Towards a general asymptotic theory for Cox model with staggered entry (Q1359419)
From MaRDI portal
scientific article
Language | Label | Description | Also known as |
---|---|---|---|
English | Towards a general asymptotic theory for Cox model with staggered entry |
scientific article |
Statements
Towards a general asymptotic theory for Cox model with staggered entry (English)
0 references
31 October 1999
0 references
The work of several authors on proportional hazards models is generalized under more practical and complex conditions. The authors consider the following situation: There are potentially infinitely many individuals, with whom are attached entry times \(\tau_i\) \((\geq 0)\), failure times \(T_i\) \((\geq 0)\), censoring times \(C_i\) \((\geq 0)\), and \(p\times 1\) covariate vector processes \(Z_i\). Let \((\tau_i,T_i, C_i,Z_i')\) be independent random vectors and suppose that the conditional hazard rate of \(T_i\) at \(s\), given \(\tau_i\), \(C_i\), and \(Z_i(u)\), \(u\leq s\), is \(\exp (\beta'Z_i(s) \lambda_0(s))\), where \(\lambda_0\) is the baseline hazard function and \(\beta\) is an unknown parameter vector. Also, suppose that only \(n\) of the \(\tau_i\)'s \((i=1,2,\dots,n)\) are being sampled, and \(\tau_i\) need not be ordered. Moreover, let \(t\) be the calendar time and \(s\) the survival time \((t\geq s)\), and the \(i\)th individual's failure time \(T_i\) be censored by \(C_i\wedge(t-\tau_i)^+\) where \(\wedge\) denotes minimum and + is the positive part. Let \(\overline T_i(t)=T_i\wedge C_i\wedge(t-\tau_i)^+\), \(N_i (t, s)=I\{T_i\leq C_i\wedge (t-\tau_i)^+ \wedge s\}\), and \(Y_i(t,s)= I\{ \overline T_i(t)\geq s\}\), where \(I\) is the usual indicator function. Let \[ \overline Z( \beta;t,s)= \sum^n_{i=1}Z_i(s) \exp\bigl(\beta'Z_i(s) \bigr) Y_i(t,s)/ \sum^n_{i =1}\exp \bigl(\beta'Z_i(s)\bigr) Y_i(t,s), \] and let a two-parameter score process be defined by \[ U(\beta;t,s) =\sum^n_{i=1} \int^s_0 \bigl[Z_i(u)- \overline Z(\beta;t,u)\bigr]N_i(t,du). \] The authors derive functional limit theorems for the basic process \(n^{-1/2}U(\beta;\,,\,)\) under the null hypothesis \((\beta=\beta_0)\) and the contiguous alternatives \((\beta= \beta_0+b/ \sqrt n)\), and for the corresponding maximum partial likelihood estimator. These results extend the works of \textit{T. Sellke} and \textit{D. Siegmund} [Biometrika 70, 315--326 (1983; Zbl 0528.62068)] and \textit{M. G. Gu} and \textit{T. L. Lai} [Ann. Stat. 19, No. 3, 1403--1433 (1991; Zbl 0762.62020)].
0 references
censoring
0 references
score function
0 references
functional central limit theorems
0 references
proportional hazards
0 references
partial likelihood
0 references
0 references
0 references