CN115714013A

CN115714013A - Construction method of clinical prediction model for pneumonia diagnosis

Info

Publication number: CN115714013A
Application number: CN202211322856.1A
Authority: CN
Inventors: 肖强; 陈司琴; 苏敏红; 雷薇; 李玺; 江佳; 张绍峰; 荣福
Original assignee: Shunde Hospital Of Southern Medical University (the First People's Hospital Of Shunde)
Current assignee: Shunde Hospital Of Southern Medical University (the First People's Hospital Of Shunde)
Priority date: 2022-10-27
Filing date: 2022-10-27
Publication date: 2023-02-24

Abstract

The invention discloses a method for constructing a clinical prediction model for pneumonia diagnosis, which relates to the technical field of the construction of pneumonia early diagnosis models and comprises the following steps: s1, performing single-factor logistic regression analysis on the inflammation indexes and the virtual variables after the differential metabolites are converted into the classification variables; s2, carrying out multi-factor logistic regression analysis on the variable with the P less than 0.05 in the single-factor analysis in the step S1 by adopting a backward stepwise regression method to obtain an independent risk factor related to CAP diagnosis; and S3, drawing a nomogram based on the multiple logistic regression model, and carrying out internal verification on the prediction model from three aspects of discrimination, calibration and clinical applicability through a bootstrap resampling method for 1000 times. The prediction model constructed by the invention has higher specificity and sensitivity, and the simple nomogram scoring system is easier to popularize in clinical practice, provides new possibility for perfecting the method for early diagnosis and severity evaluation of CAP, and has important significance for promoting accurate medicine and individualized effective treatment.

Description

Construction method of clinical prediction model for pneumonia diagnosis

Technical Field

The invention relates to the technical field of pneumonia early diagnosis model construction, in particular to a pneumonia diagnosis clinical prediction model construction method.

Background

The morbidity and mortality of community-acquired pneumonia (CAP) are steadily high in global infectious diseases, and the health of people is seriously influenced. Although community-acquired pneumonia has been recognized as a diagnostic standard, including symptoms, signs, laboratory indices and changes in imaging, accurate diagnosis of CAP at an early stage is still a considerable clinical problem, especially for patients with severe pneumonia, due to the heterogeneity of the disease and the complexity of the pathology. Early stage unobvious symptoms of the disease, nonspecific laboratory index changes, and difficulty in realizing pulmonary CT examination of severe patients all bring difficulty to early accurate diagnosis of pneumonia. If the diagnosis is not timely or wrong, the treatment of the patient may be delayed, the disease condition of the patient is aggravated, and for the non-bacterial infectious lung diseases, the wrong diagnosis also causes the abuse of antibiotic medicines, and further causes the generation of drug resistance. Therefore, finding a diagnosis method with high specificity, high sensitivity, more individuation and easy popularization has great help to improve the early diagnosis rate and the diagnosis accuracy rate of pneumonia patients, and is the key to implementing individualized treatment and accurate treatment and improving prognosis.

The biomarkers with high specificity and high sensitivity are greatly helpful for early accurate diagnosis. However, the existing serum inflammation markers, such as procalcitonin, C-reactive protein and the like, lack good specificity, can only be used as an auxiliary examination means, and cannot be used for accurately identifying pneumonia patients. Serum metabonomic analysis may help to find a diagnostic marker for pneumonia, but cannot accurately reflect the inflammatory changes of the lung. Therefore, the lower respiratory tract metabolite diagnosis marker with strong specificity and high sensitivity is screened, a more comprehensive and more individual pneumonia early diagnosis model is established, the realization of accurate medicine is promoted, and the clinical application value is important.

Therefore, the invention aims to provide a method for constructing a clinical prediction model for pneumonia diagnosis so as to solve the problems.

Disclosure of Invention

The invention aims to solve the problems and provides a method for constructing a clinical prediction model for pneumonia diagnosis.

In order to achieve the purpose, the technical scheme of the invention is as follows: a pneumonia diagnosis clinical prediction model construction method collects clinical data of research objects through a hospital case history system, converts laboratory inflammatory indexes and expression quantities of 7 differential metabolites into classification variables according to a reference range and an optimal cutoff value, constructs a clinical prediction model in three steps, and specifically comprises the following steps:

s1, performing single-factor logistic regression analysis on virtual variables after the inflammation indexes and the differential metabolites are converted into classification variables;

s2, carrying out multi-factor logistic regression analysis on the variable with the P less than 0.05 in the single-factor analysis in the step S1 by adopting a backward stepwise regression method to obtain an independent risk factor related to CAP diagnosis;

s3, drawing a nomogram based on the multiple logistic regression model, and internally verifying the prediction model from three aspects of discrimination, calibration and clinical applicability by a bootstrap resampling method for 1000 times;

statistical analysis was performed on steps S1 to S3 using SPSS26.0 and R software v.3.6.2 to obtain 3 independent risk factors for CAP diagnosis, which were: glycerophosphate (20.

Further, the method also comprises the step of carrying out metabonomic analysis on BALF samples of CAP patients and healthy people to obtain 7 metabolites which are obviously and differentially expressed in CAP patients, namely 7 differential metabolites.

Further, 7 of the differential metabolites are: dimethyl disulfide concentration (dimethyldisulphide), lysophospholipid cholate (12.

Further, the model is used to distinguish between healthy persons and CAP patients.

Compared with the prior art, the beneficial effect of this scheme:

1. the invention carries out metabonomic analysis on BALF samples of CAP patients and healthy people, determines 7 metabolites which are obviously and differentially expressed in the CAP patients, is different from the existing serum metabolites or other inflammation markers, can more directly reflect the metabolic change of the lower respiratory tract of the CAP patients, has higher specificity and sensitivity, has great potential when being applied to clinic as CAP biomarkers, and can provide new knowledge for CAP pathogenesis.

2. The invention not only excavates the biomarker more suitable for CAP, but also constructs a clinical prediction model for CAP diagnosis, the variables contained in the model not only comprise common laboratory inflammation indexes, but also incorporate newly discovered metabolic markers, and the model has good prediction capability through verification.

3. Compared with the existing diagnostic standard, the prediction model constructed by the method has higher specificity and sensitivity, and the simple nomogram scoring system is easier to popularize in clinical practice, so that new possibility is provided for perfecting the method for early diagnosis and severity evaluation of CAP, and the method has important significance in promoting accurate medicine and individualized effective treatment.

Drawings

FIG. 1 is a BALF metabolic spectrum of a pneumonia group and a healthy control group in a positive ion mode and a negative ion mode in an unsupervised principal component analysis in an embodiment of the present invention;

FIG. 2 is a graph showing the metabolic differences between the pneumonia group and the healthy control group in positive and negative ion modes in the supervised orthogonal partial least squares discriminant analysis in the examples of the present invention;

FIG. 3 is a comparison of 7 differential metabolites with a KEGG database reference spectrum in example of the present invention;

FIG. 4 is a graph showing the diagnostic efficacy of 7 different metabolites on CAP in example of the present inventionROC curve analysis;

FIG. 5 is a nomogram for visualizing a predictive model in an embodiment of the invention;

FIG. 6 is a ROC curve for a prediction model in an embodiment of the present invention;

FIG. 7 is a calibration curve of a predictive model in an embodiment of the invention;

FIG. 8 is a decision curve analysis of a predictive model in an embodiment of the invention.

Detailed Description

In order to make the technical solutions of the present invention better understood, the technical solutions of the present invention will be described in further detail below with reference to the embodiments of the present invention and the accompanying drawings. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.

It should be noted that the embodiments and features of the embodiments of the present invention may be combined with each other without conflict. The present invention will be described in detail with reference to examples.

Example (b):

the scheme provided by the embodiment of the invention is as follows: a pneumonia diagnosis clinical prediction model construction method collects clinical data of research objects through a hospital case history system, converts laboratory inflammatory indexes and expression quantities of 7 differential metabolites into classification variables according to a reference range and an optimal cutoff value, constructs a clinical prediction model in three steps, and specifically comprises the following steps:

s1, performing single-factor logistic regression analysis on the inflammation indexes and the virtual variables after the differential metabolites are converted into the classification variables;

s2, carrying out multi-factor logistic regression analysis on the variable with the P less than 0.05 in the single-factor analysis in the step S1 by adopting a backward stepwise regression method to obtain independent risk factors related to CAP diagnosis;

The specific implementation process of the embodiment scheme of the invention comprises the following steps:

1. metabonomics analysis of alveolar lavage fluid (BALF) samples

1.1BALF sample pretreatment

A sample of 50mg BALF was placed in an EP tube, 1000 μ L of extraction solvent (acetonitrile-methanol-water, 2, 1 containing internal standard) was added, homogenized for 4min at 30s,45hz with vortex oscillation, and sonicated for 5min in an ice water bath, after repeating the homogenization and sonication cycle 3 times, centrifuged at 12000rpm for 15min at-20 ℃ for 1h,4 ℃. The supernatant was transferred to an LC-MS loading flask and stored at-80 ℃ until UHPLC-QE Orbitrap/MS analysis. Quality Control (QC) samples were prepared by mixing equal amounts of supernatant from all samples.

1.2LC-MS/MS detection

Chromatographic conditions are as follows: the instrument adopts an Agilent 1290 ultra performance liquid chromatograph. And (3) chromatographic column: UPLC HSS T3 column (2.1 mm. Times.100mm, 1.8 μm). The flow rate was 0.5ml/min. The injection volume was 2. Mu.L. Mobile phase: positive ion mode: mobile phase: 0.1% aqueous formic acid (a), acetonitrile (B); negative ion mode: 55mmol/L ammonium acetate aqueous solution (A); acetonitrile (B). The elution gradient was set as follows: 0min,1% by weight of B;1 minute, 1% B;8 minutes, 99% B;10 minutes, 99% b;10.1 minutes, 1% b;12 min,1% B.

The mass spectrum conditions are as follows: MS/MS spectra were obtained using a Q active (Orbitrap MS, thermo) mass spectrometer. The ESI source conditions are set as follows: spraying voltage: 3800V (positive ion mode), -3100V (negative ion mode); capillary temperature: 320 ℃; flow rate of sheath gas: 45Arb; flow rate of auxiliary gas: 15Arb; scanning range: 70-1000m/z; first-order resolution: 70000; secondary resolution: 17500 (mm); the intensity value of the step-by-step collision energy is 3; the step collision energy was 20/40/60eV.

1.3 data processing

MS raw data files (. Raw) were converted to mzML format using protewizard software and processed by R-package XCMS (version 3.2) to obtain a data matrix consisting of Retention Time (RT), mass-to-charge ratio (m/z) values and peak intensities. After data processing using the internal MS/MS database, peak annotation was performed using OSI-SMMS (version 1.0, large connectivity chemical data solutions information technology limited).

1.4 multivariate statistical analysis

An unsupervised Principal Component Analysis (PCA) is adopted to examine the overall distribution of the sample, and the BALF metabolic spectra of the pneumonia group and the healthy control group have a relatively obvious separation trend under a positive ion mode and a negative ion mode (as shown in figure 1). Then, the metabolic differences between the pneumonia group and the healthy control group are further distinguished by adopting supervised orthogonal partial least squares discriminant analysis (OPLS-DA), and as shown in FIG. 2, the positive ion mode and the negative ion mode show a remarkable inter-group classification trend. R2X =0.62, R2Y =0.592, Q2Y =0.344 of the model in positive ion mode; the model in negative ion mode has R2X =0.937, R2Y =0.68, and Q2Y =0.614.

1.5 determination of differential metabolites

Screening two groups of significant difference metabolites by combining the VIP value of multivariate statistical analysis OPLS-DA and the P value of univariate statistical analysis student' st test, wherein the threshold value is as follows: VIP is more than or equal to 1 and P is less than 0.05. Based on the method, 7 differential metabolites in positive and negative ion modes can be obtained, and molecular formulas and names of the 7 differential metabolites are finally identified after comparison with a KEGG database reference spectrogram, as shown in the following Table 1.

TABLE 1

CAP，community-acquired pneumonia；ESI mode，election spray ionization mode；FC，fold Change；VIP，variable importance in the project ion；Status，the expression of metabolites in CAP patients increased or decreased compared with healthy controls；↑，increased；↓；decreased.

Shown in table 1, are: the dimethyl disulfide concentration (Dimethyldisulfide), lysophosphatidylcholine (12.

1.6 diagnostic potency assay for CAP by differential metabolites

To further understand the diagnostic efficacy of the 7 differential metabolites described above for CAP, ROC curve analysis was performed using SPSS26.0, as shown in table 2 below.

TABLE 2

ROC curve analyss，receiver operating cnaracteristic curve analysis；CAP，community-acquired pneumonia；AUC，area under the curve.

As can be seen from Table 2 and FIG. 4, AUC of the above 7 metabolites is > 0.7, and the corresponding sensitivities and specificities are high, which indicates that the diagnosis efficiency of CAP is high, and the 7 different metabolites combine with relevant laboratory inflammation indexes to further construct a clinical prediction model for pneumonia diagnosis.

2. Construction and verification of clinical prediction model

Clinical data of the study subjects are collected through a hospital medical record system, and the laboratory inflammatory index and the expression quantity of the 7 differential metabolites are converted into classification variables according to the reference range and the optimal cutoff value. Constructing a clinical prediction model in three steps: (1) Performing single-factor logistic regression analysis on the inflammation indexes and the virtual variables after the differential metabolites are converted into the classification variables; (2) Carrying out multi-factor logistic regression analysis on the variable with the P less than 0.05 in the single-factor analysis by adopting a backward stepwise regression method to obtain an independent risk factor related to CAP diagnosis; and (3) drawing a nomogram based on the multiple logistic regression model. And internally verifying the prediction model from three aspects of discrimination, calibration and clinical applicability by a 1000-time bootstrap resampling method. The above steps were statistically analyzed using SPSS26.0 and R software v.3.6.2.

From the above analysis, 3 independent risk factors for CAP diagnosis were obtained, as shown in Table 3 below.

TABLE 3

From table 3, it is known that: the final results were 3 independent risk factors for diagnosis of CAP: glycerophosphate (20. The ROC curve of the prediction model is shown in FIG. 6, the C-index obtained by using bootstrap for 1000 times is 0.984, and the model is used for distinguishing healthy people from CAP patients with higher accuracy. The P value of the Hosmer-Lemeshow test is 0.701, the predicted probability of the calibration curve is basically consistent with the actual probability (as shown in FIG. 7), and the result shows that the difference between the CAP predicted probability and the actual probability obtained through the histogram model is small, and the model accuracy is high. It can be seen from Decision Curve Analysis (DCA) (as shown in fig. 8) that, in a wide range, intervention in CAP patients with predictive model help diagnosis is more beneficial than total treatment or total no treatment, and the model has higher clinical applicability.

Through the above embodiments of the present invention, the present invention discovers 7 metabolites significantly differentially expressed in CAP patients through metabonomic analysis of BALF samples of CAP patients and healthy people, which are different from the existing serum metabolites or other inflammation markers, can more directly reflect the metabolic changes of the lower respiratory tract of CAP patients, have higher specificity and sensitivity, have great potential for clinical application as CAP biomarkers, and may provide new insights for CAP pathogenesis. Besides mining the biomarkers more suitable for CAP, the invention also constructs a clinical prediction model for CAP diagnosis, the variables contained in the model not only comprise common laboratory inflammation indexes, but also include the newly found metabolic markers, and the model has good prediction capability through verification. Compared with the existing diagnostic standard, the prediction model has higher specificity and sensitivity, the simple nomogram scoring system is easier to popularize in clinical practice, new possibility is provided for perfecting the early diagnosis and severity evaluation method of CAP, and the prediction model has important significance in promoting accurate medicine and individualized effective treatment.

The above embodiments are merely illustrative and not restrictive, and those skilled in the art can modify the embodiments without inventive contribution as required after reading this specification, but the invention is protected by the claims only.

Claims

1. A method for constructing a clinical prediction model for pneumonia diagnosis is characterized by comprising the following steps: clinical data of a study object are collected through a hospital medical record system, expression quantities of inflammatory indexes and 7 different metabolites in a laboratory are converted into classification variables according to a reference range and an optimal cutoff value, and a clinical prediction model is constructed in three steps, and the method specifically comprises the following steps:

2. The method of claim 1, wherein the clinical predictive model for pneumonia diagnosis is constructed by: also included is metabonomic analysis of BALF samples from CAP patients and healthy persons to obtain 7 metabolites that are significantly differentially expressed in CAP patients, i.e., 7 differential metabolites.

3. The method of claim 2, wherein the model for clinical prediction of pneumonia is: 7 of said differential metabolites are respectively: dimethyl disulfide concentration (dimethyldisulphide), lysophospholipid cholate (12.

4. The method of claim 1, wherein the model for clinical prediction of pneumonia is: the model is used to distinguish between healthy persons and CAP patients.