of ventilator, recent total parenteral nutrition (TPN), receipt of hemodialysis, HIV infection, burns, radiation therapy and chemotherapy. We combined urinary catheter and recent antibiotic use into one variable to evaluate for their synergistic effect on the development of candidemia. We categorized continuous variables and combined different levels within a variable to better model the underlying distribution.

Outcomes

We used the presence of candidemia as the primary outcome in both univariable and multivariable analyses. Candidemia was defined as having at least one positive blood culture for Candida drawn from any venous or arterial access site 48 hours prior to or after a positive urine culture. For patients with multiple episodes of candiduria or candidemia, only the first episode was incorporated in the database.

Statistical Analysis

We performed statistical analysis using SPSS V23 (IBM, Armonk, NY) and all tests were two-tailed with p ≤ 0.05 considered significant. For descriptive statistics, we used Chi-square or Fisher’s exact tests for categorical variables and Mann-Whitney U tests for continuous variables as the data were not normally distributed.

We performed univariable analysis to assess the risk of concomitant candiduria and candidemia in association with patient demographic and clinical characteristics. Criteria for evaluation in multivariable analysis included variables with p ≤ 0.2, and those with strong plausibility in previous literature, including but not limited to renal failure, solid malignancies, and bone marrow transplant^18–20. We developed the multivariable models in a parsimonious manner, adding candidate variables sequentially and retaining them in the model if they were found to be significant (p ≤ 0.05). After all relevant variables were included, those that were no longer found to be significant were sequentially removed from the model.

We then generated a c-statistic and receiver operating characteristic (ROC) curve using the final set of predictor variables.

Results

Demographics

During the study period, 4,878 hospitalized patients were diagnosed with candiduria. We excluded 638 cases from analysis due to incomplete data collection. Of the remaining 4,240 cases, 263 (6.2%) of them were positive for concurrent candidemia. A total of 3,698 (92.9%) of the 3,977 patients with candiduria alone had blood cultures drawn within 48 hours prior to or after a positive urine culture.

Demographic characteristics and relevant factors associated with concomitant candiduria and candidemia are outlined in Table 1 and Supplementary Table 1. Age and race were not associated with differences in the in the rate of candidemia among the study population. Female gender was associated with a lower risk of candidemia with candiduria compared to male gender (4.6% vs 8.0%; p < 0.001).

Table 1.

Select Baseline Characteristics of 4,240 Candiduric Patients by Presence of Candidemia, 2002–2015

Characteristica	Candidemia (N=263)	No Candidemia (N=3,977)	p-valueb	Total (N=4,240)
Demographics
Age, median (IQR)	64 (23)	64 (24)	0.486	64 (24)
Female sex	132 (50.2)	2,608 (65.6)	<0.001	2,740 (64.6)
Race			0.073
White	167 (63.5)	2,732 (68.7)		2,899 (68.4)
African American	83 (31.6)	1,054 (26.5)	0.073	1,137(26.8)
Other	13 (4.9)	191 (4.8)		204 (4.8)
Comorbidities
Hypertension	111 (42.2)	2,387 (60.0)	<0.001	2,498 (58.9)
Diabetes	67 (25.5)	1,460 (36.7)	0.004	1,527 (36.0)
Coronary artery disease	70 (26.6)	1,185 (29.8)	0.668	1,255 (29.6)
Chronic liver disease	30 (11.4)	404 (10.2)	0.310	434 (10.2)
Chronic kidney disease	57 (21.7)	921 (23.2)	0.993	978 (23.1)
Malignancy
Any	130 (49.4)	1,773 (44.6)	0.016	1,903 (44.9)
Leukemia	17 (6.5)	136 (3.4)	0.005	153 (3.6)
Hodgkin’s lymphoma	3 (1.1)	22 (0.6)	0.198	25 (0.6)
Non-Hodgkin’s lymphoma	19 (7.2)	135 (3.4)	<0.001	154 (3.6)
Hematologicc	35 (13.3)	259 (6.5)	<0.001	294 (6.9)
Gynecologicd	19 (7.2)	288 (7.2)	0.777	307 (7.2)
Other Potential
Predisposing Factors
Non-albicans species	20 (7.6)	174 (4.4)	0.040	194 (4.6)
C. glabrata	12 (4.6)	113 (2.8)		125 (2.9)
C. krusei	1 (0.4)	6 (0.2)		7 (0.2)
C. parapsilosis	4 (1.5)	19 (0.5)		23 (0.5)
C. tropicalis	2 (0.8)	27 (0.7)		29 (0.7)
Others	1 (0.4)	9 (0.2)		10 (0.2)
Hospital stay >12 days	124 (47.1)	1,218 (30.6)	<0.001	1,342 (31.7)
Bone marrow transplant	1 (0.4)	14 (0.4)	0.889	15 (0.4)
Cancer chemotherapy	8 (3.0)	80 (2.0)	0.191	88 (2.1)
Radiation therapy	6 (2.3)	60 (1.5)	0.258	66 (1.6)
TPN within prior 30 days	105 (39.9)	1,299 (32.7)	0.002	1,404 (33.1)
Gastrostomy	5 (1.9)	197 (5.0)	0.045	202 (4.8)
Devices
Ventilator	13 (4.9)	220 (5.5)	0.871	233 (5.5)
Permanent central line	109 (41.4)	1,155 (29.0)	<0.001	1,264 (29.8)
Temporary central line	114 (43.3)	1,022 (25.7)	<0.001	1,136 (26.8)
Dialysis catheter	31 (11.8)	258 (6.5)	0.002	289 (6.8)
Urinary catheter	16 (6.1)	349 (8.8)	0.224	365 (8.6)
Central venous cathetere	152 (57.8)	1,657 (41.7)	<0.001	1,809 (42.7)
Urinary catheter and antibiotic usef	15 (5.7)	276 (6.9)	0.616	291 (6.9)
Vital Signs
Temperature
>39 °C	52 (19.8)	355 (8.9)	<0.001	407 (9.6)
Heart rate >120 beats/min	121(46.0)	1,037 (26.1)	<0.001	1,158 (27.3)
Respiratory rate
20–30 breathes/min	49 (18.6)	947 (23.8)	0.225	996 (23.5)
>30 breathes/min	50 (19.0)	502 (12.6)	<0.001	552 (13.0)
Select Laboratory Values
Absolute neutrophils count
<1,500/mm3	23 (8.7)	162 (4.1)	<0.001	185 (4.4)
>9,500/mm3	82 (31.2)	1,025 (25.8)	0.002	1,107 (26.1)
Hemoglobin <9 g/dL	226 (85.9)	2,766 (69.5)	<0.001	2,992 (70.6)
Platelets <150 K/ mm3	155 (58.9)	1,703 (42.8)	<0.001	1,858 (43.8)
Creatinine >1.3 mg/dL	124 (47.1)	1,494 (37.6)	<0.001	1,618 (38.2)
Medications Ordered Within90 Days Prior to Candida BSI
Azole	19 (7.2)	268 (6.7)	0.552	287 (6.8)
Caspofungin	3 (1.1)	7 (0.2)	0.005	10 (0.2)
Corticosteroids	86 (32.7)	999 (25.1)	0.001	1,085 (25.6)
Anti-herpes antivirals	29 (11.0)	282 (7.1)	0.007	311 (7.3)
Carbapenems	58 (22.1)	585 (14.7)	<0.001	643 (15.2)
Cytotoxic agents	5 (1.9)	32 (0.8)	0.053	37 (0.9)
Glycopeptide antibiotics	131 (49.8)	1,714 (43.1)	0.003	1,845 (43.5)
Nitroimidazoles	84 (31.9)	97 (2.4)	<0.001	181 (4.3)
Penicillins	26 (9.9)	244 (6.1)	0.007	270 (6.4)
Penicillin/β-lactamase inhibitors	67 (25.5)	669 (16.8)	<0.001	736 (17.4)
Polymyxin	3 (1.1)	21 (0.5)	0.174	24 (0.6)
Any prior antifungal use	50 (19.0)	608 (15.3)	0.037	658 (15.5)
Any prior antibiotic use	239 (90.9)	3,458 (86.9)	0.001	3,697 (87.2)

^aUnless otherwise specified, characteristics are dichotomized and reported as absolute frequency (percent).

^bp-values for continuous variables were based on Mann-Whitney U statistical tests, while categorical variable p-values were obtained from either Chi-square or Fisher exact tests as appropriate.

^cIncludes any history or diagnosis of leukemia or lymphoma

^dIncludes any history or diagnosis of malignancy of the uterus, cervix, ovary, or other female genital sites

^eIncludes the presence of a permanent or temporary central line or a dialysis catheter

^fMedication was ordered within 90 days prior to Candida infection.

NOTE: Descriptive statistics for additional variables are presented in Supplementary Table 1. Abbreviations: IQR, interquartile range; TPN, total parenteral nutrition

Univariable Logistic Regression

In univariable logistic regression analyses, we found71 variables to be predictors of candidemia in the setting of candiduria (Supplementary Table 1). Candidemia was more prevalent in patients with candiduria caused by non-albicans species (odds ratio [OR], 1.9; 95% confidence interval [CI], 1.2–3.1; p = 0.007). Concomitant candidemia was also more likely to be diagnosed in patients with a hospital stay greater than 12 days (OR 2.1, 95% CI, 1.7–2.7; p < 0.001), a CVC (OR, 2.0; 95% CI, 1.6–2.6; p < 0.001), receipt of TPN (OR, 1.5; 95% CI, 1.2–2.0; p = 0.002), hematological malignancy (OR, 2.4; 95% CI, 1.6–3.5; p < 0.001), and prior antibiotic use (OR, 1.7; 95% CI, 1.3–2.4; p = 0.002). On evaluation of vital signs and laboratory values, candidemia was more prevalent among patients with an elevated heart rate (OR, 2.5; 95% CI, 1.9–3.1; p < 0.001), temperature (OR, 2.4; 95% CI, 1.8–3.0; p < 0.001) and creatinine (OR, 1.6; 95% CI, 1.2–2.0; p < 0.001), and a lower hemoglobin level(OR, 2.6; 95% CI, 1.9–3.5; p < 0.001).

Gastrostomy dependency (OR, 0.4; 95% CI, 0.2–1.0; p = 0.045) and diabetes (OR, 0.7; 95% CI, 0.5–0.9; p = 0.004), and hypertension (OR 0.6; 95% CI, 0.5–0.8; p < 0.001) were associated with a lower risk for candidemia.

Clinical Predictive Model

The multivariable logistic regression model consisted of 15 variables (Table 2). Eleven variables demonstrated an increased risk for candidemia, including hospital stay greater than 12 days (OR 1.4, 95% CI, 1.04–1.8; p = 0.027), presence of a CVC (OR, 1.8; 95% CI, 1.4–2.4; p < 0.001), receipt of TPN (OR, 1.5; 95% CI, 1.1–2.0; p = 0.005), hematological malignancy (OR, 2.1; 95% CI, 1.4–3.2; p < 0.001), and gynecologic malignancy (OR, 1.7; 95% CI, 1.03–2.9; p = 0.037). Vital signs and laboratory values included elevated heart rate (OR, 1.9; 95% CI, 1.4–2.4; p < 0.001), temperature (OR, 1.8; 95% CI, 1.3–2.3; p < 0.001) and creatinine (OR, 1.8; 95% CI, 1.4–2.3; p < 0.001), along with neutropenia (OR, 1.8; 95% CI, 1.1–3.1; p = 0.021) and neutrophilia (OR, 1.6; 95% CI, 1.2–2.1; p = 0.003). The only medication that remained significant in the final model was prior β-lactam/β-lactamase inhibitor use (OR, 1.6; 95% CI, 1.2–2.2; p = 0.002).

Table 2.

Clinical Predictive Model for Candiduria with Candidemiavs Candiduria Alone

Variable	Parameter Estimate	Odds Ratio (95% CI)	p-value
Intercept	−3.754	n/a	n
Hospital length of stay > 12 days	0.310	1.363 (1.036, 1.795)	0.027
Hematologic malignancya	0.754	2.126 (1.394, 3.242)	<0.001
Gynecologic malignancyb	0.554	1.740 (1.034, 2.927)	0.037
Central venous catheterc	0.599	1.821 (1.398, 2.372)	<0.001
Total parenteral nutrition	0.415	1.514 (1.136, 2.017)	0.005
Prior β-lactam/β-lactamase inhibitor used	0.494	1.639 (1.203, 2.233)	0.002
Heart rate >120 beats/min	0.618	1.856 (1.408, 2.446)	<0.001
Temperature
>39 °C	0.570	1.769 (1.343, 2.330)	<0.001
Absolute neutrophil count
<1,500/mm3	0.606	1.833 (1.096, 3.065)	0.021
>9,500/mm3	0.445	1.560 (1.164, 2.093)	0.003
Creatinine >1.3 mg/dL	0.583	1.791 (1.372, 2.337)	<0.001
Sex (Female)	−0.609	0.544 (0.414, 0.713)	<0.001
Diabetes mellitus	−0.445	0.641 (0.465, 0.883)	0.006
Gastrostomy	−0.984	0.374 (0.150, 0.932)	0.035
Urinary catheter and priord antibiotic use	−0.639	0.528 (0.288, 0.968)	0.039

^aIncludes any history or diagnosis of leukemia or lymphoma

^bIncludes any history or diagnosis of malignancy of the uterus, cervix, ovary, or other female genital sites

^cIncludes the presence of a permanent or temporary central line or a dialysis catheter

^dMedication was ordered within 90 days prior to Candida infection.

Abbreviations: CI, confidence interval

Variables associated with a decreased risk for candidemia in the final model included female gender (OR, 0.5; 95% CI 0.4–0.7; p < 0.001), gastrostomy dependency (OR, 0.4; 95% CI, 0.2–0.9; p = 0.035), diabetes (OR, 0.6; 95% CI, 0.5–0.9; p = 0.006), and urinary catheter with prior antibiotic use (OR, 0.5; 95% CI, 0.3–0.97; p = 0.039). The c-statistic for this model was 0.741 (95% CI, 0.710 – 0.772) (Figure 1).

Figure 1.

Receiver operating characteristic (ROC) curve for the multiple logistic regression predicting concurrent candidemia and candiduria. The c-statistic is 0.741 (95% CI 0.710 – 0.772).

Discussion

Although Candida species in the urine rarely disseminates into the bloodstream, candiduria has a positive association with candidemia⁴. Prior investigations on the risk factors for candidemia generally focus on patients in the intensive care units or as a comparison between C. albicans versus non-albicans species infection^{9, 12, 21, 22}. To the best of our knowledge, this is the first study to assess potential patient factors that influence the risk of concurrent candidemia with candiduria.

While candiduria is highly prevalent in hospitalized patients, the presence of concomitant candidemia is infrequently encountered^{1, 4}. Our study revealed that out of the 4,240 patients with candiduria, only 263 cases (6.2%) had concomitant candidemia. This aligns with current epidemiological studies, which calculate the incidence of candidemia in the presence of candiduria to be 1.3–10%^{23, 24}.

Our multivariable analysis demonstrated a decreased risk of candidemia in the setting of candiduria with female gender. Candiduria by itself is typically more prevalent in the female population because of a shorter urethra and a high likelihood (10–65%) of vulvovestibular colonization with Candida^{4, 25}. Accordingly, women with candiduria in our study outnumbered the men (63.9% vs 36.1%).

Hospital stay greater than 12 days was an independent risk factor for concomitant candidemia with candiduria. Medical interventions that invariably accompany an extended hospital course, including but not limited to exposure to antibiotics and invasive therapies, are likely driving this association rather than the length of stay itself.

Although not a variable studied before in candidemia risk assessments, our analysis identified gastrostomy to be associated with less risk for concomitant candidemia and candiduria. This could be related to the fact that those patients with a gastrostomy would likely be receiving TPN if they were without a gastrostomy tube. This finding is novel and would benefit from further study.

Other investigators have suggested both hematologic malignancies and solid tumors to be predictors of candidemia, which is consistent with our findings in candiduric patients^{19, 26}. Specifically, we found gynecologic malignancies to be the class of solid tumors associated with candidemia in the setting of candiduria. Current literature reports that candidemia in patients with solid tumors is more likely to occur in the setting of recent surgery¹⁹. Multiple studies are in consensus that abdominal surgery is the main factor driving this association due to Candida colonization and translocation from the gastrointestinal tract^{10, 27}. In contrast, there is limited research available regarding gynecologic surgeries in relation to Candida infections. Furthermore, epidemiological evidence suggests that vaginal colonization by Candida can be found in up to 20–30% of women²⁸, and surgical manipulation of the vaginal tract may disrupt the balance between host colonization and infection, subsequently allowing Candida to gain entry into the bloodstream.

On analysis of laboratory values, we found elevated creatininelevel to be independently associated with candiduria with concomitant candidemia. While few authors have investigated the relationship between creatinine level and candidemia, kidney dysfunction is a commonly recognized variable in Candida studies. Previous literature found an increased risk for candidemia in the setting of chronic renal failure (OR 4.48 −11.5), although one study was unable to adjust for other variables due to a limited sample size^{20, 29}. One cohort analysis of SICU patients suggested acute renal failure to be a risk factor for the onset of candidemia (RR 4.2)³⁰. Chronic kidney disease failed to demonstrate significance in our univariable analysis, suggesting that the elevated creatinine is more likely to represent acute renal failure.

Our multivariable analysis determined both neutropenia and neutrophilia to be clinical predictors of candiduria with candidemia. Retrospective analyses found that neutropenia was present in 8–11.6% of patients diagnosed with candidemia^{31, 32}. In comparison, neutropenia was present in 8.7% of our patient population. While neutrophils are important for early clearance of infection, prolonged neutrophilia may be associated with a high fungal burden and be indicative of an uncontrolled systemic infection.

Additional indicators of systemic infection, including elevated heart rate and temperature, were also clinical predictors of candidemia in our study. Aberrations in either white cell count, heart rate, or temperature may suggest a clinical picture consistent with a systemic inflammatory host response³³. A retrospective cohort study evaluated the inflammatory response of 60 patients with candidemia and found that greater than 90% of patients fulfilled the Systemic Inflammatory Response Syndrome (SIRS) criteria on the day of the first positive blood culture³⁴.

In regards to antibiotics, we found β-lactam/β-lactamase inhibitors to be significantly associated with elevated rates of candidemia in candiduric patients. Antibiotics can selectively alter intestinal microbial flora in such a way that facilitates the growth and dissemination of Candida species³⁵. Multiple investigations have illustrated this link between antibiotics and candidemia, yet there is a paucity of studies further delineating the class of antibiotics responsible for this association^{9, 20, 35}. The limited research available also show conflicting results; some studies showed that exposure to vancomycin and piperacillin-tazobactam was associated with C. glabrata or C. krusei candidemia yet another study found the same antibiotics to be noncontributory^{17, 36, 37}. In one study, antibiotics with anaerobic activity appeared to predispose to candidemia; however, C. glabrata accounted for the majority of the sample and the results may not apply to Candida species as a group³⁸.

Neither urinary catheter or prior antibiotic exposure contributed significantly to our model, but when combined together demonstrated a decreased risk of candidemia in candiduric patients. This finding is likely because they increase urinary colonization synergistically. Urinary catherization provides a direct portal of entry into the bladder, allowing microbes to establish infection either via an extraluminal or intraluminal pathway^{25, 39}. While ß-lactam/ß-lactamase inhibitor use is associated with an increased risk for candidemia likely by its influence on the intestinal microbial flora, prior antibiotic exposure altogether may have a greater effect on the lower genital tract, altering the microbiome and increasing Candida colonization²⁵. This factor combined with urinary catheter confers additional risk for candiduria, which translates to a decreased risk for candidemia in our final model.

Our study has several limitations. First, we used a retrospective design which allowed us to analyze a large patient sample over a broad time period, but also has inherent weaknesses and predisposes to several types of information bias and selection bias. The ICD-9 codes used to construct the database may not reflect true diagnoses, contributing to misclassification bias. Our study was also restricted to a single institution with discrepancies that may exist between our patient population and the general population.

In conclusion,we found a set of clinical comorbidities that can predict the presence of candidemia with concomitant candiduria.