Optimal timing of allogeneic hematopoietic stem cell transplantation in patients with myelodysplastic syndrome

Allogeneic hematopoietic stem cell transplantation (HSCT) represents the only curative treatment for patients with myelodysplastic syndrome (MDS), but involves non-negligible morbidity and mortality. Registry studies have shown that advanced disease stage at transplantation is associated with inferior overall survival. To define the optimal timing of allogeneic HSCT, we carried out a decision analysis by studying 660 patients who received best supportive care and 449 subjects who underwent transplantation. Risk assessment was based on both the International Prognostic Scoring System (IPSS) and the World Health Organization classification-based Prognostic Scoring System (WPSS). We used a continuous-time multistate Markov model to describe the natural history of disease and evaluate the effect of allogeneic HSCT on survival. This model estimated life expectancy from diagnosis according to treatment policy at different risk stages. Relative to supportive care, estimated life expectancy increased when transplantation was delayed from the initial stages until progression to intermediate-1 IPSS-risk or to intermediate WPSS-risk stage, and then decreased for higher risks. Modeling decision analysis on WPSS versus IPSS allowed better estimation of the optimal timing of transplantation. These observations indicate that allogeneic HSCT offers optimal survival benefits when the procedure is performed before MDS patients progress to advanced disease stages. Am. J. Hematol. 88:581–588, 2013. © 2013 Wiley Periodicals, Inc.


Descriptive statistics and survival analysis
Numerical variables are summarized by median and range, categorical variables by count and relative frequency (%) in each category. Survival analyses were performed with the Kaplan--Meier product limit method. For patients in the Pavia cohort, overall survival (OS) was defined as the time (months) between the date of diagnosis and the date of death (for cases) or last follow up (for censored patients). Patients who underwent allogeneic transplantation, acute myeloid leukemia (AML)--like chemotherapy or treatment with hypomethylating agents were censored at the time of the therapeutic procedure. IPSS and WPSS risk were analyzed as time--dependent variables. Comparisons between Kaplan--Meier curves were made using the Gehan Wilcoxon test. A competing risk analysis (Kalbfleisch--Prentice) was performed to estimate the cumulative incidence of disease progression to a higher risk category, of receiving allogeneic HSCT and of death for all causes in each IPSS and WPSS category.
In the GITMO cohort, OS was defined as the time between transplantation and death (from any cause) or last follow up (for censored observations). When estimating non--relapse mortality (NRM), any death in the absence of disease relapse was considered an event. The probability of relapse was estimated considering treatment as a failure at the time of hematologic relapse according to standardized criteria. The cumulative incidence of relapse and NRM was estimated with a competing risks approach (Kalbfleisch--Prentice).
Analyses were performed using Stata 11.2 SE software (StataCorp LP, College Station, TX, USA).

Markov model
A multi--state model describes how an individual moves between a series of states in time.
Markov models are multi--state models based on Markov processes, i.e. a stochastic process (a mathematical model for a random development in time) with the property that the probability of moving to a particular state in the future only depends on the present state, and not on past states. In other words, in Markov models the past influences the future only via the present. This may seem too strong an assumption, but in many instances it is at least plausible and, above all, it allows a relatively simple implementation of multi--state models.
Markov processes are frequently used to model chronic diseases, because model states have a natural interpretation in terms of staged progression. A commonly--used model represents a series of successively more severe disease stages or states, and an "absorbing" state, often death. The patient enters the model at a given time 0 (e.g., the time of diagnosis) and then may advance into or recover from adjacent disease stages, or die at any disease stage. The "life" of a patient is the time spent jumping between states, i.e. the time elapsed between time 0 and the patient's final transition to the absorbing state. By iterating the model a large number of times, it is possible to estimate the life expectancy of a patient.
In discrete--time Markov processes, transition from one state to the next is only allowed at discrete time--points, and a probability of transition is estimated for each allowed transition and for each time point. Discrete--time Markov processes are commonly used as decision models because they can be implemented rather easily. However, the discretization of time (e.g. allowing "jumps" between states only at fixed time points such as every given number of days, months or years) may be an oversimplification of the underlying process.
Continuous--time multi--state models allow transitions between states to take place at any point in time, and not just at the start of a discrete cycle (such as a month or year). The next state to which the individual moves, and the time of the change, are governed by a set of transition intensities, which may also depend on the time of the process, or more generally on a set of individual--specific or time--varying explanatory variables. In more detail, observations of the state Sn(t) are made on a number of individuals n at arbitrary times t, which may vary between individuals. The stages of disease may be modeled as a homogeneous continuous-time Markov process governed by a matrix Q of transition intensities qij representing the instantaneous risk of progression from the i th to the j th state. Under this model, the time spent in state i has an exponential distribution with mean --1/qii and the probability that the next state is j is --qij/qii. Multi--state Markov models may be fitted to data with irregular observation times using freely available software (Jackson CH. Multi--state models for panel data: The msm package for R. J Stat Soft. 2011;38(8):1--28).