Traditional Cox Proportional Hazards Model Research Articles

Estimating cardiovascular mortality in patients with hypertension using machine learning: The role of depression classification based on lifestyle and physical activity.

This study aims to harness machine learning techniques, particularly the Random Survival Forest (RSF) model, to assess the impact of depression on cardiovascular disease (CVD) mortality among hypertensive patients. A key objective is to elucidate the interplay between mental health, lifestyle, and physical activity while comparing the effectiveness of the RSF model against the traditional Cox proportional hazards model in predicting CVD mortality. Data from the National Health and Nutrition Examination Survey (NHANES) spanning 2007 to 2014 were used for comprehensive depression screening. The Patient Health Questionnaire-9 (PHQ-9) was employed to categorize depression severity levels among participants. The final cohort included 9271 participants, selected after excluding those with incomplete data. Participants were followed up for a median of 7.1years, and cardiovascular mortality was assessed up to December 31, 2019. We employed the RSF model to predict cardiovascular mortality with high effectiveness and precision. And to ensure comparability, we developed the traditional Cox proportional hazards model using the same set of predictors. RSF model outperformed the Cox proportional hazards model in predicting cardiovascular mortality among hypertensive patients with varying depression levels. The RSF model's integrated area under the curve (iAUC) scores were 0.842, 0.893, and 0.760 for none, mild, and severe depression, respectively, surpassing the Cox model's scores of 0.826, 0.805, and 0.746. The RSF model provides a more accurate prediction of CVD mortality among hypertensive patients with varying degrees of depression, offering a valuable tool for personalized patient care. Its ability to stratify patients into risk categories can assist healthcare professionals in making informed decisions, underscoring the potential of machine learning in public health and clinical settings. This model demonstrates particular utility in settings where detailed, patient-specific risk assessments are critical for managing long-term health outcomes. Future research should focus on external validation and integration of more diverse variables to enhance predictive power.

Prognostic importance of the Scottish inflammatory prognostic score in patients with hepatocellular carcinoma after hepatectomy: a retrospective cohort study

BackgroundThe Scottish Inflammatory Prognostic Score (SIPS), an innovative scoring system, has emerged as a promising biomarker for predicting patient outcomes following cancer therapy. This study aimed to evaluate the value of SIPS as a prognostic indicator following hepatectomy in patients with hepatocellular carcinoma (HCC).MethodsThis retrospective study included 693 HCC patients who underwent hepatectomy. Survival outcomes were compared between propensity score-matched groups. Independent prognostic factors were identified through Cox regression analysis. Additionally, both traditional Cox proportional hazards models and machine learning models based on the SIPS were developed and validated.ResultsA total of 693 HCC patients who underwent hepatectomy were included, with 102 in the high SIPS group and 591 in the low SIPS group. Following propensity score matching (1:3 ratio), both groups achieved balance, with 82 patients in the high SIPS group and 240 patients in the low SIPS group. The low SIPS group demonstrated significantly superior recurrence-free survival (RFS) (25 months vs. 21 months; P < 0.001) and overall survival (OS) (69 months vs. 58 months; P < 0.001) compared to the high SIPS group. Multivariable analysis identified SIPS as an independent adverse factor affecting both RFS and OS. The calibration curve for overall patient survival diagnosis displayed excellent predictive accuracy. Traditional COX prognostic models and machine learning models incorporating SIPS demonstrated excellent performance both the training and validation set.ConclusionThis study confirms the prognostic significance of SIPS in post-hepatectomy HCC patients, providing a practical tool for risk stratification and clinical decision-making. Further research and validation are needed to consolidate its role in prognostic assessment.

Machine learning-based individualized survival prediction model for prognosis in osteosarcoma: Data from the SEER database.

Patient outcomes of osteosarcoma vary because of tumor heterogeneity and treatment strategies. This study aimed to compare the performance of multiple machine learning (ML) models with the traditional Cox proportional hazards (CoxPH) model in predicting prognosis and explored the potential of ML models in clinical decision-making. From 2000 to 2018, 1243 patients with osteosarcoma were collected from the Surveillance, Epidemiology, and End Results (SEER) database. Three ML methods were chosen for model development (DeepSurv, neural multi-task logistic regression [NMTLR]) and random survival forest [RSF]) and compared them with the traditional CoxPH model and TNM staging systems. 871 samples were used for model training, and the rest were used for model validation. The models' overall performance and predictive accuracy for 3- and 5-year survival were assessed by several metrics, including the concordance index (C-index), the Integrated Brier Score (IBS), receiver operating characteristic curves (ROC), area under the ROC curves (AUC), calibration curves, and decision curve analysis. The efficacy of personalized recommendations by ML models was evaluated by the survival curves. The performance was highest in the DeepSurv model (C-index, 0.77; IBS, 0.14; 3-year AUC, 0.80; 5-year AUC, 0.78) compared with other methods (C-index, 0.73-0.74; IBS, 0.16-0.17; 3-year AUC, 0.73-0.78; 5-year AUC, 0.72-0.78). There are also significant differences in survival outcomes between patients who align with the treatment option recommended by the DeepSurv model and those who do not (hazard ratio, 1.88; P < .05). The DeepSurv model is available in an approachable web app format at https://survivalofosteosarcoma.streamlit.app/. We developed ML models capable of accurately predicting the survival of osteosarcoma, which can provide useful information for decision-making regarding the appropriate treatment.

Generalized Propensity Score Methods to Assess CKD-Associated Physiologic Factors and Risk of Kidney Failure in the Chronic Renal Insufficiency Cohort (CRIC) Study.

Epidemiologic studies have identified many biochemical risk factors for chronic kidney disease (CKD) progression that are correlates of kidney function, termed here 'CKD-associated physiologic factors'. Uncertainty remains if these factors are risk factors or risk markers accounting for aspects of kidney function not otherwise captured. We aimed to use flexible machine learning, a dynamic covariate history including kidney function informative markers, and generalized propensity score (GPS) weighting, to better control confounding for such exposures. We studied 3,052 adults with CKD in the Chronic Renal Insufficiency Cohort Study. We established a 2-year run-in period and assembled 90 variables that characterize variability and trends of selected CKD-associated physiologic factors and confounders. Using SuperLearner, we created a GPS for each CKD-associated physiologic factor and performed GPS-weighted Cox regressions. For context, we also evaluated results from traditional multivariable Cox proportional hazards models as in prior studies. Similar to traditional approaches, bicarbonate, calcium, potassium, hemoglobin, and PTH were each associated with risk of kidney failure using GPS weighting. The GPS approach detected non-linear associations in many factors, some of which were not detected with traditional models. We conclude that many associations between CKD-associated physiologic factors and kidney outcomes remain strong after GPS weighting.

Development and validation of accelerated failure time model for cause-specific survival and prognostication of oral squamous cell carcinoma: SEER data analysis.

Oral Squamous Cell Carcinoma is the most prevalent malignancies affecting the oral cavity. Despite progress in studies and treatment options its outlook remains grim with survival prospects greatly affected by demographic and clinical factors. Precisely predicting survival rates and prognosis plays a role in making treatment choices for the best achievable overall health outcomes. To develop and validate an accelerated failure time model as a predictive model for cause-specific survival and prognosis of Oral Squamous Cell Carcinoma patients and compare its results to the traditional Cox proportional hazard model. We screened Oral cancer patients diagnosed with Squamous Cell Carcinoma from the Surveillance Epidemiology and End Results (SEER) database between 2010 and 2020. An accelerated failure time model using the Type I generalized half logistic distribution was used to determine independent prognostic factors affecting the survival time of patients with oral squamous carcinoma. In addition, accelerated factors were estimated to assess how some variables influence the survival times of the patients. We used the Akaike Information Criterion, Bayesian Information Criterion to evaluate the model fit, the area under the curve for discriminability, Concordance Index (C-index) and Root Mean Square Error and calibration curve for predictability, to compare the type I generalized half logistic survival model to other common classical survival models. All tests are conducted at a 0.05 level of significance. The accelerated failure time models demonstrated superior effectiveness in modeling (fit and predictive accuracy) the cause-specific survival (CSS) of oral squamous cell carcinoma compared to the Cox model. Among the accelerated failure time models considered, the Type I generalized half logistic distribution exhibited the most robust model fit, as evidenced by the lowest Akaike Information Criterion (AIC = 27370) and Bayesian Information Criterion (BIC = 27415) values. This outperformed other parametric models and the Cox Model (AIC = 47019, BIC = 47177). The TIGHLD displayed an AUC of 0.642 for discrimination, surpassing the Cox model (AUC = 0.544). In terms of predictive accuracy, the model achieved the highest concordance index (C-index = 0.780) and the lowest root mean square error (RMSE = 1.209), a notable performance over the Cox model (C-index = 0.336, RMSE = 6.482). All variables under consideration in this study demonstrated significance at the 0.05 level for CSS, except for race and the time span from diagnosis to treatment, in the TIGHLD AFT model. However, differences emerged regarding the significant variations in survival times among subgroups. Finally, the results derived from the model revealed that all significant variables except chemotherapy, all TNM stages and patients with Grade II and III tumor presentations contributed to the deceleration of time to cause-specific deaths. The accelerated failure time model provides a relatively accurate method to predict the prognosis of oral squamous cell carcinoma patients and is recommended over the Cox PH model for its superior predictive capabilities. This study also underscores the importance of using advanced statistical models to improve survival predictions and outcomes for cancer patients.

Assessment of prognostic factors in patients with primary ocular adnexal lymphoma when considering competing risk elements.

Accurate prognostic factors for primary ocular adnexal lymphoma (POAL) are scarce. Survival models and prognostic factors derived without considering competing risk factors suffer from major statistical errors. This study aimed to accurately assess prognostic factors in POAL patients using competing risk models, and compare this to the traditional COX proportional hazards model. This retrospective study utilised data from the Surveillance, Epidemiology, and End Results (SEER) program 2010-2015 and included patients with B-cell POAL. The cumulative incidence function and Gray's test for cause-specific survival were calculated as univariate analysis. The competing risk models were a Fine-Gray subdistribution hazard model and a cause-specific model, and a traditional COX model was employed as a multivariate analysis. This study enrolled 846 eligible patients with POAL: 60 patients (7.09%) died from POAL and 123 patients (14.54%) died from other causes. Multivariate competing risk models indicated that age, laterality, histology subtype, the 7th edition of American Joint Committee on Cancer stage T, and radiotherapy were independent predictors for cause-specific survival of patients with POAL. There was high consistency between the two competing risk models. The COX model made several misestimations on the statistical significance and hazard ratios of prognostic factors. This study established competing risk models as a method to assess POAL prognostic factors, which was more accurate than traditional methods that do not consider competing risk elements.

The development of a prediction model based on random survival forest for the prognosis of non- Hodgkin lymphoma: A prospective cohort study in China

Background and objectiveThe pathological staging of non-Hodgkin lymphoma (NHL) is complex, the clinical manifestations are varied, and the prognosis differ considerably. To provide a useful reference for early detection and effective treatment of NHL, we developed a random survival forest (RSF) prognostic model based on machine learning (ML) algorithms using prospective cohort data collected from Chongqing Cancer Hospital from Jan 1, 2017 to Dec 31, 2019 (n = 1449) to compare with the traditional cornerstone method Cox proportional hazards (CPH) model and evaluate the predictability of the model. MethodsPatients were randomly split into a training cohort (TC) and validation cohort (VC) based on 65/35 ratio. The least absolute shrinkage and selection operator (LASSO) regression analysis was used to extracted the important features. And the RSF was modeled to explore the prognostic factors impacting the overall survival (OS) of patients with NHLs in the TC and validated in the VC. The C-index, the Integrated Brier Score (IBS), Kaplan-Meir method, the receiver operating characteristic (ROC) curve, and the area under the ROC curve (AUC) were selected to measure performances and discriminations of the models. In addition, individual survival probability predicted for NHL patients. ResultsAccording to the features extracted by LASSO model and univariable Cox model, 16 variables were selected to develop the RSF model with log-rank splitting rule, which were age, ethnicity, medical insurance, Ann Arbor stage, pathology, targeted-therapy, chemo-therapy, peripheral blood neutrophil count to lymphocyte count ratio (NLR), peripheral blood platelet count to lymphocyte count ratio (PLR), serum lactate dehydrogenase (LDH), CD4/CD8, platelet (PLT), absolute neutrophil count (ANC), lymphocyte (LYM), B-symptoms, and (CPR) were important prognostic factors. Compared to the CPH model (C-index = 0.748, IBS = 0.166), the RSF model (C-index = 0.786, IBS = 0.165) is outperformed in predictability and accuracy. The AUC of the RSF model to estimate the 1-, 3-, and 5-year OS in TC were 0.847, 0.847, and 0.809, respectively; while those in the CPH were 0.816, 0.803, and 0.750, respectively. ConclusionsTo provide practical implications for the implementation of individualized therapy, the study constructed a high-performed RSF model and reveal that it outperformed the traditional model CPH. And the RSF model ranked the risk variables. In addition, we stratified the risk of NHL patients and estimated individual survival probability based on the RSF model.

Analysis of survival-related factors in patients with endometrial cancer using a Bayesian network model.

In recent years, remarkable progress has been made in the use of machine learning, especially in analyzing prognosis survival data. Traditional prediction models cannot identify interrelationships between factors, and the predictive accuracy is lower. This study aimed to construct Bayesian network models using the tree augmented naïve algorithm in comparison with the Cox proportional hazards model. A Bayesian network model and a Cox proportional hazards model were constructed to analyze the prognostic factors of endometrial cancer. In total, 618 original cases obtained from the Surveillance, Epidemiology, and End Results database were used to construct the Bayesian network model, which was compared with the traditional Cox proportional hazards model by analyzing prognostic factors. External validation was performed using a dataset from The First Affiliated Hospital of Shandong First Medical University. The predictive accuracy, area under the receiver operating characteristic curve, and concordance index for the Bayesian network model were 74.68%, 0.787, and 0.72, respectively, compared to 68.83%, 0.723, and 0.71, respectively, for the Cox proportional hazards model. Tumor size was the most important factor for predicting survival, followed by lymph node metastasis, distant metastasis, chemotherapy, lymph node resection, tumor stage, depth of invasion, tumor grade, histological type, age, primary tumor site, radiotherapy and surgical sequence, and radiotherapy. The findings indicate that the Bayesian network model is preferable to the Cox proportional hazards model for predicting survival in patients with endometrial cancer.

Predictive Efficacy of the Advanced Lung Cancer Inflammation Index in Hepatocellular Carcinoma After Hepatectomy.

Hepatocellular carcinoma (HCC) presents a significant global health challenge due to its poor prognosis and high recurrence rates post-surgery. This study examines the predictive efficacy of the Advanced Lung Cancer Inflammation Index (ALI) in assessing the post-hepatectomy prognosis of patients with HCC. A cohort comprising 1654 HCC patients who underwent hepatectomy at Guangxi Medical University Cancer Hospital from 2013 to 2019 was enrolled. Patients were stratified into two groups according to the median ALI level, and then subjected to propensity score matching (PSM) in a 1:1 ratio. Kaplan-Meier survival curves, the traditional Cox proportional hazards (CPH) model, and machine learning (ML) models were employed to analyze and evaluate ALI's prognostic significance. Furthermore, ALI's prognostic value in digestive system tumors was validated via analysis of the National Health and Nutrition Examination Survey (NHANES) database. After applying PSM, a final cohort of 1284 patients, categorized into high and low ALI groups, revealed a significantly reduced survival time in the low ALI cohort. Univariate and multivariate Cox analyses identified ALI, BCLC stage, CK19, Hepatitis B virus (HBV) DNA, lymph node metastasis, and microvascular invasion (MVI) as independent predictors of prognosis. Both traditional CPH and ML models incorporating ALI demonstrated excellent predictive accuracy, validated through calibration curves, time-dependent ROC curves, and decision curve analysis. Furthermore, the prognostic value of ALI in digestive tumors was confirmed in the NHANES database. The ALI exhibits potential as a prognostic predictor in patients with HCC following hepatectomy, providing valuable insights into postoperative survival.

Temporal effect of imatinib adherence on time to remission in chronic myeloid leukemia patients.

Adherence to imatinib in chronic myeloid leukemia (CML) patients is estimated to be as low as 70% despite its clinical benefit, and our understanding of the impact of nonadherence in this population is limited. This study presents a novel application of the Alternating Conditional Estimation (ACE) algorithm in newly diagnosed CML patients to map the full dose-response curve (DRC) and determine how the strength of this curve varies over time. We applied the ACE algorithm alongside a backward elimination procedure to detect the presence of time dependence and nonlinearity in the relationship between imatinib adherence and time-to-remission. An extended Cox model allowing for the flexible modeling of identified effects via unpenalized B-splines was subsequently fit and assessed. The substantial improvement in model fit associated with the extended Cox approach suggests that traditional Cox proportional hazards model assumptions do not hold in this setting. Results indicate that the DRC for imatinib is non-linearly increasing, with an attenuated effect above a 74% adherence rate. The strength of this effect on remission varied over time and was strongest in the initial months of treatment, reaching a peak around 90 days post-initiation (log hazard ratio: 2.12, 95% confidence interval: 1.47 to 2.66). Most patients that achieved remission did so by 4 months (120 days) with consistently high adherence, suggesting that this could be a critical time and duration for realizing treatment benefit and patient monitoring. Findings regarding the relationship between adherence and remission can additionally help guide the design of future studies.

Predicting overall survival in chordoma patients using machine learning models: a web-app application

ObjectiveThe goal of this study was to evaluate the efficacy of machine learning (ML) techniques in predicting survival for chordoma patients in comparison with the standard Cox proportional hazards (CoxPH) model.MethodsUsing a Surveillance, Epidemiology, and End Results database of consecutive newly diagnosed chordoma cases between January 2000 and December 2018, we created and validated three ML survival models as well as a traditional CoxPH model in this population-based cohort study. Randomly, the dataset was divided into training and validation datasets. Tuning hyperparameters on the training dataset involved a 1000-iteration random search with fivefold cross-validation. Concordance index (C-index), Brier score, and integrated Brier score were used to evaluate the performance of the model. The receiver operating characteristic (ROC) curves, calibration curves, and area under the ROC curves (AUC) were used to assess the reliability of the models by predicting 5- and 10-year survival probabilities.ResultsA total of 724 chordoma patients were divided into training (n = 508) and validation (n = 216) cohorts. Cox regression identified nine significant prognostic factors (p < 0.05). ML models showed superior performance over CoxPH model, with DeepSurv having the highest C-index (0.795) and the best discrimination for 5- and 10-year survival (AUC 0.84 and 0.88). Calibration curves revealed strong correlation between DeepSurv predictions and actual survival. Risk stratification by DeepSurv model effectively discriminated high- and low-risk groups (p < 0.01). The optimized DeepSurv model was implemented into a web application for clinical use that can be found at https://hust-chengp-ml-chordoma-app-19rjyr.streamlitapp.com/.ConclusionML algorithms based on time-to-event results are effective in chordoma prediction, with DeepSurv having the best discrimination performance and calibration.

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

Impact of pre-transplantation 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 one-year mortality after pediatric heart transplantation. Data was obtained from the UNOS Database for years 2010-2020 for patients 0-17 years receiving their first heart transplant (N=4,150). 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). 7 models were tested, Hyperparameter tuning performed using Bayesian optimization and the C-index was used for model assessment. A concordance 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 five features were the most recent serum total bilirubin, the travel distance from the transplant center, the patient BMI, the deceased donor terminal SGPT/ALT and the donor PCO2. Combination of ML and expert-based methodology of selecting predictors of survival for pediatric heart transplantation provides a reasonable prediction of 1 and 3-year survival outcomes. SHAP can be an effective tool for modeling and visualizing nonlinear interactions.

P16 A Comparison of the Cox-Proportional Hazard Model and the Fine-Gray Model to Predict Survival in Patients with Non-Small Cell Lung Cancer (NSCLC)

Predicting survival for patients with non-small cell lung cancer (NSCLC) using the traditional Kaplan Meier (KM) model or Cox proportional hazard (CPH) model leads to over-simplification as competing events are ignored. The Fine-Gray (FG) competing risk model accounts for these competing events. This study compared the methodologies for analyzing the impact of various prognostic factors on survival.

Predicting the survival of patients with pancreatic neuroendocrine neoplasms using deep learning: A study based on Surveillance, Epidemiology, and End Results database.

The study aims to evaluate the performance of three advanced machine learning algorithms and a traditional Cox proportional hazard (CoxPH) model in predicting the overall survival (OS) of patients with pancreatic neuroendocrine neoplasms (PNENs). The clinicopathological dataset obtained from the Surveillance, Epidemiology, and End Results database was randomly assigned to the training set and testing set at a ratio of 7:3. The concordance index (C-index) and integrated Brier score (IBS) were used to compare the predictive performance of the models. The accuracy of the model in predicting the 5-year and 10-year survival rates was compared using the receiver operating characteristic curve, decision curve analysis (DCA) and calibration curve. This study included 3239 patients with PNENs in total. The DeepSurv model had the highest C-index of 0.7882 in the testing set and training set and the lowest IBS of 0.1278 in the testing set compared with the CoxPH, neural multitask logistic and random survival forest models (C-index = 0.7501, 0.7616, and 0.7612, respectively; IBS = 0.1397, 0.1418, and 0.1432, respectively). Moreover, the DeepSurv model had the highest accuracy in predicting 5- and 10-year OS rates (area under the curve: 0.87 and 0.90). DCA showed that the DeepSurv model had high potential for clinical decisions in 5- and 10-year OS models. Finally, we developed an online application based on the DeepSurv model for clinical use (https://whuh-ml-neuroendocrinetumor-app-predict-oyw5km.streamlit.app/). All four models analyzed above can predict the prognosis of PNENs well, among which the DeepSurv model has the best prediction performance.

Development and validation of survival prediction model for gastric adenocarcinoma patients using deep learning: A SEER-based study.

The currently available prediction models, such as the Cox model, were too simplistic to correctly predict the outcome of gastric adenocarcinoma patients. This study aimed to develop and validate survival prediction models for gastric adenocarcinoma patients using the deep learning survival neural network. A total of 14,177 patients with gastric adenocarcinoma from the Surveillance, Epidemiology, and End Results (SEER) database were included in the study and randomly divided into the training and testing group with a 7:3 ratio. Two algorithms were chosen to build the prediction models, and both algorithms include random survival forest (RSF) and a deep learning based-survival prediction algorithm (DeepSurv). Also, a traditional Cox proportional hazard (CoxPH) model was constructed for comparison. The consistency index (C-index), Brier score, and integrated Brier score (IBS) were used to evaluate the model's predictive performance. The accuracy of predicting survival at 1, 3, 5, and 10 years was also assessed using receiver operating characteristic curves (ROC), calibration curves, and area under the ROC curve (AUC). Gastric adenocarcinoma patients were randomized into a training group (n = 9923) and a testing group (n = 4254). DeepSurv showed the best performance among the three models (c-index: 0.772, IBS: 0.1421), which was superior to that of the traditional CoxPH model (c-index: 0.755, IBS: 0.1506) and the RSF with 3-year survival prediction model (c-index: 0.766, IBS: 0.1502). The DeepSurv model produced superior accuracy and calibrated survival estimates predicting 1-, 3- 5- and 10-year survival (AUC: 0.825-0.871). A deep learning algorithm was developed to predict more accurate prognostic information for gastric cancer patients. The DeepSurv model has advantages over the CoxPH and RSF models and performs well in discriminative performance and calibration.

Machine Learning at the Service of Survival Analysis: Predictions Using Time-to-Event Decomposition and Classification Applied to a Decrease of Blood Antibodies against COVID-19

The Cox proportional hazard model may predict whether an individual belonging to a given group would likely register an event of interest at a given time. However, the Cox model is limited by relatively strict statistical assumptions. In this study, we propose decomposing the time-to-event variable into “time” and “event” components and using the latter as a target variable for various machine-learning classification algorithms, which are almost assumption-free, unlike the Cox model. While the time component is continuous and is used as one of the covariates, i.e., input variables for various classification algorithms such as logistic regression, naïve Bayes classifiers, decision trees, random forests, and artificial neural networks, the event component is binary and thus may be modeled using these classification algorithms. Moreover, we apply the proposed method to predict a decrease or non-decrease of IgG and IgM blood antibodies against COVID-19 (SARS-CoV-2), respectively, below a laboratory cut-off, for a given individual at a given time point. Using train-test splitting of the COVID-19 dataset (n=663 individuals), models for the mentioned algorithms, including the Cox proportional hazard model, are learned and built on the train subsets while tested on the test ones. To increase robustness of the model performance evaluation, models’ predictive accuracies are estimated using 10-fold cross-validation on the split dataset. Even though the time-to-event variable decomposition might ignore the effect of individual data censoring, many algorithms show similar or even higher predictive accuracy compared to the traditional Cox proportional hazard model. In COVID-19 IgG decrease prediction, multivariate logistic regression (of accuracy 0.811), support vector machines (of accuracy 0.845), random forests (of accuracy 0.836), artificial neural networks (of accuracy 0.806) outperform the Cox proportional hazard model (of accuracy 0.796), while in COVID-19 IgM antibody decrease prediction, neither Cox regression nor other algorithms perform well (best accuracy is 0.627 for Cox regression). An accurate prediction of mainly COVID-19 IgG antibody decrease can help the healthcare system manage, with no need for extensive blood testing, to identify individuals, for instance, who could postpone boosting vaccination if new COVID-19 variant incomes or should be flagged as high risk due to low COVID-19 antibodies.

Comparing machine learning approaches to incorporate time-varying covariates in predicting cancer survival time

The Cox proportional hazards model is commonly used in evaluating risk factors in cancer survival data. The model assumes an additive, linear relationship between the risk factors and the log hazard. However, this assumption may be too simplistic. Further, failure to take time-varying covariates into account, if present, may lower prediction accuracy. In this retrospective, population-based, prognostic study of data from patients diagnosed with cancer from 2008 to 2015 in Ontario, Canada, we applied machine learning-based time-to-event prediction methods and compared their predictive performance in two sets of analyses: (1) yearly-cohort-based time-invariant and (2) fully time-varying covariates analysis. Machine learning-based methods—gradient boosting model (gbm), random survival forest (rsf), elastic net (enet), lasso and ridge—were compared to the traditional Cox proportional hazards (coxph) model and the prior study which used the yearly-cohort-based time-invariant analysis. Using Harrell’s C index as our primary measure, we found that using both machine learning techniques and incorporating time-dependent covariates can improve predictive performance. Gradient boosting machine showed the best performance on test data in both time-invariant and time-varying covariates analysis.

Longitudinal Data to Enhance Dynamic Stroke Risk Prediction.

Stroke risk prediction based on electronic health records is currently an important research topic. Previous research activities have generally used single-time physiological data to build static models and have focused on algorithms to improve prediction accuracy. Few studies have considered historical measurements from a data perspective to construct dynamic models. Since it is a chronic disease, the risk of having a stroke increases and the corresponding risk factors become abnormal when healthy people are diagnosed with a stroke. Therefore, in this paper, we applied longitudinal data, with the backward joint model, to the Chinese Longitudinal Healthy Longevity and Happy Family Study's dataset to monitor changes in individuals' health status precisely on time and to increase the prediction accuracy of the model. The three-year prediction accuracy of our model, considering three measurements of longitudinal parameters, is 0.926. This is higher than the traditional Cox proportional hazard model, which has a 0.833 prediction accuracy. The results obtained in this study verified that longitudinal data improves stroke risk prediction accuracy and is promising for dynamic stroke risk prediction and prevention. Our model also verified that the frequency of fruit consumption, erythrocyte hematocrit, and glucose are potential stroke-related factors.

An R-Based Landscape Validation of a Competing Risk Model.

The Cox proportional hazard model is widely applied for survival analyses in clinical settings, but it is not able to cope with multiple survival outcomes. Different from the traditional Cox proportional hazard model, competing risk models consider the presence of competing events and their combination with a nomogram, a graphical calculating device, which is a useful tool for clinicians to conduct a precise prognostic prediction. In this study, we report a method for establishing the competing risk nomogram, that is, the evaluation of its discrimination (i.e., concordance index and area under the curve) and calibration (i.e., calibration curves) abilities, as well as the net benefit (i.e., decision curve analysis). In addition, internal validation using bootstrap resamples of the original dataset and external validation using an external dataset of the established competing risk nomogram were also performed to demonstrate its extrapolation ability. The competing risk nomogram should serve as a useful tool for clinicians to predict prognosis with the consideration of competing risks.

A deep learning model with incorporation of microvascular invasion area as a factor in predicting prognosis of hepatocellular carcinoma after R0 hepatectomy.

Microvascular invasion (MVI) is a known risk factor for prognosis after R0 liver resection for hepatocellular carcinoma (HCC). The aim of this study was to develop a deep learning prognostic prediction model by incorporating a new factor of MVI area to the other independent risk factors. Consecutive patients with HCC who underwent R0 liver resection from January to December 2016 at the Eastern Hepatobiliary Surgery Hospital were included in this retrospective study. For patients with MVI detected on resected specimens, they were divided into two groups according to the size of the maximal MVI area: the small-MVI group and the large-MVI group. Of 193 patients who had MVI in the 337 HCC patients, 130 patients formed the training cohort and 63 patients formed the validation cohort. The large-MVI group of patients had worse overall survival (OS) when compared with the small-MVI group (p = 0.009). A deep learning model was developed based on the following independent risk factors found in this study: MVI stage, maximal MVI area, presence/absence of cirrhosis, and maximal tumor diameter. The areas under the receiver operating characteristic of the deep learning model for the 1-, 3-, and 5-year predictions of OS were 80.65, 74.04, and 79.44, respectively, which outperformed the traditional COX proportional hazards model. The deep learning model, by incorporating the maximal MVI area as an additional prognostic factor to the other previously known independent risk factors, predicted more accurately postoperative long-term OS for HCC patients with MVI after R0 liver resection.