Component-wise gradient boosting and false discovery control in survival analysis with high-dimensional covariates

Levodopa-induced dyskinesia in Parkinson's disease: Insights from cross-cohort prognostic analysis using machine learning

BACKGROUND

Prolonged levodopa treatment in Parkinson's disease (PD) often leads to motor complications, including levodopa-induced dyskinesia (LID). Despite continuous levodopa treatment, some patients do not develop LID symptoms, even in later stages of the disease.

OBJECTIVE

This study explores machine learning (ML) methods using baseline clinical characteristics to predict the development of LID in PD patients over four years, across multiple cohorts.

METHODS

Using interpretable ML approaches, we analyzed clinical data from three independent longitudinal PD cohorts (LuxPARK, n = 356; PPMI, n = 484; ICEBERG, n = 113) to develop cross-cohort prognostic models and identify potential predictors for the development of LID. We examined cohort-specific and shared predictive factors, assessing model performance and stability through cross-validation analyses.

RESULTS

Consistent cross-validation results for single and multiple cohort analyses highlighted the effectiveness of the ML models and identified baseline clinical characteristics with significant predictive value for the LID prognosis in PD. Predictors positively correlated with LID include axial symptoms, freezing of gait, and rigidity in the lower extremities. Conversely, the risk of developing LID was inversely associated with the occurrence of resting tremors, higher body weight, later onset of PD, and visuospatial abilities.

CONCLUSIONS

This study presents interpretable ML models for dyskinesia prognosis with significant predictive power in cross-cohort analyses. The models may pave the way for proactive interventions against dyskinesia in PD by optimizing levodopa dosing regimens and adjunct treatments with dopamine agonists or MAO-B inhibitors, and by employing non-pharmacological interventions such as dietary adjustments affecting levodopa absorption for high-risk LID patients.

Comparison of models for stroke-free survival prediction in patients with CADASIL

Cerebral autosomal dominant arteriopathy with subcortical infarcts and leukoencephalopathy, which is caused by mutations of the NOTCH3 gene, has a large heterogeneous progression, presenting with declines of various clinical scores and occurrences of various clinical event. To help assess disease progression, this work focused on predicting the composite endpoint of stroke-free survival time by comparing the performance of Cox proportional hazards regression to that of machine learning models using one of four feature selection approaches applied to demographic, clinical and magnetic resonance imaging observational data collected from a study cohort of 482 patients. The quality of the modeling process and the predictive performance were evaluated in a nested cross-validation procedure using the time-dependent Brier Score and AUC at 5 years from baseline, the former measuring the overall performance including calibration and the latter highlighting the discrimination ability, with both metrics taking into account the presence of right-censoring. The best model for each metric was the componentwise gradient boosting model with a mean Brier score of 0.165 and the random survival forest model with a mean AUC of 0.773, both combined with the LASSO feature selection method.

A double-robust test for high-dimensional gene coexpression networks conditioning on clinical information

It has been increasingly appealing to evaluate whether expression levels of two genes in a gene coexpression network are still dependent given samples' clinical information, in which the conditional independence test plays an essential role. For enhanced robustness regarding model assumptions, we propose a class of double-robust tests for evaluating the dependence of bivariate outcomes after controlling for known clinical information. Although the proposed test relies on the marginal density functions of bivariate outcomes given clinical information, the test remains valid as long as one of the density functions is correctly specified. Because of the closed-form variance formula, the proposed test procedure enjoys computational efficiency without requiring a resampling procedure or tuning parameters. We acknowledge the need to infer the conditional independence network with high-dimensional gene expressions, and further develop a procedure for multiple testing by controlling the false discovery rate. Numerical results show that our method accurately controls both the type-I error and false discovery rate, and it provides certain levels of robustness regarding model misspecification. We apply the method to a gastric cancer study with gene expression data to understand the associations between genes belonging to the transforming growth factor β signaling pathway given cancer-stage information.

Machine learning survival models trained on clinical data to identify high risk patients with hormone responsive HER2 negative breast cancer

For endocrine-positive Her2 negative breast cancer patients at an early stage, the benefit of adding chemotherapy to adjuvant endocrine therapy is not still confirmed. Several genomic tests are available on the market but are very expensive. Therefore, there is the urgent need to explore novel reliable and less expensive prognostic tools in this setting. In this paper, we shown a machine learning survival model to estimate Invasive Disease-Free Events trained on clinical and histological data commonly collected in clinical practice. We collected clinical and cytohistological outcomes of 145 patients referred to Istituto Tumori "Giovanni Paolo II". Three machine learning survival models are compared with the Cox proportional hazards regression according to time-dependent performance metrics evaluated in cross-validation. The c-index at 10 years obtained by random survival forest, gradient boosting, and component-wise gradient boosting is stabled with or without feature selection at approximately 0.68 in average respect to 0.57 obtained to Cox model. Moreover, machine learning survival models have accurately discriminated low- and high-risk patients, and so a large group which can be spared additional chemotherapy to hormone therapy. The preliminary results obtained by including only clinical determinants are encouraging. The integrated use of data already collected in clinical practice for routine diagnostic investigations, if properly analyzed, can reduce time and costs of the genomic tests.

Efficient prediction of early-stage diabetes using XGBoost classifier with random forest feature selection technique

Diabetes is one of the most common and serious diseases affecting human health. Early diagnosis and treatment are vital to prevent or delay complications related to diabetes. An automated diabetes detection system assists physicians in the early diagnosis of the disease and reduces complications by providing fast and precise results. This study aims to introduce a technique based on a combination of multiple linear regression (MLR), random forest (RF), and XGBoost (XG) to diagnose diabetes from questionnaire data. MLR-RF algorithm is used for feature selection, and XG is used for classification in the proposed system. The dataset is the diabetic hospital data in Sylhet, Bangladesh. It contains 520 instances, including 320 diabetics and 200 control instances. The performance of the classifiers is measured concerning accuracy (ACC), precision (PPV), recall (SEN, sensitivity), F1 score (F1), and the area under the receiver-operating-characteristic curve (AUC). The results show that the proposed system achieves an accuracy of 99.2%, an AUC of 99.3%, and a prediction time of 0.04825 seconds. The feature selection method improves the prediction time, although it does not affect the accuracy of the four compared classifiers. The results of this study are quite reasonable and successful when compared with other studies. The proposed method can be used as an auxiliary tool in diagnosing diabetes.

Clinical factors associated with rapid treatment of sepsis

PURPOSE

To understand what clinical presenting features of sepsis patients are historically associated with rapid treatment involving antibiotics and fluids, as appropriate.

DESIGN

This was a retrospective, observational cohort study using a machine-learning model with an embedded feature selection mechanism (gradient boosting machine).

METHODS

For adult patients (age ≥ 18 years) who were admitted through Emergency Department (ED) meeting clinical criteria of severe sepsis from 11/2007 to 05/2018 at an urban tertiary academic medical center, we developed gradient boosting models (GBMs) using a total of 760 original and derived variables, including demographic variables, laboratory values, vital signs, infection diagnosis present on admission, and historical comorbidities. We identified the most impactful factors having strong association with rapid treatment, and further applied the Shapley Additive exPlanation (SHAP) values to examine the marginal effects for each factor.

RESULTS

For the subgroups with or without fluid bolus treatment component, the models achieved high accuracy of area-under-receiver-operating-curve of 0.91 [95% CI, 0.86-0.95] and 0.84 [95% CI, 0.81-0.86], and sensitivity of 0.81[95% CI, 0.72-0.87] and 0.91 [95% CI, 0.81-0.97], respectively. We identified the 20 most impactful factors associated with rapid treatment for each subgroup. In the non-hypotensive subgroup, initial physiological values were the most impactful to the model, while in the fluid bolus subgroup, value minima and maxima tended to be the most impactful.

CONCLUSION

These machine learning methods identified factors associated with rapid treatment of severe sepsis patients from a large volume of high-dimensional clinical data. The results provide insight into differences in the rapid provision of treatment among patients with sepsis.

Cystic fibrosis-related diabetes onset can be predicted using biomarkers measured at birth

Purpose

Cystic fibrosis (CF), caused by pathogenic variants in the CF transmembrane conductance regulator (CFTR), affects multiple organs including the exocrine pancreas, which is a causal contributor to cystic fibrosis–related diabetes (CFRD). Untreated CFRD causes increased CF-related mortality whereas early detection can improve outcomes.

Methods

Using genetic and easily accessible clinical measures available at birth, we constructed a CFRD prediction model using the Canadian CF Gene Modifier Study (CGS; n = 1,958) and validated it in the French CF Gene Modifier Study (FGMS; n = 1,003). We investigated genetic variants shown to associate with CF disease severity across multiple organs in genome-wide association studies.

Results

The strongest predictors included sex, CFTR severity score, and several genetic variants including one annotated to PRSS1, which encodes cationic trypsinogen. The final model defined in the CGS shows excellent agreement when validated on the FGMS, and the risk classifier shows slightly better performance at predicting CFRD risk later in life in both studies.

Conclusion

We demonstrated clinical utility by comparing CFRD prevalence rates between the top 10% of individuals with the highest risk and the bottom 10% with the lowest risk. A web-based application was developed to provide practitioners with patient-specific CFRD risk to guide CFRD monitoring and treatment.

SurvBoost: An R Package for High-Dimensional Variable Selection in the Stratified Proportional Hazards Model via Gradient Boosting

High-dimensional variable selection in the proportional hazards (PH) model has many successful applications in different areas. In practice, data may involve confounding variables that do not satisfy the PH assumption, in which case the stratified proportional hazards (SPH) model can be adopted to control the confounding effects by stratification without directly modeling the confounding effects. However, there is a lack of computationally efficient statistical software for high-dimensional variable selection in the SPH model. In this work an R package, SurvBoost, is developed to implement the gradient boosting algorithm for fitting the SPH model with high-dimensional covariate variables. Simulation studies demonstrate that in many scenarios SurvBoost can achieve better selection accuracy and reduce computational time substantially compared to the existing R package that implements boosting algorithms without stratification. The proposed R package is also illustrated by an analysis of gene expression data with survival outcome in The Cancer Genome Atlas study. In addition, a detailed hands-on tutorial for SurvBoost is provided.

Longitudinal Risk Prediction of Chronic Kidney Disease in Diabetic Patients Using a Temporal-Enhanced Gradient Boosting Machine: Retrospective Cohort Study

BACKGROUND

Artificial intelligence-enabled electronic health record (EHR) analysis can revolutionize medical practice from the diagnosis and prediction of complex diseases to making recommendations in patient care, especially for chronic conditions such as chronic kidney disease (CKD), which is one of the most frequent complications in patients with diabetes and is associated with substantial morbidity and mortality.

OBJECTIVE

The longitudinal prediction of health outcomes requires effective representation of temporal data in the EHR. In this study, we proposed a novel temporal-enhanced gradient boosting machine (GBM) model that dynamically updates and ensembles learners based on new events in patient timelines to improve the prediction accuracy of CKD among patients with diabetes.

METHODS

Using a broad spectrum of deidentified EHR data on a retrospective cohort of 14,039 adult patients with type 2 diabetes and GBM as the base learner, we validated our proposed Landmark-Boosting model against three state-of-the-art temporal models for rolling predictions of 1-year CKD risk.

RESULTS

The proposed model uniformly outperformed other models, achieving an area under receiver operating curve of 0.83 (95% CI 0.76-0.85), 0.78 (95% CI 0.75-0.82), and 0.82 (95% CI 0.78-0.86) in predicting CKD risk with automatic accumulation of new data in later years (years 2, 3, and 4 since diabetes mellitus onset, respectively). The Landmark-Boosting model also maintained the best calibration across moderate- and high-risk groups and over time. The experimental results demonstrated that the proposed temporal model can not only accurately predict 1-year CKD risk but also improve performance over time with additionally accumulated data, which is essential for clinical use to improve renal management of patients with diabetes.

CONCLUSIONS

Incorporation of temporal information in EHR data can significantly improve predictive model performance and will particularly benefit patients who follow-up with their physicians as recommended.

An exploration of ontology-based EMR data abstraction for diabetic kidney disease prediction

Diabetic Kidney Disease (DKD) is a critical and morbid complication of diabetes and the leading cause of chronic kidney disease in the developed world. Electronic medical records (EMRs) hold promise for supporting clinical decision-making with its nationwide adoption as well as rich information characterizing patients' health care experience. However, few retrospective studies have fully utilized the EMR data to model DKD risk. This study examines the effectiveness of an unbiased data driven approach in identifying potential DKD patients in 6 months prior to onset by utilizing EMR on a broader spectrum. Meanwhile, we evaluate how different levels of data granularity of Medications and Diagnoses observations would affect prediction performance and knowledge discovery. The experimental results suggest that different data granularity may not necessarily influence the prediction accuracy, but it would dramatically change the internal structure of the predictive models.

Covariance-Insured Screening

Modern bio-technologies have produced a vast amount of high-throughput data with the number of predictors far greater than the sample size. In order to identify more novel biomarkers and understand biological mechanisms, it is vital to detect signals weakly associated with outcomes among ultrahigh-dimensional predictors. However, existing screening methods, which typically ignore correlation information, are likely to miss weak signals. By incorporating the inter-feature dependence, a covariance-insured screening approach is proposed to identify predictors that are jointly informative but marginally weakly associated with outcomes. The validity of the method is examined via extensive simulations and a real data study for selecting potential genetic factors related to the onset of multiple myeloma.

Robust clinical marker identification for diabetic kidney disease with ensemble feature selection

Objective

Diabetic kidney disease (DKD) is one of the most frequent complications in diabetes associated with substantial morbidity and mortality. To accelerate DKD risk factor discovery, we present an ensemble feature selection approach to identify a robust set of discriminant factors using electronic medical records (EMRs).

Material and Methods

We identified a retrospective cohort of 15 645 adult patients with type 2 diabetes, excluding those with pre-existing kidney disease, and utilized all available clinical data types in modeling. We compared 3 machine-learning-based embedded feature selection methods in conjunction with 6 feature ensemble techniques for selecting top-ranked features in terms of robustness to data perturbations and predictability for DKD onset.

Results

The gradient boosting machine (GBM) with weighted mean rank feature ensemble technique achieved the best performance with an AUC of 0.82 [95%-CI, 0.81-0.83] on internal validation and 0.71 [95%-CI, 0.68-0.73] on external temporal validation. The ensemble model identified a set of 440 features from 84 872 unique clinical features that are both predicative of DKD onset and robust against data perturbations, including 191 labs, 51 visit details (mainly vital signs), 39 medications, 34 orders, 30 diagnoses, and 95 other clinical features.

Discussion

Many of the top-ranked features have not been included in the state-of-art DKD prediction models, but their relationships with kidney function have been suggested in existing literature.

Conclusion

Our ensemble feature selection framework provides an option for identifying a robust and parsimonious feature set unbiasedly from EMR data, which effectively aids in knowledge discovery for DKD risk factors.

False discovery control for penalized variable selections with high-dimensional covariates

Modern bio-technologies have produced a vast amount of high-throughput data with the number of predictors much exceeding the sample size. Penalized variable selection has emerged as a powerful and efficient dimension reduction tool. However, control of false discoveries (i.e. inclusion of irrelevant variables) for penalized high-dimensional variable selection presents serious challenges. To effectively control the fraction of false discoveries for penalized variable selections, we propose a false discovery controlling procedure. The proposed method is general and flexible, and can work with a broad class of variable selection algorithms, not only for linear regressions, but also for generalized linear models and survival analysis.

Ensembling Variable Selectors by Stability Selection for the Cox Model

As a pivotal tool to build interpretive models, variable selection plays an increasingly important role in high-dimensional data analysis. In recent years, variable selection ensembles (VSEs) have gained much interest due to their many advantages. Stability selection (Meinshausen and Bühlmann, 2010), a VSE technique based on subsampling in combination with a base algorithm like lasso, is an effective method to control false discovery rate (FDR) and to improve selection accuracy in linear regression models. By adopting lasso as a base learner, we attempt to extend stability selection to handle variable selection problems in a Cox model. According to our experience, it is crucial to set the regularization region Λ in lasso and the parameter λ_min properly so that stability selection can work well. To the best of our knowledge, however, there is no literature addressing this problem in an explicit way. Therefore, we first provide a detailed procedure to specify Λ and λ_min. Then, some simulated and real-world data with various censoring rates are used to examine how well stability selection performs. It is also compared with several other variable selection approaches. Experimental results demonstrate that it achieves better or competitive performance in comparison with several other popular techniques.

Accommodating missingness in environmental measurements in gene-environment interaction analysis

For the prognosis of complex diseases, beyond the main effects of genetic (G) and environmental (E) factors, gene-environment (G-E) interactions also play an important role. Many approaches have been developed for detecting important G-E interactions, most of which assume that measurements are complete. In practical data analysis, missingness in E measurements is not uncommon, and failing to properly accommodate such missingness leads to biased estimation and false marker identification. In this study, we conduct G-E interaction analysis with prognosis data under an accelerated failure time (AFT) model. To accommodate missingness in E measurements, we adopt a nonparametric kernel-based data augmentation approach. With a well-designed weighting scheme, a nice "byproduct" is that the proposed approach enjoys a certain robustness property. A penalization approach, which respects the "main effects, interactions" hierarchy, is adopted for selection (of important interactions and main effects) and regularized estimation. The proposed approach has sound interpretations and a solid statistical basis. It outperforms multiple alternatives in simulation. The analysis of TCGA data on lung cancer and melanoma leads to interesting findings and models with superior prediction.