Enhancing explainability of random survival forests in predicting stent patency risk for malignant colonic obstruction

Hepatocellular carcinoma (HCC) exhibits a propensity for early recurrence following liver resection, resulting in a bleak prognosis. At present, majority of the predictive models for the early postoperative recurrence of HCC rely on the linear assumption of the Cox Proportional Hazard (CPH) model. However, the predictive efficacy of this model is constrained by the intricate nature of clinical data. The present study aims to investigate the efficacy of the random survival forest (RSF) model, which is a machine learning algorithm, in predicting the early postoperative recurrence of HCC, and compare its performance with that of the traditional CPH model. This analysis seeks to elucidate the potential advantages of the RSF model over the CPH model in addressing this clinical challenge. The present retrospective cohort study was conducted at a single center. After excluding 41 patients, a total of 541 patients were included in the final model construction and subsequent analysis. The patients were randomly divided into two groups at a 7:3 ratio: training group (n = 378) and validation group (n = 163). The least absolute shrinkage and selection operator (LASSO) regression was used to identify the risk factors in the training group. Then, the identified factors were used to develop the RSF and CPH regression models. The predictive ability of the model was assessed using the concordance index (C-index). The accuracy of the model predictions was evaluated using the receiver operating characteristic curve (ROC) and area under the receiver operating characteristic curve (AUC). The clinical practicality of the model was measured by decision curve analysis (DCA), and the overall performance of the model was evaluated using the Brier score. The RSF model was visually represented using the Shapley additive explanations (SHAP) framework. Then, the RSF, CPH regression, and albumin-bilirubin (ALBI) grade models were compared. The following variables were examined by LASSO regression: alpha fetoprotein (AFP), gamma-glutamyl transpeptidase to platelet ratio (GPR), blood transfusion (BT), microvascular invasion (MVI), large vessel invasion (LVI), Edmondson-Steiner (ES) grade, liver capsule invasion (LCI), satellite nodule (SN), and Barcelona clinic liver cancer (BCLC) grade. Then, a RSF model was developed using 500 trees, and the variable importance (VIMP) ranking was MVI, LCI, SN, BT, BCLC, ESG, AFP, GPR and LVI. After these aforementioned factors were applied, the RSF and CPH regression models were developed and compared using the ALBI grade model. The C-index for the RSF model (0.896 and 0.798, respectively) outperformed that of the CPH regression model (0.803 and 0.772, respectively) and ALBI grade model (0.517 and 0.515, respectively), in both the training and validation groups. Three time points were selected to assess the predictive capabilities of these models: 6, 12 and 18 months. For the training group, the AUC value for the RSF model at 6, 12 and 18 months was 0.971 (95% CI: 0.955-0.988), 0.919 (95% CI: 0.887-0.951) and 0.899 (95% CI: 0.867-0.932), respectively. For the validation cohort, the AUC value for the RSF model at 6, 12 and 18 months was 0.830 (95% CI: 0.728-0.932), 0.856 (95% CI: 0.787-0.924) and 0.832 (95% CI: 0.764-0.901), respectively. The AUC values were higher in the RSF model, when compared to the CPH regression model and ALBI grade model, in both groups. The DCA results revealed that the net clinical benefits associated to the RSF model were superior to those associated to the CPH regression model and ALBI grade model in both groups, suggesting a higher level of clinical utility in the RSF model. The Brier score for the RSF model at 6, 12 and 18 months was 0.062, 0.125 and 0.178, respectively, in the training group, and 0.111, 0.128 and 0.149, respectively, in the validation group. In summary, the RSF model demonstrated superior performance, when compared to the CPH regression model and ALBI grade model. Furthermore, the RSF model demonstrated superior predictive ability, accuracy, clinical practicality, and overall performance, when compared to the CPH regression model and ALBI grade model. In addition, the RSF model was able to successfully stratify patients into three distinct risk groups (low-risk, medium-risk and high-risk) in both groups (p < 0.001). The RSF model demonstrates efficacy in predicting early recurrence following HCC surgery, exhibiting superior performance, when compared to the CPH regression model and ALBI grade model. For patients undergoing HCC surgery, the RSF model can serve as a valuable tool for clinicians to postoperatively stratify patients into distinct risk categories, offering guidance for subsequent follow-up care.

Introduction: This study aimed to assess the long-term renal prognosis of patients with hypertensive nephropathy (HN) diagnosed through renal biopsy, utilizing the random survival forest (RSF) algorithm. Methods: From December 2010 to December 2022, HN patients diagnosed by renal biopsy in Xijing Hospital were enrolled and randomly divided into training set and testing set at a ratio of 7∶3. The study’s composite endpoint was defined as a ≥50% decline in estimated glomerular filtration rate (eGFR), end-stage renal disease, or death. RSF and Cox regression were used to establish a renal prognosis prediction model based on the factors screened by the RSF algorithm. The Concordance index (C-index), integrated Brier score, net reclassification improvement (NRI), and integrated discrimination improvement (IDI) were used to evaluate discrimination, calibration, and risk classification, respectively. Results: A total of 225 patients were included in this study, with 72 (32.0%) patients experiencing combined events after a median follow-up of 29.9 (16.6, 52.1) months. Six eligible variables (overall chronicity grade of renal pathology, eGFR, high-density lipoprotein cholesterol, hematocrit, monocyte, and stroke volume) were selected from clinical data and introduced into the RSF model. The RSF model had a higher C-index in both the training set (0.904 [95% CI: 0.842–0.938] vs. 0.831 [95% CI: 0.768–0.894], p < 0.001) and the testing set (0.893 [95% CI: 0.770–0.944] vs. 0.841 [95% CI: 0.751–0.931], p = 0.021) compared to the Cox model. NRI and IDI indicated that the RSF model outperformed the Cox model regarding risk classification. Conclusion: In this study, the RSF algorithm was employed to identify the risk factors affecting the prognosis of HN patients, and a clinical prognostic RSF model was constructed to predict the adverse outcomes of HN patients based on renal pathology. Compared to the traditional Cox regression model, the RSF model offers superior performance and can provide valuable new insights for clinical diagnosis and treatment strategies.

Random survival forests-based survival prediction for spinal chordomas.

BackgroundElderly patients with sepsis have many comorbidities, and the clinical reaction is not obvious. Thus, clinical treatment is difficult. We planned to use the laboratory test results and comorbidities of elderly patients with sepsis from a large-scale public database Medical Information Mart for Intensive Care (MIMIC) IV to build a random survival forest (RSF) model and to evaluate the model’s predictive value for these patients.MethodsClinical information of elderly patients with sepsis in MIMIC IV database was collected retrospectively. Machine learning (RSF) was used to select the top 30 variables in the training cohort to build the final RSF model. The model was compared with the traditional scoring systems SOFA, SAPSII, and APSIII. The performance of the model was evaluated by C index and calibration curve.ResultsA total of 6,503 patients were enrolled in the study. The top 30 important variables screened by RSF were used to construct the final RSF model. The new model provided a better C-index (0.731 in the validation cohort). The calibration curve described the agreement between the predicted probability of RSF model and the observed 30-day survival.ConclusionsWe constructed a prognostic model to predict a 30-day mortality risk in elderly patients with sepsis based on machine learning (RSF algorithm), and it proved superior to the traditional scoring systems. The risk factors affecting the patients were also ranked. In addition to the common risk factors of vasopressors, ventilator use, and urine output. Newly added factors such as RDW, type of ICU unit, malignant cancer, and metastatic solid tumor also significantly influence prognosis.

A novel model for predicting prognosis in patients with metastatic major salivary gland carcinoma.

BackgroundThe present study aimed to explore the application value of random survival forest (RSF) model and Cox model in predicting the progression-free survival (PFS) among patients with locoregionally advanced nasopharyngeal carcinoma (LANPC) after induction chemotherapy plus concurrent chemoradiotherapy (IC + CCRT).MethodsEligible LANPC patients underwent magnetic resonance imaging (MRI) scan before treatment were subjected to radiomics feature extraction. Radiomics and clinical features of patients in the training cohort were subjected to RSF analysis to predict PFS and were tested in the testing cohort. The performance of an RSF model with clinical and radiologic predictors was assessed with the area under the receiver operating characteristic (ROC) curve (AUC) and Delong test and compared with Cox models based on clinical and radiologic parameters. Further, the Kaplan-Meier method was used for risk stratification of patients.ResultsA total of 294 LANPC patients (206 in the training cohort; 88 in the testing cohort) were enrolled and underwent magnetic resonance imaging (MRI) scans before treatment. The AUC value of the clinical Cox model, radiomics Cox model, clinical + radiomics Cox model, and clinical + radiomics RSF model in predicting 3- and 5-year PFS for LANPC patients was [0.545 vs 0.648 vs 0.648 vs 0.899 (training cohort), and 0.566 vs 0.736 vs 0.730 vs 0.861 (testing cohort); 0.556 vs 0.604 vs 0.611 vs 0.897 (training cohort), and 0.591 vs 0.661 vs 0.676 vs 0.847 (testing cohort), respectively]. Delong test showed that the RSF model and the other three Cox models were statistically significant, and the RSF model markedly improved prediction performance (P < 0.001). Additionally, the PFS of the high-risk group was lower than that of the low-risk group in the RSF model (P < 0.001), while comparable in the Cox model (P > 0.05).ConclusionThe RSF model may be a potential tool for prognostic prediction and risk stratification of LANPC patients.

Papillary thyroid carcinoma (PTC) is a common malignancy whose incidence is three times greater in females than in males. The prognosis of ageing patients is poor. This research was designed to construct models to predict the overall survival of elderly female patients with PTC. We developed prediction models based on the random survival forest (RSF) algorithm and traditional Cox regression. The data of 4539 patients were extracted from the Surveillance, Epidemiology, and End Results (SEER) database. Twelve variables were analysed to establish the models. The C-index and the Brier score were selected to evaluate the discriminatory ability of the models. Time-dependent receiver operating characteristic (ROC) curves were also drawn to evaluate the accuracy of the models. The clinical benefits of the two models were compared on the basis of the DCA curve. In addition, the Shapley Additive Explanations (SHAP) plot was used to visualize the contribution of the variables in the RSF model. The C-index of the RSF model was 0.811, which was greater than that of the Cox model (0.781). According to the Brier score and the area under the ROC curve (AUC), the RSF model performed better than the Cox model. On the basis of the DCA curve, the RSF model demonstrated fair clinical benefit. The SHAP plot showed that age was the most important variable contributing to the outcome of PTC in elderly female patients. The RSF model we developed performed better than the Cox model and might be valuable for clinical practice.

BackgroundWilms tumor (WT) is the most common malignant renal tumor in children. Despite advances in treatment, accurate prediction of the long-term prognosis remains challenging. Various Cox regression-based models have been developed to assess WT survival rates; however, there is a pressing need for more precise tools.MethodsData from the SEER database (2000–2021) and external validation data from Chongqing Medical University Children’s Hospital were utilized. Key prognostic factors for children with WT were identified via last absolute shrinkage and selection operator (LASSO) regression, which was subsequently used to construct the Random Survival Forest(RSF) model for long-term survival prediction. SHAP were applied to enhance the interpretability of the model. The model performance was compared to that of conventional Cox models via calibration curves, the Concordance index (C-index), the net reclassification index (NRI), and the integrated discrimination index (IDI).ResultsWe included 1,629 children with WT from the SEER database and externally validated the model via data from 169 children at Children’s Hospital of Chongqing Medical University(CHCMU). Kaplan‒Meier curves revealed higher mortality rates for Chinese children with WT than for their counterparts in the United States. LASSO regression identified six key variables for the development of the RSF and Cox models. The SHAP method was utilized to rank these variables in descending order of importance: tumor stage, age, lymph node density(LND), tumor metastasis, number of positive lymph nodes, and laterality (unilateral/bilateral). The RSF model demonstrated superior predictive performance and generalizability, as indicated by Brier scores, calibration curves, AUC curves, and risk curves. Moreover, the RSF model significantly outperformed the Cox model in terms of prediction accuracy (C-index: 0.868 vs. 0.759), with substantial improvements in the NRI and IDI (P < 0.01). Decision curve analysis also revealed that the RSF model provided a greater net benefit at 3, 5, and 7 years than did the Cox model, which underscored the greater clinical utility of the RSF model. Sensitivity analysis based on imputed data confirmed the robustness of the model, with consistent predictor selection and comparable performance metrics, further supported the stability and reliability of the RSF framework.ConclusionWe successfully developed a robust machine learning model that accurately assesses key prognostic factors affecting the long-term survival of children with WT. This model offers substantial clinical value for risk stratification and can assist clinicians in making more informed treatment decisions. By applying SHAP analysis, physicians can better understand the critical factors influencing WT prognosis and tailor intervention strategies more precisely.Supplementary InformationThe online version contains supplementary material available at 10.1007/s12672-025-03645-2.

Malignant colorectal obstruction: looking for synchronous lesions with the scope through a metal stent…!

Many researches have applied machine learning methods to find associations between radiomic features and clinical outcomes. Random survival forests (RSF), as an accurate classifier, sort all candidate variables as the rank of importance values. There was no study concerning on finding radiomic predictors in patients with extremity and trunk wall soft-tissue sarcomas using RSF. This study aimed to determine associations between radiomic features and overall survival (OS) by RSF analysis. To identify radiomic features with important values by RSF analysis, construct predictive models for OS incorporating clinical characteristics, and evaluate models' performance with different method. We collected clinical characteristics and radiomic features extracted from plain and contrast-enhanced computed tomography (CT) from 353 patients with extremity and trunk wall soft-tissue sarcomas treated with surgical resection. All radiomic features were analyzed by Cox proportional hazard (CPH) and followed RSF analysis. The association between radiomics-predicted risks and OS was assessed by Kaplan-Meier analysis. All clinical features were screened by CPH analysis. Prognostic clinical and radiomic parameters were fitted into RSF and CPH integrative models for OS in the training cohort, respectively. The concordance indexes (C-index) and Brier scores of both two models were evaluated in both training and testing cohorts. The model with better predictive performance was interpreted with nomogram and calibration plots. Among all 86 radiomic features, there were three variables selected with high importance values. The RSF on these three features distinguished patients with high predicted risks from patients with low predicted risks for OS in the training set (P < 0.001) using Kaplan-Meier analysis. Age, lymph node involvement and grade were incorporated into the combined models for OS (P < 0.05). The C-indexes in both two integrative models fluctuated above 0.80 whose Brier scores maintained less than 15.0 in the training and testing datasets. The RSF model performed little advantages over the CPH model that the calibration curve of the RSF model showed favorable agreement between predicted and actual survival probabilities for the 3-year and 5-year survival prediction. The multimodality RSF model including clinical and radiomic characteristics conducted high capacity in prediction of OS which might assist individualized therapeutic regimens. Level III, prognostic study.

Early recurrence is common for hepatocellular carcinoma (HCC) after surgical resection, being the leading cause of death. Traditionally, the COX proportional hazard (CPH) models based on linearity assumption have been used to predict early recurrence, but predictive performance is limited. Machine learning models offer a novel methodology and have several advantages over CPH models. Hence, the purpose of this study was to compare random survival forests (RSF) model with CPH models in prediction of early recurrence for HCC patients after curative resection. A total of 4,758 patients undergoing curative resection from two medical centers were included. Fifteen features including age, gender, etiology, platelet count, albumin, total bilirubin, AFP, tumor size, tumor number, microvascular invasion, macrovascular invasion, Edmondson-Steiner grade, tumor capsular, satellite nodules and liver cirrhosis were used to construct the RSF model in training cohort. Discrimination, calibration, clinical usefulness and overall performance were assessed and compared with other models. Five hundred survival trees were used to generate the RFS model. The five highest Variable Importance (VIMP) were tumor size, macrovascular invasion, microvascular invasion, tumor number and AFP. In training, internal and external validation cohort, the C-index of RSF model were 0.725 [standard errors (SE) =0.005], 0.762 (SE =0.011) and 0.747 (SE =0.016), respectively; the Gönen & Heller's K of RSF model were 0.684 (SE =0.005), 0.711 (SE =0.008) and 0.697 (SE =0.014), respectively; the time-dependent AUC (2 years) of RSF model were 0.818 (SE =0.008), 0.823 (SE =0.014) and 0.785 (SE =0.025), respectively. The RSF model outperformed early recurrence after surgery for liver tumor (ERASL) model, Korean model, American Joint Committee on Cancer tumor-node-metastasis (AJCC TNM) stage, Barcelona Clinic Liver Cancer (BCLC) stage and Chinese stage. The RSF model is capable of stratifying patients into three different risk groups (low-risk, intermediate-risk, high-risk groups) in the training and two validation cohorts (all P<0.0001). A web-based prediction tool was built to facilitate clinical application (https://recurrenceprediction.shinyapps.io/surgery_predict/). The RSF model is a reliable tool to predict early recurrence for patients with HCC after curative resection because it exhibited superior performance compared with other models. This novel model will be helpful to guide postoperative follow-up and adjuvant therapy.

Cervical cancer (CCa) significantly affects female fertility and quality of life. This study aimed to construct and validate a random survival forest (RSF) model to identify the factors that affect the overall survival (OS) in patients with CCa in China and compare its performance with that of the Cox proportional hazards model (Cox model). Data on CCa patients were collected from Chongqing University Cancer Hospital. The performance and discrimination ability of the models were evaluated via the C-index, integrated Brier score (IBS), accuracy, sensitivity, specificity, and area under the receiver operating characteristic curve (AUC). The Kaplan-Meier (K-M) survival curve was used to analyze the difference in OS between patients with high and low risk predicted by RSF model. A total of 3,982 patients were included in this study. Comparing to Cox model, the RSF model ranked important variables and identified radiotherapy (RT) as an important treatment measure. A comprehensive analysis of the evaluation indices confirmed that the RSF model outperformed the Cox model (IBS: 0.152 vs. 0.162, C-index: 0.863 vs. 0.764). The RSF model metrics for the validation cohort (VC) were as follows: 1-, 3-, and 5-year AUC (0.908, 0.884, and 0.869), sensitivity (0.746), specificity (0.825), and accuracy (0.808). The OS of low-risk patients predicted by RSF was greater than that of high-risk patients. The RSF model demonstrated excellent discrimination, calibrated predictions, and stratified risk for CCa patients. Furthermore, it outperformed the Cox model in predicting risks, thus enabling the delivery of personalised treatment and follow-up strategies.

PurposeTo evaluate machine learning–based survival model roles in predicting rehospitalization after hip fractures to improve reduce the burden on the healthcare system.MethodsThis retrospective cohort study examined 718 patients with hip fractures hospitalized at the Daejeon Eulji Medical Center between January 2020 and June 2022. Demographic and clinical variables, and rehospitalization data were collected at 6 weeks and 3, 6, 12, and 24 months. Cox proportional hazards (CoxPH), random survival forest (RSF), gradient boosting (GB), and fast survival support vector machine (SVM) models were developed.Model performance was assessed using the concordance index (c-index), area under the curve (AUC), and Kaplan–Meier survival curves. Feature importance was analyzed using permutation importance, with the best model selected based on overall performance.ResultsHyperparameter tuning optimized the models. The GB model had the highest mean AUC of 0.868, followed by the RSF (0.785), SVM (0.763), and CoxPH (0.736) models. Feature importance analysis highlighted femoral neck T-score, age, body mass index, operation time, compression fracture, and total calcium as significant predictors. Feature selection improved the c-index for the RSF model from 0.742 to 0.874 and CoxPH model from 0.717 to 0.915; the GB and SVM models exhibited a c-index decline post-feature selection. The GB and RSF models predicted lower rehospitalization probabilities than Kaplan–Meier estimates; the CoxPH model’s predictions were closely aligned with the observed data.ConclusionsThe effect of feature selection on model performance highlights the need for comprehensive variable selection and model evaluation strategies to improve predictive accuracy.

e13554 Background: Quantitative 18F-NaF PET/CT imaging metrics have been shown to be prognostic in metastatic prostate cancer (mPC) patients. However, previous studies have shown conflicting results in which metrics could be prognostic. This study investigates if current methods from literature generalize to external datasets and explores which combination of features are necessary to for survival models to generalize across datasets. Methods: Imaging and progression-free survival (PFS) data from 118 patients with mPC from four separate prospective clinical trials were gathered retrospectively. Patients received 18F-NaF PET/CT imaging at baseline and at follow-up, between eight and thirteen weeks. TRAQinform IQ technology (AIQ Solutions) was used to identify, segment, and track individual lesions from baseline to follow-up. Eighty-four imaging features were extracted from each patient and sorted into baseline, follow-up, response, patient-level (no inter-lesion comparison), and intrapatient heterogeneity (comparisons between lesions). The data was split into two training and testing sets, 44 patients from one study and 73 patients from the remaining 3 studies. As they can utilize large number of inputs without overfitting, random survival forest (RSF) models were chosen to evaluate performance of feature sets in predicting PFS. Different combinations of features were used as inputs to RSF models to compare single timepoint features with response features and patient-level features with intrapatient heterogeneity features. The performance of the RSF models, together with other methods identified in literature, were evaluated in each dataset using Kaplan-Meier analysis for categorical variables and the c-index for continuous variables. Results: No patient-level imaging features highlighted by literature displayed significant association to PFS across all four clinical trials (c-index &lt; 0.62 in at least one dataset). Other criteria from literature did not generalize across all datasets (P &gt; 0.05). The RSF model trained with all features had high c-indices in all four datasets (range: 0.66-0.80). RSF models built with response features (min: 0.63) performed better on average than models built with features obtained from single timepoints (min: 0.55). Patient-level features (min: 0.56) were not sufficient in all testing scenarios as compared to intrapatient heterogeneity features (min: 0.63). Conclusions: The candidate imaging biomarkers from previous 18F-NaF PET/CT imaging studies of mPC patients did not generalize across all datasets. Incorporating response and heterogeneity features with single-timepoint and patient-level features resulted in RSF prediction models which were generalizable across all datasets. Use of such models hold promise for improving outcome prediction in mPC patients.

BackgroundAccurately predicting the occurrence of imminent new vertebral fractures (NVFs) in patients with osteoporotic vertebral compression fractures (OVCFs) undergoing vertebral augmentation (VA) is challenging with yet no effective approach. This study aim to examine a machine learning model based on radiomics signature and clinical factors in predicting imminent new vertebral fractures after vertebral augmentation.MethodsA total of 235 eligible patients with OVCFs who underwent VA procedures were recruited from two independent institutions and categorized into three groups, including training set (n = 138), internal validation set (n = 59), and external validation set (n = 38). In the training set, radiomics features were computationally retrieved from L1 or adjacent vertebral body (T12 or L2) on T1-w MRI images, and a radiomics signature was constructed using the least absolute shrinkage and selection operator algorithm (LASSO). Predictive radiomics signature and clinical factors were fitted into two final prediction models using the random survival forest (RSF) algorithm or COX proportional hazard (CPH) analysis. Independent internal and external validation sets were used to validate the prediction models.ResultsThe two prediction models were integrated with radiomics signature and intravertebral cleft (IVC). The RSF model with C-indices of 0.763, 0.773, and 0.731 and time-dependent AUC (2 years) of 0.855, 0.907, and 0.839 (p < 0.001 for all) was found to be better predictive than the CPH model in training, internal and external validation sets. The RSF model provided better calibration, larger net benefits (determined by decision curve analysis), and lower prediction error (time-dependent brier score of 0.156, 0.151, and 0.146, respectively) than the CPH model.ConclusionsThe integrated RSF model showed the potential to predict imminent NVFs following vertebral augmentation, which will aid in postoperative follow-up and treatment.