Abstract
In contrast to the popular Cox model which presents a multiplicative covariate effect specification on the time to event hazards, the semiparametric additive risks model (ARM) offers an attractive additive specification, allowing for direct assessment of the changes or the differences in the hazard function for changing value of the covariates. The ARM is a flexible model, allowing the estimation of both time-independent and time-varying covariates. It has a nonparametric component and a regression component identified by a finite-dimensional parameter. This chapter presents an efficient approach for maximum-likelihood (ML) estimation of the nonparametric and the finite-dimensional components of the model via the Minorize-Maximize (MM) algorithm for case-II interval-censored data. The operating characteristics of our proposed MM approach are assessed via simulation studies, with illustration on a breast cancer dataset via the R package MMIntAdd. It is expected that the proposed computational approach will not only provide scalability to the ML estimation scenario but may also simplify the computational burden of other complex likelihoods or models.
Access this chapter
Tax calculation will be finalised at checkout
Purchases are for personal use only
References
Bogaerts, K., Komárek, A., & Lesaffre, E. (2018). Survival analysis with interval-censored data: A practical approach with examples in R, SAS, and BUGS. CRC/Taylor & Francis Group.
Efron, B., & Tibshirani, R. J. (1993). An introduction to the bootstrap. Chapman and Hall.
Finkelstein, D. M. (1986). A proportional hazards model for interval-censored failure time data. Biometrics, 42, 845–854.
Finkelstein, D. M., & Wolfe, R. A. (1985). A semiparametric model for regression analysis of interval-censored failure time data. Biometrics, 41, 933–945.
Fletcher, R. (2013). Practical methods of optimization. John Wiley & Sons.
Huang, J. (1996). Efficient estimation for the proportional hazards model with interval censoring. Annals of Statistics, 24, 540–568.
Huang, X., Xu, J., & Tian, G. (2019). On profile MM algorithms for Gamma frailty survival models. Statistica Sinica, 29, 895–916.
Huffer, F. W., & McKeague, I. W. (1991). Weighted least squares estimation for Aalen’s additive risk model. Journal of the American Statistical Association, 86, 114–129.
Hunter, D. R., & Lange, K. (2000). Quantile regression via an MM algorithm. Journal of Computational and Graphical Statistics, 9, 60–77.
Hunter, D. R., & Lange, K. (2002). Computing estimates in the proportional odds model. Annals of the Institute of Statistical Mathematics, 54, 155–168.
Hunter, D. R., & Lange, K. (2004). A tutorial on MM algorithms. The American Statistician, 58, 30–37.
Hunter, D. R., & Li, R. (2005). Variable selection using MM algorithms. Annals of Statistics, 33, 1617–1642.
Lin, D., & Ying, Z. (1994). Semiparametric analysis of the additive risk model. Biometrika, 81, 61–71.
Martinussen, T., & Scheike, T. H. (2002). Efficient estimation in additive hazards regression with current status data. Biometrika, 89, 649–658.
Murphy, S. A., & Van der Vaart, A. W. (2000). On profile likelihood. Journal of the American Statistical Association, 95, 449–465.
Nguyen, H. D. (2017). An introduction to MM algorithms for machine learning and statistical estimation. WIREs Data Mining and Knowledge Discovery, 7(e1198).
Srivastava, S., DePalma, G., & Liu, C. (2019). An asynchronous distributed expectation maximization algorithm for massive data: the DEM algorithm. Journal of Computational and Graphical Statistics, 28, 233–243.
Wang, L., Sun, J., & Tong, X. (2010). Regression analysis of case-II interval-censored failure time data with the additive hazards model. Statistica Sinica, 20, 1709–1723.
Wang, P., Zhou, Y., & Sun, J. (2020). A new method for regression analysis of interval-censored data with the additive hazards model. Journal of the Korean Statistical Society, 49, 1131–1147.
Wang, T., He, K., Ma, W., Bandyopadhyay, D., & Sinha, S. (2022). Minorize-maximize algorithm for the generalized odds rate model for clustered current status data. Canadian Journal of Statistics. To appear.
Wu, T. T., & Lange, K. (2010). The MM alternative to EM. Statistical Science, 25, 492–505.
Zeng, D., Cai, J., & Shen, Y. (2006). Semiparametric additive risks model for interval-censored data. Statistica Sinica, 16, 287–302.
Zhang, Z., & Sun, J. (2010). Interval censoring. Statistical Methods in Medical Research, 19, 53–70.
Zhao, B., Wang, S., Wang, C., & Sun, J. (2021). New methods for the additive hazards model with the informatively interval-censored failure time data. Biometrical Journal, 63, 1507–1525.
Zhou, H., & Zhang, Y. (2012). EM vs MM: A case study. Computational Statistics & Data Analysis, 56, 3909–3920.
Acknowledgements
Bandyopadhyay acknowledges funding support from the NIH grants P20CA252717, P20CA264067, R21DE031879, R01DE031134, and P30CA016059 (VCU’s Massey Cancer Center Support Grant).
Author information
Authors and Affiliations
Corresponding author
Editor information
Editors and Affiliations
Appendix
Appendix
We shall use the second part of Lemma 1 from Wang et al. (2022) in proving Theorem 1, and we present this result in the following proposition. The proof of Proposition 1 can be found in Wang et al. (2022).
Proposition 1 (Wang et al., 2022)
For any τ, τ 0 ≥ 0
where A 1(τ 0) = exp(−τ 0)∕{1 −exp(−τ 0)} and A 2(τ 0) = exp(−τ 0)∕2{1 −exp(−τ 0)}2.
1.1 Proof of Theorem 1
In ℓ 2(λ, β) and ℓ 4(λ, β), (λ 1, …, λ m)⊤ are not entangled with β. Therefore, there is no need to develop the minorization functions for them. In the following, we show how to find the minorization functions for ℓ 1(λ, β) and ℓ 3(λ, β). Define \(u(L_i,X_i)=\sum _{k: t_k\leq L_i}\lambda _{k}+\beta ^{\top } Z_{x_i}(L_i)\), \(u(R_i,X_i)=\sum _{k: t_k\leq R_i}\lambda _{k}+\beta ^{\top } Z_{x_i}(R_i)\) and \(u(L_i, R_i,X_i)=\sum _{k: L_i<t_k\leq R_i}\lambda _{k}+\beta ^{\top } \{Z_{x_i}(R_i)-Z_{x_i}(L_i)\}\). According to our model assumption (1), u(L i, X i) > 0, u(R i, X i) > 0 and u(L i, R i, X i) > 0 for all i. Now, we can re-write
Applying Proposition 1 to the second term of the above display with τ = u(L i, X i) and τ 0 = u 0(L i, X i), we obtain
where C 1(u 0(L i, X i)) is the constant term that only depends on u 0(L i, X i), given as \(C_1(u_0(L_i,X_i))=\mbox{log}[1-\mbox{exp}\{-u_0(L_i, X_i)\}]-A_1(u_0(L_i,X_i))u_0(L_i,X_i)-A_2(u_0(L_i,X_i))u_0^2(L_i,X_i)+1.\) Next, we look into the following three terms of (7). First,
where, the inequality is obtained by applying Jensen’s inequality on the concave function f(x) = −x 2 and noting that \(\sum _{k: t_k\le L_i}\lambda _{k0}/u_0(L_i,X_i)+ \beta _0^{\top } Z_{x_i}(L_i)/u_0(L_i,X_i)=1\). Second, applying the standard inequality log(x) ≥ 1 − 1∕x for any generic x > 0, we have
and third,
where, the last inequality is obtained by applying Jensen’s inequality on the concave function f(x) = −1∕x, and noting that \(\sum _{k: t_k\le L_i}\lambda _{k0}/u_0(L_i,X_i)+ \beta _0^{\top } Z_{x_i}(L_i)/u_0(L_i,X_i)=1\). Then, applying the last three inequalities in (7), we obtain \(\ell _1(\lambda ,\beta )\ge \ell _{1,\dagger }(\lambda ,\beta |\lambda _0,\beta _0)\equiv \sum _{k=1}^m\mathcal {M}_{1,1,k}(\lambda _k|\lambda _0,\beta _0)+\mathcal {M}_{1,2}(\beta |\lambda _0,\beta _0)+\mathcal {M}_{1,3}(\lambda _0,\beta _0)\), where for k = 1, …, m,
and \( \mathcal {M}_{1,3}(\lambda _0,\beta _0)=\sum _{i=1}^n\Delta _{L,i}\{\mbox{log}[1-\mbox{exp}\{-u_0(L_i, X_i)\}]-A_1(u_0(L_i,X_i)) u_0(L_i,X_i)-A_2(u_0(L_i,X_i)) u_0^2(L_i,X_i)+1\} \). Next, consider finding the minorization function for ℓ 3(λ, β). Here, we use the same techniques as finding the minorization function for ℓ 1(λ, β). Note,
Now applying Proposition 1 to the second term of the above display with τ = u(L i, R i, X i) and τ 0 = u 0(L i, R i, X i), we obtain
where, C 1(u 0(L i, R i, X i)) is the constant term that only depends on u 0(L i, R i, X i), given by \( C_1(u_0(L_i,R_i,X_i))=\mbox{log}[1-\mbox{exp}\{-u_0(L_i,R_i, X_i)\}]-A_1(u_0(L_i,R_i,X_i)) u_0(L_i,R_i,X_i)-A_2(u_0(L_i,R_i,X_i)) u_0^2(L_i,R_i,X_i)+1.\) Similarly, we have the following three inequalities,
and
where, the first and the third inequalities are obtained by applying Jensen’s inequality on the concave function f(x) = −x 2 and f(x) = −1∕x, respectively, and the second inequality is obtained by applying the standard inequality log(x) ≥ 1 − 1∕x. Applying the above two inequalities in (8), we obtain \(\ell _3(\lambda ,\beta )\ge \ell _{3,\dagger }(\lambda ,\beta |\lambda _0,\beta _0)\equiv \sum _{k=1}^m\mathcal {M}_{3,1,k}(\lambda _k|\lambda _0,\beta _0)+\mathcal {M}_{3,2}(\beta |\lambda _0,\beta _0)+\mathcal {M}_{3,3}(\lambda _0,\beta _0)\), where
and
Finally, we obtain
where \( \mathcal {M}_{1,k}(\lambda _k|\lambda _0,\beta _0) = \mathcal {M}_{1,1,k}(\lambda _k|\lambda _0,\beta _0)+ \mathcal {M}_{3,1,k}(\lambda _k|\lambda _0,\beta _0)- \lambda _k\sum ^n_{i=1} \Delta _{I, i} I(t_k\leq L_i) \), \( \mathcal {M}_{2}(\beta |\lambda _0,\beta _0) = \mathcal {M}_{1, 2}(\beta |\lambda _0,\beta _0)+ \mathcal {M}_{3, 2}(\beta |\lambda _0,\beta _0)- \sum _{i=1}^n\Delta _{I,i}\beta ^{\top } Z_{x_i} (L_i)\), and \(\mathcal {M}_3(\lambda _0,\beta _0) = \mathcal {M}_{1, 3}(\lambda _0,\beta _0) +\mathcal {M}_{3, 3}(\lambda _0,\beta _0)\).
Rights and permissions
Copyright information
© 2022 The Author(s), under exclusive license to Springer Nature Switzerland AG
About this chapter
Cite this chapter
Wang, T., Bandyopadhyay, D., Sinha, S. (2022). Efficient Estimation of the Additive Risks Model for Interval-Censored Data. In: Sun, J., Chen, DG. (eds) Emerging Topics in Modeling Interval-Censored Survival Data. ICSA Book Series in Statistics. Springer, Cham. https://doi.org/10.1007/978-3-031-12366-5_9
Download citation
DOI: https://doi.org/10.1007/978-3-031-12366-5_9
Published:
Publisher Name: Springer, Cham
Print ISBN: 978-3-031-12365-8
Online ISBN: 978-3-031-12366-5
eBook Packages: Mathematics and StatisticsMathematics and Statistics (R0)