A method, apparatus, medium, and program product for constructing a survival model for catheter-related thrombosis based on machine learning

Abstract

The application provides a method, device, medium and program product for constructing a survival model of catheter-related thrombosis based on machine learning, and relates to the field of intelligent medical treatment. The method comprises the following steps: acquiring training set samples, which include sample basic information, catheter types, systemic drug treatment, any one or several data in tumor sites and corresponding result information of the samples; processing the data of the systemic drug treatment and the tumor sites by using a factor folding method to obtain processed data characteristics of the systemic drug treatment and the tumor sites; converting the basic information and the catheter type data into numerical features; and inputting the data characteristics, the numerical features and the result information into a machine learning model to obtain an SM-CRT model. The application collects catheter insertion data of tumor patients to construct a survival model of catheter-related thrombosis, namely an SM-CRT model, which is a survival model for providing a risk score and a catheter thrombosis risk time distribution.

Description

Technical Field

[0001] This invention relates to the field of intelligent healthcare, and more specifically, to a method, apparatus, medium, and program product for constructing a survival model of catheter-related thrombosis based on machine learning. Background Technology

[0002] Central venous catheters facilitate drug delivery and improve patient comfort, but they also introduce catheter-related complications. Catheter-related thrombosis (CRT) is one of the major complications affecting cancer patients. Previous reports indicate that the incidence of CRT in cancer patients ranges from 2.4% to 61.5%. Addressing this issue involves four consecutive tasks: first, selecting the optimal catheter type with the lowest risk for each patient (primary prevention); second, identifying high-risk catheter placement cases and applying preventative strategies such as antithrombotic drugs (primary prevention); third, optimizing catheter insertion timing to avoid insertion during the most dangerous period and applying additional preventative strategies when necessary (primary prevention); and fourth, optimizing ultrasound examination cycles based on the patient's cumulative probability to enable early detection of CRT (secondary prevention) and timely treatment (tertiary prevention). However, predictive models capable of accurately achieving all of these functions have not yet been established and remain uncharacterized. Summary of the Invention

[0003] This invention aims to at least address one of the technical problems existing in the prior art. To this end, this invention provides a method, device, medium, and program product for constructing a survival model for catheter-related thrombosis based on machine learning. The method of this invention constructs an SM-CRT by collecting catheter insertion data from cancer patients. This model provides a risk score (crank, used to assess general risk over time and calculate the c-index) and a predicted survival distribution (distr, used to guide catheter and ultrasound follow-up times).

[0004] The first aspect of this application discloses a method for constructing a survival model of catheter-related thrombosis based on machine learning, the method comprising:

[0005] S101, Obtain training set samples including basic sample information, catheter type, systemic drug treatment, tumor site, and any one or more data and corresponding result information of the samples;

[0006] S102, The factor folding method is used to process the data on systemic drug therapy and tumor site to obtain the processed data features of systemic drug therapy and tumor site; the basic information and catheter type data are converted into numerical features;

[0007] S103, the data features, numerical features and result information are input into the machine learning model for training to obtain the survival model of catheter-related thrombosis, which is the SM-CRT model.

[0008] The second aspect of this application discloses a method for identifying the risk of CRT based on a survival model of catheter-related thrombosis, the method comprising:

[0009] 201. Obtain data on the subjects, including basic information, type of catheter used, systemic drug treatment, and tumor location;

[0010] 202, Input the data into the SM-CRT model described in the first aspect of this application to calculate the risk value;

[0011] 203. Output auxiliary prediction results based on the risk value to indicate whether the subject has a high or low risk of developing CRT. Specifically, when the risk value is greater than the first threshold, output auxiliary prediction results indicating a high risk of developing CRT; when the risk value is less than the first threshold, output auxiliary prediction results indicating a low risk of developing CRT.

[0012] A third aspect of this application discloses a method for guiding catheter type selection based on a survival model of catheter-related thrombosis, the method comprising:

[0013] 301. Obtain basic information about the subject, systemic drug treatment, tumor site data, and the type of catheter to be used;

[0014] 302. Input the basic information, systemic drug treatment, tumor site data, and expected catheter type into the SM-CRT model described in the first aspect of this application, and output the predicted risk value of CRT occurring with different catheter types; determine the final catheter type to be used based on the predicted risk value.

[0015] The fourth aspect of this application discloses a method for guiding extubation time based on a survival model of catheter-related thrombosis, the method comprising:

[0016] 401. Obtain data on the subject, including basic information, type of catheter used, systemic drug treatment, and tumor location;

[0017] 402, input the data into the SM-CRT model described in the first aspect of this application to predict the high CRT risk period, and predict the extubation time of the subject based on the high CRT risk period.

[0018] The fifth aspect of this application discloses a method for guiding ultrasound follow-up cycles based on a survival model of catheter-related thrombosis, the method comprising:

[0019] 501, to obtain data on the subject including basic information, type of catheter used, systemic drug treatment, tumor location, and first-time data;

[0020] 502, Input the data into the SM-CRT model described in the first aspect of this application to calculate the cumulative distribution function (CDF) at different times;

[0021] 503. If the CDF of the second time is increased by a second threshold compared to the first time, then the second time is the predicted ultrasound follow-up time adjacent to the first time; the ultrasound follow-up time can be a specific time point; the first time is the time of the last ultrasound examination.

[0022] The sixth aspect of this application discloses a computer device, the device comprising: a memory and a processor; the memory being used to store a computer program; the processor executing the computer program to implement the steps of the above-described method.

[0023] The seventh aspect of this application discloses a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described above.

[0024] The eighth aspect of this application discloses a computer program product, including a computer program that, when executed by a processor, implements the steps of the above-described method.

[0025] This application has the following beneficial effects:

[0026] 1. This application innovatively discloses a method for constructing a survival model for catheter-related thrombosis (CRT) based on machine learning. This method collects large-scale, multi-center catheter data from cancer patients and comprehensively records variables such as general patient information, catheter information, tumor status, systemic therapy, and blood biomarkers. A machine learning model (SM-CRT) is developed to predict the general risk and temporal distribution of CRT. The performance and application of SM-CRT have been well validated on both training and testing datasets.

[0027] 2. This application innovatively applies the SM-CRT model to guide the selection of catheter type, stratifies the risk of thrombosis prevented by CRT, assists in the decision-making of catheter placement timing (PDF is used to determine the high-risk and low-risk periods of CRT at the patient level; on this basis, PDF can guide the selection of catheter removal time to avoid the most dangerous period), and optimizes the ultrasound follow-up period, ultimately improving patient prognosis. Attached Figure Description

[0028] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0029] Figure 1 This is a schematic diagram of the method flow provided in the first aspect of the present invention;

[0030] Figure 2This is a schematic diagram of the method flow provided in the second aspect of the present invention;

[0031] Figure 3 This is a schematic diagram of the method flow provided in the third aspect of the present invention;

[0032] Figure 4 This is a schematic diagram of the method flow provided in the fourth aspect of the present invention;

[0033] Figure 5 This is a schematic diagram of the method flow provided in the fifth aspect of the present invention;

[0034] Figure 6 This is a schematic diagram of a computer device provided in an embodiment of the present invention;

[0035] Figure 7 This is a schematic diagram of the architecture of an exemplary computing device provided in an embodiment of the present invention;

[0036] Figure 8 This is a schematic diagram of the storage medium provided in an embodiment of the present invention;

[0037] Figure 9 This is an overview diagram of catheter-related thrombotic (CRT) events in the training set provided in this embodiment of the invention; Figure 9 a. Kaplan-Meier curves of CRT-free survival in patients in the training set. Figure 9 b. Risk estimates for CRT events obtained using global and local bandwidth selection algorithms and boundary kernel functions. Estimated risk rates for month 6, year 1, and year 2 are marked with vertical dashed lines. CRT = catheter-related thrombosis; D11 = day 11; M6 = month 6; Y1 = year 1; Y2 = year 2;

[0038] Figure 10 This is a forest plot of univariate Cox analysis showing hazard ratios for CRT-free survival, provided in an embodiment of the present invention. HR = hazard ratio; CI = confidence interval; CRT = catheter-related thrombosis; BMI = body mass index; KPS = Karnowski performance score; ALB = albumin; Hb = hemoglobin; PLT = platelets; WBC = white blood cells; CVC = central venous catheter; FICC = femoral central catheter; PICC = peripherally inserted central venous catheter; PORT = implanted venous port; Anti.HER.mAb = human epidermal growth factor receptor monoclonal antibody; Anti.CD20.mAb = anti-CD20 monoclonal antibody;

[0039] Figure 11 This refers to the computational performance of the trained catheter-related thrombosis (CRT) survival model provided in this embodiment of the invention. Figure 11a) External ten-fold nested cross-validation (left bar chart) and the consistency index (c-index) of the model based on three external test datasets (right heatmap). The data on the left is represented as a box plot, where the middle line is the median, the left and right hinges correspond to the first and third quartiles, respectively, and the whiskers correspond to the minimum or maximum hinges not exceeding 1.5 × IQR (where IQR is the interquartile range). The median is also color-coded. Any data exceeding the whiskers is considered an outlier. The data on the right is represented as a heatmap, where the c-index is color-coded and labeled. Figure 11 b, the time-dependent c-index of SM-CRT based on three external test datasets. Advantage = prospective test queue; Ext1 = external test 1; Ext2 = external test 2;

[0040] Figure 12 The risk period stratification based on Distr prediction provided in this embodiment of the invention demonstrates the ability of SM-CRT to optimize extubation time decisions. Figure 12 a. Schematic diagram of patient risk stratification based on distribution prediction. Patient characteristics are input into the model to predict the transient CRT probability over time. Then, a smooth density estimate is performed on the distribution with a Gaussian kernel and a bandwidth of 3. Periods with a transient probability greater than the maximum 50% are considered high-risk periods. The first high-risk day (t1), the day with the highest risk, and the first day after the high-risk period (t2) are marked with red dots. The actual number of days the catheter was placed is marked with a solid colored line. If the patient's actual extubation date occurs before t1, the patient is assigned to the low-risk group; if extubation occurs on or after t1 but before t2, the patient is assigned to the high-risk group; if extubation occurs at or after t2, the patient is assigned to the long-term group. Figure 12 b. Compare the number of daily CRT events among different SMCRT distribution-based prediction groups using Fisher's exact test based on the training and test datasets. Data are represented as histograms. The p-values ​​for multiple comparisons between groups were adjusted using the Benjamini-Hochberg (BH) procedure to control for the false discovery rate (FDR). The groups are color-coded. Figure 12 c. The number of daily CRT events was compared between different SM-CRT risks and groups based on distr predictions, using Fisher's exact test based on the training and test datasets. The data is represented as a histogram. The groups are color-coded.

[0041] Figure 13 This is a model interpretation of SM-CRT provided in this embodiment of the invention. The directional marker region between the highest and lowest time survival function curves in the sensitivity analysis is represented by a histogram. Larger numbers indicate higher importance in predicting CRT. These functions are arranged by category and color-coded;

[0042] Figure 14 This is a functional diagram of the SM-CRT display when in use, provided in an embodiment of the present invention. Figure 14 a) Display a histogram of catheter type selection by comparing risk predictions for different catheter types. Figure 14 b. The Kaplan-Meier curve shows the risk identification based on SM-CRT risk stratification and risk prediction. Figure 14 c. The probability density function (PDF) determined by SM-CRT guides catheter timing by providing a high CRT risk period (between t1 and t2). Figure 14 d. The cumulative distribution function (CDF) determined by SM-CRT is used to optimize the ultrasound follow-up period, and the recommended number of subsequent days is marked with a red dot. Detailed Implementation

[0043] To enable those skilled in the art to better understand the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings.

[0044] In some of the processes described in the specification, claims, and accompanying drawings of this invention, multiple operations appearing in a specific order are included. However, it should be clearly understood that these operations may not be executed in the order they appear herein, or may be executed in parallel. The operation numbers, such as 101, 102, etc., are merely used to distinguish different operations and do not represent any execution order. Furthermore, these processes may include more or fewer operations, and these operations may be executed sequentially or in parallel. It should be noted that the descriptions such as "first," "second," etc., in this document are used to distinguish different messages, devices, modules, etc., and do not represent a sequential order, nor do they limit "first" and "second" to different types.

[0045] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0046] Figure 1 This is a schematic flowchart of a method for constructing a survival model of catheter-related thrombosis based on machine learning, provided by an embodiment of the present invention. Specifically, the method includes the following steps:

[0047] S101, Obtain training set samples including basic sample information, catheter type, systemic drug treatment, tumor site, and any one or more data and corresponding result information of the samples;

[0048] In some more specific embodiments, the outcome information includes: CRT event, event time, and catheter insertion duration; wherein, a CRT event is defined as venous thrombosis along the catheter; the time from catheter insertion to the occurrence of CRT is considered as CRT-free survival (CFS), and extubation without CRT and catheter placement without CRT until the last ultrasound follow-up are considered as review time points;

[0049] In some more specific embodiments, the basic information of the sample includes any one or more of the following: age, sex, body mass index, KPS, smoking status, drinking status, and past or comorbid diseases;

[0050] In some more specific embodiments, the pre-existing or comorbid diseases include any one or more of the following: hypertension, diabetes, coronary heart disease, cerebral infarction, hyperlipidemia, and a history of venous thrombosis;

[0051] In some more specific embodiments, the catheter type includes any one or more of the following: FICC, PICC, CVC, PORT;

[0052] In some more specific embodiments, the systemic drug treatment is recorded during CRT or extubation and mainly includes any one or more of the following: chemotherapy (platinum-based drugs, microtubule inhibitors, antimetabolites, antitumor antibiotics, nitrogen mustard, DNA topoisomerase inhibitors), targeted therapy (immunotherapy, anti-angiogenic therapy, AntiHER mAb, AntiCD20 mAb).

[0053] In some more specific embodiments, the tumor site includes any one or more of the following (rare tumor sites are grouped with the most recent site, and tumors are divided into 21 groups): lung / pleura, esophagus, mediastinum, stomach, intestine, pancreas, liver, bile duct, kidney / adrenal gland, tunica albuginea, rectum, bladder / ureter, uterus / vagina, prostate, ovary / fallopian tube, testis, breast, head / neck, trunk wall, limbs, and lymphoma;

[0054] Optionally, the data may also include experimental test data (blood markers) and tumor staging; the experimental test data (blood markers) may include any one or more of the following: albumin, hemoglobin, platelets, leukocytes, and D-dimer; the tumor staging may include I, II, III, and IV; optionally, the data may also include whether the catheter tip position is appropriate.

[0055] S102, the factor folding method is used to process the systemic drug treatment and tumor site data to obtain the processed systemic drug treatment and tumor site data features; basic information and catheter type data are converted into numerical features; various feature engineering methods are used to construct features, and the factor folding method is applied to reduce the number of sparse features; one-hot encoding is performed on the categorical variables as input to the model;

[0056] In some embodiments, the method further includes: calculating the importance of data features and numerical features in S102; the method for calculating the importance includes: using the "model_profile" function to evaluate the change of the survival function value derived by the model as the feature value changes to perform sensitivity analysis; calculating the area and direction between the highest and lowest time survival function curves, and determining the importance of feature prediction CRT based on the area and direction;

[0057] S103, the data features, numerical features and result information are input into the machine learning model for training to obtain the SM-CRT model. The SM-CRT model can predict the risk score crank and the survival distribution distr prediction type.

[0058] In some embodiments, the machine learning model uses one or more of the following methods: rfsrc, ranger, cforest, coxph, xgboost, penalized, coxboost; preferably coxph.

[0059] Figure 2 This is a schematic flowchart of a method for identifying CRT risk based on the SM-CRT model provided by an embodiment of the present invention. Specifically, the method includes the following steps:

[0060] 201. Obtain data on the subjects, including basic information, type of catheter used, systemic drug treatment, and tumor location;

[0061] 202, Input the data into the SM-CRT model described in the first aspect of this application to calculate the risk value;

[0062] 203. Output auxiliary prediction results based on the risk value to indicate whether the subject has a high or low risk of developing CRT. Specifically, when the risk value is greater than the first threshold, output auxiliary prediction results indicating a high risk of developing CRT; when the risk value is less than the first threshold, output auxiliary prediction results indicating a low risk of developing CRT.

[0063] Optionally, the method for determining the first threshold includes, but is not limited to, selecting the median of the risk ranking score predicted by the CRT model for each dataset; or customizing the cutoff value for specific practical application scenarios.

[0064] Figure 3This is a schematic flowchart of a method for guiding catheter type selection based on the SM-CRT model according to an embodiment of the present invention. Specifically, in this application scenario, systemic therapy is not comparable; systemic therapy refers to chemotherapy and targeted therapy for tumors. The selection of systemic therapy is primarily based on efficacy, rather than thrombosis risk. The method includes the following steps:

[0065] 301. Obtain basic information about the subject, systemic drug treatment, tumor site data, and the type of catheter to be used;

[0066] 302. Input the basic information, systemic drug treatment, tumor site data, and expected catheter type into the SM-CRT model described in the first aspect of this application, and output the predicted risk value of CRT occurring with different catheter types; determine the final catheter type to be used based on the predicted risk value; experimental verification: FICC and PICC have a higher risk of CRT than CVC and PORT.

[0067] Figure 4 This is a schematic flowchart of a method for guiding extubation time based on an SM-CRT model according to an embodiment of the present invention. Specifically, the method includes the following steps:

[0068] 401. Obtain data on the subject, including basic information, type of catheter used, systemic drug treatment, and tumor location;

[0069] 402. Input the data into the SM-CRT model described in the first aspect of this application to predict the high CRT risk period, and predict the extubation time of the subject based on the high CRT risk period; T1-t2 is the high thrombosis risk period, that is, the catheter placement time should be as close to t1 as possible (put the catheter in before t1), or it should be placed for a long time (more than t2, and the catheter should be removed after t2).

[0070] Optionally, the extubation time can be a specific point in time or a time interval;

[0071] Optionally, the high CRT risk period is between the first high-risk day and the second high-risk day;

[0072] Optionally, the first and second highest-risk days are determined by calculating instantaneous probabilities. The determination method includes: calculating the time distribution of the probability density function PDF of the training set samples; calculating the kernel density estimate of the time distribution with a Gaussian kernel and a bandwidth of N; calculating the instantaneous probability over time based on the kernel density estimate, and calculating the highest instantaneous probability over the first M time periods (6 months); marking the number of days with an instantaneous probability greater than the first percentage (50%) of the highest instantaneous probability as a high-risk period; marking the first day of the high-risk period as the first high-risk day t1, and the first day after the high-risk period as the second high-risk day t2; calculating the PDF uses the default function in the R language.

[0073] Optionally, the high-risk period also includes the day with the highest risk, and the time corresponding to the highest risk is the day with the highest risk.

[0074] Furthermore, the method also includes: calculating the risk value of the subject based on the SM-CRT model, and predicting the extubation time of the subject based on the risk value and the high CRT risk period; specifically, combining risk ranking prediction and distribution prediction to further group patients, by selecting the median risk ranking value based on the dataset as the cutoff value, and dividing patients into six groups (high risk ranking + high risk, high risk ranking + long-term, high risk ranking + low risk, low risk ranking + high risk, low risk ranking + long-term, low risk ranking + low risk; low risk ranking + high risk and low risk ranking + low risk are similar, both lower than high risk ranking + high risk).

[0075] Figure 5 This is a schematic flowchart of a method for guiding ultrasound follow-up cycles based on an SM-CRT model, provided by an embodiment of the present invention. Specifically, the method includes the following steps:

[0076] 501, to obtain data on the subject including basic information, type of catheter used, systemic drug treatment, tumor location, and first-time data;

[0077] 502, Input the data into the SM-CRT model described in the first aspect of this application to calculate the cumulative distribution function (CDF) at different times;

[0078] 503. If the CDF of the second time point increases by a second threshold compared to the first time point, then the second time point is the predicted ultrasound follow-up time adjacent to the first time point; the ultrasound follow-up time point can be a specific time point; the first time point is the time of the last ultrasound examination.

[0079] In some embodiments, the terms “subject” or “test subject” or “sample to be tested” as used herein refer to any animal (e.g., a mammal), including but not limited to humans, non-human primates, rodents, etc., which will become the recipient of a particular treatment. Generally, the terms “subject” and “patient” are used interchangeably herein when referring to human subjects. Preferably, the subject is a human. In some embodiments, the sample to be tested is a patient clinically used for prognostic assessment.

[0080] In some embodiments, the auxiliary prediction results include, but are not limited to, paper or electronic reports. These results are obtained by intelligent machines based on the relevant data of the subjects and are intended only as a reference for medical personnel, not as the final diagnosis results of the subjects.

[0081] In some embodiments, the first threshold or the second threshold is obtained through training on training set samples or based on prior knowledge. It can be a specific threshold or a range, and its specific form is not specifically limited in this embodiment. The letters N, M, etc., appearing in this embodiment are all rational numbers.

[0082] Figure 6 This is a schematic diagram of a computer device provided in an embodiment of the present invention, such as... Figure 6 As shown, the device 2000 may include: one or more processors 2010 and one or more memories 2020; wherein the memories store computer-readable code that, when run by the one or more processors, can perform the methods described above.

[0083] The processor in this embodiment can be an integrated circuit chip with signal processing capabilities. The processor can be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), an off-the-shelf programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the methods, operations, and logic block diagrams disclosed in this embodiment. The general-purpose processor can be a microprocessor or any conventional processor, and can be based on an x86 or ARM architecture.

[0084] In general, the various exemplary embodiments of this disclosure can be implemented in hardware or dedicated circuitry, software, firmware, logic, or any combination thereof. Some aspects can be implemented in hardware, while others can be implemented in firmware or software that can be executed by a controller, microprocessor, or other computing device. When aspects of embodiments of this disclosure are illustrated or described as block diagrams, flowcharts, or using some other graphical representation, it will be understood that the blocks, apparatuses, systems, techniques, or methods described herein can be implemented as non-limiting examples in hardware, software, firmware, dedicated circuitry or logic, general-purpose hardware or controllers or other computing devices, or some combination thereof.

[0085] For example, the method or apparatus according to embodiments of this disclosure can also be used by means of Figure 7 The architecture of the computing device 3000 shown is used for implementation. For example... Figure 7As shown, the computing device 3000 may include a bus 3010, one or more CPUs 3020, a read-only memory (ROM) 3030, a random access memory (RAM) 3040, a communication port 3050 connected to a network, an input / output component 3060, a hard disk 3070, etc. The storage devices in the computing device 3000, such as the ROM 3030 or the hard disk 3070, may store various data or files used for processing and / or communication of the methods provided in this disclosure, as well as program instructions executed by the CPU. The computing device 3000 may also include a user interface 3080. Of course, Figure 7 The architecture shown is merely exemplary and can be omitted as needed when implementing different devices. Figure 7 One or more components in the computing device shown.

[0086] This invention also includes a computer-readable storage medium, such as... Figure 8 The diagram illustrates a storage medium 4000 provided in an embodiment of the present invention. The computer storage medium 4020 stores computer-readable instructions 4010. When the computer-readable instructions 4010 are executed by a processor, the method described above according to embodiments of the present disclosure can be performed. The computer-readable storage medium in the embodiments of the present disclosure may be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. Non-volatile memory may be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. Volatile memory may be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of RAM are available, such as Static Random Access Memory (SRAM), Dynamic Random Access Memory (DRAM), Synchronous Dynamic Random Access Memory (SDRAM), Double Data Rate Synchronous Dynamic Random Access Memory (DDRSDRAM), Enhanced Synchronous Dynamic Random Access Memory (ESDRAM), Synchronous Link Dynamic Random Access Memory (SLDRAM), and Direct Memory Bus Random Access Memory (DR RAM). It should be noted that the memory used in the methods described herein is intended to include, but is not limited to, these and any other suitable types of memory.

[0087] This disclosure also provides a computer program product or system, including a computer program that, when executed by a processor, implements the steps of the above-described method.

[0088] In some embodiments, this embodiment also discloses a system for constructing an SM-CRT model based on machine learning, the system comprising:

[0089] The first acquisition module is used or configured to acquire training set samples including basic sample information, catheter type, systemic drug treatment, tumor site, or any one or more of the data and corresponding result information of the samples.

[0090] The preprocessing module is used or configured to process systemic drug treatment and tumor site data using a factor folding method to obtain processed systemic drug treatment and tumor site data features; and to convert basic information and catheter type data into numerical features.

[0091] The model training module is used or configured to input the data features, numerical features and result information into the machine learning model for training to obtain the SM-CRT model.

[0092] In some embodiments, this embodiment also discloses a system for identifying the risk of CRT occurrence based on the SM-CRT model, the system comprising:

[0093] The second acquisition module is used or configured to acquire data on the subject, including basic information, type of catheter used, systemic drug treatment, and tumor location.

[0094] A risk value first calculation module is used or configured to input the data into the SM-CRT model described in the first aspect of this application to calculate the risk value;

[0095] The CRT risk prediction module is used or configured to output auxiliary prediction results based on the risk value, indicating whether the subject has a high or low risk of developing CRT. Specifically, when the risk value is greater than a first threshold, it outputs an auxiliary prediction result indicating a high risk of developing CRT, and when the risk value is less than the first threshold, it outputs an auxiliary prediction result indicating a low risk of developing CRT.

[0096] In some embodiments, this embodiment also discloses a system for guiding catheter type selection based on the SM-CRT model, the system comprising:

[0097] The third acquisition module is used or configured to acquire basic information about the subject, systemic drug treatment, tumor site data, and the type of catheter to be used.

[0098] The catheter type selection module is used or configured to input the basic information, systemic drug treatment, tumor site data, and the expected catheter type into the SM-CRT model described in the first aspect of this application, and output a predicted risk value of CRT occurring using different catheter types; and determine the final catheter type to be used based on the predicted risk value.

[0099] In some embodiments, this embodiment also discloses a system for guiding extubation time based on an SM-CRT model, the system comprising:

[0100] The fourth acquisition module is used or configured to acquire data on the subject, including basic information, type of catheter used, systemic drug treatment, and tumor location.

[0101] The extubation time prediction module is used or configured to input the data into the SM-CRT model described in the first aspect of this application to predict the high CRT risk period, and predict the extubation time of the subject based on the high CRT risk period.

[0102] In some embodiments, this embodiment also discloses a system for guiding ultrasound follow-up cycles based on an SM-CRT model, the system comprising:

[0103] The fifth acquisition module is used or configured to acquire data on the subject, including basic information, type of catheter used, systemic drug treatment, tumor location, and first-time data.

[0104] The CDF calculation module is used or configured to input the data into the SM-CRT model described in the first aspect of this application to calculate the cumulative distribution function (CDF) at different times;

[0105] An ultrasound follow-up prediction module is used or configured to be the predicted ultrasound follow-up time adjacent to the first time if the CDF of the second time is increased by a second threshold compared to the first time. Specific Implementation

[0106] method:

[0107] Study Design and Population: Cancer patients who underwent catheterization, met inclusion criteria, and received treatment at four centers were recruited. Data were prospectively collected and retrospectively analyzed. The study comprised one training group and three test groups. Patients recruited from Center 1 and Center 2 were pooled for multicenter training and testing. Specifically, patients whose data were collected before June 30, 2023, were included in the training cohort, and those whose data were collected after July 1, 2023, were included in the prospective test cohort. Patients recruited from Center 3 and Center 4 were included in a separate external test set.

[0108] Inclusion and exclusion criteria: Eligible patients were pathologically diagnosed with malignancy; successfully underwent catheter insertion; voluntarily participated in this study and reported their data; were ≥18 years old; and underwent ultrasound-guided extubation or, if not, at least one vascular ultrasound examination was performed after catheter insertion. Key exclusion criteria included incomplete patient information, unknown primary tumor location, and lack of follow-up (excluding patients who did not record ultrasound data or attend follow-up appointments). Notably, if a patient had multiple catheters, the management of each case differed in our study.

[0109] Catheter insertion: The modified Seldinger technique under ultrasound guidance was used for the placement of central venous catheters (CVCs), femoral artery central catheters (FICCs), peripherally inserted central catheters (PICCs), and port-of-care (PORTs). All CVCs and PORTs were non-tunneled catheters, inserted via the contralateral subclavian or jugular vein (without surrounding tumors such as breast cancer). All PICCs were inserted into the superior vena cava via the contralateral basilar vein, and all FICCs were inserted into the superior vena cava via the contralateral basilar vein. The inferior vena cava was inserted based on the CVC and through the femoral vein on the uninjured side. FICCs are primarily used in patients with superior vena cava syndrome (SVCS)46 which can cause edema of the head, neck, and upper extremities, leading to hemodynamic stagnation and thrombosis47. CVCs, FICCs, and PORTs are made of polyurethane, while PICCs are made of silicone rubber. The length of CVCs, PORTs, and FICCs in the vein is approximately 18 cm, and the length of PICCs in the vein is approximately 40 cm. The outer diameter (OD) of the catheters is as follows: 4Fr for centers 1 and 2; 4Fr for the PICC at center 3; 6Fr for the PORT at center 3; and 6Fr for center 4. Following non-femoral artery catheter placement, a chest X-ray (including the ipsilateral upper limb and neck) is performed to confirm catheter orientation and tip position. All X-rays are evaluated by a physician specializing in venous catheter insertion, and at least one radiologist is responsible for the chest X-ray report. For non-femoral artery catheters, the catheter tip is typically located in the lower third of the superior vena cava, at the atrioventricular junction, or in the upper third of the right atrium. T6-T8 (6th-8th thoracic vertebrae) is considered an appropriate position when referenced to the vertebrae; otherwise, this position is considered inappropriate. For FICCs, the catheter tip is placed near the origin of the inferior vena cava of the common iliac vein, and inappropriate placement is not considered.

[0110] Data Collection: General patient information, past or comorbid diseases, tumor status, catheter-related information, and baseline laboratory data were recorded during catheter insertion. During CRT occurrence or extubation, medications used during catheter insertion and outcome information were recorded, including the CRT event, event time, and duration of catheter insertion. A CRT event was defined as venous thrombosis along the catheter. Variable Details:

[0111] The baseline variables recorded at the time of initial catheter insertion included 5 categories and 47 features, including the following: 1) General information: age, sex, body mass index (BMI), Karnowski score (KPS), smoking status, and alcohol consumption. Past or comorbid conditions: hypertension, diabetes, coronary artery disease, cerebral infarction, hyperlipidemia, and history of venous thrombosis. 2) Tumor status: tumor type and stage (I, II, III, and IV). For tumor type, tumors were grouped by location. In particular, rare tumor sites were grouped with the most recent site. Tumors were divided into 21 groups: lung / pleura, esophagus, mediastinum, stomach, intestine, pancreas, liver, bile duct, kidney / adrenal gland, peritoneum, rectum, bladder / ureter, uterus / vagina, prostate, ovary / fallopian tube, testis, breast, head / neck, trunk wall, extremities, and lymphoma. 3) Catheter-related information: catheter type (PICC, CVC, PORT, or FICC) and improper catheter tip placement. 4) Baseline laboratory data: albumin (ALB), hemoglobin (Hb), platelets (PLT), white blood cells (WBC), and D-dimer.

[0112] A class of 11 characteristics and outcome information is recorded during CRT occurrence or catheter removal. These variables include the following: 1) Medications used during catheter insertion: chemotherapy (platinum-based, microtubule inhibitors, antimetabolites, antitumor antibiotics, nitrogen mustard, DNA topoisomerase inhibitors), targeted therapy (immunotherapy, anti-angiogenic therapy, Anti-HER mAb, Anti-CD20 mAb), and nutrition obtained from medical records, whether or not via catheter. 2) Outcome information: CRT event, event time, and duration of catheter insertion. A CRT event is defined as venous thrombosis along the catheter.

[0113] Multiple feature engineering methods were used to prune features. Factor folding was applied to tumor type and chemotherapy type to reduce the number of sparse features. For example, "lung" and "pleura" were combined into "lung / pleura". One-hot encoding was performed on categorical variables as input to the model. Due to the large sample size and relatively small number of features in our dataset (58 in total), no collinearity removal or variable selection was performed to avoid potential loss in model performance.

[0114] CRT assessment: Patients are continuously monitored until catheter removal is performed as directed by the physician. Diagnosis of CRT is made using vascular Doppler ultrasound and color imaging (GELOGIQ™ 9). Ultrasound examinations are performed every 3 months after catheter insertion, during removal, and if any clinical symptoms suggestive of CRT are observed. The occurrence of CRT is considered the target event in survival analysis, and the time from catheter insertion to the occurrence of CRT is considered CRT-free survival (CFS). Removal without CRT and catheter placement without CRT up to the last ultrasound follow-up are considered review time points.

[0115] Survival Model Construction and Validation: The model was trained using the training set. To select a model with appropriate hyperparameters and evaluate its performance, a nested 10x cross-validation strategy was applied. The inner 10x cross-validation was used for hyperparameter tuning, and the outer 10x cross-validation was used for internal validation. There are 23 basic models and 21 ensemble models. The relationship between the 23 basic models and the 21 ensemble models is as follows: the best 7 basic models are combined in pairs (7*6 / 2=21). Additionally, the best 2, 3, 4, 5, 6, and 7 basic models are also combined (a total of 6). (Detailed information: Twenty-three basic models, including 'surv.akritas', 'surv.blackboost', 'surv.cforest', 'surv.coxboost', 'surv.coxph', 'surv.coxtime', 'surv.ctree', 'surv.cv_coxboost', 'surv.cv_glmnet', 'surv.deephit', 'surv.gbm', 'surv.glmboost', 'surv.glmnet') The models 'surv.kaplan', 'surv.loghaz', 'surv.nelson', 'surv.pchazard', 'surv.penalized', 'surv.ranger', 'surv.rfsrc', and 'surv.rpart' were developed, along with 'surv.xgboost.aft' and 'surv.xgboost.cox'. Additionally, six ensemble models based on the top 2 to 7 best-performing models (in external cross-validation) were developed, named surv.ens_topN (e.g., surv.ens_top2), and 21 ensemble models based on all models were developed. Possible pairwise combinations between the top 7 best-performing models (in external cross-validation) are named model1_ens_model2 (e.g., ranger_ens_coxph). These models are trained to predict risk scores (crank) and survival distributions (distr) prediction types. Crank is a risk score used to assess general risk over time and calculate the c-index. distra is a predicted survival distribution used to guide catheter and ultrasound follow-up times. The default weighted discrete distribution from the mlr3probaR package is applied to this model. The consistency index (c-index) is used as an evaluation metric to compare the performance of different models in outer 10-fold cross-validation. The c-index ranges from 0 to 1, with a value of 0.5 indicating no discriminant power (equivalent to random chance) and a value of 1.0 indicating perfect discriminant power. Brier scores are used for calibration comparisons. Brier scores range from 0 to 1, with lower values ​​indicating higher predictive accuracy.The model is then trained using all the training data and tested using the test set. Model Interpretation: While the Cox model can be easily interpreted based on feature coefficients, the coefficients may not accurately reflect feature importance and which features contribute to the model's predictive outcomes due to scale differences between features (e.g., platelet count is a continuous variable from 0 to 1000, while sex is a binary variable with a value of 0 or 1). Therefore, the importance of each feature is determined by calculating the area between the highest and lowest time-survival function curves in the sensitivity analysis, taking into account the direction of importance. The "survex" package 48 in R is used for survival model interpretation. The "model_profile" function evaluates how the model-derived survival function values ​​change with feature values ​​for sensitivity analysis. The area between the highest and lowest time-survival function curves is then calculated to assess the importance of each feature. If the highest feature value correlates with the lowest curve, the feature is considered to have a risk effect (the area is marked as positive); otherwise, it is considered to have a protective effect (the area is marked as negative). Larger numbers indicate higher importance in predicting CRT.

[0116] Statistical analysis: Kaplan-Meier survival curves are generated to estimate the temporal distribution of events. The "muhaz" package and "muhaz" function are used to smooth the estimation of the danger function of right-censored data using global and local bandwidth selection algorithms and boundary kernel functions; other parameters are set to default values.

[0117] Univariate Cox proportional hazards regression analysis was used to assess the association between the variables and CFS. The Wald test was used to assess the significance of each variable in the Cox regression model. The hazard ratio (HR) and 95% confidence interval (CI) were calculated for each variable. An HR greater than 1 indicated an increased risk of CRT, while an HR less than 1 indicated a protective effect. The log2 HR was compared between different tumor sites within different cavities using the Wilcoxon rank test. Specifically, patients with prostate cancer did not experience CRT events, and HR could not be calculated; therefore, we designated -2.5 as the log2 HR value for prostate cancer in this analysis. P-values ​​for multiple comparisons between different cavities were adjusted using the Benjamini-Hochberg (BH) procedure to control for false discovery rate (FDR).

[0118] The c-index derived from 10-fold cross-validation of different models was compared using the Wilcoxon rank test. The median SM-CRT predicted hazard value for each dataset was chosen as the cutoff for classifying patients into high-risk or low-risk groups to ensure that each group contained approximately equal numbers. Event occurrence time data were analyzed using the Kaplan-Meier method to estimate survival probabilities, and inter-group comparisons were performed using the log-rank test. A Cox proportional hazards regression model estimating hazard ratios (HRs) was employed to quantify the impact of risk-based stratification on CRT outcomes.

[0119] Fisher's exact test was used to compare differences between risk-stratified groups, and p-values ​​were adjusted using the BH procedure. To address potential confounding issues caused by imbalances in inter-group variables during risk-stratification, propensity score matching (PSM) was employed. PSM was performed by estimating propensity scores using the model's risk predictions. Nearest neighbor matching (1:1 ratio) with a caliper ratio of 0.01 was used, without substitution, to minimize imbalance. All tests were two-sided. Unless otherwise stated, p-values ​​< 0.05 were considered statistically significant. All statistical analyses were performed using R statistical software version 4.3.

[0120] result:

[0121] Patient characteristics and univariate analysis: From January 1, 2017 to June 30, 2023, we collected data from 25,899 patients in the training cohort. A total of 1,031 patients (4%) received CRT treatment. Based on Kaplan-Meier survival curves (…),… Figure 9 (a) Over time, the probability of non-CRT survival decreased from 1 to 0.786. The survival curves show that the risk increases sharply before day 11 after catheter insertion, then drops to near 0 two years after catheter insertion. Figure 9 (b) highlights the time-dependent risk distribution of CRT events.

[0122] A univariate proportional hazards regression model was used to examine potential risk factors associated with CRT. Figure 10 It is worth noting that the mediastinum and esophagus are closer to the superior vena cava than other sites and have the highest heart rate (HR) values. Figure 10This led us to speculate whether the spatial distance from the tumor to the superior vena cava was related to CRT. We found that HR values ​​were higher for thoracic tumors compared to abdominal tumors (median 2.539 vs. 0.912, p=0.017) and pelvic tumors (median 2.539 vs. 0.46, p=0.024). Furthermore, abdominal tumors had a higher HR value compared to pelvic tumors (median 0.912 vs. 0.46, p=0.0023) (the testes are considered to be located in the pelvis). These data suggest that tumor location and the distance between the tumor and the superior vena cava containing the duct are important predictors of CRT.

[0123] Survival models can accurately predict CRT events: A survival model integrating the above variables is constructed to predict CRT events (SM-CRT). Results show that surv.rfsrc, surv.ranger, surv.cforest, surv.coxph, surv.xgboost.aft, surv.penalized, and surv.coxboost have the highest c-index values ​​in the base model with ten times the outer nesting. Cross-validation (...) Figure 11 a) These high-performance base models were selected for ensemble learning. The final model was then trained using data from all patients in the training cohort and tested on three test datasets. The surv.coxph model had the highest c-index among the base models, while xgboost.aft_ens_penalized had the highest c-index among the ensemble models with prospective test datasets; however, xgboost.aft_ens_penalized was only slightly higher than surv.coxph (0.702 vs. 0.697). Considering surv.coxph's lower complexity and higher interpretability, it was chosen as the final SM-CRT model, demonstrating good performance with c-indexes of 0.703 and 0.7 on the two external test datasets, respectively. Brier scores were used for calibration comparisons, showing that surv.coxph had lower Brier scores across all test datasets. Overall, SM-CRT exhibited high performance in CRT prediction. This makes it possible to select catheter type for a given patient by comparing risk predictions for alternative types.

[0124] To further evaluate the performance of the surv.coxph model over time, we calculated the time-dependent c-exponent for each model. The results show that SM-CRT can achieve a generally high c-exponent in the early stages of catheter insertion, even reaching 0.8 in all three test datasets. Figure 11 b). Then, the c-index gradually decreases over time until it reaches the overall c-index level ( Figure 11(b) This indicates that prediction becomes more difficult over time. The median SM-CRT predicted risk score for each dataset was then chosen as the cutoff for classifying patients into high-risk or low-risk groups. In the prospective test (HR=1.86, p=0.0058) and external test 1 (HR=31.39, p<0.001), patients with higher SM-CRT risk scores were significantly at higher risk of receiving CRT than those with lower SM-CRT risk scores. Furthermore, the HR was slightly higher in the small sample size external test 2 (HR=1.94, p=0.051). These results suggest that SMCRT can accurately identify high-risk and low-risk CRT cases.

[0125] CRT distribution prediction can guide extubation time: The risk prediction of the SM-CRT model can be used to predict the general risk of patients and guide the selection of catheter type. However, the risk prediction results cannot be used to determine the patient's extubation time, which is another actionable factor and the most effective method for preventing CRT. Here, we explore whether the SM-CRT model can be used to guide the patient's extubation time based on the "distr" prediction, which provides information on the probability time distribution of CRT. Patients are stratified by risk period (low risk, high risk, and long-term) according to the SM-CRT distr prediction. Figure 12 a). We first made comparisons within each dataset. The results showed that the low-risk group (0.09% vs. 0.19%, padj < 0.001) and the long-term group (0.09% vs. 0.19%, padj < 0.001) had significantly lower daily CRT events than the high-risk group based on the training set. Similar results were obtained using the prospective test dataset (0.10% vs. 0.21%, padj = 0.107; 0.04% vs. 0.21%, padj < 0.001). However, the cumulative number of days in the low-risk group was extremely low (less than 1,000 days), and no CRT events were recorded in the two external test datasets. Therefore, we combined all three test datasets for comparison. The results showed that, based on the test dataset, the number of daily CRT events in the low-risk and long-term groups was significantly lower than that in the high-risk group (0.08% vs. 0.22%, padj = 0.0254; 0.09% vs. 0.22%, padj < 0.001). Figure 12 (b) Due to potential biases introduced by patient variables, we subsequently performed propensity score matching (PSM) to balance patient variables based on risk scores predicted by SM-CRT. Results also showed that both the low-risk and long-term groups had significantly fewer daily CRT events than the high-risk group. These results suggest that SM-CRT can be used to effectively identify high-risk CRT cycles, thereby optimizing catheter removal time in clinical practice.

[0126] Some patients may have a low-risk prediction, in which case the risk of CRT is not high even during the high-risk period defined by the Distr prediction; therefore, we further grouped patients by combining risk and distribution predictions. By selecting the median risk value based on all datasets as the cutoff, we divided patients into six groups according to different combinations of risk scores (high-risk vs. low-risk) and distribution periods (high-risk, low-risk, and long-term). The results confirmed that, in both training and testing datasets, the number of daily CRT events during the high-risk period was similar for patients with low-risk scores to those with high-risk scores during the low-risk period (0.08% vs. 0.11%, p=0.159) and for the testing dataset (0.15% vs. 0.14%, p=1). Figure 12 c). Furthermore, both were significantly lower than the high-risk period for patients with high risk scores in both the training and test datasets. Figure 12 c). Therefore, SM-CRT can provide more detailed risk stratification for complex clinical practices by combining risk and distribution predictions.

[0127] The model interpretation system reveals key factors related to CRT: our goal is to interpret SM-CRT based on all test datasets. Most results are similar to those of univariate analysis. Figure 13 For different tumor sites, the order of importance for predicting CRT is as follows: mediastinum, esophagus, bile duct, liver, breast, lung / pleura, pancreas, head / neck, kidney / adrenal gland, lymphoma, uterus / vagina, stomach, peritoneum, limbs, intestines, rectum, testis, bladder / ureter, ovary / fallopian tube, trunk wall, and prostate. Figure 13 Thoracic tumors were also shown to pose a higher risk than abdominal tumors, although this difference was not significant (median area 111.14 vs. 44.07, p=0.067), while pelvic tumors (median area 111.14 vs. -16.56, p=0.024) had a larger area. Furthermore, abdominal tumors had a larger area than pelvic tumors (median 44.07 vs. -16.56, p=0.036) (the testes are considered to be located in the pelvis). Regarding chemotherapy, the order of importance is as follows: nitrogen mustard, platinum, antimetabolites, DNA topoisomerase inhibitors, microtubule inhibitors, and antitumor antibiotics. Figure 13 Tumor staging and blood markers are not very important for CRT prediction. Figure 13 Overall, catheter type, tumor location, type of systemic therapy, and general patient characteristics are important factors associated with CRT.

[0128] Application of SM-CRT in the web-based Shiny application: The results of SM-CRT application are shown below: Function 1 guides the selection of catheter type by comparing the risk predictions of optional catheter types in SM-CRT ( Figure 14a). Please note that systemic treatments in this tool are not set to be comparable because clinical practice primarily focuses on anti-tumor efficacy rather than CRT prophylaxis. Function two is to identify high-risk patients based on SM-CRT risk prediction. Figure 14 b). The cutoff value is set to the median risk in the training dataset; however, for specific real-world applications, a customized cutoff value is required. Function 3 guides catheterization timing by providing a high-CRT risk period based on the PDF determined by the SM-CRT distr prediction. Figure 14 c). In clinical practice, it is recommended to combine risk and distribution prediction for more customized risk and time classification to guide catheter insertion time and antithrombotic drug administration. Function 4 is to optimize ultrasound follow-up cycles based on the cumulative distribution function (CDF) determined by the distr prediction of SM-CRT. Figure 14 d). If the CDF increases the cumulative CRT probability by 5% compared to the previous ultrasound examination (further adjustments may be needed based on the specific clinical situation), a new ultrasound examination is recommended.

[0129] The exemplary embodiments of this disclosure described in detail above are merely illustrative and not restrictive. Those skilled in the art will understand that various modifications and combinations can be made to these embodiments or their features without departing from the principles and spirit of this disclosure, and such modifications should fall within the scope of this disclosure.

Claims

1. A method for guiding extubation time based on a survival model of catheter-related thrombosis, characterized in that, The method includes:

401. Obtain data on the subject, including basic information, type of catheter used, systemic drug treatment, and tumor location; 402, Input the data into the survival model of catheter-related thrombosis to predict the high CRT risk period, and predict the extubation time of the subject based on the high CRT risk period; The specific method for constructing the survival model of catheter-related thrombosis is as follows: S101, Obtain training set samples including basic sample information, catheter type, systemic drug treatment, tumor site, and any one or more data and corresponding result information of the samples; S102, The factor folding method is used to process the data on systemic drug therapy and tumor site to obtain the processed data features of systemic drug therapy and tumor site; the basic information and catheter type data are converted into numerical features; S103, The data features, numerical features and result information are input into the machine learning model for training to obtain a survival model of catheter-related thrombosis; The catheter type includes any one or more of the following: FICC, PICC, CVC, PORT; The machine learning model uses one or more of the following methods: rfsrc, ranger, cforest, coxph, xgboost, penalized, and coxboost. The extubation time refers to a specific point in time or a time interval. The high CRT risk period is between the first and second highest risk days; The first and second highest-risk days are determined by calculating instantaneous probabilities. The determination method includes: calculating the time distribution of the probability density function PDF of the training set samples; calculating the kernel density estimate of the time distribution with a Gaussian kernel and a bandwidth of N; and calculating the instantaneous probability changing over time based on the kernel density estimate. The high-risk period also includes the day of highest risk, and the time corresponding to the highest risk is the day of highest risk; The method further includes: calculating the risk value of the subject based on the SM-CRT model, and predicting the extubation time of the subject based on the risk value and the high CRT risk period.

2. The method for guiding extubation time based on a survival model of catheter-related thrombosis according to claim 1, characterized in that, The machine learning model is Coxph.

3. The method for guiding extubation time based on a survival model of catheter-related thrombosis according to claim 1, characterized in that, The method further includes: calculating the importance of data features and numerical features in S102.

4. The method for guiding extubation time based on a survival model of catheter-related thrombosis according to claim 3, characterized in that, The method for calculating the importance includes: using the "model_profile" function to evaluate the sensitivity analysis of the changes in the survival function value derived from the model as the feature value changes; calculating the area and direction between the highest and lowest time survival function curves, and determining the importance of the feature prediction CRT based on the area and direction.

5. The method for guiding extubation time based on a survival model of catheter-related thrombosis according to claim 1, characterized in that, The results information includes: CRT event, event time, and catheter insertion duration; wherein, CRT event is defined as venous thrombosis along the catheter.

6. The method for guiding extubation time based on a survival model of catheter-related thrombosis according to claim 1, characterized in that, The basic information of the sample includes one or more of the following: age, gender, body mass index, kPS, smoking status, alcohol consumption status, and past or comorbid diseases.

7. The method for guiding extubation time based on a survival model of catheter-related thrombosis according to claim 6, characterized in that, The aforementioned pre-existing or comorbid conditions include any one or more of the following: hypertension, diabetes, coronary heart disease, cerebral infarction, hyperlipidemia, or a history of venous thrombosis.

8. The method for guiding extubation time based on a survival model of catheter-related thrombosis according to claim 1, characterized in that, The systemic drug treatment recorded during CRT or extubation mainly includes one or more of the following: chemotherapy, targeted therapy; The chemotherapy includes one or more of the following: platinum-based drugs, microtubule inhibitors, antimetabolites, antitumor antibiotics, nitrogen mustard, and DNA topoisomerase inhibitors; The targeted therapy includes one or more of the following: immunotherapy, anti-angiogenic therapy, AntiHER.mAb, and Anti.CD20.mAb.

9. The method for guiding extubation time based on a survival model of catheter-related thrombosis according to claim 1, characterized in that, The tumor sites include any one or more of the following, with rare tumor sites grouped with the most recent site, and the tumors are divided into 21 groups: lung / pleura, esophagus, mediastinum, stomach, intestine, pancreas, liver, bile duct, kidney / adrenal gland, tunica albuginea, rectum, bladder / ureter, uterus / vagina, prostate, ovary / fallopian tube, testis, breast, head / neck, trunk wall, limbs, and lymphoma.

10. The method for guiding extubation time based on a survival model of catheter-related thrombosis according to claim 1, characterized in that, The data also includes experimental test data and tumor staging.

11. The method for guiding extubation time based on a survival model of catheter-related thrombosis according to claim 10, characterized in that, The experimental data include any one or more of the following: albumin, hemoglobin, platelets, leukocytes, and D-dimer.

12. The method for guiding extubation time based on a survival model of catheter-related thrombosis according to claim 1, characterized in that, The tumor stages include I, II, III, and IV.

13. The method for guiding extubation time based on a survival model of catheter-related thrombosis according to claim 1, characterized in that, The data also includes whether the catheter tip position is appropriate.

14. A method for guiding ultrasound follow-up cycles based on a survival model of catheter-related thrombosis, characterized in that, The method includes: 501, to obtain data on the subject including basic information, type of catheter used, systemic drug treatment, tumor location, and first-time data; 502, Input the data into the survival model of catheter-related thrombosis to calculate the cumulative distribution function (CDF) at different times; 503. If the CDF of the second time point is increased by a second threshold compared to the first time point, then the second time point is the predicted ultrasound follow-up time adjacent to the first time point. The specific method for constructing the survival model of catheter-related thrombosis is as follows: S101, Obtain training set samples including basic sample information, catheter type, systemic drug treatment, tumor site, and any one or more data and corresponding result information of the samples; S102, The factor folding method is used to process the data on systemic drug therapy and tumor site to obtain the processed data features of systemic drug therapy and tumor site; the basic information and catheter type data are converted into numerical features; S103, The data features, numerical features and result information are input into the machine learning model for training to obtain a survival model of catheter-related thrombosis; The catheter type includes any one or more of the following: FICC, PICC, CVC, PORT; The machine learning model uses one or more of the following methods: rfsrc, ranger, cforest, coxph, xgboost, penalized, and coxboost; The second threshold is obtained by training with training set samples or based on prior knowledge.

15. The method for guiding ultrasound follow-up cycles based on a survival model of catheter-related thrombosis according to claim 14, characterized in that, The machine learning model is Coxph.

16. The method for guiding ultrasound follow-up cycles based on a survival model of catheter-related thrombosis according to claim 14, characterized in that, The method further includes: calculating the importance of data features and numerical features in S102.

17. The method for guiding ultrasound follow-up cycles based on a survival model of catheter-related thrombosis according to claim 16, characterized in that, The method for calculating the importance includes: using the "model_profile" function to evaluate the sensitivity analysis of the changes in the survival function value derived from the model as the feature value changes; calculating the area and direction between the highest and lowest time survival function curves, and determining the importance of the feature prediction CRT based on the area and direction.

18. The method for guiding ultrasound follow-up cycles based on a survival model of catheter-related thrombosis according to claim 14, characterized in that, The results information includes: CRT event, event time, and catheter insertion duration; wherein, CRT event is defined as venous thrombosis along the catheter.

19. The method for guiding ultrasound follow-up cycles based on a survival model of catheter-related thrombosis according to claim 14, characterized in that, The basic information of the sample includes one or more of the following: age, gender, body mass index, kPS, smoking status, alcohol consumption status, and past or comorbid diseases.

20. The method for guiding ultrasound follow-up cycles based on a survival model of catheter-related thrombosis according to claim 19, characterized in that, The aforementioned pre-existing or comorbid conditions include any one or more of the following: hypertension, diabetes, coronary heart disease, cerebral infarction, hyperlipidemia, or a history of venous thrombosis.

21. The method for guiding ultrasound follow-up cycles based on a survival model of catheter-related thrombosis according to claim 14, characterized in that, The systemic drug treatment recorded during CRT or extubation mainly includes one or more of the following: chemotherapy, targeted therapy; The chemotherapy includes one or more of the following: platinum-based drugs, microtubule inhibitors, antimetabolites, antitumor antibiotics, nitrogen mustard, and DNA topoisomerase inhibitors; The targeted therapy includes one or more of the following: immunotherapy, anti-angiogenic therapy, AntiHER.mAb, and Anti.CD20.mAb.

22. The method for guiding ultrasound follow-up cycles based on a survival model of catheter-related thrombosis according to claim 14, characterized in that, The tumor sites include any one or more of the following, with rare tumor sites grouped with the most recent site, and the tumors are divided into 21 groups: lung / pleura, esophagus, mediastinum, stomach, intestine, pancreas, liver, bile duct, kidney / adrenal gland, tunica albuginea, rectum, bladder / ureter, uterus / vagina, prostate, ovary / fallopian tube, testis, breast, head / neck, trunk wall, limbs, and lymphoma.

23. The method for guiding ultrasound follow-up cycles based on a survival model of catheter-related thrombosis according to claim 14, characterized in that, The data also includes experimental test data and tumor staging.

24. The method for guiding ultrasound follow-up cycles based on a survival model of catheter-related thrombosis according to claim 23, characterized in that, The experimental data include any one or more of the following: albumin, hemoglobin, platelets, leukocytes, and D-dimer.

25. The method for guiding ultrasound follow-up cycles based on a survival model of catheter-related thrombosis according to claim 14, characterized in that, The tumor stages include I, II, III, and IV.

26. The method for guiding ultrasound follow-up cycles based on a survival model of catheter-related thrombosis according to claim 14, characterized in that, The data also includes whether the catheter tip position is appropriate.

27. A computer device, characterized in that, The device includes: a memory and a processor; the memory is used to store a computer program; the processor executes the computer program to implement the method of any one of claims 1-26.

28. A computer-readable storage medium, characterized in that, It stores a computer program thereon, which, when executed by a processor, implements the method as described in any one of claims 1-26.

29. A computer program product, comprising a computer program, characterized in that, When executed by a processor, the computer program implements the method described in any one of claims 1-26.

Images