Abstract
Recurrent event and failure time data arise frequently in many clinical and observational studies. In this article, we propose a joint modeling of generalized scale-change models for the recurrent event process and the failure time, and allow the two processes to be correlated through a shared frailty. The proposed joint model is flexible in that it requires neither the Poisson assumption for the recurrent event process nor a parametric assumption on the frailty distribution. Estimating equation approaches are developed for parameter estimation, and the asymptotic properties of the resulting estimators are established. Simulation studies are conducted to evaluate the finite sample performances of the proposed method. An application to a medical cost study of chronic heart failure patients is provided.
Similar content being viewed by others
References
Andersen P, Gill R (1982) Cox’s regression model for counting processes: A large sample study. Annals of Statistics 10:1100–1120
Chen YQ, Jewell NP (2001) On a general class of semiparametric hazards regression models. Biometrika 88:687–702
Chen C-M, Shen P-S, Chuang Y (2016) The partly Aalen’s model for recurrent event data with a dependent terminal event. Statistics in Medicine 35:268–281
Cook R, Lawless J (2007) The Statistical Analysis of Recurrent Events. Springer, New York
Dong L, Sun L (2015) A flexible semiparametric transformation model for recurrent event data. Lifetime Data Analysis 21:20–41
Ghosh D (2004) Accelerated rates regression models for recurrent failure time data. Lifetime Data Analysis 10:247–261
Gill RD, Johansen S (1990) A survey of product-integration with a view toward application in survival analysis. The Annals of Statistics 18:1501–1555
Huang C-Y, Qin J, Wang M-C (2010) Semiparametric analysis for recurrent event data with time-dependent covariates and informative censoring. Biometrics 66:39–49
Huang C, Wang M (2004) Joint modeling and estimation for recurrent event processes and failure time data. Journal of the American Statistical Association 99:1153–1165
Kalbfleisch J, Schaubel D, Ye Y, Gong Q (2013) An estimating function approach to the analysis of recurrent and terminal events. Biometrics 69:366–374
Lancaster T, Intrator O (1998) Panel data with survival: hospitalization of HIV-positive patients. Journal of the American Statistical Association 93:46–53
Liang K, Zeger S (1986) Longitudinal data analysis using generalized linear models. Biometrika 73:13–22
Lin D, Wei L, Yang I, Ying Z (2000) Semiparametric regression for the mean and rate function of recurrent events. Journal of the Royal Statistical Society B 62:711–730
Lin DY, Wei LJ, Ying Z (1998) Accelerated failure time models for counting processes. Biometrika 85:605–618
Lin DY, Wei LJ, Ying Z (2001) Semiparametric transformation models for point processes. Journal of the American Statistical Association 96:620–628
Liu L, Huang X, O’Quigley J (2008) Analysis of longitudinal data in the presence of informative observational times and a dependent terminal event, with application to medical cost data. Biometrics 64:950–958
Liu L, Wolfe RA, Huang X (2004) Shared frailty models for recurrent events and a terminal event. Biometrics 60:747–756
Nocedal J, Wright S (1999) Numerical Optimization. Springer, New York
Pollard D (1990) Empirical Processes: Theory and Applications. Institute of Mathematical Statistics, Hayward, CA
Qu L, Sun L, Liu L (2017) Joint modeling of recurrent and terminal events using additive models. Statistics and Its Interface 10:699–710
Schaubel DE, Zeng D, Cai J (2006) A semiparametric additive rates model for recurrent event data. Lifetime Data Analysis 12:389–406
Sun L, Song X, Zhou J, Liu L (2012) Joint analysis of longitudinal data with informative observation times and a dependent terminal event. Journal of the American Statistical Association 107:688–700
Sun L, Su B (2008) A class of accelerated means regression models for recurrent event data. Lifetime Data Analysis 14:357–375
van der Vaart AW, Wellner JA (1996) Weak Convergence and Empirical Processes. Springer, New York
Sun Y, Chiou SH, Marr KA, Huang C-Y (2022) Statistical inference on shape and size indexes for counting processes. Biometrika 109:195–208
Wang M-C, Huang C-Y (2014) Statistical inference methods for recurrent event processes with shape and size parameters. Biometrika 101:553–566
Wang M, Qin J, Chiang C (2001) Analyzing recurrent event data with informative censoring. Journal of the American Statistical Association 96:1057–1065
Xu G, Chiou SH, Huang C-Y, Wang M-C, Yan J (2017) Joint scale-change models for recurrent events and failure time. Journal of the American Statistical Association 112:794–805
Xu G, Chiou SH, Yan J, Marr K, Huang C-Y (2020) Generalized scale-change Models for recurrent event processes under informative censoring. Statistica Sinica 30:1773–1795
Ye Y, Kalbfleisch J, Schaubel D (2007) Semiparametric analysis of correlated recurrent and terminal events. Biometrics 63:78–87
Ying Z (1993) A large sample study of rank estimation for censored regression data. The Annals of Statistics 21:76–99
Zeng D, Lin DY (2007) Semiparametric transformation models with random effects for recurrent events. Journal of the American Statistical Association 102:167–180
Zeng D, Lin DY (2009) Semiparametric transformation models with random effects for joint analysis of recurrent and terminal events. Biometrics 65:746–752
Zeng D, Lin DY (2008) Efficient resampling methods for nonsmooth estimating functions. Biostatistics 9:355–363
Acknowledgements
The authors thank the Editor, Professor Mei-Ling Ting Lee, an Associate Editor and two reviewers for their insightful comments and suggestions that greatly improved the article. This research was partly supported by the National Natural Science Foundation of China (No. 12171463).
Author information
Authors and Affiliations
Corresponding author
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Appendices
Appendix
In order to prove Theorem 1, we study the asymptotic distribution of \(n^{1/2}(\hat{\theta } - \theta _0)\), \(n^{1/2}(\hat{\xi } - \xi _0),\) \(n^{1/2}\{\widehat{\Lambda }_0(t) - \Lambda _0(t) \}\) and \(n^{1/2}\{\widehat{H}_0(t) - H_0(t) \}\).
1. Asymptotic Linearity of \(\widehat{\Lambda }_0(t; \alpha )\)
By the uniform strong law of large numbers (Pollard 1990), we obtain that \(Q_n(t; \alpha ) \rightarrow Q(t; \alpha )\) and \(R_n(t; \alpha ) \rightarrow R(t; \alpha )\) almost surely uniformly in \(\alpha \) and \(t \in [0, \tau ].\) In addition, the functional central limit theorem (Pollard 1990) implies that \(||Q_n(t; \alpha ) - Q(t; \alpha )|| = O_p(n^{-1/2})\) and \(||R_n(t; \theta ) - R(t; \theta )|| = O_p(n^{-1/2})\), where \(|| \cdot ||\) denotes the supremum norm. Then using the asymptotic properties of the product-integral (Gill and Johansen 1990), we have
Let \( \Lambda _0(t; \alpha ) =\exp \{ -\int ^\tau _t {R^{-1}(u; \alpha )} {dQ(u; \alpha )}\}\) with \(\Lambda _0(t; \alpha _0) \equiv \Lambda _0(t).\) It follows from the functional delta method that uniformly in \(\alpha \) and \(t \in [0, \tau ],\)
where
with \(E\phi _i(t; \alpha ) = 0.\) Then by (13) and (14), uniformly in \(\alpha \) and \(t \in [0, \tau ],\)
which gives an asymptotic i.i.d. representation of \(n^{1/2} \{\widehat{\Lambda }_n(t;\alpha ) - \Lambda _0(t; \alpha )\}.\)
We next show the asymptotic linearity of \(n^{1/2} \{\widehat{\Lambda }_0(t; \alpha ) - \widehat{\Lambda }_0(t; \alpha _0)\}.\) Note that
For any positive sequence \(\varepsilon _n \rightarrow 0\) and \(||\alpha - \alpha _0|| \le \varepsilon _n\), following similar arguments as in the proof of Theorem 1 of Ying (1993), we have that uniformly in \(\{\alpha : ||\alpha - \alpha _0|| \le \varepsilon _n \}\) and \(u \in [0, \tau ]\),
which implies that \(I_2(t; \alpha ) = o_p(n^{-1/2})\) uniformly in \(\alpha \) and \(t \in [0, \tau ]\). In addition, using a similar argument as in Lemma 3 of Ying (1993), we obtain that \(I_3(t; \alpha ) + I_4(t; \alpha ) = o_p(n^{-1/2})\) uniformly in \(\alpha \) and \(t \in [0, \tau ]\). Note that uniformly in \(\{\alpha : ||\alpha - \alpha _0|| \le \varepsilon _n \}\) and \(t \in [0, \tau ],\)
Then by integration by parts, \(I_ 5(t; \alpha ) = o_p(n^{-1/2})\) uniformly in \(\{\alpha : ||\alpha - \alpha _0|| \le \varepsilon _n \}\) and \(t \in [0, \tau ]\). Thus, \( I_2(t; \alpha ) + I_3(t; \alpha ) + I_4(t; \alpha ) + I_ 5(t; \alpha ) = o_p(n^{-1/2})\) uniformly in \(\{\alpha : ||\alpha - \alpha _0|| \le \varepsilon _n \}\) and \(u \in [0, \tau ]\). Furthermore, in view of the definitions of \(Q(t; \alpha )\) and \(R(t; \alpha )\), a straightforward calculation yields that uniformly in \(\alpha \) and \(t \in [0, \tau ],\)
Hence it follows from (13) and (16) that uniformly in \(\{\alpha : ||\alpha - \alpha _0|| \le \varepsilon _n \}\) and \(t \in [0, \tau ],\)
where
2. Asymptotic properties of \(\hat{\theta }\) and \(\widehat{\Lambda }_0(t) \)
We first show the asymptotic normality of \(S_n(\theta _0)\). Note that by (6) and (7),
In view of (15), the first term on the right-hand side of (18) equals
where \(V_1(x, m, t)\) is the joint probability measure of \((X_i, m_i, T_i)\) with \(x^*=(x, \Phi (x; \theta _0)^\prime )^\prime \) and \(t^*=te^{x'\alpha _0}.\) In a similar manner, we have
Thus, by combining the above results, we obtain
which implies that \(n^{1/2}S (\theta _0)\) is asymptotically normal, where
Next, we show the consistency of \(\hat{\theta }.\) Let \(S(\theta )=(S_{1}(\theta )^\prime , S_{2}(\theta )^\prime )^\prime ,\) where \(S_{1}(\theta )\) is the vector consisting of the first p components of \(S(\theta )\). Write
Note that
For any positive sequence \(\varepsilon _n \rightarrow 0,\) by the Taylor expansion and the uniform strong law of large numbers, we have that for \(||\theta - \theta _0|| \le \varepsilon _n,\) the first term on the right-hand side of (20) is
In view of (17), by following the similar argument as the above, the second term on the right-hand side of (20) is
Similarly, the third term on the right-hand side of (20) equals
Hence for any positive sequence \(\varepsilon _n \rightarrow 0\) and \(||\theta - \theta _0|| \le \varepsilon _n\),
where
Likewise, we have
where
Using (21) and (22), we have that for any sequence \(\varepsilon _n \rightarrow 0\) and \(||\theta - \theta _0|| \le \varepsilon _n\),
In a similar manner, we obtain that for any sequence \(\varepsilon _n \rightarrow 0\) and \(||\theta - \theta _0|| \le \varepsilon _n\),
where
Therefore,
where
Let \(\mathcal {S}(\theta )\) be the limit of \(n^{-1}S(\theta ).\) Note that \(\mathcal {S}(\theta _0)=0\) and \(\mathcal {S}(\theta )\ne 0\) for \(\theta \ne \theta _0.\) Then following the argument used in Theorem 2 of Lin et al. (1998), we have that \(\hat{\theta }\) is consistent. The above consistency proof is also suitable for the case when \(\hat{\theta }\) is the zero-crossing of \(S(\theta ).\)
For the asymptotic normality of \(\hat{\theta },\) it follows from (19) and (23) that
which implies that \(n^{1/2} (\hat{\theta } - \theta _0) \) is asymptotically normal with mean zero and covariance matrix \({D}^{-1}E\{\Upsilon _i \Upsilon _i^\prime \} ({D}^{-1})^\prime \).
To show the weak convergence of \(\widehat{\Lambda }_0(t),\) it follows from (15), (17) and (24) that uniformly in \(t \in [0, \tau ],\)
where \( \Psi _i(t) =\Lambda _0(t) \big \{-\varphi (t)^\prime (I_p, 0_p) {D}^{-1}\Upsilon _i +\phi _i(t; \alpha _0) \big \}, \) and \(I_p\) and \(0_p\) are the \(p \times p\) identity matrix and zero matrix, respectively. Because \(\Psi _i(t)\) \((i=1,...,n)\) are independent zero-mean random variables for each t, the multivariate central limit theorem implies that \(n^{1/2} \{\widehat{\Lambda }_0(t) - \Lambda _0(t) \}\) converges in finite-dimensional distributions to a zero-mean Gaussian process. Since \(\Psi _i(t)\) can be written as sums or products of monotone functions of t, and thus are manageable. Hence \(\Psi _i(\cdot )\) is a tight process (van der Vaart and Wellner 1996). Therefore, \(n^{1/2} \{\widehat{\Lambda }_0(t) - \Lambda _0(t) \}\) is tight and converges weakly to a zero-mean Gaussian process with covariance function \(E\{\Psi _i(s)\Psi _i(t)\}\) at (s, t).
3. Asymptotic properties of \(\hat{\xi }\) and \(\widehat{H}_0(t) \)
We first show the asymptotic normality of \( {U}(\xi ).\) For \(k=0\) and 1, define
It follows from (15), (17) and (24) that uniformly in \(t \in [0, \tau ],\) for \(k=0\) and 1,
where
and \(V_2(x, w, m, t)\) denotes the joint probability measure of \((X_i, W_i, m_i, T_i)\) with \(w^*(t)=(w^\prime , G(t, w; \xi _0)^\prime )^\prime ,\) \(t^*=te^{x'\alpha _0}\) and \(y^*=te^{w'\beta _0}.\) Let \(\mathcal {U}(\xi )\) be the limit of \(n^{-1} {U}(\xi ).\) Note that \(\mathcal {U}(\xi _0)=0.\) Then the functional delta method yields
which implies that \(n^{1/2}{U}(\xi _0)\) is asymptotically normal, where
and \(\bar{w}(t)={s}^{(1)}(t)/{s}^{(0)}(t).\)
Next, we show the consistency of \(\hat{\xi }.\) Let \( {U}(\xi )=( {U}_1(\xi )^\prime , {U}_2(\xi )^\prime )^\prime ,\) where \( {U}_1(\xi )\) is the vector consisting of the first p components of \( {U}(\xi )\). Write
By similar arguments as in Chen and Jewell (2001) and Sun and Su (2008), we obtain that for any sequence \(\varepsilon _n \rightarrow 0\) and \(||\xi - \xi _0|| \le \varepsilon _n\),
where
and \(s^{(2)}(t) = E[\nu _i e^{W_i^\prime (\gamma _0 - \beta _0)} I\{Y^*_i(\beta _0) \ge t\}W_i^{\otimes 2}]\) with \(a^{\otimes 2}=a a^\prime \) for any vector a. Thus, we have that for any sequence \(\varepsilon _n \rightarrow 0\) and \(||\xi - \xi _0|| \le \varepsilon _n\),
Let \( {\tilde{s}}^{(1)}(t) =E[ \nu _i G(t, W_i; \xi ) e^{W_i^\prime (\gamma _0 - \beta _0)} I\{Y^*_i(\beta _0) \ge t\}], \) \( \tilde{s}^{(2)}(t) =E[ \nu _i e^{W_i^\prime (\gamma _0 - \beta _0)} I\{Y^*_i(\beta _0) \ge t\} G(t, W_i; \xi ) W_i^\prime ],\) and \(\tilde{w}(t)={\tilde{s}}^{(1)}(t)/s^{(0)}(t).\) In a similar manner, we get that for any sequence \(\varepsilon _n \rightarrow 0\) and \(||\xi - \xi _0|| \le \varepsilon _n\),
where
Thus,
where
Using the consistency of \(\hat{\theta }\) and the asymptotic linearity of \(\widehat{\Lambda }_0(t; \alpha ),\) we have that \(\mathcal {U}(\xi _0)=0,\) and \(\mathcal {U}(\xi )\ne 0\) for \(\xi \ne \xi _0.\) Thus, \(\hat{\xi }\) is consistent (Lin et al. 1998). The above consistency proof is also suitable for the case when \(\hat{\xi }\) is the zero-crossing of \(U(\xi ).\)
For the asymptotic normality of \(\hat{\xi }\), it then follows from (27) and (28) that
This implies that \(n^{1/2} (\hat{\xi } - \xi _0) \) is asymptotically normal with mean zero and covariance matrix \(\tilde{D}^{-1}E\{\Gamma _i \Gamma _i^\prime \} (\tilde{D}^{-1})^\prime \).
To show the weak convergence of \(\widehat{H}_0(t),\) write
By the Taylor expansion and the uniform strong law of large numbers, we have that uniformly in \(t \in [0, \tau ],\)
By following a similar argument as in the proof of (17), it is seen that uniformly in \(t \in [0, \tau ],\)
Note that
Then by the argument used in the proofs of (26) and (27), we obtain that uniformly in \(t \in [0, \tau ],\)
where
Thus, it follows from (29) that uniformly in \(t \in [0, \tau ],\)
where
Hence, \(n^{1/2}\{\widehat{H}_0(t) - H_0(t)\}\) converges weakly to a zero-mean Gaussian process with covariance function \(E\{\psi _i(s)\psi _i(t)\}\) at (s, t). Furthermore, it follows from (24) and (29) that \(n^{1/2}(\hat{\theta } - \theta _0)\) and \(n^{1/2}( \hat{\xi } - \xi _0)\) have asymptotically a joint normal distribution with mean zero and covariance matrix \(E\{(\Upsilon _i^\prime ({D}^{-1})^\prime , \Gamma _i^\prime (\tilde{D}^{-1})^\prime )^\prime (\Upsilon _i^\prime ({D}^{-1})^\prime , \Gamma _i^\prime (\tilde{D}^{-1})^\prime ) \}.\) Also \(n^{1/2}\{\widehat{\Lambda }_0(t) - \Lambda _0(t)\}\) and \(n^{1/2}\{\widehat{H}_0(t) - H_0(t)\}\) jointly converge weakly to a zero-mean bivariate Gaussian process with covariance function \(E(\Psi _i(s), \psi _i(s))^\prime (\Psi _i(t), \psi _i(t))\}\) at (s, t).
Rights and permissions
Springer Nature or its licensor holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.
About this article
Cite this article
Wang, X., Sun, L. Joint modeling of generalized scale-change models for recurrent event and failure time data. Lifetime Data Anal 29, 1–33 (2023). https://doi.org/10.1007/s10985-022-09573-5
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s10985-022-09573-5