Peri-implantitis onset risk prediction method, system and medium

Abstract

The invention provides a peri-implantitis onset risk prediction method and system and a medium, and belongs to the technical field of auxiliary prediction models.The peri-implantitis onset risk prediction method includes the steps that metabolite components of gingival crevicular fluid samples of multiple patients are obtained, difference analysis is conducted on metabolites through an OPLS-DA analysis method, a T test method, a multiple change method and an ROC curve drawing method, and the metabolite components of the gingival crevicular fluid samples of the multiple patients are obtained; determining differential metabolites as metabolite factors; acquiring baseline level information of a plurality of patients as clinical factors; taking the screened metabolite factors and clinical factors as independent variables, forming a data matrix constructed by grade variables and classification variables, and constructing a training set; and constructing a random forest model, and training through the training set to obtain a prediction model for predicting the peri-implant disease. According to the method, metabolic factors and clinical factors are considered to construct a peri-implantitis onset risk prediction model, and better technical support is provided for auxiliary judgment of the peri-implantitis onset risk.

Description

Technical Field

[0001] The present invention relates to the technical field of auxiliary prediction model systems, and in particular to a method, system and medium for predicting the risk of peri-implantitis. Background Art

[0002] Peri-implantitis is a common complication of dental implant treatment, affecting the long-term success of at least 25% of dental implants. Treating peri-implantitis is challenging because inflammation can lead to loss of supporting bone. More importantly, there is currently no reliable method to accurately predict an individual's response to treatment, which complicates treatment decisions.

[0003] Computer-assisted prediction has great potential in this area. Through advanced machine learning algorithms, a large amount of clinical data can be analyzed, including the patient's microbial characteristics, immune status, and clinical indicators, so as to more accurately predict the risk of peri-implantitis. This prediction not only helps doctors to develop personalized treatment plans, but also improves the effectiveness of treatment and patient satisfaction. In addition, computer-assisted prediction can also provide doctors with prognostic information about treatment outcomes. This is very helpful in determining the best treatment plan for each patient. For example, when a patient suffers from severe peri-implantitis, doctors can use predictive models to evaluate the potential effects of regenerative therapy to decide whether to perform bone reconstruction or replace implants.

[0004] However, the auxiliary prediction model construction in the existing technology only considers clinical indicators, and the data sample feature points are relatively single, which often cannot accurately characterize the actual situation of the patient's current physiological and pathological changes, resulting in the low accuracy of the constructed prediction model. Summary of the invention

[0005] To solve the above problems, the present invention provides a method, system and medium for predicting the risk of peri-implantitis. The method constructs a peri-implantitis risk prediction model by considering metabolic factors and clinical factors, providing better technical support for assisting in judging the risk of peri-implantitis.

[0006] To achieve the above object, the present invention provides the following technical solutions.

[0007] A method for predicting the risk of peri-implantitis, comprising the following steps:

[0008] The metabolite components of gingival crevicular fluid samples from multiple patients were obtained, and the OPLS-DA analysis method, T test, and fold change method were used to perform differential analysis of metabolites, obtain VIP, P, and FC indicators, and determine the differential metabolites as metabolite factors;

[0009] The metabolite factors were converted into binary variables, the ROC curves of each differential metabolite and peri-implant disease were drawn, and the AUC values ​​of each differential metabolite were determined; metabolite factors with large differences were screened according to VIP, P, FC and AUC values ​​and the corresponding preset thresholds and ranges;

[0010] A plurality of baseline information of the patient is obtained as a plurality of clinical factors, and a plurality of clinical factors highly correlated with the health status around the implant are used as the clinical factors after screening;

[0011] A data matrix was constructed based on the binary variables of the screened metabolite factors and the hierarchical and categorical variables of the clinical factors; a training set was constructed based on the data matrix of multiple healthy group samples and peripheral disease group samples; a random forest model was constructed, and a prediction model for predicting peri-implant disease was obtained by training the training set.

[0012] Preferably, it also includes:

[0013] A test set is constructed based on the data matrix of multiple healthy group samples and peripheral disease group samples. Based on the prediction model, the IncMSE index and IncNodePurity index of metabolite factors and clinical related factors are tested through the test set, and the metabolite factors and clinical factors that rank at the top in terms of IncMSE index and IncNodePurity index are taken as important indicators reflecting peri-implant disease.

[0014] Preferably, the metabolite components of the gingival crevicular fluid sample are determined by UPLC-MS analysis.

[0015] Preferably, when constructing the training set, principal component analysis is used to determine the differences between healthy group samples and peripheral disease group samples, and outlier samples within a preset confidence interval threshold are eliminated.

[0016] Preferably, the determining of differential metabolites as metabolite factors comprises the following steps:

[0017] The metabolite data of the healthy group samples and peripheral disease group samples after outlier samples were removed were analyzed by OPLS-DA analysis, T test and fold change method, respectively, where metabolites with VIP>1, P<0.05, and metabolites with FC>3 or FC<1 / 3 were considered differential metabolites and used as metabolite factors;

[0018] Metabolite factors were transformed into dichotomous variables;

[0019] ROC curves of each differential metabolite and peri-implant disease were drawn, and the AUC values ​​of each differential metabolite were determined; metabolites with VIP>1, P<0.05, metabolites with FC>3 or FC<1 / 3, and multiple differential metabolites with the highest AUC values ​​were taken as the screened metabolite factors.

[0020] Preferably, the step of converting the metabolite factors into binary variables comprises the following steps:

[0021] The ROC curves of each metabolite and sample were obtained by taking the peri-implant health status of the sample as the dependent variable;

[0022] The Youden index value of the metabolite is extracted from the ROC curve, and the metabolite content of each sample is divided according to the Youden index. When the metabolite content is greater than the Youden index, the metabolite of the sample is classified as 2, and when the metabolite content is less than the Youden index, the metabolite is classified as 1;

[0023] The process is executed cyclically to obtain binary classification results of multiple metabolic factors in multiple samples.

[0024] Preferably, the screening of clinical factors comprises the following steps:

[0025] Obtain patient baseline information corresponding to healthy group samples and peripheral disease group samples after outlier samples are removed, and determine multiple clinical factors;

[0026] Identify categorical and ordinal variables in clinical factors;

[0027] Spearman correlation analysis was performed between each clinical factor and the health status around the implant, and multiple clinical factors with high correlation coefficients were determined as the clinical factors after screening.

[0028] Preferably, the patient's baseline information includes general information, general systemic condition, basic information of implants and restorations, specifically including age, gender, education level, whether suffering from cardiovascular disease, whether suffering from osteoporosis, whether suffering from diabetes, smoking habits, drinking habits, daily brushing times, rinsing habits, tartar index, soft plaque index, improved plaque index, periodontal condition, keratinized mucosa width, implant site, cause of natural tooth loss at the implant site, implant system, implant length, immediate implantation, bone grafting, gingival penetration method, abutment material, restoration method, retention method, restoration material, mesial EA, distal EA, mesial EP, distal EP, occlusal contact area, adjacent status and classification of opposing teeth.

[0029] Preferably, the determining of categorical variables and graded variables in clinical factors comprises the following steps:

[0030] The distribution of variables was evaluated to determine the categorical variables at the baseline level, including: gender, cardiovascular disease, osteoporosis, diabetes, smoking habits, drinking habits, mouthwashing habits, implant site, cause of natural tooth loss at the implant site, implant system, immediate implant, bone grafting, gingival penetration method, abutment material, restoration method, retention method, restoration material, and classification of opposing teeth;

[0031] Baseline level ordinal variables were determined, including: age, education level, number of times of daily toothbrushing, tartar index, soft plaque index, modified plaque index, periodontal status, keratinized mucosa width, implant length, mesial EA, distal EA, mesial EP, distal EP, occlusal contact area, and proximal status.

[0032] The present invention also provides a peri-implantitis risk prediction system, the system comprising:

[0033] processor;

[0034] a memory having stored thereon a computer program executable on the processor;

[0035] Wherein, when the computer program is executed by the processor, the steps of the method for predicting the risk of peri-implantitis are implemented.

[0036] The present invention also provides a computer-readable storage medium, on which a data processing program is stored. When the data processing program is executed by a processor, the steps of the method for predicting the risk of peri-implantitis are implemented.

[0037] Beneficial effects of the present invention:

[0038] The present invention proposes a method for predicting the risk of peri-implantitis, which considers the metabolic characteristics of patients to evaluate the actual situation of the patients' current physiological and pathological changes, combines the metabolic characteristics and clinical characteristics of patients, and uses OPLS-DA analysis method, T test and fold change method to perform differential analysis of metabolites, and draws ROC curves for each metabolite and diagnosing peri-implant disease, screens differential metabolites with greater differential contributions and clinical factors with higher correlation, and finally forms a data set that can characterize the significant characteristics of peri-implant disease; through data set construction and training, a random forest prediction model for peri-implant disease based on metabolites and clinical factors is obtained, which can assist in predicting the probability of patients developing peri-implant disease and provide a reference for subsequent treatment and diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 is a specific flow chart of an embodiment of the present invention;

[0040] Figure 2is a PCA score diagram of the peri-implant health status of 36 gingival sulcular fluid samples according to an embodiment of the present invention;

[0041] Figure 3 This is a distribution diagram of VIP values ​​of 20 differential metabolites between peri-implant disease and peri-implant health according to an embodiment of the present invention;

[0042] Figure 4 is a ROC curve diagram of 11 metabolites and peri-implant health status according to an embodiment of the present invention;

[0043] Figure 5 IncMSE and IncNodePurity ranking diagrams of some metabolites and clinical factors in the embodiments of the present invention;

[0044] Figure 6 This is an error rate diagram of the random forest prediction model for peri-implant disease based on metabolites and clinical factors according to an embodiment of the present invention;

[0045] Figure 7 The training set ROC of the random forest prediction model for peri-implant disease based on metabolites and clinical factors according to an embodiment of the present invention;

[0046] Figure 8 This is the ROC test set of the random forest prediction model for peri-implant disease based on metabolites and clinical factors in an embodiment of the present invention. DETAILED DESCRIPTION

[0047] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0048] Example 1

[0049] This embodiment proposes a method for predicting the risk of peri-implantitis. Figure 1 As shown, Figure 1 The flowchart of the method for predicting the risk of peri-implantitis is shown, which specifically includes the following steps:

[0050] S1: Gingival crevicular fluid samples were obtained from multiple patients, and the metabolite composition was determined by UPLC-MS analysis. The gingival crevicular fluid samples were divided into multiple healthy group samples and peripheral disease group samples.

[0051] S2: Principal component analysis is used to determine the differences between healthy group samples and peripheral disease group samples, and outlier samples within the preset confidence interval threshold are eliminated.

[0052] In this example, UPLC-MS analysis was performed on 36 gingival crevicular fluid samples. The samples were grouped according to peri-implant health and peri-implant disease. There were 12 samples in the healthy group and 24 samples in the peri-implant disease group. PCA analysis showed that the distribution of the two groups of samples had certain differences, such as Figure 2 As shown, Figure 2 Figure 2 is the PCA score diagram of peri-implant health status of 36 gingival sulcular fluid samples. PD is the peri-implant disease group, HI is the peri-implant health group. Sample Y212 was obviously outlier and was outside the 95% confidence interval, so it was excluded.

[0053] S3: The metabolite data of the healthy group samples and peripheral disease group samples after elimination were analyzed by OPLS-DA analysis method, T test and fold change method, respectively. Metabolites with VIP>1, P<0.05, and metabolites with FC>3 or FC<1 / 3 were considered differential metabolites and taken as metabolite factors.

[0054] In the two groups of samples in this example, there were 166 metabolites with VIP>1 and P<0.05, 20 metabolites with VIP>3 and P<0.01, and 3 metabolites with FC>3.

[0055] S4: Obtain the patient baseline level information corresponding to the healthy group samples and peripheral disease group samples after elimination, and determine multiple clinical factors.

[0056] Baseline information of 35 patients was collected. The data came from the epidata database constructed when the prospective cohort was established, mainly including the general information of the patients, general conditions, basic information of implants and restorations, etc. A total of 33 variables were collected in this part of the study, including age, gender, education level, cardiovascular disease, osteoporosis, diabetes, smoking habits, drinking habits, daily brushing times, mouthwash habits, tartar index, soft plaque index, modified plaque index, periodontal condition, keratinized mucosa width, implant site, cause of natural tooth loss at the implant site, implant system, implant length, immediate implant, bone grafting, gingival penetration, abutment material, restoration method, retention method, restoration material, mesial EA, distal EA, mesial EP, distal EP, occlusal contact area, adjacent status, and classification of opposing teeth.

[0057] Thus, an original data matrix was constructed with 166 differential metabolites and 33 clinical factors of 35 patients as independent variables and the peri-implant health status diagnosis results as independent variables.

[0058] S5: Identify categorical and ordinal variables in clinical factors and transform metabolite factors into binary variables.

[0059] Variable inspection and evaluation: The original data matrix was checked and no missing values ​​or abnormal variables were found. The distribution of variables was evaluated. Among the 33 clinically related factors at the baseline level, 18 categorical variables existed, including gender, cardiovascular disease, osteoporosis, diabetes, smoking habits, drinking habits, mouthwashing habits, implant site, cause of natural tooth loss at the implant site, implant system, immediate implant, bone grafting, gingival penetration, abutment material, restoration method, retention method, restoration material, and classification of opposing teeth; there were 15 ordinal variables, namely: age, education level, number of daily brushing times, tartar index, soft plaque index, improved plaque index, periodontal status, keratinized mucosa width, implant length, mesial EA, distal EA, mesial EP, distal EP, occlusal contact area, and adjacent status. Metabolite factors were the content of each differential metabolite in the 35 samples, which were continuous variables.

[0060] Metabolic factors were converted into binary variables: The metabolite factors consisted of the contents of 166 differential metabolites in 35 samples. The data was large and complex. To facilitate model construction, the metabolite contents were converted from continuous variables to binary variables. The conversion process was based on the R language tidyverse and ROCR packages, and the threshold segmentation of each metabolite content in each sample was performed according to the Youden index. First, the peri-implant health status of sample K01 was used as the dependent variable to obtain the ROC curve of a certain metabolite and K01. Then, the Youden index value of the metabolite was extracted from the curve, and the metabolite content of each sample was segmented according to the Youden index. When the metabolite content was greater than the Youden index, the metabolite of the sample was classified as 2. When the metabolite content was less than the Youden index, the metabolite was classified as 1. The loop statement was run to obtain the binary classification results of the 166 metabolic factors in the 35 samples in turn.

[0061] After preprocessing, the original data matrix is ​​transformed into a data matrix constructed by rank variables and categorical variables, which needs to be further optimized through variable screening.

[0062] S6: Draw the ROC curves of each differential metabolite and peri-implant disease, and determine the AUC value of each differential metabolite; metabolites with VIP>1, P<0.05, metabolites with FC>3 or FC<1 / 3, and the 10 differential metabolites with the highest AUC values ​​were taken as the screened metabolite factors.

[0063] Differential metabolite screening: Based on the screening of 166 differential metabolites, 20 differential metabolites with VIP>3 and t-test P<0.01 in OPLS-DA analysis were first selected, such as Figure 3 As shown in Table 1, three metabolites with FC>3 or FC<1 / 3 were added. At the same time, the ROC curves of each differential metabolite and peri-implant disease were drawn, as shown in Figure 4As shown, the 10 metabolites with the highest AUC values ​​were taken. The AUCs of the 10th and 11th metabolites were the same, but 4 of the 11 had VIP>3 and P<0.01, so a total of 30 metabolite factors were formed.

[0064] Table 1 20 different metabolites between peri-implant disease and peri-implant health

[0065]

[0066] S7: Spearman correlation analysis was performed between each clinical factor and the health status around the implant, and 20 clinical factors with high correlation coefficients were determined as the clinical factors after screening.

[0067] Clinical factor screening: Spearman correlation analysis was performed between each factor and the peri-implant health status. The results are shown in Table 2. Among them, education level, gingival penetration method, and periodontal status were correlated with the peri-implant health status. Based on the previous research results of the research group, the significant factors of the peri-implant disease in the cohort population were analyzed by univariate and multivariate analysis. Combined with clinical experience, 17 factors including gender, age, cardiovascular disease, diabetes, smoking, drinking, cause of natural tooth loss at the implant site, bone grafting, restoration type, keratinized mucosa width, adjacent status, occlusal contact area, mouthwash habits, tartar index, soft plaque index, retention method, and simplified oral hygiene index were included, and finally 20 clinical related factors were formed. The clinical factors of 35 patients were formed into an Excel table. After data preprocessing and variable screening, the implant health status, 20 clinical factors, and 30 metabolic factors finally formed 51 columns, and the factor name and 35 samples formed 36 rows, forming a 51-column and 36-row modeling data matrix. In order to compare the prediction performance of the prediction model constructed with clinical factors alone, a clinical factor data matrix of 20 clinical factors of 35 samples was constructed.

[0068] Table 2 Spearman correlation analysis between clinical factors and peri-implant health status

[0069]

[0070] *The correlation is significant at the 0.05 level (two-tailed): **The correlation is significant at the 0.01 level (two-tailed)

[0071] S8: A data matrix was constructed based on the binary variables of the metabolite factors after screening and the hierarchical variables and categorical variables of the clinical factors; a training set was constructed based on the data matrix of multiple healthy group samples and peripheral disease group samples.

[0072] S9: Construct a random forest model and obtain a prediction model for peri-implant disease prediction through training with the training set.

[0073] S10: A test set was constructed based on the data matrix of multiple healthy group samples and peripheral disease group samples. Based on the prediction model, the IncMSE index and IncNodePurity index of metabolite factors and clinical related factors were tested through the test set. The metabolite factors and clinical factors that ranked top in IncMSE index and IncNodePurity index were taken as important indicators reflecting peri-implant disease.

[0074] The random forest model was constructed using the tidyverse and randomForest packages of R software, and the data set was divided into a training set and a validation set in a ratio of 6:4. In the random forest prediction model of peri-implant disease based on metabolites and clinical factors, some metabolites and clinical factors showed a certain importance, such as Figure 5 As shown. The top 6 factors of IncMSE relative importance are polypeptide substances proline-hydroxyproline (pos-57), isoleucine-isoleucine (neg-10683), threonine-asparagine-valine-leucine (pos-11784), phenylalanine-isoleucine (pos-2532), periodontal status, valine-leucine-serine-aspartic acid (neg-11676). The top 6 factors of IncNodePurity are threonine-asparagine-valine-leucine (pos-11784), periodontal status, isoleucine-isoleucine (neg-10683), proline-proline (pos-2897), proline-hydroxyproline (pos-57), and occlusal contact area.

[0075] Error probability plot in random forest model Figure 6 In the above figure, as the number of decision trees in the model increases, the prediction error rate stabilizes at around 0.25. The training set AUC of the random forest prediction model for peri-implant disease is 1, as shown in Figure 7 , which has good classification performance.

[0076] The model performance was evaluated in the validation set. The precision of the model in the validation set was 0.889, that is, among the samples predicted to be peri-implant disease, the probability of actually suffering from peri-implant disease was 88.9%; the recall rate was 1.000, that is, all samples that actually suffered from peri-implant disease were successfully predicted. The F1 score was 0.471, the precision and recall rates were generally balanced, and the AUC was 1. Figure 8 , the model has good prediction performance.

[0077] In the present invention, peri-implant disease, as the most common complication after implant restoration, will have a great negative impact on the implant effect, causing pain and economic burden for patients. If the probability of peri-implant disease can be predicted based on the patient's metabolic characteristics and clinical features before surgery, timely intervention can be made for the patient, and a personalized maintenance treatment plan can be formulated for the patient, it will be beneficial to improve the treatment effect of implant restoration and reduce the occurrence of peri-implant disease.

[0078] The present invention has conducted a preliminary exploration of establishing a peri-implant disease risk prediction model that combines clinical factors of peri-implant disease with omics data, and proposed an optimized prediction model. For the data preprocessing of omics and clinical factors before model building, since omics data is high-dimensional and complex compared to clinical data, it is very important to properly combine the two types of data when building the model. In combination with research practice, the present invention processes two different types of data separately.

[0079] For the inclusion of metabolic factors, since there are as many as 166 differential metabolites between the two groups (OPLS-DAVIP>1, t-test P<0.05), it is very important to select representative differential metabolites from them. The present invention considers VIP value, P value, FC value, and AUC value, and further screens the differential metabolites. In the multivariate statistical analysis method, the VIP value of OPLS-DA can measure the explanatory power and influence intensity of the distribution difference of specific metabolites on the sample grouping. The larger the VIP value of the metabolite, the greater its contribution to the difference. In the unit statistical analysis, the present invention lowers the significant difference value, and when P<0.01, it is determined that the difference in the distribution of metabolites between the two groups is statistically significant. Using the method of combining VIP value with P value, 20 metabolites with VIP>3 and P<0.01 were screened. At the same time, in the analysis, we noticed that the difference multiple FC values ​​of some differential metabolites between the two groups were very large, so the present invention supplemented the factors of FC>3 and FC<1 / 3. For the construction of the model, the ROC analysis of disease diagnosis by a single factor is of great significance. The present invention draws the ROC curves of each metabolite and the diagnosis of peri-implant disease, and finds that some metabolites have good performance as classifiers for peri-implant disease, so the metabolites with the top ten AUC rankings (AUC≥0.833) are further included to form a total of 30 metabolite metabolic factors. Since the data of the metabolite content of each sample is a continuous variable, it is not intuitive to construct a classification prediction model, so the Youden index is used to segment the metabolite content after screening out the metabolite factors, and the omics data is converted into a binary variable score by a dimensionality reduction method. The screened clinical factors and metabolite factor scores are merged into the model as covariates for modeling. This data preprocessing method takes into account the characteristics of omics data and clinical data as much as possible, but may ignore the correlation or interaction structure between different types of data. In this case, the predictive ability of omics data may be overestimated. In the present embodiment, the decision tree model constructed based on metabolites and clinical factors has only one branch after debugging in a variety of ways, which may indicate that the classification ability of omics data after data conversion is too strong, covering up other factors.

[0080] In the random forest prediction model of peri-implant disease based on metabolites and clinical factors, some metabolites and clinical factors showed a certain importance. The factors with the top predictor importance indicators IncMSE and IncNodePurity included metabolite peptide substances threonine-asparagine-valine-leucine, isoleucine-isoleucine, proline-proline, glutamyl-glutamine, and clinical factors such as periodontal status, implant penetration, and occlusal contact area. The optimized random forest model is composed of peptide substances threonine-asparagine-valine-leucine and isoleucine-isoleucine, suggesting that we can use targeted metabolomics research methods for peptides and amino acids when using the model to predict peri-implant disease in the future.

[0081] This invention proposes for the first time the combined use of metabolic factors and clinical indicators to predict the occurrence of peri-implant disease, optimizes the model, and constructs a model for predicting peri-implant disease using two metabolites, which has certain exploratory significance and practical value.

[0082] The above is a peri-implantitis risk prediction method provided by one embodiment of this embodiment. Based on the same idea, this embodiment also provides a corresponding peri-implantitis risk prediction system. Each module in the above peri-implantitis risk prediction system can be implemented in whole or in part through software, hardware and a combination thereof. The above modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0083] This embodiment also provides a computer-readable storage medium, which stores a computer program, which can be used to execute the above Figure 1 Provided is a method for predicting the risk of peri-implantitis.

[0084] Those of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application may include at least one of non-volatile and volatile memory. Non-volatile memory may include read-only memory (ROM), magnetic tape, floppy disk, flash memory or optical memory, etc. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0085] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for predicting the risk of peri-implantitis, characterized in that: The following steps are involved: The metabolite components of gingival crevicular fluid samples from multiple patients were obtained, and the OPLS-DA analysis method, T test, and fold change method were used to perform differential analysis of metabolites, obtain VIP, P, and FC indicators, and determine the differential metabolites as metabolite factors; The metabolite factors were converted into binary variables, the ROC curves of each differential metabolite and peri-implant disease were drawn, and the AUC values ​​of each differential metabolite were determined; metabolite factors with large differences were screened according to VIP, P, FC and AUC values ​​and the corresponding preset thresholds and ranges; A plurality of baseline information of the patient is obtained as a plurality of clinical factors, and a plurality of clinical factors highly correlated with the health status around the implant are used as the clinical factors after screening; A data matrix was constructed based on the binary variables of the screened metabolite factors and the hierarchical and categorical variables of the clinical factors; a training set was constructed based on the data matrix of multiple healthy group samples and peripheral disease group samples; a random forest model was constructed, and a prediction model for predicting peri-implant disease was obtained by training the training set.

2. The method for predicting the risk of peri-implantitis according to claim 1, characterized in that: Also includes: A test set is constructed based on the data matrix of multiple healthy group samples and peripheral disease group samples. Based on the prediction model, the IncMSE index and IncNodePurity index of metabolite factors and clinical related factors are tested through the test set, and the metabolite factors and clinical factors that rank at the top in terms of IncMSE index and IncNodePurity index are taken as important indicators reflecting peri-implant disease.

3. The method for predicting the risk of peri-implantitis according to claim 1, characterized in that: The metabolite components of the gingival crevicular fluid samples were determined by UPLC-MS analysis.

4. The method for predicting the risk of peri-implantitis according to claim 1, characterized in that: When constructing the training set, principal component analysis is used to determine the differences between healthy group samples and peripheral disease group samples, and outlier samples within a preset confidence interval threshold are eliminated.

5. The method for predicting the risk of peri-implantitis according to claim 4, characterized in that: The step of determining the differential metabolites as metabolite factors comprises the following steps: The metabolite data of the healthy group samples and peripheral disease group samples after outlier samples were removed were analyzed by OPLS-DA analysis, T test and fold change method, respectively, where metabolites with VIP>1, P<0.05, and metabolites with FC>3 or FC<1 / 3 were considered differential metabolites and used as metabolite factors; Metabolite factors were transformed into dichotomous variables; ROC curves of each differential metabolite and peri-implant disease were drawn, and the AUC values ​​of each differential metabolite were determined; metabolites with VIP>1, P<0.05, metabolites with FC>3 or FC<1 / 3, and multiple differential metabolites with the highest AUC values ​​were taken as the screened metabolite factors.

6. The method for predicting the risk of peri-implantitis according to claim 5, characterized in that: The step of converting metabolite factors into binary variables includes the following steps: The ROC curves of each metabolite and sample were obtained by taking the peri-implant health status of the sample as the dependent variable; The Youden index value of the metabolite is extracted from the ROC curve, and the metabolite content of each sample is divided according to the Youden index. When the metabolite content is greater than the Youden index, the metabolite of the sample is classified as 2, and when the metabolite content is less than the Youden index, the metabolite is classified as 1; The process is executed cyclically to obtain binary classification results of multiple metabolic factors in multiple samples.

7. The method for predicting the risk of peri-implantitis according to claim 5, characterized in that: The screening of clinical factors comprises the following steps: Obtain patient baseline information corresponding to healthy group samples and peripheral disease group samples after outlier samples are removed, and determine multiple clinical factors; Identify categorical and ordinal variables in clinical factors; Spearman correlation analysis was performed between each clinical factor and the health status around the implant, and multiple clinical factors with high correlation coefficients were determined as the clinical factors after screening.

8. The method for predicting the risk of peri-implantitis according to claim 7, characterized in that: The patient's baseline information includes general information, general systemic condition, basic information on implants and restorations, including age, gender, education level, whether suffering from cardiovascular disease, osteoporosis, diabetes, smoking habits, drinking habits, daily brushing times, rinsing habits, tartar index, soft plaque index, improved plaque index, periodontal condition, keratinized mucosa width, implant site, cause of natural tooth loss at the implant site, implant system, implant length, immediate implant, bone grafting, gingival penetration method, abutment material, restoration method, retention method, restoration material, mesial EA, distal EA, mesial EP, distal EP, occlusal contact area, adjacent status and classification of opposing teeth.

9. The method for predicting the risk of peri-implantitis according to claim 7, characterized in that: The step of determining the categorical variables and the graded variables in the clinical factors comprises the following steps: The distribution of variables was evaluated to determine the categorical variables at the baseline level, including: gender, cardiovascular disease, osteoporosis, diabetes, smoking habits, drinking habits, mouthwashing habits, implant site, cause of natural tooth loss at the implant site, implant system, immediate implant, bone grafting, gingival penetration method, abutment material, restoration method, retention method, restoration material, and classification of opposing teeth; Baseline level ordinal variables were determined, including: age, education level, number of times of daily toothbrushing, tartar index, soft plaque index, modified plaque index, periodontal status, keratinized mucosa width, implant length, mesial EA, distal EA, mesial EP, distal EP, occlusal contact area, and proximal status.

10. A peri-implantitis risk prediction system, characterized in that: The system comprises: processor; a memory having stored thereon a computer program executable on the processor; Wherein, when the computer program is executed by the processor, the steps of the method for predicting the risk of peri-implantitis as described in any one of claims 1 to 9 are implemented.

Images