Children sirolimus administration dosage prediction method, device, medium and program product

Abstract

The embodiment of the invention provides a children sirolimus administration dosage prediction method, equipment, a medium and a program product, and relates to the field of intelligent medical treatment. The method comprises the following steps: acquiring prediction parameter information of a to-be-detected sample, wherein prediction parameters comprise weight, age, platelet count and creatinine; inputting the prediction parameter information into the dose prediction model, and calculating the predicted dose of sirolimus; the predicted dose is the daily average dose in the stable treatment period. The application develops an administration dosage prediction model aiming at three clinical problems of dosage error caused by development dynamics, high-risk group medication risk and passive decision mode in children sirolimus administration.

Description

Technical Field

[0001] This invention relates to the field of intelligent healthcare, and more specifically, to a method, device, medium, and procedure for predicting sirolimus dosage in children. Background Technology

[0002] Sirolimus, as a mammalian target of rapamycin (mTOR) inhibitor, plays an important role in the treatment of organ transplant rejection, autoimmune diseases, and vascular malformations in children. However, its narrow therapeutic window (target plasma concentration of 5-15 ng / mL) and significant individual variability, coupled with the unique pharmacokinetic characteristics of children (such as differences in liver enzyme activity and rapid changes in body composition), make precise dosage control a major clinical challenge. Currently, dosage regimens mainly rely on weight adjustment and empirical titration, with an initial dosing success rate of less than 40%, leading to repeated total drug concentration monitoring (TDM) and prolonged dose adjustment cycles, increasing treatment risks and medical burden.

[0003] Studies have confirmed that machine learning models can improve the dosage prediction accuracy of narrow therapeutic window drugs such as warfarin and tacrolimus by integrating demographic, laboratory, and genetic factors. However, research on predictive models for sirolimus in children remains lacking: existing models mostly focus on adult populations and generally ignore physiological variables specific to children (such as growth and development indicators) and the high-dimensionality of data. More importantly, traditional models (such as XGBoost) are prone to overfitting when processing small sample medical data, resulting in poor predictive accuracy when applied to the pharmaceutical field. Summary of the Invention

[0004] This invention aims to address at least one of the technical problems existing in the prior art. To this end, this invention provides a method, device, medium, and procedure for predicting sirolimus dosage in children; the method of this invention addresses three major clinical challenges in sirolimus administration in children: dosage errors due to developmental dynamics, medication risks in high-risk groups, and passive decision-making patterns, by developing a high-precision dosage prediction model.

[0005] The first aspect of this application discloses a method for predicting sirolimus dosage in children, the method comprising:

[0006] Obtain the prediction parameter information of the sample to be tested, including weight, age, platelet count (PLT), and creatinine (CREA); The predicted parameter information is input into the dose prediction model to calculate the predicted dose of sirolimus; the predicted dose is the average daily dose during the stable treatment period. The method for constructing the dose prediction model includes: obtaining a prediction parameter dataset of training set samples; iteratively training a machine learning model based on the training set samples including the prediction parameter dataset; and using the trained model as the dose prediction model when the training stops.

[0007] In some embodiments, the age is less than or equal to 14 years old.

[0008] In some embodiments, the prediction parameter information further includes at least one of disease type, albumin (ALB), and low-density lipoprotein cholesterol (LDL-C); Optionally, the disease types include vascular malformations and primary immunodeficiency.

[0009] In some embodiments, the prediction parameter information further includes at least one of aspartate aminotransferase (AST), alanine aminotransferase (ALT), and gender.

[0010] In some embodiments, the prediction parameter information further includes at least one of the following: hemoglobin level (HGB), drug concentration (TDM), urea (UREA), red blood cell count (RBC), and total bilirubin (TBIL).

[0011] In some embodiments, the machine learning model includes at least one of CatBoost, DT, FT Transformer, LGBM, Linear Regression, ResNet, Ridge, SVM, TabPFN, and XGBoost; preferably TabPFN.

[0012] In some embodiments, the prediction parameter dataset is obtained in the following manner: The training set samples include a total dataset containing demographic characteristics, laboratory test indicators, and treatment-related indicators, as well as the total sirolimus dose for the corresponding samples. The total sirolimus dose is the average daily dose during the stable treatment period. The prediction parameter dataset is obtained by filtering from the total dataset; The demographic characteristics include age, sex, and weight; the laboratory test indicators include complete blood count, liver and kidney function, and lipid profile; and the treatment-related indicators include drug concentration (TDM) and disease type (Source).

[0013] A second aspect of this application discloses a computer device, comprising: a memory and a processor; the memory is used to store a computer program; the processor executes the computer program to implement the steps of the above-described method.

[0014] A third 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.

[0015] The fourth 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.

[0016] This application deeply integrates machine learning technology with the principles of developmental pharmacology, solving three major clinical challenges in precision drug delivery in pediatrics: 1. Breaking the "dosage dilemma" in dynamic drug administration for children's development: Children are not simply miniature adults—the dynamic processes of organ maturation, metabolic enzyme activity, and weight changes make the traditional weight-based dosing model prone to errors of 30-40%. This model breaks through these limitations through dual innovations.

[0017] Capturing nonlinear developmental trajectories: ①SHAP analysis reveals the synergistic dominant effect of weight and age (contribution value >32%), accurately quantifying metabolic differences at different developmental stages.

[0018] Infancy (<6 years): Low CYP3A4 activity necessitates dose reduction (up to 0.15 mg / kg for every year of age decrease); Pre-puberty (6-12 years): The cumulative effect of weight gain and increased enzyme activity requires stepwise dose increases (dose-weight curve slope change rate up to 18%). This finding explains a common clinical phenomenon: blood drug concentration fluctuations of up to 300% in children of the same weight due to age differences (e.g., 3-year-old vs. 10-year-old). The model, through age correction, reduces the dose error in this group by 37% (MAE = 0.17 vs. 0.27 using the traditional method).

[0019] Avoiding the "weight trap": When weight > 20 kg, the rate of increase in dosage requirement slows down, which is related to changes in the ratio of body surface area to metabolic rate in children. The model automatically identifies this inflection point, avoiding the risk of overdose caused by linear extrapolation (e.g., a 30 kg child only needs 82% of the theoretical dose).

[0020] 2. Model innovation and performance advantages: TabPFN, a novel Transformer architecture specifically designed for tabular data, demonstrates superior performance on small-sample medical data through prior knowledge integration and Bayesian inference mechanisms. Compared to traditional models (such as XGBoost, RMSE=0.30, R²=0.74), TabPFN reduces prediction error by 17%. Its advantages stem from: Feature interaction capture capability: SHAP heatmaps show that the model can dynamically resolve multivariate nonlinear relationships (…). Figure 9This model demonstrates several advantages over traditional pharmacokinetic models. For example, it exhibits a synergistic effect between weight and creatinine (CREA). It also demonstrates resistance to overfitting: the lowest standard deviation in 10-fold cross-validation (MAE standard deviation of only 0.03) indicates strong robustness to data fluctuations. Furthermore, it offers advantages in learning from fewer samples: achieving R²>0.8 with only 522 training cases, addressing the data limitations often faced in clinical research. Compared to empirical dosing or traditional pharmacokinetic models (such as POP-PK), this model offers: multi-dimensional integration: simultaneously incorporating demographics, diagnostic indicators (15 key variables), and treatment context (Source); real-time prediction capability: TabPFN single prediction takes less than 1 second, supporting immediate clinical decision-making; and improved accuracy: ±30% accuracy (75.57%) surpasses reported accuracy rates for POP-PK models (typically 60-70%).

[0021] 3. Clinical significance of key predictors: SHAP analysis revealed predictors that were highly consistent with sirolimus pharmacokinetics. Figure 10 The positive dominance of weight and age (SHAP value has the highest weight): This aligns with clinical practice of administering medication to children based on surface area (daily dose range 1-2 mg / m²), and age correction is needed for differences in metabolic enzyme development before puberty; Positive correlation with creatinine (CREA): This confirms the renal excretion of sirolimus (approximately 40% clearance), and changes in renal function directly affect blood drug concentration; Negative correlation with hemoglobin (HGB): This may be related to the drug's myelosuppressive side effects, and dose reduction is necessary for patients with low HGB to avoid exacerbating anemia. These findings not only validate the biological rationale of the model but also provide an interpretable decision-making path for dose adjustment. Most importantly, the model provides the first individualized treatment plan for the traditionally "unreliable" group of children with rare diseases—primary immunodeficiency (accounting for 38.9% of the cohort), achieving an R² of 0.89 and an accuracy of 80.77% ± 30% in this subgroup.

[0022] In summary, this application constructs and validates a sirolimus dosage prediction model for children based on real-world data. It addresses core issues through the following innovations: Data level: Integrating multidimensional characteristics of children in clinical practice (including growth and development, liver and kidney function, and blood lipid indicators), systematically quantifying the contribution of child-specific variables; Algorithm level: Comparing 10 machine learning / deep learning models (including the cutting-edge TabPFN algorithm), overcoming the bottleneck of small-sample modeling; Interpretability level: Applying SHAP (SHapley Additive ex Planations) to analyze the mechanism of action of key predictive factors, promoting the clinical application of the model. The research aims to provide precise tools for personalized drug administration in children, shortening the time to target dosing and reducing the risk of adverse reactions. Attached Figure Description

[0023] 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.

[0024] Figure 1 This is a schematic diagram of the method flow provided in the first aspect of the present invention; Figure 2 This is a schematic diagram of a system for predicting sirolimus dosage in children provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of a computer device provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of the architecture of an exemplary computing device provided in an embodiment of the present invention; Figure 5 This is a schematic diagram of the storage medium provided in an embodiment of the present invention; Figure 6 This is an analysis flowchart provided in an embodiment of the present invention; Figure 7 This is the result of stepwise feature filtering provided in the embodiments of the present invention; Figure 8 These are the results of ten-fold cross-validation of different models provided in the embodiments of the present invention; Figure 9 This is a SHAP heatmap of the TabPFN model provided in this embodiment of the invention; Figure 10 This is a visualization of SHAP values ​​provided in an embodiment of the present invention; Figure 11 Some parameters provided in the embodiments of the present invention are based on the TabPFN model SHAP dependency graph, wherein, Figure 11 'a' represents body weight. Figure 11 b is age. Figure 11 c represents the disease type. Figure 11 d stands for PLT. Figure 11 e stands for CREA Figure 11 f is the AST. Figure 11 g represents ALB. Figure 11 h represents ALT; Figure 12 Another set of parameters provided for embodiments of the present invention is based on the TabPFN model SHAP dependency graph, wherein, Figure 12 i represents gender. Figure 12 j is HGB, Figure 12 k is TDM, Figure 12 l represents UREA. Figure 12 m is RBC, Figure 12n is TBIL Figure 12 o represents LDL-C. Detailed Implementation

[0025] 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.

[0026] 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.

[0027] 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.

[0028] Figure 1 This is a schematic flowchart of a method for predicting sirolimus dosage in children according to an embodiment of the present invention. Specifically, the method includes the following steps: S101: Obtain the prediction parameter information of the sample to be tested, including weight, age, platelet count (PLT), and creatinine (CREA); In some embodiments, the terms “subject” or “test subject” or “sample” 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.

[0029] In some embodiments, the age is less than or equal to 14 years old.

[0030] In some embodiments, the prediction parameter information further includes at least one of disease type, albumin (ALB), and low-density lipoprotein cholesterol (LDL-C); Optionally, the disease types include vascular malformations and primary immunodeficiency.

[0031] In some embodiments, the prediction parameter information further includes at least one of aspartate aminotransferase (AST), alanine aminotransferase (ALT), and gender.

[0032] In some embodiments, the prediction parameter information further includes at least one of the following: hemoglobin level (HGB), drug concentration (TDM), urea (UREA), red blood cell count (RBC), and total bilirubin (TBIL).

[0033] S102: Input the prediction parameter information into the dose prediction model to calculate the predicted dose of sirolimus; the predicted dose is the average daily dose during the stable treatment period; In some embodiments, the method for constructing the dose prediction model includes: obtaining a prediction parameter dataset of training set samples; iteratively training a machine learning model based on the training set samples including the prediction parameter dataset; and using the trained model as the dose prediction model when the trained model reaches the training stopping condition.

[0034] In some embodiments, the machine learning model includes at least one of CatBoost, DT, FT Transformer, LGBM, Linear Regression, ResNet, Ridge, SVM, TabPFN, and XGBoost; preferably TabPFN.

[0035] In some embodiments, since the TabPFN model is used as a black-box model, the specific formula coefficients are typically not directly known. TabPFN (Table-Based Model) is a Transformer-based black-box model with no explicit coefficient formulas: TabPFN learns "general inference rules" through pre-training on millions of synthetic datasets, rather than traditional linear or nonlinear formulas. It outputs a predicted distribution (such as a height probability distribution) rather than a regression equation with fixed coefficients. TabPFN's predictions rely on pre-trained weights and a dynamic inference process, and the coefficients are opaque. Therefore, in this embodiment, the interpretation of the dose prediction model relies on SHAP analysis.

[0036] In some embodiments, the prediction parameter dataset is obtained in the following way: The training set samples include a total dataset containing demographic characteristics, laboratory test indicators, and treatment-related indicators, as well as the total sirolimus dose for the corresponding samples. The total sirolimus dose is the average daily dose during the stable treatment period. The predicted parameter dataset is obtained by filtering from the total dataset; Demographic characteristics include age, sex, and weight; laboratory test indicators include complete blood count, liver and kidney function, and lipid profile; treatment-related indicators include blood drug concentration (TDM) and disease grouping (Source).

[0037] In some embodiments, the auxiliary prediction results output based on the risk value 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 only used as a reference for medical staff, not as the final diagnosis results of the subjects.

[0038] Figure 3 This is a schematic diagram of a computer device provided in an embodiment of the present invention, such as... Figure 3 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.

[0039] 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.

[0040] 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.

[0041] For example, the method or apparatus according to embodiments of this disclosure can also be used by means of Figure 4 The architecture of the computing device 3000 shown is used for implementation. For example... Figure 4As 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 4 The architecture shown is merely exemplary and can be omitted as needed when implementing different devices. Figure 4 One or more components in the computing device shown.

[0042] This invention also includes a computer-readable storage medium, such as... Figure 5 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 (DRRAM). 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.

[0043] 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, such as... Figure 2 As shown, the computer program product or system for predicting sirolimus dosage in children includes: The parameter acquisition module 201 is used or configured to acquire the predicted parameter information of the sample to be tested, the predicted parameters including weight, age, platelet count (PLT), and creatinine (CREA). The result output module 202 is used or configured to input the prediction parameter information into the dose prediction model to calculate the predicted dose of sirolimus; the predicted dose is the average daily dose during the stable treatment period.

[0044] The following is a study based on the specific medical records of a particular child. The specific methods and results are as follows: 1. Materials and Methods 1.1 Research Design and Data Sources This study employed a single-center retrospective cohort design, with data sourced from the hospital's electronic medical record system (January 2018 – December 2024). Based on a previously established cohort focusing on vascular malformations and primary immunodeficiency diseases, pediatric patients aged ≤18 years who received sirolimus treatment for ≥7 days and had complete dosage records and ≥2 plasma concentration monitoring records (interval ≥72 hours) were included. Exclusion criteria included patients with hepatic or renal failure (Child-Pugh C or eGFR <30 mL / min / 1.73m²) and those concurrently using potent CYP3A4 inhibitors / inducers. A total of 653 children from 23 provincial-level administrative regions in China were ultimately included. This study was approved by the institutional ethics committee (Approval No.:

[2025] -Y-059-D), and informed consent from patients was waived due to the use of desensitized retrospective data. The study process analysis is as follows: Figure 6 As shown.

[0045] 1.2 Variable Definition and Processing The primary outcome variable was the total sirolimus dose (mg), defined as the average daily dose during the stable treatment period. Predictive variables covered three main categories: demographic characteristics (age, sex, weight), laboratory indicators (including 8 complete blood count (CBC), 6 liver and kidney function tests, and 5 lipid profile tests), and treatment-related indicators (TDM blood drug concentration and disease group source). Age and weight were expressed as median, and laboratory test values ​​were the closest measurements taken within 7 days prior to the first treatment monitoring. Disease group was defined as a binary variable (vascular malformation = 0, primary immunodeficiency = 1). Among all variables, the missing rate for CBC indicators was 9.19%, the missing rate for liver and kidney function indicators was <1.5%, the missing rate for lipid profile indicators was <1.7%, and no other variables were missing.

[0046] 1.3 Data Preprocessing Flow Data preprocessing employed a systematic workflow: first, random forest imputation was performed on continuous variables, and mode imputation was used for categorical variables; second, variables with a missing rate >50% were removed; finally, extremely imbalanced variables (single-class percentage >95%) were eliminated. The feature engineering stage adopted a stepwise feature addition strategy based on XGBoost: by calculating the importance ranking of variables, features were added stepwise and the changes in the coefficient of determination (R²) were observed. Ultimately, 15 key predictive variables were identified, including weight, age, creatinine, blood drug concentration, and hemoglobin. The combination of these variables achieved the optimal R² value when k=15.

[0047] 1.4 Model Development and Validation The dataset was stratified and sampled according to disease group (Source), divided into a training set (80%, n=522) and a test set (20%, n=131). Model development covered 10 algorithms, divided into four categories: ensemble learning (XGBoost, LGBM, CatBoost, RF), deep learning (ResNet, FT-Transformer), traditional models (Linear Regression, Ridge, SVM) [18-20], and cutting-edge tabular data processing models (TabPFN). Hyperparameter optimization adopted ten-fold cross-validation grid search, with the optimization objective being to minimize the root mean square error (RMSE). Model evaluation used four metrics: the primary metrics were RMSE and coefficient of determination (R²), and the secondary metrics included mean absolute error (MAE) and prediction accuracy within the error ranges of ±10%, ±20%, ±30%, ±40%, and ±50%.

[0048] 1.5 Interpretability Analysis Method The model interpretation employs the SHAP (SHapley Additive exPlanations) framework, implemented using the shap 0.44.0 library. Global feature importance analysis is based on ranking by the absolute mean of SHAP values; feature effect directions are visualized using SHAP dependency graphs; individual prediction interpretation uses force-directed graphs to show the contribution strength and direction of key features. Specifically, honeycomb graphs and dependency graphs are generated for the optimal model (TabPFN) to reveal the dose-response curves of key variables such as weight and age.

[0049] 1.6 Subgroup Analysis Protocol To assess the robustness of the model across different populations, two subgroup analyses were pre-planned: grouping by sex (male = Gender_0, female = Gender_1) and grouping by disease origin (Source_vascular malformation vs Source_primary immunodeficiency). RMSE, R², and MAE were calculated independently for each subgroup, with ±30% accuracy. Wilcoxon signed-rank test was used to compare differences between groups, with a significance level set at α=0.05.

[0050] 1.7 Statistical Analysis Continuous variables were described using median (interquartile range), and Spearman correlation analysis was used for comparisons between groups. Categorical variables were expressed as frequencies (percentages), and chi-square tests or Fisher's exact tests were used for comparisons between groups. Paired t-tests were used to compare model performance. All analyses were performed using Python 3.10 and scipy 1.11.0 libraries. The TabPFN model was run on an NVIDIA A100 GPU, and single-pass forward propagation training was achieved by utilizing its prior data to fit the network characteristics.

[0051] 2 Results 2.1 Patient Baseline Characteristics This study ultimately included 653 children, with a median age of 6.57 years (IQR: 3.23–9.86), a median weight of 21.00 kg (IQR: 15.00–31.00), and a balanced male-to-female ratio (54.82% vs 45.18%). Disease stratification showed that 61.10% had vascular malformations (Source_0). Among laboratory parameters, the median sirolimus plasma concentration (TDM) was 6.93 ng / mL (IQR: 4.98–9.68), creatinine (CREA) was 28.10 μmol / L (IQR: 22.60–33.91), and hemoglobin (HGB) was 121.00 g / L (IQR: 108.00–132.00). Nine indicators were missing (missing rate 0.77%-9.34%), and their distributions did not shift significantly after random forest imputation (KS test p>0.05) (Table 1).

[0052] Table 1. Demographic and Characteristic Statistical Description Table

[0053] Note 1: In the "Variable Name" column of the table above, the format for categorical variables is "Variable Name, n(%)", where n represents the frequency of the category and the value in parentheses represents the percentage of the category; the format for continuous variables is "Variable Name, median(IQR)", where median represents the median and IQR (interquartile range) represents the range of the upper and lower quartiles (lower quartile - upper quartile).

[0054] Note 2: Different methods are used to calculate the p-value for different types of feature variables. When the feature variable is a continuous variable, the Spearman correlation coefficient hypothesis test is used; when the feature variable is a binary variable, the t-test is used if it follows a normal distribution, and the Mann-Whitney U test is used if it does not follow a normal distribution; when the feature variable is a multi-category variable, the Kruskal-Wallis test is used.

[0055] 2.2 Feature Selection and Key Predictive Factors Fifteen core predictor variables were identified using a stepwise feature addition strategy driven by XGBoost. Figure 7 The top five variables in terms of importance were: weight, age, disease, platelets, and creatinine. This combination improved the model's R² to 0.81, a 9.4% improvement over the all-variable model (ΔR²=0.02).

[0056] 2.3 Model Performance Comparison 2.3.1 Ten-fold cross-validation The cross-validation performance of the 10 models is as follows Figure 8 As shown in Table 2, TabPFN exhibits the best predictive ability: mean RMSE 0.35±0.09, mean R² 0.70±0.10, and mean MAE 0.23±0.03. Its standard deviation is lower than that of other models, indicating the best stability.

[0057] Table 2. Results of 10-fold cross-validation (mean ± std)

[0058] 2.3.2 Test Set Validation As shown in Table 3, TabPFN significantly outperformed other models in terms of accuracy: RMSE (0.25 vs other models 0.30-0.36), R² (0.81 vs 0.63-0.75), and MAE (0.17 vs 0.22-0.29). Its ±30% dose prediction accuracy reached 75.57% (Table 4), an improvement of 3-5 percentage points compared to the second-best models, CatBoost (72.52%) and LGBM (70.99%). Within a wide error range of ±50%, TabPFN maintained an accuracy of 87.02%.

[0059] Table 3 Test set validation results

[0060] Table 4. Accuracy Results of the Testing Machine

[0061] 2.4 Interpretability Analysis of the TabPFN Model 2.4.1 Ranking of Feature Importance SHAP analysis ( Figure 9-11 The results showed that weight and age were the most significant predictors (SHAP value contributed 32.6%), followed by creatinine (CREA), total drug concentration (TDM), and hemoglobin (HGB). Weight and age were both strongly positively correlated (SHAP value > 0), while hemoglobin was negatively correlated (SHAP value < 0).

[0062] 2.4.2 Feature Dependency like Figure 9-11 As shown, the nonlinear effects of key variables are as follows: the rate of increase in dose requirement slows down when body weight > 20 kg (a); the positive correlation of dose increases when creatinine > 30 μmol / L (e); and the negative effect is significant when hemoglobin < 110 g / L (j).

[0063] 2.5 Subgroup Validation Results 2.5.1 Sex subgroups (Tables 5 and 6) Male patients (Gender_0) had higher predictive accuracy (RMSE=0.24, R²=0.86), but the ±30% accuracy (71.01%) was lower than that of female patients (80.65%), which may be related to the more concentrated weight distribution in female patients.

[0064] 2.5.2 Disease source subgroups (Tables 5 and 6) The second group (Source_1, primary immunodeficiency) showed significantly better predictive performance than the first group: RMSE (0.18 vs 0.29), R² (0.89 vs 0.77), and ±30% accuracy (80.77% vs 72.15%), indicating the model's applicability to specific populations.

[0065] Table 5. Validation results of gender subgroups based on the TabPFN model

[0066] Table 6. Accuracy results of gender subgroups based on the TabPFN model.

[0067] This study, based on real-world data from 653 children at Beijing Children's Hospital, established for the first time a machine learning prediction model for sirolimus administration in children. Results showed that the TabPFN model (RMSE=0.25, R²=0.81, MAE=0.17) significantly outperformed nine other control models (including XGBoost and ResNet), achieving a ±30% dose prediction accuracy of 75.57%, providing an efficient tool for personalized medication in children. The core breakthrough lies in the deep integration of machine learning technology with developmental pharmacology principles, solving three major clinical challenges in precise pediatric drug administration.

[0068] It should be noted that the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0069] 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.

[0070] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0071] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection between apparatuses or units through some interfaces, and may be electrical, mechanical, or other forms.

[0072] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0073] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0074] 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 predicting sirolimus dosage in children, characterized in that, The method includes: Obtain the prediction parameter information of the sample to be tested, including weight, age, platelet count, and creatinine; The predicted parameter information is input into the dose prediction model to calculate the predicted dose of sirolimus; the predicted dose is the average daily dose during the stable treatment period. The method for constructing the dose prediction model includes: Obtain the prediction parameter dataset of the training set samples; based on the training set samples including the prediction parameter dataset, iteratively train the machine learning model; when the trained model reaches the training stopping condition, use the trained model as the dose prediction model.

2. The method for predicting sirolimus dosage in children according to claim 1, characterized in that, The age is less than or equal to 14 years old.

3. The method for predicting sirolimus dosage in children according to claim 1, characterized in that, The prediction parameter information also includes: disease type, at least one of albumin and low-density lipoprotein cholesterol; Optionally, the disease types include vascular malformations and primary immunodeficiency.

4. The method for predicting sirolimus dosage in children according to claim 1, characterized in that, The prediction parameter information also includes at least one of aspartate aminotransferase, alanine aminotransferase, and sex.

5. The method for predicting sirolimus dosage in children according to claim 1, characterized in that, The prediction parameter information also includes at least one of the following: hemoglobin content, blood drug concentration, urea, red blood cell count, and total bilirubin.

6. The method for predicting sirolimus dosage in children according to claim 1, characterized in that, The machine learning model includes at least one of CatBoost, DT, FT Transformer, LGBM, Linear Regression, ResNet, Ridge, SVM, TabPFN, and XGBoost; preferably TabPFN.

7. The method for predicting sirolimus dosage in children according to claim 1, characterized in that, The prediction parameter dataset is obtained in the following way: The training set samples include a total dataset containing demographic characteristics, laboratory test indicators, and treatment-related indicators, as well as the total sirolimus dose for the corresponding samples. The total sirolimus dose is the average daily dose during the stable treatment period. The prediction parameter dataset is obtained by filtering from the total dataset; The demographic characteristics include age, sex, and weight; the laboratory test indicators include complete blood count, liver and kidney function, and lipid profile; and the treatment-related indicators include blood drug concentration and disease type.

8. 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 steps of the method according to any one of claims 1-7.

9. A computer-readable storage medium, characterized in that, It stores a computer program that, when executed by a processor, implements the steps of the method as described in any one of claims 1-7.

10. A computer program product, comprising a computer program, characterized in that, When executed by a processor, the computer program implements the steps of the method described in any one of claims 1-7.

Images