Comparison of Cox Model Methods in A Low-dimensional Setting with Few Events

Novel nomograms predicting overall and cancer-specific survival of malignant ependymoma patients: a population-based study

BACKGROUND

Malignant ependymomas are rare cancerous tumors that are associated with increased morbidity and mortality in the affected patients. Lately, there has been a lot of controversy about the correct way to manage and predict the survival outcome of these patients. We aim in this retrospective cohort study to develop novel nomograms that can better predict the overall survival (OS) and cancer-specific survival (CSS) of these patients.

METHODS

This is a retrospective cohort study that was conducted through the Surveillance, Epidemiology, and End Results databases (SEER) between 1998 and 2016. Patients were excluded if they had an unknown diagnosis, unknown cause of death or those with survival duration less than a month. We used penalized regression models with the highest time-dependent area under the ROC curve (AUC) and most stable calibrations to construct the nomograms. By searching the SEER database and applying the eligibility criteria, we identified 3391 patients for the final analysis.

RESULTS

Nine penalized regression models were developed of which two models including adaptive elastic-net was selected for both OS and CSS. The model incorporated age, sex, year of diagnosis, site, race, radiation, chemotherapy, surgery, and type for the construction of nomograms. We aimed in this population-based cohort study to develop novel prediction tools that can help physicians estimate the survival of malignant ependymoma patients and provide better care.

CONCLUSIONS

Our nomograms appear to have high accuracy and applicability, which we hope that can predict the survival and improve the treatment and prognosis of these patients.

Prognosis of lasso-like penalized Cox models with tumor profiling improves prediction over clinical data alone and benefits from bi-dimensional pre-screening

BACKGROUND

Prediction of patient survival from tumor molecular '-omics' data is a key step toward personalized medicine. Cox models performed on RNA profiling datasets are popular for clinical outcome predictions. But these models are applied in the context of "high dimension", as the number p of covariates (gene expressions) greatly exceeds the number n of patients and e of events. Thus, pre-screening together with penalization methods are widely used for dimensional reduction.

METHODS

In the present paper, (i) we benchmark the performance of the lasso penalization and three variants (i.e., ridge, elastic net, adaptive elastic net) on 16 cancers from TCGA after pre-screening, (ii) we propose a bi-dimensional pre-screening procedure based on both gene variability and p-values from single variable Cox models to predict survival, and (iii) we compare our results with iterative sure independence screening (ISIS).

RESULTS

First, we show that integration of mRNA-seq data with clinical data improves predictions over clinical data alone. Second, our bi-dimensional pre-screening procedure can only improve, in moderation, the C-index and/or the integrated Brier score, while excluding irrelevant genes for prediction. We demonstrate that the different penalization methods reached comparable prediction performances, with slight differences among datasets. Finally, we provide advice in the case of multi-omics data integration.

CONCLUSIONS

Tumor profiles convey more prognostic information than clinical variables such as stage for many cancer subtypes. Lasso and Ridge penalizations perform similarly than Elastic Net penalizations for Cox models in high-dimension. Pre-screening of the top 200 genes in term of single variable Cox model p-values is a practical way to reduce dimension, which may be particularly useful when integrating multi-omics.

Machine learning-guided covariate selection for time-to-event models developed from a small sample of real-world patients receiving bevacizumab treatment

Therapeutic outcomes in patients with metastatic colorectal cancer (mCRC) receiving bevacizumab treatment are highly variable, and a reliable predictive factor is not available. Progression-free survival (PFS) and overall survival (OS) were recorded from an observational, prospective study after 5 years of follow-up, including 46 patients with mCRC receiving bevacizumab treatment. Three vascular endothelial growth factor (VEGF)-A and two intercellular adhesion molecule-1 genes polymorphisms, age, gender, weight, dosing scheme, and co-treatments were collected. Given the relatively small number of events (37 [80%] for the PFS and 26 [57%] for the OS), to study the effect of these covariates on PFS and OS, a covariate analysis was performed using statistical and supervised machine learning techniques, including Cox regression, penalized Cox regression techniques (least absolute shrinkage and selection operator [LASSO], ridge regression, and elastic net), survival trees, and survival forest. The predictive performance of each method was evaluated in bootstrapped samples, using prediction error curves and the area under the curve of the receiver operating characteristic. The LASSO penalized Cox-regression model showed the best overall performance. Nonlinear mixed effects (NLME) models were developed, and a conventional stepwise covariate search was performed. Then, covariates identified as important by the LASSO model were included in the base NLME models developed for PFS and OS, resulting in improved models as compared to those obtained with the stepwise covariate search. It was shown that having gene polymorphisms in VEGFA (rs699947 and rs1570360) and ICAM1 (rs1799969) are associated with a favorable clinical outcome in patients with mCRC receiving bevacizumab treatment.

Radiation and immune checkpoint inhibitor-mediated pneumonitis risk stratification in patients with locally advanced non-small cell lung cancer: role of functional lung radiomics?

BACKGROUND

Patients undergoing chemoradiation and immune checkpoint inhibitor (ICI) therapy for locally advanced non-small cell lung cancer (NSCLC) experience pulmonary toxicity at higher rates than historical reports. Identifying biomarkers beyond conventional clinical factors and radiation dosimetry is especially relevant in the modern cancer immunotherapy era. We investigated the role of novel functional lung radiomics, relative to functional lung dosimetry and clinical characteristics, for pneumonitis risk stratification in locally advanced NSCLC.

METHODS

Patients with locally advanced NSCLC were prospectively enrolled on the FLARE-RT trial (NCT02773238). All received concurrent chemoradiation using functional lung avoidance planning, while approximately half received consolidation durvalumab ICI. Within tumour-subtracted lung regions, 110 radiomics features (size, shape, intensity, texture) were extracted on pre-treatment [^99mTc]MAA SPECT/CT perfusion images using fixed-bin-width discretization. The performance of functional lung radiomics for pneumonitis (CTCAE v4 grade 2 or higher) risk stratification was benchmarked against previously reported lung dosimetric parameters and clinical risk factors. Multivariate least absolute shrinkage and selection operator Cox models of time-varying pneumonitis risk were constructed, and prediction performance was evaluated using optimism-adjusted concordance index (c-index) with 95% confidence interval reporting throughout.

RESULTS

Thirty-nine patients were included in the study and pneumonitis occurred in 16/39 (41%) patients. Among clinical characteristics and anatomic/functional lung dosimetry variables, only the presence of baseline chronic obstructive pulmonary disease (COPD) was significantly associated with the development of pneumonitis (HR 4.59 [1.69-12.49]) and served as the primary prediction benchmark model (c-index 0.69 [0.59-0.80]). Discrimination of time-varying pneumonitis risk was numerically higher when combining COPD with perfused lung radiomics size (c-index 0.77 [0.65-0.88]) or shape feature classes (c-index 0.79 [0.66-0.91]) but did not reach statistical significance compared to benchmark models (p > 0.26). COPD was associated with perfused lung radiomics size features, including patients with larger lung volumes (AUC 0.75 [0.59-0.91]). Perfused lung radiomic texture features were correlated with lung volume (adj R² = 0.84-1.00), representing surrogates rather than independent predictors of pneumonitis risk.

CONCLUSIONS

In patients undergoing chemoradiation with functional lung avoidance therapy and optional consolidative immune checkpoint inhibitor therapy for locally advanced NSCLC, the strongest predictor of pneumonitis was the presence of baseline chronic obstructive pulmonary disease. Results from this novel functional lung radiomics exploratory study can inform future validation studies to refine pneumonitis risk models following combinations of radiation and immunotherapy. Our results support functional lung radiomics as surrogates of COPD for non-invasive monitoring during and after treatment. Further study of clinical, dosimetric, and radiomic feature combinations for radiation and immune-mediated pneumonitis risk stratification in a larger patient population is warranted.

COVID-19 and risk of subsequent life-threatening secondary infections: a matched cohort study in UK Biobank

BACKGROUND

With the increasing number of people infected with and recovered from coronavirus disease 2019 (COVID-19), the extent of major health consequences of COVID-19 is unclear, including risks of severe secondary infections.

METHODS

Based on 445,845 UK Biobank participants registered in England, we conducted a matched cohort study where 5151 individuals with a positive test result or hospitalized with a diagnosis of COVID-19 were included in the exposed group. We then randomly selected up to 10 matched individuals without COVID-19 diagnosis for each exposed individual (n = 51,402). The life-threatening secondary infections were defined as diagnoses of severe secondary infections with high mortality rates (i.e., sepsis, endocarditis, and central nervous system infections) from the UK Biobank inpatient hospital data, or deaths from these infections from mortality data. The follow-up period was limited to 3 months after the initial COVID-19 diagnosis. Using a similar study design, we additionally constructed a matched cohort where exposed individuals were diagnosed with seasonal influenza from either inpatient hospital or primary care data between 2010 and 2019 (6169 exposed and 61,555 unexposed individuals). After controlling for multiple confounders, Cox models were used to estimate hazard ratios (HRs) of life-threatening secondary infections after COVID-19 or seasonal influenza.

RESULTS

In the matched cohort for COVID-19, 50.22% of participants were male, and the median age at the index date was 66 years. During a median follow-up of 12.71 weeks, the incidence rate of life-threatening secondary infections was 2.23 (123/55.15) and 0.25 (151/600.55) per 1000 person-weeks for all patients with COVID-19 and their matched individuals, respectively, which corresponded to a fully adjusted HR of 8.19 (95% confidence interval [CI] 6.33-10.59). The corresponding HR of life-threatening secondary infections among all patients with seasonal influenza diagnosis was 4.50, 95% CI 3.34-6.08 (p for difference < 0.01). Also, elevated HRs were observed among hospitalized individuals for life-threatening secondary infections following hospital discharge, both in the COVID-19 (HR = 6.28 [95% CI 4.05-9.75]) and seasonal influenza (6.01 [95% CI 3.53-10.26], p for difference = 0.902) cohorts.

CONCLUSION

COVID-19 patients have increased subsequent risks of life-threatening secondary infections, to an equal extent or beyond risk elevations observed for patients with seasonal influenza.

Identifying miRNA-mRNA Integration Set Associated With Survival Time

In the “personalized medicine” era, one of the most difficult problems is identification of combined markers from different omics platforms. Many methods have been developed to identify candidate markers for each type of omics data, but few methods facilitate the identification of multiple markers on multi-omics platforms. microRNAs (miRNAs) is well known to affect only indirectly phenotypes by regulating mRNA expression and/or protein translation. To take into account this knowledge into practice, we suggest a miRNA-mRNA integration model for survival time analysis, called mimi-surv, which accounts for the biological relationship, to identify such integrated markers more efficiently. Through simulation studies, we found that the statistical power of mimi-surv be better than other models. Application to real datasets from Seoul National University Hospital and The Cancer Genome Atlas demonstrated that mimi-surv successfully identified miRNA-mRNA integrations sets associated with progression-free survival of pancreatic ductal adenocarcinoma (PDAC) patients. Only mimi-surv found miR-96, a previously unidentified PDAC-related miRNA in these two real datasets. Furthermore, mimi-surv was shown to identify more PDAC related miRNAs than other methods because it used the known structure for miRNA-mRNA regularization. An implementation of mimi-surv is available at http://statgen.snu.ac.kr/software/mimi-surv.

Glucose-6-phosphate dehydrogenase deficiency and stroke outcomes

OBJECTIVE

To assess the risk of glucose-6-phosphate dehydrogenase (G6PD) on stroke prognosis, we compared outcomes between patients with stroke with and without G6PD deficiency.

METHODS

The study recruited 1,251 patients with acute ischemic stroke. Patients were individually categorized into G6PD-deficiency and non-G6PD-deficiency groups according to G6PD activity upon admission. The primary endpoint was poor outcome at 3 months defined by a modified Rankin Scale (mRS) score ≥2 (including disability and death). Secondary outcomes included the overall mRS score at 3 months and in-hospital death and all death within 3 months. Logistic regression and Cox models, adjusted for potential confounders, were fitted to estimate the association of G6PD deficiency with the outcomes.

RESULTS

Among 1,251 patients, 150 (12.0%) were G6PD-deficient. Patients with G6PD deficiency had higher proportions of large-artery atherosclerosis (odds ratio [OR] 1.53, 95% confidence interval [CI] 1.09-2.17) and stroke history (OR 1.93, 95% CI 1.26-2.90) compared to the non-G6PD-deficient group. The 2 groups differed significantly in the overall mRS score distribution (adjusted common OR 1.57, 95% CI 1.14-2.17). Patients with G6PD deficiency had higher rates of poor outcome at 3 months (adjusted OR 1.73, 95% CI 1.08-2.76; adjusted absolute risk increase 13.0%, 95% CI 2.4%-23.6%). The hazard ratio of in-hospital death for patients with G6PD-deficiency was 1.46 (95% CI 1.37-1.84).

CONCLUSIONS

G6PD deficiency is associated with the risk of poor outcome at 3 months after ischemic stroke and may increase the risk of in-hospital death. These findings suggest the rationality of G6PD screening in patients with stroke.

Pulse Wave Velocity and Machine Learning to Predict Cardiovascular Outcomes in Prediabetic and Diabetic Populations

Few studies have addressed the predictive value of arterial stiffness determined by pulse wave velocity (PWV) in a high-risk population with no prevalent cardiovascular disease and with obesity, hypertension, hyperglycemia, and preserved kidney function. This longitudinal, retrospective study enrolled 88 high-risk patients and had a follow-up time of 12.4 years. We collected clinical and laboratory data, as well as information on arterial stiffness parameters using arterial tonometry and measurements from ambulatory blood pressure monitoring. We considered nonfatal, incident cardiovascular events as the primary outcome. Given the small size of our dataset, we used survival analysis (i.e., Cox proportional hazards model) combined with a machine learning-based algorithm/penalization method to evaluate the data. Our predictive model, calculated with Cox regression and least absolute shrinkage and selection operator (LASSO), included body mass index, diabetes mellitus, gender (male), and PWV. We recorded 16 nonfatal cardiovascular events (5 myocardial infarctions, 5 episodes of heart failure, and 6 strokes). The adjusted hazard ratio for PWV was 1.199 (95% confidence interval: 1.09-1.37, p < 0.001). Arterial stiffness was a predictor of cardiovascular disease development, as determined by PWV in a high-risk population. Thus, in obese, hypertensive, hyperglycemic patients with preserved kidney function, PWV can serve as a prognostic factor for major adverse cardiac events.

Prognostic Significance of Pulse Pressure and Other Blood Pressure Components for Coronary Artery Disease in Type 1 Diabetes

BACKGROUND

To compare in individuals with type 1 diabetes the prediction of incident coronary artery disease (CAD) by components of resting blood pressure-systolic, diastolic, pulse pressure, and mean arterial pressure.

METHODS

In 605 participants without known CAD at baseline and followed sequentially for 25 years, we used Cox modeling built for each blood pressure component associated with incident CAD, overall and stratified by age (<35 and ≥35 years) or hemoglobin A1c (HbA1c) (<9% and ≥9%).

RESULTS

Baseline mean age and diabetes duration were 27 and 19 years, respectively. We observed an early asymptote and then fall in diastolic blood pressure in their late 30s and early 40s in this group of type 1 diabetes individuals, followed by an early rise of pulse pressure. Adjusted hazard ratios (HR) (95% con) for CAD associated with 1 SD pressure increase were 1.35 (1.17, 1.56) for systolic pressure; 1.30 (1.12, 1.51) for diastolic pressure; 1.20 (1.03, 1.39) for pulse pressure; and 1.35 (1.17, 1.56) for mean arterial pressure. Pulse pressure emerged as a strong predictor of CAD at age ≥ 35 years (HR: 1.49 [1.15, 1.94]) and for HbA1c ≥ 9% (HR: 1.32 [1.01, 1.72]).

CONCLUSIONS

Individuals with type 1 diabetes may manifest early vascular aging by an early decline in diastolic blood pressure and rise in pulse pressure, the latter parameter becoming a comparable to systolic blood pressure in predictor incident CAD in those aged over 35 years and those with poor glycemic control.

Comparison of Penalized Cox Regression Methods in Low-Dimensional Data with Few-Events: An Application to Dialysis Patients' Data

BACKGROUND

Dialysis is a dominant therapeutic method in patients with chronic renal failure. The ratio of those who experienced the event to the predictor variables is expressed as event per variable (EPV). When EPV is low, one of the common techniques which may help to manage the problem is penalized Cox regression model (PCRM). The aim of this study was to determine the survival of dialysis patients using the PCRM in low-dimensional data with few events.

STUDY DESIGN

A cross-sectional study.

METHODS

Information of 252 dialysis patients of Bandar Abbas hospitals, southern Iran, from 2010-16 were used. To deal with few mortality cases in the sample, the PCRM (lasso, ridge and elastic net, adaptive lasso) were applied. Models were compared in terms of calibration and discrimination.

RESULTS

Thirty-five (13.9%) mortality cases were observed. Dialysis data simulations revealed that the lasso had higher prediction accuracy than other models. For one unit of increase in the level of education, the risk of mortality was reduced by 0.32 (HR=0.68). The risk of mortality was 0.26 (HR=1.26) higher for the unemployed than the employed cases. Other significant factors were the duration of each dialysis session, number of dialysis sessions per week and age of dialysis onset (HR=0.93, 0.95 and 1.33).

CONCLUSION

The performance of penalized models, especially the lasso, was satisfying in low-dimensional data with low EPV based on dialysis data simulation and real data, therefore these models are the good choice for managing of this type of data.

Machine learning in cardiovascular medicine: are we there yet?

Artificial intelligence (AI) broadly refers to analytical algorithms that iteratively learn from data, allowing computers to find hidden insights without being explicitly programmed where to look. These include a family of operations encompassing several terms like machine learning, cognitive learning, deep learning and reinforcement learning-based methods that can be used to integrate and interpret complex biomedical and healthcare data in scenarios where traditional statistical methods may not be able to perform. In this review article, we discuss the basics of machine learning algorithms and what potential data sources exist; evaluate the need for machine learning; and examine the potential limitations and challenges of implementing machine in the context of cardiovascular medicine. The most promising avenues for AI in medicine are the development of automated risk prediction algorithms which can be used to guide clinical care; use of unsupervised learning techniques to more precisely phenotype complex disease; and the implementation of reinforcement learning algorithms to intelligently augment healthcare providers. The utility of a machine learning-based predictive model will depend on factors including data heterogeneity, data depth, data breadth, nature of modelling task, choice of machine learning and feature selection algorithms, and orthogonal evidence. A critical understanding of the strength and limitations of various methods and tasks amenable to machine learning is vital. By leveraging the growing corpus of big data in medicine, we detail pathways by which machine learning may facilitate optimal development of patient-specific models for improving diagnoses, intervention and outcome in cardiovascular medicine.

Performance of Firth-and logF-type penalized methods in risk prediction for small or sparse binary data

BACKGROUND

When developing risk models for binary data with small or sparse data sets, the standard maximum likelihood estimation (MLE) based logistic regression faces several problems including biased or infinite estimate of the regression coefficient and frequent convergence failure of the likelihood due to separation. The problem of separation occurs commonly even if sample size is large but there is sufficient number of strong predictors. In the presence of separation, even if one develops the model, it produces overfitted model with poor predictive performance. Firth-and logF-type penalized regression methods are popular alternative to MLE, particularly for solving separation-problem. Despite the attractive advantages, their use in risk prediction is very limited. This paper evaluated these methods in risk prediction in comparison with MLE and other commonly used penalized methods such as ridge.

METHODS

The predictive performance of the methods was evaluated through assessing calibration, discrimination and overall predictive performance using an extensive simulation study. Further an illustration of the methods were provided using a real data example with low prevalence of outcome.

RESULTS

The MLE showed poor performance in risk prediction in small or sparse data sets. All penalized methods offered some improvements in calibration, discrimination and overall predictive performance. Although the Firth-and logF-type methods showed almost equal amount of improvement, Firth-type penalization produces some bias in the average predicted probability, and the amount of bias is even larger than that produced by MLE. Of the logF(1,1) and logF(2,2) penalization, logF(2,2) provides slight bias in the estimate of regression coefficient of binary predictor and logF(1,1) performed better in all aspects. Similarly, ridge performed well in discrimination and overall predictive performance but it often produces underfitted model and has high rate of convergence failure (even the rate is higher than that for MLE), probably due to the separation problem.

CONCLUSIONS

The logF-type penalized method, particularly logF(1,1) could be used in practice when developing risk model for small or sparse data sets.