Random Survival Forest Model Research Articles

Association of the 24‐Hour National Institutes of Health Stroke Scale After Mechanical Thrombectomy With Early and Long‐Term Survival

Background The National Institutes of Health Stroke Scale (NIHSS) obtained 24 hours after ischemic stroke is a good indicator for functional outcome and early mortality, but the correlation with long‐term survival is less clear. We analyzed the correlation of the NIHSS after 24 hours (24h NIHSS) and early clinical neurological development after mechanical thrombectomy with early and long‐term survival as well as its predictive power on survival. Methods We reviewed a prospective observational registry for all patients undergoing mechanical thrombectomy between January 2010 and December 2018. Vital status was extracted from the Swiss Population Registry. Adjusted hazard ratio (aHR) and crude hazard ratios were calculated using Cox regression. To assess predictive power of the 24h NIHSS, different Random Survival Forest models were evaluated. Results We included 957 patients (median follow‐up 1376 days). Patients with lower 24h NIHSS and major early neurological improvement had substantially better survival rates. We observed significantly higher aHR for death in patients with 24h NIHSS 12 to 15 (aHR, 1.78; 95% CI, 1.1–2.89), with 24h NIHSS 16 to 21 (aHR, 2.54, 95% CI, 1.59–4.06), and with 24h NIHSS >21 (aHR, 5.74; 95% CI, 3.47–9.5). The 24h NIHSS showed the best performance predicting mortality (receiver operating characteristic area under the curve at 3 months [0.85±0.034], at 1 year [0.82±0.029], at 2 years [0.82±0.031], and at 5 years [0.83±0.035]), followed by NIHSS change. Conclusions Patients with acute ischemic stroke achieving a low 24h NIHSS or major early neurological improvement after mechanical thrombectomy had markedly lower long‐term mortality. Furthermore, 24h NIHSS had the best predictive power for early and long‐term survival in our machine learning–based prediction.

Comparison of Random Survival Forest with Accelerated Failure Time-Weibull Model for Bridge Deck Deterioration

Bridge deck deterioration modeling is critical to infrastructure management. Deterioration modeling is traditionally done using deterministic models, stochastic models, and recently basic machine learning methods. The advanced machine learning-based survival models, such as random survival forest, have not been adapted for use in infrastructure management. This paper introduces random survival forest models for bridge deck deterioration modeling and compare their performance with a commonly used traditional stochastic model, that is, the Weibull distribution-based accelerated failure time (AFT-Weibull) model. To better adapt the random survival model for bridge deck deterioration modeling, the selection of the dependent variables is discussed between two variables: time-in-rating, and cumulative truck traffic. Inspection data from about 22,000 state-owned bridge decks in Pennsylvania are used to validate and test the performance of the models. The results suggest that cumulative truck traffic is more suitable to be selected as the dependent variable when analyzing the reliability of the bridge deck. Further, the random survival forest model outperformed the AFT-Weibull model in predictive accuracy.

Modeling disruption durations of subway service via random survival forests: The case of Shanghai

Subways are the backbone of many urban transportation systems in large cities around the world. However, disruptions of subway services, such as power-supply failure or signal failures, can cause severe travel delays due to the large carrying capacities of subway trains. Accurate predictions of the disruption durations of subway services are essential for emergency response. To predict the disruption durations of subway services, previous studies primarily used parametric models (e.g., the accelerated failure time (AFT) model), and less attention has been given to machine-learning models with high potential prediction ability and fewer parameter restrictions. This paper proposes a model to predict and explore factors that affect the disruption durations of subway services in Shanghai using machine-learning models. The longitudinal data released by the subway operator from 2012 to 2021 were collected and analyzed in this study. A random survival forest (RSF) was used to describe the disruption durations of the subway service, and influential factors, such as incident reason, incident occurrence time, and line-related variables were considered in the model. Results show that the RSF model (C-Index = 0.672) achieved better prediction accuracy than the traditional AFT model (the best C-Index = 0.609) based on the collected data in Shanghai. In addition, results indicate that incident reason, disruption location, and the time of disruption factors can significantly affect subway service disruption durations. The proposed model can be used as a tool to predict the disruption durations of subway service for better disruption management of the subway system.

Development and validation of clinical prediction models for breast cancer incidence and mortality: a protocol for a dual cohort study

IntroductionBreast cancer is the most common cancer and the leading cause of cancer-related death in women worldwide. Risk prediction models may be useful to guide risk-reducing interventions (such as pharmacological...

Development and Validation of a New Multiparametric Random Survival Forest Predictive Model for Breast Cancer Recurrence with a Potential Benefit to Individual Outcomes.

PurposeBreast cancer (BC) is a multi-factorial disease. Its individual prognosis varies; thus, individualized patient profiling is instrumental to improving BC management and individual outcomes. An economical, multiparametric, and practical model to predict BC recurrence is needed.Patients and MethodsWe retrospectively investigated the clinical data of BC patients treated at the Third Affiliated Hospital of Sun Yat-sen University and Liuzhou Women and Children’s Medical Center from January 2013 to December 2020. Random forest-recursive feature elimination (run by R caret package) was used to determine the best variable set, and the random survival forest method was used to develop a predictive model for BC recurrence.ResultsThe training and validations sets included 623 and 151 patients, respectively. We selected 14 variables, the pathological (TNM) stage, gamma-glutamyl transpeptidase, total cholesterol, Ki-67, lymphocyte count, low-density lipoprotein, age, apolipoprotein B, high-density lipoprotein, globulin, neutrophil count to lymphocyte count ratio, alanine aminotransferase, triglyceride, and albumin to globulin ratio, using random survival forest (RSF)-recursive feature elimination. We developed a recurrence prediction model using RSF. Using area under the receiver operating characteristic curve and Kaplan–Meier survival analyses, the model performance was determined to be accurate. C-indexes were 0.997 and 0.936 for the training and validation sets, respectively.ConclusionThe model could accurately predict BC recurrence. It aids clinicians in identifying high-risk patients and making treatment decisions for Breast cancer patients in China. This new multiparametric RSF model is instrumental for breast cancer recurrence prediction and potentially improves individual outcomes.

Nomograms for Predicting Hepatocellular Carcinoma Recurrence and Overall Postoperative Patient Survival.

BackgroundFew studies have focused on the prognosis of patients with hepatocellular carcinoma (HCC) of Barcelona Clinic Liver Cancer (BCLC) stage 0‒C in terms of early recurrence and 5-years overall survival (OS). We sought to develop nomograms for predicting 5-year OS and early recurrence after curative resection of HCC, based on a clinicopathological‒radiological model. We also investigated whether different treatment methods influenced the OS of patients with early recurrence.MethodsRetrospective data, including clinical pathology, radiology, and follow-up data, were collected for 494 patients with HCC who underwent hepatectomy. Nomograms estimating OS and early recurrence were constructed using multivariate Cox regression analysis, based on the random survival forest (RSF) model. We evaluated the discrimination and calibration abilities of the nomograms using concordance indices (C-index), calibration curves, and Kaplan‒Meier curves. OS curves of different treatments for patients who had recurrence within 2 years after curative surgery were depicted and compared using the Kaplan–Meier method and the log-rank test.ResultsMultivariate Cox regression revealed that BCLC stage, non-smooth margin, maximum tumor diameter, age, aspartate aminotransferase levels, microvascular invasion, and differentiation were prognostic factors for OS and were incorporated into the nomogram with good predictive performance in the training (C-index: 0.787) and testing cohorts (C-index: 0.711). A nomogram for recurrence-free survival was also developed based on four prognostic factors (BCLC stage, non-smooth margin, maximum tumor diameter, and microvascular invasion) with good predictive performance in the training (C-index: 0.717) and testing cohorts (C-index: 0.701). In comparison to the BCLC staging system, the C-index (training cohort: 0.787 vs. 0.678, 0.717 vs. 0.675; external cohort 2: 0.748 vs. 0.624, 0.729 vs. 0.587 respectively, for OS and RFS; external cohort1:0.716 vs. 0.627 for RFS, all p value<0.05), and model calibration curves all showed improved performance. Patients who underwent surgery after tumor recurrence had a higher reOS than those who underwent comprehensive treatments and supportive care.ConclusionsThe nomogram, based on clinical, pathological, and radiological factors, demonstrated good accuracy in estimating OS and recurrence, which can guide follow-up and treatment of individual patients. Reoperation may be the best option for patients with recurrence in good condition.

The Application and Comparison of Machine Learning Models for the Prediction of Breast Cancer Prognosis: Retrospective Cohort Study.

BackgroundOver the recent years, machine learning methods have been increasingly explored in cancer prognosis because of the appearance of improved machine learning algorithms. These algorithms can use censored data for modeling, such as support vector machines for survival analysis and random survival forest (RSF). However, it is still debated whether traditional (Cox proportional hazard regression) or machine learning-based prognostic models have better predictive performance.ObjectiveThis study aimed to compare the performance of breast cancer prognostic prediction models based on machine learning and Cox regression.MethodsThis retrospective cohort study included all patients diagnosed with breast cancer and subsequently hospitalized in Fudan University Shanghai Cancer Center between January 1, 2008, and December 31, 2016. After all exclusions, a total of 22,176 cases with 21 features were eligible for model development. The data set was randomly split into a training set (15,523 cases, 70%) and a test set (6653 cases, 30%) for developing 4 models and predicting the overall survival of patients diagnosed with breast cancer. The discriminative ability of models was evaluated by the concordance index (C-index), the time-dependent area under the curve, and D-index; the calibration ability of models was evaluated by the Brier score.ResultsThe RSF model revealed the best discriminative performance among the 4 models with 3-year, 5-year, and 10-year time-dependent area under the curve of 0.857, 0.838, and 0.781, a D-index of 7.643 (95% CI 6.542, 8.930) and a C-index of 0.827 (95% CI 0.809, 0.845). The statistical difference of the C-index was tested, and the RSF model significantly outperformed the Cox-EN (elastic net) model (C-index 0.816, 95% CI 0.796, 0.836; P=.01), the Cox model (C-index 0.814, 95% CI 0.794, 0.835; P=.003), and the support vector machine model (C-index 0.812, 95% CI 0.793, 0.832; P<.001). The 4 models’ 3-year, 5-year, and 10-year Brier scores were very close, ranging from 0.027 to 0.094 and less than 0.1, which meant all models had good calibration. In the context of feature importance, elastic net and RSF both indicated that TNM staging, neoadjuvant therapy, number of lymph node metastases, age, and tumor diameter were the top 5 important features for predicting the prognosis of breast cancer. A final online tool was developed to predict the overall survival of patients with breast cancer.ConclusionsThe RSF model slightly outperformed the other models on discriminative ability, revealing the potential of the RSF method as an effective approach to building prognostic prediction models in the context of survival analysis.

Prediction of prognosis in elderly patients with sepsis based on machine learning (random survival forest)

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.

Machine learning predicts all-cause mortality in patients with severe aortic regurgitation

Abstract Funding Acknowledgements Type of funding sources: None. Background Aortic regurgitation (AR) is a common valvular lesion associated with left ventricular (LV) enlargement and dysfunction and increased risk of death with onset of symptoms, significant LV enlargement, or systolic dysfunction. The current guidelines are based on symptoms, LV size and LV function to evaluate the candidacy for surgery in patients with severe AR. Purpose We propose a machine-learning (ML) based algorithm to identify patients at risk for death from AR independent of aortic valve replacement (AVR) surgery (patients were censored at AVR event unless the operation occurred within 100 days of the first visit, in which case baseline AVR is included as a predictor). Methods The model was trained with 5-fold cross-validation on a dataset of 1035 patients. Model performance is reported on an independent dataset of 207 patients. Optimal predictive performance was observed with a Conditional Random Survival Forest model. A subset of 19 / 41 variables were selected for inclusion in the final model. Variable selection was performed with 10-fold cross validation using Random Survival Forest model; variables that were on an average selected by less than 6/10 splits were excluded. Results The mean age of patients was 60 ± 17 years and 198 (18%) were females. The variables included in the models were age, end-diastolic volume, end systolic dimension, body surface area, ejection fraction, NYHA class, body mass index, diastolic blood pressure, bicuspid valve, regurgitant volume, filling pressure, mid ascending aorta diameter, mitral tissue early relaxation velocity, pulmonary artery systolic pressure, and comorbidities, and the relative variable importances averaged across five splits of cross validation in each repeat were evaluated (Figure). The predictive AUC for predicting survival of the best-performing model was 0.84 at 1 year, 0.86 at 2 years, and 0.87 overall, respectively. Conclusion Using commonly observed echo parameters and patient characteristics we successfully trained a ML model to predict survival in patients with severe AR. This technique could identify high-risk patients who would benefit from early intervention, thereby improving patient outcomes. Abstract Figure.

Radiomic analysis for predicting prognosis of colorectal cancer from preoperative 18F-FDG PET/CT

BackgroundTo develop and validate a survival model with clinico-biological features and 18F- FDG PET/CT radiomic features via machine learning, and for predicting the prognosis from the primary tumor of colorectal cancer.MethodsA total of 196 pathologically confirmed patients with colorectal cancer (stage I to stage IV) were included. Preoperative clinical factors, serum tumor markers, and PET/CT radiomic features were included for the recurrence-free survival analysis. For the modeling and validation, patients were randomly divided into the training (n = 137) and validation (n = 59) set, while the 78 stage III patients [training (n = 55), and validation (n = 23)] was divided for the further experiment. After selecting features by the log-rank test and variable-hunting methods, random survival forest (RSF) models were built on the training set to analyze the prognostic value of selected features. The performance of models was measured by C-index and was tested on the validation set with bootstrapping. Feature importance and the Pearson correlation were also analyzed.ResultsRadiomics signature (containing four PET/CT features and four clinical factors) achieved the best result for prognostic prediction of 196 patients (C-index 0.780, 95% CI 0.634–0.877). Moreover, four features (including two clinical features and two radiomics features) were selected for prognostic prediction of the 78 stage III patients (C-index was 0.820, 95% CI 0.676–0.900). K–M curves of both models significantly stratified low-risk and high-risk groups (P < 0.0001). Pearson correlation analysis demonstrated that selected radiomics features were correlated with tumor metabolic factors, such as SUVmean, SUVmax.ConclusionThis study presents integrated clinico-biological-radiological models that can accurately predict the prognosis in colorectal cancer using the preoperative 18F-FDG PET/CT radiomics in colorectal cancer. It is of potential value in assisting the management and decision making for precision treatment in colorectal cancer.Trial registration The retrospectively registered study was approved by the Ethics Committee of Fudan University Shanghai Cancer Center (No. 1909207-14-1910) and the data were analyzed anonymously.

A risk scoring system based on tumor microenvironment cells to predict prognosis and immune activity in triple-negative breast cancer

The tumor microenvironment (TME) interacting with the malignant cells plays a vital role in cancer development. Herein, we aim to establish and verify a scoring system based on the characteristics of TME cells for prognosis prediction and personalized treatment guidance in patients with triple-negative breast cancer (TNBC). 158 TNBC samples from The Cancer Genome Atlas (TCGA) were included as the training cohort, and Molecular Taxonomy of Breast Cancer International Consortium (METABRIC) (N = 297), as well as GSE58812 (N = 107), were included as the validation cohort. The enrichment scores of 64 immune and stromal cells were estimated by the xCell algorithm. In the training cohort, cells with prognostic significance were found out using univariate Cox regression analysis and further applied to the random survival forest (RSF) model. Based on the scores of M2 macrophages, CD8+ T cells, and CD4+ memory T cells, a risk scoring system was constructed, which divided TNBC patients into 4 phenotypes (M2low, M2highCD8+ThighCD4+Thigh, M2highCD8+ThighCD4+Tlow, and M2highCD8+Tlow). Furthermore, types 1 and 2 patients were merged into the low-risk group, while types 3 and 4 patients were in the high-risk group. The low-risk group had superior survival outcomes than the high-risk one, which was further confirmed in the validation cohort. Moreover, in the low-risk group, immune-related pathways were significantly enriched, and a higher level of antitumoral immune cells and immune checkpoint molecules, including PD-L1, PD-1, and CTLA-4, could be observed. Additionally, consistent results were achieved in the SYSUCC cohort when the scoring system was applied. In summary, this novel scoring system might predict the survival and immune activity of patients and might serve as a potential index for immunotherapy.

Deep Learning in Prediction of Late Major Bleeding After Transcatheter Aortic Valve Replacement.

PurposeLate major bleeding is one of the main complications after transcatheter aortic valve replacement (TAVR). We aimed to develop a risk prediction model based on deep learning to predict major or life-threatening bleeding complications (MLBCs) after TAVR.Patients and MethodsThis was a retrospective study including TAVR patients from West China Hospital of Sichuan University Transcatheter Aortic Valve Replacement Registry (ChiCTR2000033419) between April 17, 2012 and May 27, 2020. A deep learning-based model named BLeNet was developed with 56 features covering baseline, procedural, and post-procedural characteristics. The model was validated with the bootstrap method and evaluated using Harrell’s concordance index (c-index), receiver operating characteristics (ROC) curve, calibration curve, and Kaplan–Meier estimate. Captum interpretation library was applied to identify feature importance. The BLeNet model was compared with the traditional Cox proportional hazard (Cox-PH) model and the random survival forest model in the metrics mentioned above.ResultsThe BLeNet model outperformed the Cox-PH and random survival forest models significantly in discrimination [optimism-corrected c-index of BLeNet vs Cox-PH vs random survival forest: 0.81 (95% CI: 0.79–0.92) vs 0.72 (95% CI: 0.63–0.77) vs 0.70 (95% CI: 0.61–0.74)] and calibration (integrated calibration index of BLeNet vs Cox-PH vs random survival forest: 0.007 vs 0.015 vs 0.019). In Kaplan–Meier analysis, BLeNet model had great performance in stratifying high- and low-bleeding risk patients (p < 0.0001).ConclusionDeep learning is a feasible way to build prediction models concerning TAVR prognosis. A dedicated bleeding risk prediction model was developed for TAVR patients to facilitate well-informed clinical decisions.

Random survival forest model identifies novel biomarkers of event-free survival in high-risk pediatric acute lymphoblastic leukemia

High-risk pediatric B-ALL patients experience 5-year negative event rates up to 25%. Although some biomarkers of relapse are utilized in the clinic, their ability to predict outcomes in high-risk patients is limited. Here, we propose a random survival forest (RSF) machine learning model utilizing interpretable genomic inputs to predict relapse/death in high-risk pediatric B-ALL patients. We utilized whole exome sequencing profiles from 156 patients in the TARGET-ALL study (with samples collected at presentation) further stratified into training and test cohorts (109 and 47 patients, respectively). To avoid overfitting and facilitate the interpretation of machine learning results, input genomic variables were engineered using a stepwise approach involving univariable Cox models to select variables directly associated with outcomes, genomic coordinate-based analysis to select mutational hotspots, and correlation analysis to eliminate feature co-linearity. Model training identified 7 genomic regions most predictive of relapse/death-free survival. The test cohort error rate was 12.47%, and a polygenic score based on the sum of the top 7 variables effectively stratified patients into two groups, with significant differences in time to relapse/death (log-rank P = 0.001, hazard ratio = 5.41). Our model outperformed other EFS modeling approaches including an RSF using gold-standard prognostic variables (error rate = 24.35%). Validation in 174 standard-risk patients and 3 patients who failed to respond to induction therapy confirmed that our RSF model and polygenic score were specific to high-risk disease. We propose that our feature selection/engineering approach can increase the clinical interpretability of RSF, and our polygenic score could be utilized for enhance clinical decision-making in high-risk B-ALL.

Predictors of colorectal cancer survival using cox regression and random survival forests models based on gene expression data.

Understanding and identifying the markers and clinical information that are associated with colorectal cancer (CRC) patient survival is needed for early detection and diagnosis. In this work, we aimed to build a simple model using Cox proportional hazards (PH) and random survival forest (RSF) and find a robust signature for predicting CRC overall survival. We used stepwise regression to develop Cox PH model to analyse 54 common differentially expressed genes from three mutations. RSF is applied using log-rank and log-rank-score based on 5000 survival trees, and therefore, variables important obtained to find the genes that are most influential for CRC survival. We compared the predictive performance of the Cox PH model and RSF for early CRC detection and diagnosis. The results indicate that SLC9A8, IER5, ARSJ, ANKRD27, and PIPOX genes were significantly associated with the CRC overall survival. In addition, age, sex, and stages are also affecting the CRC overall survival. The RSF model using log-rank is better than log-rank-score, while log-rank-score needed more trees to stabilize. Overall, the imputation of missing values enhanced the model’s predictive performance. In addition, Cox PH predictive performance was better than RSF.

Modeling lapse rates using machine learning: a comparison between survival forests and cox proportional hazards techniques

This study undertakes a comparative analysis of the performance of machine learning and traditional survival analysis techniques in the insurance industry. The techniques compared are the traditional Cox Proportional Hazards (CPH), Random Survival Forests (RSF) and Conditional Inference Forests (CIF) machine learning models. These techniques are applied in a case study of insurance portfolio of one of Ecuador's largest insurer. This study demonstrates how machine learning techniques perform better in predicting survival function measured by the C-index and Brier Score. It also demonstrates that the predictive contribution of covariates in the RSF model is consistent with the traditional CPH model.

O-OGC01 Development and validation of multivariate prediction model of long-term survival after oesophagectomy in patients with oesophageal cancer

Abstract Background Long-term survival after oesophagectomy remains poor, with recurrence a feared common outcome. Prediction tools can help clinicians identify high-risk patients and optimise treatment decisions based on their prognostic factors. This study developed and evaluated a prediction model to predict long-term survival and time-to-recurrence following surgery for oesophageal cancer. Methods Patients who underwent curative surgery between June 2009-2015 from the European iNvestigation of SUrveillance After Resection for Esophageal Cancer study were included. Prediction models were developed for overall survival (OS) and disease-free survival (DFS) using Cox proportional hazards (CPH) and Random Survival Forest (RSF). Model performance was evaluated using discrimination (time-dependent area under the curve (tAUC)) and calibration (visual comparison of predicted and observed survival probabilities). Results This study included 4719 patients with an OS of 47.7% and DFS of 48.4% at 5 years. Sixteen variables were included in the final model. CPH and RSF demonstrated good discrimination with a tAUC of 78.2% (95% CI 77.4-79.1%) and 77.1% (95% CI 76.1-78.1%) for OS and a tAUC of 79.4% (95% CI 78.5-80.2%) and 78.6% (95% CI 77.5-79.5%) respectively for DFS at 5 years. CPH showed good agreement between predicted and observed probabilities in all quintiles. RSF showed good agreement for patients with survival probabilities between 20-80% and moderate agreement in the &amp;lt;20% and &amp;gt;80% quintile groups. Conclusions This study demonstrated the ability of a statistical model to accurately predict long-term survival and time-to-recurrence after surgery for oesophageal cancer, with CPH and RSF models showing good discrimination and calibration. Identification of patient groups at risk of recurrence and poor long-term survival can improve patient outcomes by enhancing selection of treatment methods and surveillance strategies. Future work evaluating prediction-based decisions against standard decision-making is required to improve understanding of the clinical utility derived from prognostic model use.

Predicting the risk of incident dementia in older adults: The ADNI‐dementia risk score

AbstractBackgroundThere are many risk factors associated with incident dementia in older adults, however predicting the timeframe of progression to dementia is difficult. We aimed to develop a novel, machine learning‐derived model to predict the risk of incident dementia among non‐demented older adults.MethodUsing data from 1837 non‐demented participants from the Alzheimer’s Disease Neuroimaging Initiative (ADNI), we used random survival forest (RSF) methods, a nonparametric decision tree machine learning approach, to identify predictors of incident dementia. Three sets of features were used: Model 1: Demographics (age, sex, race/ethnicity, years of education, BMI, marital status); Model 2: Demographics + Neuropsychological test scores and depression (NP; Alzheimer's Disease Assessment Scale – Cognitive (ADAS‐cog), Mini‐Mental State Examination (MMSE), geriatric depression scale (GDS)); and Model 3: Demographics + NP + Biomarkers (APOE4 status, hippocampal volume, global β‐amyloid PET‐SUVR). An integer‐based score was developed to estimate the 10‐year risk of incident dementia using regression coefficients from Cox Proportional Hazard models.ResultOver up to 10 years of follow up (mean = 3.19), 377 participants (20.5%) developed incident dementia. The RSF models 2 and 3 demonstrated good discrimination (C‐index for Model 1, 0.58 [95% CI 0.53–0.59]; model 2, 0.68 [95% CI 0.65–0.71]; and model 3, 0.66 [0.62–0.69]. Performance of RSF models was similar to Cox‐based models, which used the same features. From the identified top‐performing predictors (listed in figure 1), an integer‐based risk score for 10‐year dementia incidence was created: ADNI‐Dementia risk score 1, 2, and 3. The cumulative 10‐year incidence of dementia increased in a graded fashion from quartile 1 to quartile 4 for all risk scores (Figure 2). Risk score 3, which included biomarkers, was the most specific model for detecting individuals unlikely to progress to dementia.ConclusionWe developed a new risk score that integrates readily available demographic, clinical, and biomarker data to predict the risk of dementia among non‐demented older adults. Using this simple tool is practical in clinical setting.

Machine learning-based prediction of 1-year mortality for acute coronary syndrome✰

BackgroundClinical risk assessment with quantitative formal risk scores may add to intuitive physician risk assessment and are advised by the international guidelines for the management of acute coronary syndrome (ACS) patients. Most previous studies have used the binary regression/classification approach (dead/alive) for long-term mortality post-ACS, without considering the time-to-event as in survival analysis. The use of machine learning (ML)-based survival models has yet to be validated.The primary objective was to compare survival prediction performance of 1-year mortality following ACS of two newly developed ML-based models [random survival forest (RSF) and deep learning (DeepSurv)] with the traditional Cox-proportional hazard (CPH) model. The secondary objective was external validation of the findings. MethodsThis was a retrospective, supervised learning data mining study based on the Acute Coronary Syndrome Israeli Survey (ACSIS) and the Myocardial Ischemia National Audit Project (MINAP). The ACSIS data were divided to train/test in a 70/30 fashion. Next, the models were externally validated on the MINAP data. Harrell's C-index, inverse probability of censoring weighting (IPCW), and the Brier-score were used for models’ performance comparison. ResultsRSF performed best among the three models, with Harrell's C-index on training and testing sets reaching 0.953 and 0.924 respectively, followed by CPH multivariate selected model (0.805/0.849), CPH Univariate selected model (0.828/0.806), DeepSurv model (0.801/0.804), and the traditional CPH model (0.826/0.738). The RSF model also had the highest performance on the validation data set with 0.811 for Harrell's C-index, 0.844 for IPCW, and 0.093 for Brier score. The CPH model performance on the validation set had C-index range between 0.689 to 0.790, 0.713 to 0.826 for IPCW, and 0.094 to 0.103 Brier score. ConclusionsRSF survival predictions for long-term mortality post-ACS show improved model performance compared with the classic statistical method. This may benefit patients by allowing better risk stratification and tailored therapy, however further prospective evaluations are required.

Abstract 11025: Markers of Heart Rate Variability Represent Strong and Independent Predictors of All-Cause Death in Heart Failure - Results from the Myovasc Study

Introduction: Although different measures of heart rate variability (HRV) have been proposed as markers of autonomic dysfunction (AD) in heart failure (HF), evidence on their clinical relevance is limited. Methods: Data from the MyoVasc study (NCT04064450; N=3,289), a prospective cohort study on chronic HF, were analyzed. Study participants received a highly standardized, 5-hour examination including 24-hour Holter ECG recording. RR intervals were extracted to evaluate HRV markers from time and frequency domain analyses, and non-linear markers derived via phase space reconstruction, phase-rectified signal averaging, and entropy analyses. HRV markers of interest were selected using a systematic literature screen, and a random survival forest model predicting cardiac death. Reference values were determined in a healthy subsample. To evaluate clinical relevance, HRV markers were investigated in multivariable Cox regression analyses for their relationship to all-cause death considering potential confounders and mediators. Results: Holter ECG recordings were available for analysis in 1,001 study participants (mean age 64.5±10.5 years; female sex: 35.4%). While the top 10 most frequently reported HRV markers in literature screen were from the time and frequency domains, the data-driven approach revealed predominantly non-linear HRV measures. Values outside reference ranges for SDANN, rMSSD, frequency bands (i.e., ULF, VLF, LF, total power, ULF MED wavelet ), acceleration capacity (AC), deceleration capacity (DC), and time lag, were strong predictors of all-cause death in Cox regression analysis (median follow-up time: 6.1 [IQR 4.7/7.1] years) adjusted for age, sex, clinical profile and medication. In linear Cox regression analyses of the total sample, LF/HF (hazard ratio per SD (HR SD ) 0.78 [95% confidence interval (CI) 0.62, 0.98], p=0.034), time lag (HR SD 1.23 [95%CI 1.05, 1.44], p=0.0095), AC (HR SD : 1.57 [95%CI 1.24, 1.99], p=0.0002), and DC (HR SD 0.67 [95%CI 0.53, 0.85], p=0.0012) were strong predictors for all-cause mortality independent of age, sex, clinical profile and medication. Conclusions: HRV markers reflecting AD are strong predictors of all-cause death in HF, suggesting clinical relevance and potential for intervention.

Novel Machine Learning Algorithm Predicts Disease Progression in Polycythemia Vera (PV) with Readily-Available Baseline Characteristics

Novel Machine Learning Algorithm Predicts Disease Progression in Polycythemia Vera (PV) with Readily-Available Baseline Characteristics