Introduction
Candidemia, associated with extended hospitalisation and high mortality, poses a global threat, particularly among immunocompromised patients [Reference Benedict, Jackson, Chiller and Beer1–Reference Tang, Liu, Lin and Lai4]. The reported incidence of candidemia ranged from 1.7 to approximately 10 cases per 100,000 person-years or nearly 1.22 episodes per 1,000 discharges [Reference Bouza and Muñoz5–Reference Vallabhaneni, Mody, Walker and Chiller8]. Previous studies have identified several risk factors associated with candidemia, including indwelling catheters, usage of steroids, disrupted gut or cutaneous barriers, hemodialysis, and injection drug use, especially in the United States [Reference Ishikane, Hayakawa, Kutsuna, Takeshita and Ohmagari9–Reference Zhang, Shrum, Williams, Petnic, Nadle, Johnston, Barter, Vonbank, Bonner, Hollick, Marceaux, Harrison, Schaffner, Tesini, Farley, Pierce, Phipps, Mody, Chiller, Jackson and Vallabhaneni13]. It is worth noting that the distribution of Candida species varies across different countries and patient populations with various underlying diseases [Reference Tan, Chakrabarti, Li, Patel, Watcharananan, Liu, Chindamporn, Tan, Sun, Wu, Chen and Asia Fungal Working7, Reference Sofair, Huie-White, Reiss, Harrison, Sanza, Arthington-Skaggs and Fridkin14]. The SENTRY antifungal surveillance program reported a decrease in the isolation of C. albicans to less than half but an increase in the isolation of C. glabrata and Candida parapsilosis [Reference Pfaller, Diekema, Turnidge, Castanheira and Jones15]. Nevertheless, C. tropicalis presented high rates of resistance to fluconazole in the Asia-Pacific region and was more likely to be isolated in hemato-oncology wards [Reference Tan, Chakrabarti, Li, Patel, Watcharananan, Liu, Chindamporn, Tan, Sun, Wu, Chen and Asia Fungal Working7, Reference Pfaller, Diekema, Turnidge, Castanheira and Jones15].
Non-hospital-onset (NHO) candidemia, which is defined as the onset of candidemia occurring in outpatient settings or within 2 days after hospital admission, has been previously recognised as community-onset and is an emerging issue [Reference Sofair, Huie-White, Reiss, Harrison, Sanza, Arthington-Skaggs and Fridkin14, Reference Kato, Yoshimura, Suido, Shimizu, Ide, Sugiyama, Matsuno and Nakajima16, Reference Suzuki, Perencevich, Diekema, Livorsi, Nair, Kralovic, Roselle and Goto17]. The reported proportion of NHO candidemia among all candidemia varied from 0 to 31.14%, highlighting the need for studies to address NHO candidemia due to the discrepant epidemiological results [Reference Sofair, Huie-White, Reiss, Harrison, Sanza, Arthington-Skaggs and Fridkin14, Reference Asmundsdottir, Erlendsdottir and Gottfredsson18–Reference Pfaller, Moet, Messer, Jones and Castanheira21]. Furthermore, an American study found that NHO candidemia was more likely to result from C. parapsilosis and had a lower 30-day case-fatality rate compared with those with hospital-onset (HO) candidemia [Reference Sofair, Huie-White, Reiss, Harrison, Sanza, Arthington-Skaggs and Fridkin14]. However, there is a lack of data on candidemia, especially NHO candidemia, in Asian countries. Therefore, we conducted this 4-year retrospective study to explore the characteristics of people with candidemia.
Methods
Study design, patient settings, and definitions
We designed a retrospective observational cohort study on adult patients (age ≥ 20 years old) who were hospitalised for two or more days and were diagnosed with candidemia. We reviewed all electronic medical charts and microbiological data from the laboratory management information system in the Taichung Veterans General Hospital Research Database (registered number: F20424) between 1 January 2015 and 31 December 2018. This study was approved by the Institutional Review Board (IRB) I & II of the Taichung Veterans General Hospital, and the IRB serial is CE19376A. Informed consent was waived because all data obtained from individual patients were anonymised before the analysis.
Candidemia was defined as the presence of at least one set of blood cultures of Candida species with relevant clinical symptoms and signs during hepatisation. The onset of candidemia was defined as the day when patients received blood culture tests. Patients with candidemia were further categorised as NHO candidemia (defined as patients with candidemia experiencing onset of disease outside of the hospital, in the emergency department, or within 2 days after hospital admission) or HO candidemia (defined as the first episode of candidemia occurring after 2 days of hospitalisation) [Reference Sofair, Huie-White, Reiss, Harrison, Sanza, Arthington-Skaggs and Fridkin14, Reference Kato, Yoshimura, Suido, Shimizu, Ide, Sugiyama, Matsuno and Nakajima16, Reference Suzuki, Perencevich, Diekema, Livorsi, Nair, Kralovic, Roselle and Goto17]. Bacterial concomitant bloodstream infection (BSI) was defined as one or more positive bacterial blood cultures isolated collectively or within 48 h of the time of candidemia [Reference Chen, Xu and Wu22, Reference Zhong, Zhang, Tang, Zhou, Zheng, Zhang, Cai, Zhou, Wang, Tian, Zhang, Cui, Dong and Zhang23]. The presence of non-tunnel catheters was defined as central venous catheters (CVC) being in place for more than 24 h prior to the onset of candidemia [Reference Liu, Huang, Wang, Chen, Yen, Yang, Hsiao, Liu, Chen and Chiou24]. Initial antifungal therapy was regarded as inadequate if Candida species were resistant or if no related minimum inhibitory concentration (MIC) was reported. Patients who died or were discharged in critical condition were classified as in-hospital mortality. The aim of this study was to investigate the epidemiological characteristics of candidemia, explore the involved Candida species, and identify risks for mortality in patients with candidemia.
Laboratory identification, susceptibility test of Candida species
Five to 10 millilitres of blood samples were collected from patients with related symptoms and signs by aseptic procedure into blood culture flasks (Becton, Dickinson and Company), transferred to the Clinical Microbiology Department, and then incubated appropriately. Whenever positive cultures were noted, the pathogens were further identified by using VITEK 2 (bioMérieux, Marcy l’Etoile, France) yeast identification card system (VITEK® 2 YST ID card) or matrix-assisted laser desorption/ionisation time-of-flight mass spectrometry (MALDI-TOF MS). The antifungal susceptibility testing was performed by using the VITEK® 2 Yeast Susceptibility Card, AST-YS05 (bioMérieux, Marcy l’Etoile, France) and SensiTitreTM YeastOneTM (microbroth dilution test using alamarBlue). The susceptibility of Candida species was regrouped with updated breakpoints described in the Clinical and Laboratory Standards Institute (CLSI; M60-ED2:2020 performance standards for antifungal susceptibility testing of yeasts, 2nd edition).
Statistical analyses
Continuous variables were presented as the mean ± standard deviation (SD) or median with interquartile range (IQR), and further compared using a Student’s t-test or Mann–Whitney U test whenever appropriate based on the check of the normality assumption. Categorical variables among groups were compared using counts and percentages. Statistical comparisons between groups were made by χ2 test or Fisher’s exact. The Simpson’s diversity index was applied to express the diversity of Candida species in NHO or HO candidemia [Reference Hurlbert25]. The dependent variable was all-cause in-hospital mortality. On the contrary, the independent variables were those risk factors and underlying comorbidities potentially affecting the clinical outcomes. The Kaplan–Meier survival analysis with log-rank tests was applied for survival differences of all-cause in-hospital mortality from the onset of candidemia to discharge. We incorporated variables of interest to build a full Cox proportional-hazards model to identify predictors of in-hospital mortality, and then the adjusted hazard ratios (aHRs) of each individual factor were calculated. A p-value of 0.05 was used to determine statistical significance. Rstudio (2022.02.2+485) in R version 4.1.0 (18 May 2021) with appropriate packages was used for the statistical analysis in this study.
Results
Demographics and clinical characteristics of patients with NHO and HO candidemia
A total of 339 patients were identified in this study (Figure 1), resulting in an overall incidence of 1.50 per 1,000 admission person-years. The median age of patients was 64 years old, and 64.60% of patients were male. There was no age (p = 0.10) or gender (p = 0.785) difference between NHO and HO. The onset of HO was 28.53 days after admission (see Table 1). About, 57.52% (195/339) of patients were diagnosed with at least one kind of malignancy with no significant difference between NHO and HO (48.78% vs. 60.31%, p = 0.066). The HO group had a higher proportion of patients with non-tunnel catheters compared to the NHO group (45.14% vs. 8.54%, p < 0.001). Patients with HO candidemia had a significantly higher proportion of chemo port existence but a lower proportion of chemo port removal after the onset of candidemia compared to those with NHO candidemia (16.34% vs. 21.95%, p = 0.002). The overall Charlson comorbidity index (CCI) ± SD was 5.60 ± 2.78 in this cohort, showing no significant difference between patients in the NHO and the HO group (5.73 ± 2.70 vs. 5.56 ± 2.80, p = 0.619). 90.56% (307/339) of patients with candidemia had at least one risk factor, including the presence of a non-tunnel catheter or a chemo port implant, receiving steroids within 7 days before candidemia, receiving chemotherapy within 1 month prior to the onset of candidemia, or the presence of malignancy or chronic kidney disease (CKD) or end-stage renal disease (ESRD). The risks of acquiring candidemia were significantly higher in patients with HO candidemia compared to NHO candidemia (95.72% vs. 74.39%, p < 0.001). Most characteristics of comorbidities were similar, except for a higher proportion of diabetes mellitus (DM) in NHO candidemia (42.68% vs. 26.85%, p = 0.007).
Table 1. The comparison of demographics and clinical characteristics of patients with candidemia between NHO and HO candidemia (n = 339)
Note: Data are presented as no. (%) unless indicated in the specific patient demographics.
Abbreviations: ARDS, acute respiratory distress syndrome; CKD, chronic kidney disease; CNS, central nervous system; COPD, chronic obstructive pulmonary disease; CVC, central venous catheter; ESRD, end-stage renal disease; HO, hospital-onset; IQR, interquartile range; NHO, non-hospital-onset; SD, standard deviation; TPN, Total parenteral nutrition.
a Demographics are presented at the patient level.
b The negative value presented days prior to admission.
c The risks of candidemia development include any risk below: the presence of a non-tunnel catheter, chemo port implant, receiving steroids within 7 days before candidemia, receiving chemotherapy within 1 month before candidemia, malignancy, severe skin defect, CKD or ESRD, and hollow organ perforation.
d Includes patients with burn injury and toxic epidermal necrolysis.
* p < 0.05.
Figure 1. Flows of the study design. This study was approved by the Institutional Review Board I & II of Taichung Veterans General Hospital (CE19376A), Taichung, Taiwan. During the study period from 1 January 2015 to 31 December 2018, a total of 339 adult hospitalized patients (age = 20-year-old) were identified after excluding 21 patients from a total of 764 positive blood cultures for Candida species within 11,128 positive blood cultures by the commercial identification system (VITEK® 2) in this retrospective observational study. Further epidemiological and microbiological analyses were performed for these 339 patients and 344 isolates. Kaplan-Meier survival curves with a log-rank test were applied for survival analysis. Note:a Among 339 patients, five patients were noted to have two different Candida species isolated from their blood samples, summing up 344 isolates. The criteria of breakpoint of antifungal susceptibility were on the basis of the Clinical and Laboratory Standards Institute guidelines (CLSI; M60-ED2:2020 performance standards for antifungal susceptibility testing of yeasts, 2nd edition). b Comparison of demographic, clinical, and treatment characteristics were executed between community-onset candidemia (n=83) and nosocomial candidemia (n=256).
Diversity of Candida species and antifungal susceptibility among these 339 patients with candidemia
C. albicans (52.21%, 177) accounted for the most common isolated species, followed by C. tropicalis (17.11%, 58), C. glabrata (16.22%, 55), and C. parapsilosis (8.26%, 22). No adult patients were infected by Candida krusei; however, three C. krusei infections occurred in children, but these were excluded due to the study design. Five patients (1.47%) were found to be infected by more than one species, all consisting of C. albicans.
The Simpsons diversity index showed similar species diversity between NHO and HO candidemia (0.698 vs. 0.642). However, the proportion of Candida species was significantly different between NHO and HO (Table 2, p = 0.002). In patients with NHO candidemia, a higher proportion of C. glabrata (25.61% vs. 13.23%), C. parapsilosis (10.98% vs. 7.39%), and Candida pelliculosa (6.10% vs. 2.72%) were isolated, while there were fewer isolations of C. albicans (47.56% vs. 53.70%) and C. tropicalis (4.88% vs. 21.01%) than in the HO group. By analysing further by species, we found a higher percentage of C. glabrata isolated from patients with NHO candidemia than those in the HO group (26.83% vs. 13.62, p = 0.005), whereas the isolation of C. tropicalis showed the opposite result (4.88% vs. 21.01%, p = 0.001).
Table 2. Diversity and resistance characteristics of Candida species among 339 patients with candidemia
Note: Data are presented as no. (%).
Abbreviations: HO, hospital-onset; NHO, non-hospital-onset.
* p < 0.05.
The antifungal susceptibility test results of 344 Candida species among 339 patients with candidemia are summarised in supplement Table 1. C. albicans (176, 96.7%) still demonstrated high susceptibility to fluconazole. The antifungal resistance of fluconazole (p = 0.784) and voriconazole (p = 0.598) showed no statistical significance between NHO and HO candidemia.
Treatment characteristics among 339 patients with candidemia
Echinocandin was the most commonly prescribed initial antifungal therapy, and a higher proportion of patients with NHO candidemia accepted antifungal treatment compared to those with HO candidemia (95.12% vs. 84.83%, p = 0.016; Table 3). The gap between the onset of candidemia and initiation of antifungal therapy was 2.58 days, with a longer delay in the NHO group (3.49 vs. 2.56 days, p = 0.001). Less than half (48.08%, 163/339) of patients with candidemia received antifungal therapy on or before 2 days after candidemia, and with no significant difference in antifungal therapy between groups (p = 0.366). The proportion of inadequate initial antifungal therapy due to resistance, without related MIC report or no antifungal therapy during the whole hospitalisation was significantly higher in the HO group than in the NHO group (19.07% vs. 9.76%, p = 0.05). Finally, a longer antifungal therapy duration, including initial treatment (14.15 ± 9.68 days vs. 10.46 ± 8.96 days, p = 0.002) and total therapy duration (20.51 ± 17.34 vs. 14.98 ± 13.26 days, p = 0.003), was noted in the NHO group than in the HO group.
Table 3. Treatment characteristics among 339 patients with candidemia
Note: Data are presented as no. (%) unless indicated in the specific treatment characteristic.
Abbreviations: IQR, interquartile range; SD, standard deviation.
a Demographics are presented at the patient level.
* p < 0.05.
The outcomes of 339 patients with candidemia
The all-cause in-hospital mortality rate was 55.75% (189/339, Table 1). Survival analysis showed a significantly better outcome for patients with NHO candidemia than for those with HO candidemia (Figure 2a, p < 0.001). This survival advantage of NHO candidemia remained after stratifying Candida species with C. albicans, C. tropicalis, and C. glabrata (Figure 2b–d).
Figure 2. Survival curves analysis. A log-rank test was applied to assess the statistical significance, and a p value less than 0.05 was considered statistically significant. (a) survival comparison between non-hospital-onset (NHO) candidemia and hospital-onset (HO) candidemia within all patients. (b) survival comparison between NHO candidemia and HO candidemia within groups based on isolation of C. albicans vs. other species. (c) survival comparison between NHO candidemia and HO candidemia within groups based on isolation of C. tropicalis vs. other species. (d) survival comparison between NHO candidemia and HO candidemia within groups based on isolation of C. glabrata vs. other species. Abbreviations: NHO, non-hospital-onset; HO, hospital-onset
The multivariate Cox proportional-hazards model for risk factors associated with in-hospital mortality showed that NHO candidemia was a better predictor of outcome (adjusted hazard ratio, 0.44; 95% CI, 0.29–0.67, Table 4) regardless of age, sex, and CCI. Other variables that can predict poor clinical outcomes included recent steroid use, retained chemo port, acute liver failure, and the occurrence of septic shock during hospitalisation. On the contrary, antifungal therapy initiation within 2 days was a protective factor, reducing 40% and 38% mortality risk by using either triazole or echinocandin.
Table 4. Univariate and multivariate analysis of risk factors associated with in-hospital mortality by Cox proportional-hazards model
Abbreviations: aHR, adjusted Hazard ratio; HR, hazard ratio.
* p < 0.05.
Discussion
In this retrospective study, we investigated the differences in clinical and mycological characteristics and outcomes between patients with NHO and HO candidemia. Patients with NHO candidemia were more likely to be infected by C. glabrata; nonetheless, C. tropicalis was more commonly isolated from patients with HO candidemia. After adjusting for covariates, patients with NHO candidemia were independently associated with lower in-hospital mortality. We also identified a number of mortality-relevant risk factors, including recent steroid use, presence of a chemo port without removal, acute liver failure, and septic shock. On the other hand, initiation of antifungal therapy within 2 days with either triazole or echinocandin was a protective factor for in-hospital mortality.
The global distribution of Candida species showed a decline in the isolation of C. albicans accompanied by rises in the isolation of C. parapsilosis and C. glabrata in the SENTRY antifungal surveillance program, including in Asia countries [Reference Pfaller, Diekema, Turnidge, Castanheira and Jones15]. However, our data revealed that C. albicans was still the most numerous species, followed by C. tropicalis, C. glabrata, and C. parapsilosis. Meanwhile, Sofair et al. [Reference Sofair, Huie-White, Reiss, Harrison, Sanza, Arthington-Skaggs and Fridkin14] conducted an active, population-based surveillance study involving 1,143 patients with candidemia in the United States, showing the association between strain types and epidemiologic classification. The study reported that while C. albicans accounted for the most common species, particularly in HO candidemia, C. parapsilosis was more proportionally associated with NHO candidemia, especially in cases where hospitalization occurred further in the past [Reference Sofair, Huie-White, Reiss, Harrison, Sanza, Arthington-Skaggs and Fridkin14]. In our study, we found a relatively high proportion of C. glabrata among those with NHO candidemia, which may reflect the predominance of C. glabrata colonisation among patients in the community. This is in accordance with the SENTRY program, which ranked C. glabrata as the second most predominant species between 1997 and 2016 [Reference Pfaller, Diekema, Turnidge, Castanheira and Jones15]. Furthermore, C. tropicalis appears to be predominant in patients exposed to the hospitalised environment. This high percentage of C. tropicalis could be associated with a relatively higher proportion of underlying disease with malignancy in the patients with HO candidemia compared with NHO candidemia (60.31% vs. 48.78%, p = 0.066), as shown in previous studies [Reference Pfaller and Diekema6, Reference Tan, Chakrabarti, Li, Patel, Watcharananan, Liu, Chindamporn, Tan, Sun, Wu, Chen and Asia Fungal Working7, Reference Hung, Chen, Chang, Luh and Hsieh26]. Taken together, these pieces of evidence demonstrate that C. albicans remains the most commonly isolated species in patients with candidemia, and those with NHO candidemia are more likely to be affected by C. glabrata. The differences observed between HO and NHO candidemia can aid in identifying risk factors, predicting species, selecting appropriate treatment, and ultimately improving outcomes by addressing modifiable risks.
In spite of various antifungal therapy, candidemia is still a lethal infectious disease, leading to as high as 55.75% of in-hospital mortality in our cohort, which was comparable to previous studies [Reference Hii, Chang, Lin, Lee, Liu, Liu, Chen, Cheng and Chang3, Reference Chen, Liao, Kuo, Chen, Chen, Wang, Yang and Fung27–Reference Ruan and Hsueh29]. A previous study [Reference Sofair, Huie-White, Reiss, Harrison, Sanza, Arthington-Skaggs and Fridkin14] showed that patients with NHO candidemia had a lower risk of 30-day mortality compared with those with HO candidemia (RR, 0.64; 95% CI: 0.53–0.78, p < 0.01); however, Kato reported that NHO candidemia was not a protective factor [Reference Kato, Yoshimura, Suido, Shimizu, Ide, Sugiyama, Matsuno and Nakajima16]. In our study, patients with NHO candidemia presented a 56% reduction of hazard ratio for in-hospital mortality compared with those with HO candidemia in multivariate Cox proportional-hazards models involving other variables. Despite the fact that a higher proportion of patients with malignancy and more patients in the NHO group were diagnosed with DM than those in the HO candidemia group, these two factors did not have an impact on the in-hospital mortality advantage of NHO over HO candidemia by using Kaplan–Meier survival curves, including different Candida species (log-rank tests, all p < 0.05). Whether there is an impact of different Candida species on clinical outcomes, particularly mortality, is still debatable. While some studies revealed no association between species and outcome [Reference Muderris, Kaya, Ormen, Aksoy Gokmen, Varer Akpinar and Yurtsever Gul30–Reference Tang, Liu, Lin and Lai32], others have reported that infection caused by different species might be associated with mortality [Reference Hii, Chang, Lin, Lee, Liu, Liu, Chen, Cheng and Chang3, Reference Tang, Liu, Lin and Lai4, Reference Andes, Safdar, Baddley, Playford, Reboli, Rex, Sobel, Pappas and Kullberg33]. C. parapsilosis fungemia was found to be associated with a better outcome compared with other species or mixed fungemia by the Kaplan–Meier survival curve analysis (p = 0.044). However, such a benefit was not noted in a multivariate logistic regression analysis of 30-day mortality (aOR, 0.63, 95% CI: 0.29–1.34, p = 0.23) in Japan [Reference Kato, Yoshimura, Suido, Shimizu, Ide, Sugiyama, Matsuno and Nakajima16]. Unlike Kato’s study, our study reported that candidemia due to different species did not reveal significant influences on in-hospital mortality and the superiority of NHO candidemia (Figure 2b–d).
Some recent studies revealed the incidence of candidemia had been decreasing [Reference Cleveland, Harrison, Farley, Hollick, Stein, Chiller, Lockhart and and Park34, Reference Toda, Williams, Berkow, Farley, Harrison, Bonner, Marceaux, Hollick, Zhang, Schaffner, Lockhart, Jackson and Vallabhaneni35] in comparison to previous epidemiologic reports in the 2000s. Suzuki et al. [Reference Suzuki, Perencevich, Diekema, Livorsi, Nair, Kralovic, Roselle and Goto17] reported a 77.1% reduction in incidence rates in HO candidemia since its peak in 2004, after a series of infection control interventions. Regardless, the fact that the onset of nosocomial candidemia can occur for as long as 20 days from hospital admission has highlighted the importance of healthcare-associated infection control interventions. The percentage of NHO candidemia ranged from 17% to 31.14% [Reference Sofair, Huie-White, Reiss, Harrison, Sanza, Arthington-Skaggs and Fridkin14, Reference Kato, Yoshimura, Suido, Shimizu, Ide, Sugiyama, Matsuno and Nakajima16] and varied with time and countries [Reference Pfaller, Moet, Messer, Jones and Castanheira21]. This small proportion of NHO candidemia might be highlighted after a descent of HO candidemia by introducing multifaceted infection control interventions. Thus, it is crucial to pay more attention to these distinct characteristics and clinical outcomes of NHO and HO candidemia to customise treatment strategies. The different characteristics observed in HO and NHO candidemia could be extremely useful for clinicians to judge the most possible risk factors, select the most appropriate treatment, and ultimately improve clinical outcomes.
There were several limitations in our study. First, this was a retrospective cohort study which may have introduced potential bias due to non-standardised data collection procedures. Second, no other in-vitro resistance mechanism was explored in this study. Not all isolates had antifungal breakpoints reference based on CLSI, though those strains without breakpoints were only 4.65% (16/344, Supplementary Table 1). Further studies are also necessary for elucidating the resistance mechanism of those strains.
Conclusion
In conclusion, NHO candidemia accounted for 24.18% of all candidemia cases and was associated with a lower in-hospital mortality. A higher percentage of C. glabrata was isolated from those with NHO candidemia, while C. tropicalis was more prevalent in the HO group. Patients with organ failure, retained chemo port, or septic shock, or those who received steroids had significantly higher mortality. Nonetheless, initiation of antifungal therapy within 2 days of diagnosis was a protective factor against in-hospital mortality.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/S0950268823000894.
Data availability statement
All collected data used for the analysis in this study are available from the corresponding author at reasonable request.
Acknowledgements
We would like to express our deep gratitude to Chin-Fu Lin, I-Hua Chou, and all staff in the Microbiology Laboratory of the Department of Pathology & Laboratory Medicine and Clinical Informatics Research & Development Center of Taichung Veterans General Hospital.
Author contribution
Acquisition of data (laboratory or clinical): Y.-C.C., C.-C.C.; Conception and design of the study: Y.-C.C., C.-C.C.; Data analysis and/or interpretation: Y.-C.C., M.-W.H., W.-C.C., C.-C.C.; Drafting of the manuscript and/or critical revision: Y.-C.C., M.-W.H., W.-C.C., C.-C.C.
Financial support
This study was supported by Taichung Veterans General Hospital (Grant Nos. TCVGH-1104403B and TCVGH-1114404B), Taichung, Taiwan. The funders had no role in the study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Competing interest
The authors have no competing interest to declare.
Open access
