CN117877649B

CN117877649B - Marker group and system for predicting bacteremia of tumor patient

Info

Publication number: CN117877649B
Application number: CN202311654435.3A
Authority: CN
Inventors: 李小辉; 陈树林
Original assignee: Sun Yat Sen University Cancer Center
Current assignee: Sun Yat Sen University Cancer Center
Priority date: 2023-12-05
Filing date: 2023-12-05
Publication date: 2024-05-31
Anticipated expiration: 2043-12-05
Also published as: CN117877649A

Abstract

The invention discloses a marker group for predicting bacteremia of a tumor patient, wherein the marker group comprises monocyte number, alanine aminotransferase, C-reactive protein/procalcitonin ratio and derivative neutrophil/lymphocyte ratio. In the invention, the clinical characteristics and blood inflammation indexes of individuals are analyzed, and a new marker group and a new model for predicting the occurrence of bacteremia of cancer patients are established. The new marker sets and new models improve the accuracy of predictions and the ability to distinguish bacteremia from non-bacteremia patients compared to traditional diagnostic markers WBC, CRP and PCT. The new model successfully divides these cancer patients into high-risk and low-risk groups and can predict the etiology of bacteremia, while traditional indicators predict the etiology with poor accuracy. Moreover, the model provided by the invention can be a potential prediction tool for clinicians due to low cost and good stability.

Description

Marker group and system for predicting bacteremia of tumor patient

Technical Field

The invention belongs to the technical field of biomedicine, and particularly relates to a blood biomarker for predicting bacteremia of a tumor patient.

Background

Bacteremia refers to a disease caused by pathogenic bacteria locally invading blood flow, and temporarily passing through blood circulation to reach a proper position in the body and then multiplying. Clinical manifestations of bacteremia are systemic chills, fever and hepatosplenomegaly. Severe conditions that can lead to septic shock and multiple organ failure, particularly malignant tumors, can lead to death. Despite significant advances in cancer treatment, bacteremia remains a major complication that endangers the life of cancer patients, with mortality rates of 14% -42%; and bacteremia can delay patient chemotherapy and extend their length of stay, increasing their economic pressure. The results of the study showed that the optimal time for bacteremia treatment in cancer patients was within 48 hours of blood culture positivity. Early diagnosis and effective empirical antibacterial treatment can reduce mortality from bacterial infection in cancer patients and improve the prognosis of bacteremia.

Currently, blood culture is still the gold standard for diagnosing bacteremia. However, gold standards limit early diagnosis and clinical guidance for bacteremia due to low sensitivity and long time to report sun and may be affected by a number of factors, such as the type of pathogen and antibiotic treatment. To remedy the deficiencies of blood culture, some new methods have been used for the assisted diagnosis of bacteremia, such as matrix-assisted laser desorption/ionization time-of-flight mass spectrometry (MALDI-TOF-MS), real-time fluorescent quantitative PCR and polypeptide nucleic acid fluorescent in situ hybridization. Although in some studies, the time to report the sun is earlier than in blood culture, there are still problems and limitations in practical use due to complex experimental operations, expensive equipment and uncertain use value.

When a pathogen enters the circulatory system and interacts with the body, the body will produce a series of responses against infection, including serum markers produced by immune cells. These markers, such as White Blood Cells (WBCs), procalcitonin (PCT), cytokines and chemokines and acute phase response proteins, are of value in the early diagnosis of assisted bacteremia. Traditional studies have focused on single markers that have some ability to identify only a single or specific type of bacteremia or tumor.

Disclosure of Invention

The present invention aims to overcome at least one of the deficiencies of the prior art and to provide a blood biomarker for predicting bacteremia in a patient with a tumor.

The technical scheme adopted by the invention is as follows:

in a first aspect, the present invention provides a system for predicting bacteremia in a patient with a tumor, comprising the following means:

1) A quantification device for determining the value of the monocyte count, alanine aminotransferase, C-reactive protein/procalcitonin ratio, and derived neutrophil/lymphocyte ratio in a blood sample of a patient;

2) The prediction analysis device inputs the result of the quantitative device into a risk assessment formula, and performs prediction analysis on the result of the risk assessment formula so as to distinguish bacteremia patients from non-bacteremia patients;

3) The result output device is used for outputting the result obtained by analysis of the predictive analysis device;

Wherein, the risk assessment formula is: risk score = -0.0912-0.0263 x monocyte number + 0.0012 x alanine aminotransferase + 0.0021 x C reactive protein +0.0006 x C reactive protein/procalcitonin ratio + 0.023 x derived neutrophil/lymphocyte ratio, where monocyte number units are x 10 x 9/L, alanine aminotransferase units are U/L, C reactive protein units are mg/L.

In some examples, the predictive analysis is based on a calibration curve and a nomogram.

In a second aspect, the invention provides a kit for early detection of bacteremia in a tumour patient, the kit comprising the system according to the first aspect.

The beneficial effects of the invention are as follows:

In the invention, the clinical characteristics and blood inflammation indexes of individuals are analyzed, and a new marker group and a new model for predicting the occurrence of bacteremia of cancer patients are established. The new model improves the accuracy of the predictions and the ability to distinguish bacteremic and non-bacteremic patients compared to the traditional markers WBC, CRP and PCT. The new model successfully divided these cancer patients into high-risk and low-risk groups, which differ significantly in terms of the prediction of bacteremia. Meanwhile, the new model can also predict the etiology of bacteremia, and the traditional index has poorer accuracy for predicting the etiology. Moreover, the model provided by the invention can be a potential prediction tool for clinicians due to low cost and good stability.

Drawings

FIG. 1 is a graph of the trajectory variation of each predicted variable analyzed using LASSO regression analysis, where the optimal value of B graph lambda is determined using 10-fold cross-validation of the minimum criteria.

Figure 2 is the dynamic AUC levels for the four models in the derivation (a) and validation (B) queues.

FIG. 3 shows DCA and CIC of different diagnostic models in the development queue (panels A and C) and validation queue (panels B and D).

FIG. 4 is a nomogram and calibration of predicted bacteremia in the development queues (A and B) and validation queues (B and D).

FIG. 5 shows the correlation between the prediction model and WBC, CRP, PCT, blue positive correlation, red negative correlation, A diagram shows the derived queue, and B diagram shows the verified queue.

Detailed Description

The following disclosure provides many different embodiments, or examples, for implementing different aspects of the invention.

1) Materials and methods

The invention collects the following clinical characteristics of 135 cases of blood culture positive and 116 cases of blood culture negative tumor patients in the period from 2020 month 1 to 2022 month 12 of the university of Zhongshan tumor prevention center:

Sex, age, fever, identification of blood culture (e.g., gram negative bacteria, gram positive bacteria, fungi), white Blood Cells (WBC), neutrophils (NEU), lymphocytes (LYM), monocytes (MO), hemoglobin (HGB), neutrophil/lymphocyte ratio (NLR), platelet/lymphocyte ratio (PLR), derivatized neutrophil/lymphocyte ratio (dNLR), lymphocyte/monocyte ratio (LMR), platelets (PLT), alanine Aminotransferase (ALT), oxamic Aminotransferase (AST), albumin (ALB), C-reactive protein (CRP), cystatin C (CYSC), procalcitonin (PCT), AST/transaminase ratio (SLR), ALB/CRP ratio (ACR), C-reactive protein x procalcitonin (CPT), albumin/procalcitonin ratio (APR), procalcitonin/procalcitonin ratio (CPR).

A total of 251 consecutive patients were included (152 were used as the development cohort and the other 99 were used as the validation cohort). Of these 167 men (66.5%), 84 women (33.5%), and the median age of 53 years (quartile range [ IQR ],39-62 years), all features between the derived and validated cohorts of the experimental study were similar as can be seen from table 1.

Table 1 derives (n=152) and verifies (n=99) baseline characteristics presented in the queue

Features (e.g. a character)	Deriving queue (median, IQR)	Verification queue (median, IQR)	P value
				Median age (range)	50.5 (1- 79)	54 (1- 80)	0.565
Sex, male	102 (67.10%)	65 (65.7%)	0.812
				Heating up	135 (88.8%)	87 (87.9%)	0.821
WBC (×10^⁹/L)	7.59 (3.25 - 14.45)	6.86 (3.69 - 11.84)	0.446
				NEU(×10^⁹/L)	5.99 (1.90 - 12.07)	5.83 (2.78 - 9.71)	0.488
LYM(×10^⁹/L)	0.61 (0.28 - 1.24)	0.64 (0.36 - 1.15)	0.680
				MO(×10^⁹/L)	0.36 (0.11 - 0.77)	0.37 (0.07 - 0.74)	0.554
HGB(×10^¹²/L)	96.50 (79.00 - 113.75)	96.00 (75.00 - 113.00)	0.633
				PLT(×10^⁹/L)	195.00 (105.25 - 292.50)	159.00 (93.00 - 250.00)	0.139
ALT(U/L)	21.95 (12.23 - 44.58)	22.60 (15.40 - 38.9)	0.657
				AST(U/L)	24.10 (16.40 - 44.93)	25.30 (16.00 - 45.70)	0.981
ALB(g/L)	34.45 (30.10 - 38.33)	35.20 (29.30 - 38.60)	0.660
				CYSC (mg/L)	1.04 (0.84 - 1.23)	1.03 (0.85 - 1.31)	0.700
CRP (mg/L)	75.84 (29.36 - 135.31)	83.84 (34.68 - 141.27)	0.699
				PCT (ng/ml)	0.61 (0.16 - 4.11)	0.44 (0.15 - 1.37)	0.480
CPT	34.97 (7.63 - 284.92)	33.54 (5.27 - 136.58)	0.590
				APR	58.71 (8.51 - 233.13)	80.70 (27.01 - 228.24)	0.440
CPR	90.24 (15.09 - 318.79)	130.37 (26.81 - 311.92)	0.472
				ACR	0.42 (0.24 - 1.18)	0.41 (0.22 - 1.04)	0.876
SLR	1.19 (0.78 - 1.68)	1.17 (0.82 - 1.70)	0.793
				dNLR	4.33 (1.39 - 10.52)	4.28 (2.25 - 8.11)	0.697
NLR	8.18 (2.10 - 21.95)	6.87 (4.01 - 14.39)	0.430
				PLR	306.59 (141.01 - 628.77)	235.21 (154.55 - 430.00)	0.079
LMR	1.84 (0.85 - 4.87)	2.25 (1.00 - 7.80)	0.147

The study included 10 cancers, with tumors derived primarily from lymphohematopoietic tissue (69/251, 27.5%) and digestive system tumors (62/251, 24.7%). In addition, digestive system tumors (n=41) are the most common tumor types accounting for 30.4% of bacteremia (n=135), see table 2.

Tumor type	All queues (%)	Bacteremia (%, n/135)
			Hematopoietic and lymphoid neoplasms	69 (27.5)	21 (15.6)
Digestive system tumor	62 (24.7)	41 (30.4)
			Head and neck tumor	36 (14.3)	24 (17.8)
Tumor of urinary system and male reproductive system	20 (8.0)	13 (9.6)
			Lung, pleura, thymus and heart tumors	17 (6.8)	6 (4.4)
Female genital tumor	13 (5.2)	10 (7.4)
			Soft tissue and bone tumors	12 (4.8)	9 (6.7)
Mammary gland tumor	10 (4.0)	6 (4.4)
			Tumor of central nervous system	7 (2.8)	3 (2.2)
Endocrine and neuroendocrine tumors	5 (2.0)	2 (1.5)
			Totals to	251(100.0)	135(100.0)

The study included 135 patients infected with pathogenic bacteria. Most pathogenic bacteria are gram negative bacteria (87/135, 64.4%), mainly Escherichia coli (n=32); one third (45/135, 33.3%) is gram positive, with staphylococcus aureus (n=12); of these, only 2 isolates were fungi, including filamentous fungi and Cryptococcus neoformans, and isolated pathogens are shown in Table 3.

TABLE 3 pathogenic bacteria isolated from cancer patients

Pathogens	n	%
			Gram negative bacteria (total)	87	64.44%
Escherichia coli	32	23.70%
			Pseudomonas aeruginosa	11	8.15%
Klebsiella pneumoniae	10	7.41%
			Enterobacter cloacae subspecies cloacae	7	5.19%
Burkholderia cepacia	6	4.44%
			Serratia marcescens	4	2.97%
Acetobacter oxydans	2	1.48%
			Enterobacter aerogenes	2	1.48%
Pseudomonas maltophilia	2	1.48%
			Acinetobacter baumannii	1	0.74%
Salmonella enterica subspecies enterica	1	0.74%
			Escherichia Fu	1	0.74%
Citrobacter florida	1	0.74%
			Elizabeth meningiomycemia bacteria	1	0.74%
Enterobacter cloacae (Fr.) Kummer	1	0.74%
			Vibrio cholerae	1	0.74%
Ai Kenshi rodent strains	1	0.74%
			Pi Kedi Larstoniya	1	0.74%
Canine Pasteurella	1	0.74%
			Hydrophilic aeromonas	1	0.74%
Gram positive bacteria (total)	45	33.33%
			Staphylococcus aureus	12	8.89%
Streptococcus sp	6	4.44%
			Staphylococcus spp	5	3.70%
Human staphylococcus subspecies	4	2.97%
			Streptococcus virous	4	2.97%
Streptococcus agalactiae	3	2.22%
			Enterococcus gallinarum	2	1.48%
Staphylococcus lysohaemolyticus	2	1.48%
			Gram-positive bacilli	2	1.48%
Streptococcus pneumoniae	2	1.48%
			Enterococcus faecalis	1	0.74%
Streptococcus pyogenes	1	0.74%
			Thrombin negative staphylococci	1	0.74%
Micrococcus luteus	1	0.74%
			Fungi (aggregate)	2	1.48%
Filamentous fungi	1	0.74%
			Novel cryptococcus	1	0.74%
Totals to	135	100.00%

2) Statistical analysis

Statistical analysis was performed using R software (version 3.6.1) and LASSO regression analysis was used to extract significant predictors associated with bacteremia, and lambda values were determined by 10-fold cross validation and error to select the most useful predictor. A new model is then constructed to predict bacteremia based on the coefficients of the significant predictors of LASSO regression.

Fig. 1A shows an analysis of the trajectory change of each predicted variable. The optimal value of λ was then determined by 10-fold cross-validation using a minimum standard (fig. 1B) according to which the optimal value of λ in this study was 0.0721. Its corresponding predictor is considered an important predictor of bacteremia, including Mo, ALT, CRP, CPR and dNLR.

Finally, a new model for predicting bacteremia is established from the coefficients of the 5 predictors obtained from LASSO regression. The values of the individual variables represent the original levels of the individual variables in serum, and the risk score is calculated using the following formula:

risk score = -0.0912-0.0263 xmo+0.0012 x alt+0.0021 x crp+0.0006 x cpr+0.023 x dNLR.

3) Model evaluation

3.1 Subject work characteristic curve (ROC) analysis

The ROC curve was used to compare the discrimination capability of the new model with that of WBC, CRP or PCT index. The ROC curve is plotted and its corresponding area under the ROC curve (AUC) is calculated. The greater the AUC of the ROC curve, the better the risk prediction of the model.

From the results in table 2, it is shown that AUC of the new model was higher than WBC, CRP and PCT in predicting bacteremia (fig. 2A). The results of the validation queue are similar to the derivation queue (table 2 and fig. 2B).

Table 2 derivation of AUC values for new models, WBCs, CRPs, and PCT in queues and validation queues

Models	AUC (95% CI)	P value
			Deriving queues
New model	0.804 (0.736-0.872)
			WBC	0.509 (0.416-0.602)
CRP	0.609 (0.518-0.699)
			PCT	0.745 (0.666-0.824)
New model vs WBC		<0.001*
			Novel model vs CRP		<0.001*
New model vs PCT		0.043*
			Verification queue
New model	0.758 (0.664-0.853)
			WBC	0.536 (0.418-0.654)
CRP	0.570 (0.455-0.684)
			PCT	0.682 (0.576-0.788)
New model vs WBC		0.009*
			Novel model vs CRP		0.002*
New model vs PCT		0.036*

* P <0.05, has statistical significance

3.2 Decision Curve Analysis (DCA) and Clinical Impact Curve (CIC) analysis

Different point cutting curves are made by Decision Curve Analysis (DCA), benefits and risks brought by different cut-in points in different models are different, and improvement conditions after reclassification are calculated. Clinical Impact Curves (CIC) are used to evaluate the clinical utility and net benefit of the new model and to determine if the new model has optimal predictive value.

The DCA results show that WBC and CRP curves are very near extreme, clinical value is not great, PCT benefit is higher than the extreme curves, but it is still far below the new model, which has a very high benefit range over a wide threshold (fig. 3A).

The results of CIC demonstrate that the new model shows better net benefit over a broad range of threshold probabilities and has an impact on the prognosis of the patient (fig. 3C). This means that the new model has the most practical significance for clinical practice compared to the old model. Similarly, the verification groups DCA (fig. 3B) and CIC (fig. 3D) were also compared, and the results were identical to those described above.

3.3 Net weight classification improvement index (NRI) and integrated discrimination improvement index (IDI) analysis

IDI and NRI calculations are used to compare the prediction accuracy of the new model with other models. The higher the accuracy, the lower the accuracy, the negative value, and the results are shown in Table 3.

For the development queue, IDI analysis showed that the accuracy of the new predictive model was higher than WBC, CRP, and PCT; NRI analysis showed that the accuracy of the new predictive model was higher than WBC, CRP and PCT. The results of the validation queue are similar to the derivation queue (Table 3).

Table 3 NRI and comparison of the accuracy of IDI on New model of bacteremia with WBC, CRP, PCT

Factors of	NRI	P value	IDI	P value
					Deriving queues
Predictive model vs WBC	0.399	<0.001*	0.284	<0.001*
					Predictive model vs CRP	0.294	<0.001*	0.241	<0.001*
Predictive model vs PCT	0.115	0.114	0.243	<0.001*
					Verification queue
Predictive model vs WBC	0.170	0.172	0.153	<0.001*
					Predictive model vs CRP	0.270	0.025*	0.157	<0.001*
Predictive model vs PCT	0.300	0.002*	0.169	<0.001*

NRI, net weight classification improvement index, IDI, comprehensive identification improvement index P < 0.05

3.4 New model, WBC, CRP and PCT predicts susceptibility, specificity, PPV (positive predictive value) and NPV (negative predictive value) of bacteremia

The predictive effect of each model is evaluated. In the derivation cohort, a risk score of 0.235 was used as the demarcation point between the high and low risk groups, predicting bacteremia susceptibility and specificity of 60.5% and 87.3%, respectively, PPV of 84.5%, and NPV of 66.0% (table 4). The results of the other indices are shown in Table 4.

The results show that the new model has the highest specificity and PPV in predicting bacteremia, while PCT has the highest sensitivity and NPV. The results of the validation queue are similar to the derivation queue (see table 4).

TABLE 4 infection markers predict bacteremia sensitivity, specificity, positive predictive value and negative predictive value

	Cut-off	Sensitivity	Specificity	PPV	NPV
						Deriving queues
Predictive models (Risk scoring)	0.235	60.5%	87.3%	84.5%	66.0%
						WBC(×10^9/L)	12.33	35.8%	74.6%	61.7%	50.5%
CRP(mg/L)	135.11	35.8%	87.3%	76.3%	54.4%
						PCT(ng/ml)	0.295	80.2%	57.7%	68.4%	71.9%
Verification queue
						Predictive models (Risk scoring)	0.235	50.0%	84.4%	79.4%	58.5%
WBC(×10^9/L)	12.33	77.8%	24.4%	55.3%	47.8%
						CRP(mg/L)	135.11	35.2%	82.2%	70.4%	51.4%
PCT(ng/ml)	0.295	74.1%	51.1%	64.5%	62.2%

3.5 Modal graph construction of new model

Gram-negative bacteria are the most common isolated pathogens, particularly in studies where antibiotic prophylaxis is rarely used. Monosomycosis is an important predictor of mortality in patients with hematological malignancies, particularly gram-negative bacteria, and remains a challenge in reducing factors associated with bacteremia in cancer patients due to antibiotic resistance. Given the ubiquity of current antibiotic resistance, appropriate empirical antibiotic treatments are effective, and thus new models are necessary to predict the type of pathogen that causes bacteremia in cancer patients.

A predictive alignment map based on risk scores in the validation queue was established to predict the risk of bacteremia (fig. 4A). Each patient is assigned a score and the likelihood of prediction corresponding to the score is used to predict bacteremia in the cancer patient. The higher the score, the higher the probability of bacteremia occurring.

The calibration curve results showed that the prediction results were consistent with the observations better (fig. 4B). The calibration curve shows that the actual observations are very consistent with the predicted bacteremia of the nomogram (Hosmer-laboratory test, p=0.907). The results of the validation queue are similar to the derivation queue.

In addition to distinguishing bacteremia from non-bacteremia, the risk score of the model is used to predict the etiology of bacteremia. The whole cohort was divided into four groups, gram positive microbiome, gram negative microbiome, fungal and non-bacteremia.

Support vector machine analysis showed that the new model was most accurate in predicting bacteremia etiology (0.606), whereas WBC was 0.462, crp was 0.518, pct was 0.522. The results indicate that the new model is more accurate than other markers of inflammation.

3.6 Correlation between the New model and WBC, CRP and PCT

Correlation between the new model and other models was evaluated (fig. 5). In this figure, green is a negative correlation and red is a positive correlation. The size and color intensity of the circle is proportional to the correlation coefficient. The Pearson correlation coefficient is used to determine the linear correlation between the variables.

The results show that the new model is positively correlated with CRP (correlation coefficient=0.31, P < 0.001) and PCT (correlation coefficient=0.33, P < 0.001) except for deriving WBCs in the queue. The relevant trends and differences for the validation queue are the same as shown in figure B.

3.7 Bacteremia risk stratification based on predictive model

A risk score of 0.235 was used as a demarcation point to divide patients into high-risk and low-risk groups. The difference in each significant predictor between the high risk group (n=92) and the low risk group (n=159) was compared.

From the results in table 5, it is shown that the higher risk patients have lower monocyte counts, lower CPR levels, and higher ALT, CRP, and dNLR levels than the lower risk patients.

TABLE 5 high risk and low risk group infection markers

Marking	High risk group (median, IQR)	Low risk group (median IQR)	P value
				Deriving queues
MO (×10^9/L)	0.24 (0.07 - 0.69)	0.42 (0.15 - 0.79)	0.020*
				ALT (U/L)	35.00 (19.28 - 74.53)	16.85 (10.55 - 32.55)	0.000*
CRP (mg/L)	133.25 (72.86 - 172.89)	49.95 (21.11 - 100.09)	0.000*
				CPR	24.43 (5.52 - 71.98)	199.00 (72.76 - 583.66)	0.000*
dNLR	11.36 (7.16 - 18.34)	2.74 (0.96 - 5.10)	0.000*
				Verification queue
MO (×10^9/L)	0.24 (0.05- 0.76)	0.39 (0.07 - 0.73)	0.488
				ALT (U/L)	25.25 (15.95 - 48.50)	22.3 (15.35 - 33.90)	0.290
CRP (mg/L)	139.03 (73.95 - 72.59)	60.56 (22.41 - 109.34)	0.000*
				CPR	34.00 (6.82 - 114.23)	221.00 (60.34 - 446.05)	0.000*
dNLR	10.67 (4.78 - 14.32)	3.58 (0.85 - 5.27)	0.000*

3.8 Comparison of the ability to predict bacteremia in febrile and non-febrile patients

Malignant patients often develop fever, which affects their quality of life. Common causes are fever caused by infection (inflammatory fever), tumor fever and fever caused by drugs. In view of the specificity of tumor patients, it is not possible to determine whether the cause of fever is caused by bacteremia or by tumor fever itself. Therefore, a need exists for a rapid and accurate authentication tool

In this study, more than 85% of patients (bacteremia group 86.3%, non-bacteremia group 91.8%) had fever symptoms, and in predicting bacteremia in both febrile and non-febrile patients, AUC of the new model was higher than WBC, CRP and PCT, whereas in febrile patients, the new model was statistically significant compared to WBC, CRP; in non-febrile patients, the new model is statistically significant compared to PCT. (see Table 6)

TABLE 6 predictive model, WBC, CRP and PCT vs. ability to predict bacteremia in febrile and non-febrile patients

	AUC (95% CI)	P value
			Fever patient
New model	0.800 (0.738-0.862)
			WBC	0.523 (0.438-0.609)
CRP	0.586 (0.506-0.666)
			PCT	0.751 (0.678-0.823)
New model vs WBC		<0.001*
			Novel model vs CRP		<0.001*
New model vs PCT		0.074

Patient without fever
			New model	0.886 (0.757-1.00)
WBC	0.743 (0.528-0.958)
			CRP	0.857 (0.675-1.00)
PCT	0.643 (0.399-0.887)
			New model vs WBC		0.271
Novel model vs CRP		0.792
			New model vs PCT		0.008*

CI = confidence interval; * P < 0.05.

The above description of the present invention is further illustrated in detail and should not be taken as limiting the practice of the present invention. It is within the scope of the present invention for those skilled in the art to make simple deductions or substitutions without departing from the concept of the present invention.

Claims

1. A system for predicting bacteremia in a patient with a tumor, comprising the following means:

2. The system of claim 1, wherein the predictive analysis is based on a calibration curve and a nomogram.

3. A kit for early detection of bacteremia in a patient with a tumor, said kit comprising the system of claim 1.