key: cord-0947309-l9j2opyd
authors: Li, Yaling; Wu, Yutong; Gao, Yali; Niu, Xueli; Li, Jingyi; Tang, Mingsui; Fu, Chang; Qi, Ruiqun; Song, Bing; Chen, Hongduo; Gao, Xinghua; Yang, Ying; Guan, Xiuhao
title: Machine-learning based prediction of prognostic risk factors in patients with invasive candidiasis infection and bacterial bloodstream infection: a singled centered retrospective study
date: 2022-02-13
journal: BMC Infect Dis
DOI: 10.1186/s12879-022-07125-8
sha: 0502bdb4d7f7e949b49b9f93ddde681ed32aec71
doc_id: 947309
cord_uid: l9j2opyd

BACKGROUND: Invasive candidal infection combined with bacterial bloodstream infection is one of the common nosocomial infections that is also the main cause of morbidity and mortality. The incidence of invasive Candidal infection with bacterial bloodstream infection is increasing year by year worldwide, but data on China is still limited. METHODS: We included 246 hospitalised patients who had invasive candidal infection combined with a bacterial bloodstream infection from January 2013 to January 2018; we collected and analysed the relevant epidemiological information and used machine learning methods to find prognostic factors related to death (training set and test set were randomly allocated at a ratio of 7:3). RESULTS: Of the 246 patients with invasive candidal infection complicated with a bacterial bloodstream infection, the median age was 63 years (53.25–74), of which 159 (64.6%) were male, 109 (44.3%) were elderly patients (> 65 years), 238 (96.7%) were hospitalised for more than 10 days, 168 (68.3%) were admitted to ICU during hospitalisation, and most patients had records of multiple admissions within 2 years (167/246, 67.9%). The most common blood index was hypoproteinemia (169/246, 68.7%), and the most common inducement was urinary catheter use (210/246, 85.4%). Moreover, the most frequently infected fungi and bacteria were Candida parapsilosis and Acinetobacter baumannii, respectively. The main predictors of death prognosis by machine learning method are serum creatinine level, age, length of stay, stay in ICU during hospitalisation, serum albumin level, C-Reactive protein (CRP), leukocyte count, neutrophil count, Procalcitonin (PCT), and total bilirubin level. CONCLUSION: Our results showed that the most common candida and bacteria infections were caused by Candida parapsilosis and Acinetobacter baumannii, respectively. The main predictors of death prognosis are serum creatinine level, age, length of stay, stay in ICU during hospitalisation, serum albumin level, CRP, leukocyte count, neutrophil count, PCT and total bilirubin level. SUPPLEMENTARY INFORMATION: The online version contains supplementary material available at 10.1186/s12879-022-07125-8.

Invasive candidal infection and bacterial bloodstream infections are common in hospitals [1, 2] . It is reported that approximately 1.5 million patients with invasive candidal infection die every year worldwide [3, 4] . Bacterial bloodstream infection is the seventh leading cause of death in North America and Europe [5] . In addition, in these regions, the average annual mortality is 29 cases per 100,000 population, and the total mortality is between 13 and 20% [5] . However, the epidemiological characteristics of both invasive candidal infection and bacterial bloodstream infections differ with geographic location and time [6] [7] [8] [9] [10] [11] [12] . In the past two decades, the incidence of non-Candida albicans infection has increased. A recent study from Japan revealed that Candida albicans was the infectious agent in 58.2% of all candidiasis cases in 2003 but only in 30% of cases in 2014. In addition, a study from China reported that Candida albicans was the causative agent in only 44.9% of invasive candidal infection cases [13] , which is consistent with our previous study, which revealed that Candida albicans is no longer the most common invasive fungus [14] . The risk of death owing to invasive fungal and bacterial bloodstream infections puts enormous pressure on healthcare services, leading to a shortage of intensive care resources. However, in previous studies, the epidemiological characteristics and risk factors of invasive fungal infections complicated with bacterial bloodstream infections were rarely discussed, possibly owing to the limited capacity of some methods in analysing large datasets.

Machine learning techniques have the unique ability to deal with extensive data because they can process large datasets in a flexible and trainable manner and understand the complex relationship between variables [15] . Owing to their improved processing ability, various machine learning and artificial intelligence techniques are widely used to identify risk and prognostic factors of disease in patients to help clinicians. Therefore, we conducted a retrospective analysis of patients with invasive candidal infection concomitant with bacterial bloodstream infection and identified the prognostic indicators of death using machine learning methods.

Patients were selected as previously described [14] . We collected all data on Candida, Cryptococcus and other yeast isolates recovered from the blood, ascitic fluid, peritoneal dialysate fluid, pus and tissues of patients with invasive candidal infection (2008 version of EORTC/ MSG criteria). The onset of bacterial bloodstream infection or invasive candidal infection was defined as the date when the first positive result of blood culture was obtained. The data collected included patient characteristics at baseline, haematological diagnoses and chemotherapy, risk factors for invasive candidal infection, clinical features of invasive candidal infection, Candida test results, bacterial test results, antifungal prophylaxis and treatment and survival status at discharge. In addition, data regarding the management of patients receiving antifungal prophylaxis or therapy were recorded, including the date and nature of the change in treatment and survival status at discharge. The hospitalisation of each patient represented one event, and if a patient was re-hospitalised and received another round of treatment, he/she was considered a new event. Persistent candidal infections was defined as persistent if positive blood culture results were obtained for the same Candida species 7 days after the initiation of appropriate antifungal therapy [16] . We excluded non-Candida yeast samples and samples from non-sterile sources, such as faeces, urine, sputum, pharyngeal swabs and pus.

Aseptic humoral samples (8-10 mL) were collected and cultured for 5 days. Samples with positive results were transferred to blood agar plates, and subsequently, bacterial and fungal isolates were cultured at 35 °C for 48-72 h. Gram staining and microscopic examination were performed simultaneously. Strains (bacterial and fungal isolates) were identified on a VITEK 2 Compact system (Bio-Merieux SA, Marcy l 'etoile, France), and susceptibility tests were performed using the ATB FUN-GUS 3 kit (Bio-Merieux SA, Marcy l 'etoile, France).

The minimum inhibitory concentration (MIC) was determined according to the CLSI m27-a3 and m27-s4 antifungal susceptibility test standards. The quality Conclusion: Our results showed that the most common candida and bacteria infections were caused by Candida parapsilosis and Acinetobacter baumannii, respectively. The main predictors of death prognosis are serum creatinine level, age, length of stay, stay in ICU during hospitalisation, serum albumin level, CRP, leukocyte count, neutrophil count, PCT and total bilirubin level.

Keywords: Bacterial bloodstream infection, China, Epidemic, Invasive candidal infection, Machine learning control strains used were Candida ATCC6258 and Candida albicans ATCC90028.

We pre-processed the data and deleted missing cases with 50% features, and the mean value of missing values was filled. The dataset was randomly divided into the training and test sets (7:3), with 70% patients in the training set and 30% patients in the test set. We used the random forest, logistic regression and supportvector machine algorithms to build a prediction model. Subsequently, the trained random forest model was analysed to evaluate the feature importance ranking.

The IBM SPSS Statistics for Windows version 20.0 software (IBM Corp., Armonk, NY, USA) was used for statistical analysis. Non-normally distributed quantitative data were expressed as median and quartile ranges [M (P25, P75)] and analysed using the Mann-Whitney test for intergroup comparisons. Qualitative data were represented by relative numbers, and the chi-square test was used for intergroup comparisons.

ICU, intensive care unit; SDD, susceptible-dosedependent; PCT, procalcitonin; CRP, C-reactive protein; BDG, 1-3-β-d-glucan.

Prolonged hospitalisation was defined as hospital stay longer than 10 days. Surgery was defined as thoracic and abdominal surgeries. Recent surgery was defined as surgery performed 14 days before the first diagnosis of Candida infection. Abdominal surgery was defined as any surgery involving organs including the stomach, small intestine, colon or rectum, gallbladder, liver, pancreas, spleen and appendix. Concerning laboratory results, renal failure was defined as creatinine clearance < 60 mL/min, hypoalbuminaemia was defined as serum albumin concentration < 30 g/L and leukopaenia was defined as peripheral white blood cell count < 4 × 10 9 cells/L. Prolonged ICU stay was defined as ICU stay for more than 10 days. Long-term and combined use of multiple antibiotics were defined as the use of antibiotics for more than 14 days and the simultaneous use of more than 2 antibiotics, respectively. Multiple bacterial infections were defined as infections with more than two types of bacteria simultaneously. Multiple fungal infections were defined as infections with more than two types of fungi simultaneously. (2), Candida streptococcus (1) and Cryptococcus neoformans (1) (Fig. 1A) .

Furthermore, 28 species of bacteria were isolated; of which, 15 (53.6%) were Gram-positive and 13 ( 1B ). In addition, there were 73 (29.7%) cases of single and 173 (70.3%) cases of multiple bacterial bloodstream infections. Detailed data can be found in Additional file 2: Table S2 .

We obtained 239 isolates from 246 patients with drug sensitivity; of which, 19 (7.9%) isolates were resistant to at least one antifungal agent. Amphotericin B showed excellent results, as all strains were sensitive to it. The drug sensitivity of voriconazole and fluorouracil was good, achieving an efficiency of 96.6% (230/238) and 98.3% (235/239), respectively. In addition, the drug sensitivity of fluconazole and itraconazole was 85.8% (205/239) and 90.4% (216/239), respectively. Candida glabrata isolates were highly susceptible to fluconazole (18/26, 69.2%) and itraconazole (10/26, 38.5%) in a dose-dependent manner. Candida tropicalis isolates exhibited considerable resistance to fluconazole (5/17, 29.4%) and voriconazole (5/17, 29.4%), whereas Candida krusei isolates exhibited strong resistance to fluconazole (3/4, 75%). Detailed data can be found in Additional file 3: Table S3 .

The demographic and clinical characteristics of patients with Candida albicans and non-Candida albicans infections are shown in Furthermore, the C-reactive protein (CRP) and procalcitonin (PCT) levels were markedly elevated in patients with invasive candidal infection complicated with bacterial bloodstream infection, especially in those with Candida albicans infection. In addition, the CRP and PCT levels were higher in patients with Candida albicans infection than in patients with non-Candida albicans infection, and the difference was statistically significant. However, both leukocyte and lymphocyte counts were within the normal range. Detailed data are provided in Table 1 . 

The demographic and clinical characteristics of patients with persistent and non-persistent Candida infection are shown in Table 2 . Persistent Candida infection was associated with diabetes, longer stay in the ICU and renal failure. Differences were statistically significant.

Of the 246 patients with both Candida and bloodstream infections, 70 (28.45%) had multi-candidal infection and 176 (71.55%) patients had single candidal infection, and the demographic and clinical characteristics of patients are shown in Table 3 .

Furthermore, the duration of hospital and ICU stays was longer in patients with multi-candidal infection than in patients with single candidal infection (hospital stay: 57 versus 42 days, respectively, based on the median; ICU stay: 22.5 versus 7 days, respectively, based on the median). In addition, patients with multicandidal infection were more likely to have diabetes and develop septic shock. Furthermore, more than half (51.70%, 91/176) and approximately one-third (30%, 21/70) of post-surgical patients had multi-candidal infection. Moreover, when infected patients (not only post-surgical patients) were considered, 77.14% (84/176) patients with multi-candidal infection and 47.73% (54/70) patients with single candidal infection developed persistent infection, with increased CRP and PCT levels. The lymphocyte count was distinctly reduced in patients with single candidal infection (0.75 × 10 9 /L based on the median) but only slightly reduced in patients with multi-candidal infection Table 1 Risk factors for Candida albicans and non-Candida albicans infections a Is described by median and quartile, and the statistic was the Z value; other items were described as numbers (n-%) and the statistic was the χ 2 (1.02 × 10 9 /L based on the median). Differences were statistically significant.

We used random forest, logistic regression and support-vector machine algorithms to develop a prediction model, and the performance evaluation is shown in Table 4 . Figure 2 demonstrates the ROC curves of the prediction model. Based on analysis and training, it was found that the random forest model exhibited the best performance. A random forest model is usually used to examine the importance of different features. The most predictive characteristics of invasive candidal infection concomitant with bacterial bloodstream infection were identified to be serum creatinine, serum albumin, CRP, PCT and total bilirubin levels; age; length of stay in the hospital; stay in ICU during hospitalisation and leukocyte and neutrophil counts (Table 5 ).

To date, there have been a few epidemiological studies on patients with concomitant invasive candidal infection and bacterial bloodstream infections. We included 246 patients with concomitant invasive candidal infection and bacterial bloodstream infections admitted to a provincial medical centre in northeast China between January 2013 and January 2018. Using machine learning techniques, we found that the main predictors of death were serum creatinine, serum albumin, CRP, PCT and total bilirubin levels; age; length of stay in the hospital; stay in ICU during hospitalisation and leukocyte and neutrophil counts. The random forest model with these 10 features showed satisfactory performance, and the AUC value in the training and test sets was 0.919. Table 2 Risk factors in patients with persistent and non-persistent candidal infections a Is described by median and quartile, and the statistic was the Z value; other items were described as numbers (n-%) and the statistic was the χ 2 value b Statistic was the Fisher χ 2 value Furthermore, the epidemiological survey revealed that 96.7% (238/246) patients were hospitalised for more than 10 days, and 68.3% (168/246) patients were admitted to ICU. Most patients had multiple admissions in the past 2 years (167/246, 67.9%) and had hypoproteinaemia (169/246, 68.7%). These conditions reflect the physical characteristics of patients, which are similar to those reported in recent studies [17, 18] suggesting that infections may be associated with invasive medical operations, especially owing to long-term catheter retention. Similar results have been reported in recent studies as well [19] [20] [21] [22] [23] [24] [25] [26] .

Candida parapsilosis was the most common causative fungal agent (92/246, 37.4% patients), followed by Candida guilliermondi (53/246, 21.5% patients), Candida albicans (49/246, 19 .9% patients), Candida glabrata (26/246, 10.6% patients), Candida tropicalis (18/246, Table 3 Analysis of risk factors in patients with single candidal infection and multiple candidal infections a Is described by median and quartile, and the statistic was the Z value; other items were described as numbers (n-%) and the statistic was the χ 2 value, [27, 28] . In this study, the main predictors of death were serum creatinine, serum albumin, CRP, PCT and total bilirubin levels; age; length of stay in the hospital; stay in ICU during hospitalisation and leukocyte and neutrophil counts. High serum creatinine level is a risk factor for bacterial bloodstream infection and may be associated with renal insufficiency [29, 30] . In addition, age is a significant prognostic risk factor for nosocomial infections, and elderly patients are more likely to present with underlying diseases, low immunity and decreased organ function, which makes them more susceptible to invasive candidal infection/bacterial bloodstream infection [31, 32] . The length of stay in the hospital is an important index influenced by many factors, including the demographic characteristics, treatment complexity, complications and discharge plan of patients, and can be used as a predictor of death [33, 34] . Studies have shown that the overall mortality rate of hospitalised patients increases with the increasing duration of ICU stay, possibly owing to complications resulting from long-term intensive care [35, 36] . In addition, serum albumin level is a nutritional index and an important indicator of morbidity and mortality in critically ill patients. Low serum albumin level is an important and unique predictor of mortality [37, 38] . CRP is a classic indicator of infection. Previous studies have shown that CRP can also be used as a prognostic indicator for hospitalised patients [39, 40] . In addition, this study shows that increased leukocyte counts indicate increased mortality in hospitalised patients with infection. Similar studies have shown that the death rate of patients with cancer and dengue increases with increasing leukocyte counts [41, 42] . Infections destroy the dynamic balance of the immune system and cause significant changes in the neutrophil count, which are closely related to mortality [43, 44] . Furthermore, PCT is a classic indicator of infection, and recent studies have shown that PCT can also be used as a prognostic indicator for hospitalised patients with infection [45, 46] . In addition, total bilirubin levels can be used as a prognostic indicator in patients with coronavirus infection, respiratory tract infection and cardiogenic shock, and increased serum bilirubin levels are independently associated with mortality [47] [48] [49] . However, this study had some limitations. First, this is a single-centre study. Therefore, the results and conclusions may be affected by geographical location, hospital management strategies, infection control policies and susceptibility models. Second, owing to a retrospective design, some key factors of concomitant invasive candidal infection and bacterial bloodstream infections may have been ignored. In addition, to the best of our knowledge, machine learning was used for the first time in this study to predict the risk factors of death and prognosis of concomitant invasive candidal infection and bacterial bloodstream infections. Moreover, the relatively small sample size may affect the credibility of the results. Therefore, further large-scale, multi-centre prospective studies should be conducted to validate the results of this study.

The most common Candida and bacterial species in patients with concomitant Candida and bacterial bloodstream infections in the First Hospital of the China Medical University were Candida parapsilosis and Acinetobacter baumannii, respectively. The main predictors of death were serum creatinine, serum albumin, CRP, PCT and total bilirubin levels; age; length of stay in the hospital; stay in ICU during hospitalisation and leukocyte and neutrophil counts.

Trends among pathogens reported as causing bacteraemia in England

Invasive candidiasis

Hidden killers: human fungal infections

Superficial fungal infections

Overall burden of bloodstream infection and nosocomial bloodstream infection in North America and Europe

Nosocomial fungal infections: epidemiology, infection control, and prevention

Nosocomial fungal infections: epidemiology, diagnosis, and treatment

Epidemiology and outcomes of bloodstream infections in 177 severe burn patients from an industrial disaster: a multicentre retrospective study

Placing the burden of bacteraemia in perspective

Trends in bacteremia over 2 decades in the top end of the northern territory of Australia

Population-based bloodstream infection surveillance in rural Thailand

Bacterial and fungal pathogens isolated from patients with bloodstream infection: frequency of occurrence and antimicrobial susceptibility patterns from the SENTRY Antimicrobial Surveillance Program

Five-year national surveillance of invasive candidiasis: species distribution and azole susceptibility from the China hospital invasive fungal surveillance net (CHIF-NET) Study

A 5-year review of invasive fungal infection at an academic medical center

How machine learning will transform biomedicine

Clinical impact of time to positivity for Candida species on mortality in patients with candidaemia

Clinical characteristics, risk factors and outcomes of mixed Candida albicans/bacterial bloodstream infections

Mixed bloodstream infections involving bacteria and Candida spp

Prevention of central line-associated bloodstream infections

Catheter-associated urinary tract infection, Clostridioides difficile Colitis, central line-associated bloodstream infection, and methicillin-resistant Staphylococcus aureus

A prospective, multicenter case control study of risk factors for acquisition and mortality in Enterobacter species bacteremia

Candida urinary tract infections in adults

Epidemiology, species distribution, and predictive factors for mortality of candidemia in adult surgical patients

Catheter-related bloodstream infections

Clinical management of catheterrelated infections

Burden of serious fungal infections in the Netherlands

Follow-up creatinine level is an important predictive factor of in-hospital mortality in cirrhotic patients with spontaneous bacterial peritonitis

Systematic review on the definition and predictors of severe Clostridiodes difficile infection

Bacteremia caused by Acinetobacter baumannii among patients in critical care

Nosocomial bloodstream infections due to Candida spp. in the USA: species distribution, clinical features and antifungal susceptibilities

Patient length of stay and mortality prediction: a survey

Mortality, readmission and length of stay have different relationships using hospital-level versus patient-level data: an example of the ecological fallacy affecting hospital performance indicators

Relationship between ICU length of stay and long-term mortality for elderly ICU survivors

Mortality and functional status at one-year of follow-up in elderly patients with prolonged ICU stay

Serum albumin trend is a predictor of mortality in ICU patients with sepsis

Trends in admission serum albumin and mortality in patients with hospital readmission

Association between severity grading score and acute phase reactants in patients with Crimean Congo hemorrhagic fever

Infection biomarkers in assisting the judgement of blood stream infection and patient prognosis: a retrospective study incorporating principal components analysis

Prognostic model for patients with advanced cancer using a combination of routine blood test values

Early hematological parameters as predictors for outcomes in children with dengue in northern India: a retrospective analysis

The trajectory of alterations in immunecell counts in severe-trauma patients is related to the later occurrence of sepsis and mortality: retrospective study of 917 cases

Severe influenza with invasive pulmonary aspergillosis in immunocompetent hosts: a retrospective cohort study

Early procalcitonin assessment in the emergency department in patients with intraabdominal infection: an excess or a need

National trends in the distribution of Candida species causing candidemia in Japan from

Association of inflammatory biomarkers with subsequent clinical course in suspected late onset sepsis in preterm neonates

Bilirubin levels as potential indicators of disease severity in coronavirus disease patients: a retrospective cohort study

convenient online submission • thorough peer review by experienced researchers in your field • rapid publication on acceptance • support for research data, including large and complex data types • gold Open Access which fosters wider collaboration and increased citations maximum visibility for your research: over 100M website views per year submit your research ?

Prognostic significance of end-stage liver diseases, respiratory tract infection, and chronic kidney diseases in symptomatic acute hepatitis E

Bilirubin-a possible prognostic mortality marker for patients with ECLS

Publisher's Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations

The authors thank the patients, their families, and all investigators who participated in the study.

The online version contains supplementary material available at https:// doi. org/ 10. 1186/ s12879-022-07125-8.Additional file 1. Detailed data of patients with invasive Candida infection complicated with bacterial bloodstream infection.Additional file 2. Detailed data on bacterial species in patients with bacterial bloodstream infection.Additional file 3. Detailed data of drug sensitivity results.Authors' contributions LYL and GXH were responsible for the study concept and design. LYL and WYT were responsible for data acquisition and data extraction. Data analysis was performed by LYL, WYT and LJY. The paper was drafted by LYL and GXH. All authors supervised the study. All authors read and approved the final manuscript.

National Science and Technology Major Projects of China, Grant/Award Number (2018ZX10101003 and 2018ZX10712001).

The data supporting the findings of this study from the corresponding author upon request. If someone wants to request the data from this study, please contact Xiuhao Guan.

The study was conducted in accordance with the Declaration of Helsinki. This study was approved by The Human Ethics Review Committee of the First Hospital of China Medical University (No. 2021-260). The ethics review board of the First Hospital of China Medical University exempted the acquisition of informed consent because this was a retrospective study. Patients' data confidentiality was fully respected during data collection and the preparation of the manuscript.

Not applicable.

The authors declare that they have no competing interests.