Bayesian network-based survival prediction model for patients having undergone post-transjugular intrahepatic portosystemic shunt for portal hypertension

BACKGROUND

Portal hypertension (PHT), primarily induced by cirrhosis, manifests severe symptoms impacting patient survival. Although transjugular intrahepatic portosystemic shunt (TIPS) is a critical intervention for managing PHT, it carries risks like hepatic encephalopathy, thus affecting patient survival prognosis. To our knowledge, existing prognostic models for post-TIPS survival in patients with PHT fail to account for the interplay among and collective impact of various prognostic factors on outcomes. Consequently, the development of an innovative modeling approach is essential to address this limitation.

AIM

To develop and validate a Bayesian network (BN)-based survival prediction model for patients with cirrhosis-induced PHT having undergone TIPS.

METHODS

The clinical data of 393 patients with cirrhosis-induced PHT who underwent TIPS surgery at the Second Affiliated Hospital of Chongqing Medical University between January 2015 and May 2022 were retrospectively analyzed. Variables were selected using Cox and least absolute shrinkage and selection operator regression methods, and a BN-based model was established and evaluated to predict survival in patients having undergone TIPS surgery for PHT.

RESULTS

Variable selection revealed the following as key factors impacting survival: age, ascites, hypertension, indications for TIPS, postoperative portal vein pressure (post-PVP), aspartate aminotransferase, alkaline phosphatase, total bilirubin, prealbumin, the Child-Pugh grade, and the model for end-stage liver disease (MELD) score. Based on the above-mentioned variables, a BN-based 2-year survival prognostic prediction model was constructed, which identified the following factors to be directly linked to the survival time: age, ascites, indications for TIPS, concurrent hypertension, post-PVP, the Child-Pugh grade, and the MELD score. The Bayesian information criterion was 3589.04, and 10-fold cross-validation indicated an average log-likelihood loss of 5.55 with a standard deviation of 0.16. The model's accuracy, precision, recall, and F1 score were 0.90, 0.92, 0.97, and 0.95 respectively, with the area under the receiver operating characteristic curve being 0.72.

CONCLUSION

This study successfully developed a BN-based survival prediction model with good predictive capabilities. It offers valuable insights for treatment strategies and prognostic evaluations in patients having undergone TIPS surgery for PHT.

Comparison of nine trauma scoring systems in prediction of inhospital outcomes of pediatric trauma patients: a multicenter study

Hereby, we aimed to comprehensively compare different scoring systems for pediatric trauma and their ability to predict in-hospital mortality and intensive care unit (ICU) admission. The current registry-based multicenter study encompassed a comprehensive dataset of 6709 pediatric trauma patients aged ≤ 18 years from July 2016 to September 2023. To ascertain the predictive efficacy of the scoring systems, the area under the receiver operating characteristic curve (AUC) was calculated. A total of 720 individuals (10.7%) required admission to the ICU. The mortality rate was 1.1% (n = 72). The most predictive scoring system for in-hospital mortality was the adjusted trauma and injury severity score (aTRISS) (AUC = 0.982), followed by trauma and injury severity score (TRISS) (AUC = 0.980), new trauma and injury severity score (NTRISS) (AUC = 0.972), Glasgow coma scale (GCS) (AUC = 0.9546), revised trauma score (RTS) (AUC = 0.944), pre-hospital index (PHI) (AUC = 0.936), injury severity score (ISS) (AUC = 0.901), new injury severity score (NISS) (AUC = 0.900), and abbreviated injury scale (AIS) (AUC = 0.734). Given the predictive performance of the scoring systems for ICU admission, NTRISS had the highest predictive performance (AUC = 0.837), followed by aTRISS (AUC = 0.836), TRISS (AUC = 0.823), ISS (AUC = 0.807), NISS (AUC = 0.805), GCS (AUC = 0.735), RTS (AUC = 0.698), PHI (AUC = 0.662), and AIS (AUC = 0.651). In the present study, we concluded the superiority of the TRISS and its two derived counterparts, aTRISS and NTRISS, compared to other scoring systems, to efficiently discerning individuals who possess a heightened susceptibility to unfavorable consequences. The significance of these findings underscores the necessity of incorporating these metrics into the realm of clinical practice.

A risk score model based on endoplasmic reticulum stress related genes for predicting prognostic value of osteosarcoma

BACKGROUND

We aimed to establish an osteosarcoma prognosis prediction model based on a signature of endoplasmic reticulum stress-related genes.

METHODS

Differentially expressed genes (DEGs) between osteosarcoma with and without metastasis from The Cancer Genome Atlas (TCGA) database were mapped to ERS genes retrieved from Gene Set Enrichment Analysis to select endoplasmic reticulum stress-related DEGs. Subsequently, we constructed a risk score model based on survival-related endoplasmic reticulum stress DEGs and a nomogram of independent survival prognostic factors. Based on the median risk score, we stratified the samples into high- and low-risk groups. The ability of the model was assessed by Kaplan-Meier, receiver operating characteristic curve, and functional analyses. Additionally, the expression of the identified prognostic endoplasmic reticulum stress-related DEGs was verified using real-time quantitative PCR (RT-qPCR).

RESULTS

In total, 41 endoplasmic reticulum stress-related DEGs were identified in patients with osteosarcoma with metastasis. A risk score model consisting of six prognostic endoplasmic reticulum stress-related DEGs (ATP2A3, ERMP1, FBXO6, ITPR1, NFE2L2, and USP13) was established, and the Kaplan-Meier and receiver operating characteristic curves validated their performance in the training and validation datasets. Age, tumor metastasis, and the risk score model were demonstrated to be independent prognostic clinical factors for osteosarcoma and were used to establish a nomogram survival model. The nomogram model showed similar performance of one, three, and five year-survival rate to the actual survival rates. Nine immune cell types in the high-risk group were found to be significantly different from those in the low-risk group. These survival-related genes were significantly enriched in nine Kyoto Encyclopedia of Genes and Genomes pathways, including cell adhesion molecule cascades, and chemokine signaling pathways. Further, RT-qPCR results demonstrated that the consistency rate of bioinformatics analysis was approximately 83.33%, suggesting the relatively high reliability of the bioinformatics analysis.

CONCLUSION

We established an osteosarcoma prediction model based on six prognostic endoplasmic reticulum stress-related DEGs that could be helpful in directing personalized treatment.

A Gallbladder Cancer Survival Prediction Model Based on Multimodal Fusion Analysis

BACKGROUND

Gallbladder cancer is the sixth most common malignant gastrointestinal tumor. Radical surgery is currently the only effective treatment, but patient prognosis is poor, with a 5-year survival rate of only 5-10%. Establishing an effective survival prediction model for gallbladder cancer patients is crucial for disease status assessment, early intervention, and individualized treatment approaches. The existing gallbladder cancer survival prediction model uses clinical data-radiotherapy and chemotherapy, pathology, and surgical scope-but fails to utilize laboratory examination and imaging data, limiting its prediction accuracy and preventing sufficient treatment plan guidance.

AIMS

The aim of this work is to propose an accurate survival prediction model, based on the deep learning 3D-DenseNet network, integrated with multimodal medical data (enhanced CT imaging, laboratory test results, and data regarding systemic treatments).

METHODS

Data were collected from 195 gallbladder cancer patients at two large tertiary hospitals in Shanghai. The 3D-DenseNet network extracted deep imaging features and constructed prognostic factors, from which a multimodal survival prediction model was established, based on the Cox regression model and incorporating patients' laboratory test and systemic treatment data.

RESULTS

The model had a C-index of 0.787 in predicting patients' survival rate. Moreover, the area under the curve (AUC) of predicting patients' 1-, 3-, and 5-year survival rates reached 0.827, 0.865, and 0.926, respectively.

CONCLUSIONS

Compared with the monomodal model based on deep imaging features and the tumor-node-metastasis (TNM) staging system-widely used in clinical practice-our model's prediction accuracy was greatly improved, aiding the prognostic assessment of gallbladder cancer patients.

Identification and validation of a novel survival prediction model based on the T-cell phenotype in the tumor immune microenvironment and peripheral blood for gastric cancer prognosis

BACKGROUND

The correlation and difference in T-cell phenotypes between peripheral blood lymphocytes (PBLs) and the tumor immune microenvironment (TIME) in patients with gastric cancer (GC) is not clear. We aimed to characterize the phenotypes of CD8⁺ T cells in tumor infiltrating lymphocytes (TILs) and PBLs in patients with different outcomes and to establish a useful survival prediction model.

METHODS

Multiplex immunofluorescence staining and flow cytometry were used to detect the expression of inhibitory molecules (IMs) and active markers (AMs) in CD8⁺TILs and PBLs, respectively. The role of these parameters in the 3-year prognosis was assessed by receiver operating characteristic analysis. Then, we divided patients into two TIME clusters (TIME-A/B) and two PBL clusters (PBL-A/B) by unsupervised hierarchical clustering based on the results of multivariate analysis, and used the Kaplan-Meier method to analyze the difference in prognosis between each group. Finally, we constructed and compared three survival prediction models based on Cox regression analysis, and further validated the efficiency and accuracy in the internal and external cohorts.

RESULTS

The percentage of PD-1⁺CD8⁺TILs, TIM-3⁺CD8⁺TILs, PD-L1⁺CD8⁺TILs, and PD-L1⁺CD8⁺PBLs and the density of PD-L1⁺CD8⁺TILs were independent risk factors, while the percentage of TIM-3⁺CD8⁺PBLs was an independent protective factor. The patients in the TIME-B group showed a worse 3-year overall survival (OS) (HR: 3.256, 95% CI 1.318-8.043, P = 0.006), with a higher density of PD-L1⁺CD8⁺TILs (P < 0.001) and percentage of PD-1⁺CD8⁺TILs (P = 0.017) and PD-L1⁺CD8⁺TILs (P < 0.001) compared to the TIME-A group. The patients in the PBL-B group showed higher positivity for PD-L1⁺CD8⁺PBLs (P = 0.042), LAG-3⁺CD8⁺PBLs (P < 0.001), TIM-3⁺CD8⁺PBLs (P = 0.003), PD-L1⁺CD4⁺PBLs (P = 0.001), and LAG-3⁺CD4⁺PBLs (P < 0.001) and poorer 3-year OS (HR: 0.124, 95% CI 0.017-0.929, P = 0.015) than those in the PBL-A group. In our three survival prediction models, Model 3, which was based on the percentage of TIM-3⁺CD8⁺PBLs, PD-L1⁺CD8⁺TILs and PD-1⁺CD8⁺TILs, showed the best sensitivity (0.950, 0.914), specificity (0.852, 0.857) and accuracy (κ = 0.787, P < 0.001; κ = 0.771, P < 0.001) in the internal and external cohorts, respectively.

CONCLUSION

We established a comprehensive and robust survival prediction model based on the T-cell phenotype in the TIME and PBLs for GC prognosis.

Development and validation of a survival prediction model in elder patients with community-acquired pneumonia: a MIMIC-population-based study

BACKGROUND

To develop a prediction model predicting in-hospital mortality of elder patients with community-acquired pneumonia (CAP) admitted to the intensive care unit (ICU).

METHODS

In this cohort study, data of 619 patients with CAP aged ≥ 65 years were obtained from the Medical Information Mart for Intensive Care III (MIMIC III) 2001-2012 database. To establish the robustness of predictor variables, the sample dataset was randomly partitioned into a training set group and a testing set group (ratio: 6.5:3.5). The predictive factors were evaluated using multivariable logistic regression, and then a prediction model was constructed. The prediction model was compared with the widely used assessments: Sequential Organ Failure Assessment (SOFA), Pneumonia Severity Index (PSI), systolic blood pressure, oxygenation, age and respiratory rate (SOAR), CURB-65 scores using positive predictive value (PPV), negative predictive value (NPV), accuracy (ACC), area under the curve (AUC) and 95% confidence interval (CI). The decision curve analysis (DCA) was used to assess the net benefit of the prediction model. Subgroup analysis based on the pathogen was developed.

RESULTS

Among 402 patients in the training set, 90 (24.63%) elderly CAP patients suffered from 30-day in-hospital mortality, with the median follow-up being 8 days. Hemoglobin/platelets ratio, age, respiratory rate, international normalized ratio, ventilation use, vasopressor use, red cell distribution width/blood urea nitrogen ratio, and Glasgow coma scales were identified as the predictive factors that affect the 30-day in-hospital mortality. The AUC values of the prediction model, the SOFA, SOAR, PSI and CURB-65 scores, were 0.751 (95% CI 0.749-0.752), 0.672 (95% CI 0.670-0.674), 0.607 (95% CI 0.605-0.609), 0.538 (95% CI 0.536-0.540), and 0.645 (95% CI 0.643-0.646), respectively. DCA result demonstrated that the prediction model could provide greater clinical net benefits to CAP patients admitted to the ICU. Concerning the pathogen, the prediction model also reported better predictive performance.

CONCLUSION

Our prediction model could predict the 30-day hospital mortality in elder patients with CAP and guide clinicians to identify the high-risk population.

Developing and validating a survival prediction model based on blood exosomal ceRNA network in patients with PAAD

BACKGROUND

Among the most lethal cancers, pancreatic adenocarcinoma (PAAD) is an essential component of digestive system malignancies that still lacks effective diagnosis and treatment methods. As exosomes and competing endogenous RNA (ceRNA) regulatory networks in tumors go deeper, we expect to construct a ceRNA regulatory network derived from blood exosomes of PAAD patients by bioinformatics methods and develop a survival prediction model based on it.

METHODS

Blood exosome sequencing data of PAAD patients and normal controls were downloaded from the exoRbase database, and the expression profiles of exosomal mRNA, lncRNA, and circRNA were differentially analyzed by R. The related mRNA, circRNA, lncRNA, and their corresponding miRNA prediction data were imported into Cytoscape software to visualize the ceRNA network. Then, we conducted GO and KEGG enrichment analysis of mRNA in the ceRNA network. Genes that express differently in pancreatic cancer tissues compared with normal tissues and associate with survival (P < 0.05) were determined as Hub genes by GEPIA. We identified optimal prognosis-related differentially expressed mRNAs (DEmRNAs) and generated a risk score model by performing univariate and multivariate Cox regression analyses.

RESULTS

205 DEmRNAs, 118 differentially expressed lncRNAs (DElncRNAs), and 98 differentially expressed circRNAs (DEcircRNAs) were screened out. We constructed the ceRNA network, and a total of 26 mRNA nodes, 7 lncRNA nodes, 6 circRNA nodes, and 16 miRNA nodes were identified. KEGG enrichment analysis showed that the DEmRNAs in the regulatory network were mainly enriched in Human papillomavirus infection, PI3K-Akt signaling pathway, Osteoclast differentiation, and ECM-receptor interaction. Next, six hub genes (S100A14, KRT8, KRT19, MAL2, MYO5B, PSCA) were determined through GEPIA. They all showed significantly increased expression in cancer tissues compared with control groups, and their high expression pointed to adverse survival. Two optimal prognostic-related DEmRNAs, MYO5B (HR = 1.41, P < 0.05) and PSCA (HR = 1.10, P < 0.05) were included to construct the survival prediction model.

CONCLUSION

In this study, we successfully constructed a ceRNA regulatory network in blood exosomes from PAAD patients and developed a two-gene survival prediction model that provided new targets which shall aid in diagnosing and treating PAAD.

Surgical Efficacy and Prognosis of 54 Cases of Spinal Metastases from Breast Cancer

OBJECTIVE

To analyze the efficacy and complications of spinal metastasis surgery for breast cancer; to understand the survival and the influencing factors; and to verify the predictive ability of the currently used spinal metastasis cancer survival prediction scoring system on 1 year postoperative survival.

METHODS

A retrospective study was conducted of 54 patients with spinal metastases from breast cancer who underwent open surgery after multidisciplinary consultation in our hospital from January 2017 to October 2020. Patient demographic-related variables, breast cancer-related variables, spinal disorder-related variables, and treatment-related variables were collected. Survival curves were plotted using the Kaplan-Meier method, 1-way tests were performed using the log-rank method for factors that might affect prognosis, and candidate variables were included in the Cox model for multifactor analysis. The Tomita score, modified Tokuhashi score, modified Bauer score, and modified Katagiri score were examined by plotting the subject operating characteristic curve and calculating the area under the curve. The area under the curve was used to test the predictive ability of the SORG (Skeletal Oncology Research Group) original version, SORG line graph version, and SORG Web version for 1-year postoperative survival in patients with spinal metastases from breast cancer.

RESULTS

The average age was 51.3 ± 8.6 years in 54 patients. Twenty-one patients underwent vertebral body debulking surgery, 32 patients underwent palliative canal decompression, and 1 patient underwent vertebral en bloc resection, with an operative time of 229.3 ± 87.6 minutes and intraoperative bleeding of 1018.1 ± 931.1 mL. Postoperatively, the patient experienced significant pain relief and gradual recovery from nerve injury. Major surgical complications included cerebrospinal fluid leakage, secondary spinal cord injury, spinal tumor progression, and broken fixation. The mean survival was 32.2 months, including a 6-month survival of 90.7%, a 1-year survival of 77.8%, and a 2-year survival of 60.3%. Univariate analysis showed that preoperation with neurologic deficits, hormone-insensitive type, with brain metastases were potential risk factors for poor prognosis. Multifactorial analysis showed that hormone-insensitive type and concomitant brain metastasis were independent risk factors associated with poor prognosis. The SORG Web version had good ability to predict 1-year postoperative survival in patients with spinal metastases from breast cancer.

CONCLUSIONS

Spinal metastasis from breast cancer has good surgical efficacy, low postoperative recurrence rate, and relatively long survival after surgery. Patients with hormone-insensitive type, with brain metastasis, have a poor prognosis, and SORG Web version can predict patients' 1-year survival more accurately.

[The progress on survival prediction model of gallbladder carcinoma]

Gallbladder carcinoma (GBC) is the most common malignancy of the biliary tract, radical resection is the only effective treatment for GBC at present. However, the postoperative effect is still poor. Therefore, identifying the key prognostic factors and establishing an individual and accurate survival prediction model for GBC are critical to prognosis assessment, treatment options and clinical decision support in patients with GBC. The prediction value of current commonly used TNM staging system is limited. Cox regression model is the most commonly used classical survival analysis method, but it is difficult to establish the association between prognostic variables. Nomogram and machine learning techniques including Bayesian network have been used to establish survival prediction model of GBC in recent years, which representing a certain degree of advancement, however, the model precision and clinical application still need to be further verified. The establishment of more accurate survival prediction models for GBC based on machine learning algorithm from Chinese multicenter large sample database to guide the clinical decision-making is the main research direction in the future.

Spatially varying effects of predictors for the survival prediction of nonmetastatic colorectal Cancer

Background

An increasing number of studies have identified spatial differences in colorectal cancer survival. However, little is known about the spatially varying effects of predictors in survival prediction modeling studies of colorectal cancer that have focused on estimating the absolute survival risk for patients from a wide range of populations. This study aimed to demonstrate the spatially varying effects of predictors of survival for nonmetastatic colorectal cancer patients.

Methods

Patients diagnosed with nonmetastatic colorectal cancer from 2004 to 2013 who were followed up through the end of 2013 were extracted from the Surveillance Epidemiology End Results registry (Patients: 128061). The log-rank test and the restricted mean survival time were used to evaluate survival outcome differences among spatial clusters corresponding to a widely used clinical predictor: stage determined by AJCC 7th edition staging system. The heterogeneity test, which is used in meta-analyses, revealed the spatially varying effects of single predictors. Then, considering the above predictors in a standard survival prediction model based on spatially clustered data, the spatially varying coefficients of these models revealed that some covariate effects may not be constant across the geographic regions of the study. Then, two types of survival prediction models (a statistical model and a machine learning model) were built; these models considered the predictors and enabled survival prediction for patients from a wide range of geographic regions.

Results

Based on univariate and multivariate analysis, some prognostic factors, such as “TNM stage”, “tumor size” and “age at diagnosis,” have significant spatially varying effects among different regions. When considering these spatially varying effects, machine learning models have fewer assumption constraints (such as proportional hazard assumptions) and better predictive performance compared with statistical models. Upon comparing the concordance indexes of these two models, the machine learning model was found to be more accurate (0.898[0.895,0.902]) than the statistical model (0.732 [0.726, 0.738]).

Conclusions

Based on this study, it’s recommended that the spatially varying effect of predictors should be considered when building survival prediction models involving large-scale and multicenter research data. Machine learning models that are not limited by the requirement of a statistical hypothesis are promising alternative models.

Electronic supplementary material

The online version of this article (10.1186/s12885-018-4985-2) contains supplementary material, which is available to authorized users.

[The survival prediction model of advanced gallbladder cancer based on Bayesian network: a multi-institutional study]

Objective: To investigate the clinical value of Bayesian network in predicting survival of patients with advanced gallbladder cancer(GBC)who underwent curative intent surgery. Methods: The clinical data of patients with advanced GBC who underwent curative intent surgery in 9 institutions from January 2010 to December 2015 were analyzed retrospectively.A median survival time model based on a tree augmented naïve Bayes algorithm was established by Bayesia Lab software.The survival time, number of metastatic lymph nodes(NMLN), T stage, pathological grade, margin, jaundice, liver invasion, age, sex and tumor morphology were included in this model.Confusion matrix, the receiver operating characteristic curve and area under the curve were used to evaluate the accuracy of the model.A priori statistical analysis of these 10 variables and a posterior analysis(survival time as the target variable, the remaining factors as the attribute variables)was performed.The importance rankings of each variable was calculated with the polymorphic Birnbaum importance calculation based on the posterior analysis results.The survival probability forecast table was constructed based on the top 4 prognosis factors. The survival curve was drawn by the Kaplan-Meier method, and differences in survival curves were compared using the Log-rank test. Results: A total of 316 patients were enrolled, including 109 males and 207 females.The ratio of male to female was 1.0∶1.9, the age was (62.0±10.8)years.There was 298 cases(94.3%) R0 resection and 18 cases(5.7%) R1 resection.T staging: 287 cases(90.8%) T3 and 29 cases(9.2%) T4.The median survival time(MST) was 23.77 months, and the 1, 3, 5-year survival rates were 67.4%, 40.8%, 32.0%, respectively.For the Bayesian model, the number of correctly predicted cases was 121(≤23.77 months) and 115(>23.77 months) respectively, leading to a 74.86% accuracy of this model.The prior probability of survival time was 0.503 2(≤23.77 months) and 0.496 8(>23.77 months), the importance ranking showed that NMLN(0.366 6), margin(0.350 1), T stage(0.319 2) and pathological grade(0.258 9) were the top 4 prognosis factors influencing the postoperative MST.These four factors were taken as observation variables to get the probability of patients in different survival periods.Basing on these results, a survival prediction score system including NMLN, margin, T stage and pathological grade was designed, the median survival time(month) of 4-9 points were 66.8, 42.4, 26.0, 9.0, 7.5 and 2.3, respectively, there was a statistically significant difference in the different points(P<0.01). Conclusions: The survival prediction model of GBC based on Bayesian network has high accuracy.NMLN, margin, T staging and pathological grade are the top 4 risk factors affecting the survival of patients with advanced GBC who underwent curative resection.The survival prediction score system based on these four factors could be used to predict the survival and to guide the decision making of patients with advanced GBC.

Survival benefit for patients with diffuse intrinsic pontine glioma (DIPG) undergoing re-irradiation at first progression: A matched-cohort analysis on behalf of the SIOP-E-HGG/DIPG working group

BACKGROUND

Overall survival (OS) of patients with diffuse intrinsic pontine glioma (DIPG) is poor. The purpose of this study is to analyse benefit and toxicity of re-irradiation at first progression.

METHODS

At first progression, 31 children with DIPG, aged 2-16 years, underwent re-irradiation (dose 19.8-30.0 Gy) alone (n = 16) or combined with systemic therapy (n = 15). At initial presentation, all patients had typical symptoms and characteristic MRI features of DIPG, or biopsy-proven high-grade glioma. An interval of ≥3 months after upfront radiotherapy was required before re-irradiation. Thirty-nine patients fulfilling the same criteria receiving radiotherapy at diagnosis, followed by best supportive care (n = 20) or systemic therapy (n = 19) at progression but no re-irradiation, were eligible for a matched-cohort analysis.

RESULTS

Median OS for patients undergoing re-irradiation was 13.7 months. For a similar median progression-free survival after upfront radiotherapy (8.2 versus 7.7 months; P = .58), a significant benefit in median OS (13.7 versus 10.3 months; P = .04) was observed in favour of patients undergoing re-irradiation. Survival benefit of re-irradiation increased with a longer interval between end-of-radiotherapy and first progression (3-6 months: 4.0 versus 2.7; P < .01; 6-12 months: 6.4 versus 3.3; P = .04). Clinical improvement with re-irradiation was observed in 24/31 (77%) patients. No grade 4-5 toxicity was recorded. On multivariable analysis, interval to progression (corrected hazard ratio = .27-.54; P < .01) and re-irradiation (corrected hazard ratio = .18-.22; P < .01) remained prognostic for survival. A risk score (RS), comprising 5 categories, was developed to predict survival from first progression (ROC: .79). Median survival ranges from 1.0 month (RS-1) to 6.7 months (RS-5).

CONCLUSIONS

The majority of patients with DIPG, responding to upfront radiotherapy, do benefit of re-irradiation with acceptable tolerability.