A class of mean residual life regression models with censored survival data

Size: px
Start display at page:

Download "A class of mean residual life regression models with censored survival data"

Transcription

1 1 A class of mean residual life regression models with censored survival data LIUQUAN SUN Institute of Applied Mathematics, Academy of Mathematics and Systems Science, Chinese Academy of Sciences, Beijing, 18, P. R. China. slq@amt.ac.cn QIANG ZHAO Department of Mathematics Texas State University, San Marcos, Texas, 78666, U.S.A. qiang.zhao@txstate.edu Abstract When describing a failure time distribution, the mean residual life is sometimes preferred to the survival or hazard rate. Regression analysis making use of the mean residual life function has recently drawn a great deal of attention. In this paper, a class of mean residual life regression models are proposed for censored data, and estimation procedures and a goodness-of-fit test are developed. Both asymptotic and finite sample properties of the proposed estimators are established, and the proposed methods are applied to a cancer data set from a clinic trial. Keywords: Censored data; Estimating equation; Failure time; Goodness-of-fit test; Mean residual life. Corresponding author: Liuquan Sun, Institute of Applied Mathematics, Academy of Mathematics and Systems Science, Chinese Academy of Sciences, Beijing 18, P.R.China. slq@amt.ac.cn.

2 2 1. INTRODUCTION The mean residual life function (MRLF) is of interest in many fields such as reliability, survival analysis, actuarial studies, etc. For example, it is sometimes more informative to tell a prostate cancer patient how long he can survive or live without disease recurrence, in expectation, given his current situation (which of course includes the fact that he has survived or lived without the disease so far). As another example, a customer may be interested in knowing how much longer his or her computer can be used, given that the computer has worked normally for, say, t years. For a nonnegative survival time T with finite expectation, the MRLF at time t is m(t) = E(T t T > t). To assess the effects of covariates on the mean residual life, the proportional mean residual life model by Oakes and Dasu (199) may be used: m(t Z) = m (t) exp(β Z), (1) where m(t Z) is the MRLF corresponding to the p-vector covariate Z, m (t) is some unknown baseline MRLF when Z =, and β is an unknown vector of regression parameters. Previous work on the MRLF has focused on single-sample and two-sample cases (Oakes and Dasu, 199). For regression analysis, Maguluri and Zhang (1994) used the underlying proportional hazards structure of the model to develop estimation procedures for β in model (1), and Yuen, Zhu and Tang (23) proposed a goodness-of-fit test for model (1), when there was no censoring involved. In the presence of censoring, Chen and Cheng (25) used counting process theory to develop semiparametric inference procedures for β in model (1), and Chen, et al. (25) extended an estimation procedure of Maguluri and Zhang (1994) to censored survival data using inverse probability of censoring weighting techniques (Robins and Rotnitzky, 1992). Recently, Chen and Cheng (26) and Chen (27) proposed a new

3 3 class of additive mean residual life model and discussed various estimation methodologies with or without right censoring. However, other regression forms may be more natural or descriptive in some applications. In this paper, we consider a more general class of mean residual life regression models given by m(t Z) = m (t)g(β Z), (2) where g(t) is pre-specified and assumed to be continuous almost everywhere and twice differentiable. Examples of possible link function include g(x) = 1 + x, g(x) = e x and g(x) = log(1 + e x ). Selection of an appropriate link function may be based on prior data or the resulting interpretation of the regression parameters. In the next section, we will first discuss the situation where the censoring time is independent of T and Z, and a general inference procedure based on estimating functions is proposed. The procedure can be easily implemented numerically and the asymptotic properties of the proposed estimates of regression parameters are established. Section 3 generalizes the methods to the situation where the censoring time may depend on Z through the proportional hazards model. In Section 4, we develop test procedures for checking the adequacy of model (2) under both independent and dependent censoring scenarios based on an appropriate stochastic process which is asymptotically Gaussian. Section 5 reports some results from simulation studies conducted for evaluating the proposed methods. In Section 6, we apply the methodology to a data set from a cancer clinic trial and some concluding remarks are given in Section INFERENCE WITH INDEPENDENT CENSORING TIMES In this section, let C be the potential censoring time, and assume that C is independent of T and Z. To avoid lengthy technical discussion of the tail behaviour of the limiting

4 4 distributions, we further assume that P rc τ} >, where < τ = inft : P r(t t) = } <. Let T i, C i, Z i } (i = 1,..., n) be independent replicates of T, C, Z} and suppose that we observe X i, δ i, Z i ; i = 1,..., n}, where X i = min(t i, C i ), δ i = I(T i C i ). Here I( ) is the indicator function. Define M i (t) = N i (t) t Y i (u)dλ(u Z i ), i = 1,..., n, (3) where N i (t) = I(X i t, δ i = 1), Y i (t) = I(X i t), and Λ(t Z i ) is the cumulative hazard function of T i given Z i. It is well known that M i (t) (i = 1,..., n) are zero-mean martingale with respect to the σ-filtration σn i (u), Y i (u+), Z i : u t, i = 1,..., n}. Note that the survival function of T given Z is S(t Z) = m( Z) m(t Z) exp Then under model (2), we have t } du. m(u Z) m (t)dλ(t Z i ) = g(β Z i ) 1 dt + dm (t). (4) Thus, in view of (3) and (4), for given β, a reasonable estimator for m (t) is the solution to [ m (t)dn i (t) Y i (t) g(β Z i ) 1 dt + dm (t) } =, t τ. (5) Denote this estimator by ˆm a (t; β). Straightforward algebra on (5) leads to ˆm a (t; β) = Φ n (t) 1 t Φ n (u) Y i(u)g(β Z i ) 1 n Y du, (6) i(u) where Φ n (t) = exp t dn i(u)/ Y i(u)}, which is the Nelson-Aalen estimator of the survival function for the pooled observations with independent censoring times. estimate β, using the generalized estimating equation methods (Liang and Zeger, 1986; Cai and Schaubel, 24; Chen and Cheng, 25), we propose the following class of estimating equations for β, g (1) (β Z i ) g(β Z i ) Z [ i ˆma (t; β)dn i (t) Y i (t) g(β Z i ) 1 dt + d ˆm a (t; β) } =, To

5 5 where g (1) (x) = dg(x)/dx. In view of (5), the above estimating equations are equivalent to U a (β) = n 1 where h(x) = g (1) (x)/g(x), and h(β Z i )Z i Z a (t; β) } [ ˆm a (t; β)dn i (t) Y i (t)g(β Z i ) 1 dt =, (7) Z a (t; β) = Y i(t)h(β Z i )Z i n Y. i(t) Let ˆβ a denote the solution to U a (β) = and ˆm a (t) ˆm a (t; ˆβ a ) the corresponding estimator of the unknown baseline mean residual life m (t). Following the arguments of Chen and Cheng (25) and Lin, Wei and Ying (21), we can check that both ˆβ a and ˆm a (t) always exist and are unique and consistent. To study the asymptotic distribution of ˆβ a, we show in Appendix A.1 that n 1/2 U a (β ) is asymptotically normal with mean zero and covariance matrix that can be consistently estimated by ˆΣ a, where ˆΣ a = n 1 ˆµ(t) = Z a (t; ˆβ a ) + Φ n(t) π n (t) h( ˆβ a Z i )Z i ˆµ(t) } 2 ˆma (t) 2 dn i (t), t n 1 [ h( ˆβ az i )Z i Z a (u; ˆβ dni (u) a ) Φ n (u), and π n (t) = n 1 Y i(t). Here for a vector v, v = 1, v 1 = v and v 2 = vv. Then it follows that n 1/2 ( ˆβ a β ) is asymptotically normal with zero mean and covariance matrix that can be consistently estimated by  1 ˆΣa  1, where  = n 1 h( ˆβ az i )Z i Z a (t; ˆβ } 2 a ) Yi (t)g( ˆβ az i ) 1 dt. We also show in Appendix A.2 that n 1/2 ˆm a (t) m (t)} ( t τ) converges weakly to a zero-mean Gaussian process whose covariance function at (s, t) can be estimated consistently by ˆΓ a (s, t) = n 1 n ˆϕ i(s) ˆϕ i (t), where ˆϕ i (t) = 1 Φ n (u) [ ˆm a (u)dn i (u) Y i (u) g( Φ n (t) t π n (u) ˆβ az i ) 1 du + d ˆm a (u) } + ˆm a (t) Z a (t; ˆβ a )  1 h( ˆβ a Z i )Z i ˆµ(u) }[ ˆm a (u)dn i (u)

6 6 Y i (u) g( ˆβ az i ) 1 du + d ˆm a (u) }. The asymptotic normality for ˆm a (t), together with the consistent variance estimator ˆΓ a (t, t), enables us to construct pointwise confidence intervals for m (t). Since m (t) is nonnegative, one may want to use the log transformation for the construction of its confidence intervals. To construct simultaneous confidence bands for m (t) over a time interval of interest [t 1, t 2 ( < t 1 < t 2 τ), we need to evaluate the distribution of the supremum of a related process over [t 1, t 2. It is not possible to evaluate such distributions analytically because the limiting process of n 1/2 ˆm a (t) m (t)} does not have an independent increments structure. To handle this problem, we use a resampling scheme to approximate the distribution of n 1/2 ˆm a (t) m (t)}. Define Ŵ a (t) = n 1/2 ˆϕ i (t)ω i, where (Ω 1,..., Ω n ) are independent standard normal variables which are independent of the data X i, δ i, Z i ; i = 1,..., n}. According to the arguments of Lin et al. (2), the distribution of the process n 1/2 ˆm a (t) m (t)} can be approximated by that of the zero-mean Gaussian process Ŵa(t). To approximate the distributions of n 1/2 ˆm a (t) m (t)}, we obtain a large number of realizations from Ŵa(t) by repeatedly generating the normal random sample (Ω 1,..., Ω n ) while fixing the data X i, δ i, Z i ; i = 1,..., n} at their observed values. Using this simulation method, we may determine an approximate 1 α simultaneous confidence bands for m (t) over a time interval of interest [t 1, t INFERENCE WITH DEPENDENT CENSORING TIMES Now we consider the situation where T, C and Z may depend on each other, but given Z, we assume that T is independent of C. Also we assume that the hazard function of C given Z has the form λ c (t Z) = λ (t) expγ Z}, (8)

7 7 where λ (t) is an unspecified baseline hazard function and γ is a vector of unknown regression parameters. Note that the model from Section 2 is the model of Section 3 with γ =. Of course, γ is usually unknown. A natural estimate of γ, which is efficient under model (8), is given by the maximum partial likelihood estimate defined as the solution to U r (γ) = Z i Z r (t; γ)}dn c i (t) = (9) (Cox, 1972), where Ni c (t) = I(X i t, δ i = ), and Z r (t; γ) = S (1) (t; γ)/s () (t; γ), S (k) (t; γ) = Y i(t)z k i expγ Z i } for k =, 1, 2. Let ˆγ denote the estimator given by U r (γ) =, and ˆΛ (t) be the Breslow estimator of Λ (t) = t λ (u)du, where ˆΛ (t) = t dn c i (u) Y i(u) expˆγ Z i }. Consider a hypothetical equilibrium renewal process formed by renewals following the same survival distribution as S(t Z). The forward recurrence time V is defined as the time from a fixed time to the next immediate renewal. Then under model (2), it follows from Cox (1962) that its hazard function is λ v (t Z) = m(t Z) 1 = m (t) 1 g(β Z) 1, which is a proportional hazards model. When there is no censoring, the following partial score equation can be used to estimate β (Prentice and Self, 1983; Cai and Schaubel, 24), Êh(β Z)Z} Ê[h(β Z)Zg(β Z) 1 I(V t) d ˆF v (t) =, (1) Ê[g(β Z) 1 I(V t) where Ê and ˆF v (t) are their empirical estimates of the expectation E and F v (t), respectively. Here F v (t) is the distribution function of V. However, this equality is only theoretical, since we cannot observe V. To use the sample of T s in (1), following the arguments of Maguluri and Zhang (1994) and Cheng et al. (25), we have that for any function w(z), Ew(Z)I(V t)} = m () 1 Ew(Z)g(β Z) 1 (T t) + },

8 8 where (T t) + denotes (T t)i(t t). As a result, df v (t) = Eg(β Z) 1 I(T > t)} dt. Eg(β Z) 1 T } Replacing the respective terms in (1), we obtain the following estimating equation for β based on T s, n 1 h(β Z i )Z i h(β Z i )Z i g(β Z i ) 2 (T i t) + g(β Z i ) 2 (T i t) + g(β Z i ) 1 I(T i > t) g(β Z i ) 1 T i dt =. (11) Let G i (t; γ, Λ ) be the censoring survival distribution of C i given Z i under model (8), that is, G i (t; γ, Λ ) = exp Λ (t) exp(γ Z i ) }. Then for any well-defined function of ν, } [ ν(xi, Z i, t)δ i ν(ti, Z i, t)δ } i E = E E Z i = Eν(T i, Z i, t)}. (12) G i (X i ; γ, Λ ) G i (T i ; γ, Λ ) In view of (11) and (12), using inverse probability of censoring weighting techniques (Robins and Rotnitzky, 1992), we propose the following class of estimating equations for β when the censoring time C i may depend on Z i under model (8), where U b (β) = n 1 Z i (β, γ, Λ) = L n (t; β, γ, Λ) = δ i G i (X i ; ˆγ, ˆΛ h(β Z i )Z i Z i (β, ˆγ, ˆΛ } ) =, (13) ) h(β Z i )Z i g(β Z i ) 2 (X i t) + L n (t; β, γ, Λ)dt, L 1n (t; β, γ, Λ) L 2n (t; β, γ, Λ)L 3n (t; β, γ, Λ), and L kn (t; β, γ, Λ) = n 1 V ki(t; β)g i (X i ; γ, Λ) 1, k = 1, 2, 3. Here V 1i (t; β) = g(β Z i ) 1 I(X i > t)δ i, V 2i (t; β) = g(β Z i ) 2 (X i t) + δ i, and V 3i (t; β) = g(β Z i ) 1 X i δ i. Let ˆβ b denote the solution to U b (β) =. It can be shown in Appendix A.3 that ˆβ b is consistent and unique in a neighborhood of β. To study the asymptotic distribution of ˆβ b,

9 9 we first show that n 1/2 U b (β ) is asymptotically normal with zero mean and covariance matrix that can be consistently estimated by ˆΣ b, where ˆΣ b = n 1 [ ˆξ i + R n (t) τ S () (t; ˆγ) d ˆM i c (t) + B n Dn 1 Zi Z r (t; ˆγ) } 2 d ˆM i c (t), (14) ˆξ i = h( ˆβ bz i )Z i Z i ( ˆβ b, ˆγ, ˆΛ } ) δ i G i (X i ; ˆγ, ˆΛ ) 1 [ τ ˆξ 1i (t) Q n (t) ˆL 2n (t)ˆl 3n (t) ˆL 1n (t)ˆξ 2i (t) ˆL 2n (t) 2 ˆL 3n (t) R n (t) = n 1 B n = n 1 Q n (t) n 1 Q n (t) = n 1 ˆL 1n (t)ˆξ 3i (t) ˆL 2n (t)ˆl 3n (t) 2 h( ˆβ bz i )Z i Z i ( ˆβ b, ˆγ, ˆΛ } ) expˆγ Z i }δ i G i (X i ; ˆγ, ˆΛ ) 1 Y i (t) [ R 1n (t, u) ˆL 2n (t)ˆl 3n (t) ˆL 1n (t)r 2n (t, u) ˆL 2n (t) 2 ˆL 3n (t) ˆL 1n (t)r 3n (t, u) du, ˆL 2n (t)ˆl 3n (t) 2 h( ˆβ bz i )Z i Z i ( ˆβ b, ˆγ, ˆΛ ) }ˆΛ (X i ) expˆγ Z i }Z iδ i G i (X i ; ˆγ, ˆΛ ) 1 h( ˆβ bz i )Z i Z i ( ˆβ b, ˆγ, ˆΛ } ) expˆγ Z i }δ i G i (X i ; ˆγ, ˆΛ ) 1 Y i (t) Z r (t; ˆγ)} dˆλ (t) [ Q n (t) P 1n (t) ˆL 2n (t)ˆl 3n (t) ˆL 1n (t)p 2n (t) ˆL 2n (t) 2 ˆL 3n (t) ˆL 1n (t)p 3n (t) ˆL 2n (t)ˆl 3n (t) 2 h( ˆβ bz i )Z i g( ˆβ bz i ) 2 (X i t) + δ i G i (X i ; ˆγ, ˆΛ ) 1, ˆξ ki (t) = V ki (t; ˆβ b )G i (X i ; ˆγ, ˆΛ ) 1 ˆL kn (t), k = 1, 2, 3, R kn (t, u) = n 1 V ki (t; ˆβ b )G i (X i ; ˆγ, ˆΛ ) 1 expˆγ Z i }Y i (u), P kn (t) = n 1 V ki (t; ˆβ b )G i (X i ; ˆγ, ˆΛ ) 1 ˆΛ (X i ) expˆγ Z i }Z i ˆM c i (t) = N c i (t) t R kn (t, u) Z r (u; ˆγ) dˆλ (u), Y i (u) expˆγ Z i }dˆλ (u), dt, dt,

10 1 ˆL kn (t) = L kn (t; ˆβ b, ˆγ, ˆΛ ), and D n = U r (ˆγ)/ γ. Then it follows that n 1/2 ( ˆβ b β ) is asymptotically normal with zero mean and covariance matrix that can be consistently estimated by U b ( ˆβ } 1 b ) U b ( ˆβ } 1 b ) β ˆΣb. β To estimate the baseline mean residual life m (t), define Mi (t) = δ ii(x i > t) [ (X i t) m (t)g(β G(X i ; γ, Λ ) Z i ), i = 1,..., n. Under models (2) and (8), M i (t) are zero-mean stochastic processes. Thus, for given β, a reasonable estimator for m (t) is the solution to δ i I(X i > t) G i (X i ; ˆγ, ˆΛ ) [ (X i t) m (t)g(β Z i ) =, t τ. Denote this estimator by ˆm b (t; β), which can be expressed as ˆm b (t; β) = (X i t) + δ i G i (X i ; ˆγ, ˆΛ ) 1 I(X i > t)g(β Z i )δ i G i (X i ; ˆγ, ˆΛ. (15) ) 1 Let ˆm b (t) ˆm b (t; ˆβ b ) be the corresponding estimator of the unknown baseline mean residual life m (t) under models (2) and (8). Following the arguments of Appendix A.2 and A.3, we can check that ˆm b (t) is consistent, and that n 1/2 ˆm b (t) m (t)} ( t τ) converges weakly to a zero-mean Gaussian process whose covariance function at (s, t) can be estimated consistently by ˆΓ b (s, t) = n 1 n ˆψ i (s) ˆψ i (t), where ˆψ i (t) = ˆm b (t) Z b (t; ˆβ b ) U b ( ˆβ } 1 [ b ) ˆξ β i + B n Dn 1 Z b (t; β) = + R n (u) S () (u; ˆγ) d ˆM i c (u) + Ψ n (t; ˆβ b ) 1[ ˆM i (t) + Z i Z r (u; ˆβ b )}d ˆM i c (u), +B n(t)d 1 n I(X i > t)h(β Z i )Z i g(β Z i )δ i G(X i ; ˆγ, ˆΛ ) 1, nψ n (t; β) Z i Z r (u; ˆβ b )}d ˆM c i (u) r n (t, u) S () (u; ˆγ) d ˆM c i (u)

11 11 Ψ n (t; β) = n 1 I(X i > t)g(β Z i )δ i G i (X i ; ˆγ, ˆΛ ) 1, [ ˆM i (t) = I(X i > t) (X i t) ˆm b (t)g( ˆβ bz i ) δ i G i (X i ; ˆγ, ˆΛ ) 1, r n (t, u) = n 1 ˆM i (t) expˆγ Z i }Y i (u), Bn(t) = n 1 ˆM i (t)ˆλ (X i ) expˆγ Z i }Z i t r n (t, u) Z r (u; ˆγ) dˆλ (u), and ˆξ i, B n, R n (u) and D n are defined in (14). Note that the limiting process of n 1/2 ˆm b (t) m (t)} is quite complicated, and its properties are difficult to obtain analytically. As discussed in Section 2, we can show that the distribution of the process n 1/2 ˆm b (t) m (t)} can be approximated by that of the zero-mean Gaussian process Ŵb(t), where Ŵ b (t) = n 1/2 ˆψ i (t)ω i, and (Ω 1,..., Ω n ) are independent standard normal variables which are independent of the data X i, δ i, Z i ; i = 1,..., n}. 4. MODEL CHECKING TECHNIQUES In this section, we develop testing procedures to check the adequacy of model (2) for both independent and dependent cases. Beginning with the independent case where censoring time C is independent of T and Z, let G(t) be the survival function of C, and Ĝ(t) be the Kaplan-Meier estimate of G(t) based on X i, 1 δ i, i = 1,..., n}, where Ĝ(t) = s t 1 dn } i c (s) Y. i(s)

12 12 Define H 1 (t, z) = P X i t, Z i z, δ i = 1}, and H(t, z) = P X i t, Z i z}, where the notation Z i z means that each component of Z i is less than or equal to the corresponding component of z. After some algebraic manipulation, model (2) leads to m (t) = 1 τ z H(t, z) t (s t)g(t) g(β w)g(s) H 1(ds, dw), (16) where z stands for z 1... z p. Let us denote the right-hand side of (16) by V (t, z). Note that the left-hand side is independent of the variable z. As a measure of fit for model (2), we estimate V (t, z) by V n (t, z) and obtain the process θ n (t, z) = n 1/2 V n (t, z) V n (t, z u )}, (17) where z u is the vector of upper bounds for Z, 1 τ V n (t, z) = H n (t, z) t z } (s t)ĝ(t) g( ˆβ aw)ĝ(s)h 1n(ds, dw), and H n and H 1n are the empirical counterparts of H and H 1, respectively. That is, H n (t, z) = n 1 I(X i t, Z i z) and H 1n (t, z) = n 1 I(X i t, Z i z, δ i = 1). Under model (2), the process θ n (t, z) equals φ n (t, z) φ n (t, z u ), where φ n (t, z) = n 1/2 V n (t, z) V (t, z)} is the standardized mean residual life process. Hence, based on (17), the Kolmogorov-Smirnov (KS) type test statistic F (1) n may be used to check the adequacy of model (2), where F (1) n = sup θ n (t, z). t,z Under model (2), we show in Appendix A.4 that θ n (t, z) converges to a zero-mean Gaussian process W (t, z) whose covariance function at (t, z) and (t, z ) can be estimated consistently by ˆσ(t, z; t, z ) = n 1 n ˆη i(t, z)ˆη i (t, z ), where ˆη i (t, z) = ˆρ i (t, z) ˆρ i (t, z u ), ˆρ i (t, z) = Ĝ(t) [ τ z s t H n (t, z) t u Ĝ(s)g( ˆβ aw) H d M i c (u) 1n(ds, dw) π n (u) δ i (X i + t)ĝ(t) H n (t, z)ĝ(x i)g( ˆβ az i ) I(X i t, Z i z) V n(t, z) H n (t, z) I(X i t, Z i z)

13 13 z + Ĝ(t) (s t)h( ˆβ aw)w H n (t, z) t Ĝ(s)g( ˆβ H 1n (ds, aw) dw)â 1 h( ˆβ a Z i )Z i ˆµ(u) } [ ˆm a (u)dn i (u) Y i (u) g( ˆβ az i ) 1 du + d ˆm a (u) }, (18) d ˆm a (u) = j=1 [ ˆm a(u)dn j (u) Y j (u)g( ˆβ Z j ) 1 du j=1 Y, j(u) d M c i (u) = dn c i (u) Y i (u)dλ c n(u), and dλ c n(u) = n 1 dn c i (u)/π n (u). Consequently, F n converges in distribution to F, where F = sup W (t, z). t,z Obviously, the complicated structure of the covariance function (18) does not allow for an analytic treatment of the involved distributions. As discussed in Sections 2 and 3, we can show that the distribution of the process W (t, z) can be approximated by that of the zero-mean Gaussian process W (t, z), where W (t, z) = n 1/2 ˆη i (t, z)ω i, and (Ω 1,..., Ω n ) are independent standard normal variables which are independent of the data X i, δ i, Z i ; i = 1,..., n}. Thus, the distributions of F can be approximated by F, where F = sup W (t, z). t,z To approximate the distribution of F, we obtain a large number, say M, of realizations from F by repeatedly generating the normal random sample (Ω 1,..., Ω n ) while fixing the data X i, δ i, Z i ; i = 1,..., n} at their observed values. Then using this simulation method, we may determine an approximate critical value of the test. Specifically, the p-value of the test can be obtained as follows, p = 1 M M I( F k > F n ), k=1 where F k (k = 1,..., M) are M realizations from F.

14 14 For the dependent case where C depends on Z, an analogous procedure can be developed. Let G(t z) be the censoring survival distribution of C given Z = z, and Ĝ(t z) = exp ˆΛ (t) exp(ˆγ z) }. After some algebraic manipulation, model (2) leads to m (t) = 1 τ z H(t, z) t (s t)g(t w) g(β w)g(s w) H 1(ds, dw). (19) Let us denote the right-hand side of (19) by V (t, z). Note again that the left-hand side is independent of the variable z, and V (t, z) can be estimated by Vn (t, z), where Vn 1 τ } z (s (t, z) = t)ĝ(t w) H n (t, z) t g( ˆβ b w)ĝ(s w)h 1n(ds, dw). Similarly, for check the adequacy of model (2) under the dependent case, we use the Kolmogorov- Smirnov type test statistic F n (2), where F (2) n and θ n(t, z) = n 1/2 V n (t, z) V n (t, z u )}. = sup θn(t, z), t,z Under model (2), we can also show that θ n(t, z) converges to a zero-mean Gaussian process W (t, z) whose covariance function at (t, z) and (t, z ) can be estimated consistently by ˆσ (t, z; t, z ) = n 1 n ˆη i (t, z)ˆη i (t, z ), where ˆη i (t, z) = ˆρ i (t, z) ˆρ i (t, z u ), [ ˆρ 1 τ z (s t) expˆγ i (t, z) = w}ĝ(t w) d H n (t, z) t u g( ˆβ b w)ĝ(s w) H 1n (ds, dw) ˆM i c (u) S () (u; ˆγ) + 1 z (s t) expˆγ w}ĝ(t w) [ s H n (t, z) t g( ˆβ b w)ĝ(s w) (w Z r (v; ˆβ b ) dˆλ (v) t H 1n (ds, dw)d 1 n Zi Z r (u; ˆβ b ) } d ˆM c i (u) δ i (X i + t)ĝ(t Z i) H n (t, z)ĝ(x i Z i )g( ˆβ b Z i) I(X i t, Z i z) V n(t, z) H n (t, z) I(X i t, Z i z)

15 + 1 H n (t, z) [ ˆξ i + Consequently, F (2) n z t (s t)h( ˆβ b w)ĝ(t w)w U b ( Ĝ(s w)g( ˆβ b w) H 1n (ds, dw) ˆβ } 1 b ) β Zi Z r (u; ˆβ b ) } d ˆM i c (u). (2) R n (u) S () (u; ˆγ) d ˆM i c (u) + B n Dn 1 converges in distribution to F = sup t,z W (t, z). As in the independent case, we can show that the distribution of the process W (t, z) can be approximated by that of the zero-mean Gaussian process W (t, z) = n 1/2 ˆη i (t, z)ω i based on (2). Thus, the distributions of F can be approximated by F = sup t,z W (t, z), and the p-value of the test can be obtained in the same way as before SIMULATION STUDIES We conducted simulation studies to assess the performance of the estimation procedure proposed in Sections 2 and 3 with the focus on estimating β. In the study, the survival time T was generated from model (2) with β = or.5, and the baseline mean residual life function was taken to be m (t) =.5t + 1, which corresponds to a rescaled beta distribution (Oakes and Dasu, 199). The covariate Z was assumed to be a Bernoulli random variable with success probability.5. We considered three choices for the link function g(x): g 1 (x) = 1 + x, g 2 (x) = e x and g 3 (x) = log(1 + e x ). The censoring time C was generated from the exponential distribution with hazard rate λ e γz for γ = or 1, and λ was chosen to result in two censoring percentages of approximately 1% and 3%. Note that γ = corresponds to independent censoring times, while γ = 1 gives dependent censoring times. The results presented below are based on n = 1 or 2 with 2 replications. Table 1 shows the results for independent censoring (γ = ). It can been seen that the bias for estimating β is very small and the standard error of the estimator is very accurate for all settings. The 95% empirical coverage probability based on normal approximation are

16 16 reasonable, and the results become better when the sample size increases from 1 to 2. Table 2 shows similar results for dependent censoring (γ = 1). To investigate the asymptotical normality of the proposed estimates of β under both independent and dependent censoring, we provide some QQ-plots in Figure 1, which suggest reasonable normal approximations to the finite-sample distributions of the proposed estimators. We also considered other models and set-ups and obtained similar results. 6. AN APPLICATION We applied the proposed estimation procedures to a data set from a clinic trial on lung cancer that has previous been analyzed by others (Lad et al., 1988, Piantadosi, 25, and Chen et al. 25). The purpose of the trial is to assess the impact of systematic combination chemotherapy on patients survival. Specifically, survival time of interest includes both time to death and disease-free survival time. Between November 1979 and May 1985, 172 patients were randomized to receive either postoperative radiotherapy (RT) alone or postoperative RT plus chemotherapy with Cytoxan, Adriamycin, and Platinol (RT + CAP) for 6 months and followed until death. The mean follow-up time is 1.5 years. Only 164 patients were eligible for analysis, among which 86 patients were in RT and 78 in RT + CAP group. In our analysis, we consider examining the effect of treatment and cell type (squamous vs. nonsquamous/mixed) on patients disease-free survival. For treatment, we let Z 1 = 1 if the patient is in RT + CAP group and otherwise. For cell type, we let Z 2 = 1 if the patient had the squamous cell type and otherwise. We first fit model (8) containing both covariates to the data to determine whether dependent or independent case should be considered. The logrank test shows that the overall effect of treatment and cell type on the censoring time is insignificant with a p-value of.876. The Kaplan-Meier estimates of survival functions of the censoring time for four subgroups were plotted in Figure 2 (a). Thus, for the illustration

17 17 purpose, we then fit model (2) to the data only under the independent censoring situation. Table 3 shows that the estimation and test of hypothesis results for the effect of each of the covariates by using three different functions for g. The results show that both treatment and cell type have significant effect on the patients disease-free survival after adjusting the effect of the other. More specifically, patients in RT + CAP group have significantly longer mean residual disease-free life than those in the RT group, and patients having squamous cell type have significantly longer mean residual disease-free life than those having nonsquamous/mixed cell type. Figure 2 (b) and (c) show the difference in survival functions between the treatment groups and two cell type groups, respectively. This is consistent with the results from Chen et al. (25) under the proportional mean residual life model and from Piantadosi (25) under the proportional hazards model. Note that the three functions for g yield similar results, and the result from g(x) = e x is the least conservative based on the p-values. We also checked the adequacy of model (2) with both covariate under the three functions of g(x). Based on 5 realizations of F, the KS-type test statistics with p-values in parentheses, are (.966), (.946) and (.958) for g(x) to be 1 + x, e x and log(1 + e x ), respectively. These results indicate that model (2) fits the data adequately. 7. CONCLUDING REMARKS In this article we have studied a class of mean residual life regression models under both independent and dependent censoring. The proposed models are generalization of the proportional mean residual life model with more choices of the link function g(x). Estimation procedures were proposed for the model parameters, and asymptotic properties of the estimators were derived. The methodology was applied to a cancer data set from a clinic trial, and the simulation results show that the proposed methods work well for the situations

18 18 considered. As it is well-known, model checking is always an important issue in regression analysis, because most regression models have limitations. We proposed a goodness of fit test for model (2) based on the KS type test statistics. In addition, the Cramér-von Mises type test statistics can also be used to check the adequacy of model (2): F n (3) = θ n (t, z) 2 H n(dt, dz), which converges in distribution to F (3) = W (t, z) 2 H (dt, dz), where H and H n are the distribution function and empirical distribution function of (X i, Z i ), respectively. Similar to the KS-type test statistics, F (3) = W (t, z) 2 H n(dt, dz) can be used to approximate the distribution of F (3). For dependent censoring, the proportional hazards model was used as the working model for the censoring time. Of course, we can also choose some other semiparametric regression models as the working model for censoring. For example, we may use the proportional mean residual life model or the additive mean residual life model, then we can obtain the estimators of the censoring model parameters using the approach of Chen and Cheng (25) or Chen and Cheng (26). Thus, the estimator of the censoring survival distribution G(t z) can be obtained using the following inversion formula G(t z) = m G( z) m G (t z) exp t } m G (u z) 1 du, where m G (t z) = E(C t z, C > t) is the MRLF of C at t given z. Thus, the unknown parameter in model (2) can be estimated by using the procedure in Section 3. Since estimating functions (7) and (13) were given in a somewhat ad hoc fashion using the generalized estimating equation methods, it would be worthwhile to further investigate

19 19 the efficiencies of the proposed estimators. In principle, it might be possible to estimate β and m ( ) more efficiently by the nonparametric maximum likelihood approach, and the resulting inference procedure would be much more complicated. Another issue is that the estimates of m (t) may be not monotonic, and there is no guarantee that the finite-sample estimator ˆm a (t) + t or ˆm b (t) + t would maintain the necessary monotonicity at some time point. The incorporation of the pooled-adjacent-violators algorithm may help solving the problem as mentioned in Chen and Cheng (25). ACKNOWLEDGEMENTS This research was partly supported by the National Natural Science Foundation of China Grants (No and 17311) and the National Basic Research Program of China (973 Program) (No. 27CB81492). APPENDIX: PROOFS OF ASYMPTOTIC PROPERTIES Using the uniform strong law of large numbers (Pollard, 199, p.41), we have z a (t) = lim n Za (t; β ), and s (k) (t; γ) = lim n S (k) (t; γ) (k =, 1) uniformly in t [, τ. Let z r (t) = s (1) (t; γ )/s () (t; γ ). In addition, assume that A defined below in (A.4) is nonsingular matrix. A.1. ASYMPTOTIC NORMALITY OF U a (β ) AND ˆβ a Note that [ m (t)dn i (t) Y i (t) g(β Z i ) 1 dt + dm (t) } = m (t) dm i (t),

20 2 and [ ˆma (t; β )dn i (t) Y i (t) g(β Z i ) 1 dt + d ˆm a (t; β ) } =. Then it follows that ˆm a (t; β ) m (t)} dn i (t) nπ n d ˆm a (t; β ) m (t)} = m (t) dm i (t), which is a first-order linear ordinary differential equation in ˆm a (t; β ) m (t). It thus has the closed-form solution given by Write U a (β ) = n 1 ˆm a (t; β ) m (t) = Φ n (t) 1 +n 1 h(β Z i )Z i Z a (t; β ) } m (t)dm i (t) t Φ n (u)m (u) dm i (u). nπ n (u) h(β Z i )Z i Z a (t; β ) } [ ˆm a (t; β) m (t) dn i (t). (A.1) Using the uniform strong law of large numbers and (A.1), the second term in the right-hand side of the above equation is equivalent to where n 1 µ (t)m (t)dm i (t) + o p (n 1/2 ), µ (t) = S(t) t 1 [ π(t) S(u) E h(β Z i )Z i z a (u)}dn i (u), π(t) = EY 1 (t), and S(t) is the marginal survival function of T. Therefore, n 1/2 U a (β ) = n 1/2 h(β Z i )Z i µ(t)} m (t)dm i (t) + o p (1), where µ(t) = z a (t) + µ (t). As a result, n 1/2 U a (β ) converges in distribution to zero-mean normal distribution with covariance matrix Σ a, where [ 2m Σ a = E h(β Z i )Z i µ(t)} (t) 2 dn i (t), (A.2)

21 21 which can be consistently estimated by ˆΣ a defined in Section 2. Since the censoring time C is independent of T and Z, and t S(u Z)g(β Z) 1 du = m (t)s(t Z), under model (2), it follows from the uniform strong law of large numbers that ˆm a (t; β ) β = Φ n (t) 1 t Φ n (u) π n (u) [n 1 Y i (u)h(β Z i )g(β Z i ) 1 Z i du = 1 [h(β S(t) E Z i )Z i S(u Z i )g(β Z i ) 1 du t + o p (1) = m (t) z a (t) + o p (1). (A.3) Let  = n 1 U(β )/ β, and h (1) (x) = dh(x)/dx. Then it follows from (A.3) that  = n 1 h (1) (β Z i )Z 2 i Y i(t)h (1) (β Z i )Z 2 } i Y i(t) +n 1 = n 1 where n 1 [ ˆm a (t; β )dn i (t) Y i (t)g(β Z i ) 1 dt h(β Z i )Z i Z a (t; β ) } [ ˆm a (t; β ) β h (1) (β Z i )Z 2 i Y i(t)h (1) (β Z i )Z 2 i Y i(t) } +Y i (t)dm (t) = A + o p (1), dn i (t) + Y i (t)h(β Z i )Z ig(β Z i ) 1 dt } [ m (t)dm i (t) + Y i (t)dm (t) h(β Z i )Z i Z a (t; β ) } [ z a (t) m (t)dm i (t) + Y i (t)g(β Z i ) 1 dt Y i (t)h(β Z i )Z ig(β Z i ) 1 dt + o p (1) [ A = E h(β Z i )Z i z a (t)} 2 Y i (t)g(β Z i ) 1 dt. (A.4) Thus, the asymptotic distribution of ˆβ a follows from a Taylor series expansion of U a ( ˆβ a ) at β. For future reference, we display the asymptotic approximation n 1/2 ( ˆβ a β ) = A 1 n 1/2 h(β Z i )Z i µ(t)} m (t)dm i (t) + o p (1). (A.5).

22 22 A.2. WEAK CONVERGENCE OF ˆm a (t) To show the weak convergence of n 1/2 ˆm a (t) m (t)}, we first note that n 1/2 ˆm a (t) m (t)} = n 1/2 ˆm a (t; β ) m (t)} + n 1/2 ˆm a (t; ˆβ a ) ˆm (t; β )}. It follows from (A.1) and the uniform strong law of large numbers that n 1/2 ˆm a (t; β ) m (t)} = S(t) 1 n 1/2 t S(u)m (u) dm i (u) + o p (1). π(u) Using the Taylor expansion of ˆm a (t; ˆβ a ) together with (A.3), we have n 1/2 ˆm a (t; ˆβ a ) ˆm a (t; β )} = m (t) z a (t) n 1/2 ˆβ a β } + o p (1). Thus, it follows from (A.5) that n 1/2 ˆm a (t) m (t)} = n /2 ϕ i (t) + o p (1), where ϕ i (t) = S(t) 1 t S(u)m (u) τ dm i (u)+m (t) z a (t) A 1 h(β π(u) Z i )Z i µ(u)} m (u)dm i (u). Because ϕ i (i = 1,..., n) are independent zero-mean random variables for each t, the multivariate central limit theorem implies that n 1/2 ˆm a (t) m (t)} ( t τ) converges in finite-dimensional distributions to zero-mean Gaussian process. Using the modern empirical theory as Lin et al. (2) and Lin, Wei and Ying (21), we can show that n 1/2 ˆm a (t) m (t)} is tight and converges weakly to zero-mean Gaussian process with covariance function Γ a (s, t) = Eϕ i (s)ϕ i (t)} at (s, t), which can be estimated by ˆΓ a (s, t) given in Section 2.

23 23 A.3. ASYMPTOTIC NORMALITY OF U b (β ) AND ˆβ b It can be checked that U b (β ) = n 1 h(β Z i )Z i Z } i δi G(X i ; γ, Λ ) 1 +n 1 h(β Z i )Z i Z } i δi [Ĝi (X i ; ˆγ, ˆΛ ) 1 G(X i ; γ, Λ ) 1 Q(t) L n (t; β, ˆγ, ˆΛ } ) L(t) dt + o p (n 1/2 ), (A.6) where Q(t) = lim n Q n (t), L(t) = L 1 (t)/(l 2 (t)l 3 (t)), L k (t) = lim n L kn (t; β, γ, Λ ) (k = 1, 2, 3), and Z i = h(β Z i )Z i g(β Z i ) 2 (X i t) + L(t)dt. It is well known that (Fleming and Harrington, 1991, p.299) t ˆΛ (t) Λ (t) = n 1 dmi c (u) t s () (u; γ ) z r (u) dλ (u)(ˆγ γ ) + o p (n 1/2 ), ˆγ γ = D 1 n 1 M c i (t) = N c i (t) t and D = lim n D n. Thus, Zi z r (u) } dm c i (u) + o p (n 1/2 ), Y i (u) expγ Z i }dλ (u), L kn (t; β, ˆγ, ˆΛ ) L kn (t; β, γ, Λ ) = n 1 and L kn (t; β, γ, Λ ) L k (t) = n 1 R k (t, u) s () (u) dm c i (u) + P k (t)(ˆγ γ ) + o p (n 1/2 ), ξ ki (t) + o p (n 1/2 ), where ξ ki (t) = V ki (t; β )G i (t; γ, Λ ) 1 L k (t), and R k (t, u) and P k (t) are the limits of R kn (t, u) and P kn (t), respectively. Therefore, using the functional Delta-method, it follows from (A.6) that n 1/2 U b (β ) = n 1/2 [ ξ i + R(t) τ S () (t; ˆγ) dm i c (t) + BD 1 Zi z r (t) } dmi c (t) + o p (1),

24 24 where ξ i = δ i h(β Z i )Z i Z } i G i (X i ; γ, Λ ) [ ξ 1i (t) Q(t) L 2 (t)ˆl 3 (t) L 1(t)ξ 2i(t) L 2 (t) 2 L 3 (t) L 1(t)ξ 3i (t) dt, L 2 (t)l 3 (t) 2 and R(t) and B are the limits of R n (t) and B n given in (14), respectively. Utilizing the multivariate central limit theorem, n 1/2 U b (β ) is asymptotically normal with mean zero and covariance matrix Σ b, where Σ b = E [ ξ i + R(t) τ S () (t; ˆγ) dm i c (t) + BD 1 Zi z r (t) } 2 dmi c (t). An empirical covariance estimator ˆΣ b defined by (14), in which all unknown quantities are replaced with their observed counterparts, converges in probability to Σ b. It can be checked that U b (β) converges almost surely uniformly in a closed set of β to u b (β), and u b (β ) =, where u b (β) = Eh(β Z)Z} Eh(β Z)Zg(β Z) 2 (T t) + } Eg(β Z) 2 (T t) + } Eg(β Z) 1 I(T > t)} dt. Eg(β Z) 1 T } For any function w(z), define E t,β w(z)} = Ew(Z)g(β Z) 2 (T t) + }, Eg(β Z) 2 (T t) + } and Then E β w(z)} = Ew(Z)g(β Z) 1 T }. Eg(β Z) 1 T } u b (β ) τ = 2 β 1 m () Var t,β h(β Z)Z}E β S(t Z)m( Z) 1 } Cov β h(β Z)Z, S(t Z)} Cov t,β h(β Z)Z, g(β Z) 1 }dt. We observe that S(t Z) is decreasing function of g(β Z) 1, which implies that h(β Z), S(t Z)} and cov t,β h(β Z)Z, g(β Z) 1 } must take opposite signs (Maguluri and Zhang, 1994). This gives that u b (β )/ β is positive definite. Thus, it follows that ˆβ b is consistent and unique

25 25 in a neighborhood of β. A Taylor series expansion of U b ( ˆβ b ) yields that n 1/2 ( ˆβ b β ) is asymptotically normal with mean zero and covariance matrix given by } 1 } ub (β ) ub (β ) 1 Σ b. β β A.4. WEAK CONVERGENCE OF θ n (t, z) It can be checked that φ n (t, z) = n1/2 B n (t, z) B(t, z)} H(t, z) B(t, z) H(t, z) 2 n1/2 H n (t, z) H(t, z)} + o p (1), (A.7) where and B n (t, z) = B(t, z) = z t z t (s t)ĝ(t) g( ˆβ aw)ĝ(s)h 1n(ds, dw), (s t)g(t) g(β w)g(s) H 1(ds, dw). Consider the martingale representation of the Kaplan-Meier estimator (Fleming and Harrington, 1991, p.97) Ĝ(t) G(t) G(t) t = Ĝ(u ) n M i c (u), (A.8) G(u) nπ n (u) where M c i (t) = N c i (t) t I(X i u)dλ c (u) and Λ c (t) = log(g(t)) is the cumulative hazard function of the censoring times. It is well known that M c i (t) (i = 1,..., n) are martingales with respect to the σ-filtration σi(x i u), I(X i u, δ i = ), Z i : u t, i = 1,..., n}. It follows from (A.8) and a Taylor series expansion that z ( B n (t, z) B(t, z) = n 1 (s t)g(t) s g(β w)g(s) t z + t t dm c i (u) π(u) ) H 1 (ds, dw) (s t)g(t) g(β w)g(s) [H 1n(ds, dw) H 1 (ds, dw)

26 26 z t Thus, combining (A.5), (A.7) and (A.9), we have (s t)h(β w)w G(t) H g(β w)g(s) 1 (ds, dw)( ˆβ a β ) + o p (1). θ n (t, z) = φ n (t, z) φ n (t, z u ) = n 1/2 where η i (t, z) = ρ i (t, z) ρ i (t, z u ), and η i (t, z) + o p (1), (A.9) ρ i (t, z) = G(t) H(t, z) t [ u z s t dm c G(s)g(β w) H 1(ds, dw) i (u) π(u) + δ i(x i t)g(t) H(t, z)g(x i )g(β Z i ) I(X V (t, z) i t, Z i z) H(t, z) I(X i t, Z i z) + G(t) H(t, z) z t (s t)h(β w)w H g(β w)g(s) 1 (ds, dw) A 1 h(β Z i )Z i µ(u)} m (u)dm i (u). Thus, by the same arguments as those of Appendix A.5 in Lin et al. (2), θ n (t, z) converges weakly to zero-mean Gaussian process with covariance function σ(t, z; t, z ) = Eη i (t, z)η i (t, z )} at (t, z) and (t, z ), which can be consistently estimated by ˆσ(t, z; t, z ) given in Section 4. REFERENCES Cai, J., Schaubel, D. E. (24). Marginal mean/rates models for multiple type recurrent event data. Lifetime Data Analysis, 1, Chen, Y. Q. (27). Additive Regression of Expectancy. Journal of the American Statistical Association, 12, Chen, Y. Q., Jewell, N. P., Lei, X., Cheng, S. C. (25). Semiparametric estimation of proportional mean residual life model in presence of censoring. Biometrics, 61,

27 27 Chen, Y. Q., Cheng, S. C. (25). Semiparametric regression analysis of mean residual life with censored survival data. Biometrika, 92, Chen, Y. Q. and Cheng, S. C. (26). Linear Life Expectancy Regression with Censored Data. Biometrika, 93, Cox, D. R. (1962). Renewal Theory. London: Spottiswoode Ballantyne. Cox, D. R. (1972). Regression Models and Life-Tables (with Discussion). Journal of the Royal Statistical Society, Ser. B, 34, Fleming, T. R., Harrington, D. P. (1991). Counting processes and survival analysis. New York: John Wiley. Lad, T., Rubinstein, L, and Sadeghi, A. (1988). The benefit of adjuvant treatment for resected locally advanced non-small-cell lung cancer. Journal of Clinical Oncology, 6, Liang, K. Y., Zeger, S. L. (1986). Longitudinal data analysis using generalized linear models. Biometrika, 73, Lin, D. Y., Wei, L. J., Yang, I. and Ying, Z. (2). Semiparametric Regression for the Mean and Rate Functions of Recurrent Events. Journal of the Royal Statistical Society, Ser. B, 62, Lin, D. Y., Wei, L. J. and Ying, Z. (21). Semiparametric Transformation Models for Point Processes. Journal of the American Statistical Association, 96, Maguluri, G. and Zhang, C.-H. (1994). Estimation in the Mean Residual Life Regression Model. Journal of the Royal Statistical Society, Ser. B, 56, Oakes, D., Dasu, T. (199). A note on residual life. Biometrika, 77, Piantadosi, S. (25). Clinical Trials: A Methodologic Perspective, 2nd Edition. New York: John Wiley.

28 28 Prentice, R. L., Self, S. G. (1983). Asymptotic distribution theory for Cox-type regression models with general relative risk form. The Annals of Statistics, 11, Pollard, D. (199). Empirical Processes: Theory and Applications. Hayward, CA: Institute of Mathematical Statistics. Robins, J. M., Rotnitzky, A. (1992). Recovery of information and adjustment for dependent censoring using surrogate markers. In AIDS Epidemiology-Methodologic Issues, eds. N. Jewell, K. Dietz, and V. Farewell, Boston: Birkhauser, Yuen, K. C., Zhu, L. X., Tang, N. Y. (23). On the mean residual life regression model. Journal of Statistical Planning and Inference, 113,

29 Table 1: Simulation Results (Independent Censoring) g1(x) = 1 + x g2(x) = e x g3(x) = log(1 + e x ) n β p ˆβa SE SEE CP ˆβa SE SEE CP ˆβa SE SEE CP 1. 1% % % % % % % % p represents proportion of right-censoring; ˆβa represents the mean of the point estimates of β; SE represents sample standard error of ˆβa; SEE represents the mean of the standard error of ˆβa; CP represents the empirical 95% coverage probability. 29

30 Table 2: Simulation Results (Dependent Censoring) g1(x) = 1 + x g2(x) = e x g3(x) = log(1 + e x ) n β p ˆβb SE SEE CP ˆβb SE SEE CP ˆβb SE SEE CP 1. 1% % % % % % % % p represents proportion of right-censoring; ˆβb represents the mean of the point estimates of β; SE represents sample standard error of ˆβb; SEE represents the mean of the standard error of ˆβb; CP represents the empirical 95% coverage probability. 3

31 31 Table 3: Estimation of the effects for the lung cancer data Covariates g(x) Parameter estimate SEE p-value 1 + x Treatment (Z 1 ) e x log(1 + e x ) x Cell type (Z 2 ) e x log(1 + e x ) Note: SEE is the standard error estimate; p-value pertains to testing no covariate effect.

32 32 g1; indep censoring (3%); beta =.5; n = 2 g1; dep censoring (1%); beta =.; n = 1 Standardized Estimate Standardized Estimate Normal Quantiles Normal Quantiles g2; indep censoring (1%); beta =.5; n = 1 g2; dep censoring (3%); beta =.; n = 2 Standardized Estimate Standardized Estimate Normal Quantiles Normal Quantiles g3; indep censoring (3%); beta =.; n = 2 g3; dep censoring (1%); beta =.5; n = 1 Standardized Estimate Standardized Estimate Normal Quantiles Normal Quantiles Figure 1: Normal Q-Q Plots

33 33 (a) Survival Function RT, nonsquamous/mixed RT, squamous RT + CAP, nonsquamous/mixed RT + CAP, squamous Censoring time (b) Survival Function RT RT + CAP Time (c) Survival Function nonsquamous/mixed squamous Time Figure 2: Kaplan-Meier Estimates of Survival Functions

University of California, Berkeley

University of California, Berkeley University of California, Berkeley U.C. Berkeley Division of Biostatistics Working Paper Series Year 24 Paper 153 A Note on Empirical Likelihood Inference of Residual Life Regression Ying Qing Chen Yichuan

More information

Hypothesis Testing Based on the Maximum of Two Statistics from Weighted and Unweighted Estimating Equations

Hypothesis Testing Based on the Maximum of Two Statistics from Weighted and Unweighted Estimating Equations Hypothesis Testing Based on the Maximum of Two Statistics from Weighted and Unweighted Estimating Equations Takeshi Emura and Hisayuki Tsukuma Abstract For testing the regression parameter in multivariate

More information

STAT Sample Problem: General Asymptotic Results

STAT Sample Problem: General Asymptotic Results STAT331 1-Sample Problem: General Asymptotic Results In this unit we will consider the 1-sample problem and prove the consistency and asymptotic normality of the Nelson-Aalen estimator of the cumulative

More information

Other Survival Models. (1) Non-PH models. We briefly discussed the non-proportional hazards (non-ph) model

Other Survival Models. (1) Non-PH models. We briefly discussed the non-proportional hazards (non-ph) model Other Survival Models (1) Non-PH models We briefly discussed the non-proportional hazards (non-ph) model λ(t Z) = λ 0 (t) exp{β(t) Z}, where β(t) can be estimated by: piecewise constants (recall how);

More information

FULL LIKELIHOOD INFERENCES IN THE COX MODEL

FULL LIKELIHOOD INFERENCES IN THE COX MODEL October 20, 2007 FULL LIKELIHOOD INFERENCES IN THE COX MODEL BY JIAN-JIAN REN 1 AND MAI ZHOU 2 University of Central Florida and University of Kentucky Abstract We use the empirical likelihood approach

More information

Package Rsurrogate. October 20, 2016

Package Rsurrogate. October 20, 2016 Type Package Package Rsurrogate October 20, 2016 Title Robust Estimation of the Proportion of Treatment Effect Explained by Surrogate Marker Information Version 2.0 Date 2016-10-19 Author Layla Parast

More information

UNIVERSITY OF CALIFORNIA, SAN DIEGO

UNIVERSITY OF CALIFORNIA, SAN DIEGO UNIVERSITY OF CALIFORNIA, SAN DIEGO Estimation of the primary hazard ratio in the presence of a secondary covariate with non-proportional hazards An undergraduate honors thesis submitted to the Department

More information

Linear life expectancy regression with censored data

Linear life expectancy regression with censored data Linear life expectancy regression with censored data By Y. Q. CHEN Program in Biostatistics, Division of Public Health Sciences, Fred Hutchinson Cancer Research Center, Seattle, Washington 98109, U.S.A.

More information

Tests of independence for censored bivariate failure time data

Tests of independence for censored bivariate failure time data Tests of independence for censored bivariate failure time data Abstract Bivariate failure time data is widely used in survival analysis, for example, in twins study. This article presents a class of χ

More information

Chapter 2 Inference on Mean Residual Life-Overview

Chapter 2 Inference on Mean Residual Life-Overview Chapter 2 Inference on Mean Residual Life-Overview Statistical inference based on the remaining lifetimes would be intuitively more appealing than the popular hazard function defined as the risk of immediate

More information

Lecture 2: Martingale theory for univariate survival analysis

Lecture 2: Martingale theory for univariate survival analysis Lecture 2: Martingale theory for univariate survival analysis In this lecture T is assumed to be a continuous failure time. A core question in this lecture is how to develop asymptotic properties when

More information

Part III. Hypothesis Testing. III.1. Log-rank Test for Right-censored Failure Time Data

Part III. Hypothesis Testing. III.1. Log-rank Test for Right-censored Failure Time Data 1 Part III. Hypothesis Testing III.1. Log-rank Test for Right-censored Failure Time Data Consider a survival study consisting of n independent subjects from p different populations with survival functions

More information

Survival Analysis for Case-Cohort Studies

Survival Analysis for Case-Cohort Studies Survival Analysis for ase-ohort Studies Petr Klášterecký Dept. of Probability and Mathematical Statistics, Faculty of Mathematics and Physics, harles University, Prague, zech Republic e-mail: petr.klasterecky@matfyz.cz

More information

Goodness-of-fit test for the Cox Proportional Hazard Model

Goodness-of-fit test for the Cox Proportional Hazard Model Goodness-of-fit test for the Cox Proportional Hazard Model Rui Cui rcui@eco.uc3m.es Department of Economics, UC3M Abstract In this paper, we develop new goodness-of-fit tests for the Cox proportional hazard

More information

ANALYSIS OF COMPETING RISKS DATA WITH MISSING CAUSE OF FAILURE UNDER ADDITIVE HAZARDS MODEL

ANALYSIS OF COMPETING RISKS DATA WITH MISSING CAUSE OF FAILURE UNDER ADDITIVE HAZARDS MODEL Statistica Sinica 18(28, 219-234 ANALYSIS OF COMPETING RISKS DATA WITH MISSING CAUSE OF FAILURE UNDER ADDITIVE HAZARDS MODEL Wenbin Lu and Yu Liang North Carolina State University and SAS Institute Inc.

More information

Lecture 5 Models and methods for recurrent event data

Lecture 5 Models and methods for recurrent event data Lecture 5 Models and methods for recurrent event data Recurrent and multiple events are commonly encountered in longitudinal studies. In this chapter we consider ordered recurrent and multiple events.

More information

Goodness-Of-Fit for Cox s Regression Model. Extensions of Cox s Regression Model. Survival Analysis Fall 2004, Copenhagen

Goodness-Of-Fit for Cox s Regression Model. Extensions of Cox s Regression Model. Survival Analysis Fall 2004, Copenhagen Outline Cox s proportional hazards model. Goodness-of-fit tools More flexible models R-package timereg Forthcoming book, Martinussen and Scheike. 2/38 University of Copenhagen http://www.biostat.ku.dk

More information

Exercises. (a) Prove that m(t) =

Exercises. (a) Prove that m(t) = Exercises 1. Lack of memory. Verify that the exponential distribution has the lack of memory property, that is, if T is exponentially distributed with parameter λ > then so is T t given that T > t for

More information

Published online: 10 Apr 2012.

Published online: 10 Apr 2012. This article was downloaded by: Columbia University] On: 23 March 215, At: 12:7 Publisher: Taylor & Francis Informa Ltd Registered in England and Wales Registered Number: 172954 Registered office: Mortimer

More information

Quantile Regression for Residual Life and Empirical Likelihood

Quantile Regression for Residual Life and Empirical Likelihood Quantile Regression for Residual Life and Empirical Likelihood Mai Zhou email: mai@ms.uky.edu Department of Statistics, University of Kentucky, Lexington, KY 40506-0027, USA Jong-Hyeon Jeong email: jeong@nsabp.pitt.edu

More information

Pairwise rank based likelihood for estimating the relationship between two homogeneous populations and their mixture proportion

Pairwise rank based likelihood for estimating the relationship between two homogeneous populations and their mixture proportion Pairwise rank based likelihood for estimating the relationship between two homogeneous populations and their mixture proportion Glenn Heller and Jing Qin Department of Epidemiology and Biostatistics Memorial

More information

Panel Count Data Regression with Informative Observation Times

Panel Count Data Regression with Informative Observation Times UW Biostatistics Working Paper Series 3-16-2010 Panel Count Data Regression with Informative Observation Times Petra Buzkova University of Washington, buzkova@u.washington.edu Suggested Citation Buzkova,

More information

Statistics 262: Intermediate Biostatistics Non-parametric Survival Analysis

Statistics 262: Intermediate Biostatistics Non-parametric Survival Analysis Statistics 262: Intermediate Biostatistics Non-parametric Survival Analysis Jonathan Taylor & Kristin Cobb Statistics 262: Intermediate Biostatistics p.1/?? Overview of today s class Kaplan-Meier Curve

More information

Attributable Risk Function in the Proportional Hazards Model

Attributable Risk Function in the Proportional Hazards Model UW Biostatistics Working Paper Series 5-31-2005 Attributable Risk Function in the Proportional Hazards Model Ying Qing Chen Fred Hutchinson Cancer Research Center, yqchen@u.washington.edu Chengcheng Hu

More information

STAT331. Cox s Proportional Hazards Model

STAT331. Cox s Proportional Hazards Model STAT331 Cox s Proportional Hazards Model In this unit we introduce Cox s proportional hazards (Cox s PH) model, give a heuristic development of the partial likelihood function, and discuss adaptations

More information

PENALIZED LIKELIHOOD PARAMETER ESTIMATION FOR ADDITIVE HAZARD MODELS WITH INTERVAL CENSORED DATA

PENALIZED LIKELIHOOD PARAMETER ESTIMATION FOR ADDITIVE HAZARD MODELS WITH INTERVAL CENSORED DATA PENALIZED LIKELIHOOD PARAMETER ESTIMATION FOR ADDITIVE HAZARD MODELS WITH INTERVAL CENSORED DATA Kasun Rathnayake ; A/Prof Jun Ma Department of Statistics Faculty of Science and Engineering Macquarie University

More information

Efficiency Comparison Between Mean and Log-rank Tests for. Recurrent Event Time Data

Efficiency Comparison Between Mean and Log-rank Tests for. Recurrent Event Time Data Efficiency Comparison Between Mean and Log-rank Tests for Recurrent Event Time Data Wenbin Lu Department of Statistics, North Carolina State University, Raleigh, NC 27695 Email: lu@stat.ncsu.edu Summary.

More information

Frailty Models and Copulas: Similarities and Differences

Frailty Models and Copulas: Similarities and Differences Frailty Models and Copulas: Similarities and Differences KLARA GOETHALS, PAUL JANSSEN & LUC DUCHATEAU Department of Physiology and Biometrics, Ghent University, Belgium; Center for Statistics, Hasselt

More information

Comparing Distribution Functions via Empirical Likelihood

Comparing Distribution Functions via Empirical Likelihood Georgia State University ScholarWorks @ Georgia State University Mathematics and Statistics Faculty Publications Department of Mathematics and Statistics 25 Comparing Distribution Functions via Empirical

More information

From semi- to non-parametric inference in general time scale models

From semi- to non-parametric inference in general time scale models From semi- to non-parametric inference in general time scale models Thierry DUCHESNE duchesne@matulavalca Département de mathématiques et de statistique Université Laval Québec, Québec, Canada Research

More information

FULL LIKELIHOOD INFERENCES IN THE COX MODEL: AN EMPIRICAL LIKELIHOOD APPROACH

FULL LIKELIHOOD INFERENCES IN THE COX MODEL: AN EMPIRICAL LIKELIHOOD APPROACH FULL LIKELIHOOD INFERENCES IN THE COX MODEL: AN EMPIRICAL LIKELIHOOD APPROACH Jian-Jian Ren 1 and Mai Zhou 2 University of Central Florida and University of Kentucky Abstract: For the regression parameter

More information

1 Glivenko-Cantelli type theorems

1 Glivenko-Cantelli type theorems STA79 Lecture Spring Semester Glivenko-Cantelli type theorems Given i.i.d. observations X,..., X n with unknown distribution function F (t, consider the empirical (sample CDF ˆF n (t = I [Xi t]. n Then

More information

Modelling Survival Events with Longitudinal Data Measured with Error

Modelling Survival Events with Longitudinal Data Measured with Error Modelling Survival Events with Longitudinal Data Measured with Error Hongsheng Dai, Jianxin Pan & Yanchun Bao First version: 14 December 29 Research Report No. 16, 29, Probability and Statistics Group

More information

MAS3301 / MAS8311 Biostatistics Part II: Survival

MAS3301 / MAS8311 Biostatistics Part II: Survival MAS3301 / MAS8311 Biostatistics Part II: Survival M. Farrow School of Mathematics and Statistics Newcastle University Semester 2, 2009-10 1 13 The Cox proportional hazards model 13.1 Introduction In the

More information

Full likelihood inferences in the Cox model: an empirical likelihood approach

Full likelihood inferences in the Cox model: an empirical likelihood approach Ann Inst Stat Math 2011) 63:1005 1018 DOI 10.1007/s10463-010-0272-y Full likelihood inferences in the Cox model: an empirical likelihood approach Jian-Jian Ren Mai Zhou Received: 22 September 2008 / Revised:

More information

Introduction to Empirical Processes and Semiparametric Inference Lecture 25: Semiparametric Models

Introduction to Empirical Processes and Semiparametric Inference Lecture 25: Semiparametric Models Introduction to Empirical Processes and Semiparametric Inference Lecture 25: Semiparametric Models Michael R. Kosorok, Ph.D. Professor and Chair of Biostatistics Professor of Statistics and Operations

More information

Linear rank statistics

Linear rank statistics Linear rank statistics Comparison of two groups. Consider the failure time T ij of j-th subject in the i-th group for i = 1 or ; the first group is often called control, and the second treatment. Let n

More information

Power and Sample Size Calculations with the Additive Hazards Model

Power and Sample Size Calculations with the Additive Hazards Model Journal of Data Science 10(2012), 143-155 Power and Sample Size Calculations with the Additive Hazards Model Ling Chen, Chengjie Xiong, J. Philip Miller and Feng Gao Washington University School of Medicine

More information

MODELING THE SUBDISTRIBUTION OF A COMPETING RISK

MODELING THE SUBDISTRIBUTION OF A COMPETING RISK Statistica Sinica 16(26), 1367-1385 MODELING THE SUBDISTRIBUTION OF A COMPETING RISK Liuquan Sun 1, Jingxia Liu 2, Jianguo Sun 3 and Mei-Jie Zhang 2 1 Chinese Academy of Sciences, 2 Medical College of

More information

log T = β T Z + ɛ Zi Z(u; β) } dn i (ue βzi ) = 0,

log T = β T Z + ɛ Zi Z(u; β) } dn i (ue βzi ) = 0, Accelerated failure time model: log T = β T Z + ɛ β estimation: solve where S n ( β) = n i=1 { Zi Z(u; β) } dn i (ue βzi ) = 0, Z(u; β) = j Z j Y j (ue βz j) j Y j (ue βz j) How do we show the asymptotics

More information

Nonparametric two-sample tests of longitudinal data in the presence of a terminal event

Nonparametric two-sample tests of longitudinal data in the presence of a terminal event Nonparametric two-sample tests of longitudinal data in the presence of a terminal event Jinheum Kim 1, Yang-Jin Kim, 2 & Chung Mo Nam 3 1 Department of Applied Statistics, University of Suwon, 2 Department

More information

Multivariate Survival Data With Censoring.

Multivariate Survival Data With Censoring. 1 Multivariate Survival Data With Censoring. Shulamith Gross and Catherine Huber-Carol Baruch College of the City University of New York, Dept of Statistics and CIS, Box 11-220, 1 Baruch way, 10010 NY.

More information

Analysis of transformation models with censored data

Analysis of transformation models with censored data Biometrika (1995), 82,4, pp. 835-45 Printed in Great Britain Analysis of transformation models with censored data BY S. C. CHENG Department of Biomathematics, M. D. Anderson Cancer Center, University of

More information

Approximation of Survival Function by Taylor Series for General Partly Interval Censored Data

Approximation of Survival Function by Taylor Series for General Partly Interval Censored Data Malaysian Journal of Mathematical Sciences 11(3): 33 315 (217) MALAYSIAN JOURNAL OF MATHEMATICAL SCIENCES Journal homepage: http://einspem.upm.edu.my/journal Approximation of Survival Function by Taylor

More information

Efficiency of Profile/Partial Likelihood in the Cox Model

Efficiency of Profile/Partial Likelihood in the Cox Model Efficiency of Profile/Partial Likelihood in the Cox Model Yuichi Hirose School of Mathematics, Statistics and Operations Research, Victoria University of Wellington, New Zealand Summary. This paper shows

More information

PhD course in Advanced survival analysis. One-sample tests. Properties. Idea: (ABGK, sect. V.1.1) Counting process N(t)

PhD course in Advanced survival analysis. One-sample tests. Properties. Idea: (ABGK, sect. V.1.1) Counting process N(t) PhD course in Advanced survival analysis. (ABGK, sect. V.1.1) One-sample tests. Counting process N(t) Non-parametric hypothesis tests. Parametric models. Intensity process λ(t) = α(t)y (t) satisfying Aalen

More information

STAT331. Combining Martingales, Stochastic Integrals, and Applications to Logrank Test & Cox s Model

STAT331. Combining Martingales, Stochastic Integrals, and Applications to Logrank Test & Cox s Model STAT331 Combining Martingales, Stochastic Integrals, and Applications to Logrank Test & Cox s Model Because of Theorem 2.5.1 in Fleming and Harrington, see Unit 11: For counting process martingales with

More information

Investigation of goodness-of-fit test statistic distributions by random censored samples

Investigation of goodness-of-fit test statistic distributions by random censored samples d samples Investigation of goodness-of-fit test statistic distributions by random censored samples Novosibirsk State Technical University November 22, 2010 d samples Outline 1 Nonparametric goodness-of-fit

More information

Survival Analysis Math 434 Fall 2011

Survival Analysis Math 434 Fall 2011 Survival Analysis Math 434 Fall 2011 Part IV: Chap. 8,9.2,9.3,11: Semiparametric Proportional Hazards Regression Jimin Ding Math Dept. www.math.wustl.edu/ jmding/math434/fall09/index.html Basic Model Setup

More information

Least Absolute Deviations Estimation for the Accelerated Failure Time Model. University of Iowa. *

Least Absolute Deviations Estimation for the Accelerated Failure Time Model. University of Iowa. * Least Absolute Deviations Estimation for the Accelerated Failure Time Model Jian Huang 1,2, Shuangge Ma 3, and Huiliang Xie 1 1 Department of Statistics and Actuarial Science, and 2 Program in Public Health

More information

asymptotic normality of nonparametric M-estimators with applications to hypothesis testing for panel count data

asymptotic normality of nonparametric M-estimators with applications to hypothesis testing for panel count data asymptotic normality of nonparametric M-estimators with applications to hypothesis testing for panel count data Xingqiu Zhao and Ying Zhang The Hong Kong Polytechnic University and Indiana University Abstract:

More information

Product-limit estimators of the survival function with left or right censored data

Product-limit estimators of the survival function with left or right censored data Product-limit estimators of the survival function with left or right censored data 1 CREST-ENSAI Campus de Ker-Lann Rue Blaise Pascal - BP 37203 35172 Bruz cedex, France (e-mail: patilea@ensai.fr) 2 Institut

More information

Median regression using inverse censoring weights

Median regression using inverse censoring weights Median regression using inverse censoring weights Sundarraman Subramanian 1 Department of Mathematical Sciences New Jersey Institute of Technology Newark, New Jersey Gerhard Dikta Department of Applied

More information

Analysis of Time-to-Event Data: Chapter 6 - Regression diagnostics

Analysis of Time-to-Event Data: Chapter 6 - Regression diagnostics Analysis of Time-to-Event Data: Chapter 6 - Regression diagnostics Steffen Unkel Department of Medical Statistics University Medical Center Göttingen, Germany Winter term 2018/19 1/25 Residuals for the

More information

Residuals and model diagnostics

Residuals and model diagnostics Residuals and model diagnostics Patrick Breheny November 10 Patrick Breheny Survival Data Analysis (BIOS 7210) 1/42 Introduction Residuals Many assumptions go into regression models, and the Cox proportional

More information

Empirical Likelihood in Survival Analysis

Empirical Likelihood in Survival Analysis Empirical Likelihood in Survival Analysis Gang Li 1, Runze Li 2, and Mai Zhou 3 1 Department of Biostatistics, University of California, Los Angeles, CA 90095 vli@ucla.edu 2 Department of Statistics, The

More information

Rank Regression Analysis of Multivariate Failure Time Data Based on Marginal Linear Models

Rank Regression Analysis of Multivariate Failure Time Data Based on Marginal Linear Models doi: 10.1111/j.1467-9469.2005.00487.x Published by Blacwell Publishing Ltd, 9600 Garsington Road, Oxford OX4 2DQ, UK and 350 Main Street, Malden, MA 02148, USA Vol 33: 1 23, 2006 Ran Regression Analysis

More information

POWER AND SAMPLE SIZE DETERMINATIONS IN DYNAMIC RISK PREDICTION. by Zhaowen Sun M.S., University of Pittsburgh, 2012

POWER AND SAMPLE SIZE DETERMINATIONS IN DYNAMIC RISK PREDICTION. by Zhaowen Sun M.S., University of Pittsburgh, 2012 POWER AND SAMPLE SIZE DETERMINATIONS IN DYNAMIC RISK PREDICTION by Zhaowen Sun M.S., University of Pittsburgh, 2012 B.S.N., Wuhan University, China, 2010 Submitted to the Graduate Faculty of the Graduate

More information

Bayesian Nonparametric Inference Methods for Mean Residual Life Functions

Bayesian Nonparametric Inference Methods for Mean Residual Life Functions Bayesian Nonparametric Inference Methods for Mean Residual Life Functions Valerie Poynor Department of Applied Mathematics and Statistics, University of California, Santa Cruz April 28, 212 1/3 Outline

More information

Empirical Processes & Survival Analysis. The Functional Delta Method

Empirical Processes & Survival Analysis. The Functional Delta Method STAT/BMI 741 University of Wisconsin-Madison Empirical Processes & Survival Analysis Lecture 3 The Functional Delta Method Lu Mao lmao@biostat.wisc.edu 3-1 Objectives By the end of this lecture, you will

More information

11 Survival Analysis and Empirical Likelihood

11 Survival Analysis and Empirical Likelihood 11 Survival Analysis and Empirical Likelihood The first paper of empirical likelihood is actually about confidence intervals with the Kaplan-Meier estimator (Thomas and Grunkmeier 1979), i.e. deals with

More information

Lecture 3. Truncation, length-bias and prevalence sampling

Lecture 3. Truncation, length-bias and prevalence sampling Lecture 3. Truncation, length-bias and prevalence sampling 3.1 Prevalent sampling Statistical techniques for truncated data have been integrated into survival analysis in last two decades. Truncation in

More information

Multistate Modeling and Applications

Multistate Modeling and Applications Multistate Modeling and Applications Yang Yang Department of Statistics University of Michigan, Ann Arbor IBM Research Graduate Student Workshop: Statistics for a Smarter Planet Yang Yang (UM, Ann Arbor)

More information

DAGStat Event History Analysis.

DAGStat Event History Analysis. DAGStat 2016 Event History Analysis Robin.Henderson@ncl.ac.uk 1 / 75 Schedule 9.00 Introduction 10.30 Break 11.00 Regression Models, Frailty and Multivariate Survival 12.30 Lunch 13.30 Time-Variation and

More information

Cox s proportional hazards model and Cox s partial likelihood

Cox s proportional hazards model and Cox s partial likelihood Cox s proportional hazards model and Cox s partial likelihood Rasmus Waagepetersen October 12, 2018 1 / 27 Non-parametric vs. parametric Suppose we want to estimate unknown function, e.g. survival function.

More information

ST745: Survival Analysis: Nonparametric methods

ST745: Survival Analysis: Nonparametric methods ST745: Survival Analysis: Nonparametric methods Eric B. Laber Department of Statistics, North Carolina State University February 5, 2015 The KM estimator is used ubiquitously in medical studies to estimate

More information

Empirical likelihood and self-weighting approach for hypothesis testing of infinite variance processes and its applications

Empirical likelihood and self-weighting approach for hypothesis testing of infinite variance processes and its applications Empirical likelihood and self-weighting approach for hypothesis testing of infinite variance processes and its applications Fumiya Akashi Research Associate Department of Applied Mathematics Waseda University

More information

SEMIPARAMETRIC METHODS FOR ESTIMATING CUMULATIVE TREATMENT EFFECTS IN THE PRESENCE OF NON-PROPORTIONAL HAZARDS AND DEPENDENT CENSORING

SEMIPARAMETRIC METHODS FOR ESTIMATING CUMULATIVE TREATMENT EFFECTS IN THE PRESENCE OF NON-PROPORTIONAL HAZARDS AND DEPENDENT CENSORING SEMIPARAMETRIC METHODS FOR ESTIMATING CUMULATIVE TREATMENT EFFECTS IN THE PRESENCE OF NON-PROPORTIONAL HAZARDS AND DEPENDENT CENSORING by Guanghui Wei A dissertation submitted in partial fulfillment of

More information

Asymptotic Distributions for the Nelson-Aalen and Kaplan-Meier estimators and for test statistics.

Asymptotic Distributions for the Nelson-Aalen and Kaplan-Meier estimators and for test statistics. Asymptotic Distributions for the Nelson-Aalen and Kaplan-Meier estimators and for test statistics. Dragi Anevski Mathematical Sciences und University November 25, 21 1 Asymptotic distributions for statistical

More information

Survival Analysis. Lu Tian and Richard Olshen Stanford University

Survival Analysis. Lu Tian and Richard Olshen Stanford University 1 Survival Analysis Lu Tian and Richard Olshen Stanford University 2 Survival Time/ Failure Time/Event Time We will introduce various statistical methods for analyzing survival outcomes What is the survival

More information

On Estimation of Partially Linear Transformation. Models

On Estimation of Partially Linear Transformation. Models On Estimation of Partially Linear Transformation Models Wenbin Lu and Hao Helen Zhang Authors Footnote: Wenbin Lu is Associate Professor (E-mail: wlu4@stat.ncsu.edu) and Hao Helen Zhang is Associate Professor

More information

Statistical Methods for Alzheimer s Disease Studies

Statistical Methods for Alzheimer s Disease Studies Statistical Methods for Alzheimer s Disease Studies Rebecca A. Betensky, Ph.D. Department of Biostatistics, Harvard T.H. Chan School of Public Health July 19, 2016 1/37 OUTLINE 1 Statistical collaborations

More information

AFT Models and Empirical Likelihood

AFT Models and Empirical Likelihood AFT Models and Empirical Likelihood Mai Zhou Department of Statistics, University of Kentucky Collaborators: Gang Li (UCLA); A. Bathke; M. Kim (Kentucky) Accelerated Failure Time (AFT) models: Y = log(t

More information

Improving Efficiency of Inferences in Randomized Clinical Trials Using Auxiliary Covariates

Improving Efficiency of Inferences in Randomized Clinical Trials Using Auxiliary Covariates Improving Efficiency of Inferences in Randomized Clinical Trials Using Auxiliary Covariates Anastasios (Butch) Tsiatis Department of Statistics North Carolina State University http://www.stat.ncsu.edu/

More information

Harvard University. Harvard University Biostatistics Working Paper Series

Harvard University. Harvard University Biostatistics Working Paper Series Harvard University Harvard University Biostatistics Working Paper Series Year 2008 Paper 85 Semiparametric Maximum Likelihood Estimation in Normal Transformation Models for Bivariate Survival Data Yi Li

More information

Simple techniques for comparing survival functions with interval-censored data

Simple techniques for comparing survival functions with interval-censored data Simple techniques for comparing survival functions with interval-censored data Jinheum Kim, joint with Chung Mo Nam jinhkim@suwon.ac.kr Department of Applied Statistics University of Suwon Comparing survival

More information

Goodness-of-fit tests for randomly censored Weibull distributions with estimated parameters

Goodness-of-fit tests for randomly censored Weibull distributions with estimated parameters Communications for Statistical Applications and Methods 2017, Vol. 24, No. 5, 519 531 https://doi.org/10.5351/csam.2017.24.5.519 Print ISSN 2287-7843 / Online ISSN 2383-4757 Goodness-of-fit tests for randomly

More information

Chapter 7 Fall Chapter 7 Hypothesis testing Hypotheses of interest: (A) 1-sample

Chapter 7 Fall Chapter 7 Hypothesis testing Hypotheses of interest: (A) 1-sample Bios 323: Applied Survival Analysis Qingxia (Cindy) Chen Chapter 7 Fall 2012 Chapter 7 Hypothesis testing Hypotheses of interest: (A) 1-sample H 0 : S(t) = S 0 (t), where S 0 ( ) is known survival function,

More information

Statistical Analysis of Competing Risks With Missing Causes of Failure

Statistical Analysis of Competing Risks With Missing Causes of Failure Proceedings 59th ISI World Statistics Congress, 25-3 August 213, Hong Kong (Session STS9) p.1223 Statistical Analysis of Competing Risks With Missing Causes of Failure Isha Dewan 1,3 and Uttara V. Naik-Nimbalkar

More information

Semiparametric maximum likelihood estimation in normal transformation models for bivariate survival data

Semiparametric maximum likelihood estimation in normal transformation models for bivariate survival data Biometrika (28), 95, 4,pp. 947 96 C 28 Biometrika Trust Printed in Great Britain doi: 1.193/biomet/asn49 Semiparametric maximum likelihood estimation in normal transformation models for bivariate survival

More information

Lecture 6 PREDICTING SURVIVAL UNDER THE PH MODEL

Lecture 6 PREDICTING SURVIVAL UNDER THE PH MODEL Lecture 6 PREDICTING SURVIVAL UNDER THE PH MODEL The Cox PH model: λ(t Z) = λ 0 (t) exp(β Z). How do we estimate the survival probability, S z (t) = S(t Z) = P (T > t Z), for an individual with covariates

More information

Estimation and Inference of Quantile Regression. for Survival Data under Biased Sampling

Estimation and Inference of Quantile Regression. for Survival Data under Biased Sampling Estimation and Inference of Quantile Regression for Survival Data under Biased Sampling Supplementary Materials: Proofs of the Main Results S1 Verification of the weight function v i (t) for the lengthbiased

More information

TESTINGGOODNESSOFFITINTHECOX AALEN MODEL

TESTINGGOODNESSOFFITINTHECOX AALEN MODEL ROBUST 24 c JČMF 24 TESTINGGOODNESSOFFITINTHECOX AALEN MODEL David Kraus Keywords: Counting process, Cox Aalen model, goodness-of-fit, martingale, residual, survival analysis. Abstract: The Cox Aalen regression

More information

CENSORED QUANTILE REGRESSION WITH COVARIATE MEASUREMENT ERRORS

CENSORED QUANTILE REGRESSION WITH COVARIATE MEASUREMENT ERRORS Statistica Sinica 21 211, 949-971 CENSORED QUANTILE REGRESSION WITH COVARIATE MEASUREMENT ERRORS Yanyuan Ma and Guosheng Yin Texas A&M University and The University of Hong Kong Abstract: We study censored

More information

Cox Regression in Nested Case Control Studies with Auxiliary Covariates

Cox Regression in Nested Case Control Studies with Auxiliary Covariates Biometrics DOI: 1.1111/j.1541-42.29.1277.x Cox Regression in Nested Case Control Studies with Auxiliary Covariates Mengling Liu, 1, Wenbin Lu, 2 and Chi-hong Tseng 3 1 Division of Biostatistics, School

More information

BIAS OF MAXIMUM-LIKELIHOOD ESTIMATES IN LOGISTIC AND COX REGRESSION MODELS: A COMPARATIVE SIMULATION STUDY

BIAS OF MAXIMUM-LIKELIHOOD ESTIMATES IN LOGISTIC AND COX REGRESSION MODELS: A COMPARATIVE SIMULATION STUDY BIAS OF MAXIMUM-LIKELIHOOD ESTIMATES IN LOGISTIC AND COX REGRESSION MODELS: A COMPARATIVE SIMULATION STUDY Ingo Langner 1, Ralf Bender 2, Rebecca Lenz-Tönjes 1, Helmut Küchenhoff 2, Maria Blettner 2 1

More information

Accelerated Failure Time Models: A Review

Accelerated Failure Time Models: A Review International Journal of Performability Engineering, Vol. 10, No. 01, 2014, pp.23-29. RAMS Consultants Printed in India Accelerated Failure Time Models: A Review JEAN-FRANÇOIS DUPUY * IRMAR/INSA of Rennes,

More information

Chapter 4 Fall Notations: t 1 < t 2 < < t D, D unique death times. d j = # deaths at t j = n. Y j = # at risk /alive at t j = n

Chapter 4 Fall Notations: t 1 < t 2 < < t D, D unique death times. d j = # deaths at t j = n. Y j = # at risk /alive at t j = n Bios 323: Applied Survival Analysis Qingxia (Cindy) Chen Chapter 4 Fall 2012 4.2 Estimators of the survival and cumulative hazard functions for RC data Suppose X is a continuous random failure time with

More information

UNIVERSITÄT POTSDAM Institut für Mathematik

UNIVERSITÄT POTSDAM Institut für Mathematik UNIVERSITÄT POTSDAM Institut für Mathematik Testing the Acceleration Function in Life Time Models Hannelore Liero Matthias Liero Mathematische Statistik und Wahrscheinlichkeitstheorie Universität Potsdam

More information

4. Comparison of Two (K) Samples

4. Comparison of Two (K) Samples 4. Comparison of Two (K) Samples K=2 Problem: compare the survival distributions between two groups. E: comparing treatments on patients with a particular disease. Z: Treatment indicator, i.e. Z = 1 for

More information

A note on L convergence of Neumann series approximation in missing data problems

A note on L convergence of Neumann series approximation in missing data problems A note on L convergence of Neumann series approximation in missing data problems Hua Yun Chen Division of Epidemiology & Biostatistics School of Public Health University of Illinois at Chicago 1603 West

More information

Analysis of Time-to-Event Data: Chapter 2 - Nonparametric estimation of functions of survival time

Analysis of Time-to-Event Data: Chapter 2 - Nonparametric estimation of functions of survival time Analysis of Time-to-Event Data: Chapter 2 - Nonparametric estimation of functions of survival time Steffen Unkel Department of Medical Statistics University Medical Center Göttingen, Germany Winter term

More information

1 Introduction. 2 Residuals in PH model

1 Introduction. 2 Residuals in PH model Supplementary Material for Diagnostic Plotting Methods for Proportional Hazards Models With Time-dependent Covariates or Time-varying Regression Coefficients BY QIQING YU, JUNYI DONG Department of Mathematical

More information

Estimation of Conditional Kendall s Tau for Bivariate Interval Censored Data

Estimation of Conditional Kendall s Tau for Bivariate Interval Censored Data Communications for Statistical Applications and Methods 2015, Vol. 22, No. 6, 599 604 DOI: http://dx.doi.org/10.5351/csam.2015.22.6.599 Print ISSN 2287-7843 / Online ISSN 2383-4757 Estimation of Conditional

More information

Logistic regression model for survival time analysis using time-varying coefficients

Logistic regression model for survival time analysis using time-varying coefficients Logistic regression model for survival time analysis using time-varying coefficients Accepted in American Journal of Mathematical and Management Sciences, 2016 Kenichi SATOH ksatoh@hiroshima-u.ac.jp Research

More information

USING MARTINGALE RESIDUALS TO ASSESS GOODNESS-OF-FIT FOR SAMPLED RISK SET DATA

USING MARTINGALE RESIDUALS TO ASSESS GOODNESS-OF-FIT FOR SAMPLED RISK SET DATA USING MARTINGALE RESIDUALS TO ASSESS GOODNESS-OF-FIT FOR SAMPLED RISK SET DATA Ørnulf Borgan Bryan Langholz Abstract Standard use of Cox s regression model and other relative risk regression models for

More information

Introduction to Empirical Processes and Semiparametric Inference Lecture 01: Introduction and Overview

Introduction to Empirical Processes and Semiparametric Inference Lecture 01: Introduction and Overview Introduction to Empirical Processes and Semiparametric Inference Lecture 01: Introduction and Overview Michael R. Kosorok, Ph.D. Professor and Chair of Biostatistics Professor of Statistics and Operations

More information

Typical Survival Data Arising From a Clinical Trial. Censoring. The Survivor Function. Mathematical Definitions Introduction

Typical Survival Data Arising From a Clinical Trial. Censoring. The Survivor Function. Mathematical Definitions Introduction Outline CHL 5225H Advanced Statistical Methods for Clinical Trials: Survival Analysis Prof. Kevin E. Thorpe Defining Survival Data Mathematical Definitions Non-parametric Estimates of Survival Comparing

More information

Practical considerations for survival models

Practical considerations for survival models Including historical data in the analysis of clinical trials using the modified power prior Practical considerations for survival models David Dejardin 1 2, Joost van Rosmalen 3 and Emmanuel Lesaffre 1

More information

Smooth nonparametric estimation of a quantile function under right censoring using beta kernels

Smooth nonparametric estimation of a quantile function under right censoring using beta kernels Smooth nonparametric estimation of a quantile function under right censoring using beta kernels Chanseok Park 1 Department of Mathematical Sciences, Clemson University, Clemson, SC 29634 Short Title: Smooth

More information