Survival Models Research Articles

Improving survival models in healthcare: a novel matching approach.

We present, to our knowledge, the first methodological study aimed at enhancing the prognosticpower of Cox regression models, widely used in survival analysis, through optimized data selection. Ourapproach employs a novel two-stage mechanism: by framing the prognostic stratum matching problemintuitively, we select prognostically representative patient observations to create a more balanced trainingset. This enables the model to assign equal attention to distinct prognostic subgroups. We demonstratethe methodology using an observational dataset of 1,799 patients with resected colorectal cancer livermetastases, 1,197 of whom received adjuvant chemotherapy and 602 who did not. In our study, as is current standard practice, the comparator was training prognostic models on the entire cohort (referred to as "model 1"). Models trained on the untreated and treated subgroups, matched through our approach (referred to as "model 3A and 3B", respectively), showed an improvement of up to 20% in bootstrapped C-indices compared tomodel 1. Notably, model 3 exhibited superior calibration, with a 6- to 10-fold improvement over model 1. Additional performance metrics aligned with these findings, and robustness was confirmed through biascorrectedbootstrapping. Given the ongoing development of numerous linear prognostic models and thegeneral applicability of our approach to any observational data, this method holds significant potentialto impact biomedical research and clinical practice where prognostic models are utilized.

Stunting and academic trajectory in urban settings of Burkina Faso.

Impaired growth in childhood can lead to poor cognitive development and low school performance. However, literature on the effects of stunting on school trajectory is very limited. The primary objective of this research was to estimate the age at which children start school according to levels of height-for-age z-score (stunting). A second objective was to estimate the gain in terms of age at school entry associated with an improvement in height-for-age z-score. A third objective was to explore the relationship between stunting, grade repetition, and school dropout. We used longitudinal data from the Ouagadougou (Burkina Faso) Health and Demographic Surveillance System. Data from a 2010 health survey of children under 5 years of age were merged with subsequent longitudinal schooling data. The study included 767 children globally who participated in the health and education surveys. Education data allowed us to apprehend academic trajectory measured by age at school entry, repetition, and dropout. The health survey gathered anthropometric information that was used to measure stunting. The adjusted age at school entry was estimated using a Poisson model. The gain represents the difference in adjusted age at school entry for different values of height-for-age. The relationship between stunting and grade repetition and dropout was studied using a discrete-time survival model. Results showed that children entered school on average at 5.7 years old, and the incidence of grade repetition and dropout was 17.7 and 6.6 per 100 persons-years, respectively. The adjusted age at school entry of severely stunted children was 6.2 years [95% confidence intervals (CI): 6.1; 6.3] compared to 5.1 years [95% CI: 5.0; 5.3] for children who had normal growth. The difference (gain) in adjusted age at school entry between severely and non-stunted children was thus 1.06 [95% CI: 0.87; 1.25] years. If a child's growth changed from severe stunting to normal growth, their risk of repeating a grade decreased by 5.0 [95% CI: 0.0; 9.0] per 100 persons-years. We did not observe a relationship between height-for-age and dropout. The results show that schooling is affected in several ways for children who are stunted. The age at school entry of stunted children is more likely to be delayed. Also, being stunted is associated with higher incidence of grade repetition. However, the relationship between stunting and dropout was inconclusive.

Camrelizumab plus rivoceranib compared sorafenib as first-line therapeutic options for advanced hepatocellular carcinoma in China: a cost-effectiveness analysis

AimThe objective of this research is to assess the cost-effectiveness of combining camrelizumab with rivoceranib in comparison to sorafenib as first-line therapeutic options for advanced hepatocellular carcinoma from the Chinese...

Leveraging Tumor Mutation Profiles to Forecast Immune Checkpoint Blockade Resistance in Melanoma, Lung, Head and Neck, Bladder and Renal Cancers

Immune checkpoint blockade (ICB), radiotherapy, chemotherapy and surgery are currently used as therapeutic strategies against melanoma, lung, bladder and renal cancers, but their efficacy is limited. Thus, I need to predict treatment response and resistance to address this challenge. In this study, I analyzed 350 lung cancer, 320 melanoma, 215 bladder cancer, 139 head and neck cancer and 151 renal carcinoma patients treated with ICB to identify tumor mutations associated with response and resistance to treatment. I identified several tumor mutations linked with a difference in survival outcomes following ICB. In lung cancer, missense mutations in ABL1, ASXL1, EPHA3, EPHA5, ERBB4, MET, MRE11A, MSH2, NOTCH1, PAK7, PAX5, PGR, ZFHX3, PIK3C3 and REL genes were indicative of favorable responses to ICB. Conversely, mutations in TGFBR2, ARID5B, CDKN2C, HIST1H3I, RICTOR, SMAD2, SMAD4 and TP53 genes were associated with shorter overall survival post-ICB treatment. In melanoma, mutations in FBXW7, CDK12, CREBBP, CTNNB1, NOTCH1 and RB1 genes predict resistance to ICB, whereas missense mutations in FAM46C and RHOA genes are associated with extended overall survival. In bladder cancer, mutations in HRAS genes predict resistance to ICB, whereas missense mutations in ERBB2, GNAS, ATM, CDKN2A and LATS1 genes, as well as nonsense mutations in NCOR1 and TP53 genes, are associated with extended overall survival. In head and neck cancer, mutations in genes like PIK3CA and KRAS correlated with longer survival, while mutations in genes like TERT and TP53 were linked to shorter survival. In renal carcinoma, mutations such as EPHA5, MGA, PIK3R1, PMS1, TSC1 and VHL were linked to prolonged overall survival, while others, including total splice mutations and mutations in B2M, BCOR, JUN, FH, IGF1R and MYCN genes were associated with shorter overall survival following ICB. Then, I developed predictive survival models by machine learning that correctly forecasted cancer patient survival following ICB within an error between 5 and 8 months based on their distinct tumor mutational attributes. In conclusion, this study advocates for personalized immunotherapy approaches in cancer patients.

Clinical predictive models for recurrence and survival in treated laryngeal and hypopharyngeal cancer: a systematic review and meta-analysis.

The limitations of the traditional TNM system have spurred interest in multivariable models for personalized prognostication in laryngeal and hypopharyngeal cancers (LSCC/HPSCC). However, the performance of these models depends on the quality of data and modelling methodology, affecting their potential for clinical adoption. This systematic review and meta-analysis (SR-MA) evaluated clinical predictive models (CPMs) for recurrence and survival in treated LSCC/HPSCC. We assessed models' characteristics and methodologies, as well as performance, risk of bias (RoB), and applicability. Literature searches were conducted in MEDLINE (OVID), Embase (OVID) and IEEE databases from January 2005 to November 2023. The search algorithm used comprehensive text word and index term combinations without language or publication type restrictions. Independent reviewers screened titles and abstracts using a predefined Population, Index, Comparator, Outcomes, Timing and Setting (PICOTS) framework. We included externally validated (EV) multivariable models, with at least one clinical predictor, that provided recurrence or survival predictions. The SR-MA followed PRISMA reporting guidelines, and PROBAST framework for RoB assessment. Model discrimination was assessed using C-index/AUC, and was presented for all models using forest plots. MA was only performed for models that were externally validated in two or more cohorts, using random-effects model. The main outcomes were model discrimination and calibration measures for survival (OS) and/or local recurrence (LR) prediction. All measures and assessments were preplanned prior to data collection. The SR-MA identified 11 models, reported in 16 studies. Seven models for OS showed good discrimination on development, with only one excelling (C-index >0.9), and three had weak or poor discrimination. Inclusion of a radiomics score as a model parameter achieved relatively better performance. Most models had poor generalisability, demonstrated by worse discrimination performance on EV, but they still outperformed the TNM system. Only two models met the criteria for MA, with pooled EV AUCs 0.73 (95% CI 0.71-0.76) and 0.67 (95% CI 0.6-0.74). RoB was high for all models, particularly in the analysis domain. This review highlighted the shortcomings of currently available models, while emphasizing the need for rigorous independent evaluations. Despite the proliferation of models, most exhibited methodological limitations and bias. Currently, no models can confidently be recommended for routine clinical use. https://www.crd.york.ac.uk/prospero/display_record.php?ID=CRD42021248762, identifier CRD42021248762.

Development and application of a predictive model for survival and drug therapy based on COVID-19-related lncRNAs in non-small cell lung cancer.

Numerous studies have substantiated the pivotal role of long non-coding RNAs (lncRNAs) in the progression of non-small cell lung cancer (NSCLC) and the prognosis of afflicted patients. Notably, individuals with NSCLC may exhibit heightened vulnerability to the novel coronavirus disease (COVID-19), resulting in a more unfavorable prognosis subsequent to infection. Nevertheless, the impact of COVID-19-related lncRNAs on NSCLC remains unexplored. The aim of our study was to develop an innovative model that leverages COVID-19-related lncRNAs to optimize the prognosis of NSCLC patients. Pertinent genes and patient data were procured from reputable databases, including TCGA, Finngen, and RGD. Through co-expression analysis, we identified lncRNAs associated with COVID-19. Subsequently, we employed univariate, LASSO, and multivariate COX regression techniques to construct a risk model based on these COVID-19-related lncRNAs. The validity of the risk model was assessed using KM analysis, PCA, and ROC. Furthermore, functional enrichment analysis was conducted to elucidate the functional pathways linked to the identified lncRNAs. Lastly, we performed TME analysis and predicted the drug sensitivity of the model. Based on risk scores, patients were categorized into high- and low-risk subgroups, revealing distinct clinicopathological factors, immune pathways, and chemotherapy sensitivity between the subgroups. Four COVID-19-related lncRNAs (AL161431.1, AC079949.1, AC123595.1, and AC108136.1) were identified as potential candidates for constructing prognostic prediction models for NSCLC. We also observed a positive correlation between risk score and MDSC, exclusion, and CAF. Additionally, two immune pathways associated with high-risk and low-risk subgroups were identified. Our findings further support the association between COVID-19 infection and neuroactive ligand-receptor interaction, as well as steroid metabolism in NSCLC. Moreover, we identified several highly sensitive chemotherapy drugs for NSCLC treatment. The developed model holds significant value in predicting the prognosis of NSCLC patients and guiding treatment decisions.

Effects of initial peritoneal dialysis prescription on clinical outcomes in Japanese peritoneal dialysis patients: a cohort study

Effects of the initial peritoneal dialysis (PD) prescription on clinical outcomes are unknown in Japan. We conducted a cohort study using data from Peritoneal Dialysis Outcomes and Practice Patterns Study. The patients were divided into two groups by the volume of the initial PD prescription (≤ 4 L/day or > 4 L/day). Cause-specific Cox proportional hazards survival models were used to model the association between different PD prescriptions and the clinical outcomes. The outcomes included transfer to HD, mortality, the composite of mortality and transfer to HD, peritonitis, hospitalization, and the patient-reported outcomes (PROs). Of the 342 patients, 98 were prescribed ≤ 4 L/day, and 244 were prescribed > 4 L/day. Patients prescribed ≤ 4 L/day were older with a lower percentage being male, had more cardiovascular and cerebrovascular disease but lower diabetes prevalence, were more likely to be receiving CAPD, used more assisted PD, and had lower BMI and mean serum creatinine levels. There were no significant differences between groups in terms of transfer to HD, mortality, transfer to HD or mortality, hospitalization, incidence of peritonitis, and PROs. Patients with initial PD prescriptions of ≤ 4 L/day compared to > 4 L/day had similar clinical outcomes. This practice may provide health economic benefits in Japan.

Cytokine profiles as predictors of HIV incidence using machine learning survival models and statistical interpretable techniques

HIV remains a critical global health issue, with an estimated 39.9 million people living with the virus worldwide by the end of 2023 (according to WHO). Although the epidemic’s impact varies significantly across regions, Africa remains the most affected. In the past decade, considerable efforts have focused on developing preventive measures, such as vaccines and pre-exposure prophylaxis, to combat sexually transmitted HIV. Recently, cytokine profiles have gained attention as potential predictors of HIV incidence due to their involvement in immune regulation and inflammation, presenting new opportunities to enhance preventative strategies. However, the high-dimensional, time-varying nature of cytokine data collected in clinical research, presents challenges for traditional statistical methods like the Cox proportional hazards (PH) model to effectively analyze survival data related to HIV. Machine learning (ML) survival models offer a robust alternative, especially for addressing the limitations of the PH model’s assumptions. In this study, we applied survival support vector machine (SSVM) and random survival forest (RSF) models using changes or means in cytokine levels as predictors to assess their association with HIV incidence, evaluate variable importance, measure predictive accuracy using the concordance index (C-index) and integrated Brier score (IBS) and interpret the model’s predictions using Shapley additive explanations (SHAP) values. Our results indicated that RSFs models outperformed SSVMs models, with the difference covariate model performing better than the mean covariate model. The highest C-index for SSVM was 0.7180 under the difference covariate model, while for RSF, it reached 0.8801 under the difference covariate model using the log-rank split rule. Key cytokines identified as positive predictors of HIV incidence included TNF-A, BASIC-FGF, IL-5, MCP-3, and EOTAXIN, while 29 cytokines were negative predictors. Baseline factors such as condom use frequency, treatment status, number of partners, and sexual activity also emerged as significant predictors. This study underscored the potential of cytokine profiles for predicting HIV incidence and highlighted the advantages of RSFs models in analyzing high-dimensional, time-varying data over SSVMs. It further through ablation studies emphasized the importance of selecting key features within mean and difference based covariate models to achieve an optimal balance between model complexity and predictive accuracy.

Intraoperative Tranexamic Acid in Radical Cystectomy: Impact on Bleeding, Thromboembolism, and Survival Outcomes.

Perioperative blood transfusion has been reported in > 50% of patients undergoing radical cystectomy (RC). Unfortunately, perioperative blood transfusion (PBT) in patients undergoing RC has been associated with poor oncological outcomes. Tranexamic acid (TXA) use has been proposed to decrease the need for PBT. Here, we seek to investigate the impact of intraoperative TXA on the risk of perioperative bleeding and venous thromboembolism (VTE) in patients undergoing RC. We also investigate its long-term impact on overall survival (OS) and cancer-specific survival (CSS) outcomes. We queried the prospectively maintained Mayo Clinic Radical Cystectomy registry and identified all RC performed for bladder cancer between 1990 and 2021. Primary outcomes assessed include the risk of perioperative bleeding, the need for blood transfusion, and the risk of VTE. Secondary outcomes include the impact of using TXA on OS and CSS. Of 2862 patients with complete available data, 468 received TXA (TXA recipient) and were matched 1:1 for age, neoadjuvant chemotherapy, pathologic staging, and preoperative hemoglobin with a group who did not receive TXA (TXA nonrecipient). TXA recipients experienced less estimated blood loss intraoperatively (median of 600 vs 650) cc and were less likely to need PBT (31% vs 50%, P value < .001) compared with TXA nonrecipients. There was no difference between groups in deep venous thrombosis and pulmonary embolism rates within 90 days of RC. In the adjusted survival model, use of TXA was not independently associated with significant impact on OS or CSS. However, perioperative blood transfusion was associated with poor OS and CSS (P value < .001). TXA use was associated with a significant reduction in estimated blood loss and PBT without increased risk of VTE. In univariable analyses, we observed an association between TXA use and improved OS as well as CSS. However, in multivariable analyses, TXA itself was not independently associated with improved OS or CSS; instead, PBT was. Further studies are warranted to explore strategies for minimizing PBTs and their impact on survival outcomes.

From seedlings to adults: Linking survival and leaf functional traits over ontogeny.

As long-lived tropical trees grow into the multi-layered canopy and face different environmental conditions, the relationships between leaf traits and whole-plant survival can vary over ontogeny. We tested the strength and direction of the relationships between leaf traits and long-term survival data across life stages for woody species from a subtropical forest in Puerto Rico. Trait-survival relationships were largely consistent across ontogeny with conservative traits leading to higher survival rates. The stage-specific relationship R2 increased by up to one order of magnitude compared to studies not considering ontogenetic trait variations. Stage-specific traits were significant predictors of their corresponding stage-specific survival: Seedlings traits were better predictors of seedling survival than adult traits, and adult traits were better predictors of maximum adult survival than seedling traits. Our results suggest that stage-specific leaf traits reflect different strategies over ontogeny and can substantially improve predictability of survival models in tropical forests.

Deep Multiple Instance Learning Model to Predict Outcome of Pancreatic Cancer Following Surgery.

Background/Objectives: Pancreatic ductal adenocarcinoma (PDAC) is a cancer with very poor prognosis despite early surgical management. To date, only clinical variables are used to predict outcome for decision-making about adjuvant therapy. We sought to generate a deep learning approach based on hematoxylin and eosin (H&E) or hematoxylin, eosin and saffron (HES) whole slides to predict patients' outcome, compare these new entities with known molecular subtypes and question their biological significance; Methods: We used as a training set a retrospective private cohort of 206 patients treated by surgery for PDAC cancer and a validation cohort of 166 non-metastatic patients from The Cancer Genome Atlas (TCGA) PDAC project. We estimated a multi-instance learning survival model to predict relapse in the training set and evaluated its performance in the validation set. RNAseq and exome data from the TCGA PDAC database were used to describe the transcriptomic and genomic features associated with deep learning classification; Results: Based on the estimation of an attention-based multi-instance learning survival model, we identified two groups of patients with a distinct prognosis. There was a significant difference in progression-free survival (PFS) between these two groups in the training set (hazard ratio HR = 0.72 [0.54;0.96]; p = 0.03) and in the validation set (HR = 0.63 [0.42;0.94]; p = 0.01). Transcriptomic and genomic features revealed that the poor prognosis group was associated with a squamous phenotype. Conclusions: Our study demonstrates that deep learning could be used to predict PDAC prognosis and offer assistance in better choosing adjuvant treatment.

Deep Survival Analysis for Alzheimer’s Disease Based on Longitudinal Neuroimaging Data

AbstractBackgroundAlzheimer’s Disease (AD) accounts for more than half of dementia cases and is a leading cause of death in the United States among elderly population. Early risk prediction is crucial to ensure effective and timely intervention plans. Deep learning‐based survival analysis based on cross‐sectional neuroimaging data has shown promising results in predicting the future progression of the dementia onset at an early stage. This study aims to derive deep survival analysis using multi‐timepoint longitudinal neuroimaging data to achieve a more accurate dementia onset prediction.MethodData: Longitudinal MRI of 1724 subjects (average age was 73.8) from the Alzheimer’s Disease Neuroimaging Initiative was used. To derive dementia onset event and duration information for the survival analysis, censored data (i.e. dementia conversion happened before or after the study period) was excluded (19.6% left‐censored, 63.2% right‐censored), resulting in 297 subjects. The input features included 89 FreeSurfer‐parcellated brain structural volumes, and demographic factors such as age and sex. Deep survival analysis was conducted using DeepSurv Pytorch library, using a multi‐layer perceptron (MLP) model architecture as the feature extractor. Image features from three timepoints were concatenated before feeding into the deep survival model for predicting future AD conversion. For comparison, ablation studies were conducted to include single timepoint neuroimaging features for each subject (derived from the last timepoints from the longitudinal features). The survival models’ performance was evaluated using the concordance index (C‐index) using 10‐fold cross‐validation.ResultThe accuracy of the three‐timepoint longitudinal data (C‐index=97.1%) is higher than the single timepoint study (C‐index=75.6%) (Figure 1). Figure 2 shows the histogram of the difference between the actual and predicted dementia onset time (derived by taking the 0.5 threshold cutoff point from the C‐index, Positive value means delayed prediction). Single timepoint prediction showed more false negatives (delayed prediction) while longitudinal prediction showed relatively even distribution of false positive (earlier prediction) and false negative.ConclusionIn this study, we demonstrated the of longitudinal neuroimaging data in deep survival analysis to improve the prediction the time of conversion for future dementia, with less false negative, promoting early prediction and potential early intervention prior to the dementia.

An Evaluation of an Algorithm for the Selection of Flexible Survival Models for Cancer Immunotherapies: Pass or Fail?

Accurately extrapolating survival beyond trial follow-up is essential in a health technology assessment where model choice often substantially impacts estimates of clinical and cost effectiveness. Evidence suggests standard parametric models often provide poor fits to long-term data from immuno-oncology trials. Palmer et al. developed an algorithm to aid the selection of more flexible survival models for these interventions. We assess the usability of the algorithm, identify areas for improvement and evaluate whether it effectively identifies models capable of accurate extrapolation. We applied the Palmer algorithm to the CheckMate-649 trial, which investigated nivolumab plus chemotherapy versus chemotherapy alone in patients with gastroesophageal adenocarcinoma. We evaluated the algorithm's performance by comparing survival estimates from identified models using the 12-month data cut to survival observed in the 48-month data cut. The Palmer algorithm offers a systematic procedure for model selection, encouraging detailed analyses and ensuring that crucial stages in the selection process are not overlooked. In our study, a range of models were identified as potentially appropriate for extrapolating survival, but only flexible parametric non-mixture cure models provided extrapolations that were plausible and accurately predicted subsequently observed survival. The algorithm could be improved with minor additions around the specification of hazard plots and setting out plausibility criteria. The Palmer algorithm provides a systematic framework for identifying suitable survival models, and for defining plausibility criteria for extrapolation validity. Using the algorithm ensures that model selection is based on explicit justification and evidence, which could reduce discordance in health technology appraisals.

Real-World and Clinical Trial Validation of a Deep Learning Radiomic Biomarker for PD-(L)1 Immune Checkpoint Inhibitor Response in Advanced Non-Small Cell Lung Cancer.

This study developed and validated a novel deep learning radiomic biomarker to estimate response to immune checkpoint inhibitor (ICI) therapy in advanced non-small cell lung cancer (NSCLC) using real-world data (RWD) and clinical trial data. Retrospective RWD of 1,829 patients with advanced NSCLC treated with PD-(L)1 ICIs were collected from 10 academic and community institutions in the United States and Europe. The RWD included data sets for discovery (Data Set A-Discovery, n = 1,173) and independent test (Data Set B, n = 458). A radiomic pipeline, containing a deep learning feature extractor and a survival model, generated the computed tomography (CT) response score (CTRS) applied to the pretreatment routine CT/positron emission tomography (PET)-CT scan. An enhanced CTRS (eCTRS) also incorporated age, sex, treatment line, and lesion annotations. Performance was evaluated against progression-free survival (PFS) and overall survival (OS). Biomarker generalizability was further evaluated using a secondary analysis of a prospective clinical trial (ClinicalTrials.gov identifier: NCT02573259) evaluating the PD-1 inhibitor sasanlimab in second or later line of treatment (Data Set C, n = 54). In RWD Test Data Set B, the CTRS identified patients with a high probability of response to ICI with a PFS hazard ratio (HR) of 0.46 (95% CI, 0.26 to 0.82) and an OS HR of 0.50 (95% CI, 0.28 to 0.92) in the first-line ICI monotherapy cohort, after adjustment for baseline covariates including the PD-L1 tumor proportion score. In Clinical Trial Data Set C, the CTRS demonstrated an adjusted PFS HR of 1.03 (95% CI, 0.43 to 2.47) and an OS HR of 0.33 (95% CI, 0.14 to 0.91). The CTRS and eCTRS outperformed traditional imaging biomarkers of lesion size in PFS and OS for RWD Test Data Set B and in OS for the Clinical Trial Data Set. The study developed and validated a deep learning radiomic biomarker using pretreatment routine CT/PET-CT scans to identify ICI benefit in advanced NSCLC.

Supervised clustering of MRI measures informed by incident cognitive impairment via aggregated survival model SHAP values

AbstractBackgroundAlzheimer’s disease (AD) is a complex heterogeneous neurodegenerative disease. Unsupervised clustering techniques have been used to identify disease subtypes, but such approaches are limited since subtypes may not directly be related to disease progression. Herein, we implement a novel supervised clustering approach that aims to identify MRI‐derived subtypes that are likely to experience incident cognitive impairment (ICI).MethodsWe used data from 822 cognitively normal individuals (68.4 ± 9.5 years old) with Clinical Dementia Rating® (CDR®) of 0 at baseline, from the Knight ADRC. We performed supervised clustering of eight volumetric MRI regional measures at baseline. MRI measures were adjusted for age, sex, and education. Mean aggregated time‐dependent Shapley Additive exPlanation (SHAP) values were derived from random survival forest models with an ICI outcome based on time‐to‐conversion from a CDR=0 to CDR&gt;0 over 10 years of follow‐up. K‐means clustering was then applied to SHAP values. Analysis of Variance was used to assess cluster differences and linear regression was used to understand longitudinal trajectories of MRI volumetrics and CSF biomarkers.ResultsIn the entire sample, hippocampus, entorhinal, and amygdala SHAP values were determined to be the most important MRI measures in terms of estimated contribution to the model prediction with the hippocampus differentiating from other regional measure about 24 months from baseline (Figure 1). The optimal number of clusters was determined to be four. Cluster characteristics and significant differences among them are presented in Table 1 and Figure 1. Cluster 4 is characterized by the largest mean hippocampus SHAP value with approximately 89% of participants experiencing ICI. Figure 2 shows longitudinal linear trajectories for MRI and cerebrospinal measures. Cluster 1 was the healthiest across all longitudinal biomarkers. Whereas, Cluster 4 had the smallest hippocampal volume and abnormal CSF measures both at baseline and over time.ConclusionsWe implemented a novel application of SHAP values from survival models in a supervised clustering approach to identifying MRI‐subtypes with different probabilities of incident cognitive impairment. This approach could be used to elucidate the heterogeneity inherent in the relationships among AD biomarkers and longitudinal cognitive changes.

Comparison of Different Machine Learning Models for Predicting Long-Term Overall Survival in Non-metastatic Colorectal Cancers.

In recent years, machine learning (ML) methods have gained significant popularity among medical researchers interested in cancer. We aimed to test different (ML) models to predict both overall survival and survival at specific time points in patients with non-metastatic colorectal cancer (CRC). The clinicopathological and treatment data of non-metastatic CRC patients with more than 10 years of follow-up at a single center were retrospectively reviewed. 1, 2, 3, 5, and 10-year survival rates for all patients and stages I-III were statistically calculated using the Kaplan-Meier method. Five distinct machine-learning algorithms were employed to develop predictive models for patient survival at five designated time points. A total of 498 patients were included in the study. The decision tree model had the highest area under the curve (AUC) for 1-year survival prediction (0.89). The ensemble model had the highest AUC for predicting 2-year, 3-year, and 5-year survival predictions (0.86, 0.92, and 0.89, respectively), while the support vector machine model had the highest AUC (0.84) for predicting 10-year survival. When considering the stages separately and assessing survival for the designated time intervals, the accuracy of all five models was found to be similar, ranging around 70% or higher. ML models can predict short- and long-term survival in patients with CRC, both for the overall patient population and when stratified by stage.

Local Recurrence and Survival in Patients With Melanoma >2 mm in Thickness at Difficult Sites Treated With 1-cm Versus 2-cm Margins.

Melanoma guidelines recommend surgical excision with 2-cm margins for melanomas >2 mm in thickness. However, this procedure may be problematic at critical anatomic sites. We aimed to compare the outcomes of wide (2 cm) versus narrow (1 cm) excision margins in patients with melanoma >2 mm in thickness near critical structures. We retrospectively examined 736 patients undergoing excision with wide versus narrow margins at the National Cancer Institute in Milan, Italy, between 2001 and 2015. A total of 265 (36.0%) patients received a wide local excision-82 (30.9%) with linear repair and 183 (69.1%) with flap or graft reconstruction. A total of 471 (64.0%) patients received a narrow excision-320 (67.9%) with linear repair and 151 (32.1%) with flap or graft reconstruction (P<.001). The 10-year overall survival rate was 69.5% (95% CI, 63.3%-76.2%) in the wide group and 68.7% (95% CI, 63.8%-74.0%) in the narrow group (P=.462); 10-year crude cumulative incidence (CCI) of local recurrence was 5.4% (95% CI, 3.2%-9.2%) in the wide and 8.8% (95% CI, 6.4%-12.1%) in the narrow group (P=.150). Multivariable Fine-Gray modeling of the CCI of local recurrence showed that Breslow thickness (P=.010) was the only statistically significant parameter. Multivariable Cox models for overall survival showed that age (P<.001), Breslow thickness (P<.001), and sentinel lymph node status (P=.019) were statistically significant covariates. Excision margin was not a significant parameter affecting patients' outcome. Wide local excision with 1-cm margins for melanoma >2 mm in thickness was not associated with an increased risk of local recurrence and did not affect overall survival.

Development, Validation, and Clinical Utility of Electronic Patient-Reported Outcome Measure-Enhanced Prediction Models for Overall Survival in Patients With Advanced Non-Small Cell Lung Cancer Receiving Immunotherapy.

Electronic patient-reported outcome measures (ePROMs) are increasingly collected routinely in clinical practice and may be prognostic for survival in adults with advanced non-small cell lung cancer (NSCLC) in addition to clinical data. This study developed ePROM-enhanced models for predicting 1-year overall survival in patients with advanced NSCLC at the start of immunotherapy. This is a single-center study using consecutive patients from a tertiary cancer hospital in England. Using Cox proportional hazards models, we developed one clinical factor-only model and three ePROM-enhanced models, each including one of the following factors: quality of life (as measured by EuroQoL five-dimension five-level utility score) and overall symptom burden and number of moderate-to-severe symptoms (as measured by patient-reported version of Common Terminology Criteria for Adverse Events). Predictive performance was evaluated and compared through bootstrapping internal validation, and clinical utility was determined via decision curve analysis. The clinical factor-only model contained age, histology, performance status, and neutrophile-to-lymphocyte ratio. While calibration was similar between the clinical factor-only and ePROM-enhanced models, the latter showed improved discrimination by 0.020 (95% CI, 0.011 to 0.024), 0.024 (95% CI, 0.016 to 0.031), and 0.024 (95% CI, 0.014 to 0.029) when enhanced with ePROMs on quality of life, overall symptom burden, and number of moderate-to-severe symptoms, respectively. If care decisions are to be made at risk thresholds between 25% and 75%, the ePROM-enhanced models led to higher net benefit than the clinical factor-only model and the default strategies of intervention for all and intervention for none. The ePROM-enhanced models outperformed the clinical factor-only model in predicting 1-year overall survival for patients with advanced NSCLC receiving immunotherapy and showed potential clinical utility for informing decisions in this population. Future studies should focus on validating the models in external data sets.

Elevated CA19-9 within the normal range suggests poorer prognosis in stage II CRC: A retrospective analysis of a large sample in a single center.

Carbohydrate antigen 19-9 (CA19-9) and carcinoembryonic antigen (CEA) serve as pivotal tumor markers in colorectal cancer (CRC). However, uncertainty persists regarding the prognostic significance of the two tumor markers when falling within the normal range. We attempt to compare the prognostic differences of tumor markers at different levels within the reference range. This retrospective study scrutinized 2,167 cases of stage II CRC verified by pathology after surgery at the Fudan University Shanghai Cancer Center. Using R software to calculate the optimal critical value to compare the clinical and pathological characteristics and prognosis of different levels of tumor markers. The survival and regression modeling strategies packages of R software drew the nomograms. Utilizing R software, the optimal critical value of CA19-9 was determined as 12.12 U/mL and that of CEA as 1.89 U/mL. Kaplan-Meier survival analysis unveiled that, within the normal range, higher levels of CEA were linked to poorer overall survival (OS) [HR = 1.829 (1.280, 2.989), P = 0.0033] and disease-free survival (DFS) [HR = 1.472 (1.114, 1.944), P = 0.0444]. Similarly, heightened levels of CA19-9 also indicated inferior OS [HR = 1.750 (1.203, 2.455), P = 0.0076] and DFS [HR = 1.361 (1.098, 1.686), P = 0.0049]. Furthermore, multivariate analysis identified CA19-9 as an independent risk factor for OS (HR = 1.49,95% CI: 1.086-2.045, P = 0.014) and DFS (HR = 1.327,95% CI: 1.070-1.647, P = 0.01), while the impact of CEA on OS and DFS was not statistically significant. A nomogram constructed based on the Cox regression model can effectively evaluate the prognosis of CRC patients. Although within the normal range, elevated CA19-9 was associated with an inferior prognosis, chemotherapy decisions of different intensities can be adjusted based on nomograms. This work will contribute to standardizing the diagnosis and treatment of stage II CRC and provide clinicians with essential insights for chemotherapy decisions.

Utility of an Echocardiographic Machine Learning Model to Predict Outcomes in Intensive Cardiac Care Unit Patients.

The risk stratification at admission to the intensive cardiac care unit (ICCU) is crucial and remains challenging. We aimed to investigate the accuracy of a machine learning (ML)-model based on initial transthoracic echocardiography (TTE) to predict in-hospital major adverse events (MAEs) in a broad spectrum of patients admitted to ICCU. All consecutive patients hospitalized in ICCUs with a complete TTE performed within the first 24hours of admission were included in this prospective multicenter study (39 centers). Sixteen TTE parameters were evaluated. The ML model involved automated feature selection by random survival forest and model building with an extreme gradient boosting (XGBoost) algorithm. The primary outcome was in-hospital MAEs defined as all-cause death, resuscitated cardiac arrest, or cardiogenic shock. Of 1,499 consecutive patients (63±15years, 70% male), MAEs occurred in 67 patients (4.5%). The 5 TTE parameters selected in the model were left ventricular outflow tract velocity-time integral, E/e' ratio, systolic pulmonary artery pressure, tricuspid annular plane systolic excursion, and left ventricular ejection fraction. Using the XGBoost, the ML model exhibited a higher area under the receiver operating curve compared with any existing scores (ML model, 0.83 vs logistic regression, 0.76, ACUTE-HF score:,0.66; thrombolysis in myocardial infarction score, 0.60; Global Registry of Acute Coronary Events score, 0.58, all P<.001). The ML model had an incremental prognostic value for predicting MAE over a traditional model including clinical and biological data (C index 0.80 vs 0.73, P=.012; chi-square 59.7 vs 32.4; P<.001). The ML model based on initial TTE exhibited a higher prognostic value to predict in-hospital MAEs compared with existing scores or clinical and biological data in the ICCU.