International Journal of Molecular Medical Science 2025, Vol.15 http://medscipublisher.com/index.php/ijmms © 2025 MedSci Publisher, registered at the publishing platform that is operated by Sophia Publishing Group, founded in British Columbia of Canada. All Rights Reserved. -
International Journal of Molecular Medical Science 2025, Vol.15 http://medscipublisher.com/index.php/ijmms © 2025 MedSci Publisher, registered at the publishing platform that is operated by Sophia Publishing Group, founded in British Columbia of Canada. All Rights Reserved. MedSci Publisher is an international Open Access publisher specializing in disease therapy, clinical pharmacology, clinical biochemistry, vaccines, immunology, microbiology at the publishing platform that is operated by Sophia Publishing Group (SPG), founded in British Columbia of Canada. Publisher MedSci Publisher Editedby Editorial Team of International Journal of Molecular Medical Science Email: edit@ijmms.medscipublisher.com Website: http://medscipublisher.com/index.php/ijmms Address: 11388 Stevenston Hwy, PO Box 96016, Richmond, V7A 5J5, British Columbia Canada International Journal of Molecular Medical Science (ISSN 1927-6656) is an open access, peer reviewed journal published online by MedSci Publisher. The journal publishes scientific articles like original research articles, case reports, review articles, editorials, short communications and correspondence of the high quality pertinent to all aspects of human biology, pathophysiology and molecular medical science, including genomics, transcriptomics, proteomics, metabolomics of disease therapy, clinical pharmacology, clinical biochemistry, vaccines, immunology, microbiology, epidemiology, aging, cancer biology, infectious diseases, neurological diseases and myopathies, stem cells and regenerative medicine, vascular and cardiovascular biology, as well as the important implications for human health and clinical practice research. All the articles published in International Journal of Molecular Medical Science are Open Access, and are distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. MedSci Publisher uses CrossCheck service to identify academic plagiarism through the world’s leading plagiarism prevention tool, iParadigms, and to protect the original authors’ copyrights.
International Journal of Molecular Medical Science (online), 2025, Vol. 15, No. 2 ISSN 1927-6656 http://medscipublisher.com/index.php/ijmms © 2025 MedSci Publisher, registered at the publishing platform that is operated by Sophia Publishing Group, founded in British Columbia of Canada. All Rights Reserved. Latest Content The Predictive Model For The Risk of Venous Thromboembolism In Children: A Systematic Review of and Meta-Analysis Bochen Wang, Zixuan Liu, Yang Li, Xinqi Shi International Journal of Molecular Medical Science, 2025, Vol. 15, No. 2, 54-68 Research Progress on the Main Active Ingredients and Pharmacological Mechanisms of Food-Medicine Homologous Substances YanLou International Journal of Molecular Medical Science, 2025, Vol. 15, No. 2, 69-79 Study on the Role and Molecular Mechanisms of Yam Polysaccharides in Pre-Diabetes Intervention Lihui Xu, Tingting Feng, Keyan Fang International Journal of Molecular Medical Science, 2025, Vol. 15, No. 2, 80-88 Structural Characterization and Immunomodulatory Functions of Cordyceps Polysaccharides Yinchen Zhao, Haomin Chen International Journal of Molecular Medical Science, 2025, Vol. 15, No. 2, 89-97 The Study on the Impact of the Ingredient Ratio of Compound Dietary Fiber Preparations on Gut Microecological Regulation Function Huixian Li, Jingqiang Wang International Journal of Molecular Medical Science, 2025, Vol. 15, No. 2, 98-106
International Journal of Molecular Medical Science, 2025, Vol.15, No.2, 54-68 http://medscipublisher.com/index.php/ijmms 54 Research Report Open Access The Predictive Model For The Risk of Venous Thromboembolism In Children: A Systematic Review of and Meta-Analysis Bochen Wang, Zixuan Liu, Yang Li , Xinqi Shi School of Nursing, Peking Union Medical College, Beijing, 100144, Beijing, China Corresponding author: liyang3413@sina.com International Journal of Molecular Medical Science, 2025, Vol.15, No.2 doi: 10.5376/ijmms.2025.15.0006 Received: 28 Nov., 2024 Accepted: 19 Feb., 2025 Published: 02 Mar., 2025 Copyright © 2025 Wang et al., This is an open access article published under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. Preferred citation for this article: Wang B.C., Liu Z.X., Li Y., and Shi X.Q., 2025, The predictive model for the risk of venous thromboembolism in children: a systematic review of and meta-analysis, International Journal of Molecular Medical Science, 15(2): 54-68 (doi: 10.5376/ijmms.2025.15.0006) Abstract Through systematic review and meta-analysis, this study systematically evaluated the predictive efficacy and heterogeneity of the risk prediction model for venous thromboembolism (VTE) in children, providing evidence-based evidence for the optimization and application of the clinical model. A search of the Chinese and English databases (as of April 16, 2024) included 13 studies. The results showed that the current model was dominated by Logistic regression, central venous catheter (CVC) was the most commonly used predictor (inclusion rate 61.5%), and the combined AUC was 0.84 (95% CI: 0.80-0.88). However, heterogeneity in the included models was significant (I²=96%), especially when the predictors exceeded 5, mainly due to differences in variable selection (e.g., regional differences in medical practice, unadjusted for age stratification effects). Seven studies had a high risk of bias, which was mainly reflected in the opacity of missing data processing and the absence of blind predictors and outcome variables. In this study, a "core + extended" model architecture was proposed for the first time: age-specific thresholds for biomarkers (e.g., D-dimer, CRP) were vertically integrated, and cross-regional core variable sets (≤5 items, e.g., CVC, age, surgery) were constructed horizontally to balance prediction accuracy with clinical universality. In the future, the application value of dynamic biomarkers and machine learning algorithms in children's VTE stratification should be verified by multi-center cohort. Keywords Children; Venous thromboembolism; Prediction model; Systematic evaluation; meta-Analysis 1 Background Venous thromboembolism (VTE) refers to the formation of blood clots, and thrombi or a part of a thrombus detaches and causes embolism in the veins, resulting in partial or complete blockage of the blood vessels. The blockage leads to impaired venous return, which includes deep vein thrombosis and pulmonary embolism (Tan and Zhi, 2023). According to a study conducted by the American Academy of Pediatrics, the hospitalization rate of pediatric VTE has increased by 130% from 2008 to 2019 (from 46 per 10 000 cases to 106 per 10 000 cases) (O’Brien et al., 2022). There are no typical clinical features of VTE on children in the early stage. Delay in diagnosis and treatment poses a great health risk to children, increasing the chance of pulmonary embolism, which could lead to higher mortality rate, increased risk of VTE recurrence and the development of post-thrombotic syndrome (White et al., 2021). Most of the current risk prediction models for venous thromboembolism are used in adults. However, according to the American College of Chest Physicians’ “Antithrombotic Therapy and Prevention of Thrombosis in Neonates and Children Guidelines” published in February 2012, it is known that pediatric patients (defined as ≤21 years old according to the seventh edition of Pediatrics Nursing) differ from adult patients in epidemiology, pharmacokinetics, coagulation system, and comorbidities, resulting in lower applicability of adult VTE risk prediction models to children (Monagle et al., 2012; Cui and Zhang, 2021). At present, risk assessment for adult patients is more mature. The application of adult VTE risk prediction models is also considered to significantly reduce the incidence of VTE in adults (Biss, 2016). However, for children, there is a lack of reliable methods for preventing VTE. Currently, the occurrence of VTE in children largely depends on pediatricians' high suspicion, and there is a lack of practical risk assessment tools in clinical practice (Yang and Hao, 2020; Walker et al., 2023).
International Journal of Molecular Medical Science, 2025, Vol.15, No.2, 54-68 http://medscipublisher.com/index.php/ijmms 55 In the “American College of Chest Physicians Evidence-Based Clinical Practice Guidelines”, it is suggested to use chemoprophylaxis for children at risk of VTE, such as using low molecular weight heparin drugs for thrombosis prevention, by Chalmers et al. (2011) and others. However, the safety and efficacy of this chemical prevention are still controversial (Chalmers et al., 2011; Rühle and Stoll, 2018). Therefore, it is crucial to establish high-quality pediatric venous thromboembolism risk prediction model. At present, the children's VTE risk prediction model is rarely used clinically, with a variety of research subjects and poor homogeneity (Yang et al., 2022). Therefore, the present study focused on venous thromboembolism (VTE) in children, aiming to explore new influencing factors for VTE and analyze differences in the occurrence of VTE in different regions through systematic review and meta-analysis of published risk prediction models. At the same time, the predictive ability of the combined AUC evaluation model provides a strong basis for the selection and development of relevant models for clinical medical staff. 2Methods The research protocol has been registered in PROSPERO (registration number: CRD42024537499). 2.1 Literature search strategy The search databases are Wanfang, CNKI, CSTJ, PubMed, Web of Science, Embase, and Cochrane Library. The search period is from the establishment of the database to April 16, 2024. The English search terms include: child, pediatric, venous thromboembolism, deep vein thrombosis, pulmonary embolism, predict, predictive model, predictive score, risk factor, risk prediction, risk assessment, model. The search strategy is based on PubMed, as shown in Figure 1. The literature was searched by 2 researchers, and if there was a disagreement, the 2 researchers would discuss it, and if necessary, a third researcher would make the judgment. Figure 1 PubMed retrieval mode In addition, we used the PICOTS system in the systematic review to help us define the purpose of this study, the search strategy, and the inclusion and exclusion criteria. The main items of the systematic review are described as follows: P (Population): Pediatric patients with venous thromboembolism aged ≤21 years. I (Intervention model): Developed and published pediatric venous thromboembolism risk prediction model. C (Comparator): No competition model. Outcome: The research findings indicate the formation of venous thromboembolism. T (Timing): Basic information at admission, results of clinical scoring scales, and laboratory-based predictive indicators. S (Setting): The expected use of the risk prediction model is to predict the risk of venous thromboembolism in children, improve the prediction rate of VTE in children by clinical staff, and reduce the incidence of VTE in children. 2.2 Inclusion and exclusion criteria The inclusion criteria for the study are: ① Study types: cohort studies and case-control studies; ② Study population: children, aged ≤21 years, including premature infants; ③Study content: construction of the risk prediction model for pediatric venous thromboembolism.
International Journal of Molecular Medical Science, 2025, Vol.15, No.2, 54-68 http://medscipublisher.com/index.php/ijmms 56 Exclusion criteria: ①Non-Chinese or English language literature; ②Studies that have not established risk prediction models; ③Repetitive publications, such as those published in both conferences and journals; ④ Incomplete data, such as missing methods for constructing models or lack of outcome indicator data. 2.3 Literature screening and data extraction Two researchers independently conducted literature screening and data extraction, cross-checked each other. Initially, duplicate literature was removed, then preliminary screening was done based on the title and abstract of the literature, finally, inclusion and exclusion criteria were introduced to evaluate and extract data from the full text. If there were any disagreements between the two researchers in the literature screening, a third researcher would be involved in the discussion to reach a consensus. The extracted data information includes the author, publication year, research design, study population, data source, sample size, variable selection method, model development method, model validation type, model performance measure, handling of missing data, and treatment method for continuous variables. 2.4 Quality evaluation The two researchers will independently assess the risk of bias in the studies included using the PROBAST tool (Wolff et al., 2019). In case of discrepancies in the assessment of bias between the two researchers, they will resolve it through discussion, and if necessary, the two researchers will discuss it together with a third researcher. 2.5 Data synthesis and statistical analysis In R4.3.1 software, the "meta" package is mainly used for meta-analysis operations. First, the data included in the study was imported into the R environment in a standardized format, including the basic information of the study, the AUC and its standard error and other key data. The meta-analysis was performed using the "metagen" function, which can set parameters for different data types and study designs. Heterogeneity was assessed using I² values and P values. When P<0.05, there was no significant heterogeneity among the studies. I²>75%, 25-75% and <25% represent high, medium and low heterogeneity levels, respectively. If P>0.1 and I²<50%, the heterogeneity of each study was acceptable, and the fixed-effect model was used for meta-analysis. On the contrary, if P≤0.1 and I²≥50%, it indicates that there is a large heterogeneity among the studies. Then, the random effects model is used for meta-analysis and sensitivity analysis is conducted to find the source of heterogeneity. Rigorous treatment strategies were employed to address potentially missing data from included studies. Firstly, the missing data is randomly processed by using the multiple interpolation method in the "mouse" package. The input data set of missing values is generated by several simulations, and the missing values are estimated by combining the known data information. For the non-random missing data whose mechanism is difficult to determine, the relevant data are carefully excluded without affecting the reliability of the research conclusions, and the possible impact of this processing method is explained in the research results and discussion. 3Results 3.1 Literature selection process and results A total of 3 056 relevant literatures were obtained, and 13 literatures (Sharathkumar et al., 2012; Kerlin et al., 2015; Connelly et al., 2016; Marquez and Shabanova et al., 2016; Spavor et al., 2016; Yen et al., 2016; Cairo et al., 2018; Kerris et al., 2020; Jaffray et al., 2021; Walker et al., 2021; Jaffray et al., 2022; Papillon et al., 2023; Tiratrakoonseree et al., 2024) were finally included after screening and re-screening. The literature screening process and results are shown in Figure 2. 3.2 Research characteristics Table 1 summarizes the basic characteristics of the 13 included studies. These studies were published between 2012 and 2024, primarily conducted in the United States, with only two conducted in other countries, namely Canada and Thailand. Among the included studies, there were 11 retrospective cohort studies and 2 prospective cohort studies. The earliest VTE study was conducted by Sharathkumar in 2012, with the most publications in
International Journal of Molecular Medical Science, 2025, Vol.15, No.2, 54-68 http://medscipublisher.com/index.php/ijmms 57 2016, totaling 4. In terms of study populations, 5 studies focused on pediatric patients, 3 on pediatric critically ill patients, 3 on pediatric trauma patients, 1 on pediatric cancer patients, and 1 on pediatric surgical patients. With regards to the source of study populations, 3 were from single-center studies and 10 were from multi-center studies. In terms of outcome prediction, 10 studies were on VTE, 2 on DVT, and 1 on PE. The total sample size ranged from 104 to 536 423 cases. Table 2 summarizes the predictive model information from the 13 included studies. Three studies reported missed data, one study was unclear about missing data, and three studies reported methods for dealing with missing data. Logistic regression analysis was used for model construction in all 13 studies. Eight studies converted continuous variables into categorical variables. The most commonly used predictor in all models was CVC, which was included as the final predictor in 8 studies. Other commonly used predictors included age and surgery, both of which were used in 6 studies. The reported AUC or C statistics ranged from 0.67 to 0.954 4, with some studies reporting positive predictive value, negative predictive value, sensitivity, and specificity. Out of the 13 studies, 7 conducted calibration, with the Hosmer-Lemeshow test being the most commonly used method, employed 5 times. Figure 2 Document screening flow chart 3.3 Model validation The included 13 studies have all undergone internal or external validation, with 12 studies undergoing internal validation and 1 study undergoing external validation. 3.4 Quality evaluation results According to the PROBAST tool, Table 3 summarizes the applicability and risk of bias of the included studies. In terms of applicability studies, all studies in various fields and overall were rated as low risk. In the field of the study subjects, all included studies were deemed low risk; in the predictor domain, all included studies were deemed unclear risk of bias, primarily due to not specifying whether predictor variables were assessed without knowledge of the outcome; in the outcome domain, all included studies were deemed unclear risk of bias, mainly because it was not stated whether outcomes were determined without knowledge of predictor variables; in the statistical analysis domain, 7 studies were rated as high risk, with 5 studies not mentioning how missing data were handled specifically, and 3 studies not explicitly stating whether overfitting and underfitting were considered. During the quality evaluation process, it was found that the majority of studies did not mention how to handle missing data, leading to an increased risk of bias. This result suggests that model developers need to be more rigorous and detailed when designing and describing research processes.
International Journal of Molecular Medical Science, 2025, Vol.15, No.2, 54-68 http://medscipublisher.com/index.php/ijmms 58 Table 1 Basic characteristics of the 13 studies included Author (Year) Country Type of study Object of Study Data sources Predicted outcomes Sample size Sharathkumar et al. (2012) America Retrospective Study Child Patient, ≤Age 20 Two tertiary care pediatric hospitals VTE 519 Yen et al. (2016) America Retrospective Study Child Patient with Trauma, ≤Age 21 John Hopkins Preliminary Adult and Child Trauma Center VTE 17366 Cairo et al. (2018) America Retrospective Study Child Operation Patient, ≤Age 18 NSQIP-P Database VTE 218432 Marquez et al. (2016) America Prospective Cohort Study Child Patient with Severe Disease, <Age 18 Three Children’s Hospital located in the northeast of America DVT 175 Kerris et al. (2020) America Retrospective Study Child Patient with Severe Heart Disease, ≤Age 18 Tertiary care referral to the Children's Hospital cardiac ICU VTE 2204 Jaffray et al. (2021) America Retrospective Study Child Patient, ≤Age 21 Children's Hospital-Acquired Thrombosis Registry VTE 1567 Kerlin et al. (2015) America Retrospective Study Child Patient, ≤Age 18 Monocentric VTE 389 Papillon et al. (2023) America Retrospective Study Child Patient with Trauma, ≤Age 18 Trauma Quality Improvement Program(TQIP)Database VTE 347576 Spavor et al. (2016) Canada Retrospective Study Child Cancer Patient, ≤Age 18 Multi-center Study DVT 218 Walker et al. (2021) America Retrospective Study Child Patient, ≤Age 21 Little Monroe-Carrier Children's Society Hospital VTE 111352 Jaffray et al. (2022) America Retrospective Study Child Patient with Severe Disease, ≤Age 21 Multi-institutional Children's School Hospital-Acquired Thrombosis Registry VTE 735 Connelly et al. (2016) America Retrospective Study Child Patient with Trauma, ≤Age 21 National Trauma Database VTE 536423 Tiratrakoonseree et al. (2024) America Retrospective Study Child Patient, ≤Age 20 Two tertiary care pediatric hospitals VTE 519 Note: VTE stands for Venous Thromboembolism; DVT stands for Deep Vein Thrombosis; PE stands for Pulmonary Embolism
International Journal of Molecular Medical Science, 2025, Vol.15, No.2, 54-68 http://medscipublisher.com/index.php/ijmms 59 Table 2 Modeling of the 13 studies included Author (Year) Missing data processing Model development method Calibration method Proof technique Number of predictor Ultimate predictor Model performance Sharathkumar et al. (2012) No information is mentioned Logistic Regression - Internal Verification 6 Blood flow infection, CVC, admission to ICU or NICU, hospitalization ≥7 days, immobilization >72 h, avoid using pregnancy drugs AUC 0.852, the Sensitivity was 70.0%, specificity was 80.0%, the Positive Predictive Value was 2.5%, and the Negative Predictive Value was 99.7% Yen et al. (2016) No information is mentioned Logistic Regression HosmerLemeshow Internal Verification 6 Age, GCS, ISS, Intubation, Blood Transfusion, Major Surgery AUC 0.91, the Sensitivity was 87.0% and the Specificity was 81.0% Cairo et al. (2018) No information is mentioned Logistic Regression - Internal Verification 5 Age, Duration of Anesthesia, ASA Grade, Renal Failure, and Septic Shock AUC 0.907, the Sensitivity was 84.4%, and Specificity was 88.2% Marquez et al. (2016) No information is mentioned Logistic Regression Akaike Information Guide Internal Verification 4 Age, Recent Surgery, Subclavian CVC, and Blood Transfusion AUC0.80 Kerris et al. (2020) - Logistic Regression HosmerLemeshow Internal Verification 4 CVC, Sepsis, Single Ventricular Disease, and Extracorporeal Membrane Oxygenation Support AUC0.830 Jaffray et al. (2021) No information is mentioned Logistic Regression HosmerLemeshow Internal Verification 11 <1 year old or 10-22 years old, cancer, congenital heart disease, other high-risk diseases, recent hospitalization, immobility, platelet count, CVC placement before admission, recent surgery, steroid use on admission, intubation, and CVC placement on admission AUC0.78 Kerlin et al. (2015) - Logistic Regression HosmerLemeshow Internal Verification 5 Male, CVC, Limb Abnormality, Active Tumor, Alternative Diagnosis AUC0.73 Papillon et al. (2023) Multiple Interpolation Method Logistic Regression - Internal Verification 15 Intubation, need for supplemental oxygen, spinal injury, pelvic fracture, multiple long bone fractures, major surgery, age, need for blood transfusion, placement of intracranial pressure monitor or extracardiac drainage tube, low Glasgow coma score AUC0.9544
International Journal of Molecular Medical Science, 2025, Vol.15, No.2, 54-68 http://medscipublisher.com/index.php/ijmms 60 Continued Table 2 Author (Year) Missing data processing Model development method Calibration method Proof technique Number of predictor Ultimate predictor Model performance Walker et al. (2021) Median Interpolation Method Logistic Regression - Internal Verification 11 CVC, VTE history, heart disease, blood gas test, infection, age, mean erythrocyte hemoglobin concentration, cancer, erythrocyte distribution width, lactic acid, surgery AUC 0.908, the Positive Predictive Value was 20.1% and the Negative Predictive Value was 99.5% Jaffray et al. (2022) Multiple Interpolation Method Logistic Regression - Internal Verification 5 Recent CVC implantation, immobility, congenital heart disease, length of stay ≥3 days prior to ICU admission, autoimmune disease/inflammatory status, or current history of infection AUC0.79 Connelly et al. (2016) - Logistic Regression Bayesian Information Guide External Verification 10 GCS, age, sex, intubation, ICU admission, blood transfusion, CVC, pelvic fracture, lower limb fracture, major surgery AUC0.945 Tiratrakoonseree et al. (2024) - Logistic Regression HosmerLemeshow Internal Verification 5 Congenital heart disease, known thrombosis, history of VTE and nephrotic syndrome, and a clinically significant presentation of chest pain AUC 0.809, the Sensitivity was 74.4%, and the specificity was 75.4% Note: “-”, not reported; KNN, the nearest neighbor algorithm of “k”; ISS: Trauma severity score; GCS: Glasgow Coma Scale; ASA: American Society of Anesthesiologists; we believe that AUC=0.5-0.7 is poor discrimination, 0.7-0.8 is moderate discrimination, 0.8-0.9 is good discrimination, and 0.9-1.0 is excellent discrimination
International Journal of Molecular Medical Science, 2025, Vol.15, No.2, 54-68 http://medscipublisher.com/index.php/ijmms 61 Table 3 PROBAST results included in the study Author (Year) Type of study Bias risk Applicability Totality Object of study Result Analysis Object of study Predictive factor Result Bias risk Applicability Sharathkumar et al. (2012) B + ？ ？ - + + + - + Yen et al. (2016) B + ？ ？ + + + + ? + Cairo et al. (2018) B + ？ ？ - + + + - + Marquez et al. (2016) B + ？ ？ + + + + ? + Kerris et al. (2020) B + ？ ？ - + + + - + Jaffray et al. (2021) B + ？ ？ + + + + ? + Kerlin et al. (2015) B + ？ ？ - + + + - + Papillon et al. (2023) B + ？ ？ + + + + ? + Spavor et al. (2016) B + ？ ？ - + + + - + Walker et al. (2021) B + ？ ？ + + + + ? + Jaffray et al. (2022) B + ？ ？ + + + + ? + Connelly et al. (2016) B + ？ ？ - + + + - + Tiratrakoonseree et al. (2024) B + ？ ？ - + + + - + Note: PROBAST, Prediction model Risk Of Bias Assessment Tool; A stands for “development only”, and B stands for “development and validation in the same study”; “+” indicates low risk of bias/low applicability concern, “-” Indicates high risk of bias/high applicability concern, and “?” Indicates unclear bias risk/unclear application 3.5 Includes meta-analysis to validate the model in the overview Including 13 studies in the review, due to inadequate reporting of model details and the absence of 95% confidence intervals in some studies, only 10 studies were included in the meta-analysis (Figure 3). The combined AUC was calculated using a random effects model, with a result of 0.84 (95% confidence interval: 0.80~0.88). The I² value was 96% (P<0.01). Due to the high heterogeneity in the results of meta-analysis, this study further explored the sources of heterogeneity by means of two methods of retention cross-validation and subgroup analysis (Figure 4; Figure 5; Figure 6; Figure 7). Firstly, the residual one method is used to carry out the analysis. With the help of the “metainf” function of the “meta” package in R software, we carried out a one-by-one deletion operation on the included documents to identify whether one or several documents had a significant impact on the overall heterogeneity. However, we did not find a clear cause of heterogeneity after completing the residual cross-validation. Based on this, we decided to conduct a follow-up subgroup analysis. 3.6 Subgroup analysis In the study on the risk prediction model of VTE in children, subgroup analysis was conducted according to the study population and the number of predictors of the risk prediction model (Figure 5; Figure 6; Figure 7). In this study, we analyzed the subgroup data and heterogeneity of the pediatric medical group, pediatric surgery group, and ICU group, and found that there were differences in the efficacy and heterogeneity of the model in each subgroup.
International Journal of Molecular Medical Science, 2025, Vol.15, No.2, 54-68 http://medscipublisher.com/index.php/ijmms 62 Figure 3 10 forest maps included in the model Figure 4 Leave-One-Out Figure 5 The forest map of participants was analyzed Figure 6 The forest map of whether the number of predictors is greater than 5 was analyzed
International Journal of Molecular Medical Science, 2025, Vol.15, No.2, 54-68 http://medscipublisher.com/index.php/ijmms 63 Figure 7 The forest map of different number of predictors was analyzed The pooled AUC of the pediatric internal medicine group was 0.82, and the prediction efficiency was above average, but the heterogeneity was extremely high. This is mainly due to the variety of variables selected and the unadjusted age stratification effect. Subsequent model optimization needs to strengthen the standardization of core variables. The pooled AUC of the pediatric surgery group was 0.92, with excellent prediction efficiency and low heterogeneity. Due to the concentration of patient risk factors and high consistency in the selection of model variables, the application value of targeted models in the perioperative period was highlighted. The pooled AUC of the ICU group was 0.81, which showed moderate prediction efficiency and very low heterogeneity. However, the AUC was relatively low, suggesting the need to integrate dynamic biomarkers to improve prediction accuracy. In addition, the study also found that the model with more than 5 predictive factors, because the information covered is more comprehensive, the various factors can complement and cooperate, so the model efficiency is significantly higher; On the contrary, models with predictive factors less than 5 lack consistency due to the large differences in factor selection and data samples among different studies, which leads to greater heterogeneity. 4 Discussion Childhood VTE risk prediction model is a tool to evaluate the possibility of VTE in children by collecting various relevant data of children and using statistical analysis. Building on comprehensive and accurate predictors and using the tool effectively can significantly reduce the risk of VTE in children. This study finds that in the process of developing risk prediction models, most studies have a large gap in the inclusion of predictive factors. As mentioned earlier, CVC is the most common predictive factor in children's VTE risk prediction models, but was not used in the study by Tiratrakoonseree et al. (2024). The model developed in this study reported an AUC value of 0.809, indicating high performance.This suggests that even without the inclusion of the typically prominent CVC factor, the model was still able to achieve a good level of discriminatory power in predicting VTE risk among children. It was noted that risk prediction models that are applicable to Western populations may not be applicable to Thailand, as significant risk factors may differ between populations (Milford et al., 2020). Previous studies on children's VTE risk prediction models have mainly focused on North America, particularly the United States. Due to limited resources and low usage of CVC in countries like Thailand, this predictive factor is not significant in
International Journal of Molecular Medical Science, 2025, Vol.15, No.2, 54-68 http://medscipublisher.com/index.php/ijmms 64 Thailand. Therefore, it can be concluded that the performance of children's VTE models studied in the United States may be affected in regions like Thailand. Since risk factors can be highly influenced by the environment and vary from region to region, a series of targeted strategies are needed to better adapt the model to different healthcare Settings (Sheng et al., 2022). First, region-specific risk factors should be incorporated. For example, in Southeast Asian countries, after in-depth study of the pathogenesis characteristics of local children VTE, blood hypercoagulability caused by tropical diseases and infection-related factors caused by differences in sanitary conditions can be included in the model. For example, in malaria-endemic areas, the effect of malaria infection on the blood system of children increases the risk of VTE and can be considered as an important risk factor. Secondly, variable weights should be optimized. The weights of existing risk factors should be re-evaluated according to regional differences. In Southeast Asia, where CVC usage is low, the weight of CVC in the model should be appropriately reduced, while the weight of factors more relevant to local conditions should be increased, such as the influence of local high incidence of congenital diseases on VTE. This needs to be achieved through big data analysis and local clinical studies to ensure that the model is more realistic. Finally, it is important to carry out multi-center research and model validation. Through cross-regional multi-center research, data from different regions such as Western and Southeast Asian countries are collected to jointly participate in model construction and verification. A number of representative medical centers in Southeast Asia and the West were selected to collect child VTE case data, and these rich data resources were used to optimize the model and verify and adjust it in different regions, so as to build a more universal model to ensure accurate prediction of VTE risk in different medical Settings. In addition, this study found that BMI could be used as a predictor in future VTE risk prediction models for children. In Sharathkumar et al. (2012) study, it was found that BMI is becoming one of the risk factors for children's VTE, especially in the adolescent group. Three studies from 2018 to 2024 (Cairo et al., 2018; Walker et al., 2021; Tiratrakoonseree et al., 2024), included heart disease and kidney disease as risk factors for children's VTE. According to research findings, childhood obesity increases the risk of cardiovascular and kidney diseases and can lead to increased blood viscosity, thereby increasing the likelihood of thrombosis (Varra et al., 2024). Furthermore, obesity has been classified as a fundamental disease by the World Health Organization (Leung et al., 2024). VTE is rare in healthy children but its incidence is increasing in children with underlying diseases (Witmer and Raffini, 2020). BMI is an objective indicator for evaluating childhood obesity and overweight. A study in the United States showed a significant association between obesity, overweight, and the diagnosis of venous thromboembolism in children aged 2 to 18, but further research is needed to fully define this relationship (Halvorson et al., 2016). Therefore, researchers are advised to increase research on the correlation between BMI and children's VTE. During the data extraction process, we found that age is the second largest predictor after CVC. In the study by Jaffray et al. (2021), 728 pediatric patients aged 0-21 with VTE were included, with <1 year olds and adolescents accounting for 39% and 34%, respectively. However, the study by Cairo et al. (2018) found that the probability of venous thromboembolism (VTE) in children aged 6 to 15 years is small, making age a difficult predictor in pediatric VTE risk prediction models. Based on this, we can infer that the neonatal period and adolescence are high-risk stages for pediatric VTE. The high incidence of VTE in neonates and infants is due to differences in their hemostatic systems. Neonates do not pass coagulation proteins through the placenta, leading to lower concentrations of procoagulants, along with factors such as the hypercoagulable state due to disease and increased invasive treatments. Therefore, special attention should be paid to the presence of risk factors such as sepsis, prematurity, and CVC during the care of children in this stage (Makatsariya et al., 2022). In neonates with sepsis, activation of leukocytes and platelets in the body can lead to venous stasis, keeping the blood in a hypercoagulable state and promoting the occurrence of VTE (Fort et al., 2022). Premature infants have lower levels of natural anticoagulants (such as antithrombin and
International Journal of Molecular Medical Science, 2025, Vol.15, No.2, 54-68 http://medscipublisher.com/index.php/ijmms 65 protein C), and the decrease in procoagulant factors is related to the degree of prematurity. An excessive decrease in procoagulant factors can promote the occurrence of VTE (Fort et al., 2022). The high incidence of VTE in adolescents is attributed to their thin vascular walls, high blood viscosity, and high activity levels. In terms of diseases, inflammatory bowel disease and trauma are the main VTE risk factors; in daily habits, obesity and prolonged sitting are key factors. Inflammatory bowel disease damages the intestinal mucosa, causing ulceration, platelet aggregation, and coagulation system activation. Traumatized patients also need platelet aggregation and coagulation system activation for tissue repair. Adolescents' high blood viscosity further elevates the VTE risk. Obese children's blood clots more easily than that of normal children. Besides, adolescents' long-term sitting for study, video-watching, or gaming leads to venous stasis and VTE. Poor daily habits not only trigger VTE but also promote its recurrence in children (Biss et al., 2016; Srivaths and Dietrich, 2016). Thus, healthcare providers should monitor adolescents' disease status and lifestyle, enhance health education during treatment, assist them in correcting unhealthy habits, and reduce VTE incidence and recurrence. Model-building researchers can focus on childhood's high-risk periods, create targeted risk prediction models, and promote children's health. At present, the construction of risk prediction models for venous thromboembolism (VTE) in children faces a dynamic balance between the dominance of clinical variables and the application potential of biomarkers. Existing models are mostly based on clinical variables (such as central venous catheter, age, surgery, etc.), and the integration of VTE-related biomarkers is still in the exploratory stage. Studies have shown that D-dimer, as a fibrin degradation product, presents a specific increase in the occurrence of VTE in children, and its dynamic monitoring can reflect the real-time coagulation status. C-reactive protein (CRP) complements risk assessment in children with specific inflammation-related diseases by characterizing the association between inflammation and thrombosis. The inclusion of these biomarkers is expected to break the bottleneck of current model prediction efficiency, but at the same time, it also aggravates the challenges of model complexity and heterogeneity. Notably, when the model had more than 5 predictors, the meta-analysis showed a significant increase in heterogeneity. This phenomenon is due to the interweaving of multiple factors: first, the combination of biomarkers and clinical variables needs to account for physiological differences in the age stratification (newborn to adolescent), such as the normal range of D-dimer changes with age; Secondly, differences in global medical practice (such as the use of central venous catheters between East and West) lead to shifts in the weight of predictors. Furthermore, variable selection bias (such as focusing on disease factors or treatments) exacerbates incomparability between models. For example, simultaneous integration of D-dimer, CRP, and clinical variables requires both addressing the interaction effects of biomarkers and age and reconciliating geographic heterogeneity in the importance of variables in different healthcare Settings. In future studies, researchers need to adopt a dual strategy: at the longitudinal level, establish age-specific biomarker thresholds through multicenter cohorts, and analyze their nonlinear associations with clinical variables using machine learning; At the horizontal level, a core predictive factor subset (e.g., limited to 5 key variables) is used as a cross-regional baseline, while allowing region-specific expansion variables (e.g., strengthening infection-related indicators in Southeast Asia). This “core + extension” model architecture can not only maintain the homogeneity of meta-analysis, but also retain the advantage of biomarkers to improve prediction accuracy, and ultimately achieve the optimal balance between precision medicine and clinical universality of VTE risk assessment tools in children. In future clinical work, based on the results of this study, the following five clinical recommendations are put forward for medical staff: ①The use of central venous catheter (CVC) should be strictly controlled, the retention time should be reduced and monitoring should be done, especially in resource-limited areas, alternative risk factors should be paid attention to. ②For children of different ages, such as newborns, infants and adolescents, targeted management should be conducted according to their physiological characteristics and risk factors. ③The risk of VTE should be evaluated comprehensively in children with surgery and trauma, and preventive measures should be reasonably selected and blood transfusion management should be standardized. ④Dynamic monitoring
International Journal of Molecular Medical Science, 2025, Vol.15, No.2, 54-68 http://medscipublisher.com/index.php/ijmms 66 of inflammatory markers combined with clinical variables to avoid misjudgment. ⑤Attach importance to obesity assessment, and jointly develop intervention programs for children with obesity. In short, clinical staff should be alert to the synergistic effect of multiple risk factors, and accurately prevent and manage VTE in children according to regionalized and age-stratified strategies to ensure the health of children. 5 Conclusion This systematic review and meta-analysis comprehensively evaluated the predictive efficacy and heterogeneity of venous thromboembolism (VTE) risk prediction models in children. The pooled AUC of 0.84 (95% CI: 0.80~0.88) demonstrated moderate-to-good discriminative performance across existing models, predominantly based on logistic regression, with central venous catheter (CVC) as the most frequent predictor. However, significant heterogeneity (I²=96%) was observed, driven by variability in predictor selection, regional differences in medical practices, and insufficient adjustment for age stratification. Seven studies exhibited a high risk of bias, primarily due to opaque handling of missing data and lack of blinding in predictor-outcome assessments. The proposed "core + extended" model architecture integrates age-specific biomarker thresholds (e.g., D-dimer, CRP) vertically and cross-regional core variables (≤5 items) horizontally, aiming to harmonize accuracy with clinical applicability. Future research should prioritize multicenter cohorts to validate dynamic biomarkers and machine learning algorithms, address regional disparities in predictor relevance (e.g., CVC in low-resource settings), and refine age-stratified strategies for neonates and adolescents. Clinical recommendations emphasize judicious CVC use, obesity assessment, and dynamic monitoring of inflammatory markers. By addressing these gaps, future models can enhance precision while maintaining generalizability, ultimately improving VTE prevention and management in pediatric populations. Acknowledgement We would like to thank Professor Yang Li, Professor Xin Zhang and Professor Yeli Huang for their guidance and inspiration on the topic selection, ideas, views and demonstration of this research. We thank Peking Union Medical College for supporting this study. Conflict of Interest Disclosure The authors affirm that this research was conducted without any commercial or financial relationships that could be construed as a potential conflict of interest. References Biss T., Alikhan R., Payne J., Alamelu J., Williams M., Richards M., Mathias M., Tunstall O., and Chalmers E., 2016, Venous thromboembolism occurring during adolescence, Archives of Disease in Childhood, 101(5): 427-432. https://doi.org/10.1136/archdischild-2015-309875 Biss T.T., 2016, Venous thromboembolism in children: is it preventable?, Seminars in Thrombosis and Hemostasis, 42(6): 603-611. https://doi.org/10.1055/s-0036-1581100 Cairo S.B., Lautz T.B., Schaefer B.A., Yu G., and Rothstein D.H., 2018, Risk factors for venous thromboembolic events in pediatric surgical patients: defining indications for prophylaxis, Journal of Pediatric Surgery, 53(10): 1996-2002. https://doi.org/10.1016/j.jpedsurg.2017.12.016 Chalmers E., Ganesen V., Liesner R., Maroo S., Nokes T., Saunders D., and Williams M., 2011, Guideline on the investigation, management and prevention of venous thrombosis in children, British Journal of Haematology, 154(2): 196-207. https://doi.org/10.1111/j.1365-2141.2010.08543.x Connelly C.R., Laird A., Barton J.S., Fischer P.E., Krishnaswami S., Schreiber M.A., Zonies D.H., and Watters J.M., 2016, A clinical tool for the prediction of venous thromboembolism in pediatric trauma patients, JAMA Surgery, 151(1): 50-57. https://doi.org/10.1001/jamasurg.2015.2670 Cui Y., and Zhang Y.X., eds., 2021, Pediatric Nursing (7th ed.), People's Medical Publishing House, Beijing, China, pp. 442. Fort P., Beg K., Betensky M., Kiskaddon A., and Goldenberg N.A., 2022, Venous thromboembolism in premature neonates, Seminars in Thrombosis and Hemostasis, 48(4): 422-433. https://doi.org/10.1055/s-0041-1740267 Halvorson E.E., Ervin S.E., Russell T.B., Skelton J.A., Davis S., and Spangler J., 2016, Association of obesity and pediatric venous thromboembolism, Hospital Pediatrics, 6(1): 22-26. https://doi.org/10.1542/hpeds.2015-0039
International Journal of Molecular Medical Science, 2025, Vol.15, No.2, 54-68 http://medscipublisher.com/index.php/ijmms 67 Jaffray J., Branchford B., Goldenberg N., Malvar J., Croteau S.E., Silvey M., Fargo J.H., Cooper J.D., Bakeer N., Sposto R., Ji L.Y., Zakai N.A., Faustino E.V.S., Stillings A., Krava E., Young G., and Mahajerin A., 2021, Development of a risk model for pediatric hospital-acquired thrombosis: a report from the Children's Hospital-Acquired Thrombosis Consortium, The Journal of Pediatrics, 228: 252-259. https://doi.org/10.1016/j.jpeds.2020.09.016 Jaffray J., Mahajerin A., Branchford B., Nguyen A.T.H., Faustino E.V.S., Silvey M., Croteau S.E., Fargo J.H., Cooper J.D., Bakeer N., Zakai N.A., Stillings A., Krava E., Amankwah E.K., Young G., and Goldenberg N.A., 2022, A new risk assessment model for hospital-acquired venous thromboembolism in critically ill children: a report from the Children's Hospital-Acquired Thrombosis Consortium, Pediatric Critical Care Medicine, 23(1): e1-e9. https://doi.org/10.1097/PCC.0000000000002826 Kerlin B.A., Stephens J.A., Hogan M.J., Smoyer W.E., and O'Brien S.H., 2015, Development of a pediatric-specific clinical probability tool for diagnosis of venous thromboembolism: a feasibility study, Pediatric Research, 77(3): 463-471. https://doi.org/10.1038/pr.2014.198 Kerris E.W., Sharron M., Zurakowski D., Staffa S.J., Yurasek G., and Diab Y., 2020, Hospital-associated venous thromboembolism in a pediatric cardiac ICU: a multivariable predictive algorithm to identify children at high risk, Pediatric Critical Care Medicine, 21(6): e362-e368. https://doi.org/10.1097/PCC.0000000000002293 Leung A.K., Wong A.H., and Hon K.L., 2024, Childhood obesity: an updated review, Current Pediatric Reviews, 20(1): 2-26. https://doi.org/10.2174/1573396318666220801093225 Makatsariya A., Bitsadze V., Khizroeva J., Vorobev A., Makatsariya N., Egorova E., Mischenko A., Mashkova T., and Antonova A., 2022, Neonatal thrombosis, The Journal of Maternal-Fetal & Neonatal Medicine, 35(6): 1169-1177. https://doi.org/10.1080/14767058.2020.1743668 Marquez A., Shabanova V., and Faustino E.V.S., 2016, Prediction of catheter-associated thrombosis in critically ill children, Pediatric Critical Care Medicine, 17(11): e521-e528. https://doi.org/10.1097/PCC.0000000000000958 Milford K., von Delft D., Majola N., and Cox S., 2020, Long-term vascular access in differently resourced settings: a review of indications, devices, techniques, and complications, Pediatric Surgery International, 36: 551-562. https://doi.org/10.1007/s00383-020-04640-0 Monagle P., Chan A.K., Goldenberg N.A., Ichord R.N., Journeycake J.M., Nowak-Göttl U., and Vesely S.K., 2012, Antithrombotic therapy in neonates and children: antithrombotic therapy and prevention of thrombosis: American college of chest physicians evidence-based clinical practice guidelines, Chest, 141(2): e737S-e801S. https://doi.org/10.1378/chest.11-2308 O'Brien S.H., Stanek J.R., Witmer C.M., and Raffini L., 2022, The continued rise of venous thromboembolism across US children's hospitals, Pediatrics, 149(3): e2021054649. https://doi.org/10.1542/peds.2021-054649 Papillon S.C., Pennell C.P., Master S.A., Turner E.M., Arthur L.G., Grewal H., and Aronoff S.C., 2023, Derivation and validation of a machine learning algorithm for predicting venous thromboembolism in injured children, Journal of Pediatric Surgery, 58(6): 1200-1205. https://doi.org/10.1016/j.jpedsurg.2023.02.040 Rühle F., and Stoll M., 2018, Advances in predicting venous thromboembolism risk in children, British Journal of Haematology, 180(5): 654-665. https://doi.org/10.1111/bjh.15060 Sharathkumar A.A., Mahajerin A., Heidt L., Doerfer K., Heiny M., Vik T., Fallon R., and Rademaker A., 2012, Risk‐prediction tool for identifying hospitalized children with a predisposition for development of venous thromboembolism: Peds‐Clot clinical decision rule, Journal of Thrombosis and Haemostasis, 10(7): 1326-1334. https://doi.org/10.1111/j.1538-7836.2012.04779.x Sheng Y., Wu T., Fan C., Hao H., and Gao W., 2022, Factors influencing the optimal selection of central venous access devices: a qualitative study of health care team members' perspectives, International Journal of Nursing Sciences, 9(4): 445-452. https://doi.org/10.1016/j.ijnss.2022.09.006 Spavor M., Halton J., Dietrich K., Israels S., Shereck E., Yong J., Yasui Y., and Mitchell L.G., 2016, Age at cancer diagnosis, non-O blood group and asparaginase therapy are independently associated with deep venous thrombosis in pediatric oncology patients: a risk model, Thrombosis Research, 144: 27-31. https://doi.org/10.1016/j.thromres.2016.05.015 Srivaths L., and Dietrich J.E., 2016, Prothrombotic risk factors and preventive strategies in adolescent venous thromboembolism, Clinical and Applied Thrombosis/Hemostasis, 22(6): 512-519. https://doi.org/10.1177/1076029616631428 Tan Z.Y., and Zhi K.K., 2023, Current status and progress in the diagnosis and treatment of venous thromboembolism in children, Journal of Vascular and Endovascular Surgery, 9(4): 454-459. Tiratrakoonseree T., Charoenpichitnun S., Natesirinilkul R., Songthawee N., Komvilaisak P., Pongphitcha P., Vaewpanich J., and Sirachainan N., 2024, Clinical prediction tool to identify children at risk of pulmonary embolism, Thrombosis Research, 234: 151-157. https://doi.org/10.1016/j.thromres.2024.01.006
RkJQdWJsaXNoZXIy MjQ4ODYzNA==