Method for predicting accurate dosage of polymyxin B for multi-drug-resistant severe infection patient

Abstract

The invention relates to the technical field of drug information, in particular to a method for predicting the accurate dosage of polymyxin B for a multi-drug-resistant severe infection patient, which comprises the following steps: acquiring model establishment data, establishing a basic model, incorporating covariables, evaluating the model, performing external verification and simulating based on the model. According to the method provided by the invention, accurate prediction of medication of the polymyxin B is realized, and death of a patient with severe infection caused by anti-infection failure due to too low dosage of the polymyxin B is reduced; the situation that the health of a severe patient is seriously damaged due to serious renal toxicity and neurotoxicity side effects caused by overlarge dosage of the polymyxin B is avoided, and the method has important significance on precise medication of clinically refractory severe infection patients.

Description

Technical Field

[0001] The present invention relates to the field of drug information technology, and in particular to a method for accurately predicting the dosage of polymyxin B for patients with multidrug-resistant severe infections. Background Art

[0002] Antimicrobial resistance is receiving increasing attention. Polymyxin B is an antibiotic that has inhibitory effects on a variety of negative bacteria, including Pseudomonas aeruginosa, Escherichia coli, Klebsiella, and Haemophilus. It is also sensitive to strains resistant to other antibiotics. However, polymyxin B has severe nephrotoxicity and nerve blockade, so its use and dosage need to be very cautious.

[0003] In recent years, polymyxin B has been widely used in clinical anti-infection treatment. In clinical practice, polymyxin B shows a narrow therapeutic window, that is, the minimum effective concentration is close to the minimum toxic concentration. If the dose is too large, it is easy to cause toxic reactions, and there are significant individual differences. Therapeutic concentration monitoring (TDM) is needed in clinical practice to ensure its rational use. At present, most clinicians rely on experience to set the initial dose and adjust the dose of polymyxin B, but the individual differences in blood drug concentration are large, which often leads to anti-infection failure and even causes severe nephrotoxicity and neurotoxicity in patients, seriously damaging the health of patients. Although some medical institutions have promoted the individualized treatment of this drug through HPLC-MS / MS, the cycle of "patient medication treatment-sampling after steady state-sample processing-drug concentration analysis-obtaining results-dose regimen adjustment" has a significant time lag. In addition, due to the high cost of this technology and the high requirements for operators, it is difficult for most medical institutions to implement it, resulting in the current situation that clinicians can only rely on experience to treat patients with anti-infection, which also aggravates the problem of clinical anti-infection failure.

[0004] Therefore, it is urgent to develop a method for accurately predicting the dosage of polymyxin B. Summary of the invention

[0005] In order to solve the problem of anti-infection failure caused by large individual differences in blood drug concentration, the present invention provides a method for predicting the precise dosage of polymyxin B in patients with multidrug-resistant severe infections. By establishing a population pharmacokinetic model (Pop-PK model) of polymyxin B in patients with multidrug-resistant severe infections, the establishment of the model includes the inclusion of basic models and covariates, model evaluation, external validation of the model, and simulation based on the model. The population pharmacokinetic model (Pop-PK model) of polymyxin B constructed by the present invention is used to guide the precise medication method of polymyxin B in patients with multidrug-resistant severe infections, and realizes the prediction of dosage and frequency of administration, which helps the clinic to more efficiently use polymyxin B in patients with multidrug-resistant severe infections and realize individualized treatment.

[0006] The present invention provides a method for accurately predicting the dosage of polymyxin B for patients with multidrug-resistant severe infections, comprising the following steps:

[0007] Collect personal data of patients with multidrug-resistant severe infections and clinical drug use data of polymyxin B in patients with multidrug-resistant severe infections as data for model establishment;

[0008] Based on the model establishment data, a structural model is established through compartmental model fitting and fixed effect parameter prediction to determine the absorption and disposal of the drug. Then, a random effect model is used to quantify the variation that cannot be explained by the fixed effect parameters through random effect parameters. Then, covariates are determined by analyzing and screening the continuous variables and categorical variables of the model establishment data. The basic model is incorporated with covariates to obtain a population pharmacokinetic model to clarify the effects of relevant factors on drug pharmacokinetics and pharmacodynamics. After that, the population pharmacokinetic model is evaluated, including goodness of fit plots, visualization tests, and bootstrapping methods to test and evaluate the predictive ability and stability of the population pharmacokinetic model.

[0009] External validation: It is considered to be one of the most rigorous model evaluation methods, mainly examining the ability of the model to be extended and applied to other conditions; external validation of the population pharmacokinetic model based on collected independent external data, including predictive performance diagnosis, simulation performance diagnosis and Bayesian prediction; and measuring the inaccuracy and bias of the predictive ability of the population pharmacokinetic model after computational test evaluation;

[0010] Model-based simulation: The separately collected external data were used as the simulation data set. After stratification by covariates, the population pharmacokinetic model was used to predict the steady-state blood drug concentration of patients under different dosage regimens. The requirements of the guidelines for pharmacokinetic / pharmacodynamic (PK / PD) parameters were used as the basis for determining the recommended regimen, and the recommendations were summarized to obtain the precise dosage recommendation of polymyxin B for patients with multidrug-resistant severe infections.

[0011] Further, it also includes data processing;

[0012] Among them, the collected model building data was edited using Microsoft Excel to organize the data files for Phoenix NLME software analysis. The organized items included TIME, DV, ADDL, II, AMT, EVID, MVD and covariates, and were saved in .csv format that can be recognized by Phoenix NLME. The organized data can be directly used to establish the basic model.

[0013] Furthermore, the compartment model includes a one-compartment model and a two-compartment model.

[0014] Furthermore, the random effect model includes determining the inter-individual random effect model and the residual model. The inter-individual random effect model includes additive, proportional and exponential forms; the residual model includes additive, proportional and mixed forms; the basic model is determined by selecting the minimum AIC value, the minimum -2LL value and the goodness of fit diagram.

[0015] The present invention also provides a population pharmacokinetic model for accurately predicting the dosage of polymyxin B, which is obtained by the method for accurately predicting the dosage of polymyxin B in patients with multidrug-resistant severe infections. The population pharmacokinetic model for accurately predicting the dosage of polymyxin B is:

[0016] V(L)=52.09*exp(ηV)

[0017] V2(L)=28.60*exp(ηV2)

[0018]

[0019] Q(L / h)=23.86*exp(ηQ);

[0020] Wherein, V is the central compartment distribution volume, unit: liter; exp is the exponential function with the natural constant e as the base; η is a random variable that obeys the normal distribution; V2 is the peripheral compartment step volume; CL is the central compartment clearance rate, unit: liter / hour; CrCl is the serum creatinine clearance rate; Q is the inter-compartmental clearance rate.

[0021] Compared with the prior art, the present invention provides a method for accurately dosing polymyxin B for patients with multidrug-resistant severe infections, and the specific beneficial effects are:

[0022] The prediction method of the present invention is used to help clinicians accurately predict the individual initial medication dose and medication frequency for patients with multidrug-resistant severe infections, thereby reducing the death of patients with severe infections due to anti-infection failure caused by too low a dose of polymyxin B; it also avoids severe damage to the health of severe patients due to severe nephrotoxicity and neurotoxic side effects caused by excessive doses of polymyxin B, which is of great significance for the precise medication of patients with clinically refractory severe infections.

[0023] The method for predicting the precise dosage of polymyxin B in patients with multidrug-resistant severe infections provided by the present invention guides the precise medication of polymyxin B in patients with multidrug-resistant severe infections by establishing a population pharmacokinetic model (Pop-PK model) of polymyxin B. Specifically, by collecting blood drug concentration data of 80 patients with multidrug-resistant severe infections, separating target pathogens, and combining clinical drug combination, it is proved by model verification that creatinine clearance is the main covariate affecting the pharmacokinetic process of polymyxin B in vivo, and the individualized administration of polymyxin B is considered from the perspective of population pharmacokinetics, providing a theoretical basis for adjusting the medication regimen, and realizing accurate prediction of drug dosage and administration frequency through the blood creatinine data of individual patients. The main evaluation index of the precise medication of the polymyxin B model in the present invention is derived from creatinine (Ccr), which solves the defect that the traditional medication strategy with weight as an indicator is difficult to obtain for patients with weight changes or large weight changes in a short period of time, and is difficult to accurately use medication.

[0024] The population pharmacokinetic model for accurately predicting the dosage of polymyxin B provided by the present invention can realize the prediction of dosage and frequency of administration, and help to more efficiently and accurately use polymyxin B for patients with multidrug-resistant severe infections in the clinic, thereby achieving individualized treatment. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0026] Figure 1 This is a graph showing the blood concentration-time relationship of polymyxin B after clinical use in patients with multidrug-resistant severe infections collected for model establishment.

[0027] Figure 2 is a variable correlation analysis diagram; in the figure, A is the correlation analysis diagram of white blood cell count (WBC) and neutrophil count (NEU); B is the correlation analysis diagram of total bilirubin (TBIL) and direct bilirubin (DBIL); C is the correlation analysis diagram of red blood cell count (RBC) and hemoglobin (HGB); D is the correlation analysis diagram of γ-glutamyl transferase (GGT) and alkaline phosphatase (ALP); E is the correlation analysis diagram of serum creatinine (Ccr) and uric acid (UA); F is the correlation analysis diagram of total protein (TP) and globulin (GLOB); G is the correlation analysis diagram of alanine aminotransferase (ALT) and aspartate aminotransferase (AST); H is the correlation analysis diagram of serum urea nitrogen (BU) and uric acid (UA); I is the correlation analysis diagram of serum creatinine (Ccr) and serum urea nitrogen (BU).

[0028] Figure 3 It is a goodness of fit plot for model evaluation; in the figure, A is the goodness of fit plot of individual predicted concentration and observed concentration; B is the goodness of fit plot of group predicted concentration and observed concentration; C is the goodness of fit plot of time after administration and conditional weighted residual; D is the goodness of fit plot of group predicted concentration and conditional weighted residual.

[0029] Figure 4 Visual prediction test plot for model evaluation.

[0030] Figure 5 A visualization of the prediction test results for external validation.

[0031] Figure 6 These are the Bayesian prediction results under different scenarios, 0: no TDM detection value; C0: one prior TDM steady-state trough concentration is known; C2: one prior TDM steady-state peak concentration is known; C0+C2: two prior TDM concentrations (trough concentration and peak concentration) are known.

[0032] Figure 7 Recommendations for the model for patients with different levels of renal function;

[0033] In the figure, A is the model recommendation for people with normal renal function;

[0034] B is the model recommendation for the hyperrenal function group;

[0035] C is the model recommendation for people with mild renal impairment;

[0036] D is the model recommendation for people with moderate renal impairment;

[0037] E is the model recommendation for people with severe renal impairment.

[0038] Figure 8 This is an architecture diagram of a clinical decision support strategy for accurately predicting the dosage of polymyxin B for patients with severe infections according to an embodiment of the present invention. DETAILED DESCRIPTION

[0039] The specific implementation of the present invention is described in detail below, but it should be understood that the protection scope of the present invention is not limited by the specific implementation. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0040] The present invention provides a method for accurately predicting the dosage of polymyxin B for patients with multidrug-resistant severe infections, including data collection and collation, basic model and covariate inclusion, model evaluation, model external validation and model-based simulation.

[0041] The basic model includes a structural model and a random effects model; the development of the structural model mainly includes the fitting of the compartment model for determining the absorption and disposal of drugs and the prediction of fixed effect parameters, and the compartment model includes a one-compartment model and a two-compartment model; the one-compartment model and the two-compartment model are used to simulate the obtained model development data set, respectively, and the changes in AIC between different compartment models are compared, and the one with the smallest AIC value is selected for inclusion.

[0042] The random effect model quantifies the part that cannot be explained by the fixed effect parameters by determining the random effect parameters, including the inter-individual variation model and the intra-individual variation model. The inter-individual variation refers to the deviation of the individual parameter value from the typical value of the group; the intra-individual variation (residual variation) refers to the deviation of the individual predicted value from the actual observed value, which is used to explain the difference between the observed value and the individual model predicted value caused by other factors except the inter-individual variation; finally, the basic model is determined using the Akaike information criterion value (AIC), -2 times the logarithm of the maximum likelihood value (-2LL), and the goodness-of-fit plot.

[0043] Covariates were divided into continuous variables and categorical variables. In the preliminary screening, R language and Rstudio were used to analyze scatter plots and Pearson correlation coefficients to retain only one variable with collinearity. Then, a stepwise method, including forward inclusion and reverse elimination, was used to determine the population pharmacokinetic model based on the change in the objective function value.

[0044] Model evaluation: Based on the model establishment data set, the predictive ability and stability of the population pharmacokinetic model were tested and evaluated using goodness-of-fit plots, visualization tests, and bootstrapping methods.

[0045] External validation: Measuring the inaccuracy and bias of the computational model’s predictive and simulation capabilities based on the collection of independent external data.

[0046] Model-based simulation: The separately collected external data are used as the simulation data set. After stratification by covariates, the model is used to predict the patient's steady-state blood drug concentration under different dosage regimens. The pharmacokinetic parameters are summarized to obtain the precise dosage recommendation of polymyxin B for patients with multidrug-resistant severe infections.

[0047] Example 1: Method for Predicting the Accurate Dosage of Polymyxin B for Patients with Multidrug-Resistant Severe Infections

[0048] 1. Collect patient personal data and clinical drug use data of polymyxin B in patients with multidrug-resistant severe infections as data for model establishment.

[0049] 1. Data Collection

[0050] The demographic information, laboratory examination indicators before and after medication, microbial culture and drug sensitivity test results (PD indicators), and clinical use data of polymyxin B in patients with severe multidrug-resistant bacteria infections in the Department of Critical Care Medicine were collected. The clinical use data of polymyxin B included specific dosing regimens, steady-state blood drug concentration data (PK indicators), and other relevant clinical data.

[0051] The relationship between the steady-state blood drug concentration data (PK index) and the time after the administration of polymyxin B from the patient's PK blood is shown in the figure below: Figure 1 0h refers to before administration, 2h, 4h, 6h and 8h refer to the blood drug concentration 2h, 4h, 6h and 8h after administration respectively.

[0052] Study population

[0053] 1. Selection criteria

[0054] Patients included in the present invention should meet the following criteria: (1) age ≥ 18 years old, regardless of gender; (2) in addition to confirming pneumonia, the patient's radiological examination should have at least one of the following clinical symptoms: 1) new cough and sputum, or aggravation of the original respiratory disease with the appearance of purulent sputum; 2) fever or hypothermia; fever is manifested as: axillary temperature greater than 38°C, rectal temperature > 38.5°C or tympanic membrane temperature > 38.5°C; hypothermia is manifested as axillary temperature ≤ 35°C; 3) signs of pulmonary consolidation and / or moist rales; 4) white blood cell count greater than 10×10 9 / L or <4×10 9 / L, with or without left shift of the nucleus; 5) dyspnea, tachypnea or hypoxemia; the criteria are: oxygen saturation <90% or PO2 <60 mmHg measured by arterial blood gas when the subject breathes room air at standard atmospheric pressure; 6) patients need to undergo Gram staining and culture of respiratory specimens, and the source of the specimens must be confirmed to be the lower respiratory tract before enrollment, and the bacterial culture results indicate infection with multidrug-resistant Gram-negative bacteria; (3) severe infection is known or the researcher believes that it is caused by microorganisms sensitive to polymyxin B; (4) voluntary participation and signing of written informed consent. Confirmed pneumonia means CT shows new or progressive pulmonary infiltration, which is consistent with bacterial pneumonia.

[0055] 2. Exclusion criteria

[0056] Patients with any of the following should be excluded from the present invention:

[0057] (1) Patients with a history of allergy to the investigational drug polymyxin B, or known allergies to more than two drugs or foods; (2) Patients with pulmonary infiltrates due to non-infectious causes, such as pulmonary embolism, aspiration chemical pneumonia, hypersensitivity pneumonitis, congestive heart failure, etc.; (3) Patients suspected of Gram-positive or Gram-negative coccal infection based on clinical manifestations, signs, imaging examinations, and laboratory tests; (4) Patients with known or suspected atypical pathogen infection based on epidemiological evidence, such as Chlamydia pneumoniae, Mycoplasma pneumoniae, and Legionella; (5) Patients who cannot stop taking other nephrotoxic drugs during medication, such as bacitracin, streptomycin, neomycin, kanamycin, buprenomycin, gentamicin, tobramycin, cephalosporin, paromomycin, puromycin, and colistin; (6) Subjects must not receive polymyxin B treatment for more than 48 hours, and polymyxin B must be injected intravenously for no less than 4 times.

[0058] 3. Dosage regimen

[0059] Polymyxin B sulfate was injected intravenously, once every 12 hours, and it was recommended that the infusion be completed within 1 hour. The specific dosage regimen, infusion time, and treatment maintenance time were determined by the clinical team. Each subject provided polymyxin B for no more than 10 days, a total of 20 doses, and the clinician's dosing regimen was collected.

[0060] 4. Collection of biological samples

[0061] For the included subjects, 4 mL of blood samples were collected at 0 h before administration, 1 h±5 min, 3 h±5 min, and 8 h±5 min after the end of infusion after the 5th intravenous infusion, respectively. The samples were centrifuged at 4 °C and 1700 g for 10 min, and the upper plasma was extracted and transferred to labeled cryotubes marked with the protocol number, subject number, and collection time point.

[0062] 5. Biological sample analysis

[0063] The LC-MS / MS detection platform was used to detect blood drug concentrations, including polymyxin B1 and polymyxin B2. The LC-MS / MS instrument was an Agilent 1260 high-performance liquid chromatograph connected in series with an AB SCIEX QTRAP 5500 mass spectrometer. The detection method was to use the MRM transitions in positive electrospray ionization mode for quantification m / z 402.2→101.1 (PB1, DP 80V, CE 10V), 397.5→101.1 (PB2, DP 80V, CE 10V) and 386.2→101.0 (IS, DP 70V, CE 22V).

[0064] The mobile phase was 0.1% formic acid-water (solvent A) and 0.1% formic acid-acetonitrile (solvent B); a gradient elution program was used: 0.00-0.10 min, 95% solvent A, 5% solvent B; 0.10-1.00 min, 95% A-40% solvent A, 5%-60% solvent B; 1.00-1.40 min, 40%-10% solvent A, 60%-90% solvent B; 1.40-2.20 min, 10% solvent A, 90% solvent B; 2.20-2.70 min, 10% A-95% solvent A, 90%-5% solvent B; 2.70-3.00 min, 95% solvent A, 5% solvent B. The quantification of polymyxin B was based on the peak area ratio of polymyxin B1 and B2 and IS. Use 1.6.3 and Microsoft Excel for data analysis and graphing.

[0065] 6. Microbiological specimen collection

[0066] Baseline specimens were collected before the use of polymyxin B, and visit specimens were collected after the polymyxin B treatment regimen was completed as required by the protocol. Other time periods can be collected as needed. Bacteriological test specimens include blood, lower respiratory tract sputum, bronchial secretions or bronchoalveolar lavage fluid, clean midstream urine, pus, wound secretions or puncture aspiration fluid, peritoneal pus, cerebrospinal fluid, pleural and peritoneal puncture fluid, etc.;

[0067] 7. Microbiological examination

[0068] The isolation and identification methods and operating procedures of pathogens should comply with the technical standards and operating procedures of the health industry for clinical microbiology testing, and must be based on the requirements of clinical microbiology experimental technology for clinical trials of antimicrobial drugs, and the species must be identified to the "species" level. PD indicators should be obtained.

[0069] Sputum smear specimens are subjected to Gram staining and microscopic examination. The results of microscopic examination show that the average number of white blood cells per field of view under low-power microscope is ≥25, and the number of squamous epithelial cells is ≤10, and then sputum bacterial culture is performed; urine specimens are tested for urine colony count; abdominal pus specimens are mostly mixed infections, and aerobic and anaerobic bacteria cultures are required at the same time, and they are sent for examination immediately in an anaerobic environment; blood specimens should be taken from two different parts of the body as a routine blood culture specimen, and both aerobic and anaerobic cultures are required.

[0070] 8. Data Collection

[0071] 8.1 Data Sources

[0072] Inpatient electronic medical record system data: mainly includes HIS, LIS, PACS, electronic medical record data, nursing records and discharge settlement system related data.

[0073] 8.2 Data Elements

[0074] The data of this study mainly include: patient demographic characteristics, polymyxin B-related information, pathogen infection information, laboratory examination indicators, laboratory blood drug concentration detection indicators, concomitant medications, underlying diseases, etc.

[0075] 8.2.1 Basic characteristics of patients

[0076] (1) Patient number: the pinyin abbreviation of the patient's name. (2) Demographic characteristics: age, gender, weight, race, and place of origin. (3) Target pathogen isolation: multidrug-resistant Klebsiella pneumoniae, multidrug-resistant Acinetobacter baumannii, and multidrug-resistant Pseudomonas aeruginosa. (4) Comorbidities: heart failure, renal failure, diabetes, malignant tumors, cerebral infarction, and chronic obstructive pulmonary disease.

[0077] 8.2.2 Information about polymyxin B

[0078] (1) Drug information: Polymyxin B Sulfate for injection (trade name: Yale), a product of Shanghai No. 1 Biochemical Pharmaceutical Co., Ltd., 500,000 units per bottle. (2) Polymyxin B dosing regimen: Each subject provided polymyxin B for no more than 10 days (20 doses), and blood samples were collected from the included subjects at 0h±5h before the fifth or more intravenous infusions, 1h±5min, 3h±5min, and 8h±5min after the end of the infusion. (3) Standard product information: Polymyxin B1 standard (batch number GC111601, purity 96.40%, PB1) was purchased from GLPBIO, USA; Polymyxin B2 standard (batch number P039-01, purity 92.76%, PB2) was purchased from TOKU-E, Japan; Polymyxin E2 standard (internal standard, batch number 7-JOB-133-2, purity 91.17%, IS) was purchased from TRC, Canada. Methanol and acetonitrile were purchased from Merck, Germany, and formic acid was purchased from Enco Chemical, USA.

[0079] 8.2.3 Obtaining drug concentration data of polymyxin B from biological sample analysis

[0080] A liquid chromatography-tandem mass spectrometer (LC-MS / MS) analysis method for polymyxin B was established. The linear range of polymyxin B (including polymyxin B1 and polymyxin B2) was 200-6400 ng / mL, and the average regression equation of the standard curve was: y=0.00418x+0.0197 (correlation coefficient r=0.9962). Quality control was performed using a low-concentration quality control of 400 ng / mL, a medium-concentration quality control of 1700 ng / mL, and a high-concentration quality control of 4800 ng / mL. The relative errors of the low, medium, and high quality control concentrations did not exceed ±15%. When the quality control samples did not meet the standards, the analytical batch data were not adopted.

[0081] 8.2.4 Microbiological examination to obtain drug sensitivity test results

[0082] Pathogen Isolation: Pathogens were isolated from all baseline and post-dose specimens.

[0083] Drug sensitivity test: The diameter of the inhibition zone is determined by the paper diffusion method, and the corresponding drug sensitivity test specifications and quality control requirements are followed in the test. The MIC values ​​of all baseline isolates in the drug sensitivity test are issued, and the MIC values ​​of all antimicrobial drugs involved in the drug sensitivity test are listed. 50 and MIC 90 MIC analysis of polymyxin B against clinical isolates by bacterial species 50 and MIC 90 values ​​and MIC ranges, sensitivity rates and resistance rates; if necessary, group classification and statistics can be performed according to the different resistance phenotypes of bacteria, or bacteria that are confirmed to have unique resistance patterns and / or resistance mechanisms, such as carbapenem-resistant Enterobacteriales, non-sugar-fermenting Gram-negative bacteria such as Acinetobacter baumannii and Pseudomonas aeruginosa.

[0084] 8.2.5 Inspection index testing

[0085] The test indicators should collect the samples corresponding to the day of treatment concentration monitoring. If the test indicators are not tested on the same day, the data values ​​within ±2 days of concentration monitoring can be collected.

[0086] (1) Liver function: total protein, albumin, total bilirubin, direct bilirubin, alanine aminotransferase, aspartate aminotransferase, glutamyl transpeptidase, alkaline phosphatase; (2) Renal function: serum creatinine, creatinine clearance, serum urea nitrogen, uric acid; (3) Routine blood test: white blood cell count, red blood cell count, hemoglobin, platelets, lymphocytes, monocytes, neutrophils, eosinophils, basophils, C-reactive protein.

[0087] 8.2.6 The patient's creatinine clearance (CrCL) is calculated using the Cockcroft-Gault formula:

[0088] Male: CrCL = [(140-age)*weight] / [72*serum creatinine value (mg / mL)];

[0089] Female: CrCL = [(140-age)*weight×0.85] / [72*serum creatinine value (mg / mL)].

[0090] 8.2.7 Concomitant medication

[0091] For combined medication, collect the drugs taken continuously within 7 days before concentration monitoring (focus on recording the use of antimicrobial drugs, such as cephalosporins, tetracyclines, glycopeptides, aminoglycosides, and quinolones), and record the specific drugs, dosage, and frequency of medication used by the patient.

[0092] 9. Data management and statistical analysis

[0093] 9.1 Data Management

[0094] This includes collecting data in the form of original record forms in a unified format, and organizing the data that can be included according to the editing requirements of PhoenixNLME software. If uncertain results are encountered during the case collection process, the clinical team members will discuss and give a one-time conclusion.

[0095] 9.2 Statistical analysis

[0096] SPSS22.0 was used to perform statistical analysis on the data. Demographic data (gender, age, height, weight, etc.) were analyzed descriptively. For continuous variables, a normal distribution test was first used. If the data conformed to the normal distribution, the mean ± standard deviation (Mean ± SD) was used for description; if not, the median and upper and lower quartiles were used for description. For normally distributed data, the Student's t-test was used for comparison between the two groups; for non-normally distributed data, the Mann-Whitney U-test was used for comparison between the two groups.

[0097] 2. Based on the model establishment data, the basic model was established by using compartmental model fitting and fixed effect parameter prediction; then the covariates were determined by analyzing the continuous variables and categorical variables of the model establishment data, and the covariates were incorporated into the basic model to obtain the population pharmacokinetic model. The predictive ability and stability of the population pharmacokinetic model were then tested and evaluated using goodness-of-fit plots, visualization tests, and bootstrap methods.

[0098] The nonlinear mixed-effects modeling method of Phoenix 8.3 was used to establish a population pharmacokinetic model of polymyxin B in patients with severe infection. The establishment of the population pharmacokinetic model included data collation, inclusion of basic model and covariates, and model evaluation.

[0099] 1. Data collation

[0100] According to the method of Example 1, relevant information of hospitalized patients with severe lung infections caused by multidrug-resistant Gram-negative bacteria from November 2020 to June 2021 (50 cases in total) was collected as data for model establishment.

[0101] The model was built with data, and the original record form with a unified format was used to enter the data and formulate unified entry rules. The quality control was carried out by double checking (the information collector and the checker could not be completed by the same person). If errors or omissions were found, they were corrected and supplemented in time. The included data were edited by Microsoft Excel as required for the data file for Phoenix NLME software analysis. The items to be sorted included TIME, DV, ADDL, II, AMT, EVID, MVD and covariates. The data were transferred to the .csv format recognizable by Phoenix NLME.

[0102] 2. Development of basic model

[0103] The basic model includes a structural model and a random effect model; the development of the structural model mainly includes the fitting of the compartment model for determining the absorption and disposal mode of the drug and the prediction of the fixed effect parameters. The one-compartment model and the two-compartment model are used to simulate the obtained model development data set, and the changes in AIC between different compartment models are compared. The model with the smallest AIC value is selected as the criterion; the random effect model quantifies the part that cannot be explained by the fixed effect parameter by determining the random effect parameter, including the inter-individual variation model and the intra-individual variation model. The inter-individual variation refers to the deviation of the individual parameter value from the typical value of the group, which is expressed in a single or combined form such as addition, proportion, index, etc. According to the results reported in the previous literature, it is described in the present invention by an exponential model; the intra-individual variation (residual variation) refers to the deviation of the individual predicted value from the actual observed value. Finally, the basic model is determined using the minimum principle of Akaike information criterion value (AIC) and the logarithm of the maximum likelihood value -2 times (-2LL). The results are shown in Table 1. The two-compartment model is selected as the optimal model, the clearance rate (CL) and the apparent distribution volume (V) are used as fixed effects, and the proportional model is used to describe the residual variation.

[0104] The inter-individual variability is represented by an exponential model:

[0105] Individual parameter = group typical value * e 个体随机效效应 , where e represents a natural constant.

[0106] Table 1 Basic model screening

[0107]

[0108] 3. Covariate inclusion

[0109] The purpose of incorporating covariates is to identify the relevant factors that affect drug PK and PD and to improve the predictive power of the model. Covariates can be divided into continuous variables and categorical variables according to data type. Figure 2 As shown, in the preliminary screening, with the help of R language and Rstudio, scatter plots and Pearson correlation coefficients (screening criteria: scatter plots were linear and r>0.7) were used to exclude variables with collinearity, and only one of them was left for covariate screening. A stepwise method was used, including forward inclusion and reverse exclusion, and the population pharmacokinetic model was determined by the change in the objective function value (OFV).

[0110] Forward inclusion criterion: the difference of -2 times the maximum likelihood estimate (Δ-2LL)>3.84 (P<0.05, df=1); backward exclusion criterion: Δ-2LL>10.83 (P<0.001, df=1).

[0111] According to the covariate screening requirements, as shown in Table 2, creatinine clearance was used as a covariate for central compartment clearance; the results of the population pharmacokinetic model are as follows:

[0112] V(L)=52.09*exp(nV)

[0113] V2(L)=28.60*exp(nV2)

[0114]

[0115] Q(L / h)=23.86*exp(ηQ);

[0116] Wherein, V, central compartment distribution volume, unit: liter; exp, exponential function with natural constant e as base; η, random variable obeying normal distribution; V2, peripheral compartment distribution volume; CL, central compartment clearance, L / h, unit: liter / hour; CrCl, serum creatinine clearance; Q, inter-compartmental clearance.

[0117] The results showed that the two-compartment model with creatinine clearance as the covariate of central compartment clearance was able to fully describe the collected pharmacokinetic data. The exponential model was used to describe the inter-individual variation, and the proportional model was used to describe the residual variation. For the central compartment, the population typical values ​​of its distribution volume and clearance were 52.09 L and 28.60 L / h, respectively; the population typical value of the peripheral compartment distribution volume was 28.60 L, and the population typical value of the inter-compartment clearance was 23.86 L / h.

[0118] Table 2 Covariate screening process

[0119] Model Model composition -2LL Δ-2LL Model 1 Base model (2-compartment proportion residual variance model) 3128 -- Model 2 Model 1 + central compartment clearance - creatinine clearance 3097 -31 Model 3 Model 2 + intercompartmental clearance - creatine kinase isoenzymes 3089 -39 Model 4 Model 3 + Central Compartmental Clearance - Total Bilirubin 3080 -48 Model 5 Model 4 + Volume of Distribution in Peripheral Compartment - Glutathione Reductase 3073 -55 Model 6 Model 5 + peripheral distribution volume - glutathione reductase 3080 -48 Model 7 Model 6+Internal clearance-total bilirubin 3089 -39 Model 8 Model 7+ Intercompartmental clearance-creatine kinase isoenzymes 3097 -31

[0120] 4. Population pharmacokinetic model evaluation

[0121] The evaluation data of the population pharmacokinetic model came from the model establishment data set, and the evaluation methods used goodness-of-fit plots, visual prediction tests, and bootstrapping methods.

[0122] The fit diagram includes a model diagnostic diagram based on prediction and a model diagnostic diagram based on residuals. The model diagnostic diagram based on prediction includes dependent variable-group predicted value and dependent variable-individual predicted value, which are used to evaluate the consistency between the observed value and the group / individual predicted value. The higher the overlap between the reference line and the trend line, the better the fit of the model; the model diagnostic diagram based on residuals includes conditional weighted residual-time and conditional weighted residual-group predicted value, which are used to judge whether the setting of the residual model is reasonable. In the present invention, time is displayed as time after administration. If the model fits well, the value of the conditional weighted residual should be distributed between ±2 and symmetrically distributed on both sides of the reference line; if Figure 3 As shown, the observed blood drug concentration of the final model maintains a high consistency with the group predicted value or the individual predicted value, and the conditional weighted residual and the time after administration or the group predicted value are symmetrically distributed on both sides of the reference line (y=0), and there is no obvious trend change within the range of ±2, which proves that the final two-compartment model can fully describe the pharmacokinetic data of the modeling group.

[0123] Visual test evaluates the fitting degree of the model by generating simulated data sets and comparing the distribution characteristics of the 5% quantile, 50% quantile and 95% quantile of the observed data sets and the simulated data sets at each time point. When the 5% quantile, median and 95% quantile of the actual value data set in the figure fall within the 95% confidence interval corresponding to the simulated data, the simulated data is considered to be consistent with the observed data distribution; Figure 4 As shown, most of the observed values ​​are distributed within the 90% confidence interval corresponding to the 5%, 50%, and 95% prediction percentiles, indicating that the predicted values ​​of the simulated data are consistent with the observed values, that is, the prediction accuracy of the final model is good.

[0124] The bootstrap method generates 1000 bootstrap data sets by resampling, fits the model with the bootstrap generated data set, and estimates the model parameters. When the 95% confidence interval (2.5% to 97.5%) of the estimated parameters of the bootstrap data set contains the estimated parameters of the original model, and the parameters are successfully estimated more than 800 times in 1000 bootstrap data sets, that is, the robustness rate is >80%, the model is considered to be verified by the bootstrap method. The results of the bootstrap method are shown in Table 3. The final model parameter estimates all fall within the 95% confidence interval of the estimated parameters of the bootstrap data set, which indicates that the established model has good stability.

[0125] Table 3 Bootstrap results of the final population pharmacokinetic model

[0126]

[0127]

[0128] 3. External Validation

[0129] The data for model evaluation comes from the data set used to build the model, so the results fail to effectively reflect the performance of the model when expanded and applied to other conditions. In principle, external validation is the most stringent model evaluation method. Therefore, in accordance with the requirements for model building data, independent external data are collected to externally validate the final Pop-PK model; Process for collecting independent external data: According to the method in Example 1, relevant data of 16 patients who received PB treatment in the hospital from September 2022 to January 2023 were collected as an independent external data validation set.

[0130] The first step is to diagnose the prediction performance of the model. The relative prediction error (PE%) is calculated using the following formula to compare the predicted concentration of the population with the actual value. If the relative prediction error deviates from the normal distribution, the median prediction error (MDPE) is used to evaluate the accuracy, and the median absolute prediction error (MAPE) is used to evaluate the precision. The percentage of prediction errors falling within ±20% (F 20 , -20% ≤ PE% ≤ 20%) and the percentage of prediction errors falling within ±30% (F 30 , -30% ≤ PE% ≤ 30%) represents a comprehensive index of accuracy and precision. When the median prediction error ≤ ±15%, the median absolute prediction error ≤ 30%, the percentage of prediction errors falling within ±20% > 35%, and the percentage of prediction errors falling within ±30% > 50%, the candidate model is considered clinically acceptable. As shown in Table 4, the median prediction error, the median absolute prediction error, the percentage of prediction errors falling within ±20%, and the percentage of prediction errors falling within ±30% of the current model are 0.89%, 23.89%, 40.00% and 57.14%, respectively, which meet the clinical acceptance criteria.

[0131]

[0132] Table 4 Diagnosis results of the model based on prediction

[0133]

[0134]

[0135] The second step is to diagnose the simulation performance of the model through the visual prediction test of prediction and variation correction. Under the condition of setting 2000 simulations, the 5th, 50th and 95th simulation values ​​and their corresponding 90% confidence intervals are calculated and compared with the actual values. In theory, the observed values ​​outside the 90% interval of the simulation value should not exceed 10%. The results are shown in Figure 2. Figure 5As shown, the 50th and 95th percentiles of the observed values ​​are in good agreement with the simulated values. In addition, there are no test values ​​that exceed the 90% confidence interval of the simulated values, which indicates that the simulation results of the Pop-PK model fully capture the test results of TDM.

[0136] Finally, Bayesian prediction was used to evaluate the impact of prior concentration on model predictability using maximum a posteriori Bayesian (MAPB) prediction, which is a method of evaluating individual patient PK parameters using the Pop-PK model and actual observations. For each patient, after assigning 0 to 2 prior concentration data in the data set, the patient's steady-state blood drug concentration was predicted by the model and compared with the actual value, and the individual relative prediction error (IPRED) was calculated. The median individual prediction error (MDIPE), the median individual absolute prediction error (MAIPE), and the percentage of individual prediction errors falling within ±20% (IF 20 , -20% ≤ IPRED% ≤ 20%) and the percentage of prediction errors falling within ±30% (IF 30 , -30% ≤ IPRED% ≤ 30%) to evaluate the overall prediction performance of the model. The clinical acceptance criteria of each evaluation index are the same as those of “simulated performance diagnosis”; Figure 6 As shown in Table 5, the median of individual prediction errors ranged from -8.06% to 0.89%, the median of individual absolute prediction errors ranged from 9.97% to 23.89%, the percentage of individual prediction errors within ±20% ranged from 40.00% to 80.00%, and the percentage of individual prediction errors within ±30% ranged from 57.14% to 94.29%, all of which met the acceptance criteria, that is, the predictive ability of the current final model met the clinical acceptance criteria.

[0137]

[0138] Table 5 Bayesian prediction results

[0139]

[0140] 4. Model-based simulation

[0141] The relevant data of 46 patients with severe infection who received PB treatment in the hospital from February to November 2023 were collected to form a simulated data set.

[0142] The results of model development showed that creatinine clearance (Crcl) was a covariate affecting the final two-compartment model clearance CL. Therefore, 46 patients with severe infection who received PB treatment were stratified according to renal function, and Phoenix NLME software was used to predict the patients' renal clearance at different dosages (50WU q12 h, 75WU q12 h, 100WU LD+50WU MD q12 h, 100WU LD+75WU MD q12 h, 150WU LD+75WU MD q12 h). h), steady-state trough concentrations and steady-state peak concentrations at different injection times (1h, 2h, 2h), and pharmacokinetic parameters of different individuals were obtained. Then, the requirements for steady-state blood drug concentration monitoring in the "Chinese Expert Consensus on the Clinical Application of Polymyxin" were used, that is, the area under the drug-time curve within 24h at steady state was between 50 and 100mg·h / L, which was equivalent to maintaining the average steady-state blood drug concentration at 2-4mg / L. This was selected as the basis for establishing the recommended regimen. Then, the recommended regimen was summarized through the prediction parameters based on the final model, and the dosing regimen recommendation based on renal function status was established. The results are as follows Figure 7 .

[0143] First, for patients with normal renal function (80≤Crcl≤120mL / min). Among the 46 patients included, 7 had normal renal function. The individual pharmacokinetic parameters and final recommended regimen based on the Pop-PK model are as follows: Figure 7 As shown in A, the pharmacokinetic parameters of patients are as follows: central distribution volume (V) is between 30.23 and 57.71 L, peripheral distribution volume (V2) is between 13.74 and 44.81 L, central clearance (CL) is between 2.19 and 7.76 L / h, and inter-compartmental clearance (Q) is 23.86 L / h; for patients with central distribution volume V≤40.25 L and central clearance CL≤3.65 L / h, the model recommends 75 WU 2hq12 h (a maintenance dose of 75 WU intravenously injected within 2 hours, with a dosing interval of 12 hours), 75 WU 2h q12 h, and 100 WU_75 WU q12 h (a loading dose of 100 WU is given first, followed by a maintenance dose of 75 WU every 12 h); for patients with central volume of distribution V>40.25 L and central clearance CL>3.65 L / h, the recommended regimen is 100 WU q12 h.

[0144] For the 7 patients with hyperrenal function (Crcl>120mL / min), the final recommended regimen based on the Pop-PK model was as follows Figure 7As shown in B, the pharmacokinetic parameters are: central compartment distribution volume (V) is between 20.92 and 41.21 L, peripheral compartment distribution volume (V2) is between 5.59 and 43.11 L, central compartment clearance (CL) is between 2.61 and 5.87 L / h, and intercompartmental clearance (Q) is 23.86 L / h; for these patients, the model recommends three main regimens, including 75WU 2h q12 h, 100WU_75WU 2h q12 h, and 100WU_75WU 2h q12 h.

[0145] For the 6 patients with mild renal impairment (51≤Crcl≤79mL / min), the final recommended regimen based on the Pop-PK model was as follows Figure 7 As shown in C, pharmacokinetic parameters: central distribution volume (V) is between 15.60 and 80.86 L, peripheral distribution volume (V2) is between 4.39 and 24.31 L, central clearance (CL) is between 1.47 and 5.32 L / h, and inter-compartmental clearance (Q) is 23.86 L / h. For patients with central distribution volume (V) ≤ 60.51 L and central clearance (CL) ≤ 2.42 L / h, the model recommended regimens include 50 WU 2h q12 h, 75 WU 2h q12 h, and 100 WU_50 WU q12 h; for patients with central distribution volume (V) > 60.51 and central clearance (CL) > 2.42 L / h, the model recommended regimen is 100 WU q12 h.

[0146] For 8 patients with moderate renal impairment (31≤Crcl≤50mL / min), individual pharmacokinetic parameters and final recommended regimens are as follows Figure 7 As shown in D, the pharmacokinetic parameters are: central compartment distribution volume (V) is between 18.03 and 80.86 L, peripheral compartment distribution volume (V2) is between 4.43 and 38.55 L, central compartment clearance (CL) is between 1.46 and 5.32 L / h, and intercompartmental clearance (Q) is 23.86 L / h. For these patients, the recommended regimens based on the Pop-PK model include: for patients with V≤43.33L and CL≤1.92L / h, 50WU 2h_q12 h or 100WU_50WU_q12 h is recommended; for patients with 43.33<V≤64.21L and 1.92<CL≤3.26L / h, 75WU 1h_q12 h or 100WU_75WU_q12 h is recommended; when CL>3.26L / h, the 100WU q12 h dosage regimen is recommended.

[0147] Finally, for 20 patients with severe renal impairment (Crcl ≤ 30 mL / min), the final recommended regimen based on the Pop-PK model is as follows: Figure 7As shown in E, the pharmacokinetic parameters are: central distribution volume (V) is between 13.89 and 142.76 L, peripheral distribution volume (V2) is between 4.54 and 72.84 L, central clearance (CL) is between 0.89 and 4.02 L / h, and inter-compartmental clearance (Q) is 23.86 L / h; for patients with CL≤2.24 L / h, the main recommended regimens are 50 WU q12 h, 100 WU_50 WU 3 h q12 h, 75 WU 3 h q12 h, and 100 WU_75 WU q12 h; for patients with 2.24<CL≤4.02 L and V≤104.57 L, the recommended regimens are 75 WU 1 h q12 h, 75 WU 2 h q12 h, 100 WU_75 WU 1 hq12 h h; for patients with V>104.57L, the recommended regimen is 100WU q12 h.

[0148] 5. Model Application

[0149] A case study of early precision medication of polymyxin B in patients with multidrug-resistant severe infections guided by the dosing regimen recommended by the population pharmacokinetic model of polymyxin B.

[0150] Case 1: This is a real case of using the accurate dosage prediction of polymyxin B in patients with multidrug-resistant severe infections to guide the accurate drug administration of systemic infections, which is completely different from the existing polymyxin B prediction model that has no clinical application. The details are as follows:

[0151] The patient, male, 56 years old, was reported by his family members to have been "unconscious for 5 hours" and was admitted to the emergency department of the First Affiliated Hospital of Nanchang University on April 13, 2023 for "intracranial hemorrhage". He underwent craniotomy and hematoma removal and was transferred to the neurosurgery ICU after surgery. Admission diagnosis: 1. Brain herniation (primary diagnosis), 2. Intracerebral hemorrhage.

[0152] Treatment process: After admission to the neurosurgery ICU, relevant examinations were completed, and CT showed: 1. After the left temporoparietal lobe-basal ganglia hemorrhage broke into the ventricle, the left frontal and temporal bones were partially absent, a small amount of fluid, blood and air accumulation in the surgical area, and related brain tissue swelling; 2. Small nodules in the anterior segment of the right upper lobe, which were suspected to be proliferative lesions; 3. Strip-like slightly high-density shadows between the two lower lungs, which were suspected to be infectious lesions; 4. A small amount of pleural effusion on both sides; 5. Liver cysts. Transferred to the Department of Critical Care Medicine on April 18, and on May 8, inflammatory indicators: white blood cell count 4.15×10 9 / L, C-reactive protein: 16.64mg / L, procalcitonin 0.16ng / mL, sputum culture reported multidrug-resistant Klebsiella pneumoniae. After admission, he was given anti-infection, nutritional support, hemostasis, dehydration to reduce intracranial pressure, acid suppression and stomach protection, and anti-epileptic treatment.

[0153] After consultation with the pharmacist, the physician considered giving polymyxin B and combining it with tigecycline for anti-infection treatment. The routine dosing time was determined: 8:00 am and 20:00 pm. The patient's individual information and dosing time were input to simulate the patient's steady-state blood drug concentration under the two dosing regimens of LD 100WU+MD50WU q12 h; LD 100WU+MD 75WU q12 h. The model prediction results are shown in Table 6. Considering the patient's etiological diagnosis, the final dosing regimen of "first load 100WU, followed by a maintenance dose of 75WU, intravenous infusion for 2 hours, twice a day" was determined. According to the patient's monitoring results (Table 7), the patient's steady-state trough concentration was normal, while the peak concentration was slightly higher. The average steady-state blood drug concentration and the area under the drug-time curve (AUCss, 24h) within 24 hours at steady state met the requirements. According to the definition of acute kidney injury in the KDIGO Clinical Practice Guidelines for Acute Kidney Injury, the patient's creatinine clearance decreased by 3μmol / L (no more than 26.5μmol / L) within 48 hours, which means that he did not have acute kidney injury, so the dosage did not need to be adjusted. After 14 days of treatment (May 22), the patient's blood culture results showed no bacterial growth, no anaerobic bacteria growth, and sputum culture results showed no multidrug-resistant bacteria. The patient's creatinine clearance was measured to be 135mL / min, and no acute kidney injury occurred. Therefore, polymyxin B was discontinued, and imipenem-cilastatin sodium was continued for anti-infection.

[0154] Table 6 Predicted values ​​of steady-state blood drug concentrations under different dosage regimens

[0155]

[0156] Note: 100WU, 50WU 1h_12h: First give a loading dose of 100WU, then give a maintenance dose of 50WU, complete the injection within 1 hour, and the interval between doses is 12 hours; 100WU, 50WU 2h_12h: First give a loading dose of 100WU, then give a maintenance dose of 50WU, complete the injection within 2 hours, and the interval between doses is 12 hours; 100WU, 50WU 3h_12h: First give a loading dose of 100WU, then give a maintenance dose of 50WU, complete the injection within 3 hours, and the interval between doses is 12 hours; 100WU, 75WU1h_12h: First give a loading dose of 100WU, then give a maintenance dose of 75WU, complete the injection within 1 hour, and the interval between each dose is 12 hours; 100WU, 75WU 2h_12h: First give a loading dose of 100WU, then give a maintenance dose of 75WU, complete the injection within 2 hours, and each administration interval is 12 hours; 100WU, 75WU 3h_12h: First give a loading dose of 100WU, then give a maintenance dose of 75WU, complete the injection within 3 hours, and each administration interval is 12 hours; C ss,min : Steady-state trough concentration; Css,max : Steady-state peak concentration.

[0157] Table 7 Comparison of steady-state blood drug concentration predicted by Pop-PK model and actual TDM detection value

[0158] project <![CDATA[C ss,min (μg / mL)]]> <![CDATA[C ss,max (μg / mL)]]> <![CDATA[C ss,avg (μg / mL)]]> <![CDATA[AUC ss,24h (mg·h / L)]]> <![CDATA[PRED model ]]> 2.34 5.07 3.70 85.09 <![CDATA[OBS TDM ]]> 2.34 5.51 3.97 93.02 IPE(%) 0.00 -7.98 -5.61 -8.52

[0159] Note: PRED model :Prediction results of steady-state blood drug concentration based on Pop-PK model; OBS TDM : actual TDM detection value; IPE: individual relative prediction error; C ss,min : Steady-state trough concentration; C ss,max : Steady-state peak concentration; C ss,avg : Mean steady-state plasma concentration; AUC ss,24h : The area under the plasma concentration-time curve within 24 hours at steady state.

[0160] Case 2: Application of the accurate prediction model for the dosage of polymyxin B for multidrug-resistant severe infections. This is a real case to guide the precise drug administration for intracranial infection, which is completely different from other polymyxin B prediction models that have no clinical application.

[0161] Patient Xu, male, 58 years old, was admitted to the Department of Critical Care Medicine, First Affiliated Hospital of Nanchang University on May 11, 2023 due to "sudden nausea and vomiting with impaired consciousness for 17 days". Admission diagnosis: 1. Intracranial infection, 2. Basal ganglia hemorrhage (primary diagnosis), 3. Hypertension stage 3 (extremely high risk);

[0162] Diagnosis and treatment process: After admission, relevant examinations were completed. Routine cerebrospinal fluid examination showed: turbid and yellow cerebrospinal fluid, Pan's test 2+, 90% neutrophils. On May 13, 2023, the neurosurgery consultation performed ventricular drilling and drainage under general anesthesia, and at the same time, ventricular and lumbar drainage tubes were left in place for pus drainage. The culture results of cerebrospinal fluid specimens showed multidrug-resistant Acinetobacter baumannii, and the culture results of blood specimens showed multidrug-resistant Staphylococcus urealyticum. Comprehensive treatment was given, including anti-infection, brain protection, enteral nutrition, vasoactive drugs to maintain blood pressure, mechanical ventilation, and correction of hypoproteinemia.

[0163] After consultation with the pharmacist, the clinician considered giving polymyxin B for infection prevention, and added meropenem and tigecycline by intrathecal injection. The regular administration time was determined to be 20:00 and 08:00, and the pharmacist asked the nurse to accurately grasp the time. Considering the patient's renal function (creatinine clearance: 152mL / min), the model simulated the blood drug concentration of polymyxin B under three types of dosage regimens: 100WU_50WU q12 h and 75WU q12 h, 100WU_75WU q12 h, and the results are shown in Table 8. Considering the patient's infection: cerebrospinal fluid biochemistry showed a total white blood cell count of 1100 / uL: neutrophil segmented granulocytes 95%. It was finally determined that a loading dose of 100WU was given for the first time, followed by a maintenance dose of 75WU (25mL / h) intravenously injected with a micropump, with a dosing interval of 12h. According to the results in Table 9, the patient's average steady-state blood drug concentration and the area under the drug-time curve (AUCss, 24h) within 24 hours at steady state meet the requirements; in addition, according to the definition of acute kidney injury in the KDIGO Clinical Practice Guidelines for Acute Kidney Injury, by monitoring the patient's renal function, it was found that the serum creatinine value within 48 hours after medication decreased by 17.9μmol / L (<26.5), and the urine volume for 6 hours was >0.5mL / kg / h. Therefore, it was considered that no adverse reactions of renal toxicity occurred and no adjustment of the dosing regimen was required. On May 20, the cerebrospinal fluid biochemistry results showed that the total white blood cell count was 150 / uL, the neutrophil segmented granulocytes were 90%, and the inflammatory indicators showed a downward trend; in addition, the 48-hour cerebrospinal fluid culture results reported no sterile growth, and the sputum culture results showed no multidrug-resistant bacteria, so polymyxin B was discontinued.

[0164] Table 8 Predicted values ​​of steady-state blood drug concentrations under different dosage regimens

[0165]

[0166] Note: 100WU, 50WU 1h_12h: First give a loading dose of 100WU, then give a maintenance dose of 50WU, complete the injection within 1 hour, and the interval between doses is 12 hours; 100WU, 50WU 2h_12h: First give a loading dose of 100WU, then give a maintenance dose of 50WU, complete the injection within 2 hours, and the interval between doses is 12 hours; 100WU, 50WU 3h_12h: first give a loading dose of 100WU, then give a maintenance dose of 50WU, and complete the injection within 3 hours, with a dosing interval of 12 hours; 100WU, 75WU1h_12h: first give a loading dose of 100WU, then give a maintenance dose of 75WU, and complete the injection within 1 hour, with a dosing interval of 12 hours; 100WU, 75WU2h_12h: first give a loading dose of 100WU, then give a maintenance dose of 75WU, and complete the injection within 2 hours, with a dosing interval of 12 hours; 100WU, 75WU 3h_12h: first give a loading dose of 100WU, then give a maintenance dose of 75WU, and complete the injection within 3 hours, with a dosing interval of 12 hours; C ss,min : Steady-state trough concentration; C ss,max : Steady-state peak concentration.

[0167] Table 9 Comparison of steady-state blood drug concentration predicted by Pop-PK model and actual TDM detection value

[0168] project <![CDATA[C ss,min (μg / mL)]]> <![CDATA[C ss,max (μg / mL)]]> <![CDATA[C ss,avg (μg / mL)]]> <![CDATA[AUC ss,24h (mg·h / L)]]> <![CDATA[PRED model ]]> 1.96 5.13 3.54 79.57 <![CDATA[OBS TDM ]]> 1.74 6.00 3.73 89.50 IPE(%) -12.97 -14.40 -8.53 -11.09

[0169] Note: PRED model :Prediction results of steady-state blood drug concentration based on Pop-PK model; OBS TDM : actual TDM detection value; IPE: individual relative prediction error; C ss,min : Steady-state trough concentration; C ss,max : Steady-state peak concentration; C ss,avg : Steady-state mean blood concentration; AUC ss,24h : The area under the plasma concentration-time curve within 24 hours at steady state.

[0170] Example 3: Establishment of a clinical decision support strategy for accurate prediction of polymyxin B dosage for patients with severe infections

[0171] like Figure 8 As shown, the present invention provides a method for establishing a clinical decision support system for accurately predicting the dosage of polymyxin B for patients with severe infections. The clinical decision support system for accurately predicting the dosage of polymyxin B for patients with severe infections includes a data support layer, a model development layer, a model verification layer, and a model application layer.

[0172] 1. Data support layer

[0173] Based on the data of patients with severe infections in the intensive care unit (ICU) of a large tertiary medical center in the past five years, a polymyxin B database was formed. The integrated patient information database was used for retrieval, cohort discovery, online exploratory analysis, and export of required data analysis.

[0174] Through set operation techniques such as inclusion and exclusion, accurate screening of the research population is achieved, and a prospective cohort study is conducted. The data scope of sample information collection comes from the hospital's HIS, LIS, PACS, electronic medical record data, data mined from text in unstructured data in electronic medical records, patient follow-up information, etc., and certain combinations are made according to needs, such as basic patient information, medical history collection, microbial infection, comorbidities, and concomitant medications.

[0175] Through the arbitrary selection of customized research cohorts and parameters, exploratory analysis is achieved, supporting one-way frequency analysis, chi-square test analysis, general linear correlation analysis, ordinal variable correlation analysis, t-test, and one-way analysis of variance; by constructing indicator analysis of any research cohort, including mean, standard deviation, percentile table, frequency distribution diagram, change trend diagram, correlation analysis, matrix correlation analysis, etc.

[0176] 2. Model development layer

[0177] Based on the data analysis system, the mirror data of the precision medication database for multidrug-resistant severe infections was integrated, and a clinical decision support model for the precision medication of polymyxin B was established based on the comprehensive application of classification algorithms such as Bayes, decision tree, SVM, clustering algorithms such as Isodata, and association algorithms such as improved Apriori method.

[0178] The multi-category data sources ensure the universality and accuracy of the model. The system's built-in Json data storage technology is easy for people to read and write, and is also easy for machines to parse and generate, and effectively improves network transmission efficiency. The system loads classification algorithms such as decision trees and Svm, clustering algorithms such as Isodata, and improved association algorithms such as the Apriori method, which optimize connection operations, improve algorithm efficiency, and increase human-computer interaction structures.

[0179] 3. Model Validation Layer

[0180] By continuously collecting large clinical samples, testing data analysis, TDM monitoring technology, drug-resistant bacteria analysis, etc., the model is continuously applied and repeatedly verified, so that massive data can be more effectively optimized and data mined under the combination of machine learning algorithms and data access technology.

[0181] 4. Model application layer

[0182] It supports the implementation and management of multiple scientific research projects at the same time, not limited to one clinical department or one scientific research project. It can realize the centralized and unified management of clinical scientific research in the entire hospital and personalized support for a single scientific research project.

[0183] The model supports the use of serum creatinine, a clinically available detection indicator, as the main covariate to prospectively guide the individualized medication of polymyxin B. The core technology in the present invention is the application of population pharmacokinetic research methods, which takes the population as the object, analyzes all the data of the same population, and combines the classical pharmacokinetic model with the population statistical model. It is a new pharmacokinetic research method that studies different variations with the population mean and variance of the population pharmacokinetic parameters. It can quantitatively examine the effects of different physiological factors, pathophysiological factors, and drug interactions on the in vivo process of the drug, and quantify the factors that affect the drug pharmacokinetic process; at the same time, the Bayesian feedback method in population pharmacokinetics can be used to predict the individual pharmacokinetic parameters of patients to achieve accurate prediction and dose titration of drug dosage.

[0184] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0185] Although preferred embodiments of the present invention have been described, additional changes and modifications may occur to these embodiments once those skilled in the art are aware of the basic inventive concepts.

[0186] Obviously, those skilled in the art may make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A method for accurately predicting the dosage of polymyxin B for patients with multidrug-resistant severe infections, characterized in that: The following steps are involved: Collect personal data of patients with multidrug-resistant severe infections and clinical drug use data of polymyxin B in patients with multidrug-resistant severe infections as data for model establishment; Based on the model establishment data, the basic model was established by using the compartmental model fitting, fixed effect parameter prediction and random effect model; then the covariates were determined by analyzing the continuous variables and categorical variables of the model establishment data, and the covariates were included in the basic model to obtain the population pharmacokinetic model, and then the prediction ability and stability of the population pharmacokinetic model were tested and evaluated using the goodness of fit plot, visualization test and bootstrap method; External validation: measures the inaccuracy and bias of the predictive ability of the population pharmacokinetic model after computational testing based on the collection of independent external data; Model-based simulation: The separately collected external data were used as the simulation data set. After stratification by covariates, the population pharmacokinetic model was used to predict the steady-state blood drug concentration of patients under different dosage regimens. The different dosage regimens were summarized according to the pharmacokinetic parameters to obtain the precise dosage recommendation of polymyxin B for patients with multidrug-resistant severe infections.

2. The method for accurately predicting the dosage of polymyxin B for patients with multidrug-resistant severe infections according to claim 1, characterized in that: External validation included diagnostics of the predictive performance of the population pharmacokinetic model after test evaluation, diagnostics of the simulation performance of the population pharmacokinetic model, and Bayesian prediction; The diagnosis of the predictive performance of the population pharmacokinetic model after the test and evaluation is as follows: the relative prediction error of the population pharmacokinetic model is calculated and tested and evaluated; if the relative prediction error deviates from the normal distribution, the median prediction error is used to evaluate the accuracy, and the median absolute prediction error is used to evaluate the precision; when the median prediction error is ≤±15%, the median absolute prediction error is ≤30%, the percentage of prediction errors falling within ±20% is >35%, and the percentage of prediction errors falling within ±30% is >50%, the candidate model is considered to be clinically acceptable.

3. The method for accurately predicting the dosage of polymyxin B for patients with multidrug-resistant severe infections according to claim 2, characterized in that: The relative prediction error calculation formula is:

4. The method for accurately predicting the dosage of polymyxin B for patients with multidrug-resistant severe infections according to claim 2, characterized in that: The simulation performance of the population pharmacokinetic model is diagnosed by visual prediction test of prediction and variation correction.

5. The method for accurately predicting the dosage of polymyxin B for patients with multidrug-resistant severe infections according to claim 2, characterized in that: Bayesian predictions were: Maximum a posteriori Bayesian predictions were used to assess the impact of prior concentrations on the predictability of population pharmacokinetic models.

6. The method for accurately predicting the dosage of polymyxin B for patients with multidrug-resistant severe infections according to claim 1, characterized in that: The simulated data set includes the personal data of patients with severe infections who received polymyxin B treatment in the hospital and the clinical medication data of polymyxin B.

7. The method for accurately predicting the dosage of polymyxin B for patients with multidrug-resistant severe infections according to claim 1, characterized in that: It also includes data processing; Among them, Microsoft Excel was used to organize the data files used for Phoenix NLME software analysis for model building data. The organized items included TIME, DV, ADDL, II, AMT, EVID, MVD and covariates, and saved them in .csv format recognizable by Phoenix NLME for the establishment of the basic model.

8. The method for accurately predicting the dosage of polymyxin B for patients with multidrug-resistant severe infections according to claim 7, characterized in that: The compartment model includes a one-compartment model and a two-compartment model.

9. The method for accurately predicting the dosage of polymyxin B for patients with multidrug-resistant severe infections according to claim 8, characterized in that: The prediction of fixed effect parameters includes determining the inter-individual random effect model and the residual model. The inter-individual random effect model includes additive, proportional and exponential forms; the residual model includes additive, proportional and mixed forms; the basic model is determined according to the minimum AIC and -2LL minimum principles.

10. A population pharmacokinetic model for accurate prediction of polymyxin B dosage, characterized in that: The method for accurately predicting the dosage of polymyxin B for patients with multidrug-resistant severe infections according to claim 1 is obtained, and the population pharmacokinetic model for accurately predicting the dosage of polymyxin B is: V(L)=52.09*exp(ηηV) V2(L)=28.60*exp(ηV2) Q(L / h)=23.86*exp(ηηQ) Wherein, V, central compartment distribution volume, unit: liter; exp, exponential function with natural constant e as base; η, random variable obeying normal distribution; V2, peripheral compartment distribution volume, unit: liter; CL, central compartment clearance rate, unit: liter / hour; CrCl, serum creatinine clearance rate; Q, inter-compartmental clearance rate, unit: liter / hour.

Images