Regression analysis of multivariate interval-censored failure time data with informative censoring

Regression analysis of multivariate interval-censored failure time data has been discussed by many authors1-6. For most of the existing methods, however, one limitation is that they only apply to the situation where the censoring is non-informative or the failure time of interest is independent of the censoring mechanism. It is apparent that this may not be true sometimes and as pointed out by some authors, the analysis that does not take the dependent censoring into account could lead to biased or misleading results7,8. In this study, we consider regression analysis of multivariate interval-censored data arising from the additive hazards model and propose an estimating equation-based approach that allows for the informative censoring. The method can be easily implemented and the asymptotic properties of the proposed estimator of regression parameters are established. Also we perform a simulation study for the evaluation of the proposed method and it suggests that the method works well for practical situations. Finally, the proposed approach is applied to a set of real data.

A Deep CNN Approach For Predicting Cumulative Incidence Based On Pseudo-Observations

Background: Prognostic models are of high relevance in many medical application domains. However, many common machine learning methods have not been developed for direct applicability to right-censored outcome data. Recently there have been adaptations of these methods to make predictions based on only structured data (such as clinical data). Pseudo-observations has been suggested as a data pre-processing step to address right-censoring in deep neural network. There is a theoretical backing for the use of pseudo-observations to replace the right-censored response outcome, and this allows for algorithms and loss functions designed for continuous, non-censored data to be used. Medical images have been used to predict time-to-event outcomes applying deep convolutional neural network (CNN) methods using a Cox partial likelihood loss function under the assumption of proportional hazard. We propose a method to predict the cumulative incidence from images and structured clinical data by integrating (or combining) pseudo-observations and convolutional neural networks.Results: The performance of the proposed method is assessed in simulation studies and a real data example in breast cancer from The Cancer Genome Atlas (TCGA). The results are compared to the existing convolutional neural network with Cox loss. Our simulation results show that our proposed method performs similar to or even outperforms the comparator, particularly in settings where both the dependent censoring and the survival time do not follow proportional hazards in large sample sizes. The results found in the application in the TCGA data are consistent with the results found in the simulation for small sample settings, where both methods perform similarly. Conclusions: The proposed method facilitates the application of deep CNN methods to time-to-event data and allows for the use of simple and easy to modify loss functions thus contributing to modern image-based precision medicine.

KG 4/2015: A randomized, controlled, multicenter, open-label phase III clinical trial of GV1001 with gemcitabine/capecitabine in previous untreated, eotaxin-high patients with advanced pancreatic ductal adenocarcinoma.

4020 Background: In the TeloVac study, GV1001 with Gemcitabine/capecitabine (G/C) did not show increased overall survival (OS) than G/C in patients (pts) with advanced pancreatic ductal adenocarcinoma (PDA). But cytokine examination suggested high serum eotaxin level may predict improved survivals in pts received GV1001 with G/C. This phase III trial was designed to assess the efficacy of GV1001 with G/C for previous untreated eotaxin-high Korean pts with advanced PDA. Methods: Eligible pts with histologically proven locally advanced and metastatic PDA (except peritoneal carcinomatosis), age > 18 years, and ECOG PS 0–2 were recruited. Pts were randomly assigned (1:1) to receive either G/C or G/C with GV1001 (G/C/GV). All pts receiving G/C/GV were with high serum eotaxin level (≥81.02 ng/mL), and the pts receiving G/C were randomly assigned again (1:1) to eotaxin-high and eotaxin-low pts. Study was designed according to Korean MFDS guidance for approval of clinical trial. G/C treatment included G (1000 mg/m2, 30 min IVF, D 1, 8, & 15) and C (830 mg/m2 BID for 21 days per month (m). G/C/GV treatment included an intradermal injection of GM-CSF (75 μg) and GV1001 (0.56 mg; D 1, 3, & 5, once on week 2–4, & 6, then monthly thereafter) from the start of G/C. The primary endpoint was OS. The secondary endpoints included time to progression (TTP), objective response rate, and safety. Survival data was analyzed using the copula graphic estimate method under dependent censoring. The response was independently assessed per RECIST v1.1. Under the one-sided significance level of 2.5% and to achieve the power of 80% of the statistical significance with the median OS difference from 7.9 to 14.9 m (HR = 0.53), 85 events and 118 registrations needed. Considering 20% drop-outs, 148 registrations were required. Results: Between Nov 2015 and Apr 2020, of 511 pts screened in 16 centers, eotaxin-high pts were identified as 34.7% (174 / 502 pts). 148 pts randomly assigned to G/C/GV (n = 75; all eotaxine-high) and G/C (n = 73; 37 eotaxine-high, 36 eotaxine-low). Median OS was significantly improved in the G/C/GV group with 11.3m [95% CI 8.6-14.0] than G/C group with 7.5 m [95% CI 5.1-10.0] (p = 0.021). Also, median TTP was significantly improved in the G/C/GV group (7.3 m [95% CI 5.0-9.7]) than in the G/C group (4.5 m [95% CI 3.2-5.8], p = 0.021). In other secondary endpoints, no statistical significance was confirmed between the two groups. Grade 3-4 treatment-emergent adverse events were reported in 49 pts (73.13%) vs. 58 pts (77.33%) in the G/C and G/C/GV group, without significant differences (p = 0.562). Conclusions: G/C/GV treatments significantly extend OS and TTP in advanced PDA than G/C, and specific safety-related issues had not been found. GV1001 should be considered as one of the options in PDA pts with high serum eotaxin levels. Clinical trial information: NCT02854072.

On semiparametric modelling, estimation and inference for survival data subject to dependent censoring

When modelling survival data, it is common to assume that the survival time T is conditionally independent of the censoring time C given a set of covariates. However, there are numerous situations in which this assumption is not realistic. The goal of this paper is therefore to develop a semiparametric normal transformation model, which assumes that after a proper nonparametric monotone transformation, the vector (T, C) follows a linear model, and the vector of errors in this bivariate linear model follows a standard bivariate normal distribution with possibly non-diagonal covariance matrix. We show that this semiparametric model is identifiable, and propose estimators of the nonparametric transformation, the regression coefficients and the correlation between the error terms. It is shown that the estimators of the model parameters and the transformation are consistent and asymptotically normal. We also assess the finite sample performance of the proposed method by comparing it with an estimation method under a fully parametric model. Finally, our method is illustrated using data from the AIDS Clinical Trial Group 175 study.

Doubly robust estimator of risk in the presence of censoring dependent on time-varying covariates: application to a primary prevention trial for coronary events with pravastatin

Background: In the presence of dependent censoring even after stratification of baseline covariates, the Kaplan–Meier estimator provides an inconsistent estimate of risk. To account for dependent censoring, time-varying covariates can be used along with two statistical methods: the inverse probability of censoring weighted (IPCW) Kaplan–Meier estimator and the parametric g-formula estimator. The consistency of the IPCW Kaplan–Meier estimator depends on the correctness of the model specification of censoring hazard, whereas that of the parametric g-formula estimator depends on the correctness of the models for event hazard and time-varying covariates. Methods: We combined the IPCW Kaplan–Meier estimator and the parametric g-formula estimator into a doubly robust estimator that can adjust for dependent censoring. The estimator is theoretically more robust to model misspecification than the IPCW Kaplan–Meier estimator and the parametric g-formula estimator. We conducted simulation studies with a time-varying covariate that affected both time-to-event and censoring under correct and incorrect models for censoring, event, and time-varying covariates. We applied our proposed estimator to a large clinical trial data with censoring before the end of follow-up. Results: Simulation studies demonstrated that our proposed estimator is doubly robust, namely it is consistent if either the model for the IPCW Kaplan–Meier estimator or the models for the parametric g-formula estimator, but not necessarily both, is correctly specified. Simulation studies and data application demonstrated that our estimator can be more efficient than the IPCW Kaplan–Meier estimator. Conclusions: The proposed estimator is useful for estimation of risk if censoring is affected by time-varying risk factors.