AAC Accepted Manuscript Posted Online 6 March 2017 Antimicrob. Agents Chemother. doi:10.1128/AAC.02091-16 Copyright © 2017 American Society for Microbiology. All Rights Reserved.
1
A Risk Score for Fluconazole Failure among Patients with Candidemia
2
Luis Ostrosky-Zeichner,a,#, Rachel Harringtonb, Nkechi Azieb, Hongbo Yangc, Nanxin Lic, Jing
4
Zhaoc, Valerie Kooc, Eric Q. Wuc
5
Division of Infectious Diseases, McGovern Medical School, Houston, Texas, USAa; Astellas
6
Pharma US, Northbrook, IL, USAb; Analysis Group, Inc., Boston, MA, USAc
7 8
Running Head: Fluconazole Failure Risk in Candidemia
9 10
# Address correspondence to Luis Ostrosky-Zeichner,
[email protected]
11
For submission to: Antimicrobial Agents and Chemotherapy
12 13
Number of Tables and Figures: 4 tables + 1 figure
14
Number of references: 40
15
Key Words: Candida, failure, fluconazole, candidemia, risk score
16
Word Count: 2,638 words/no limit
17
CONFIDENTIAL
PAGE 1
Downloaded from http://aac.asm.org/ on March 16, 2019 by guest
3
ABSTRACT
19
Word Count: 248/250
20
Background: This study aimed to develop a prediction model to identify patients with
21
candidemia at high risk of failing fluconazole.
22
Methods: Adult patients in the US with candidemia who received fluconazole during
23
hospitalization were selected from the Cerner Health Facts® database (04/2004-03/2013).
24
Fluconazole failure was defined as switching/adding another antifungal, positive Candida culture
25
≥10 days after fluconazole initiation, or death during hospitalization. Patients were randomized
26
into modeling and validation samples. Using the modeling sample, a least-absolute-shrinkage-
27
and-selection-operator regression was used to select risk predictors of fluconazole failure
28
(demographics, Candida species, initiation of fluconazole before positive culture and after
29
admission, and comorbidities, procedures, and treatments during the 6-months before admission
30
to fluconazole initiation). The prediction model was evaluated using the validation sample.
31
Results: Of 987 identified patients (average age of 61 years, 51% male, and 72% white), 49%
32
did and 51% did not fail fluconazole. Of those who failed, 70% switched or added another
33
antifungal, 21% had a second positive Candida test, and 42% died during hospitalization. Nine
34
risk factors were included in the prediction model: days to start fluconazole after admission,
35
Candida glabrata or krusei infection, hematological malignancies, venous thromboembolism
36
(VTE), enteral nutrition, use of non-operative intubation/irrigation, mechanical ventilation, and
37
other antifungal use. All but VTE were associated with a higher risk of failure. The model’s c-
38
statistic was 0.65, with a Hosmer-Lemeshow test p=0.23.
CONFIDENTIAL
PAGE 2
Downloaded from http://aac.asm.org/ on March 16, 2019 by guest
18
39
Summary: This prediction model identified patients with a high risk of fluconazole failure,
40
illustrating the potential value and feasibility of personalizing candidemia treatment.
41
Downloaded from http://aac.asm.org/ on March 16, 2019 by guest
CONFIDENTIAL
PAGE 3
42
INTRODUCTION Candida is the most common cause of hospital-acquired bloodstream infections in the US
44
and a major cause of morbidity and mortality worldwide (1-3). In addition to high morbidity and
45
mortality, candidemia is also associated with high economic burden of nearly $300 million
46
annually in the US (4). Compared to a matched non-candidemia-exposed cohort, adult patients
47
with candidemia incurred higher hospital charges, by $39,331, and had longer stays, of 10.1 days
48
in length (5).
49
Echinocandins (e.g., caspofungin, micafungin, and anidulafungin), fluconazole, and
50
liposomal amphotericin B (AmB) are currently the recommended first-line therapies for the
51
treatment of Candida infections (6, 7). Historically, fluconazole was the recommended first-line
52
therapy in the Infectious Diseases Society of America’s (IDSA) treatment guideline for Candida
53
infections, and thus was the most commonly used treatment (8). In recent years, however,
54
fluconazole-resistant species such as Candida krusei and Candida glabrata have been isolated
55
with increasingly frequency (9, 10). This is cause for concern (11-13), and recent reports have
56
found that approximately 7% of all Candida bloodstream isolates were tested as fluconazole
57
resistant (14). Furthermore, these strains appear to be increasingly resistant. An analysis of 313
58
isolates of Candida glabrata over a 10-year period demonstrated an increase in resistance to
59
fluconazole from 18% to 30% (15, 16). In response to these findings, along with a patient-level
60
clinical trial review showing a mortality benefit when using echinocandins (17), the 2016 update
61
to the IDSA guidelines for the treatment of candidiasis now recommends echinocandins as first-
62
line therapy (6). Fluconazole is considered an acceptable alternative first-line therapy in patients
63
who are not critically ill and who are not likely to have a fluconazole-resistant Candida strain, CONFIDENTIAL
PAGE 4
Downloaded from http://aac.asm.org/ on March 16, 2019 by guest
43
and the IDSA recommends transition from first-line echinocandin or AmB to fluconazole,
65
usually within a week, for patients who have stabilized and have non-resistant fluconazole strains
66
(6). The latest IDSA guideline may potentially lead to a gradual shift from first-line fluconazole
67
to first-line echinocandin; however, presently, fluconazole is still widely prescribed (4).
68
An important question in the clinical management of candidemia is how to determine the
69
choice of an effective first-line therapy for an individual patient, i.e. how to determine which
70
patients should use fluconzole vs. an echinocandin or other therapy. While microbiological
71
testing (speciation and antifugnal susceptability testing) of admitted patients’ Candida cultures
72
can most accurately inform treatment decisions (empirical therapy), waiting for the results of
73
such tests can delay potentially life-saving treatment (18). Some studies have reported a
74
correlation between higher patient mortality and the delay of appropriate antifungal treatment,
75
with earlier initiation associated with lower mortality (19, 20). The odds of mortality increased
76
by 50% with an antifungal therapy initiation delay of just 1 day from the onset of symptoms (19).
77
However, a retrospective analysis assessing mortality rates and the time to initiation of
78
appropriate therapy for candidemia found no correlation with mortality (21). Nevertheless,
79
nosocomial invasive fungal disease has one of the highest rates of inappropriate therapy, which
80
is associated with increased mortality, thus early and accurate identification of the proper therapy
81
is of paramount importance (22, 23).
82
The objective of the current study was to develop a risk score model, based on patients’
83
clinical profiles, to predict the risk of failing fluconazole among patients with candidemia. Such
84
a model will identify patients who are not suitable for fluconazole, allow early initiation of
CONFIDENTIAL
PAGE 5
Downloaded from http://aac.asm.org/ on March 16, 2019 by guest
64
alternative therapies, and may lead to improved clinical outcomes and potential reductions in
86
healthcare costs among these high-risk patients.
87
METHODS
88
Data Source
89
This study used the Cerner Health Facts® Hospital Database, a commercially-available
90
electronic de-identified health record database used in over 150 hospitals throughout the US. The
91
database contains 110 million records spanning from 04/01/2004 to 03/31/2013, and includes
92
patient demographic information, as well as medical (admission, discharge, diagnoses,
93
procedures, etc.), hospital pharmacy, and laboratory data (including microbiology and pathogen
94
information). Data collected was compliant with the Health Insurance Portability and
95
Accountability Act and the Declaration of Helsinki (1964, amended 2008); no ethical review was
96
required.
97
Sample Selection and Construction
98
Adult patients aged 18 years and older who had at least one positive blood culture
99
indicating Candida infection during a hospitalization were selected. Patients were further
100
required to have initiated intravenous fluconazole treatment during their hospital stay, and no
101
more than 5 days before the positive blood culture reading for Candida infection. Included
102
patients were also required to be over 18 years old and have at least one non-missing underlying
103
diagnosis during their hospitalization. For patients who met all sample selection criteria and had
104
multiple hospitalization records, one hospitalization per patient was randomly selected. This was CONFIDENTIAL
PAGE 6
Downloaded from http://aac.asm.org/ on March 16, 2019 by guest
85
to allow capture of both the early and late use of fluconazole during the course of the Candida
106
infection, and to avoid confounding in the results due to correlation if all observations from same
107
patient were included.
108
Key Definitions
109
The index hospitalization was defined as a hospitalization encounter that met all sample
110
selection criteria. The index date was defined as the date of initiation of fluconazole treatment as
111
the first-line treatment during the index hospitalization. The baseline period was defined as the
112
period 6 months prior to the index hospital admission through the index date (fluconazole
113
initiation). Patients with fluconazole failure were defined as patients who met any of the
114
following criteria: 1) switched to or added on another antifungal therapy following fluconazole
115
initiation; 2) had a subsequent positive blood culture for Candida infection during the index
116
hospitalization and at least 10 days following fluconazole initiation; or 3) died during the period
117
from index date to end of index hospitalization. For criterion 2), the IDSA treatment guidelines
118
do not specify a time for the definition of fluconazole failure (8). The National Committee on
119
Clinical Laboratory Standards has proposed to define failure as lack of response to systematic
120
fluconazole therapy within 14 days (24). Given that the initiation of fluconazole was allowed
121
within 5 days of a positive blood culture in the current study; thus, a timeframe of 10 days was
122
used to evaluate the recovery from candidemia based on a blood culture result.
123
Study Measures
124 125
Patient-level measures were collected from the database and included demographics (age, sex, and race), baseline period risk factors (Charlson Comorbidity Index [CCI] (25), diagnoses, CONFIDENTIAL
PAGE 7
Downloaded from http://aac.asm.org/ on March 16, 2019 by guest
105
procedures, and treatments received) within 6 months prior to the index date, and characteristics
127
of the index hospitalizations (index year, Candida species, time to initiate fluconazole from
128
admission, and prior fluconazole use). All study measures were compared between patients with
129
and without fluconazole failure using Wilcoxon rank-sum tests for continuous variables and Chi-
130
square tests for categorical variables.
131
Development of a Prediction Model for Risk of Fluconazole Failure
132
An optimal set of risk factors for predicting fluconazole failure were selected using the
133
least absolute shrinkage and selection operator (LASSO) approach from a list of candidate risk
134
factors (26), which permits simultaneous selection of a set of risk factors instead of pre-defining
135
risk factors or step-wise selection. Candidate risk factors included patient demographics such as
136
age, gender, and race, index year, index hospitalization characteristics such as Candida species
137
during the index hospitalization, time to start fluconazole from admission, prior fluconazole use,
138
and conditions (diagnoses, procedures, and treatments) during the baseline period. Baseline
139
conditions with frequencies more than 1% in the entire sample were considered as candiate risk
140
factors.
141
Once the optimal list of risk factors was selected, a prediction model was then developed
142
using a cross-validation approach. Total samples were partitioned into modeling and validation
143
samples stratified by fluconazole failure status at a ratio of 2 to 1. A logistic regression model
144
was developed in the modeling sample, and model performance was evaluated by the C-statistic
145
and the Hosmer-Lemeshow test (27) in the validation sample. Sensitivity, specificity, positive
146
predictive value (PPV), and negative predictive value (NPV) were estimated for various cut-offs
147
of the predictive failure rate that defined fluconazole failure using the validation sample (28). CONFIDENTIAL
PAGE 8
Downloaded from http://aac.asm.org/ on March 16, 2019 by guest
126
148 149
Statistical Analyses Statistical analyses were performed in the program SAS version 9.4(SAS Institute Inc. Cary, NC) and R version 3.1.2 (open source, The R Foundation for Statistical Computing). A two-
151
tailed p-value of 0.05 was used to determine significance.
152
RESULTS
153
Baseline Characteristics
154
Among 6,051 patients who had at least one positive blood culture indicating a Candida
155
infection during the index hospitalization, a total of 987 patients met all inclusion criteria and
156
were included in the final study sample (Figure 1). Of those, 488 (49%) patients experienced
157
fluconazole failure. Among the patients who experienced fluconazole failure, 70% had switched
158
or added another antifungal therapy, 21% had a subsequent positive blood test for Candida, and
159
42% died in the hospital. Patient demographics were similar between patients with and without
160
fluconazole failure (Table 1). Both groups had mean age of 61 years, were ~51% male, and 70%
161
Caucasian. The percentage of patients with Candida albicans was 55.4%, with Candida glabrata
162
was 22.1%, with Candida krusei was 0.8%, with Candida parapsilosis was 11.1%, and with
163
other Candida species was 11.3%. Patients with fluconazole failure, compared to those without
164
failure, had a higher rate of Candida glabrata infection (27.5% versus 16.8%, p