Exploratory subgroup identification in the heterogeneous Cox model: A relatively simple procedure.

For survival analysis applications we propose a novel procedure for identifying subgroups with large treatment effects, with focus on subgroups where treatment is potentially detrimental. The approach, termed forest search, is relatively simple and flexible. All-possible subgroups are screened and selected based on hazard ratio thresholds indicative of harm with assessment according to the standard Cox model. By reversing the role of treatment one can seek to identify substantial benefit. We apply a splitting consistency criteria to identify a subgroup considered "maximally consistent with harm." The type-1 error and power for subgroup identification can be quickly approximated by numerical integration. To aid inference we describe a bootstrap bias-corrected Cox model estimator with variance estimated by a Jacknife approximation. We provide a detailed evaluation of operating characteristics in simulations and compare to virtual twins and generalized random forests where we find the proposal to have favorable performance. In particular, in our simulation setting, we find the proposed approach favorably controls the type-1 error for falsely identifying heterogeneity with higher power and classification accuracy for substantial heterogeneous effects. Two real data applications are provided for publicly available datasets from a clinical trial in oncology, and HIV.

Model Checking Techniques for Assessing Functional Form Specifications in Censored Linear Regression Models.

In this paper we develop model checking techniques for assessing functional form specifications of covariates in censored linear regression models. These procedures are based on a censored data analog to taking cumulative sums of "robust" residuals over the space of the covariate under investigation. These cumulative sums are formed by integrating certain Kaplan-Meier estimators and may be viewed as "robust" censored data analogs to the processes considered by Lin, Wei & Ying (2002). The null distributions of these stochastic processes can be approximated by the distributions of certain zero-mean Gaussian processes whose realizations can be generated by computer simulation. Each observed process can then be graphically compared with a few realizations from the Gaussian process. We also develop formal test statistics for numerical comparison. Such comparisons enable one to assess objectively whether an apparent trend seen in a residual plot reects model misspecification or natural variation. We illustrate the methods with a well known dataset. In addition, we examine the finite sample performance of the proposed test statistics in simulation experiments. In our simulation experiments, the proposed test statistics have good power of detecting misspecification while at the same time controlling the size of the test.

Estimation of treatment effects in weighted log-rank tests.

Non-proportional hazards have been observed in clinical trials. The log-rank test loses power and the standard Cox model generally produces biased estimates under such conditions. Weighted log-rank tests have been utilized to increase the test power; however, it is not intuitive how to interpret the test result in terms of the clinical effect. We propose a Cox-model based time-varying treatment effect estimate to complement the weighted log-rank test. The score test from the proposed model is equivalent to the weighted log-rank test, and a time-profile of the treatment effect can be obtained by fitting a time-varying covariate Cox model. Simulation results show that the proposed model preserves type-I error and achieve higher power than log-rank tests under non-proportional hazards scenarios. Whereas the standard Cox model produces biased effect estimates, the proposed model produces unbiased estimates if the weight function is correctly specified. It also achieves a better model fit and an enhanced flexibility to accommodate non-proportional hazards compared to the standard Cox model. The proposed approach makes the assumptions of the weighted log-rank test explicit and the validity of assumptions can be assessed based on prior knowledge or model goodness-of-fit. It also helps to translate the weighted log-rank test results into quantitative estimates of the treatment effect with intuitive interpretation. The proposed method can be routinely conducted to complement weighted log-rank tests, especially in the setting where non-proportional hazards are expected.

Functional form diagnostics for Cox's proportional hazards model.

We propose a new type of residual and an easily computed functional form test for the Cox proportional hazards model. The proposed test is a modification of the omnibus test for testing the overall fit of a parametric regression model, developed by Stute, González Manteiga, and Presedo Quindimil (1998, Journal of the American Statistical Association93, 141-149), and is based on what we call censoring consistent residuals. In addition, we develop residual plots that can be used to identify the correct functional forms of covariates. We compare our test with the functional form test of Lin, Wei, and Ying (1993, Biometrika80, 557-572) in a simulation study. The practical application of the proposed residuals and functional form test is illustrated using both a simulated data set and a real data set.