Child obesity intervention decision system based on stable typing and knowledge base association and construction method thereof

Abstract

The application discloses a child obesity intervention decision system based on stable typing and knowledge base association and a construction method thereof, and comprises the following steps: screening a core index set from candidate indexes by combining a correlation analysis and a random forest feature selection method and manual selection; the candidate indexes comprise a plurality of static indexes and a plurality of dynamic indexes; based on the core index set, an obesity metabolic subtype classification model is constructed by using an unsupervised clustering algorithm; a knowledge base associated with each obesity subtype is constructed; and the obesity metabolic subtype classification model and the knowledge base are integrated. According to the scheme, the core index set for obesity typing is screened by combining the manual and algorithm feature selection methods, obesity children are divided into different subtypes based on the core index set, the stability and reliability of typing are realized, the problem that the typing has no clinical guiding significance is overcome, meanwhile, the stable and reliable typing result combined with the constructed knowledge base can provide precise decision suggestions for child obesity intervention.

Description

Technical Field

[0001] This invention belongs to the field of childhood obesity technology, and in particular relates to a childhood obesity intervention decision system and its construction method based on stable typing and knowledge base association. Background Technology

[0002] Childhood obesity has become a major public health concern worldwide. Statistics show that the number of overweight or obese children and adolescents globally exceeds 340 million, and the obesity rate among children and adolescents in China continues to rise. Obesity not only affects children's physical and mental health but is also closely related to various complications such as hypertension, type 2 diabetes, and metabolic-related fatty liver disease. Furthermore, approximately 82.3% of obese children will maintain their obesity into adulthood, significantly increasing the long-term risk of cardiovascular and metabolic diseases. Therefore, early risk stratification and precise intervention for obese children have significant clinical implications.

[0003] Currently, the clinical diagnosis of childhood obesity mainly relies on body mass index (BMI) or its derivatives (such as the BMIZ score). However, BMI only reflects the overall degree of overweight and cannot reveal the complex pathophysiological heterogeneity behind obesity. Clinical practice shows that obese children with the same BMI level may have significant differences in metabolic status, risk of complications, and response to interventions. To overcome this limitation, researchers have begun to attempt to classify obese populations based on multidimensional clinical data in order to achieve more refined risk assessment. However, existing classification studies still have the following shortcomings:

[0004] First, existing studies often suffer from arbitrariness or incompleteness in determining the indicators used for typing. For example, some studies directly include all available clinical indicators, leading to severe collinearity among indicators and clustering results affected by redundant variables. Other studies select only some indicators based on experience, ignoring the impact of gender differences and physiological changes during children's growth and development on metabolic indicators. This results in typing results that are difficult to reproduce across different cohorts, poor clustering stability, and limited reference value of the typing results. Summary of the Invention

[0005] The purpose of this invention is to address the problems existing in the prior art by proposing a decision-making system for childhood obesity intervention based on stable classification and knowledge base association, and its construction method. First, the stability of classification is ensured through indicator screening. Second, a knowledge base corresponding to each classification is constructed by combining multiple intervention cohorts to achieve accurate decision support based on stable classification.

[0006] To achieve the above objectives, the present invention adopts the following technical solutions:

[0007] The above-mentioned method for constructing a decision-making system for childhood obesity intervention based on stable typing and knowledge base association includes:

[0008] From the candidate indicators of the childhood obesity cohort, a core indicator set for obesity classification was selected by using a feature selection method that combines correlation analysis and random forest.

[0009] The candidate indicators include several static indicators and several dynamic indicators, and the dynamic indicators include the Matsuda index, the area under the curve of insulin increment in the oral glucose tolerance test (OGTT), and the area under the curve of blood glucose increment in the OGTT.

[0010] Based on the core indicator set, an unsupervised clustering algorithm was used to construct an obesity metabolic subtype classification model on the discovery queue, which identified multiple obesity subtypes with clinical differences.

[0011] Analyze the differences in response to intervention measures among different obesity subtypes and construct a knowledge base associated with each obesity subtype;

[0012] By integrating the obesity metabolic subtype classification model and the knowledge base, when the core indicator set of the individual to be evaluated is obtained, the obesity subtype to which the individual to be evaluated belongs and the corresponding personalized intervention suggestions are output.

[0013] In the above-mentioned method for constructing a decision-making system for childhood obesity intervention based on stable classification and knowledge base association, the feature selection method combining correlation analysis and random forest to screen out the core indicator set for obesity classification specifically includes:

[0014] Collinearity among all candidate indicators was assessed using the Spearman correlation matrix.

[0015] Indicator pairs whose absolute correlation coefficients are higher than the coefficient threshold are identified as highly collinear indicator pairs;

[0016] Remove one index from each pair of highly collinearity indices;

[0017] Random forest models with body mass index as the dependent variable were constructed for both male and female groups.

[0018] Based on the importance evaluation index values ​​output by the random forest model for each candidate indicator, several candidate indicators whose importance evaluation index values ​​for male and female targets are greater than the importance index threshold are extracted from the remaining candidate indicators after correlation analysis.

[0019] Take the union of the indicators extracted from both, and select the first indicator set as the core indicator set.

[0020] In the above-mentioned method for constructing a decision-making system for childhood obesity intervention based on stable classification and knowledge base association, the coefficient threshold is 0.8;

[0021] The importance evaluation index mentioned is the %IncMSE index;

[0022] The threshold for the important index is 8;

[0023] By incorporating triglycerides and age as supplementary indicators into the first indicator set, a core indicator set containing triglycerides and age is obtained.

[0024] In the above-mentioned method for constructing a decision-making system for childhood obesity intervention based on stable typing and knowledge base association, for each pair of highly collinear indicators, indicators with low clinical accessibility or weak interpretability are removed.

[0025] The candidate indicators include systolic blood pressure, diastolic blood pressure, serum uric acid, alanine aminotransferase, triglycerides, total cholesterol, high-density lipoprotein cholesterol, low-density lipoprotein cholesterol, glycated hemoglobin, insulin resistance index, fasting insulin level, fasting blood glucose, Matsuda index, area under the curve of OGTT insulin increment, and area under the curve of OGTT blood glucose increment.

[0026] After screening using feature selection methods, age and triglycerides were added to the core indicator set.

[0027] The core indicator set includes: age, systolic blood pressure, high-density lipoprotein cholesterol, triglycerides, alanine aminotransferase, serum uric acid, insulin resistance index, Matsuda index, area under the curve of OGTT insulin increment, and area under the curve of OGTT blood glucose increment.

[0028] In the above-mentioned method for constructing a decision-making system for childhood obesity intervention based on stable classification and knowledge base association, this method also includes performing principal component analysis to reduce the dimensionality of the core indicator set;

[0029] In the dimensionality-reduced space, K-means clustering, and / or hierarchical clustering, and / or self-organizing neural network algorithms were used for clustering. The clustering of the discovery cohort identified four obesity subtypes, including simple obesity, visceral fat storage tendency, isolated insulin resistance, and multiple metabolic disorders.

[0030] In the above-mentioned method for constructing a decision-making system for childhood obesity intervention based on stable typing and knowledge base association, after constructing the obesity metabolic subtype classification model, a confidence quantification step is also included:

[0031] Based on the prototype / cluster features of each subtype obtained by the unsupervised clustering algorithm, the fuzzy C-means algorithm is used to calculate the membership probability of each individual belonging to each subtype and output the classification confidence level.

[0032] When the core indicator set of the individual to be evaluated is obtained, the output also includes the obesity subtype to which the individual to be evaluated belongs and its classification confidence level;

[0033] In the above-mentioned method for constructing a decision-making system for childhood obesity intervention based on stable typing and knowledge base association, the differences in the response to intervention measures include:

[0034] The extent of improvement in metabolic indicators after standardized lifestyle intervention varied among different obesity subtypes;

[0035] The degree of improvement in metabolic indicators after drug intervention varied among different obesity subtypes.

[0036] In the above-mentioned method for constructing a decision-making system for childhood obesity intervention based on stable subtypes and knowledge base association, the differences in responses of different subtypes to different intervention measures are obtained in the following way:

[0037] Prepare drug intervention cohorts and lifestyle intervention cohorts;

[0038] Based on the aforementioned obesity metabolic subtype classification model, obesity was classified into two cohorts: the drug intervention cohort and the lifestyle intervention cohort. Four classification results were obtained for each cohort.

[0039] The same drug intervention trial was used for each of the four subtypes in the drug intervention cohort;

[0040] The same standardized lifestyle intervention was used for each of the four subtypes of the lifestyle intervention cohort;

[0041] Based on the results of drug intervention, the response of the four subtypes in the drug intervention cohort was analyzed under drug intervention, and the differences in the degree of improvement of metabolic indicators of each obesity subtype under drug intervention were obtained and stored in the knowledge base.

[0042] Based on the results of lifestyle intervention, the response of the four subtypes of the lifestyle cohort under standardized lifestyle intervention was analyzed, and the differences in the degree of improvement of metabolic indicators of each obesity subtype under standardized lifestyle intervention were obtained and stored in the knowledge base.

[0043] A decision-making system for childhood obesity intervention constructed by the method, characterized in that it includes:

[0044] The data acquisition module is used to acquire clinical indicator data of the individual to be evaluated, including several static indicators and several dynamic indicators;

[0045] The static indicators include age, systolic blood pressure, high-density lipoprotein cholesterol, triglycerides, alanine aminotransferase, serum uric acid, and insulin resistance index.

[0046] The dynamic indicators include the Matsuda index, the area under the curve of OGTT insulin increment, and the area under the curve of OGTT blood glucose increment.

[0047] The subtyping module, connected to the data acquisition module, has an embedded obesity metabolic subtype classification model, which is used to output the obesity subtype to which the individual to be evaluated belongs based on the input clinical indicator data.

[0048] The knowledge base module stores clinical knowledge data associated with different obesity subtypes, including expected intervention response data for each obesity subtype.

[0049] The decision output module is connected to the subtyping module and the knowledge base module respectively. It is used to retrieve the corresponding clinical knowledge data from the knowledge base module according to the obesity subtype to which the individual to be evaluated belongs, generate personalized intervention suggestions for the individual to be evaluated, and output a visual report.

[0050] In the aforementioned childhood obesity intervention decision system, the typing module further includes a confidence calculation unit, which is used to calculate the membership probability of the individual to be evaluated belonging to the output obesity subtype based on the fuzzy C-means algorithm, and output the classification confidence; the visualization report generated by the decision output module also includes the classification confidence.

[0051] The intervention response expectation data stored in the knowledge base module includes:

[0052] Data on the differences in the degree of improvement of metabolic indicators in different obesity subtypes after standardized lifestyle intervention;

[0053] Data on the differences in the degree of improvement of metabolic indicators in different obesity subtypes after drug intervention.

[0054] The advantages of this invention are as follows: A core set of indicators for obesity subtyping is selected through a combination of manual and algorithmic feature selection methods. Based on this core set of indicators, obese children are divided into different subtypes through data-driven cluster analysis. Multi-dimensional analysis verifies the stability and reliability of the subtyping scheme, overcoming the problem of subtyping lacking clinical guidance significance. Based on multiple independent childhood obesity cohorts (including drug intervention and lifestyle intervention cohorts), a differentiated knowledge base of responses to intervention measures for different subtypes is established. Finally, a childhood obesity subtype determination and clinical decision support system is developed that can quantify the confidence level of subtyping attribution and match personalized intervention recommendations based on high-confidence subtyping results, providing more accurate decision-making suggestions for childhood obesity intervention. Attached Figure Description

[0055] Figure 1 This is an overall flowchart of constructing a decision-making system for childhood obesity intervention in an embodiment of the present invention;

[0056] Figure 2 This is a schematic diagram of the metabolic variable screening process used for cluster analysis in an embodiment of the present invention;

[0057] a) Heatmap of Spearman correlation coefficients among all candidate indicators;

[0058] b) Ranking of variable importance based on the %IncMSE index of the male subgroup random forest model;

[0059] c) Ranking of variable importance based on the %IncMSE index of the female subgroup random forest model;

[0060] Figure 3 This is a schematic diagram illustrating the characteristic analysis of four subtypes of childhood obesity in an embodiment of the present invention;

[0061] a) A heatmap displaying the clinical indicator characteristics of each subtype;

[0062] b) Stacked diagram of the proportion of each subtype in the discovery queue;

[0063] c) Radar chart of the prevalence of obesity-related complications in each subtype, with subtype 1 as a reference;

[0064] d) Dynamic blood glucose change curves of each subtype during the oral glucose tolerance test;

[0065] e) Dynamic changes in insulin secretion of each subtype during the oral glucose tolerance test;

[0066] f) A forest plot showing the interaction between BMIZ score and metabolic variables within each subtype;

[0067] g) Schematic diagram of the continuous spectrum of subtype disease severity inferred from master curve analysis;

[0068] Figure 4 This is a schematic diagram illustrating the methodological validation and repeatability analysis of obesity subtypes in this embodiment of the invention;

[0069] a) Discover the standardized mean difference forest plot of the self-organizing neural mapping network and K-means clustering results in the queue;

[0070] b) A bar chart showing the consistency between the self-organizing neural mapping network and the K-means clustering assignment in the queue;

[0071] c) Discover the forest plot of standardized mean difference between hierarchical clustering and K-means clustering results in the queue;

[0072] d) A bar chart showing consistency between hierarchical clustering and K-means clustering assignments in the queue;

[0073] e) The maximum membership probability distribution map for evaluating the deterministic assignment of K-means clustering based on fuzzy C-means analysis;

[0074] f) A circular plot of the four subtype distributions in the validation queue within the center;

[0075] g) A circular plot of the four subtype distributions in the off-center validation cohort;

[0076] Figure 5 This is a schematic diagram illustrating the body composition characteristics and drug intervention response of obesity subtypes in an embodiment of the present invention;

[0077] a) Schematic diagram of covariance analysis of differences in interbody component indices among the four obesity subtypes;

[0078] b) Forest plot of the interaction between BMIZ score and body composition index within each subtype;

[0079] c) Bar chart showing the difference in HOMA-IR between the metformin treatment group and the control group;

[0080] d) Bar chart showing the difference in Matsuda Index between the metformin treatment group and the control group;

[0081] e) Bar chart showing the difference in insulin_inc_auc between the metformin treatment group and the control group;

[0082] Figure 6 This is a schematic diagram illustrating the overall effect and subtype-specific response after lifestyle intervention in an embodiment of the present invention;

[0083] a) Bar chart showing the percentage change of overall key metabolic variables relative to baseline after intervention;

[0084] b) Bar graph showing the changes in liver function and uric acid levels as a function of BMIZ score units in the four subgroups after intervention;

[0085] c) Bar graph showing the changes in blood lipid indicators with BMI Z-score units in the four subgroups after intervention;

[0086] d) Bar graph showing the changes in glucose metabolism indicators as a function of BMIZ score units in the four subgroups after intervention;

[0087] e) Bar graph showing the percentage change in overall body composition indicators after intervention;

[0088] f) Bar graph showing the changes in absolute mass and proportionate body composition indices as a function of BMIZ score units in the four subgroups after intervention;

[0089] g) Bar graph showing the change of relative fat distribution index with BMI Z-score units in the four subgroups after intervention;

[0090] Figure 7 This is a system architecture block diagram of the childhood obesity intervention decision system in an embodiment of the present invention. Detailed Implementation

[0091] 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.

[0092] like Figure 1 As shown, this embodiment uses data from five independent childhood obesity cohorts, as detailed below:

[0093] (1) Discovery cohort: A total of 3017 obese children were included. Data were obtained from the Department of Endocrinology of a children's hospital. The inclusion criterion was a body mass index (BMI) greater than the 95th percentile of children of the same age and sex. The exclusion criteria were: lack of key research indicators, or having known endocrine disorders, genetic diseases, viral hepatitis, and other chronic or infectious diseases.

[0094] (2) In-center validation cohort: A total of 465 obese children were included, with data obtained from the Department of Endocrinology of a children's hospital. The inclusion and exclusion criteria were the same as those of the discovery cohort. All participants in this cohort underwent body composition analysis based on dual-energy X-ray absorptiometry (DXA).

[0095] (3) External validation cohort: a total of 305 obese children, data from the Department of Pediatric Endocrinology of a People's Hospital and a Women and Children's Hospital, with inclusion and exclusion criteria the same as the discovery cohort.

[0096] (4) Drug intervention follow-up cohort: a total of 106 obese children, data from the Department of Endocrinology of a children's hospital, median follow-up time was 8.5 months, and inclusion criteria were the same as above.

[0097] (5) Lifestyle intervention cohort: A total of 45 overweight / obese children were included in the analysis after data cleaning and exclusion of individuals who did not meet the indicators required for this study.

[0098] The baseline characteristics of the five cohort datasets are shown in Table 1. The sex distribution, developmental stage (Tanner stage), BMI and BMIZ score of each population are similar.

[0099] Table 1 Baseline characteristics of the obese children dataset

[0100]

[0101] BMI, Body Mass Index; BMIZ score, standardized Z-score of BMI; WHtR, waist-to-height ratio; WHR, waist-to-hip ratio; SBP, systolic blood pressure; DBP, diastolic blood pressure; BUA, uric acid; ALT, alanine aminotransferase; TG, triglycerides; TC, total cholesterol; HDL, high-density lipoprotein cholesterol; LDL, low-density lipoprotein cholesterol; HbA1c, glycated hemoglobin; HOMA-IR, insulin resistance index.

[0102] Key Indicator Collection and Definition Standards

[0103] Diagnostic criteria for obesity-related complications

[0104] Abnormal glucose metabolism: including prediabetes and diabetes, specifically defined as impaired fasting glucose (fasting blood glucose ≥6.1 and <7.0 mmol / L), impaired glucose tolerance (2-hour OGTT blood glucose ≥7.8 and <11.0 mmol / L), or diabetes (meeting any of the following criteria: fasting blood glucose ≥7.0 mmol / L, 2-hour OGTT blood glucose ≥11.0 mmol / L, random blood glucose ≥11.1 mmol / L with corresponding symptoms, or HbA1c ≥6.5%).

[0105] Abnormal lipid metabolism: Total cholesterol (TC) ≥ 5.17 mmol / L; Low-density lipoprotein cholesterol (LDL) ≥ 3.36 mmol / L; High-density lipoprotein cholesterol (HDL) < 1.03 mmol / L; Non-HDL cholesterol ≥ 3.74 mmol / L; Triglycerides (TG), ≥ 1.12 mmol / L for children under 10 years old, and ≥ 1.46 mmol / L for children over 10 years old.

[0106] Hypertension: defined as a systolic blood pressure (SBP) and / or a diastolic blood pressure (DBP) above the 95th percentile for children of the same age, sex, and height.

[0107] Hyperuricemia: defined as a serum uric acid (BUA) level higher than 420 μmol / L.

[0108] Metabolic fatty liver disease: diagnosed by clinicians based on medical history and ultrasound or MRI results.

[0109] Metabolic syndrome:

[0110] The criteria for determining central obesity are as follows:

[0111] Girls: Waist-to-height ratio ≥ 0.46;

[0112] Boys: For those under 10 years old, the waist-to-height ratio is ≥0.46; for those ≥10 years old, the waist-to-height ratio is ≥0.48.

[0113] At least two of the following indicators must be abnormal:

[0114] High blood sugar includes impaired fasting glucose, impaired glucose tolerance, and type 2 diabetes.

[0115] hypertension;

[0116] High-density lipoprotein cholesterol (HDL) < 1.03 mmol / L, or non-HDL cholesterol ≥ 3.76 mmol / L;

[0117] Triglycerides (TG) ≥ 1.47 mmol / L.

[0118] Clinical indicator measurement

[0119] All participants had their height, weight, waist circumference, hip circumference, and other body measurements measured according to a standard protocol, and their BMI, BMIZ score, waist-to-hip ratio, and waist-to-height ratio were calculated. Their sexual development was assessed according to the Tanner stage.

[0120] Laboratory parameters were collected and tested in the morning on an empty stomach, including liver function (ALT, aspartate aminotransferase AST, gamma-glutamyl transferase GGT), lipid profile (TC, TG, HDL, LDL, apolipoprotein ApoA1, ApoB), BUA, etc., and were completed using a standard laboratory testing system.

[0121] Insulin sensitivity was assessed using a standard oral glucose tolerance test (OGTT, glucose load 1.75 g / kg, maximum 75 g). Blood samples were collected at 0, 30, 60, 90, and 120 minutes to measure blood glucose and insulin levels. Insulin resistance and sensitivity were quantified using the HOMA-IR and Matsuda Index, assessed using a homeostatic model. The area under the curve (AUC) for insulin and blood glucose increments during the OGTT was calculated using the trapezoidal method. Serum glucose levels were measured using the standard glucose oxidase method and a glucose analyzer. Plasma insulin levels were measured using radioimmunoassay.

[0122] Cluster analysis

[0123] In this embodiment, when selecting clustering indicators, the availability of indicators and data completeness are comprehensively considered, and several static indicators and several dynamic indicators are included in the preliminary analysis. Figure 2As shown in Figure a, a total of 15 candidate indicators were selected: systolic blood pressure (SBP), diastolic blood pressure (DBP), serum uric acid (BUA), alanine aminotransferase (ALT), triglycerides (TG), total cholesterol (TC), high-density lipoprotein cholesterol (HDL), low-density lipoprotein cholesterol (LDL), glycated hemoglobin (HbA1c), insulin resistance index (HOMA-IR), fasting insulin (insulin), fasting blood glucose (glucose), Matsuda Index, area under the curve of insulin increment in OGTT (insulin_inc_auc), and area under the curve of blood glucose increment in OGTT (glucose_inc_auc).

[0124] Spearman correlation matrix was used to assess collinearity among the indicators. For indicator pairs with correlation coefficients higher than 0.8, one indicator was excluded, with preference given to indicators with low clinical accessibility or weak interpretability. For indicator pairs related to glucose metabolism, OGTT-derived indicators were retained first; for indicator pairs related to lipid metabolism, indicators with higher clinical representativeness were retained first. Analysis revealed two highly collinear indicator pairs among the 15 indicators: TC and LDL, and insulin 0 and HOMA-IR. LDL and insulin 0 were removed, leaving 13 indicators.

[0125] Subsequently, a random forest model was constructed with Body Mass Index (BMI) as the dependent variable, categorized by gender. The variables were ranked according to their importance in the model output. In this embodiment, indicators with an importance greater than 8 were obtained for both men and women, and the union of these two values ​​was taken. Figure 2 bc.

[0126] The random forest model consists of multiple independent decision trees, each trained by random sampling and feature selection from the input sample set. The model uses the remaining candidate indicators (systolic blood pressure, diastolic blood pressure, uric acid, alanine aminotransferase, triglycerides, total cholesterol, high-density lipoprotein cholesterol, glycated hemoglobin, fasting blood glucose, insulin resistance index, Matsuda index, area under the curve of insulin increment in the OGTT, and area under the curve of blood glucose increment in the OGTT) as independent variables and BMI as the dependent variable. Random forest regression models are constructed and trained separately for all samples or a subset of the cohort, such as grouping the discovery cohort by gender. The core idea is to use the underlying logic that the greater the contribution of an indicator to accurately predicting BMI, the stronger its association with obesity. By employing a random forest training process to calculate the variable importance score of each indicator, and ranking the indicators for male and female groups based on these scores, the indicators with higher importance from both groups were selected. Ultimately, eight core indicators were identified based on their contribution to BMI prediction: systolic blood pressure (SBP), serum uric acid (BUA), alanine aminotransferase (ALT), high-density lipoprotein cholesterol (HDL), insulin resistance index (HOMA-IR), Matsuda index, area under the OGTT insulin increment curve (insulin_inc_auc), and area under the OGTT glucose increment curve (glucose_inc_auc). This approach aims to identify indicators with a high correlation to obesity.

[0127] Meanwhile, considering the clinical importance of triglycerides (TG) and the different physiological characteristics of children at different ages, both triglycerides and age were included. Ultimately, a core set of 10 indicators was determined for clustering: age, systolic blood pressure (SBP), high-density lipoprotein cholesterol (HDL), triglycerides (TG), alanine aminotransferase (ALT), serum uric acid (BUA), the homeostasis model assessment of insulin resistance index (HOMA-IR), the Matsuda Index, the area under the OGTT insulin increment curve (insulin_inc_auc), and the area under the OGTT glucose increment curve (glucose_inc_auc).

[0128] Considering gender differences, all subsequent cluster analyses in this embodiment were performed separately for each gender. First, Principal Component Analysis (PCA) was used to reduce the dimensionality of the 10 indicators, selecting the first two principal components. Then, K-means clustering was used to construct an obesity metabolic subtype classification model on the discovery cohort. The optimal number of clusters was determined to be 4, taking into account both silhouette coefficient and clinical interpretability. Each subtype was named according to the characteristic distribution of its clinical indicators. The final obesity metabolic subtype classification model consisted of four cluster centers and a distance-based decision rule. Of course, similar methods, such as K-medians, binary K-means, or K-means++, could also be used in practical applications; no specific restrictions were imposed.

[0129] Based on the above 10 indicators, cluster analysis was performed on the discovery cohort, identifying four distinct obesity subtypes, such as... Figure 3 As shown in b, among the four obesity subtypes, Cluster 1 showed near-normal metabolic indicators with only mild insulin resistance; Cluster 4 exhibited severely abnormal metabolic indicators, demonstrating extreme metabolic disturbances across multiple systems; Cluster 3 showed significant isolated insulin resistance, with other metabolic indicators normal or only slightly abnormal; and Cluster 2 showed mildly abnormal metabolic indicators, with relatively insignificant characteristics compared to the other three subtypes. In terms of subtype distribution, Cluster 1 accounted for approximately one-quarter of the total population, while Cluster 4 accounted for a relatively low proportion, approximately one-fifth. This indicates that the classification results achieved accurate classification of the metabolic characteristics of childhood obesity. Analysis suggests that this is because the core indicators of this scheme simultaneously include the Matsuda index, the area under the OGTT insulin increment curve, and the area under the OGTT blood glucose increment curve, which can reconstruct the entire glucose metabolism process of glucose regulation-insulin secretion-insulin action, capture the dynamic metabolic response characteristics of children under glucose load, and identify specific glucose metabolism characteristics.

[0130] Demographic analysis showed that the Cluster 1 population was generally younger, mainly in the Tanner 1 stage of development; the Cluster 4 population was older, mainly in the later stages of development; and the Cluster 2 and Cluster 3 were in the middle range. For details, please refer to Table 2.

[0131] Odds ratio analysis of complication prevalence showed that, using Cluster 1 as a reference, the odds ratios of all complications in the other three types were significantly higher. Among them, Cluster 4 had the highest prevalence of all complications in Cluster 4, significantly higher than the other three types; the prevalence of hyperuricemia in Cluster 2 was particularly prominent, being 12.2 times that of Cluster 1. Figure 3 As shown in c, this reveals the differences in the prevalence risk of obesity-related complications among different obesity subtypes.

[0132] The OGTT results showed that children in Cluster 1 had the lowest insulin secretion levels but the best glucose tolerance among the four types; conversely, children in Cluster 4 had the highest insulin secretion levels but the worst glycemic control; the glycemic curve of Cluster 3 was similar to that of Cluster 2, but its insulin secretion was significantly higher than that of Cluster 2, consistent with its significant insulin resistance metabolic characteristics. Figure 3 As shown in de. Further analysis was conducted by introducing a linear regression model with interaction terms to examine whether there were differences in the impact of increased BMI Z-score on metabolic indicators among different subtypes.

[0133] The results showed that in Cluster 4, the increase in HOMA-IR was greatest for every 1 unit increase in BMIZ score; in Cluster 2, the area under the OGTT blood glucose curve (glucose_inc_auc) decreased with increasing BMIZ score, indicating improved glucose tolerance, while the increase in serum uric acid (BUA) levels was significant. Figure 3 As shown in f. Finally, based on the above 10 clinical indicators, a master curve analysis was performed to reveal the disease severity gradient of each subtype. The analysis results show that Cluster 4 has the highest disease severity, Cluster 1 has the lowest, Cluster 2 and Cluster 3 are at an intermediate level, and Cluster 3 is slightly higher than Cluster 2, as shown in f. Figure 3 As shown in g.

[0134] The above analysis verifies the rationality of the classification scheme and reveals the clinical and metabolic patterns of different subtypes under the classification results of this scheme, which can provide data support for early risk assessment and personalized intervention.

[0135] Table 2 reveals demographic characteristics of each obesity subtype in the cohort.

[0136]

[0137] Furthermore, this embodiment assesses cluster stability by calculating the Jaccard similarity index. Through 2000 resampling evaluations, the Jaccard indices for male and female clusters reached 0.938 and 0.907 respectively, eliminating cluster randomness. The discriminability of the model was validated by constructing a random forest, and the average accuracy calculated using 10-fold cross-validation reached 0.932.

[0138] In addition, this embodiment also uses two other unsupervised clustering methods, namely Self-Organizing Neural Maps (SOM) and hierarchical clustering, to reproduce the K-means clustering results. SOM creates a 1×4 linear topological grid, uses PCA-reduced data as input, and sets 1000 iterations for training, with the learning rate linearly decreasing from 0.05 to 0.01. Hierarchical clustering uses Minkowski distance to calculate the distance matrix between samples and Ward's minimum variance method for hierarchical clustering. Classification consistency and Kappa coefficients are calculated to assess subtype stability. Simultaneously, Cohen's d-value is calculated to evaluate the differences in specific indicators between different clustering methods for the same subtype; a d-value higher than 0.5 is considered to indicate a significant difference.

[0139] The results show that both SOM and hierarchical clustering can obtain clustering results for four obesity subtypes similar to K-means. Among them, the four subtypes obtained by SOM show very little difference from K-means in the distribution of specific index values, with Cohen d values ​​≤0.08 and classification accuracy exceeding 0.92 for each subtype. Figure 4 As shown in a and b, the weighted kappa coefficient reaches 0.974; the four subtypes obtained by hierarchical clustering do not show significant differences from K-means in the distribution of specific index values, with Cohen d values ​​≤ 0.3, Cluster1 accuracy reaching 0.962, and Cluster2-4 accuracy above 0.7. Figure 4 As shown in c and d, the weighted kappa coefficient reaches 0.852, indicating that both SOM and hierarchical clustering can reproduce the clustering results obtained by K-means well, demonstrating the robustness of this embodiment at the clustering method level.

[0140] Furthermore, this embodiment employs a soft clustering method using fuzzy C-means clustering to calculate the certainty of each individual's classification according to the K-means group. The proportion of individuals with a certainty below 0.6 in each subtype is less than 17%, and the average certainty of each subtype is above 0.81, indicating that the K-means hard clustering results have high reliability. Figure 4 As shown in e.

[0141] To evaluate the reproducibility and generalization ability of this typing result, this embodiment reproduces the K-means results in both an independent cohort within the center and an external cohort. Using the centroid obtained from the discovery cohort as the standard, the distance from each individual to the centroid is calculated, and clustering is performed according to the minimum distance principle. Simultaneously, the self-clustering results of each independent cohort are used as the gold standard to calculate sensitivity and specificity. The independent cohort within the center has a sample size of 465 individuals, with 313 males, a mean BMI of 28.55 ± 4.45 kg / m², and a mean BMI Z-score of 3.11 ± 1.04. The distribution of each cluster obtained from self-clustering and allocation in the discovery cohort is shown below. Figure 4As shown in f, the average sensitivity and specificity were 0.891 and 0.958, respectively. The sample size of the independent cohort outside the center was 305 people, 208 of whom were male, with a mean BMI of 29.01 ± 4.47 kg / m2 and a mean BMI Z-score of 3.1 ± 0.96. The distribution of each cluster obtained from self-clustering and discovery cohort assignment is shown in f. Figure 4 As shown in g, the average sensitivity and specificity are 0.762 and 0.895, respectively, indicating that the genotyping pattern was well reproduced in both independent cohorts.

[0142] To reveal the body composition characteristics of different obesity subtypes, this embodiment analyzed body composition data from independent cohorts within the center, including six indicators: total lean mass, total fat mass, total % fat, lean body mass / fat ratio, android / gynoid ratio, and trunk / limb fat ratio. After adjusting for age, sex, and Tanner stage, analysis of covariance was performed. The results showed that Cluster 1 had the lowest total fat mass, the lowest total fat percentage, and the lowest degree of fat centrality (lowest android / gynoid ratio and lowest trunk / limb fat ratio). Cluster 4 had the highest total fat mass, the highest total fat percentage, and the highest degree of fat centrality (highest android / gynoid ratio and highest trunk / limb fat ratio). Figure 5 As shown in a. Further analysis using a linear regression model incorporating interaction terms revealed that with the increase in BMIZ score, total fat mass and total body fat percentage increased in all subtypes, while lean body mass ratio decreased. Cluster 4 showed the fastest increase in total fat mass and total body fat percentage, while Cluster 2 showed the fastest increase in total lean body mass. Simultaneously, Cluster 2 exhibited the fastest increase in fat centripetal density (higher increases in fat-to-hip ratio and fat-to-truncal-to-limb ratio), as shown in a. Figure 5 As shown in b, the body composition analysis results further revealed the heterogeneity among different subtypes.

[0143] Based on this, this embodiment evaluated the differences in response among different obesity subtypes under different interventions. The drug intervention cohort consisted of 106 participants, including 74 males, with a mean BMI of 28.33 ± 4.46 kg / m² and a mean BMI score of 2.96 ± 0.96. The intervention drug was metformin, and the median follow-up time was 8.5 months. A linear mixed-effects model was used to analyze the changes in metabolic indicators between the intervention and non-intervention groups for each subtype, and an interaction term was introduced. The results showed that, regarding glycemic homeostasis, metformin only significantly improved HOMA-IR in Cluster 4, and had no significant effect on insulin resistance in Clusters 1-3. Figure 5 As shown in c; it can significantly increase the Matsuda index of Clusters 2-4 and increase insulin sensitivity, with the most significant improvement observed in Cluster 3, such as Figure 5 As shown in d; metformin can significantly reduce the amount of insulin secreted throughout the Cluster 2-4 OGTT, and the area under the OGTT insulin curve is reduced, as shown in d. Figure 5 As shown in e. Overall, metformin can bring certain benefits to all subtypes, but the benefit to Cluster1 is the smallest, and no significant effect was found on Cluster1 in helping to maintain blood glucose homeostasis.

[0144] As can be seen above, metformin shows a wide range of effects on blood glucose improvement across different subtypes: it only significantly improves insulin resistance HOMA-IR in Cluster 4, while offering almost no benefit to blood glucose homeostasis in Cluster 1. Furthermore, within the same subtype, the effectiveness / ineffectiveness and degree of improvement of metformin are highly consistent among individuals. For example, HOMA-IR is significantly improved in Cluster 4 individuals, insulin sensitivity increases in Clusters 2-4 individuals, and Cluster 1 individuals show the least benefit with no significant improvement in blood glucose. This demonstrates that the subtype results of this protocol have clinical value and can guide intervention measures based on subtype classification.

[0145] Furthermore, this embodiment evaluated whether there were differences in the response levels of different subtypes under a standard lifestyle intervention. The standard lifestyle intervention in this embodiment consisted of a balanced diet and two hours of high-intensity training daily. This embodiment analyzed the results of a clinical trial initiated by a children's hospital. After data cleaning, individuals who did not meet the indicators required for this study were excluded, and the remaining 45 individuals were analyzed. Among them, 32 were male, with a mean BMI of 25.51±3.31 kg / m² and a mean BMIZscore of 2.38±0.7. They received a 3-week standard lifestyle training program. Changes in metabolic and body composition indicators before and after training were analyzed to assess whether there were differences in the response levels of different subtypes. Overall, the 3-week training significantly reduced the participants' weight, with a 3.8% decrease in BMI and a 9.3% decrease in BMIZscore compared to before training. In terms of metabolic indicators, the training greatly improved the participants' metabolic disorders, with the most significant improvement in HOMA-IR, which decreased to 29% of its pre-training level. Figure 6 As shown in a.

[0146] In subgroup analysis, this embodiment standardized the changes in indicators by dividing the changes by the difference in BMIZ score for each group before comparison. Analysis showed that Cluster 4 exhibited the most significant improvement in various metabolic indicators, including rapid correction of ALT, AST, GGT, TC, LDL, and HOMA-IR. Cluster 1 showed the smallest metabolic benefit per unit, while Clusters 2-3 also showed varying degrees of metabolic benefit. Figure 6 As shown in Figure bd, this result indicates that the classification scheme of this embodiment can distinguish obese subgroups with different patterns of metabolic index changes. Regarding body composition indicators, after training, total fat mass and total % fat decreased, while the lean body mass / fat ratio increased. Simultaneously, the android / gynoid ratio and trunk / limb fat ratio decreased, suggesting a reduction in total body fat, particularly visceral fat, but a decrease in total lean body mass, indicating a negative nitrogen balance. Figure 6 As shown in e.

[0147] Subgroup analysis showed that among the four subtypes, only Cluster 1 did not decrease in total lean body mass, but it experienced the largest decrease in total body fat percentage, the largest increase in lean body fat ratio, and the largest decreases in fat-to-hip ratio and fat-to-truncal-to-limb ratio, demonstrating a healthy trend in body composition changes. Cluster 4, while experiencing the largest decrease in total fat mass, also saw the largest decrease in total lean body mass. Cluster 2 experienced a significant decrease in total lean body mass, but the smallest decrease in total body fat percentage, fat-to-hip ratio, and fat-to-truncal-to-limb ratio, suggesting that this subtype has difficulty losing fat and tends to reduce peripheral fat first, indicating poor fat loss results. Figure 6 As shown in fg.

[0148] Similarly, the above-mentioned body composition change results show that the obesity subtype classification constructed in this embodiment can effectively distinguish obese subgroups with significant heterogeneity in body composition change patterns. Different subtypes exhibit significant differences in the trends and extent of improvement in lean body mass, fat content, and body fat distribution, and individuals within the same subtype show highly consistent body composition change patterns. Specifically, Cluster 1 shows the best trend in healthy body composition change, Cluster 4 shows significant fat reduction accompanied by substantial lean body mass loss, and Cluster 2 shows difficulty in fat loss and poor improvement in body fat distribution. This provides guidance for subsequent intervention measures and can be used to build a knowledge base. It also confirms that the classification results of this scheme are not simply mathematical clustering, but rather a classification standard with clear phenotypic differences in body composition and physiological significance. It can predict the potential for body composition improvement in different obesity subtypes, providing an objective basis for individualized body composition regulation and precise intervention management for obese patients.

[0149] Based on the differences in metabolic and body composition indicators among the various types, as well as their responses to metformin and standard lifestyle interventions, the naming and basis for each cluster are summarized as follows:

[0150] Cluster 1 is a simple obesity type with generally normal metabolic indicators, predominantly Tanner stage 1, and the lowest prevalence of various complications. It has the lowest body fat percentage, with fat distribution predominantly peripheral. Fat accumulation is minimal when BMI increases. Metformin has no significant effect on glucose metabolism, and lifestyle interventions are the most effective for fat loss.

[0151] Cluster 2 is characterized by a tendency towards visceral fat storage, with mild abnormalities in various indicators and a higher risk of hyperuricemia. Resting body composition is unremarkable, but the centripetal distribution of fat is most pronounced as BMI increases. Metformin can improve insulin sensitivity, but lifestyle interventions for weight loss tend to result in muscle loss and are less effective at reducing fat.

[0152] Cluster 3 is an isolated insulin-resistant type, characterized by significant insulin resistance but normal other indicators. It exhibits no obvious distribution characteristics in the Tanner staging and no special body composition characteristics. Metformin is the most effective treatment for improving insulin sensitivity, while lifestyle interventions offer no particular benefit.

[0153] Cluster 4 is characterized by multiple metabolic disorders, with all indicators showing significant abnormalities. It is predominantly Tanner stage 4-5, with the highest prevalence of various complications, the highest body fat percentage, and the highest degree of centripetal fat distribution. Fat accumulation is fastest when BMIZ score increases. Metformin has the most significant effect on improving HOMA-IR. Lifestyle interventions provide the greatest metabolic benefits, but muscle and fat decrease simultaneously during weight loss.

[0154] By analyzing the differences in body composition characteristics and responses to interventions among various obesity subtypes, a knowledge base associated with each obesity subtype is constructed. The knowledge base data may include:

[0155] Reference values ​​for complication prevalence: derived from the above analysis of the discovery cohort, storing the complication prevalence odds ratio / complication prevalence for each subtype.

[0156] Reference values ​​for body composition characteristics: derived from the analysis of body composition cohorts within the center used, storing typical values ​​or ranges for each subtype in terms of total body fat mass, body fat percentage, and fat distribution ratio.

[0157] Expected outcomes of intervention response: derived from the analysis of the intervention cohort used, storing the expected magnitude of improvement or response pattern of each subtype in key metabolic indicators in response to standardized lifestyle intervention or metformin treatment.

[0158] The clinical decision-making system integrates the obesity metabolic subtype classification model and knowledge base constructed using the cohort. When the core indicator set of the individual to be evaluated is obtained, the system uses the center coordinates of the four subtypes determined by K-means clustering of the cohort in the PCA space as a benchmark. It then calculates the Euclidean distance after projecting the individual's indicator data to determine the subtype, and determines the subtype based on the minimum distance principle. The system outputs the individual's obesity subtype, classification confidence level, and corresponding personalized intervention recommendations. The output may also include typical body composition characteristics corresponding to the individual's obesity subtype, complication prevalence ratio / complication prevalence, etc.

[0159] Examples of personalized intervention recommendations for each subtype are as follows:

[0160] For individuals with simple obesity: mild insulin resistance and good metabolic status, it is recommended to adopt only standardized lifestyle interventions and drug treatment is not recommended;

[0161] Visceral fat storage type: Poor response to standardized lifestyle interventions, prone to lean body loss. It is recommended to adopt personalized lifestyle interventions on the basis of insulin sensitizer (metformin) intervention, focusing on increasing resistance exercise and reducing simple weight loss diet control, so as to control fat while protecting lean body mass.

[0162] Isolated insulin resistance: The core abnormality is insulin resistance. It is recommended to prioritize the use of insulin sensitizers to improve systemic insulin sensitivity.

[0163] Multiple metabolic disorders: This type has the highest risk of complications and the most severe metabolic disorders. It is recommended to use insulin sensitizers for treatment, combined with standardized lifestyle interventions, close monitoring of complications, and multidisciplinary comprehensive management and treatment when necessary.

[0164] The above intervention recommendations are merely examples. Specific recommendations should be freely compiled by those skilled in the art based on the response of different subtypes in the two cohorts under drug and lifestyle interventions, and are not subject to any restrictions here.

[0165] In summary, this approach, based on five childhood and adolescent obesity datasets, combined feature selection methods with human assistance to identify 10 core indicators for a systematic analysis of obese children and adolescents. The analysis revealed the existence of obesity subtypes with different pathophysiological characteristics. These subtypes exhibited significant differences in metabolic indicators, body composition distribution characteristics, and responses to metformin and standard lifestyle interventions. This subtype model can be used for early risk stratification of obese children and provides a basis for developing individualized intervention strategies, contributing to the advancement of precision medicine for childhood obesity.

[0166] Specifically, such as Figure 7 As shown, the childhood obesity intervention decision system provided in this solution includes:

[0167] Data acquisition module 1 is used to acquire clinical indicator data of the individual to be evaluated, including several static indicators and several dynamic indicators;

[0168] Static indicators include age, systolic blood pressure, high-density lipoprotein cholesterol, triglycerides, alanine aminotransferase, serum uric acid, and insulin resistance index;

[0169] Dynamic indicators include the Matsuda index, area under the curve of OGTT insulin increment, and area under the curve of OGTT blood glucose increment;

[0170] Subtyping module 2, connected to data acquisition module 1, contains an obesity metabolic subtype classification model, which is used to output the obesity subtype of the individual to be evaluated based on the input clinical indicator data.

[0171] Knowledge base module 3 stores clinical knowledge data associated with different obesity subtypes, including expected intervention response data for each obesity subtype.

[0172] The decision output module 4 is connected to the subtyping module 2 and the knowledge base module 3 respectively. It is used to retrieve the corresponding clinical knowledge data from the knowledge base module according to the obesity subtype to which the individual to be evaluated belongs, generate personalized intervention suggestions for the individual to be evaluated, and output a visual report.

[0173] Preferably, the classification module 2 further includes a confidence calculation unit 5, which is used to calculate the probability of the individual to be evaluated belonging to the output obesity subtype based on the fuzzy C-means algorithm, and output the classification confidence; the visualization report generated by the decision output module 4 also includes the classification confidence.

[0174] Compared with existing technologies, the childhood obesity classification system and its supporting clinical decision system, which have been validated by multiple datasets, show significant advantages in terms of classification stability, clinical interpretability, and intervention guidance value. They can more accurately reflect the pathophysiological heterogeneity of childhood obesity and provide reliable technical means for implementing stratified management and precise intervention in clinical practice.

[0175] The specific embodiments described herein are merely illustrative of the spirit of the invention. Those skilled in the art to which this invention pertains may make various modifications or additions to the described specific embodiments or use similar methods to substitute them, without departing from the spirit of the invention or exceeding the scope defined by the appended claims.

Claims

1. A method for constructing a decision-making system for childhood obesity intervention based on stable classification and knowledge base association, characterized in that, include: From the candidate indicators of the childhood obesity cohort, a core indicator set for obesity classification was selected by using a feature selection method that combines correlation analysis and random forest. The candidate indicators include several static indicators and several dynamic indicators, and the dynamic indicators include the Matsuda index, the area under the curve of OGTT insulin increment, and the area under the curve of OGTT blood glucose increment. Based on the core indicator set, an unsupervised clustering algorithm was used to construct an obesity metabolic subtype classification model on the discovery queue, which identified multiple obesity subtypes with clinical differences. Analyze the differences in response to intervention measures among different obesity subtypes and construct a knowledge base associated with each obesity subtype; By integrating the obesity metabolic subtype classification model and the knowledge base, when the core indicator set of the individual to be evaluated is obtained, the obesity subtype to which the individual to be evaluated belongs and the corresponding personalized intervention suggestions are output.

2. The method for constructing a decision-making system for childhood obesity intervention based on stable typing and knowledge base association as described in claim 1, characterized in that, The feature selection method combining correlation analysis and random forest to screen out the core indicator set for obesity classification specifically includes: Collinearity among all candidate indicators was assessed using the Spearman correlation matrix. Indicator pairs whose absolute correlation coefficients are higher than the coefficient threshold are identified as highly collinear indicator pairs; Remove one index from each pair of highly collinearity indices; Random forest models with body mass index as the dependent variable were constructed for both male and female groups. Based on the importance evaluation index values ​​output by the random forest model for each candidate indicator, several candidate indicators whose importance evaluation index values ​​for male and female targets are greater than the importance index threshold are extracted from the remaining candidate indicators after correlation analysis. Take the union of the indicators extracted from both, and select the first indicator set as the core indicator set.

3. The method for constructing a childhood obesity intervention decision-making system based on stable classification and knowledge base association as described in claim 2, characterized in that, The threshold value for the coefficient is 0.8; The importance evaluation index mentioned is the %IncMSE index; The threshold for the important index is 8; By incorporating triglycerides and age as supplementary indicators into the first indicator set, a core indicator set containing triglycerides and age is obtained.

4. The method for constructing a decision-making system for childhood obesity intervention based on stable typing and knowledge base association as described in claim 3, characterized in that, For each pair of highly collinear indicators, indicators with low clinical accessibility or weak interpretability were removed. The candidate indicators include systolic blood pressure, diastolic blood pressure, serum uric acid, alanine aminotransferase, triglycerides, total cholesterol, high-density lipoprotein cholesterol, low-density lipoprotein cholesterol, glycated hemoglobin, insulin resistance index, fasting insulin level, fasting blood glucose, Matsuda index, area under the curve of OGTT insulin increment, and area under the curve of OGTT blood glucose increment. After screening using feature selection methods, age and triglycerides were added to the core indicator set. The core indicator set includes: age, systolic blood pressure, high-density lipoprotein cholesterol, triglycerides, alanine aminotransferase, serum uric acid, insulin resistance index, Matsuda index, area under the curve of OGTT insulin increment, and area under the curve of OGTT blood glucose increment.

5. The method for constructing a decision-making system for childhood obesity intervention based on stable typing and knowledge base association as described in claim 1, characterized in that, This method also includes performing principal component analysis to reduce the dimensionality of the core indicator set; In the dimensionality-reduced space, K-means clustering, and / or hierarchical clustering, and / or self-organizing neural network algorithms were used for clustering. The clustering of the discovery cohort identified four obesity subtypes, including simple obesity, visceral fat storage tendency, isolated insulin resistance, and multiple metabolic disorders.

6. The method for constructing a decision-making system for childhood obesity intervention based on stable typing and knowledge base association as described in claim 1, characterized in that, After constructing the obesity metabolic subtype classification model, a confidence quantification step is also included: Based on the prototype / cluster features of each subtype obtained by the unsupervised clustering algorithm, the fuzzy C-means algorithm is used to calculate the membership probability of each individual belonging to each subtype and output the classification confidence level. When the core indicator set of the individual to be evaluated is obtained, the output also includes the obesity subtype to which the individual to be evaluated belongs and its classification confidence level.

7. The method for constructing a decision-making system for childhood obesity intervention based on stable classification and knowledge base association as described in claim 5, characterized in that, The differences in response to the intervention measures include: The extent of improvement in metabolic indicators after standardized lifestyle intervention varied among different obesity subtypes; The degree of improvement in metabolic indicators after drug intervention varied among different obesity subtypes.

8. The method for constructing a decision-making system for childhood obesity intervention based on stable classification and knowledge base association as described in claim 7, characterized in that, Differences in responses to different interventions among different subtypes were obtained using the following methods: Prepare drug intervention cohorts and lifestyle intervention cohorts; Based on the aforementioned obesity metabolic subtype classification model, obesity was classified into two cohorts: the drug intervention cohort and the lifestyle intervention cohort. Four classification results were obtained for each cohort. The same drug intervention trial was used for each of the four subtypes in the drug intervention cohort; The same standardized lifestyle intervention was used for each of the four subtypes of the lifestyle intervention cohort; Based on the results of drug intervention, the response of the four subtypes in the drug intervention cohort was analyzed under drug intervention, and the differences in the degree of improvement of metabolic indicators of each obesity subtype under drug intervention were obtained and stored in the knowledge base. Based on the results of lifestyle intervention, the response of the four subtypes of the lifestyle cohort under standardized lifestyle intervention was analyzed, and the differences in the degree of improvement of metabolic indicators of each obesity subtype under standardized lifestyle intervention were obtained and stored in the knowledge base.

9. A decision-making system for childhood obesity intervention constructed by the method of any one of claims 1 to 8, characterized in that, include: The data acquisition module is used to acquire clinical indicator data of the individual to be evaluated, including several static indicators and several dynamic indicators; The static indicators include age, systolic blood pressure, high-density lipoprotein cholesterol, triglycerides, alanine aminotransferase, serum uric acid, and insulin resistance index. The dynamic indicators include the Matsuda index, the area under the curve of OGTT insulin increment, and the area under the curve of OGTT blood glucose increment. The subtyping module, connected to the data acquisition module, has an embedded obesity metabolic subtype classification model, which is used to output the obesity subtype to which the individual to be evaluated belongs based on the input clinical indicator data. The knowledge base module stores clinical knowledge data associated with different obesity subtypes, including expected intervention response data for each obesity subtype. The decision output module is connected to the subtyping module and the knowledge base module respectively. It is used to retrieve the corresponding clinical knowledge data from the knowledge base module according to the obesity subtype to which the individual to be evaluated belongs, generate personalized intervention suggestions for the individual to be evaluated, and output a visual report.

10. The childhood obesity intervention decision system according to claim 9, characterized in that, The classification module also includes a confidence calculation unit, which is used to calculate the probability of the individual to be evaluated belonging to the output obesity subtype based on the fuzzy C-means algorithm, and output the classification confidence; the visualization report generated by the decision output module also includes the classification confidence. The intervention response expectation data stored in the knowledge base module includes: Data on the differences in the degree of improvement of metabolic indicators in different obesity subtypes after standardized lifestyle intervention; Data on the differences in the degree of improvement of metabolic indicators in different obesity subtypes after drug intervention.

Images