Machine learning of clinical performance in a pancreatic cancer database

ObjectiveWe consider predictive models for clinical performance of pancreatic cancer patients based on machine learning techniques. The predictive performance of machine learning is compared with that of the linear and logistic regression techniques that dominate the medical oncology literature.Methods and materialsWe construct predictive models over a clinical database that we have developed for the University of Massachusetts Memorial Hospital in Worcester, Massachusetts, USA. The database contains retrospective records of 91 patient treatments for pancreatic tumors. Classification and regression targets include patient survival time, Eastern Cooperative Oncology Group (ECOG) quality of life scores, surgical outcomes, and tumor characteristics. The predictive performance of several techniques is described, and specific models are presented.ResultsWe show that machine learning techniques attain a predictive performance that is as good as, or better than, that of linear and logistic regression, for target attributes that include tumor N and T stage, survival time, and ECOG quality of life scores. Bayesian techniques are found to provide the best performance overall. For tumor size as the target attribute, however, logistic regression (respectively linear regression in the case of a numerical as opposed to discrete target) performs best. Preprocessing in the form of attribute selection and supervised attribute discretization improves predictive performance for most of the predictive techniques and target attributes considered.ConclusionMachine learning provides techniques for improved prediction of clinical performance. These techniques therefore merit consideration as valuable alternatives to traditional multivariate regression techniques in clinical medical studies.     

A prediction model based on clinical and histological features for predicting recurrence in patients with stage I-II endometrial cancer after surgical treatment

ObjectiveThe study aimed to develop a prediction model combining clinical and histological features to predict recurrence in patients with stage I-II endometrial cancer (EC) after surgical treatment.MethodsA total of 746 stage I-II EC patients who had received primary surgical treatment at Taizhou People's Hospital between 2014 and 2018 were included and randomly divided as a Training cohort (n = 520) and a Validation cohort (n = 226) at a 7:3 ratio. Clinical features including age, body mass index, comorbidities, lymphadenectomy, and adjuvant treatment, and histological features including histologic type, myometrial invasion, cervical stromal invasion, and expression levels of Ki67, estrogen receptor (ER), progesterone receptor (PR), whey acidic protein 4-disulphide core domain 2 (WFDC2), and p53 were used to develop a prediction model for EC recurrence in the Training cohort using a multivariable Cox regression model. Model discrimination and calibration were further evaluated in the Validation cohort.ResultsEC recurrence was observed in 60 (11.54%) patients in the Training cohort with a median length of follow-up of 39 months. Age, adjuvant treatment, histologic type, cervical stromal invasion, and expression levels of Ki67, ER, PR, and WFDC2 were factors significantly associated with EC recurrence based on univariable Cox regression analysis. After a model selection by AIC in a stepwise algorithm, the final model incorporated the above predictors showed a C-index of 0.85 and fair calibration in the Training cohort. In the Validation cohort, the model still showed good discrimination power (C-index 0.80) but moderate calibration.ConclusionsThe developed prediction model combining clinical and histological features can help to predict the EC recurrence in patients with stage I-II EC after surgical treatment.     

Epidermal Growth Factor Receptor Is Related to Poor Survival in Glioblastomas: Single-Institution Experience

Purpose

There are conflicting results surrounding the prognostic significance of epidermal growth factor receptor (EGFR) status in glioblastoma (GBM) patients. Accordingly, we attempted to assess the influence of EGFR expression on the survival of GBM patients receiving postoperative radiotherapy.

Materials and Methods

Thirty three GBM patients who had received surgery and postoperative radiotherapy at our institute, between March 1997 and February 2006, were included. The evaluation of EGFR expression with immunohistochemistry was available for 30 patients. Kaplan-Meier survival analysis and Cox regression were used for statistical analysis.

Results

EGFR was expressed in 23 patients (76.7%), and not expressed in seven (23.3%). Survival in EGFR expressing GBM patients was significantly less than that in non-expressing patients (median survival: 12.5 versus 17.5 months, p=0.013). Patients who received more than 60 Gy showed improved survival over those who received up to 60 Gy (median survival: 17.0 versus 9.0 months, p=0.000). Negative EGFR expression and a higher radiation dose were significantly correlated with improved survival on multivariate analysis. Survival rates showed no differences according to age, sex, and surgical extent.

Conclusion

The expression of EGFR demonstrated a significantly deleterious effect on the survival of GBM patients. Therefore, approaches targeting EGFR should be considered in potential treatment methods for GBM patients, in addition to current management strategies.     

基于3D-ResNet深度影像特征的胆囊癌生存预测模型

建立一个精准的个体化胆囊癌患者生存预测模型,分析、寻找新的胆囊癌预后因素,对于患者预后评估、治疗模式选择、手术患者筛选、术后辅助治疗方案确定及医疗资源合理使用均具有重要意义。本文提出一种基于3D-ResNet提取深度影像特征建立胆囊癌患者生存预后模型的方法,通过迁移学习以及训练3D-ResNet自动提取患者CT的深度特征,并利用提取的深度影像特征,通过Cox比例风险回归模型建立胆囊癌患者的生存预测模型。实验结果表明,基于深度影像特征建立的胆囊癌患者预后因子在预测患者生存时的C指数达到0.734,利用深度影像特征预后因子预测患者的1、3、5年存活率AUC分别达到0.833、0.791、0.813。本方法对胆囊癌预后预测有着良好的指示作用。     

Identification of a multidimensional transcriptome signature for survival prediction of postoperative glioblastoma multiforme patients

Background

Glioblastoma multiform (GBM) is a devastating brain tumor with maximum surgical resection, radiotherapy plus concomitant and adjuvant temozolomide (TMZ) as the standard treatment. Diverse clinicopathological and molecular features are major obstacles to accurate predict survival and evaluate the efficacy of chemotherapy or radiotherapy. Reliable prognostic biomarkers are urgently needed for postoperative GBM patients.

Methods

The protein coding genes (PCGs) and long non-coding RNA (lncRNA) gene expression profiles of 233 GBM postoperative patients were obtained from The Cancer Genome Atlas (TCGA), TANRIC and Gene Expression Omnibus (GEO) database. We randomly divided the TCGA set into a training (n?=?76) and a test set (n?=?77) and used GSE7696 (n?=?80) as an independent validation set. Survival analysis and the random survival forest algorithm were performed to screen survival associated signature.

Results

Six PCGs (EIF2AK3, EPRS, GALE, GUCY2C, MTHFD2, RNF212) and five lncRNAs (CTD-2140B24.6, LINC02015, AC068888.1, CERNA1, LINC00618) were screened out by a risk score model and formed a PCG-lncRNA signature for its predictive power was strongest (AUC?=?0.78 in the training dataset). The PCG-lncRNA signature could divide patients into high- risk or low-risk group with significantly different survival (median 7.47 vs. 18.27 months, log-rank test P?<?0.001) in the training dataset. Similar result was observed in the test dataset (median 11.40 vs. 16.80 months, log-rank test P?=?0.001) and the independent set (median 8.93 vs. 16.22 months, log-rank test P?=?0.007). Multivariable Cox regression analysis verified that it was an independent prognostic factor for the postsurgical patients with GBM. Compared with IDH mutation status, O-(6)-methylguanine DNA methyltransferase promoter methylation status and age, the signature was proved to have a superior predictive power. And stratified analysis found that the signature could further separated postoperative GBM patients who received TMZ-chemoradiation into high- and low-risk groups in TCGA and GEO dataset.

Conclusions

The PCG-lncRNA signature was a novel prognostic marker to predict survival and TMZ-chemoradiation response in GBM patients after surgery.

     

Creating hospital-specific customized clinical pathways by applying semantic reasoning to clinical data

ObjectiveClinical pathways (CPs) are widely studied methods to standardize clinical intervention and improve medical quality. However, standard care plans defined in current CPs are too general to execute in a practical healthcare environment. The purpose of this study was to create hospital-specific personalized CPs by explicitly expressing and replenishing the general knowledge of CPs by applying semantic analysis and reasoning to historical clinical data.MethodsA semantic data model was constructed to semantically store clinical data. After querying semantic clinical data, treatment procedures were extracted. Four properties were self-defined for local ontology construction and semantic transformation, and three Jena rules were proposed to achieve error correction and pathway order recognition. Semantic reasoning was utilized to establish the relationship between data orders and pathway orders.ResultsA clinical pathway for deviated nasal septum was used as an example to illustrate how to combine standard care plans and practical treatment procedures. A group of 224 patients with 11,473 orders was transformed to a semantic data model, which was stored in RDF format. Long term order processing and error correction made the treatment procedures more consistent with clinical practice. The percentage of each pathway order with different probabilities was calculated to declare the commonality between the standard care plans and practical treatment procedures. Detailed treatment procedures with pathway orders, deduced pathway orders, and orders with probability greater than 80% were provided to efficiently customize the CPs.ConclusionsThis study contributes to the practical application of pathway specifications recommended by the Ministry of Health of China and provides a generic framework for the hospital-specific customization of standard care plans defined by CPs or clinical guidelines.     

Sarcopenia as a Predictor of Prognosis in Early Stage Ovarian Cancer

BackgroundTo identify sarcopenia as a predictive prognostic factor of ovarian cancer in terms of survival outcome in patients with early-stage ovarian cancer.MethodsData of Konkuk University Medical Center from March 2002 to December 2017 were reviewed retrospectively. Eighty-two patients who underwent surgery due to early-stage (International Federation of Gynecology and Obstetrics stage I/II) ovarian cancer and had computed tomography (CT) images taken at the initial diagnosis were included. The initial CT scan images were analyzed with SliceOmatic software (TomoVision). A sarcopenia cutoff value was defined as a skeletal muscle index of ≤ 38.7 cm²/m². Overall survival (OS) times were compared according to the existence of sarcopenia, and subgroup analyses were performed.ResultsA Kaplan-Meier analysis showed a significant survival disadvantage for patients with early-stage ovarian cancer when they had sarcopenia (P < 0.001; log-rank test). Sarcopenia remained a significant prognostic factor for OS in early-stage ovarian cancer, in a Cox proportional hazards model regression analysis (HR, 21.9; 95% CI, 2.0–199.9; P = 0.006).ConclusionThis study demonstrated that sarcopenia was predictive of OS in patients with early-stage ovarian cancer. Further prospective studies with a larger number of patients are warranted to determine the extent to which sarcopenia can be used as a prognostic factor in ovarian cancer.     

Predicting the graft survival for heart-lung transplantation patients: An integrated data mining methodology

Background

Predicting the survival of heart-lung transplant patients has the potential to play a critical role in understanding and improving the matching procedure between the recipient and graft. Although voluminous data related to the transplantation procedures is being collected and stored, only a small subset of the predictive factors has been used in modeling heart-lung transplantation outcomes. The previous studies have mainly focused on applying statistical techniques to a small set of factors selected by the domain-experts in order to reveal the simple linear relationships between the factors and survival. The collection of methods known as ‘data mining’ offers significant advantages over conventional statistical techniques in dealing with the latter's limitations such as normality assumption of observations, independence of observations from each other, and linearity of the relationship between the observations and the output measure(s). There are statistical methods that overcome these limitations. Yet, they are computationally more expensive and do not provide fast and flexible solutions as do data mining techniques in large datasets.

Purpose

The main objective of this study is to improve the prediction of outcomes following combined heart-lung transplantation by proposing an integrated data-mining methodology.

Methods

A large and feature-rich dataset (16,604 cases with 283 variables) is used to (1) develop machine learning based predictive models and (2) extract the most important predictive factors. Then, using three different variable selection methods, namely, (i) machine learning methods driven variables—using decision trees, neural networks, logistic regression, (ii) the literature review-based expert-defined variables, and (iii) common sense-based interaction variables, a consolidated set of factors is generated and used to develop Cox regression models for heart-lung graft survival.

Results

The predictive models’ performance in terms of 10-fold cross-validation accuracy rates for two multi-imputed datasets ranged from 79% to 86% for neural networks, from 78% to 86% for logistic regression, and from 71% to 79% for decision trees. The results indicate that the proposed integrated data mining methodology using Cox hazard models better predicted the graft survival with different variables than the conventional approaches commonly used in the literature. This result is validated by the comparison of the corresponding Gains charts for our proposed methodology and the literature review based Cox results, and by the comparison of Akaike information criteria (AIC) values received from each.

Conclusions

Data mining-based methodology proposed in this study reveals that there are undiscovered relationships (i.e. interactions of the existing variables) among the survival-related variables, which helps better predict the survival of the heart-lung transplants. It also brings a different set of variables into the scene to be evaluated by the domain-experts and be considered prior to the organ transplantation.     

Risk assessment in localized primary cutaneous melanoma: a Southwest Oncology Group study evaluating nine factors and a test of the Clark logistic regression prediction model

We studied 9 clinical and pathologic factors in 259 patients using Cox model regression analysis to determine which factors have independent predictive value. Median follow-up time in all patients still alive was 12.3 years (range, 1.7 to 16.7 years). Tumor-infiltrating lymphocytes (P = .005), primary site (P = .006), and thickness (P = .02) had independent predictive value. Ulceration (P = .06) and age (P = .07) had marginal value. We used 6 of those factors to test the Clark logistic regression prediction model, which accurately predicted 8-year survival in 121 (72.9%) of 166 patients and accurately predicted melanoma-specific mortality in 32 (43%) of 74 patients. The combined or overall accuracy of the Clark model was only 64%.     

Mining association patterns of drug-interactions using post marketing FDA’s spontaneous reporting data

Background and objectivesPharmacovigilance (PhV) is an important clinical activity with strong implications for population health and clinical research. The main goal of PhV is the timely detection of adverse drug events (ADEs) that are novel in their clinical nature, severity and/or frequency. Drug interactions (DI) pose an important problem in the development of new drugs and post marketing PhV that contribute to 6–30% of all unexpected ADEs. Therefore, the early detection of DI is vital. Spontaneous reporting systems (SRS) have served as the core data collection system for post marketing PhV since the 1960s. The main objective of our study was to particularly identify signals of DI from SRS. In addition, we are presenting an optimized tailored mining algorithm called “hybrid Apriori”.MethodsThe proposed algorithm is based on an optimized and modified association rule mining (ARM) approach. A hybrid Apriori algorithm has been applied to the SRS of the United States Food and Drug Administration’s (U.S. FDA) adverse events reporting system (FAERS) in order to extract significant association patterns of drug interaction-adverse event (DIAE). We have assessed the resulting DIAEs qualitatively and quantitatively using two different triage features: a three-element taxonomy and three performance metrics. These features were applied on two random samples of 100 interacting and 100 non-interacting DIAE patterns. Additionally, we have employed logistic regression (LR) statistic method to quantify the magnitude and direction of interactions in order to test for confounding by co-medication in unknown interacting DIAE patterns.ResultsHybrid Apriori extracted 2933 interacting DIAE patterns (including 1256 serious ones) and 530 non-interacting DIAE patterns. Referring to the current knowledge using four different reliable resources of DI, the results showed that the proposed method can extract signals of serious interacting DIAEs. Various association patterns could be identified based on the relationships among the elements which composed a pattern. The average performance of the method showed 85% precision, 80% negative predictive value, 81% sensitivity and 84% specificity. The LR modeling could provide the statistical context to guard against spurious DIAEs.ConclusionsThe proposed method could efficiently detect DIAE signals from SRS data as well as, identifying rare adverse drug reactions (ADRs).     

Mobile Phone-Based Unobtrusive Ecological Momentary Assessment of Day-to-Day Mood: An Explorative Study

BackgroundEcological momentary assessment (EMA) is a useful method to tap the dynamics of psychological and behavioral phenomena in real-world contexts. However, the response burden of (self-report) EMA limits its clinical utility.ObjectiveThe aim was to explore mobile phone-based unobtrusive EMA, in which mobile phone usage logs are considered as proxy measures of clinically relevant user states and contexts.MethodsThis was an uncontrolled explorative pilot study. Our study consisted of 6 weeks of EMA/unobtrusive EMA data collection in a Dutch student population (N=33), followed by a regression modeling analysis. Participants self-monitored their mood on their mobile phone (EMA) with a one-dimensional mood measure (1 to 10) and a two-dimensional circumplex measure (arousal/valence, –2 to 2). Meanwhile, with participants’ consent, a mobile phone app unobtrusively collected (meta) data from six smartphone sensor logs (unobtrusive EMA: calls/short message service (SMS) text messages, screen time, application usage, accelerometer, and phone camera events). Through forward stepwise regression (FSR), we built personalized regression models from the unobtrusive EMA variables to predict day-to-day variation in EMA mood ratings. The predictive performance of these models (ie, cross-validated mean squared error and percentage of correct predictions) was compared to naive benchmark regression models (the mean model and a lag-2 history model).ResultsA total of 27 participants (81%) provided a mean 35.5 days (SD 3.8) of valid EMA/unobtrusive EMA data. The FSR models accurately predicted 55% to 76% of EMA mood scores. However, the predictive performance of these models was significantly inferior to that of naive benchmark models.ConclusionsMobile phone-based unobtrusive EMA is a technically feasible and potentially powerful EMA variant. The method is young and positive findings may not replicate. At present, we do not recommend the application of FSR-based mood prediction in real-world clinical settings. Further psychometric studies and more advanced data mining techniques are needed to unlock unobtrusive EMA’s true potential.     

DNA methyltransferase 3a rs1550117 genetic polymorphism predicts poor survival in gastric cancer patients

DNA methyltransferase 3a (DNMT3a) have been suggested to play a crucial role in human cancer prognosis. Single nucleotide polymorphisms (SNPs) in DNMT3a genes may have an impact on the prognosis of cancers. This study aimed to investigate the association between SNPs of DNMT3a gene and prognosis of gastric cancer (GC). Two sites of DNMT3a SNPs, rs1550117 and rs13420827 were selected and genotyped using TaqMan assay in 447 GC patients who received gastrectomy. Effects of genotypes on clinical outcomes of GC were calculated by Kaplan-Meier survival analysis and Cox regression model. We found that the AG or AA genotype of rs1550117 was associated with significantly poorer survival and increased death risk of GC compared with GG genotype (dominant model: HR=1.35, 95% CI=1.01-1.80, P=0.043). Further multivariate Cox regression analysis revealed that in addition to the known factors including male, larger tumor sizes and high clinical stage, rs1550117 variant was an independently predictive factor for survival in GC patients. No significant association was found between rs13420827 genetic variants and GC prognosis. Our findings first demonstrated that DNMT3a rs1550117 polymorphism may be a potential biomarker in predicting overall survival of GC patients.     

Molecular subtyping for lung adenocarcinoma and a novel prognostic model based on ligand-receptor pairs

PurposeLung adenocarcinoma (LUAD) is a leading cause of cancer death worldwide. Ligands and receptors play important roles in cell communication. This study aimed to demonstrate the importance of ligand-receptor (LR) pairs in LUAD development through constructing molecular subtypes and a prognostic model based on LR pairs.Materials and methodsA total of 1110 LUAD samples with clinical and expression data were obtained from public databases. Unsupervised consensus clustering was applied to construct molecular subtypes based on LR pairs. Least absolute shrinkage and selection operator (LASSO) Cox regression and stepwise Akaike information criterion (stepAIC) were conducted to build a prognostic model.ResultsThree molecular subtypes (C1, C2 and C3) were constructed based on 17 prognosis-related LR pairs. C1 subtype had the worst prognosis, while C3 subtype had the optimal prognosis. Oncogenic pathways such as epithelial-mesenchymal transition (EMT) were activated in C1 subtype. A prognostic model was built based on 8 LR pairs, and could classify samples into high- and low-LR score groups. Two groups had distinct overall survival and tumor microenvironment (TME). High-LR score group was more sensitive to chemotherapeutic drugs, while low-LR score group could benefit much from anti-PD-1/PD-L1 therapy.ConclusionsThe study showed that LR pairs played critical roles in LUAD development. The prognostic model could predict prognosis and guide personalized therapy for LUAD patients.     

Predicting Postoperative Complications and Long-Term Survival After Lung Cancer Surgery Using Eurolung Risk Score

BackgroundThis study aimed to assess the clinical relevance of the parsimonious Eurolung risk scoring system for predicting postoperative morbidity, mortality, and long-term survival in Korean patients with surgically resected non-small cell lung cancer.MethodsThis retrospective analysis used the data of patients who underwent anatomical resection for non-small cell lung cancer between 2004 and 2018 at a single institution. The parsimonious aggregate Eurolung score was calculated for each patient. The Cox regression model was used to determine the ability of the Eurolung scoring system for predicting long-term outcomes.ResultsOf the 7,278 patients in the study, cardiopulmonary complications and mortality occurred in 687 (9.4%) and 53 (0.7%) patients, respectively. The rate of cardiopulmonary complications and mortality gradually increased with the increase in the Eurolung risk scores (all P < 0.001). When risk scores were grouped into four categories, the Eurolung scoring system showed a stepwise deterioration of overall survival with the increase in risk scores, and this association was statistically significant (P < 0.001). Multivariate Cox analysis showed that the Eurolung scoring system, classified into four categories, was a significant prognostic factor of overall survival even after adjusting for covariates such as tumor histology and pathological stage (P < 0.001).ConclusionStratification based on the parsimonious Eurolung scoring system showed good discriminatory ability for predicting postoperative morbidity, mortality, and long-term survival in South Korean patients with surgically resected non-small cell lung cancer. This might help clinicians to provide a detailed prognosis and decide the appropriate treatment option for high-risk patients with non-small cell lung cancer.     

Prediction of COVID-19-related Mortality and 30-Day and 60-Day Survival Probabilities Using a Nomogram

BackgroundPrediction of mortality in patients with coronavirus disease 2019 (COVID-19) is a key to improving the clinical outcomes, considering that the COVID-19 pandemic has led to the collapse of healthcare systems in many regions worldwide. This study aimed to identify the factors associated with COVID-19 mortality and to develop a nomogram for predicting mortality using clinical parameters and underlying diseases.MethodsThis study was performed in 5,626 patients with confirmed COVID-19 between February 1 and April 30, 2020 in South Korea. A Cox proportional hazards model and logistic regression model were used to construct a nomogram for predicting 30-day and 60-day survival probabilities and overall mortality, respectively in the train set. Calibration and discrimination were performed to validate the nomograms in the test set.ResultsAge ≥ 70 years, male, presence of fever and dyspnea at the time of COVID-19 diagnosis, and diabetes mellitus, cancer, or dementia as underling diseases were significantly related to 30-day and 60-day survival and mortality in COVID-19 patients. The nomogram showed good calibration for survival probabilities and mortality. In the train set, the areas under the curve (AUCs) for 30-day and 60-day survival was 0.914 and 0.954, respectively; the AUC for mortality of 0.959. In the test set, AUCs for 30-day and 60-day survival was 0.876 and 0.660, respectively, and that for mortality was 0.926. The online calculators can be found at https://koreastat.shinyapps.io/RiskofCOVID19/.ConclusionThe prediction model could accurately predict COVID-19-related mortality; thus, it would be helpful for identifying the risk of mortality and establishing medical policies during the pandemic to improve the clinical outcomes.     

The association of hepatitis C virus infection with the prognosis of chronic hemodialysis patients: A retrospective study of 3,064 patients between 1999 and 2010

The prevalence of hepatitis C virus (HCV) infection is high among patients receiving chronic hemodialysis. To clarify the risk ratio of HCV infection with respect to mortality and prognosis in chronic hemodialysis patients, a retrospective longitudinal cohort study was conducted in 2010 and involved 3,064 patients receiving chronic hemodialysis at nine dialysis facilities in Hiroshima, Japan, who were recruited from 1999 to 2003. Logistic regression and Cox hazards models were used to estimate the mortality risk among hemodialysis patients. Among the patients, 422 (14.0%) were positive for HCV RNA. HCV RNA positivity was associated with death in the logistic model (adjusted odds ratio = 1.79; P < 0.001). However, it was not a risk factor for the reduced of survival rate in the Cox proportional hazard model (adjusted risk ratio = 1.07; P = 0.4138). In summary, among hemodialysis patients, HCV RNA is correlated with the mortality rate; however, it is not significantly correlated with prognosis in terms of survival time. On the other hand, diabetes and age at dialysis onset are significantly correlated with survival. Diabetes control treatment should be preferentially selected for hemodialysis patients, and antiviral therapy for HCV should be introduced based on the clinical state of the patient. J. Med. Virol. 87:1558–1564, 2015. © 2015 The Authors. Journal of Medical Virology published by Wiley Periodicals, Inc.     

Prognostic implications of immune-related eight-gene signature in pediatric brain tumors

Genomic studies have provided insights into molecular subgroups and oncogenic drivers of pediatric brain tumors (PBT) that may lead to novel therapeutic strategies. Participants of the cohort Pediatric Brain Tumor Atlas: CBTTC (CBTTC cohort), were randomly divided into training and validation cohorts. In the training cohort, Kaplan-Meier analysis and univariate Cox regression model were applied to preliminary screening of prognostic genes. The LASSO Cox regression model was implemented to build a multi-gene signature, which was then validated in the validation and CBTTC cohorts through Kaplan-Meier, Cox, and receiver operating characteristic curve (ROC) analyses. Also, gene set enrichment analysis (GSEA) and immune infiltrating analyses were conducted to understand function annotation and the role of the signature in the tumor microenvironment. An eight-gene signature was built, which was examined by Kaplan-Meier analysis, revealing that a significant overall survival difference was seen, either in the training or validation cohorts. The eight-gene signature was further proven to be independent of other clinic-pathologic parameters via the Cox regression analyses. Moreover, ROC analysis demonstrated that this signature owned a better predictive power of PBT prognosis. Furthermore, GSEA and immune infiltrating analyses showed that the signature had close interactions with immune-related pathways and was closely related to CD8 T cells and monocytes in the tumor environment. Identifying the eight-gene signature (CBX7, JADE2, IGF2BP3, OR2W6P, PRAME, TICRR, KIF4A, and PIMREG) could accurately identify patients'' prognosis and the signature had close interactions with the immunodominant tumor environment, which may provide insight into personalized prognosis prediction and new therapies for PBT patients.