Skip to main content
Log in

An EM algorithm for the destructive COM-Poisson regression cure rate model

  • Published:
Metrika Aims and scope Submit manuscript

Abstract

In this paper, we consider a competitive scenario and assume the initial number of competing causes to undergo a destruction after an initial treatment. This brings in a more realistic and practical interpretation of the biological mechanism of the occurrence of tumor since what is recorded is only from the undamaged portion of the original number of competing causes. Instead of assuming any particular distribution for the competing cause, we assume the competing cause to follow a Conway–Maxwell Poisson distribution which brings in flexibility as it can handle both over-dispersion and under-dispersion that we usually encounter in count data. Under this setup and assuming a Weibull distribution to model the time-to-event, we develop the expectation maximization algorithm for such a flexible destructive cure rate model. An extensive simulation study is carried out to demonstrate the performance of the proposed estimation method. Finally, a melanoma data is analyzed for illustrative purpose.

This is a preview of subscription content, log in via an institution to check access.

Access this article

Price excludes VAT (USA)
Tax calculation will be finalised during checkout.

Instant access to the full article PDF.

Fig. 1
Fig. 2
Fig. 3

Similar content being viewed by others

References

  • Balakrishnan N, Pal S (2012) EM algorithm-based likelihood estimation for some cure rate models. J Stat Theory Pract 6(4):698–724

    Article  MathSciNet  Google Scholar 

  • Balakrishnan N, Pal S (2013) Lognormal lifetimes and likelihood-based inference for flexible cure rate models based on COM-Poisson family. Comput Stat Data Anal 67:41–67

    Article  MathSciNet  Google Scholar 

  • Balakrishnan N, Pal S (2015) An EM algorithm for the estimation of parameters of a flexible cure rate model with generalized gamma lifetime and model discrimination using likelihood-and information-based methods. Comput Stat 30(1):151–189

    Article  MathSciNet  MATH  Google Scholar 

  • Balakrishnan N, Pal S (2016) Expectation maximization-based likelihood inference for flexible cure rate models with Weibull lifetimes. Stat Methods Med Res 25(4):1535–1563

    Article  MathSciNet  Google Scholar 

  • Berkson J, Gage RP (1952) Survival curve for cancer patients following treatment. J Am Stat Assoc 47(259):501–515

    Article  Google Scholar 

  • Borges P, Rodrigues J, Balakrishnan N (2012) Correlated destructive generalized power series cure rate models and associated inference with an application to a cutaneous melanoma data. Comput Stat Data Anal 56(6):1703–1713

    Article  MathSciNet  MATH  Google Scholar 

  • Cancho VG, de Castro M, Rodrigues J (2012) A Bayesian analysis of the Conway–Maxwell–Poisson cure rate model. Stat Pap 53(1):165–176

    Article  MathSciNet  MATH  Google Scholar 

  • Chen MH, Ibrahim JG, Sinha D (1999) A new Bayesian model for survival data with a surviving fraction. J Am Stat Assoc 94(447):909–919

    Article  MathSciNet  MATH  Google Scholar 

  • Conway RW, Maxwell WL (1962) A queuing model with state dependent service rates. J Ind Eng 12(2):132–136

    Google Scholar 

  • Dunn PK, Smyth GK (1996) Randomized quantile residuals. J Comput Graph Stat 5(3):236–244

    Google Scholar 

  • Gallardo DI, Bolfarine H, Pedroso-de Lima AC (2016) An EM algorithm for estimating the destructive weighted Poisson cure rate model. J Stat Comput Simul 86(8):1497–1515

    Article  MathSciNet  Google Scholar 

  • Kokonendji CC, Mizere D, Balakrishnan N (2008) Connections of the Poisson weight function to overdispersion and underdispersion. J Stat Plan Inference 138(5):1287–1296

    Article  MathSciNet  MATH  Google Scholar 

  • Kuk AYC, Chen CH (1992) A mixture model combining logistic regression with proportional hazards regression. Biometrika 79(3):531–541

    Article  MATH  Google Scholar 

  • Lange K (1995) A gradient algorithm locally equivalent to the EM algorithm. J R Stat Soc Ser B 57:425–437

    MathSciNet  MATH  Google Scholar 

  • Li CS, Taylor JMG (2002) A semi-parametric accelerated failure time cure model. Stat Med 21(21):3235–3247

    Article  Google Scholar 

  • Li CS, Taylor JMG, Sy JP (2001) Identifiability of cure models. Stat Probab Lett 54(4):389–395

    Article  MathSciNet  MATH  Google Scholar 

  • Louis TA (1982) Finding the observed information matrix when using the EM algorithm. J R Stat Soc Ser B (Methodol) 44(2):226–233

    MathSciNet  MATH  Google Scholar 

  • Lu W, Ying Z (2004) On semiparametric transformation cure models. Biometrika 91(2):331–343

    Article  MathSciNet  MATH  Google Scholar 

  • Maller RA, Zhou X (1996) Survival analysis with long-term survivors. Wiley, New York

    MATH  Google Scholar 

  • Pal S, Balakrishnan N (2015) Likelihood inference based on EM algorithm for the destructive length-biased Poisson cure rate model with Weibull lifetime. Commun Stat Simul Comput. https://doi.org/10.1080/03610918.2015.1053918

    Google Scholar 

  • Pal S, Balakrishnan N (2016) Destructive negative binomial cure rate model and EM-based likelihood inference under Weibull lifetime. Stat Probab Lett 116:9–20

    Article  MathSciNet  MATH  Google Scholar 

  • Pal S, Balakrishnan N (2017a) An EM type estimation procedure for the destructive exponentially weighted Poisson regression cure model under generalized gamma lifetime. J Stat Comput Simul 87(6):1107–1129

    Article  MathSciNet  Google Scholar 

  • Pal S, Balakrishnan N (2017b) Likelihood inference for the destructive exponentially weighted Poisson cure rate model with Weibull lifetime and an application to melanoma data. Comput Stat 32(2):429–449

    Article  MathSciNet  MATH  Google Scholar 

  • Peng Y, Zhang J (2008) Identifiability of a mixture cure frailty model. Stat Probab Lett 78(16):2604–2608

    Article  MATH  Google Scholar 

  • Rigby RA, Stasinopoulos DM (2005) Generalized additive models for location, scale and shape. J Roy Stat Soc Ser C (Appl Stat) 54(3):507–554

    Article  MathSciNet  MATH  Google Scholar 

  • Rodrigues J, Cancho VG, de Castro M, Louzada-Neto F (2009a) On the unification of long-term survival models. Stat Probab Lett 79(6):753–759

    Article  MathSciNet  MATH  Google Scholar 

  • Rodrigues J, de Castro M, Cancho VG, Balakrishnan N (2009b) COM-Poisson cure rate survival models and an application to a cutaneous melanoma data. J Stat Plan Inference 139(10):3605–3611

    Article  MathSciNet  MATH  Google Scholar 

  • Rodrigues J, de Castro M, Balakrishnan N, Cancho VG (2011) Destructive weighted Poisson cure rate models. Lifetime Data Anal 17(3):333–346

    Article  MathSciNet  MATH  Google Scholar 

  • Rodrigues J, Balakrishnan N, Cordeiro GM, de Castro M, Cancho VG (2015) Latent cure rate model under repair system and threshold effect. J Stat Comput Simul 85(14):2860–2873

    Article  MathSciNet  Google Scholar 

  • Shmueli G, Minka TP, Kadane JB, Borle S, Boatwright P (2005) A useful distribution for fitting discrete data: revival of the Conway–Maxwell–Poisson distribution. J Roy Stat Soc Ser C (Appl Stat) 54(1):127–142

    Article  MathSciNet  MATH  Google Scholar 

  • Tsodikov AD, Ibrahim JG, Yakovlev AY (2003) Estimating cure rates from survival data. J Am Stat Assoc 98(464):1063–1078

    Article  Google Scholar 

  • Yakovlev AY, Tsodikov AD (1996) Stochastic models of tumor latency and their biostatistical applications. World Scientific, Singapore

    Book  MATH  Google Scholar 

  • Yu B, Tiwari RC, Cronin KA, Feuer EJ (2004) Cure fraction estimation from the mixture cure models for grouped survival data. Stat Med 23(11):1733–1747

    Article  Google Scholar 

Download references

Acknowledgements

Our sincere thanks go to the reviewers for their constructive comments and suggestions on earlier versions of this manuscript which led to this improved version. The third author also expresses his thanks to the Natural Sciences and Engineering Research Council of Canada for funding this research through an individual discovery grant.

Author information

Authors and Affiliations

Authors

Corresponding author

Correspondence to Suvra Pal.

Ethics declarations

Conflicts of interest

On behalf of all authors, the corresponding author states that there is no conflict of interest.

Electronic supplementary material

Appendix A: First- and second-order derivatives of the Q-function

Appendix A: First- and second-order derivatives of the Q-function

1.1 Destructive Bernoulli cure rate model

$$\begin{aligned} Q({\varvec{\theta }},\varvec{\pi }^{(k)})\propto & {} \displaystyle \sum _{I_1} \log \left( \frac{\eta _ip_if(t_i)}{1+\eta _i}\right) + \displaystyle \sum _{I_0} (1-\pi _i^{(k)}) \, \log \left( \frac{1+\eta _i(1-p_i)}{1+\eta _i}\right) \\&+\displaystyle \sum _{I_0}\pi _i^{(k)} \log \left( \frac{\eta _ip_iS(t_i)}{1+\eta _i}\right) .\\ \frac{\partial Q}{\partial \beta _{1l}}= & {} \displaystyle \sum _{I_1}\frac{x_{1il}}{1+\eta _i}-\displaystyle \sum _{I_0}(1-\pi _i^{(k)})\frac{x_{1il}\eta _ip_i}{(1+\eta _i)\{1+\eta _i(1-p_i)\}}\\&+\displaystyle \sum _{I_0}\pi _i^{(k)}\frac{x_{1il}}{1+\eta _i},\\ \frac{\partial Q}{\partial \beta _{2k}}= & {} \displaystyle \sum _{I_1}x_{2ik}(1-p_i)-\displaystyle \sum _{I_0}(1-\pi _i^{(k)})\frac{x_{2ik}\eta _ip_i(1-p_i)}{\{1+\eta _i(1-p_i)\}}\\&+ \displaystyle \sum _{I_0}\pi _i^{(k)}x_{2ik}(1-p_i),\\ \frac{\partial Q}{\partial \gamma _1}= & {} \displaystyle \sum _{I_1}E_i+\displaystyle \sum _{I_0}\pi _i^{(k)}\frac{\log (\gamma _2t_i)(\gamma _2t_i)^{\frac{1}{\gamma _1}}}{\gamma ^2_1},\\ \frac{\partial Q}{\partial \gamma _2}= & {} \displaystyle \sum _{I_1}H_i-\displaystyle \sum _{I_0}\pi _i^{(k)}\frac{(\gamma _2t_i)^{\frac{1}{\gamma _1}}}{\gamma _1\gamma _2},\\ \frac{\partial ^2 Q}{\partial \beta _{1l}\partial \beta _{1l^{\prime }}}= & {} -\displaystyle \sum _{I_1}\frac{x_{1il}x_{1il^{\prime }}\eta _i}{(1+\eta _i)^2}-\displaystyle \sum _{I_0}(1-\pi _i^{(k)})\frac{x_{1il}x_{1il^{\prime }}\eta _ip_i\{1-\eta ^2_i(1-p_i)\}}{(1+\eta _i)^2\{1+\eta _i(1-p_i)\}^2}\\&-\displaystyle \sum _{I_0}\pi _i^{(k)}\frac{x_{1il}x_{1il^{\prime }}\eta _i}{(1+\eta _i)^2},\\ \frac{\partial ^2 Q}{\partial \beta _{2k}\partial \beta _{2k^{\prime }}}= & {} -\displaystyle \sum _{I_1}x_{2ik}x_{2ik^{\prime }}p_i(1-p_i)\\ \end{aligned}$$
$$\begin{aligned}&-\displaystyle \sum _{I_0}(1-\pi _i^{(k)})\frac{x_{2ik}x_{2ik^{\prime }}\eta _ip_i(1-p_i)^2\{1+\eta _i-\frac{p_i}{(1-p_i)}\}\{1+\eta _i(1-p_i)\}}{\{1+\eta _i(1-p_i)\}^2}\\&- \displaystyle \sum _{I_0}\pi _i^{(k)}x_{2ik}x_{2ik^{\prime }}p_i(1-p_i),\\ \frac{\partial ^2 Q}{\partial \gamma ^2_1}= & {} -\displaystyle \sum _{I_1}\left\{ \frac{2E_i}{\gamma _1}+\frac{1}{\gamma _1^2}+\frac{(\log (\gamma _2t_i))^2(\gamma _2t_i)^{\frac{1}{\gamma _1}}}{\gamma _1^4}\right\} \\&-\displaystyle \sum _{I_0}\pi _i^{(k)}\frac{\log (\gamma _2t_i)(\gamma _2t_i)^{\frac{1}{\gamma _1}}\{\log (\gamma _2t_i)+2\gamma _1\}}{\gamma ^4_1},\\ \frac{\partial ^2 Q}{\partial \gamma ^2_2}= & {} -\displaystyle \sum _{I_1}\left\{ \frac{H_i}{\gamma _2} + \frac{(\gamma _2t_i)^{\frac{1}{\gamma _1}}}{\gamma _1^2\gamma _2^2}\right\} +\displaystyle \sum _{I_0}\pi _i^{(k)}\frac{(\gamma _2t_i)^{\frac{1}{\gamma _1}}(\gamma _1-1)}{\gamma ^2_1\gamma ^2_2},\\ \frac{\partial ^2 Q}{\partial \beta _{1l}\partial \beta _{2k}}= & {} -\displaystyle \sum _{I_0}(1-\pi _i^{(k)})\frac{x_{1il}x_{2ik}\eta _ip_i(1-p_i)}{\{1+\eta _i(1-p_i)\}^2},\\ \frac{\partial ^2 Q}{\partial \gamma _1\partial \gamma _2}= & {} \displaystyle \sum _{I_1}\left\{ \frac{(\gamma _2t_i)^{\frac{1}{\gamma _1}}\log (\gamma _2t_1)}{\gamma ^2_1\gamma _2}-\frac{H_i}{\gamma _1}\right\} + \displaystyle \sum _{I_0}\pi _i^{(k)}\frac{(\gamma _2t_i)^{\frac{1}{\gamma _1}}}{\gamma ^2_1\gamma _2}\left( 1+\frac{\log (\gamma _2t_1)}{\gamma _1}\right) ,\\ \frac{\partial ^2 Q}{\partial \beta _{1l}\partial \gamma _s}= & {} 0,\\ \frac{\partial ^2 Q}{\partial \beta _{2k}\partial \gamma _s}= & {} 0\\ \end{aligned}$$

for \(s=1,2\), where

$$\begin{aligned} E_i = \frac{\log (\gamma _2t_i)(\gamma _2t_i)^{\frac{1}{\gamma _1}}}{\gamma ^2_1} - \frac{\log (\gamma _2t_i)}{\gamma ^2_1} - \frac{1}{\gamma _1}\quad \text {and} \quad H_i=\frac{1-(\gamma _2t_i)^{\frac{1}{\gamma _1}}}{\gamma _1\gamma _2}. \end{aligned}$$

1.2 Destructive Poisson cure rate model

$$\begin{aligned} Q({\varvec{\theta }},\varvec{\pi }^{(k)})\propto & {} \displaystyle \sum _{I_1}\left\{ \log (\eta _ip_if(t_i))-\eta _ip_iF(t_i)\right\} - \, \displaystyle \sum _{I_0} (1-\pi _i^{(k)})\eta _ip_i\\&+\displaystyle \sum _{I_0}\pi _i^{(k)} \log \left\{ \exp (-\eta _ip_iF(t_i))-\exp (-\eta _ip_i)\right\} .\\ \frac{\partial Q}{\partial \beta _{1l}}= & {} \displaystyle \sum _{I_1}x_{1il}(1-\eta _ip_iF(t_i))-\displaystyle \sum _{I_0}(1-\pi _i^{(k)})x_{1il}\eta _ip_i - \displaystyle \sum _{I_0}\pi _i^{(k)}x_{1il}\eta _ip_iB_i,\\ \frac{\partial Q}{\partial \beta _{2k}}= & {} \displaystyle \sum _{I_1}x_{2ik}(1-p_i)(1-\eta _ip_iF(t_i))-\displaystyle \sum _{I_0}(1-\pi _i^{(k)})x_{2ik}\eta _ip_i(1-p_i)\\&- \displaystyle \sum _{I_0}\pi _i^{(k)}x_{2ik}\eta _ip_i(1-p_i)B_i,\\ \frac{\partial Q}{\partial \gamma _1}= & {} \displaystyle \sum _{I_1}\left( E_i -\eta _ip_iF_1^{\prime }(t_i)\right) -\displaystyle \sum _{I_0}\pi _i^{(k)}\frac{\eta _ip_iF_1^{\prime }(t_i)}{C_i},\\ \frac{\partial Q}{\partial \gamma _2}= & {} \displaystyle \sum _{I_1}\left( H_i-\eta _ip_iF_2^{\prime }(t_i)\right) - \displaystyle \sum _{I_0}\pi _i^{(k)}\frac{\eta _ip_iF_2^{\prime }(t_i)}{C_i},\\ \frac{\partial ^2 Q}{\partial \beta _{1l}\partial \beta _{1l^{\prime }}}= & {} -\displaystyle \sum _{I_1}x_{1il}x_{1il^{\prime }}\eta _ip_iF(t_i)-\displaystyle \sum _{I_0}(1-\pi _i^{(k)})x_{1il}x_{1il^{\prime }}\eta _ip_i\\&+\displaystyle \sum _{I_0}\pi _i^{(k)}x_{1il}x_{1il^{\prime }}\eta _ip_i\left\{ \eta _ip_i\left( B_i^*-B_i^2\right) +B_i\right\} ,\\ \frac{\partial ^2 Q}{\partial \beta _{2k} \partial \beta _{2k^{\prime }}}= & {} -\displaystyle \sum _{I_1}x_{2ik}x_{2ik^{\prime }}p_i(1-p_i)\left\{ \eta _i(1-p_i)F(t_i) + 1-\eta _ip_iF(t_i)\right\} \\&+\displaystyle \sum _{I_0}(1-\pi _i^{(k)})x_{2ik}x_{2ik^{\prime }}\eta _ip_i(1-p_i)(2p_i-1)\\&+\displaystyle \sum _{I_0}\pi _i^{(k)}x_{1il}x_{2ik}\eta _ip_i(1-p_i)^2\left\{ \eta _ip_i(B_i^*-B_i^2)+\frac{(2p_i-1)B_i}{1-p_i}\right\} ,\\ \frac{\partial ^2 Q}{\partial \gamma ^2_1}= & {} -\displaystyle \sum _{I_1}\left\{ \frac{2E_i}{\gamma _1}+\frac{1}{\gamma _1^2}+\frac{(\log (\gamma _2t_i))^2(\gamma _2t_i)^{\frac{1}{\gamma _1}}}{\gamma _1^4}+\eta _ip_iF_{11}^{\prime \prime }(t_i)\right\} \end{aligned}$$
$$\begin{aligned}&-\displaystyle \sum _{I_0}\pi _i^{(k)}\left\{ \frac{\eta _ip_iF_{11}^{\prime \prime }(t_i)}{C_i}+\frac{\left( \eta _ip_iF_1^{\prime }(t_i)\right) ^2(1-C_i)}{C_i^2}\right\} ,\\ \frac{\partial ^2 Q}{\partial \gamma ^2_2}= & {} -\displaystyle \sum _{I_1}\left\{ \frac{H_i}{\gamma _2} + \frac{(\gamma _2t_i)^{\frac{1}{\gamma _1}}}{\gamma _1^2\gamma _2^2} + \eta _ip_iF_{22}^{\prime \prime }(t_i)\right\} \\&-\displaystyle \sum _{I_0}\pi _i^{(k)}\left\{ \frac{\eta _ip_iF_{22}^{\prime \prime }(t_i)}{C_i}+\frac{\eta _i^2p_i^2\left( F_2^{\prime }(t_i)\right) ^2 (1-C_i)}{C_i^2}\right\} ,\\ \frac{\partial ^2 Q}{\partial \beta _{1l}\partial \beta _{2k}}= & {} -\displaystyle \sum _{I_1}x_{1il}x_{2ik}\eta _ip_i(1-p_i)F(t_i) - \displaystyle \sum _{I_0}(1-\pi _i^{(k)})x_{1il}x_{2ik}\eta _ip_i(1-p_i)\\&+\displaystyle \sum _{I_0}\pi _i^{(k)}x_{1il}x_{2ik}\eta _ip_i(1-p_i) \left\{ \eta _ip_i\left( B_i^*-B_i^2\right) -B_i\right\} ,\\ \frac{\partial ^2 Q}{\partial \beta _{1l}\partial \gamma _{s}}= & {} -\displaystyle \sum _{I_1}x_{1il}\eta _ip_iF_s^{\prime }(t_i) - \displaystyle \sum _{I_0}\pi _i^{(k)}\frac{x_{1il}\eta _ip_iF_s^{\prime }(t_i)}{C_i}\left\{ 1-\frac{\eta _ip_iS(t_i)(1-C_i)}{C_i}\right\} ,\\ \frac{\partial ^2 Q}{\partial \beta _{2k}\partial \gamma _{s}}= & {} -\displaystyle \sum _{I_1}x_{2ik}\eta _ip_i(1-p_i)F_s^{\prime }(t_i) - \displaystyle \sum _{I_0}\pi _i^{(k)}\frac{x_{2ik}\eta _ip_i(1-p_i)F_s^{\prime }(t_i)}{C_i}\\&\times \left\{ 1-\frac{\eta _ip_iS(t_i)(1-C_i)}{C_i}\right\} ,\\ \frac{\partial ^2 Q}{\partial \gamma _1\partial \gamma _2}= & {} \displaystyle \sum _{I_1}\left\{ \frac{(\gamma _2t_i)^{\frac{1}{\gamma _1}}\log (\gamma _2t_i)}{\gamma _1^3\gamma _2}-\frac{H_i}{\gamma _1} - \eta _ip_iF_{12}^{\prime \prime }(t_i)\right\} - \displaystyle \sum _{I_0}\pi _i^{(k)}\frac{\eta _ip_iF_{12}^{\prime \prime }(t_i)}{C_i} \\&+\displaystyle \sum _{I_0}\pi _i^{(k)}\frac{\eta ^2_ip^2_iF_1^{\prime }(t_i)F_2^{\prime }(t_i)\left( C_i-1\right) }{C_i^2} \end{aligned}$$

for \(s=1, 2\), where \(E_i\) and \(H_i\) are as defined before with

$$\begin{aligned} B_i= & {} \frac{S_{\text {pop}}(t_i)F(t_i)-p_{0i}}{S_{\text {pop}}(t_i)-p_{0i}}, \quad B_i^*=\frac{S_{\text {pop}}(t_i)F^2(t_i)-p_{0i}}{S_{\text {pop}}(t_i)-p_{0i}},\\ C_i= & {} 1-\exp (-\eta _ip_iS(t_i)),\\ F_1^{\prime }(t_i)= & {} \frac{\partial F(t_i)}{\partial \gamma _1}=-\frac{S(t_i)(\gamma _2t_i)^{\frac{1}{\gamma _1}}\log (\gamma _2t_i)}{\gamma _1^2}, \qquad F_2^{\prime }(t_i)=\frac{\partial F(t_i)}{\partial \gamma _2}=\frac{S(t_i)(\gamma _2t_i)^{\frac{1}{\gamma _1}}}{\gamma _1\gamma _2},\\ F_{11}^{\prime \prime }(t_i)= & {} \frac{\partial ^2 F(t_i)}{\partial \gamma _1^2}=F_1^{\prime }(t_i)\left( E_i-\frac{1}{\gamma _1}\right) , \quad F_{22}^{\prime \prime }(t_i)=\frac{\partial ^2 F(t_i)}{\partial \gamma _2^2}=F_2^{\prime }(t_i)\left( H_i-\frac{1}{\gamma _2}\right) ,\\ F_{12}^{\prime \prime }(t_i)= & {} \frac{\partial ^2 F(t_i)}{\partial \gamma _1\partial \gamma _2}=F_2^{\prime }(t_i)E_i. \end{aligned}$$

1.3 Destructive COM-Poisson cure rate model

$$\begin{aligned} \mathbf Q ({\varvec{\theta }},\varvec{\pi }^{(k)})\propto & {} \displaystyle \sum _{I_1}\left[ \log (p_if(t_i)) + \log z_{f10}-\log \{z(1-p_iF(t_i))\}\right] \\&+\displaystyle \sum _{I_0}(1-\pi _i^{(k)})\left( \log z_p-\log z\right) +\displaystyle \sum _{I_0}\pi _i^{(k)}\log \left( \frac{z_f-z_p}{z}\right) .\\ \frac{\partial Q}{\partial \beta _{1l}}= & {} \displaystyle \sum _{I_1}x_{1il}\left\{ \frac{z_{f20}}{z_{f10}} - \frac{z_1}{z}\right\} +\displaystyle \sum _{I_0}(1-\pi _i^{(k)})x_{1il}\left\{ \frac{z_{p10}}{z_p} - \frac{z_1}{z}\right\} \\&+\displaystyle \sum _{I_0}\pi _i^{(k)} x_{1il}\left\{ \frac{z_{f10} - z_{p10} }{z_f-z_p}-\frac{z_1}{z}\right\} ,\\ \frac{\partial Q}{\partial \beta _{2k}}= & {} \displaystyle \sum _{I_1}x_{2ik}(1-p_i)\left\{ \frac{1}{1-p_iF(t_i)}-p_iF(t_i)\left( \frac{z_{f21}}{z_{f10}}\right) \right\} \\&-\displaystyle \sum _{I_0}(1-\pi _i^{(k)})\frac{x_{2ik}p_i(1-p_i)z_{p11}}{z_p}-\displaystyle \sum _{I_0}\pi _i^{(k)}x_{2ik}p_i(1-p_i) \end{aligned}$$
$$\begin{aligned}&\times \left\{ \frac{ F(t_i)z_{f11} - z_{p11}}{z_f-z_p}\right\} ,\\ \frac{\partial Q}{\partial \gamma _1}= & {} \displaystyle \sum _{I_1}\left\{ E_i - p_iF_1^{\prime }(t_i)\left( \frac{z_{f21}}{z_{f10}} - \frac{1}{1-p_iF(t_i)}\right) \right\} - \displaystyle \sum _{I_0}\pi _i^{(k)}\frac{p_iF_1^{\prime }(t_i)z_{f11}}{z_f-z_p},\\ \frac{\partial Q}{\partial \gamma _2}= & {} \displaystyle \sum _{I_1}\left\{ H_i-p_iF_2^{\prime }(t_i)\left( \frac{z_{f21}}{z_{f10}}-\frac{1}{1-p_iF(t_i)}\right) \right\} -\displaystyle \sum _{I_0}\pi _i^{(k)}\frac{p_iF_2^{\prime }(t_i)z_{f11}}{z_f-z_p},\\ \frac{\partial ^2 Q}{\partial \beta _{1l}\partial \beta _{1l^{\prime }}}= & {} \displaystyle \sum _{I_1}x_{1il}x_{1il^{\prime }}\left\{ \frac{z_{f30}}{z_{f10}}-\left( \frac{z_{f20}}{z_{f10}} \right) ^2+\left( \frac{z_1}{z}\right) ^2-\frac{z_2}{z}\right\} \\&+\displaystyle \sum _{I_0}(1-\pi _i^{(k)})x_{1il}x_{1il^{\prime }}\left\{ \frac{z_{p20}}{z_p} - \left( \frac{z_{p10}}{z_p}\right) ^2+\left( \frac{z_1}{z}\right) ^2-\frac{z_2}{z}\right\} \end{aligned}$$
$$\begin{aligned}&+\displaystyle \sum _{I_0}\pi _i^{(k)} x_{1il}x_{1il^{\prime }}\left\{ \frac{z_{f20} - z_{p20}}{z_f-z_p}-\left( \frac{z_{f10} - z_{p10}}{z_f-z_p}\right) ^2+\left( \frac{z_1}{z}\right) ^2 - \frac{z_2}{z}\right\} ,\\ \frac{\partial ^2 Q}{\partial \beta _{2k}\partial \beta _{2k^{\prime }}}= & {} \displaystyle \sum _{I_1}\frac{x_{2ik}x_{2ik^{\prime }}(1-p_i)^2}{(1-p_iF(t_i))^2}\left\{ p_iF(t_i) - \frac{p_i(1-p_iF(t_i))}{1-p_i}\right\} \\&+\displaystyle \sum _{I_1}\frac{x_{2ik}x_{2ik^{\prime }}p^2_iF(t_i)(1-p_i)z_{f21}}{z_{f10}}\\&+\displaystyle \sum _{I_1}\frac{x_{2ik}x_{2ik^{\prime }}p_iF(t_i)(1-p_i)^2}{z_{f10}}\left\{ p_iF(t_i)\left( z_{f22}-\frac{z_{f21}^2}{z_{f10}}\right) -z_{f21}{}\right\} \\&+\displaystyle \sum _{I_0}(1-\pi _i^{(k)})\frac{x_{2ik}x_{2ik^{\prime }}(1-p_i)^2p_i(p_iz_{p12}-z_{p11})}{z_p}\\&+\displaystyle \sum _{I_0}(1-\pi _i^{(k)})\frac{x_{2ik}x_{2ik^{\prime }}p_i^2(1-p_i)z_{p11}}{z_p}\left\{ 1-(1-p_i)\frac{z_{p11}}{z_p}\right\} \\&-\displaystyle \sum _{I_0}\pi _i^{(k)}\frac{x_{2ik}x_{2ik^{\prime }}p_i(1-p_i)^2}{z_f-z_p}\\&\times \left\{ F(t_i)z_{f11}-z_{p11}-p_i (F^2(t_i)z_{f12} - z_{p12})\right\} \\&+\displaystyle \sum _{I_0}\pi _i^{(k)}\frac{x_{2ik}x_{2ik^{\prime }}p^2_i(1-p_i)(F(t_i)z_{f11}-z_{p11})}{z_f-z_p}\\&-\displaystyle \sum _{I_0}\pi _i^{(k)}\frac{x_{2ik}x_{2ik^{\prime }}p^2_i(1-p_i)^2(F(t_i)z_{f11}-z_{p11})^2}{(z_f-z_p)^2}, \end{aligned}$$
$$\begin{aligned} \frac{\partial ^2 Q}{\partial \gamma _1^2}= & {} -\displaystyle \sum _{I_1}\left\{ \frac{2E_i}{\gamma _1}+\frac{(\log (\gamma _2t_i))^2(\gamma _2t_i)^{\frac{1}{\gamma _1}}}{\gamma ^4_1}+\frac{1}{\gamma ^2_1}+p_iF_{11}^{\prime \prime }(t_i)\frac{z_{f21}}{z_{f10}}\right\} \\&+\displaystyle \sum _{I_1}\left[ p^2_i(F_1^{\prime }(t_i))^2\left\{ \frac{z_{f22}}{z_{f10}} - \left( \frac{z_{f21}}{z_{f10}}\right) ^2\right\} + \frac{p_i}{1-p_iF(t_i)}\right. \\&\left. \times \left\{ F_{11}^{\prime \prime }+ \frac{p_i\left( F_1^{\prime }(t_i)\right) ^2}{1-p_iF(t_i)}\right\} \right] \\&-\displaystyle \sum _{I_0}\pi _i^{(k)}\frac{p_i}{z_f-z_p}\left\{ F_{11}^{\prime \prime }(t_i)z_{f11}-p_i\left( F_1^{\prime }(t_i)\right) ^2\left( z_{f12}-\frac{z_{f11}^2}{z_f-z_p}\right) \right\} , \end{aligned}$$
$$\begin{aligned} \frac{\partial ^2 Q}{\partial \gamma _2^2}= & {} -\displaystyle \sum _{I_1}\left\{ \frac{H_i}{\gamma _2}+\frac{(\gamma _2t_i)^{\frac{1}{\gamma _1}}}{\gamma _1^2\gamma _2^2}+p_iF_{22}^{\prime \prime }(t_i)\frac{z_{f21}}{z_{f10}}\right\} \\&+\displaystyle \sum _{I_1}\left[ p^2_i(F_2^{\prime }(t_i))^2\left\{ \frac{z_{f22}}{z_{f10}}-\left( \frac{z_{f21}}{z_{f10}}\right) ^2\right\} +\frac{p_i}{1-p_iF(t_i)}\right. \\&\times \left. \left\{ F_{22}^{\prime \prime }+\frac{p_i\left( F_2^{\prime }(t_i)\right) ^2}{1-p_iF(t_i)}\right\} \right] \\&-\displaystyle \sum _{I_0}\pi _i^{(k)}\frac{p_i}{z_f-z_p}\left\{ F_{22}^{\prime \prime }(t_i)z_{f11}-p_i\left( F_2^{\prime }(t_i)\right) ^2\left( z_{f12}-\frac{z_{f11}^2}{z_f-z_p}\right) \right\} ,\\ \frac{\partial ^2 Q}{\partial \beta _{1l}\partial \beta _{2k}}= & {} \displaystyle \sum _{I_1}\frac{x_{1il}x_{2ik}p_iF(t_i)(1-p_i)}{z_{f10}}\left( \frac{z_{f21} z_{f20}}{z_{f10}}-z_{f31}\right) \\&+\displaystyle \sum _{I_0}(1-\pi _i^{(k)})x_{1il}x_{2ik}p_i(1-p_i)\left\{ \frac{z_{p11}z_{p10}}{z_p^2} - \frac{z_{p21}}{z_p}\right\} \\&+\displaystyle \sum _{I_0}\pi _i^{(k)}\frac{x_{1il}x_{2ik}p_i(1-p_i)}{\left( z_f-z_p\right) ^2}\left[ z_{f10}\left\{ F(t_i)z_{f11}-z_{p11}\right\} \right. \\&\left. -z_{p10}\left\{ F(t_i)z_{f11}-z_{p11}\right\} \right] \\&-\displaystyle \sum _{I_0}\pi _i^{(k)}\frac{x_{1il}x_{2ik}p_i(1-p_i) (F(t_i)z_{f21}-z_{p21})}{z_f-z_p},\\ \frac{\partial ^2 Q}{\partial \beta _{1l}\partial \gamma _s}= & {} \displaystyle \sum _{I_1}x_{1il}p_iF_s^{\prime }(t_i)\left\{ (1-p_iF(t_i))\left( \frac{z_{f21}}{z_{f10}}\right) ^2 - \frac{z_{f31}}{z_{f10}}\right\} \\&-\displaystyle \sum _{I_0}\pi _i^{(k)}\frac{x_{1il}p_i(1-p_i)F_s^{\prime }(t_i)z_{f11} z_{p11}}{\left( z_f-z_p\right) ^2}\\&+ \displaystyle \sum _{I_0}\pi _i^{(k)}\frac{x_{1il}p_i(1-p_iF(t_i))F_s^{\prime }(t_i)z_{f11}^2}{\left( z_f-z_p\right) ^2}\\&-\displaystyle \sum _{I_0}\pi _i^{(k)}\frac{x_{1il}p_i(1-p_i)F_s^{\prime }(t_i)z_{f21}}{z_f-z_p},\\ \frac{\partial ^2 Q}{\partial \beta _{2k}\partial \gamma _s}= & {} -\displaystyle \sum _{I_1}x_{2ik}p_iF_s^{\prime }(t_i)(1-p_i)\left\{ \frac{z_{f21}}{z_{f10}} - p_iF(t_i)\frac{z_{f22}}{z_{f10}}\right\} \\&-\displaystyle \sum _{I_1}x_{2ik}p_iF_s^{\prime }(t_i)(1-p_i)\left\{ p_iF(t_i)\left( \frac{z_{f21}}{z_{f10}}\right) ^2 - \frac{1}{\left( 1-p_iF(t_i)\right) ^2}\right\} \\&-\displaystyle \sum _{I_0}\pi _i^{(k)}\frac{x_{2ik}p^2_iF_s^{\prime }(t_i)(1-p_i)}{(z_f-z_p)^2}\left\{ F(t_i)z_{f11}^2-z_{f11}z_{p11}\right\} \end{aligned}$$
$$\begin{aligned}&+\displaystyle \sum _{I_0}\pi _i^{(k)}\frac{x_{2ik}p_iF_s^{\prime }(t_i)(1-p_i)}{z_f-z_p}\left\{ p_iF(t_i)z_{f12}-z_{f11}\right\} ,\\ \frac{\partial ^2 Q}{\partial \gamma _1\partial \gamma _2}= & {} \displaystyle \sum _{I_1}\left\{ \frac{p_iF_{12}^{\prime \prime }(t_i)}{(1-p_iF(t_i))}-\frac{H_i}{\gamma _1}\right\} \\&+\displaystyle \sum _{I_1}\left\{ \frac{(\gamma _2t_i)^{\frac{1}{\gamma _1}}\log (\gamma _2t_i)}{\gamma ^3_1\gamma _2} + \frac{p^2_iF_1^{\prime }(t_i)F_2^{\prime }(t_i)}{(1-p_iF(t_i))^2}\right\} \\&+\displaystyle \sum _{I_1}p^2_iF_1^{\prime }(t_i)F_2^{\prime }(t_i)\frac{z_{f22}}{z_{f10}}-\displaystyle \sum _{I_1}p_iF_{12}^{\prime \prime }(t_i)\frac{z_{f21}}{z_{f10}}\\&-\displaystyle \sum _{I_1}p^2_iF_1^{\prime }(t_i)F_2^{\prime }(t_i)\left( \frac{z_{f21}}{z_{f10}}\right) ^2\\&+\displaystyle \sum _{I_0}\pi _i^{(k)}\frac{p^2_iF_1^{\prime }(t_i)F_2^{\prime }(t_i)}{z_f-z_p}\left\{ z_{f12} - \frac{z_{f11}^2}{z_f-z_p}\right\} -\displaystyle \sum _{I_0}\pi _i^{(k)}\frac{p_iF_{12}^{\prime \prime }(t_i)z_{f11}}{z_f-z_p} \end{aligned}$$

for \(s=1, 2\), where \(E_i, \;H_i,\;F_{1}^{\prime }(t_i), \;F_{2}^{\prime }(t_i), \;F_{11}^{\prime \prime }(t_i), \;F_{12}^{\prime \prime }(t_i) \; \text {and} \; F_{22}^{\prime \prime }(t_i)\) are all as defined before with

$$\begin{aligned} z= & {} Z(\eta _i ,\phi ),\quad z_p=Z[\eta _i(1-p_i),\phi ],\quad z_F=Z[\eta _i(1-p_iF(t_i)),\phi ],\\ z_{2}= & {} \sum ^{\infty }_{j=1}\frac{j^2\eta _i^j}{(j!)^{\phi }},\quad z_{1}=\sum ^{\infty }_{j=1}\frac{j\eta _i^j}{(j!)^{\phi }},\\ z_{p10}= & {} \sum ^{\infty }_{j=1}\frac{j[\eta _i(1-p_i)]^j}{(j!)^{\phi }},\\ z_{p11}= & {} \sum ^{\infty }_{j=1}\frac{j\eta _i^j(1-p_i)^{j-1}}{(j!)^{\phi }},\quad z_{p20}=\sum ^{\infty }_{j=1}\frac{j^2[\eta _i(1-p_i)]^j}{(j!)^{\phi }},\\ z_{p21}= & {} \sum ^{\infty }_{j=1}\frac{j^2\eta _i^j(1-p_i)^{j-1}}{(j!)^{\phi }},\\ z_{p12}= & {} \sum ^{\infty }_{j=2}\frac{j(j-1)\eta _i^j(1-p_i)^{j-2}}{(j!)^{\phi }},\quad z_{f10}=\sum ^{\infty }_{j=1}\frac{j[\eta _i(1-p_iF(t_i))]^j}{(j!)^{\phi }},\\ z_{f11}= & {} \sum ^{\infty }_{j=1}\frac{j\eta _i^j(1-p_iF(t_i))^{j-1}}{(j!)^{\phi }},\\ z_{f20}= & {} \sum ^{\infty }_{j=1}\frac{j^2[\eta _i(1-p_iF(t_i))]^j}{(j!)^{\phi }},\quad z_{f21}=\sum ^{\infty }_{j=1}\frac{j^2\eta _i^j(1-p_iF(t_i))^{j-1}}{(j!)^{\phi }},\\ z_{f12}= & {} \sum ^{\infty }_{j=2}\frac{j(j-1)\eta _i^j(1-p_iF(t_i))^{j-2}}{(j!)^{\phi }},\\ z_{f22}= & {} \sum ^{\infty }_{j=2}\frac{j^2(j-1)\eta _i^j(1-p_iF(t_i))^{j-2}}{(j!)^{\phi }},\quad z_{f31}=\sum ^{\infty }_{j=1}\frac{j^3\eta _i^j(1-p_iF(t_i))^{j-1}}{(j!)^{\phi }},\\ z_{f30}= & {} \sum ^{\infty }_{j=1}\frac{j^3[\eta _i(1-p_iF(t_i))]^j}{(j!)^{\phi }}. \end{aligned}$$

Rights and permissions

Reprints and permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Pal, S., Majakwara, J. & Balakrishnan, N. An EM algorithm for the destructive COM-Poisson regression cure rate model. Metrika 81, 143–171 (2018). https://doi.org/10.1007/s00184-017-0638-8

Download citation

  • Received:

  • Published:

  • Issue Date:

  • DOI: https://doi.org/10.1007/s00184-017-0638-8

Keywords

Navigation