Abstract
Background
Survival analysis methods such as the KaplanMeier method, logrank test, and Cox proportional hazards regression (Cox regression) are commonly used to analyze data from randomized withdrawal studies in patients with major depressive disorder. However, unfortunately, such common methods may be inappropriate when a longterm censored relapsefree time appears in data as the methods assume that if complete followup were possible for all individuals, each would eventually experience the event of interest.
Methods
In this paper, to analyse data including such a longterm censored relapsefree time, we discuss a semiparametric cure regression (Cox cure regression), which combines a logistic formulation for the probability of occurrence of an event with a Cox proportional hazards specification for the time of occurrence of the event. In specifying the treatment's effect on diseasefree survival, we consider the fraction of longterm survivors and the risks associated with a relapse of the disease. In addition, we develop a treebased method for the time to event data to identify groups of patients with differing prognoses (cure survival CART). Although analysis methods typically adapt the logrank statistic for recursive partitioning procedures, the method applied here used a likelihood ratio (LR) test statistic from a fitting of cure survival regression assuming exponential and Weibull distributions for the latency time of relapse.
Results
The method is illustrated using data from a sertraline randomized withdrawal study in patients with major depressive disorder.
Conclusions
We concluded that Cox cure regression reveals facts on who may be cured, and how the treatment and other factors effect on the cured incidence and on the relapse time of uncured patients, and that cure survival CART output provides easily understandable and interpretable information, useful both in identifying groups of patients with differing prognoses and in utilizing Cox cure regression models leading to meaningful interpretations.
Background
In a clinical study involving patients with major depressive disorder (MDD), a longterm placebo treatment is not acceptable due to an increased risk of suicide. In such a situation, a randomized withdrawal design is one of the most useful approaches for comparing the longterm efficacy and safety of a drug and a placebo. A typical randomized withdrawal study consists of an initial phase during which all patients are given openlabel active treatment followed by randomization of the responders to continued doubleblind treatment with either active or placebo treatment. The advantage of this study design is the reduced duration of placebo treatment and an earlyescape endpoint, such as a relapse of signs or symptoms of the disease or a lack of efficacy.
Table 1 provides a summary of the design and rates of relapse in MDD patients from several recent randomized withdrawal studies [15]. Common features among the studies include longterm followup, relatively large sample size, high incidence of nonrelapsed patients, and nonexcessive censoring from lost to followup during the period when relapse events often occur. In addition, the KaplanMeier method and logrank test were most often used to examine the efficacy of the test drug compared with a placebo. The results from these randomized withdrawal studies indicate that most individuals with MDD often did not experience relapse events during the treatment period. The proportion of nonrelapse patients ranged from 74.0% to 91.5% in the drug groups and 58.0% to 80.5% in the placebo groups. The proportion of patients who responded favourably to treatment, showing no subsequent signs or symptoms of the disease, were considered "cured", while the remaining patients may eventually "relapse". As a result, a longterm censored relapsefree time may appear in data. In this situation, unfortunately, standard survival analysis methods such as the logrank test and Cox proportional hazards regression (Cox regression) may be inappropriate when analyzing such longterm censored data, as they assume that if complete followup were possible for all individuals, each would eventually experience the event of interest.
Table 1. Study designs and relapse rates from several randomized withdrawal studies in patients with MDD
In this paper, two methods of analysis were considered: a semiparametric cure regression (Cox cure regression) and a treebased method, known as survival CART (Classification And Regression Trees) [6]. The two methods are generally used for the analysis of timetoevent censored data and may provide the findings different from the standard Cox regression by assuming that there are cured individuals in the data. The former method combines a logistic formulation for the probability of occurrence of an event with a proportional hazards specification for the time of occurrence of the event [713], so that the effects of the treatment and other factors can be interpreted separately into those on the proportion of cured patients and the failure time of uncured patients. In the latter method, termed cure survival CART, the probability of occurrence of an event with the exponential or Weibull distribution for the time of the event occurrence is modelled using binary treestructure. Our approach was to incorporate the LR test statistic from the fitting of the cure survival regressions for recursive partitioning procedures, although the standard survival CART method adapts the method of CART paradigm using the logrank statistic [14]. Cure survival CART was performed to supplement the results of Cox cure regression and to provide a simple and useful interpretation. These findings would provide valuable information for the future treatment of patients with MDD. This paper is structured as follows: first, we describe the models, parameter estimations, and algorithms of the two methods. We then illustrate some aspects of the two methods, using data collected in a sertraline randomized withdrawal study in patients with MDD [3], and end with a discussion.
Methods
Cox cure regression
The model
Suppose that the data (T_{i}, Δ_{i}, X_{i}, Z_{i}) are available for an individual i = 1,..., n. T_{i }denotes the time to the occurrence of the event defined by min(, U_{i}), where and U_{i }are the random variables of true survival and censoring, respectively. Δ_{i }denotes the censoring indicator Δ_{i }= I ( = T_{i}), where I(·) is the indicator function. X_{i }and Z_{i }are the covariate vectors related to the cured incidence and uncured survival, respectively.
Let λ_{i}(·) be the hazard function for an individual i. Suppose that λ_{i }holds for the Cox proportional hazards model if individual i is uncured and λ_{i }has zerohazard otherwise. Therefore, we can write
where λ_{0}(t) is the baseline hazard function, β is the pdimensional parameter vector corresponding to Z_{i}, and η_{i }is the indicator η_{i }= 1 if individual i eventually experiences the event (uncured) and η_{i }= 0 if individual i never experiences the event, with the cured incidence c_{i }= Pr(η_{i }= 0  X_{i}). Here we suppose that c_{i }is the logistic model given by
where α is the qdimensional parameter corresponding to X_{i}, which usually consists of the form (1, X_{i1},..., X_{iq1})^{T }with the intercept term X_{i0 }= 1.
Parameter estimation
In general, although observing η_{i }= 0 of being cured is not possible, only η_{i }= 1 of being uncured when Δ_{i }= 1 is known. For this reason, a marginal type of full likelihood considered by Boag [15] is available for Cox cure regression. Then, the logarithm of the marginal full likelihood for Cox cure regression is
where , S_{i}(t) = exp[exp(β^{T }Z_{i})Λ_{0}(t)] and θ = (α^{T}, β^{T})^{T}
The estimate of (θ, Λ_{0}) is obtained by maximizing l_{f }over (θ, Λ_{0}). This maximization is performed using the EM (ExpectationMaximization) algorithm, given a suitable stating value (θ^{(0)}, ) of (θ, Λ_{0}). We prepare a suitable θ^{(0) }using the Monte Carlo method [7,8]. Once we have θ^{(0)}, a suitable corresponding to θ^{(0) }is computed by a NewtonRaphson method discussed in Sugimoto and Hamasaki [13]. Although the optimization technique based on the EM algorithm is computationally fast, it easily fails to converge depending on starting values in Cox cure regression [10]. Therefore, providing an appropriate (θ^{(0)}, ) prudently in advance using these methods, then we go ahead to the EM algorithm.
Using the concept of Taylor [16], the EM algorithm for Cox cure regression has been developed [810]. The mth Estep in the EM algorithm transforms l_{f }into the form for the completedata log likelihood
for a given , where w^{(m) }is computed by current parameter values as and
In the m th Mstep algorithm, it is found that a Breslow's estimate of Λ_{0 }[17],
can be utilized for the maximization of . Substituting into Λ_{0 }of leads to a form of log partial likelihood
Then, the Mstep of over (β, Λ_{0}) is replaced by maximizing only over β in treating Λ_{0 }as a nonparametric function. Therefore, the Mstep is easily performed by maximizing and over α and β, respectively, via a NewtonRaphson method.
Now, if θ^{(m) }= (α^{(m)}, β^{(m}) are the current estimates obtained by the Mstep for a given w^{(m)}, then current estimate of Λ_{0 }is written as . For the next (m +1) th Estep, w^{(m) }is updated to w^{(m+1) }by substituting (θ^{(m}, ) into (θ, Λ_{0}) in (1). Finally, (), which provides becomes an estimate of (θ, Λ_{0}) in a series of the EMiteration. From our limited experiences with real data and simulation, it is recommended that a variety of starting value θ^{(0) }be used to ensure that a global maximum is found. For further discussions, please see Sugimoto and Goto [8], Peng and Dear [9], Sy and Taylor [10] and Sugimoto et al. [12]
Cure survival CART
In general, treebased methods provide classifications of patients with differing prognoses that may help clarify the association of patient characteristics and survival. In this section, a treebased method for censored data including longterm survivors is developed, based on cure survival regression. With respect to the cost of calculations in estimating the parameters, rather than the straightforward use of the semiparametric Cox cure regression discussed in the previous section, the fitting of parametric cure regression is more reasonable, where the exponential distribution is assumed to be an underlying distribution for the latency time of relapse. However, since an assumption of exponential distribution is restrictive, the Weibull distribution is also considered. First, the model will be described and then the algorithm will be discussed.
The model
It is assumed that the treated patient population is composed of both cured and uncured patients. A sample consists of vectors (T_{i}, Δ_{i}, X_{i}), i = 1,..., n, where (T_{i}, Δ_{i}, X_{i}) are the same as the definitions in Cox cure regression. The full likelihood of the tree constructed in this sample can be described as
where is a set of terminal nodes, {i: X_{i }∈ } is a set of individual labels which belong to node h, and represents a covariate space which provides node h; λ_{h}(t), S_{h }(t), and c_{h }are the hazard function, survival function, and cure rate, respectively, for the individuals in node h. For the Weibull model, we write and with unknown parameters μ_{h }and ρ_{h}, while the exponential case is reduced to those of ρ = 1. w_{h }(T_{i}) is the conditional probability that the i th individual in node h will eventually belong to the uncured category given that no event has occurred by time T_{i}, such that
The major area of interest was the examining of cure rates and the identification of MDD patients with different prognoses. Parameter transformations such as log{c_{h}/(1  c_{h})} = α_{h}, and were adapted and the EM algorithm, described in the previous section, is used to estimate parameters c_{h}, μ_{h }and ρ_{h}. For the Weibull model, it is recommended that a suitable starting point of ρ_{h }is sought in advance.
The algorithm
The treebased method requires the splitting, pruning, and selection of a pruned subtree to be specified [6]. Since highdimensional data may give a complicated treestructure, the following specifications were considered to simplify the tree structure, and the results were subsequently interpreted.
Splitting
Two disjointed treebased graphs are developed by recursively splitting data into two regions. Each split is evaluated for each of the variables, and a single variable (X_{i}) and the split value resulting in the greatest reduction in impurity is selected by likelihoodratio test statistic of cure exponential regression, to best reduce error. When growing trees, the improvement measure for a split (s) at node h into left and right daughter nodes, l (h) and r (h), respectively, is measured by R(s, h) = R (h)  {R(l (h))R +r (h)}, where)
represents the 2 times maximum log likelihood divided by the sample size n (deviance residual) obtained for the data of node ħ = (= h, l(h), r (h)); , and are the maximum likelihood estimates (MLEs) of c_{ħ}, μ_{ħ }and ρ_{ħ }obtained for node ħ,
In the exponential case, holds and the hazard function is simply expressed as
A split satisfying over s ∈ is selected, where the set is composed of all possible binary splitting rules considered at node h. The two new daughter nodes, l (h) and r (h), determined by are generated, and the tree as a whole grows. Such a generation of nodes is performed successively at any node until a prespecified stopping condition (e.g., the maximal number of terminal nodes or the minimal number of observations in terminal nodes) is satisfied.
Pruning
The "Pruning procedure" [6] determines the order how to remove superfluous subtrees of a largest tree to build a good fitting simple treebased model. This process is repeated until only a small fraction of the data remains. The final tree expresses a logical rule representing an extreme outcome group.
Methods based on withinnode deviance are hired for pruning. In the CART algorithm, the performance of a tree is based on the cost complexity measure, defined by
For the binary tree, is the number of terminal nodes, and γ is a penalty per terminal node . A subtree (a tree obtained by removing branches) (γ) is the smallest optimally pruned subtree for any penalty γ of the largest tree if
where indicates that is a subtree of ; and is the smallest optimally pruned subtree such that for every optimally pruned subtree .
The cost complexity pruning algorithm allows the optimally pruned subtree for any γ to be obtained. This algorithm finds the sequence of optimally pruned subtrees by repeatedly deleting branches of the tree for which the average reduction in impurity per split in the branch is smallest, where
The pruning algorithm is necessary for finding optimal subtrees because the number of possible subtrees grows very rapidly as a function of tree size.
Selection of a pruned subtree
The selection of a pruned subtree can be based on a resampling technique (crossvalidation) to correct for overoptimism due to split point optimization [6]. The most popular method for obtaining treebased estimates of prediction error (or expected deviance) is Vfold crossvalidation. The data are divided up into V sets and subsamples of about equal size, and trees are grown from the subsamples . For any γ, an optimally pruned subtree (γ) and MLEs and in are obtained.
The average crossvalidated deviance residual over V subsamples is
Let γ_{k* }be the value of that minimizes R^{cv}(γ) of , k = 0,1,2,⋯. Then a tree corresponding to γ_{k* }is selected.
We adopt V = 10 because a smaller value of V is preferred to V = n in the application of CART [18]. While 10fold crossvalidation is a standard method for selecting tree size, it is subject to considerable variability. Therefore, in the application to this data, we performed 500 replications of 10fold crossvalidation and then determined γ_{k* }to minimize the average of 500 pairs of . We will be able to use the "1SE (Standard Error)" rule [6] to choose a simpler tree, where such an SE is directly estimated by the variation of R^{cv }(γ) with 500 replications in this application.
Results
Study design and major results of a sertraline randomized withdrawal study
A multicenter, placebocontrolled, randomized withdrawal study was used to evaluate the efficacy and safety of sertraline in Japanese patients with MDD [3]. Following a 1week observational period for washout, only those patients who responded after 8 weeks of openlabel sertraline treatment were randomly assigned to receive one of two subsequent 16week doubleblind treatments with sertraline or a placebo. Patients who did not respond after 8 weeks of the openlabel treatment were discontinued from the study. The primary variable was a relapse of the disease during the doubleblind phase, which was defined as either (i) having a score of Hamilton rating scale for depression (HAMD) (17 items) of 18 point or greater and a clinical global impression (compared to baseline of the openlabel phase) of "nochange" or "worse", during two consecutive visits, or (ii) being unable to continue treatment due to insufficient efficacy.
A total of 415 patients were screened, of which 361 received sertraline in the openlabel phase of the study. From this group, a total of 235 were randomized to the sertraline (n_{s }= 117) or placebo (n_{p }= 118) treatment groups during the doubleblind phase. All analyses were based on the intenttotreat principle. Relapse rates of 8.5% (10 of 117) and 19.5% (23 of 118) in the sertraline and placebo treatment groups, respectively were observed, which represented a statistically significant difference (Chisquared test: p = 0.0158). Examination of the time to relapse using the KaplanMeier methods showed that the relapsefree rate curve for the sertraline treatment group was significantly higher than in placebo treatment group throughout the doubleblind phase (logrank test: p = 0.0261) (Figure 1). For more details, please refer to Kamijima et al. [3].
Figure 1. KaplanMeier estimates for time to relapse in the sertraline and placebo treatment groups.
The standard logistic and Cox regressions were used to identify the effect of variables on the incidence of cured and uncured patients, which were not given in Kamijima et al. [3]. Each model included the treatment (Sertraline = 1), baseline HAMD score at openlabel phase (OP), baseline HAMD score at doubleblind phase (DP), gender (Female = 1), age, the number of episodes, duration from the first episode, duration of this episode, interval from the previous episode, and complication of MDD (complication) (Yes = 1) as covariates (Table 2). The results from both analyses suggest significant effects of treatment and gender on relapse events and time to relapse.
Table 2. Result of standard logistic and Cox regressions to a sertraline randomized withdrawal study in patients with MDD
Application of Cox cure regression and cure survival CART to sertraline data
Cox cure regression, including influential covariates for both cured incidence and uncured survivals, provides flexibility in model building. However, this approach may open discussions on the possibility of an overparameterization of models and the identifiability between the parameters of cured incidence and uncured survivals [10], although the parameters of the standard cure model are identifiable in some sense [19]. In our situation, there was more scientific and practical interest in estimating the cured incidence as the objective of the study was to show how well the drug prevents an eventual episode of recurring illness (cured incidence), compared with placebo and, further, how other covariates influence cured incidence. Thus, in selecting the covariates, we considered that, giving priority to cured incidence over uncured survivals, a minimal number or no covariate in uncured survivals may be appropriate, and then we suggested the following guidelines on variable selection in our situation: (1) treatment is always factored into cured incidence, (2) either a minimal number or no covariates are included into uncured survivals, (3) a covariate already included into cured incidence is not included into uncured survivals. Following these guidelines, the "best" subset of all possible combinations of covariates can be selected by a minimum value of Akaike's information criterion (AIC) [20], given by is (θ, Λ_{0}) that maximizes l_{f }(θ, Λ_{0}), discussed in the previous Methods section.
The result of Cox cure regression after variable selection is summarized in Table 3. The value of AIC suggested the best subset of treatment and baseline HAMD score at DP, which were included into cured incidence, and gender and complication were included into uncured survivals. For the cured incidence, the positive estimate for treatment indicated a higher cure rate for patients who received sertraline treatment, while the negative estimate for baseline HAMD score at DP indicated a lower cured incidence for patients with a higher value of the score. In contrast, for uncured survivals, the positive estimate for gender indicated an earlier occurrence of relapse for female patients, and the negative estimate for complication indicated a prolonged occurrence of relapse for patients with any complications of MDD.
Table 3. Result of best subset for Cox cure regression
The result of best subset of covariates for standard logistic and Cox regressions with the minimum value of AIC is shown in Table 4. Standard logistic and Cox regressions suggested the effects of treatment and gender, which were in agreement with the result of Cox cure regression. In addition to the baseline HAMD score at DP and complication, the standard logistic regression indicated a weak effect on the number of episodes. On the other hand, standard Cox regression was not able to detect the effect of baseline HAMD score at DP, but it indicated a weak effect on the number of episodes. This was a major discrepancy between Cox cure regression and standard Cox regression. Compared with the best subset of covariates of standard logistic and Cox regressions, Cox cure regression suggested the covariates that were more important for cured incidence but less important for the uncured survivals, and vice versa. The subset for Cox cure regression provided a smaller value of AIC than that of standard Cox regression. The standard Cox regression is a special case of Cox cure regression with an infinitely small intercept in cured incidence [10]. The estimate and SE for the intercept were 1.571 and 2.870, respectively. These results support the use of Cox cure regression rather than standard Cox regression.
Table 4. Result of best subset for standard logistic and Cox regressions
Next, the cure survival CART was used to identify groups of patients with differing prognoses, and to refine the model previously obtained by Cox cure regression with meaningful interpretation. Four covariates (treatment, baseline HAMD score at DP, gender and complication), which were identified by Cox cure regression, were selected to characterize the time to relapse for the sertraline data. In the application of the cure survival CART with exponential distribution, based on the average of 500 replications, the minimum crossvalidated deviance residual was R^{cv }(γ_{k*}) = 1.753 (SE = 0.083), and the corresponding tree was quite large. The 1SE rule was used to choose a simpler tree, shown in Figure 2, which provides the crossvalidated estimate of R^{cv }(γ) = 1.791. Similarly, for the cure survival CART with Weibull distribution, the minimum crossvalidated estimate based on the average of 500 replications is R^{cv }(γ_{k*}) = 1.981 (SE = 0.229), and the corresponding tree is shown in Figure 3. As seen in Figure 2, the cure survival CART with exponential distribution showed that the primary split was gender and the secondary was treatment for the subgroup of female patients. Further partitioning of the tree was based on the baseline HAMD score at DP of 6 points for the subgroup of female patients assigned to the sertraline treatment group. On the other hand, as seen in Figure 3, the cure survival CART with Weibull distribution provided a simpler tree structure compared with that seen in the cure survival CART with exponential distribution; there was no partitioning of the tree based on the baseline HAMD score at DP for the subgroup of female patients assigned to the sertraline treatment group. By comparison with the crossvalidated estimates between exponential and Weibull distributions, we could find that the cure survival CART with exponential distribution provided a better fit than that with Weibull distribution.
Figure 2. Tree structure determined by cure survival CART analysis with exponential distribution.
Figure 3. Tree structure determined by cure survival CART analysis with Weibull distribution.
From the results obtained by the two cure survival CARTs, refined Cox cure regression was reperformed; the model included treatment, the baseline HAMD score at DP and the interaction between the treatment and the baseline HAMD score at DP into cured incidence, and gender and complication into uncured survivals. The baseline HAMD score at DP was categorized into two groups: baseline HAMD score at DP >6 ( = 1) and baseline HAMD score at DP ≤ 6 (= 0).
The result of refined Cox cure regression is shown in Table 5. For cured incidence, a major difference between original and refined Cox cure regressions was a significant effect of the interaction between the treatment and the categorized baseline HAMD score at DP. Refined Cox cure regression provided the negative estimates for the interaction, which indicated a higher cure rate for patients with a lower value of the score who received sertraline treatment, but a lower cure rate for patients with a higher value of the score who received the placebo. For uncured survival, there was no major difference between the original and refined Cox cure regressions. Also, as refined Cox cure regression provided a smaller AIC value than original Cox cure regression, refined Cox cure regression would lead to an improved fit. Figure 4 shows estimated curves for the time to relapse for each combination of the treatment and categorized baseline HAMD score at DP, adjusted by gender and complication, using refined Cox regression. The important differences in cured incidence were observed between the sertraline and placebo groups with regard to the baseline HAMD score.
Discussion
The results obtained by Cox cure regression and cure exponential CART agree with findings reported by several authors [2123]. For example, Nierenberg et al. [23] reported that a greater number of residual symptoms and higher HAMD scores are associated with a higher probability of relapse. Although there are several inconsistencies regarding the gender difference in the course of the relapse, Kuehner [21] reported that female patients have a higher risk of earlier occurrence of relapse. In addition, comorbidity of MMD with other illnesses has been widely reported [22]. The variable of complication used in the analysis was a binary variable of "yes" or "no", but original data included more detailed information on the number and type of complications for each patient. Therefore, further investigation on the effect of the number and type of complications on uncured survivals (or cured incidence) is necessary. Furthermore, the cutoff point for the baseline HAMD score at DP at 6 points suggested by cure survival CART was nearly equal to the HAMD definition of full remission (a score of 7 points or less). It is generally considered that the final goal of treatment for MDD is to achieve and maintain remission and the prevention of relapse.
Cox cure regression (or survival cure CART) generally requires a longterm followup [10,24]. The sertraline study discussed in this paper was intended to contribute to a new drug application in Japan, so the length of the followup period (16 weeks) in the study was minimized to merely detect the drug's effect, in order to reduce the duration of unnecessary exposure of patients to the drug or placebo. However, in the application of Cox cure regression to real data, before the formal analysis, an assessment of whether or not the length of followup is sufficient would be useful for interpreting the result. To confirm this for the sertraline data, the q_{n}test discussed by Maller and Zhou [24] was performed for the sertraline and placebo groups, constructed by the estimated cure rates and censoring distribution for this data. For the sertraline group, the observed value of 0.08547 of q_{n }was between 94% and 96% critical points of the test, which supported that the length of followup for the sertraline group was acceptably minimal and the data had levelled off. On the other hand, for the placebo group, the observed value of 0.0085 of q_{n }was much smaller than the value of 0.068 for the 95% point, which did not support that the length of followup for the placebo group was sufficient and that the data had levelled off. According to the results of two q_{n}tests, we could conclude that the length of followup period in the sertraline study was sufficient at least to detect the drug's effect compared with placebo.
In variable selection for the fitting of Cox cure regression to the sertraline data, we gave priority to cured incidence over uncured survivals. However, if there was more scientific interest in when the illness may recur rather than in the eventual cure, giving priority to uncured survivals over cured incidence could be appropriate. There are several aspects of variable selection, depending on the applications of interest.
In the paper, we discussed the two methods of cure Cox regression and cure survival CART. As described in Method section, the former method is a semiparametric regression, but the latter method use a parametric cure regression. Although the cure survival CART output provided information in refining Cox cure regression leading to meaningful interpretations for the sertraline data, note that there is the potential inconsistency between the two regressions when they consider both for data as the estimates from the semiparametric Cox cure regression and those from the parametric Cox cure regression are usually sensitive to the baseline distribution assumption. Our future challenge is to develop the semiparametric cure survival CART with fewer amounts of computations.
Conclusions
In this study, a semiparametric cure regression was used to investigate the latency time of recurrence observed in a sertraline randomized withdrawal study in patients with MDD. In specifying the treatment's effect on diseasefree survival, account was taken of the fraction of longterm survivors and the risks associated with the relapse of the disease. In addition, a treebased method, i.e., the cure survival CART, was used to analyze the time to event data in order to identify groups of patients with differing prognoses. The following are the main findings: (1) Cox cure regression reveals facts on who may be cured, and how the treatment and other factors effect on the cured incidence and on the relapse time of uncured patients. (2) Cure survival CART output provides easily understandable and interpretable information, useful both in identifying groups of patients with differing prognoses and in utilizing Cox cure regression leading to meaningful interpretations.
The methods discussed in this paper could be applied to the development of stratification schemes for future clinical studies and the identification of patients suitable for studies involving therapy targeted at a specific prognostic group. This is would be beneficial as it is often desirable to understand the correlation between a patient's characteristics and relapse times to aid in the design of clinical studies.
Abbreviations
CART: Classification and regression trees; CI: Confidence interval; Doubleblind phase: DP; EM: ExpectationMaximization; HAMD: Hamilton rating scale for depression; Likelihood ratio: LR; MDD: Major depressive disorder; MLE: Maximum likelihood estimate; Openlabel phase: OP; SE: Standard error.
Competing interests
Tomoyuki Sugimoto and Yuko Ohno declare no competing interests. Ichiro Arano is an employee of Pfizer Japan, and Toshimitsu Hamasaki was an employee of Pfizer Japan between 1997 and 2004.
Authors' contributions
ST developed the methods and FORTRAN programs to perform the methods. IA performed the analyses supervised by TS, TH and YO. IA and TH wrote the paper with contributions from ST. All authors read and approved the final manuscript.
Acknowledgements
The authors are grateful to two referees and editors for their constructive comments and valuable suggestions. The authors also thank Pfizer Japan for giving permission to use the data from the sertraline randomized withdrawal trial for this research. This research was financially supported by a program grant from The Ministry of Education, Culture, Sports, Science and Technology, Japan (GrantinAid for Scientific Research (C), NO.20500255) and The 2008 Inamori Research Grants Program.
References

Rapaort MH, Bose A, Zheng H: Escitalopram continuation treatment prevents relapse of depressive episodes.
J Clin Psychiatry 2004, 65:4449. Publisher Full Text

Keller MB, Ruwe FJL, Janssens CJJG, Sitsen JMAD, Jokinen R, Janczewski J: Relapse prevention with Gepirone ER in outpatients with major depression.

Kamijima K, Burt T, Cohen G, Arano I, Hamasaki T: A placebocontrolled, randomized withdrawal study of sertraline for major depressive disorder in Japan.
Int Clin Psychopharmacol 2006, 21:19. PubMed Abstract  Publisher Full Text

Perahia DG, Gilaberte I, Wang F, Wiltse CG, Huckins SA, Clemens JW, Montgomery SA, Montejo AL, Detke MJ: Duloxetine in the prevention of relapse of major depressive disorder: Doubleblind placebocontrolled study.
Br J Psychiatry 2006, 188:346353. PubMed Abstract  Publisher Full Text

Kocsis JH, Thase ME, Trivedi MH, Shelton RC, Kornstein SG, Nemeroff CB, Friedman ES, Gelenberg AJ, Dunner DL, Hirschfeld RMA, Rothschild AJ, Ferguson JM, Schatzberg AF, Zajecka JM, Pedersen RD, Yan B, Ahmed S, Musgnung J, Ninan PT, Keller MB: Prevention of recurrent episodes of depression with Venlafaxine ER in a 1Year maintenance phase from the PREVENT study.
J Clin Psychiatry 2007, 68:10141023. PubMed Abstract  Publisher Full Text

Breiman L, Freidman JH, Olshen RA, Stone CJ: Classification and Regression Trees. Belmont, CA: Wadsworth; 1984.

Kuk AYC, Chen CH: A mixture model combining logistic regression and proportional hazards regression.
Biometrika 1992, 79:531541. Publisher Full Text

Sugimoto T, Goto M: Adjustment of the proportional hazards model via characteristics of cases and its inference.
Bull Comput Statist Jpn 1999, 12:85106.
(in Japanese)

Peng Y, Dear KBG: A nonparametric mixture model for cure rate estimation.
Biometrics 2000, 56:237243. PubMed Abstract  Publisher Full Text

Sy JP, Taylor JMG: Estimation in a Cox proportional hazards cure model.
Biometrics 2000, 56:227236. PubMed Abstract  Publisher Full Text

Lu W, Ying Z: On semiparametric transformation cure models.
Biometrika 2004, 91:331343. Publisher Full Text

Sugimoto T, Hamasaki T, Goto M: Estimation from pseudo partial likelihood in a semiparametric cure model.
Journal of the Japanese Society of Computational Statistics 2005, 18(1):3346.

Sugimoto T, Hamasaki T: Properties of estimators for baseline hazard functions in a semiparametric cure model.
Ann Inst Statist Math 2006, 58:647674. Publisher Full Text

Segal MR: Regression trees for censored data.
Biometrics 1988, 44:3548. Publisher Full Text

Boag JW: Maximum likelihood estimations of the proportion of patients cured by cancer therapy.

Taylor JMG: Semiparametric estimation in failure time mixture models.
Biometrics 1995, 51:899907. PubMed Abstract  Publisher Full Text

Breiman L: Heuristics of instability and stabilization in model selection.
Ann Statist 1996, 24:23502383. Publisher Full Text

Li CS, Taylor JMG, Sy JP: Identifiability of cure models.
Stat Prob Lett 2001, 54:389395. Publisher Full Text

Akaike H: A new look at the statistical model identification.
IEEE Tran Auto Cont 1974, 19:716723. Publisher Full Text

Kuehner C: Gender differences in the shortterm course of unipolar depression in a followup sample of depressed inpatients.
J Affect Disord 1999, 56:127139. PubMed Abstract  Publisher Full Text

Evans DL, Charney DS, Lewis L, Golden RN, Gorman JM, Krishnan KR, Nemeroff CB, Bremner JD, Carney RM, Coyne JC, Delong MR, FrasureSmith N, Glassman AH, Gold PW, Grant I, Gwyther L, Ironson G, Johnson RL, Kanner AM, Katon WJ, Kaufmann PG, Keefe FJ, Ketter T, Laughren TP, Leserman J, Lyketsos CG, McDonald WM, McEwen BS, Miller AH, Musselman D, O'Connor C, Petitto JM, Pollock BG, Robinson RG, Roose SP, Rowland J, Sheline Y, Sheps DS, Simon G, Spiegel D, Stunkard A, Sunderland T, Tibbits P Jr, Valvo WJ: Mood disorders in the medically ill: scientific review and recommendations.
Biol Psychiatry 2005, 58:175189. PubMed Abstract  Publisher Full Text

Nierenberg AA, Husain MM, Trivedi MH, Fava M, Warden D, Wisniewski SR, Miyahara S, Rush AJ: Residual symptoms after remission of major depressive disorder with citalopram and risk of relapse: a STAR*D report.
Psychol Med 2009, 40:4150. PubMed Abstract  Publisher Full Text

Maller R, Zhou X: Survival Analysis with LongTerm Survivors. Chichester: John Wily & Sons; 1996.
Prepublication history
The prepublication history for this paper can be accessed here: