Assessment and comparison of prognostic classification schemes for survival data

Prediction of recurrence risk in endometrial cancer with multimodal deep learning

Predicting distant recurrence of endometrial cancer (EC) is crucial for personalized adjuvant treatment. The current gold standard of combined pathological and molecular profiling is costly, hampering implementation. Here we developed HECTOR (histopathology-based endometrial cancer tailored outcome risk), a multimodal deep learning prognostic model using hematoxylin and eosin-stained, whole-slide images and tumor stage as input, on 2,072 patients from eight EC cohorts including the PORTEC-1/-2/-3 randomized trials. HECTOR demonstrated C-indices in internal (n = 353) and two external (n = 160 and n = 151) test sets of 0.789, 0.828 and 0.815, respectively, outperforming the current gold standard, and identified patients with markedly different outcomes (10-year distant recurrence-free probabilities of 97.0%, 77.7% and 58.1% for HECTOR low-, intermediate- and high-risk groups, respectively, by Kaplan-Meier analysis). HECTOR also predicted adjuvant chemotherapy benefit better than current methods. Morphological and genomic feature extraction identified correlates of HECTOR risk groups, some with therapeutic potential. HECTOR improves on the current gold standard and may help delivery of personalized treatment in EC.

Prediction of disease-free survival for precision medicine using cooperative learning on multi-omic data

In precision medicine, both predicting the disease susceptibility of an individual and forecasting its disease-free survival are areas of key research. Besides the classical epidemiological predictor variables, data from multiple (omic) platforms are increasingly available. To integrate this wealth of information, we propose new methodology to combine both cooperative learning, a recent approach to leverage the predictive power of several datasets, and polygenic hazard score models. Polygenic hazard score models provide a practitioner with a more differentiated view of the predicted disease-free survival than the one given by merely a point estimate, for instance computed with a polygenic risk score. Our aim is to leverage the advantages of cooperative learning for the computation of polygenic hazard score models via Cox's proportional hazard model, thereby improving the prediction of the disease-free survival. In our experimental study, we apply our methodology to forecast the disease-free survival for Alzheimer's disease (AD) using three layers of data. One layer contains epidemiological variables such as sex, APOE (apolipoprotein E, a genetic risk factor for AD) status and 10 leading principal components. Another layer contains selected genomic loci, and the last layer contains methylation data for selected CpG sites. We demonstrate that the survival curves computed via cooperative learning yield an AUC of around $0.7$, above the state-of-the-art performance of its competitors. Importantly, the proposed methodology returns (1) a linear score that can be easily interpreted (in contrast to machine learning approaches), and (2) a weighting of the predictive power of the involved data layers, allowing for an assessment of the importance of each omic (or other) platform. Similarly to polygenic hazard score models, our methodology also allows one to compute individual survival curves for each patient.

Artificial intelligence-based prognostic model accurately predicts the survival of patients with diffuse large B-cell lymphomas: analysis of a large cohort in China

BACKGROUND

Diffuse large B-cell lymphomas (DLBCLs) display high molecular heterogeneity, but the International Prognostic Index (IPI) considers only clinical indicators and has not been updated to include molecular data. Therefore, we developed a widely applicable novel scoring system with molecular indicators screened by artificial intelligence (AI) that achieves accurate prognostic stratification and promotes individualized treatments.

METHODS

We retrospectively enrolled a cohort of 401 patients with DLBCL from our hospital, covering the period from January 2015 to January 2019. We included 22 variables in our analysis and assigned them weights using the random survival forest method to establish a new predictive model combining bidirectional long-short term memory (Bi-LSTM) and logistic hazard techniques. We compared the predictive performance of our "molecular-contained prognostic model" (McPM) and the IPI. In addition, we developed a simplified version of the McPM (sMcPM) to enhance its practical applicability in clinical settings. We also demonstrated the improved risk stratification capabilities of the sMcPM.

RESULTS

Our McPM showed superior predictive accuracy, as indicated by its high C-index and low integrated Brier score (IBS), for both overall survival (OS) and progression-free survival (PFS). The overall performance of the McPM was also better than that of the IPI based on receiver operating characteristic (ROC) curve fitting. We selected five key indicators, including extranodal involvement sites, lactate dehydrogenase (LDH), MYC gene status, absolute monocyte count (AMC), and platelet count (PLT) to establish the sMcPM, which is more suitable for clinical applications. The sMcPM showed similar OS results (P < 0.0001 for both) to the IPI and significantly better PFS stratification results (P < 0.0001 for sMcPM vs. P = 0.44 for IPI).

CONCLUSIONS

Our new McPM, including both clinical and molecular variables, showed superior overall stratification performance to the IPI, rendering it more suitable for the molecular era. Moreover, our sMcPM may become a widely used and effective stratification tool to guide individual precision treatments and drive new drug development.

Intertumoral Heterogeneity Based on MRI Radiomic Features Estimates Recurrence in Hepatocellular Carcinoma

BACKGROUND

Hepatocellular carcinoma (HCC) heterogeneity impacts prognosis, and imaging is a potential indicator.

PURPOSE

To characterize HCC image subtypes in MRI and correlate subtypes with recurrence.

STUDY TYPE

Retrospective.

POPULATION

A total of 440 patients (training cohort = 213, internal test cohort = 140, external test cohort = 87) from three centers.

FIELD STRENGTH/SEQUENCE

1.5-T/3.0-T, fast/turbo spin-echo T2-weighted, spin-echo echo-planar diffusion-weighted, contrast-enhanced three-dimensional gradient-recalled-echo T1-weighted with extracellular agents (Gd-DTPA, Gd-DTPA-BMA, and Gd-BOPTA).

ASSESSMENT

Three-dimensional volume-of-interest of HCC was contoured on portal venous phase, then coregistered with precontrast and late arterial phases. Subtypes were identified using non-negative matrix factorization by analyzing radiomics features from volume-of-interests, and correlated with recurrence. Clinical (demographic and laboratory data), pathological, and radiologic features were compared across subtypes. Among clinical, radiologic features and subtypes, variables with variance inflation factor above 10 were excluded. Variables (P < 0.10) in univariate Cox regression were included in stepwise multivariate analysis. Three recurrence estimation models were built: clinical-radiologic model, subtype model, hybrid model integrating clinical-radiologic characteristics, and subtypes.

STATISTICAL TESTS

Mann-Whitney U test, Kruskal-Wallis H test, chi-square test, Fisher's exact test, Kaplan-Meier curves, log-rank test, concordance index (C-index). Significance level: P < 0.05.

RESULTS

Two subtypes were identified across three cohorts (subtype 1:subtype 2 of 86:127, 60:80, and 36:51, respectively). Subtype 1 showed higher microvascular invasion (MVI)-positive rates (53%-57% vs. 26%-31%), and worse recurrence-free survival. Hazard ratio (HR) for the subtype is 6.10 in subtype model. Clinical-radiologic model included alpha-fetoprotein (HR: 3.01), macrovascular invasion (HR: 2.32), nonsmooth tumor margin (HR: 1.81), rim enhancement (HR: 3.13), and intratumoral artery (HR: 2.21). Hybrid model included alpha-fetoprotein (HR: 2.70), nonsmooth tumor margin (HR: 1.51), rim enhancement (HR: 3.25), and subtypes (HR: 5.34). Subtype model was comparable to clinical-radiologic model (C-index: 0.71-0.73 vs. 0.71-0.73), but hybrid model outperformed both (C-index: 0.77-0.79).

CONCLUSION

MRI radiomics-based clustering identified two HCC subtypes with distinct MVI status and recurrence-free survival. Hybrid model showed superior capability to estimate recurrence.

LEVEL OF EVIDENCE

3 TECHNICAL EFFICACY STAGE: 2.

A transformer model for cause-specific hazard prediction

BACKGROUD

Modelling discrete-time cause-specific hazards in the presence of competing events and non-proportional hazards is a challenging task in many domains. Survival analysis in longitudinal cohorts often requires such models; notably when the data is gathered at discrete points in time and the predicted events display complex dynamics. Current models often rely on strong assumptions of proportional hazards, that is rarely verified in practice; or do not handle sequential data in a meaningful way. This study proposes a Transformer architecture for the prediction of cause-specific hazards in discrete-time competing risks. Contrary to Multilayer perceptrons that were already used for this task (DeepHit), the Transformer architecture is especially suited for handling complex relationships in sequential data, having displayed state-of-the-art performance in numerous tasks with few underlying assumptions on the task at hand.

RESULTS

Using synthetic datasets of 2000-50,000 patients, we showed that our Transformer model surpassed the CoxPH, PyDTS, and DeepHit models for the prediction of cause-specific hazard, especially when the proportional assumption did not hold. The error along simulated time outlined the ability of our model to anticipate the evolution of cause-specific hazards at later time steps where few events are observed. It was also superior to current models for prediction of dementia and other psychiatric conditions in the English longitudinal study of ageing cohort using the integrated brier score and the time-dependent concordance index. We also displayed the explainability of our model's prediction using the integrated gradients method.

CONCLUSIONS

Our model provided state-of-the-art prediction of cause-specific hazards, without adopting prior parametric assumptions on the hazard rates. It outperformed other models in non-proportional hazards settings for both the synthetic dataset and the longitudinal cohort study. We also observed that basic models such as CoxPH were more suited to extremely simple settings than deep learning models. Our model is therefore especially suited for survival analysis on longitudinal cohorts with complex dynamics of the covariate-to-outcome relationship, which are common in clinical practice. The integrated gradients provided the importance scores of input variables, which indicated variables guiding the model in its prediction. This model is ready to be utilized for time-to-event prediction in longitudinal cohorts.

Deep Survival Analysis With Latent Clustering and Contrastive Learning

Survival analysis is employed to analyze the time before the event of interest occurs, which is broadly applied in many fields. The existence of censored data with incomplete supervision information about survival outcomes is one key challenge in survival analysis tasks. Although some progress has been made on this issue recently, the present methods generally treat the instances as separate ones while ignoring their potential correlations, thus rendering unsatisfactory performance. In this study, we propose a novel Deep Survival Analysis model with latent Clustering and Contrastive learning (DSACC). Specifically, we jointly optimize representation learning, latent clustering and survival prediction in a unified framework. In this way, the clusters distribution structure in latent representation space is revealed, and meanwhile the structure of the clusters is well incorporated to improve the ability of survival prediction. Besides, by virtue of the learned clusters, we further propose a contrastive loss function, where the uncensored data in each cluster are set as anchors, and the censored data are treated as positive/negative sample pairs according to whether they belong to the same cluster or not. This design enables the censored data to make full use of the supervision information of the uncensored samples. Through extensive experiments on four popular clinical datasets, we demonstrate that our proposed DSACC achieves advanced performance in terms of both C-index (0.6722, 0.6793, 0.6350, and 0.7943) and Integrated Brier Score (IBS) (0.1616, 0.1826, 0.2028, and 0.1120).

CAMP-B score predicts the risk of hepatocellular carcinoma in patients with chronic hepatitis B after HBsAg seroclearance

BACKGROUND

Risk of hepatocellular carcinoma (HCC) persists after hepatitis B surface antigen (HBsAg) seroclearance in patients with chronic hepatitis B (CHB).

AIMS

To identify risk factors and construct a predictive model for HCC development.

METHODS

We retrospectively analysed patients with CHB with HBsAg seroclearance. Primary outcome was HCC development. Factors identified from a multivariate Cox model in the training cohort, consisting of 3476 patients from two Korean hospitals, were used to construct the prediction model. External validation was performed using data from 5255 patients in Hong Kong.

RESULTS

In the training cohort, HCC occurred in 102 patients during 24,019 person-years of observation (0.43%/year). Risk scores were assigned to cirrhosis (C:3), age ≥50 years (A:2), male sex (M:3) and platelet count <150,000/mm3 (P:1); all were independently associated with an increased risk of HCC in multivariate analysis The time-dependent area under receiver operating characteristic curves for 5, 10 and 15 years in the training and validation cohorts were 0.782, 0.817 and 0.825 and 0.785, 0.771 and 0.796, respectively. In the validation cohort, 85 patients developed HCC (0.24%/year). The corresponding incidence of HCC in the low-, intermediate- and high-risk groups were 0.07%, 0.37% and 0.90%, respectively.

CONCLUSIONS

The CAMP-B score (cirrhosis, age ≥50 years, male sex and platelet count <150,000/mm3/L) was significantly associated with HCC development after HBsAg seroclearance. CAMP-B score can be easily implemented in real-world clinical practice and helps stratify HCC risk in patients with CHB following HBsAg seroclearance.

Prediction Under Interventions: Evaluation of Counterfactual Performance Using Longitudinal Observational Data

Predictions under interventions are estimates of what a person's risk of an outcome would be if they were to follow a particular treatment strategy, given their individual characteristics. Such predictions can give important input to medical decision-making. However, evaluating the predictive performance of interventional predictions is challenging. Standard ways of evaluating predictive performance do not apply when using observational data, because prediction under interventions involves obtaining predictions of the outcome under conditions that are different from those that are observed for a subset of individuals in the validation dataset. This work describes methods for evaluating counterfactual performance of predictions under interventions for time-to-event outcomes. This means we aim to assess how well predictions would match the validation data if all individuals had followed the treatment strategy under which predictions are made. We focus on counterfactual performance evaluation using longitudinal observational data, and under treatment strategies that involve sustaining a particular treatment regime over time. We introduce an estimation approach using artificial censoring and inverse probability weighting that involves creating a validation dataset mimicking the treatment strategy under which predictions are made. We extend measures of calibration, discrimination (c-index and cumulative/dynamic AUCt) and overall prediction error (Brier score) to allow assessment of counterfactual performance. The methods are evaluated using a simulation study, including scenarios in which the methods should detect poor performance. Applying our methods in the context of liver transplantation shows that our procedure allows quantification of the performance of predictions supporting crucial decisions on organ allocation.

Machine Learning-Based Survival Prediction Models for Progression-Free and Overall Survival in Advanced-Stage Hodgkin Lymphoma

PURPOSE

Patients diagnosed with advanced-stage Hodgkin lymphoma (aHL) have historically been risk-stratified using the International Prognostic Score (IPS). This study investigated if a machine learning (ML) approach could outperform existing models when it comes to predicting overall survival (OS) and progression-free survival (PFS).

PATIENTS AND METHODS

This study used patient data from the Danish National Lymphoma Register for model development (development cohort). The ML model was developed using stacking, which combines several predictive survival models (Cox proportional hazard, flexible parametric model, IPS, principal component, penalized regression) into a single model, and was compared with two versions of IPS (IPS-3 and IPS-7) and the newly developed aHL international prognostic index (A-HIPI). Internal model validation was performed using nested cross-validation, and external validation was performed using patient data from the Swedish Lymphoma Register and Cancer Registry of Norway (validation cohort).

RESULTS

In total, 707 and 760 patients with aHL were included in the development and validation cohorts, respectively. Examining model performance for OS in the development cohort, the concordance index (C-index) for the ML model, IPS-7, IPS-3, and A-HIPI was found to be 0.789, 0.608, 0.650, and 0.768, respectively. The corresponding estimates in the validation cohort were 0.749, 0.700, 0.663, and 0.741. For PFS, the ML model achieved the highest C-index in both cohorts (0.665 in the development cohort and 0.691 in the validation cohort). The time-varying AUCs for both the ML model and the A-HIPI were consistently higher in both cohorts compared with the IPS models within the first 5 years after diagnosis.

CONCLUSION

The new prognostic model for aHL on the basis of ML techniques demonstrated a substantial improvement compared with the IPS models, but yielded a limited improvement in predictive performance compared with the A-HIPI.

Survival Analysis of High-Dimensional Data With Graph Convolutional Networks and Geometric Graphs

This article proposes a survival model based on graph convolutional networks (GCNs) with geometric graphs directly constructed from high-dimensional features. First, we clarify that the graphs used in GCNs play an important role in processing the relational information of samples, and the graphs that align well with the underlying data structure could be beneficial for survival analysis. Second, we show that sparse geometric graphs derived from high-dimensional data are more favorable compared with dense graphs when used in GCNs for survival analysis. Third, from this insight, we propose a model for survival analysis based on GCNs. By using multiple sparse geometric graphs and a proposed sequential forward floating selection algorithm, the new model is able to simultaneously perform survival analysis and unveil the local neighborhoods of samples. The experimental results on real-world datasets show that the proposed survival analysis approach based on GCNs outperforms a variety of existing methods and indicate that geometric graphs can aid survival analysis of high-dimensional data.

Cardiovascular Risk Associated With Social Determinants of Health at Individual and Area Levels

Importance

The benefit of adding social determinants of health (SDOH) when estimating atherosclerotic cardiovascular disease (ASCVD) risk is unclear.

Objective

To examine the association of SDOH at both individual and area levels with ASCVD risks, and to assess if adding individual- and area-level SDOH to the pooled cohort equations (PCEs) or the Predicting Risk of CVD Events (PREVENT) equations improves the accuracy of risk estimates.

Design, Setting, and Participants

This cohort study included participants data from 4 large US cohort studies. Eligible participants were aged 40 to 79 years without a history of ASCVD. Baseline data were collected from 1995 to 2007; median (IQR) follow-up was 13.0 (9.3-15.0) years. Data were analyzed from September 2023 to February 2024.

Exposures

Individual- and area-level education, income, and employment status.

Main outcomes and measures

ASCVD was defined as the composite outcome of nonfatal myocardial infarction, death from coronary heart disease, and fatal or nonfatal stroke.

Results

A total of 26 316 participants were included (mean [SD] age, 61.0 [9.1] years; 15 494 women [58.9%]; 11 365 Black [43.2%], 703 Chinese American [2.7%], 1278 Hispanic [4.9%], and 12 970 White [49.3%]); 11 764 individuals (44.7%) had at least 1 adverse individual-level SDOH and 10 908 (41.5%) had at least 1 adverse area-level SDOH. A total of 2673 ASCVD events occurred during follow-up. SDOH were associated with increased risk of ASCVD at both the individual and area levels, including for low education (individual: hazard ratio [HR], 1.39 [95% CI, 1.25-1.55]; area: HR, 1.31 [95% CI, 1.20-1.42]), low income (individual: 1.35 [95% CI, 1.25-1.47]; area: HR, 1.28 [95% CI, 1.17-1.40]), and unemployment (individual: HR, 1.61 [95% CI, 1.24-2.10]; area: HR, 1.25 [95% CI, 1.14-1.37]). Adding area-level SDOH alone to the PCEs did not change model discrimination but modestly improved calibration. Furthermore, adding both individual- and area-level SDOH to the PCEs led to a modest improvement in both discrimination and calibration in non-Hispanic Black individuals (change in C index, 0.0051 [95% CI, 0.0011 to 0.0126]; change in scaled integrated Brier score [IBS], 0.396% [95% CI, 0.221% to 0.802%]), and improvement in calibration in White individuals (change in scaled IBS, 0.274% [95% CI, 0.095% to 0.665%]). Adding individual-level SDOH to the PREVENT plus area-level social deprivation index (SDI) equations did not improve discrimination but modestly improved calibration in White participants (change in scaled IBS, 0.182% [95% CI, 0.040% to 0.496%]), Black participants (0.187% [95% CI, 0.039% to 0.501%]), and women (0.289% [95% CI, 0.115% to 0.574%]).

Conclusions and Relevance

In this cohort study, both individual- and area-level SDOH were associated with ASCVD risk; adding both individual- and area-level SDOH to the PCEs modestly improved discrimination and calibration for estimating ASCVD risk for Black individuals, and adding individual-level SDOH to PREVENT plus SDI also modestly improved calibration. These findings suggest that both individual- and area-level SDOH may be considered in future development of ASCVD risk assessment tools, particularly among Black individuals.

Incorporation of emergent symptoms and genetic covariates improves prediction of aromatase inhibitor therapy discontinuation

Objectives

Early discontinuation is common among breast cancer patients taking aromatase inhibitors (AIs). Although several predictors have been identified, it is unclear how to simultaneously consider multiple risk factors for an individual. We sought to develop a tool for prediction of AI discontinuation and to explore how predictive value of risk factors changes with time.

Materials and Methods

Survival machine learning was used to predict time-to-discontinuation of AIs in 181 women who enrolled in a prospective cohort. Models were evaluated via time-dependent area under the curve (AUC), c-index, and integrated Brier score. Feature importance was analysis was conducted via Shapley Additive Explanations (SHAP) and time-dependence of their predictive value was analyzed by time-dependent AUC. Personalized survival curves were constructed for risk communication.

Results

The best-performing model incorporated genetic risk factors and changes in patient-reported outcomes, achieving mean time-dependent AUC of 0.66, and AUC of 0.72 and 0.67 at 6- and 12-month cutoffs, respectively. The most significant features included variants in ESR1 and emergent symptoms. Predictive value of genetic risk factors was highest in the first year of treatment. Decrease in physical function was the strongest independent predictor at follow-up.

Discussion and Conclusion

Incorporation of genomic and 3-month follow-up data improved the ability of the models to identify the individuals at risk of AI discontinuation. Genetic risk factors were particularly important for predicting early discontinuers. This study provides insight into the complex nature of AI discontinuation and highlights the importance of incorporating genetic risk factors and emergent symptoms into prediction models.

Predicting time-to-intubation after critical care admission using machine learning and cured fraction information

Intubation for mechanical ventilation (MV) is one of the most common high-risk procedures performed in Intensive Care Units (ICUs). Early prediction of intubation may have a positive impact by providing timely alerts to clinicians and consequently avoiding high-risk late intubations. In this work, we propose a new machine learning method to predict the time to intubation during the first five days of ICU admission, based on the concept of cure survival models. Our approach combines classification and survival analysis, to effectively accommodate the fraction of patients not at risk of intubation, and provide a better estimate of time to intubation, for patients at risk. We tested our approach and compared it to other predictive models on a dataset collected from a secondary care hospital (AZ Groeninge, Kortrijk, Belgium) from 2015 to 2021, consisting of 3425 ICU stays. Furthermore, we utilised SHAP for feature importance analysis, extracting key insights into the relative significance of variables such as vital signs, blood gases, and patient characteristics in predicting intubation in ICU settings. The results corroborate that our approach improves the prediction of time to intubation in critically ill patients, by using routinely collected data within the first hours of admission in the ICU. Early warning of the need for intubation may be used to help clinicians predict the risk of intubation and rank patients according to their expected time to intubation.

Hybridizing mechanistic mathematical modeling with deep learning methods to predict individual cancer patient survival after immune checkpoint inhibitor therapy

We present a study where predictive mechanistic modeling is used in combination with deep learning methods to predict individual patient survival probabilities under immune checkpoint inhibitor (ICI) therapy. This hybrid approach enables prediction based on both measures that are calculable from mechanistic models (but may not be directly measurable in the clinic) and easily measurable quantities or characteristics (that are not always readily incorporated into predictive mechanistic models). The mechanistic model we have applied here can predict tumor response from CT or MRI imaging based on key mechanisms underlying checkpoint inhibitor therapy, and in the present work, its parameters were combined with readily-available clinical measures from 93 patients into a hybrid training set for a deep learning time-to-event predictive model. Analysis revealed that training an artificial neural network with both mechanistic modeling-derived and clinical measures achieved higher per-patient predictive accuracy based on event-time concordance, Brier score, and negative binomial log-likelihood-based criteria than when only mechanistic model-derived values or only clinical data were used. Feature importance analysis revealed that both clinical and model-derived parameters play prominent roles in neural network decision making, and in increasing prediction accuracy, further supporting the advantage of our hybrid approach. We anticipate that many existing mechanistic models may be hybridized with deep learning methods in a similar manner to improve predictive accuracy through addition of additional data that may not be readily implemented in mechanistic descriptions.

Generation of synthetic whole-slide image tiles of tumours from RNA-sequencing data via cascaded diffusion models

Training machine-learning models with synthetically generated data can alleviate the problem of data scarcity when acquiring diverse and sufficiently large datasets is costly and challenging. Here we show that cascaded diffusion models can be used to synthesize realistic whole-slide image tiles from latent representations of RNA-sequencing data from human tumours. Alterations in gene expression affected the composition of cell types in the generated synthetic image tiles, which accurately preserved the distribution of cell types and maintained the cell fraction observed in bulk RNA-sequencing data, as we show for lung adenocarcinoma, kidney renal papillary cell carcinoma, cervical squamous cell carcinoma, colon adenocarcinoma and glioblastoma. Machine-learning models pretrained with the generated synthetic data performed better than models trained from scratch. Synthetic data may accelerate the development of machine-learning models in scarce-data settings and allow for the imputation of missing data modalities.

ResDeepSurv: A Survival Model for Deep Neural Networks Based on Residual Blocks and Self-attention Mechanism

Survival analysis, as a widely used method for analyzing and predicting the timing of event occurrence, plays a crucial role in the medicine field. Medical professionals utilize survival models to gain insight into the effects of patient covariates on the disease, and the correlation with the effectiveness of different treatment strategies. This knowledge is essential for the development of treatment plans and the enhancement of treatment approaches. Conventional survival models, such as the Cox proportional hazards model, require a significant amount of feature engineering or prior knowledge to facilitate personalized modeling. To address these limitations, we propose a novel residual-based self-attention deep neural network for survival modeling, called ResDeepSurv, which combines the benefits of neural networks and the Cox proportional hazards regression model. The model proposed in our study simulates the distribution of survival time and the correlation between covariates and outcomes, but does not impose strict assumptions on the basic distribution of survival data. This approach effectively accounts for both linear and nonlinear risk functions in survival data analysis. The performance of our model in analyzing survival data with various risk functions is on par with or even superior to that of other existing survival analysis methods. Furthermore, we validate the superior performance of our model in comparison to currently existing methods by evaluating multiple publicly available clinical datasets. Through this study, we prove the effectiveness of our proposed model in survival analysis, providing a promising alternative to traditional approaches. The application of deep learning techniques and the ability to capture complex relationships between covariates and survival outcomes without relying on extensive feature engineering make our model a valuable tool for personalized medicine and decision-making in clinical practice.

Clinical Feature Ranking Based on Ensemble Machine Learning Reveals Top Survival Factors for Glioblastoma Multiforme

Glioblastoma multiforme (GM) is a malignant tumor of the central nervous system considered to be highly aggressive and often carrying a terrible survival prognosis. An accurate prognosis is therefore pivotal for deciding a good treatment plan for patients. In this context, computational intelligence applied to data of electronic health records (EHRs) of patients diagnosed with this disease can be useful to predict the patients' survival time. In this study, we evaluated different machine learning models to predict survival time in patients suffering from glioblastoma and further investigated which features were the most predictive for survival time. We applied our computational methods to three different independent open datasets of EHRs of patients with glioblastoma: the Shieh dataset of 84 patients, the Berendsen dataset of 647 patients, and the Lammer dataset of 60 patients. Our survival time prediction techniques obtained concordance index (C-index) = 0.583 in the Shieh dataset, C-index = 0.776 in the Berendsen dataset, and C-index = 0.64 in the Lammer dataset, as best results in each dataset. Since the original studies regarding the three datasets analyzed here did not provide insights about the most predictive clinical features for survival time, we investigated the feature importance among these datasets. To this end, we then utilized Random Survival Forests, which is a decision tree-based algorithm able to model non-linear interaction between different features and might be able to better capture the highly complex clinical and genetic status of these patients. Our discoveries can impact clinical practice, aiding clinicians and patients alike to decide which therapy plan is best suited for their unique clinical status.

Causes of death among people living with metastatic cancer

Studying survivorship and causes of death in patients with advanced or metastatic cancer remains an important task. We characterize the causes of death among patients with metastatic cancer, across 13 cancer types and 25 non-cancer causes and predict the risk of death after diagnosis from the diagnosed cancer versus other causes (e.g., stroke, heart disease, etc.). Among 1,030,937 US (1992-2019) metastatic cancer survivors, 82.6% of patients (n = 688,529) died due to the diagnosed cancer, while 17.4% (n = 145,006) died of competing causes. Patients with lung, pancreas, esophagus, and stomach tumors are the most likely to die of their metastatic cancer, while those with prostate and breast cancer have the lowest likelihood. The median survival time among patients living with metastases is 10 months; our Fine and Gray competing risk model predicts 1 year survival with area under the receiver operating characteristic curve of 0.754 (95% CI [0.754, 0.754]). Leading non-cancer deaths are heart disease (32.4%), chronic obstructive and pulmonary disease (7.9%), cerebrovascular disease (6.1%), and infection (4.1%).

Risk factors affecting patients survival with colorectal cancer in Morocco: survival analysis using an interpretable machine learning approach

The aim of our study was to assess the overall survival rates for colorectal cancer at 3 years and to identify associated strong prognostic factors among patients in Morocco through an interpretable machine learning approach. This approach is based on a fully non-parametric survival random forest (RSF), incorporating variable importance and partial dependence effects. The data was povided from a retrospective study of 343 patients diagnosed and followed at Hassan II University Hospital. Covariate selection was performed using the variable importance based on permutation and partial dependence plots were displayed to explore in depth the relationship between the estimated partial effect of a given predictor and survival rates. The predictive performance was measured by two metrics, the Concordance Index (C-index) and the Brier Score (BS). Overall survival rates at 1, 2 and 3 years were, respectively, 87% (SE = 0.02; CI-95% 0.84-0.91), 77% (SE = 0.02; CI-95% 0.73-0.82) and 60% (SE = 0.03; CI-95% 0.54-0.66). In the Cox model after adjustment for all covariates, sex, tumor differentiation had no significant effect on prognosis, but rather tumor site had a significant effect. The variable importance obtained from RSF strengthens that surgery, stage, insurance, residency, and age were the most important prognostic factors. The discriminative capacity of the Cox PH and RSF was, respectively, 0.771 and 0.798 for the C-index while the accuracy of the Cox PH and RSF was, respectively, 0.257 and 0.207 for the BS. This shows that RSF had both better discriminative capacity and predictive accuracy. Our results show that patients who are older than 70, living in rural areas, without health insurance, at a distant stage and who have not had surgery constitute a subgroup of patients with poor prognosis.

A Bayesian survival treed hazards model using latent Gaussian processes

Survival models are used to analyze time-to-event data in a variety of disciplines. Proportional hazard models provide interpretable parameter estimates, but proportional hazard assumptions are not always appropriate. Non-parametric models are more flexible but often lack a clear inferential framework. We propose a Bayesian treed hazards partition model that is both flexible and inferential. Inference is obtained through the posterior tree structure and flexibility is preserved by modeling the log-hazard function in each partition using a latent Gaussian process. An efficient reversible jump Markov chain Monte Carlo algorithm is accomplished by marginalizing the parameters in each partition element via a Laplace approximation. Consistency properties for the estimator are established. The method can be used to help determine subgroups as well as prognostic and/or predictive biomarkers in time-to-event data. The method is compared with some existing methods on simulated data and a liver cirrhosis dataset.

Machine learning-based clinical decision support system for treatment recommendation and overall survival prediction of hepatocellular carcinoma: a multi-center study

The treatment decisions for patients with hepatocellular carcinoma are determined by a wide range of factors, and there is a significant difference between the recommendations of widely used staging systems and the actual initial treatment choices. Herein, we propose a machine learning-based clinical decision support system suitable for use in multi-center settings. We collected data from nine institutions in South Korea for training and validation datasets. The internal and external datasets included 935 and 1750 patients, respectively. We developed a model with 20 clinical variables consisting of two stages: the first stage which recommends initial treatment using an ensemble voting machine, and the second stage, which predicts post-treatment survival using a random survival forest algorithm. We derived the first and second treatment options from the results with the highest and the second-highest probabilities given by the ensemble model and predicted their post-treatment survival. When only the first treatment option was accepted, the mean accuracy of treatment recommendation in the internal and external datasets was 67.27% and 55.34%, respectively. The accuracy increased to 87.27% and 86.06%, respectively, when the second option was included as the correct answer. Harrell's C index, integrated time-dependent AUC curve, and integrated Brier score of survival prediction in the internal and external datasets were 0.8381 and 0.7767, 91.89 and 86.48, 0.12, and 0.14, respectively. The proposed system can assist physicians by providing data-driven predictions for reference from other larger institutions or other physicians within the same institution when making treatment decisions.

Investigation of factors affecting physical activity level in patients with primary Sjögren's syndrome

OBJECTIVES

This study aimed to determine physical activity levels and understand the factors influencing an active lifestyle among patients with primary Sjögren's syndrome (pSS).

METHODS

Ninety-seven patients participated in this multicentric study. Physical activity levels were assessed using the International Physical Activity Questionnaire-Short Form (IPAQ-SF). The Inflammatory Arthritis Facilitators and Barriers (IFAB) questionnaire was used to evaluate perceived barriers and facilitators to physical activity.

RESULTS

Forty-six patients were physically inactive and the rest of them were moderately active. Commonly identified barriers included a lack of motivation, fatigue, and pain. Conversely, knowledge of the health and mood benefits for physical activity emerged as a key motivator. Patients with better scores on facilitators and lower scores on barriers exhibited higher physical activity levels (p < 0.05). Notably, a high level of perceived facilitators of physical activity (odds ratio [OR]: 1.02; 95% confidence interval [CI], 1.00 – 1.05) and reduced pain (OR: 0.81; 95% CI: 0.69 – 0.95) were linked to an active lifestyle.

CONCLUSIONS

This study emphasizes the role of motivation and awareness of the benefits of physical activity for health and mood in driving physical activity for patients with primary Sjögren’s syndrome. Tailored physical activity programs that address psychological aspects and disease-related pain, and fatigue should be designed to counter sedentary lifestyles in pSS patients.

Assessing model prediction performance for the expected cumulative number of recurrent events

In a recurrent event setting, we introduce a new score designed to evaluate the prediction ability, for a given model, of the expected cumulative number of recurrent events. This score can be seen as an extension of the Brier Score for single time to event data but works for recurrent events with or without a terminal event. Theoretical results are provided that show that under standard assumptions in a recurrent event context, our score can be asymptotically decomposed as the sum of the theoretical mean squared error between the model and the true expected cumulative number of recurrent events and an inseparability term that does not depend on the model. This decomposition is further illustrated on simulations studies. It is also shown that this score should be used in comparison with a reference model, such as a nonparametric estimator that does not include the covariates. Finally, the score is applied for the prediction of hospitalisations on a dataset of patients suffering from atrial fibrillation and a comparison of the prediction performances of different models, such as the Cox model, the Aalen Model or the Ghosh and Lin model, is investigated.

Dynamic updating of clinical survival prediction models in a changing environment

BACKGROUND

Over time, the performance of clinical prediction models may deteriorate due to changes in clinical management, data quality, disease risk and/or patient mix. Such prediction models must be updated in order to remain useful. In this study, we investigate dynamic model updating of clinical survival prediction models. In contrast to discrete or one-time updating, dynamic updating refers to a repeated process for updating a prediction model with new data. We aim to extend previous research which focused largely on binary outcome prediction models by concentrating on time-to-event outcomes. We were motivated by the rapidly changing environment seen during the COVID-19 pandemic where mortality rates changed over time and new treatments and vaccines were introduced.

METHODS

We illustrate three methods for dynamic model updating: Bayesian dynamic updating, recalibration, and full refitting. We use a simulation study to compare performance in a range of scenarios including changing mortality rates, predictors with low prevalence and the introduction of a new treatment. Next, the updating strategies were applied to a model for predicting 70-day COVID-19-related mortality using patient data from QResearch, an electronic health records database from general practices in the UK.

RESULTS

In simulated scenarios with mortality rates changing over time, all updating methods resulted in better calibration than not updating. Moreover, dynamic updating outperformed ad hoc updating. In the simulation scenario with a new predictor and a small updating dataset, Bayesian updating improved the C-index over not updating and refitting. In the motivating example with a rare outcome, no single updating method offered the best performance.

CONCLUSIONS

We found that a dynamic updating process outperformed one-time discrete updating in the simulations. Bayesian updating offered good performance overall, even in scenarios with new predictors and few events. Intercept recalibration was effective in scenarios with smaller sample size and changing baseline hazard. Refitting performance depended on sample size and produced abrupt changes in hazard ratio estimates between periods.

Development and Validation of the CANHEART Population-Based Laboratory Prediction Models for Atherosclerotic Cardiovascular Disease

BACKGROUND

Prediction of atherosclerotic cardiovascular disease (ASCVD) in primary prevention assessments exclusively with laboratory results may facilitate automated risk reporting and improve uptake of preventive therapies.

OBJECTIVE

To develop and validate sex-specific prediction models for ASCVD using age and routine laboratory tests and compare their performance with that of the pooled cohort equations (PCEs).

DESIGN

Derivation and validation of the CANHEART (Cardiovascular Health in Ambulatory Care Research Team) Lab Models.

SETTING

Population-based cohort study in Ontario, Canada.

PARTICIPANTS

A derivation and internal validation cohort of adults aged 40 to 75 years without cardiovascular disease from April 2009 to December 2015; an external validation cohort of primary care patients from January 2010 to December 2014.

MEASUREMENTS

Age and laboratory predictors measured in the outpatient setting included serum total cholesterol, high-density lipoprotein cholesterol, triglycerides, hemoglobin, mean corpuscular volume, platelets, leukocytes, estimated glomerular filtration rate, and glucose. The ASCVD outcomes were defined as myocardial infarction, stroke, and death from ischemic heart or cerebrovascular disease within 5 years.

RESULTS

Sex-specific models were developed and internally validated in 2 160 497 women and 1 833 147 men. They were well calibrated, with relative differences less than 1% between mean predicted and observed risk for both sexes. The c-statistic was 0.77 in women and 0.71 in men. External validation in 31 697 primary care patients showed a relative difference less than 14% and an absolute difference less than 0.3 percentage points in mean predicted and observed risks for both sexes. The c-statistics for the laboratory models were 0.72 for both sexes and were not statistically significantly different from those for the PCEs in women (change in c-statistic, -0.01 [95% CI, -0.03 to 0.01]) or men (change in c-statistic, -0.01 [CI, -0.04 to 0.02]).

LIMITATION

Medication use was not available at the population level.

CONCLUSION

The CANHEART Lab Models predict ASCVD with similar accuracy to more complex models, such as the PCEs.

PRIMARY FUNDING SOURCE

Canadian Institutes of Health Research.

Machine learning model for predicting physical activity related bleeding risk in Chinese boys with haemophilia A

BACKGROUND

Physical activity is a crucial part of an active lifestyle for haemophiliac children. However, the fear of bleeds has been identified as barriers to participating physical activity for haemophiliac children even with prophylaxis. Lack of evidence and metrics driven by data is key problem.

OBJECTIVES

We aim to develop machine learning models based on clinical data with multiple potential factors considered to predict risk of physical activity bleeding for haemophilia children with prophylaxis.

METHODS

From this cohort study, we collected information on 98 haemophiliac children with adequate prophylaxis (trough FVIII:C level > 1 %). The involved potential predictor variables include demographic information, treatment information, physical activity, joint evaluation, and pharmacokinetic parameters, etc. We applied CoxPH, Random Survival Forests (RSF) and DeepSurv to construct prediction models for the risk of bleeding during physical activities. All three survival analysis models were internally and externally validated.

RESULTS

A total of 98 patients were enrolled in this study. Their median age was 7.9 (5.5, 10.2) years. The CoxPH, RSF and DeepSurv models' discriminative and calibration abilities were all high, and the RSF model had the best performance (Internal validation: C-index, 0.7648 ± 0.0139; Brier Score, 0.1098 ± 0.0015; External validation: C-index, 0.7260 ± 0.0154; Brier Score, 0.0930 ± 0.0018). The prediction curves demonstrated that the developed RSF model can distinguish the risks well between bleeding and non-bleeding patients, as well as patients with different levels of physical activity. Meanwhile, the feature importance analysis confirmed that physical activity bleeding was deduced by comprehensive effects of various factors, and the importance of different factors on bleeding outcome is discrepant.

CONCLUSIONS

This study revealed from the mechanism that it is necessary to incorporate multiple factors to accurately predict physical activity related bleeding risk. In clinical practice, the designed machine learning models can provide guidance for children with haemophilia A to positively participate in physical activity.

Risk Stratification Model for Hepatocellular Cancer in Patients With Cirrhosis

BACKGROUND & AIMS

The available risk stratification indices for hepatocellular cancer (HCC) have limited applicability. We developed and externally validated an HCC risk stratification index in U.S. cohorts of patients with cirrhosis.

METHODS

We used data from 2 prospective U.S. cohorts to develop the risk index. Patients with cirrhosis were enrolled from 8 centers and followed until development of HCC, death, or December 31, 2021. We identified an optimal set of predictors with the highest discriminatory ability (C-index) for HCC. The predictors were refit using competing risk regression and its predictive performance was evaluated using the area under the receiver-operating characteristic curve (AUROC). External validation was performed in a cohort of 21,550 patients with cirrhosis seen in the U.S Veterans Affairs system between 2018 and 2019 with follow-up through 2021.

RESULTS

We developed the model in 2431 patients (mean age 60 years, 31% women, 24% cured hepatitis C, 16% alcoholic liver disease, and 29% nonalcoholic fatty liver disease). The selected model had a C-index of 0.77 (95% confidence interval [CI], 0.73-0.81), and the predictors were age, sex, smoking, alcohol use, body mass index, etiology, α-fetoprotein, albumin, alanine aminotransferase, and platelet levels. The AUROCs were 0.75 (95% CI, 0.65-0.85) at 1 year and 0.77 (95% CI, 0.71-0.83) at 2 years, and the model was well calibrated. In the external validation cohort, the AUROC at 2 years was 0.70 with excellent calibration.

CONCLUSION

The risk index, including objective and routinely available risk factors, can differentiate patients with cirrhosis who will develop HCC and help guide discussions regarding HCC surveillance and prevention. Future studies are needed for additional external validation and refinement of risk stratification.

A prognostic model for predicting progression-free survival in patients with advanced non-small cell lung cancer after image-guided microwave ablation plus chemotherapy

OBJECTIVES

This study aimed to build and validate a prediction model that can predict progression-free survival (PFS) in patients with advanced non-small cell lung cancer (NSCLC) after image-guided microwave ablation (MWA) plus chemotherapy.

METHODS

Data from a previous multi-center randomized controlled trial (RCT) was used and assigned to either the training data set or the external validation data set according to the location of the centers. Potential prognostic factors were identified by multivariable analysis in the training data set and used to construct a nomogram. After bootstraps internal and external validation, the predictive performance was evaluated by concordance index (C-index), Brier Score, and calibration curves. Risk group stratification was conducted using the score calculated by the nomogram. Then a simplified scoring system was built to make risk group stratification more convenient.

RESULTS

In total, 148 patients (training data set: n = 112; external validation data set: n = 36) were enrolled for analysis. Six potential predictors were identified and entered into the nomogram, including weight loss, histology, clinical TNM stage, clinical N category, tumor location, and tumor size. The C-indexes were 0.77 (95% CI, 0.65-0.88, internal validation) and 0.64 (95% CI, 0.43-0.85, external validation). The survival curves of different risk groups also displayed significant distinction (p < 0.0001).

CONCLUSIONS

We found weight loss, histology, clinical TNM stage, clinical N category, tumor location, and tumor size were prognostic factors of progression after receiving MWA plus chemotherapy and constructed a prediction model that can predict PFS.

CLINICAL RELEVANCE STATEMENT

The nomogram and scoring system will assist physicians to predict the individualized PFS of their patients and decide whether to perform or terminate MWA and chemotherapy according to the expected benefits.

KEY POINTS

• Build and validate a prognostic model using the data from a previous randomized controlled trial to predict progression-free survival after receiving MWA plus chemotherapy. • Weight loss, histology, clinical TNM stage, clinical N category, tumor location, and tumor size were prognostic factors. • The nomogram and scoring system published by the prediction model can be used to assist physicians to make clinical decisions.

Survival analysis for pediatric heart transplant patients using a novel machine learning algorithm: A UNOS analysis

BACKGROUND

Impact of pretransplantation risk factors on mortality in the first year after heart transplantation remains largely unknown. Using machine learning algorithms, we selected clinically relevant identifiers that could predict 1-year mortality after pediatric heart transplantation.

METHODS

Data were obtained from the United Network for Organ Sharing Database for years 2010-2020 for patients 0-17 years receiving their first heart transplant (N = 4150). Features were selected using subject experts and literature review. Scikit-Learn, Scikit-Survival, and Tensorflow were used. A train:test split of 70:30 was used. N-repeated k-fold validation was performed (N = 5, k = 5). Seven models were tested, Hyperparameter tuning performed using Bayesian optimization and the concordance index (C-index) was used for model assessment.

RESULTS

A C-index above 0.6 for test data was considered acceptable for survival analysis models. C-indices obtained were 0.60 (Cox proportional hazards), 0.61 (Cox with elastic net), 0.64 (gradient boosting), 0.64 (support vector machine), 0.68 (random forest), 0.66 (component gradient boosting), and 0.54 (survival trees). Machine learning models show an improvement over the traditional Cox proportional hazards model, with random forest performing the best on the test set. Analysis of the feature importance for the gradient boosted model found that the top 5 features were the most recent serum total bilirubin, the travel distance from the transplant center, the patient body mass index, the deceased donor terminal Serum glutamic pyruvic transaminase/Alanine transaminase (SGPT/ALT), and the donor PCO2.

CONCLUSIONS

Combination of machine learning and expert-based methodology of selecting predictors of survival for pediatric heart transplantation provides a reasonable prediction of 1- and 3-year survival outcomes. SHapley Additive exPlanations can be an effective tool for modeling and visualizing nonlinear interactions.

Distance-Metric Learning for Personalized Survival Analysis

Personalized time-to-event or survival prediction with right-censored outcomes is a pervasive challenge in healthcare research. Although various supervised machine learning methods, such as random survival forests or neural networks, have been adapted to handle such outcomes effectively, they do not provide explanations for their predictions, lacking interpretability. In this paper, an alternative method for survival prediction by weighted nearest neighbors is proposed. Fitting this model to data entails optimizing the weights by learning a metric. An individual prediction of this method can be explained by providing the user with the most influential data points for this prediction, i.e., the closest data points and their weights. The strengths and weaknesses in terms of predictive performance are highlighted on simulated data and an application of the method on two different real-world datasets of breast cancer patients shows its competitiveness with established methods.

Colorectal cancer risk stratification on histological slides based on survival curves predicted by deep learning

Studies have shown that colorectal cancer prognosis can be predicted by deep learning-based analysis of histological tissue sections of the primary tumor. So far, this has been achieved using a binary prediction. Survival curves might contain more detailed information and thus enable a more fine-grained risk prediction. Therefore, we established survival curve-based CRC survival predictors and benchmarked them against standard binary survival predictors, comparing their performance extensively on the clinical high and low risk subsets of one internal and three external cohorts. Survival curve-based risk prediction achieved a very similar risk stratification to binary risk prediction for this task. Exchanging other components of the pipeline, namely input tissue and feature extractor, had largely identical effects on model performance independently of the type of risk prediction. An ensemble of all survival curve-based models exhibited a more robust performance, as did a similar ensemble based on binary risk prediction. Patients could be further stratified within clinical risk groups. However, performance still varied across cohorts, indicating limited generalization of all investigated image analysis pipelines, whereas models using clinical data performed robustly on all cohorts.

GNN-surv: Discrete-Time Survival Prediction Using Graph Neural Networks

Survival prediction models play a key role in patient prognosis and personalized treatment. However, their accuracy can be improved by incorporating patient similarity networks, which uncover complex data patterns. Our study uses Graph Neural Networks (GNNs) to enhance discrete-time survival predictions (GNN-surv) by leveraging relationships in these networks. We build these networks using cancer patients' genomic and clinical data and train various GNN models on them, integrating Logistic Hazard and PMF survival models. GNN-surv models exhibit superior performance in survival prediction across two urologic cancer datasets, outperforming traditional MLP models. They maintain robustness and effectiveness under varying graph construction hyperparameter μ values, with performance boosts of up to 14.6% and 7.9% in the time-dependent concordance index and reductions in the integrated brier score of 26.7% and 24.1% in the BLCA and KIRC datasets, respectively. Notably, these models also maintain their effectiveness across three different types of GNN models, suggesting potential adaptability to other cancer datasets. The superior performance of our GNN-surv models underscores their wide applicability in the fields of oncology and personalized medicine, providing clinicians with a more accurate tool for patient prognosis and personalized treatment planning. Future studies can further optimize these models by incorporating other survival models or additional data modalities.

BAYESIAN INFERENCE AND DYNAMIC PREDICTION FOR MULTIVARIATE LONGITUDINAL AND SURVIVAL DATA

Alzheimer's disease (AD) is a complex neurological disorder impairing multiple domains such as cognition and daily functions. To better understand the disease and its progression, many AD research studies collect multiple longitudinal outcomes that are strongly predictive of the onset of AD dementia. We propose a joint model based on a multivariate functional mixed model framework (referred to as MFMM-JM) that simultaneously models the multiple longitudinal outcomes and the time to dementia onset. We develop six functional forms to fully investigate the complex association between longitudinal outcomes and dementia onset. Moreover, we use the Bayesian methods for statistical inference and develop a dynamic prediction framework that provides accurate personalized predictions of disease progressions based on new subject-specific data. We apply the proposed MFMM-JM to two large ongoing AD studies: the Alzheimer's Disease Neuroimaging Initiative (ADNI) and National Alzheimer's Coordinating Center (NACC), and identify the functional forms with the best predictive performance. our method is also validated by extensive simulation studies with five settings.

Tackling dynamic prediction of death in patients with recurrent cardiovascular events

In the field of cardiovascular disease, recurrent events such as stroke or myocardial infarction (MI) are often encountered, leading to an increase in the risk of death. Accurately evaluating the prognosis of patients and dynamically predicting the risk of death by considering the historical recurrent events can improve medical decisions and lead to better health care outcomes. Recently proposed joint modeling approaches within the Bayesian framework have inspired the development of a dynamic prediction tool, which can be applied for subject-level prediction of death with implementation in software packages. The prediction model incorporates subject heterogeneity with subject-level random effects that account for unobserved time-invariant factors and an extra copula function capturing the part caused by unmeasured time-dependent factors. Thereafter, given the prespecified landmark timet ' $$ {t}^{\prime } $$ , the survival probability for a prediction horizon time of interestt $$ t $$ can be estimated for each individual. The prediction accuracy is assessed by time-dependent receiving operating characteristic curve and the area under the curve and the Brier score with calibration plots is compared to traditional joint frailty models. Finally, the tool is applied to patients with multiple attacks of stroke or MI in the Cardiovascular Health study and the Atherosclerosis Risk in Communities study for illustration.

Comparing predictions among competing risks models with rare events: application to KNOW-CKD atudy-a multicentre cohort study of chronic kidney disease

A prognostic model to determine an association between survival outcomes and clinical risk factors, such as the Cox model, has been developed over the past decades in the medical field. Although the data size containing subjects' information gradually increases, the number of events is often relatively low as medical technology develops. Accordingly, poor discrimination and low predicted ability may occur between low- and high-risk groups. The main goal of this study was to evaluate the predicted probabilities with three existing competing risks models in variation with censoring rates. Three methods were illustrated and compared in a longitudinal study of a nationwide prospective cohort of patients with chronic kidney disease in Korea. The prediction accuracy and discrimination ability of the three methods were compared in terms of the Concordance index (C-index), Integrated Brier Score (IBS), and Calibration slope. In addition, we find that these methods have different performances when the effects are linear or nonlinear under various censoring rates.

Pretherapy investigations using highly robust visualized biomarkers from CT imaging by multiple machine-learning techniques toward its prognosis prediction for ALK-inhibitor therapy in NSCLC: a feasibility study

PURPOSE

Molecularly targeted therapy has revolutionized the therapeutic landscape and is emerging as the first-line treatment option for ALK-rearranged non-small-cell lung cancer (NSCLC). In this study, the highly informative and robust biomarkers based on pre-treatment CT images and clinicopathologic features will be developed and validated to predict the prognosis for ALK-inhibitor therapy in NSCLC patients.

METHODS

A total of 161 ALK-positive NSCLC patients treated with ALK inhibitors were retrospectively collected as training, validation and test sets from multi-center institutions. Cox proportional hazard regression (CPH) penalized by LASSO and random survival forest (RSF) coupled with recursive feature elimination (RFE) were used for radiomics and clinical features identification and model construction. An overlapping post-processing method was extra added to training process to investigate the stronger biomarker on the whole set.

RESULTS

123 of the collected cases progressed after a median follow-up of 15.5 months (IQR, 8.3-25.3). The T and M staging, pericardial effusion, age and ALK inhibitor-alectinib were determined as significant predictors in the survival analysis. Furthermore, we visualized the finally retained 4 radiomics feature. The RSF models built from overlapping-processed clinical and radiomics features respectively reached the maximum C-index of 0.68 and 0.75,but the combination of them,radioclinical signature, improved the score to 0.78. The model on the validation and external test datasets yielded the C-index of 0.73 and 0.79, with the iAUC of 0.76 and 0.83, the IBS of 0.119 and 0.112.

CONCLUSION

With respect to a simple selection strategy of overlapping optimal radiomics and clinical features from different survival models may promote better progression-free survival(PFS) prediction than conventional survival analysis, which provides a potential method for guiding personalized pre-treatment options of NSCLC.

Spatial cellular architecture predicts prognosis in glioblastoma

Intra-tumoral heterogeneity and cell-state plasticity are key drivers for the therapeutic resistance of glioblastoma. Here, we investigate the association between spatial cellular organization and glioblastoma prognosis. Leveraging single-cell RNA-seq and spatial transcriptomics data, we develop a deep learning model to predict transcriptional subtypes of glioblastoma cells from histology images. Employing this model, we phenotypically analyze 40 million tissue spots from 410 patients and identify consistent associations between tumor architecture and prognosis across two independent cohorts. Patients with poor prognosis exhibit higher proportions of tumor cells expressing a hypoxia-induced transcriptional program. Furthermore, a clustering pattern of astrocyte-like tumor cells is associated with worse prognosis, while dispersion and connection of the astrocytes with other transcriptional subtypes correlate with decreased risk. To validate these results, we develop a separate deep learning model that utilizes histology images to predict prognosis. Applying this model to spatial transcriptomics data reveal survival-associated regional gene expression programs. Overall, our study presents a scalable approach to unravel the transcriptional heterogeneity of glioblastoma and establishes a critical connection between spatial cellular architecture and clinical outcomes.

Deep Learning Approaches for Glioblastoma Prognosis in Resource-Limited Settings: A Study Using Basic Patient Demographic, Clinical, and Surgical Inputs

BACKGROUND

Glioblastoma (GBM) is the most common brain tumor in the United States, with an annual incidence rate of 3.21 per 100,000. It is the most aggressive type of diffuse glioma and has a median survival of months after treatment. This study aims to assess the accuracy of different novel deep learning models trained on a set of simple clinical, demographic, and surgical variables to assist in clinical practice, even in areas with constrained health care infrastructure.

METHODS

Our study included 37,095 patients with GBM from the SEER (Surveillance Epidemiology and End Results) database. All predictors were based on demographic, clinicopathologic, and treatment information of the cases. Our outcomes of interest were months of survival and vital status. Concordance index (C-index) and integrated Brier scores (IBS) were used to evaluate the performance of the models.

RESULTS

The patient characteristics and the statistical analyses were consistent with the epidemiologic literature. The models C-index and IBS ranged from 0.6743 to 0.6918 and from 0.0934 to 0.1034, respectively. Probabilistic matrix factorization (0.6918), multitask logistic regression (0.6916), and logistic hazard (0.6916) had the highest C-index scores. The models with the lowest IBS were the probabilistic matrix factorization (0.0934), multitask logistic regression (0.0935), and logistic hazard (0.0936). These models had an accuracy (1-IBS) of 90.66%; 90.65%, and 90.64%, respectively. The deep learning algorithms were deployed on an interactive Web-based tool for practical use available via https://glioblastoma-survanalysis.herokuapp.com/.

CONCLUSIONS

Novel deep learning algorithms can better predict GBM prognosis than do baseline methods and can lead to more personalized patient care regardless of extensive electronic health record availability.

A novel prognostic index for diffuse large B-cell lymphoma combined baseline metabolic tumour volume with clinical and pathological risk factors

OBJECTIVES

This study aimed to develop a novel prognostic index integrating baseline metabolic tumour volume (MTV) along with clinical and pathological parameters for diffuse large B-cell lymphoma (DLBCL).

METHODS

This prospective trial enrolled 289 patients with newly diagnosed DLBCL. The predictive value of novel prognostic index was compared with Ann Arbor staging and National Comprehensive Cancer Network International Prognostic Index (NCCN-IPI). We used the concordance index (C-index) and a calibration curve to determine its predictive capacity.

RESULTS

Multivariate analysis revealed high MTV (>191 cm 3 ), Ann Arbor stage (III-IV) and MYC/BCL2 double expression lymphoma (DEL) to be independently associated with inferior progression-free survival (PFS) and overall survival (OS). Ann Arbor stage and DEL could be stratified by MTV. Our index, combining MTV with Ann Arbor stage and DEL status, identified four prognostic groups: group 1 (no risk factors,), group 2 (one risk factor), group 3 (two risk factors), and group 4 (three risk factors). The 2-year PFS rates were 85.5, 73.9, 53.6, and 13.9%; 2-year OS rates were 94.6, 87.0, 67.5, and 24.2%, respectively. The C-index values of the novel index were 0.697 and 0.753 for PFS and OS prediction, which was superior to Ann Arbor stage and NCCN-IPI.

CONCLUSION

The novel index including tumour burden and clinicopathological features may help predict outcome of DLBCL (clinicaltrials.gov identifier: NCT02928861).

Time-related survival prediction in molecular subtypes of breast cancer using time-to-event deep-learning-based models

Background

Breast cancer (BC) survival prediction can be a helpful tool for identifying important factors selecting the effective treatment reducing mortality rates. This study aims to predict the time-related survival probability of BC patients in different molecular subtypes over 30 years of follow-up.

Materials and methods

This study retrospectively analyzed 3580 patients diagnosed with invasive breast cancer (BC) from 1991 to 2021 in the Cancer Research Center of Shahid Beheshti University of Medical Science. The dataset contained 18 predictor variables and two dependent variables, which referred to the survival status of patients and the time patients survived from diagnosis. Feature importance was performed using the random forest algorithm to identify significant prognostic factors. Time-to-event deep-learning-based models, including Nnet-survival, DeepHit, DeepSurve, NMLTR and Cox-time, were developed using a grid search approach with all variables initially and then with only the most important variables selected from feature importance. The performance metrics used to determine the best-performing model were C-index and IBS. Additionally, the dataset was clustered based on molecular receptor status (i.e., luminal A, luminal B, HER2-enriched, and triple-negative), and the best-performing prediction model was used to estimate survival probability for each molecular subtype.

Results

The random forest method identified tumor state, age at diagnosis, and lymph node status as the best subset of variables for predicting breast cancer (BC) survival probabilities. All models yielded very close performance, with Nnet-survival (C-index=0.77, IBS=0.13) slightly higher using all 18 variables or the three most important variables. The results showed that the Luminal A had the highest predicted BC survival probabilities, while triple-negative and HER2-enriched had the lowest predicted survival probabilities over time. Additionally, the luminal B subtype followed a similar trend as luminal A for the first five years, after which the predicted survival probability decreased steadily in 10- and 15-year intervals.

Conclusion

This study provides valuable insight into the survival probability of patients based on their molecular receptor status, particularly for HER2-positive patients. This information can be used by healthcare providers to make informed decisions regarding the appropriateness of medical interventions for high-risk patients. Future clinical trials should further explore the response of different molecular subtypes to treatment in order to optimize the efficacy of breast cancer treatments.

Buckley-James boosting model based on extreme learning machine and random survival forests

Buckley-James (BJ) model is a typical semiparametric accelerated failure time model, which is closely related to the ordinary least squares method and easy to be constructed. However, traditional BJ model built on linearity assumption only captures simple linear relationships, while it has difficulty in processing nonlinear problems. To overcome this difficulty, in this paper, we develop a novel regression model for right-censored survival data within the learning framework of BJ model, basing on random survival forests (RSF), extreme learning machine (ELM), and L2 boosting algorithm. The proposed method, referred to as ELM-based BJ boosting model, employs RSF for covariates imputation first, then develops a new ensemble of ELMs-ELM-based boosting algorithm for regression by ensemble scheme of L2 boosting, and finally, uses the output function of the proposed ELM-based boosting model to replace the linear combination of covariates in BJ model. Due to fitting the logarithm of survival time with covariates by the nonparametric ELM-based boosting method instead of the least square method, the ELM-based BJ boosting model can capture both linear covariate effects and nonlinear covariate effects. In both simulation studies and real data applications, in terms of concordance index and integrated Brier sore, the proposed ELM-based BJ boosting model can outperform traditional BJ model, two kinds of BJ boosting models proposed by Wang et al., RSF, and Cox proportional hazards model.

Supervised two-dimensional functional principal component analysis with time-to-event outcomes and mammogram imaging data

Screening mammography aims to identify breast cancer early and secondarily measures breast density to classify women at higher or lower than average risk for future breast cancer in the general population. Despite the strong association of individual mammography features to breast cancer risk, the statistical literature on mammogram imaging data is limited. While functional principal component analysis (FPCA) has been studied in the literature for extracting image-based features, it is conducted independently of the time-to-event response variable. With the consideration of building a prognostic model for precision prevention, we present a set of flexible methods, supervised FPCA (sFPCA) and functional partial least squares (FPLS), to extract image-based features associated with the failure time while accommodating the added complication from right censoring. Throughout the article, we hope to demonstrate that one method is favored over the other under different clinical setups. The proposed methods are applied to the motivating data set from the Joanne Knight Breast Health cohort at Siteman Cancer Center. Our approaches not only obtain the best prediction performance compared to the benchmark model, but also reveal different risk patterns within the mammograms.

Elicited clinician knowledge did not improve dementia risk prediction in individuals with mild cognitive impairment

OBJECTIVES

This study aims to develop and validate a Bayesian risk prediction model that combines research cohort data with elicited expert knowledge to predict dementia progression in people with mild cognitive impairment (MCI).

STUDY DESIGN AND SETTING

This is a prognostic risk prediction modeling study based on cohort data (Alzheimer's disease neuroimaging initiative [ADNI]; n = 365) of research participants with MCI and elicited expert data. Bayesian Cox models were used to combine expert knowledge and ADNI data to predict dementia progression in people with MCI. Posterior distributions were obtained based on Gibbs sampler and the predictive performance was evaluated using ten-fold cross-validation via c-index, integrated calibration index (ICI), and integrated brier score (IBS).

RESULTS

365 people with MCI were included, mean age was 73 years (SD = 7.5), and 39% developed dementia within 3 years. When expert knowledge was incorporated, the c-index, ICI, and IBS values were 0.74 (95% CI 0.70-0.79), 0.06 (95% CI 0.05-0.08), and 0.17 (95% CI 0.14-0.19), respectively. These were similar to the model without expert knowledge data.

CONCLUSION

The addition of expert knowledge did not improve model accuracy in this ADNI sample to predict dementia progression in individuals with MCI.

Mitigating Membership Inference in Deep Survival Analyses with Differential Privacy

Deep neural networks have been increasingly integrated in healthcare applications to enable accurate predicative analyses. Sharing trained deep models not only facilitates knowledge integration in collaborative research efforts but also enables equitable access to computational intelligence. However, recent studies have shown that an adversary may leverage a shared model to learn the participation of a target individual in the training set. In this work, we investigate privacy-protecting model sharing for survival studies. Specifically, we pose three research questions. (1) Do deep survival models leak membership information? (2) How effective is differential privacy in defending against membership inference in deep survival analyses? (3) Are there other effects of differential privacy on deep survival analyses? Our study assesses the membership leakage in emerging deep survival models and develops differentially private training procedures to provide rigorous privacy protection. The experimental results show that deep survival models leak membership information and our approach effectively reduces membership inference risks. The results also show that differential privacy introduces a limited performance loss, and may improve the model robustness in the presence of noisy data, compared to non-private models.

Multivariate functional mixed model with MRI data: An application to Alzheimer's disease

Alzheimer's Disease (AD) is the leading cause of dementia and impairment in various domains. Recent AD studies, (ie, Alzheimer's Disease Neuroimaging Initiative (ADNI) study), collect multimodal data, including longitudinal neurological assessments and magnetic resonance imaging (MRI) data, to better study the disease progression. Adopting early interventions is essential to slow AD progression for subjects with mild cognitive impairment (MCI). It is of particular interest to develop an AD predictive model that leverages multimodal data and provides accurate personalized predictions. In this article, we propose a multivariate functional mixed model with MRI data (MFMM-MRI) that simultaneously models longitudinal neurological assessments, baseline MRI data, and the survival outcome (ie, dementia onset) for subjects with MCI at baseline. Two functional forms (the random-effects model and instantaneous model) linking the longitudinal and survival process are investigated. We use Markov Chain Monte Carlo (MCMC) method based on No-U-Turn Sampling (NUTS) algorithm to obtain posterior samples. We develop a dynamic prediction framework that provides accurate personalized predictions of longitudinal trajectories and survival probability. We apply MFMM-MRI to the ADNI study and identify significant associations among longitudinal outcomes, MRI data, and the risk of dementia onset. The instantaneous model with voxels from the whole brain has the best prediction performance among all candidate models. The simulation study supports the validity of the estimation and dynamic prediction method.

Impacts of removing race from the calculation of the kidney donor profile index

The kidney donor risk index (KDRI), standardized as the kidney donor profile index (KDPI), estimates graft failure risk for organ allocation and includes a coefficient for the Black donor race that could create disparities. This study used the Scientific Registry of Transplant Recipients data to recalculate KDRI coefficients with and without the Black race variable for deceased donor kidney transplants from 1995 to 2005 (n = 69 244). The recalculated coefficients were applied to deceased kidney donors from 2015 to 2021 (n = 72 926) to calculate KDPI. Removing the Black race variable had a negligible impact on the model's predictive ability. When the Black race variable was removed, the proportion of Black donors above KDPI 85%, a category with a higher risk of organ nonuse, declined from 31.09% to 17.75%, closer to the 15.68% above KDPI 85% among non-Black donors. KDPI represents percentiles relative to all other donors, so the number of Black donors moving below KDPI 86% was roughly equal to the number of non-Black donors moving above KDPI 85%. Removing the Black donor indicator from KDRI/KDPI may improve equity without substantial overall impact on the transplantation system, though further improvement may require the use of absolute measures of donor risk KDRI rather than relative measures of risk KDPI.

Comparison of linear and non-linear machine learning models for time-dependent readmission or mortality prediction among hospitalized heart failure patients

Although many models are available to predict prognosis of heart failure patients, most tools combining survival analysis are based on proportional hazard model. Non-linear machine learning algorithms would overcome the limitation of the time-independent hazard ratio assumption and provide more information in readmission or mortality prediction among heart failure patients. The present study collected the clinical information of 1796 hospitalized heart failure patients surviving during hospitalization in a Chinese clinical center from December 2016 to June 2019. A traditional multivariate Cox regression model and three machine learning survival models were developed in derivation cohort. Uno's concordance index and integrated Brier score in validation cohort were calculated to evaluate the discrimination and calibration of different models. Time-dependent AUC and Brier score curves were plotted to assess the performance of models at different time phases.

The Accuracy of Initial U.S. Heart Transplant Candidate Rankings

BACKGROUND

The U.S. heart allocation system ranks candidates with only 6 treatment-based categorical "statuses" and ignores many objective patient characteristics.

OBJECTIVES

This study sought to determine the effectiveness of the standard 6-status ranking system and several novel prediction models in identifying the most urgent heart transplant candidates.

METHODS

The primary outcome was death before receipt of a heart transplant. The accuracy of the 6-status system was evaluated using Harrell's C-index and log-rank tests of Kaplan-Meier estimated survival by status for candidates listed postpolicy (November 2018 to March 2020) in the Scientific Registry of Transplant Recipients data set. The authors then developed Cox proportional hazards models and random survival forest models using prepolicy data (2010-2017). The predictor variables included age, diagnosis, laboratory measurements, hemodynamics, and supportive treatment at the time of listing. The performance of these models was compared with the candidate's 6-status ranking in the postpolicy data.

RESULTS

Since policy implementation, the 6-status ranking at listing has had moderate ability to rank-order candidates (C-index: 0.67). Statuses 4 and 6 had no significant difference in survival (P = 0.80), and status 5 had lower survival than status 4 (P < 0.001). Novel multivariable prediction models derived with prepolicy data ranked candidates correctly more often than the 6-status rankings (Cox proportional hazards model C-index: 0.76; random survival forest model C-index: 0.74). Objective physiologic measurements, such as glomerular filtration rate, had high variable importance.

CONCLUSIONS

The treatment-based 6-status heart allocation system has only moderate ability to rank-order candidates by medical urgency. Predictive models that incorporate physiologic measurements can more effectively rank-order heart transplant candidates by urgency.

The combination of supervised and unsupervised learning based risk stratification and phenotyping in pulmonary arterial hypertension-a long-term retrospective multicenter trial

BACKGROUND

Accurate risk stratification in pulmonary arterial hypertension (PAH), a devastating cardiopulmonary disease, is essential to guide successful therapy. Machine learning may improve risk management and harness clinical variability in PAH.

METHODS

We conducted a long-term retrospective observational study (median follow-up: 67 months) including 183 PAH patients from three Austrian PAH expert centers. Clinical, cardiopulmonary function, laboratory, imaging, and hemodynamic parameters were assessed. Cox proportional hazard Elastic Net and partitioning around medoid clustering were applied to establish a multi-parameter PAH mortality risk signature and investigate PAH phenotypes.

RESULTS

Seven parameters identified by Elastic Net modeling, namely age, six-minute walking distance, red blood cell distribution width, cardiac index, pulmonary vascular resistance, N-terminal pro-brain natriuretic peptide and right atrial area, constituted a highly predictive mortality risk signature (training cohort: concordance index = 0.82 [95%CI: 0.75 - 0.89], test cohort: 0.77 [0.66 - 0.88]). The Elastic Net signature demonstrated superior prognostic accuracy as compared with five established risk scores. The signature factors defined two clusters of PAH patients with distinct risk profiles. The high-risk/poor prognosis cluster was characterized by advanced age at diagnosis, poor cardiac output, increased red cell distribution width, higher pulmonary vascular resistance, and a poor six-minute walking test performance.

CONCLUSION

Supervised and unsupervised learning algorithms such as Elastic Net regression and medoid clustering are powerful tools for automated mortality risk prediction and clinical phenotyping in PAH.